·
Economia ·
Estatística Aplicada para Finanças
Send your question to AI and receive an answer instantly
Recommended for you
4
Analise de Dividendos Calculo de Media Desvio Padrao e Coeficiente de Variacao
Estatística Aplicada para Finanças
FGV
37
Harvard Case Estatística
Estatística Aplicada para Finanças
FGV
5
Questão Central
Estatística Aplicada para Finanças
FGV
7
Job Task 1 Estatistica Aplicada
Estatística Aplicada para Finanças
FGV
6
Job Task 02 Estatistica
Estatística Aplicada para Finanças
FGV
2
Exercícios Resolvidos de Probabilidade e Estatística: Defeitos em Fabricação e Medidas Estatísticas
Estatística Aplicada para Finanças
FGV
1
Estatística - Cálculo do Tamanho da Amostra e Plano Amostral
Estatística Aplicada para Finanças
UFV
2
Analise do Impacto da Pandemia no Turismo da Bahia - Estudo de Correlação e Modelagem Preditiva
Estatística Aplicada para Finanças
UFV
11
Fórmulas e Testes Estatísticos de Correlação
Estatística Aplicada para Finanças
FAAP
1
Atividade sobre Amostragem e Métodos Estatísticos
Estatística Aplicada para Finanças
UFV
Preview text
EFICIÊNCIA OPERACIONAL NOS PRINCIPAIS BANCOS PÚBLICOS DO BRASIL OPERATIONAL EFFICIENCY IN THE MAIN PUBLIC BANKS OF BRAZIL Carlos Viegas Neto Ângela Rozane Leal de Souza RESUMO O estudo analisa a eficiência operacional dos principais bancos públicos do Brasil no período de 2010 a 2018 utilizando a metodologia de Análise Envoltória de Dados DEA e o Índice de Eficiência Operacional IEO A pesquisa caracterizase como quantitativa descritiva e documental com nove bancos compondo a amostra do estudo A análise da eficiência sob o prisma do IEO revelou dois grupos de bancos com patamares distintos de eficiência Além disso sinalizou uma tendência de melhora da eficiência no período Pela análise DEA verificouse que os bancos Banco do Brasil Banco do Estado de Sergipe SA e Banco do Estado do Pará SA tiveram os melhores resultados ficando na fronteira de eficiência em todos os anos avaliados Ainda pela DEA verificouse os pesos atribuídos às variáveis e o distanciamento dos bancos ineficientes da fronteira de eficiência Na comparação entre as duas ferramentas de análise percebeuse que há resultados análogos e divergentes destacandose a divergência de resultados nos bancos Banco do Estado do Espírito Santo SA e Banco do Estado do Pará SA por conta das variáveis utilizadas O estudo contribui a partir das análises com a gestão de eficiência dos bancos envolvidos pois fornece informações que podem direcionar a atuação dos bancos na busca por melhorias e também para seus stakeholders Palavraschave Eficiência Análise Envoltória de Dados Banco ABSTRACT The study analyzes the operational efficiency of the main Brazilian public banks from 2010 to 2018 using the Data Envelopment Analysis DEA methodology and the Operational Efficiency Index IEO The research is characterized as quantitative descriptive and documentary with nine banks composing the study sample The efficiency analysis from the IEO perspective revealed two groups of banks with different levels of efficiency In addition it signaled a trend of improved efficiency in the period Through the DEA analysis it was verified that the banks Banco do Brasil Banco do Estado de Sergipe SA and Banco do Estado do Pará SA had the best results remaining on the efficiency frontier in all evaluated years Also by the DEA the weights attributed to the variables and the distancing of the inefficient banks from the efficiency frontier were verified In the comparison between the two tools of analysis it was noticed that there are similar and divergent results highlighting the divergence of results in the banks Banco do Estado do Espírito Santo SA and Banco do Estado do Pará SA due to the variables used The study contributes from the analyzes to the Trabalho de Conclusão de Curso apresentado no primeiro semestre de 2019 ao Departamento de Ciências Contábeis e Atuariais da Faculdade de Ciências Econômicas da Universidade Federal do Rio Grande do Sul UFRGS como requisito parcial para obtenção do título de Bacharel em Ciências Contábeis Graduando em Ciências Contábeis na Universidade Federal do Rio Grande do Sul carlosvnetogmailcom Mestra em Ciências Contábeis pela Universidade do Vale do Rio dos Sinos UNISINOS Doutora em Agronegócio pela Universidade Federal do Rio Grande do Sul UFRGS Professora do Departamento de Ciências Contábeis e Atuarias da UFRGS 2 efficiency management of the banks involved since it provides information that can guide the performance of the banks in the search for improvements and also for their stakeholders Keywords Efficiency Data Envelopment Analysis Bank 1 INTRODUÇÃO A intermediação financeira exercida predominantemente pelos bancos possui papel central na economia brasileira CLEMENTE KÜHL 2006 Os cinco bancos com maior participação de mercado movimentaram o total de 28 trilhões de reais em operações de crédito durante o ano de 2017 DEPARTAMENTO INTERSINDICAL DE ESTATÍSTICAS E ESTUDOS SOCIOECONÔMICOS DIEESE 2017 Esse valor representa quase a metade do PIB brasileiro no mesmo ano o que denota a relevância do segmento O mercado bancário é composto por instituições públicas e privadas que atuam no varejo Os bancos públicos no Brasil possuem variados tamanhos e alguns deles abrangência nacional Embora sejam bancos com viés social e por vezes atuem como agentes de políticas de estado precisam ser competitivos em seus processos para fazer frente à forte concorrência do segmento e garantir os resultados esperados ARAÚJO CINTRA 2011 Essas instituições possuem o nível operacional com amplo espectro de atuação bancos múltiplos Segundo Fortuna 2011 os bancos múltiplos operam e administram várias carteiras de clientes sendo que uma delas deve obrigatoriamente ser comercial ou de investimentos Nesse contexto para alcançar seus clientes esses bancos tradicionalmente utilizamse de agências e correspondentes bancários espalhados por todo o país Ocorre entretanto que os canais de atendimento que não exigem a estrutura física de um ponto de venda passaram a ser muito utilizados pelas empresas financeiras Não por acaso verificase um forte movimento de encerramento de agências físicas BANCO CENTRAL DO BRASIL BACEN 2017 Mais que isso com o advento das fintechs financeiras de tecnologia as quais fornecem atendimento e realizam negócios com baixos custos operacionais exclusivamente por plataformas virtuais tornouse imperioso que os cuidados com a eficiência operacional subissem o patamar de importância nos bancos convencionais No modelo de negócios de pagamentos por exemplo as fintechs captam clientes oriundos dos bancos tradicionais sem necessitar de grandes investimentos ARAÚJO 2018 Tratar de eficiência conforme Périco Rebelatto e Santana 2008 significa obter ganhos de produtividade sem aumentar a quantidade de dispêndios Ou seja é a capacidade de otimizar o atingimento de resultados mediante uso de tecnologia ou melhoria de processos No caso específico dos bancos um indicador importante apontado nos relatórios de estabilidade financeira é o Índice de Eficiência Operacional IEO BACEN 2016 o qual informa o quanto o banco precisa dispender para gerar receitas Assim diante do crescimento da concorrência no mercado de negócios financeiros e a evolução tecnológica do segmento a eficiência nos bancos públicos ganha relevo e a seguinte questão é levantada qual o comportamento dos resultados obtidos pelos principais bancos públicos do Brasil no tocante à eficiência operacional no período de 2010 a 2018 Ante essa indagação o estudo tem por objetivo analisar a eficiência operacional dos principais bancos públicos do Brasil no período de 2010 a 2018 utilizando a metodologia Análise Envoltória de Dados DEA e o IEO A avaliação da eficiência nas organizações é objeto de estudo recorrente na academia e no meio empresarial BERGER MESTER 1997 Isso acontece porque a melhoria contínua nos processos das empresas é condição para sua perenidade Existem contudo variados métodos para avaliação de produtividade dentre os quais está a DEA Essa técnica utilizase 3 de recursos matemáticos para avaliar a obtenção de resultados por unidades produtivas a partir de processos tecnológicos parecidos CASADO 2007 Tratandose de bancos públicos o tema ganha destaque por se tratar de interesse difuso Araújo e Cintra 2011 explicam que os bancos públicos exercem papel diferenciado e entre outras funções têm o papel de atuar na oferta anticíclica de crédito Assim o estudo contribui com a gestão de eficiência dos bancos e fornece informação útil não só para as organizações envolvidas mas também para seus stakeholders 2 REFERENCIAL TEÓRICO Nesta seção é apresentada a configuração do sistema financeiro nacional bem como o papel dos bancos nesse contexto Os conceitos associados à eficiência em todos os ambientes de modo geral e no prisma do mercado bancário são abordados na sequência Depois disso apresentase aspectos relacionados ao conceito e ao modelo de avaliação DEA os quais são utilizados como ferramenta para a pesquisa em apreço Por fim são apresentados outros estudos que buscaram investigar a eficiência bancária com enfoques semelhantes à presente pesquisa 21 SISTEMA FINANCEIRO O principal instrumento regulatório do sistema financeiro nacional é a Lei n 4595 de 31 de dezembro de 1964 Nela é estabelecido os principais atores do sistema financeiro a saber Conselho Monetário Nacional CMN Banco Central do Brasil BACEN Banco do Brasil SA Banco Nacional do Desenvolvimento Econômico e demais instituições financeiras públicas e privadas Todas essas organizações em conjunto são responsáveis pelo adequado funcionamento desse sistema o qual possibilita as diversas transações eou operações financeiras entre os participantes do mercado Fortuna 2011 p 16 define sistema financeiro como sendo Um conjunto de instituições que se dedicam de alguma forma ao trabalho de propiciar condições satisfatórias para a manutenção de um fluxo de recursos entre poupadores e investidores O mercado financeiro onde se processam essas transações permite que um agente econômico qualquerum indivíduo ou empresa sem perspectivas de aplicação em algum investimento próprio da poupança que é capaz de gerar denominado como um agente econômico superavitário seja colocado em contato com outro cujas perspectivas de investimento superam as respectivas disponibilidades de poupança denominado como agente econômico deficitário Em outras palavras podese dizer que é um sistema que permite a circulação de forma regrada confiável e dinâmica de valores financeiros entre os diversos agentes da sociedade Ele é composto por agentes reguladores supervisores e operadores O Quadro 1 ilustra essa configuração 4 Quadro 1 Organização Sistema Financeiro Nacional Moeda Crédito Capitais e Câmbio Seguros Privados Previdência Fechada Reguladores Conselho Monetário Nacional Conselho Nacional de Seguros Privados Conselho Nacional de Previdência Complementar Supervisores Banco Central Comissão de Valores Mobiliários Superintendência de Seguros Privados Superintendência Nacional de Previdência Complementar Operadores Bancos e Caixas Econômicas Administradores de Consórcios Bolsa de Valores Seguradoras e Resseguradoras Fundos de Pensão Cooperativas de Créditos Corretoras e distribuidoras Bolsa de Mercadorias e Futuro Entidades Abertas de Previdência Instituições de Pagamento Demais Instituições não Bancárias Sociedades de Capitalização Fonte adaptado BACEN 2019 Conforme podese ver no Quadro 1 há uma segregação detalhada dos agentes envolvidos no sistema Além de separar as entidades por funções também o faz por área de atuação Nessa divisão os bancos e caixas econômicas estão dentre os operadores do sistema no que tange à moeda crédito capitais e câmbio De maneira geral são essas últimas instituições que materializam o sistema financeiro para a maioria das pessoas e empresas através de seus serviços FORTUNA 2011 Embora ofereçam uma gama vasta de serviços os bancos e também as caixas econômicas que se comportam como bancos múltiplos na atualidade têm como principal papel a intermediação financeira A realização desse serviço possibilita a melhor eficiência do sistema por diversos motivos conforme explica Rosa 2014 Os bancos conseguem reduzir os custos das transações em razão da escala com que operam e com os investimentos em tecnologia para melhorar o processo o que seria inviável se os agentes atuassem individualmente Além disso concentrando as operações em bancos há uma curva de aprendizagem que possibilita reduzir a assimetria de informações Com isso a avaliação de risco tornase qualificada e os recursos captados tendem a ser direcionados de forma otimizada para os agentes deficitários Mesmo a Lei n 459564 sendo o principal instrumento regulatório ela não esgota a legislação sobre o assunto Existem diversas leis que também foram marcos na regulação como as leis 638576 Lei da CVM e 640476 Lei das SA posteriormente alterada pela Lei n 1030301 5 22 EFICIÊNCIA Do ponto de vista econômico um dos grandes desafios enfrentados pelas organizações e pela sociedade como um todo é a adequada utilização dos fatores de produção capital terra trabalho tecnologia tendo em vista a sua disponibilidade limitada e a necessidade de atender demandas insaciáveis Seguindo esse raciocínio conforme Rossetti 2003 p 242 a escassez confrontada com as necessidades ilimitáveis torna onerosos tanto o processo produtivo quanto o de escolha sobre o que e quanto produzir Nessa esteira percebese que os economistas concentram grandes esforços na direção de encontrar opções para melhorar o aproveitamento dos recursos MATTOS TERRA 2015 Tratar os recursos limitados de forma a otimizar os resultados produzidos está associado ao conceito de eficiência o qual pode ser compreendido por diferentes ramificações Algumas dessas maneiras de enxergar a eficiência são classificadas como eficiência técnica eficiência alocativa e eficiência social MATTOS TERRA 2015 No que concerne ao setor bancário é usual trabalhar com a chamada eficiência operacional a qual está associada com despesas de funcionamento da instituição financeira PRIMO et al 2013 Essa medida não é apenas um instrumento de gestão mas sim um dos principais influenciadores para lucratividade de instituições bancárias VENNET 2002 A eficiência operacional enquanto uma relação de despesas operacionais e receitas evidencia o quanto de estrutura operacional é necessária par fazer frente às operações cotidianas ASSAF NETO 2015 Ressaltase contudo conforme lembram Neves Júnior et al 2007 que a eficiência operacional não é calculada de forma uniforme pelos bancos Ainda no tocante à eficiência em bancos Spong Sulivam e DeYoung 1995 entendem que a eficiência passa pela geração de produtos e serviços que gerem renda e ao mesmo tempo sejam realizados com controle de custos Destacam como fatores determinantes também a estrutura organizacional e de pessoas da instituição SPONG SULIVAM DEYOUNG 1995 A natureza jurídica dos bancos público privado ou economia mista também pode influenciar a eficiência da instituição dadas as suas peculiaridades Apesar disso é importante destacar que Altunbas Evans e Molyneux 2001 não percebem uma supremacia dos bancos privados em relação à eficiência considerando seus estudos com bancos alemães Esse mesmo pensamento é compartilhado por Staub Souza e Tabak 2010 os quais realizaram suas pesquisas considerando bancos brasileiros 23 ANÁLISE ENVOLTÓRIA DE DADOS DEA Medidas de eficiência podem ser aferidas por diferentes caminhos Casado 2007 explica que em economia é usual valerse de regressão por mínimos quadrados ordinários sendo portanto um recurso que utiliza uma função de média Uma alternativa a esse modelo é a avaliação com técnicas não paramétricas as quais permitem a obtenção da fronteira eficiente sem a necessidade de pressupostos sobre as variáveis ou funções de produção eou de custo CAMARGO JÚNIOR MATIAS 2005 p 3 Além disso é um instrumento importante para mensurar o desempenho relativo de instituições MAÇADA 2001 Macedo Santos e Silva 2006 acrescentam também que a técnica não paramétrica faz uso de ferramentas de programação linear no lugar de estatística A DEA é uma técnica não paramétrica e é conceituada da seguinte maneira segundo Casado 2007 p 60 A Análise Envoltória de Dados DEA é uma técnica nãoparamétrica que emprega programação matemática para construir fronteiras de produção de unidades produtivas DMUs que empregam processos tecnológicos semelhantes para 6 transformar múltiplos insumos em múltiplos produtos Tais fronteiras são empregadas para avaliar a eficiência relativa dos planos de operação executados pelas DMUs e servem também como referência para o estabelecimento de metas eficientes para cada unidade produtiva É importante observar que as DMUs citadas no conceito são qualquer unidade produtiva podendo ser uma empresa um setor um grupo etc A única condição é que tenham tarefasatividades semelhantes e com os mesmos objetivos para permitir uma comparação consistente conforme esclarecem Mello et al 2005 A avaliação de DMUs utilizando a metodologia DEA pode seguir diferentes abordagens Dois modelos CCR e BCC destacamse por serem amplamente utilizados observam Périco Rebelatto e Santana 2008 O modelo pioneiro CCR nome oriundo das iniciais de seus criadores tem como característica uma relação proporcional entre as variáveis de entrada e saída Ou seja assume que as DMUs apresentam retornos constantes de escala Charnes Cooper e Rhodes 1978 Já o modelo BCC cujo nome também é proveniente das iniciais de seus criadores rompe a necessidade da premissa de retornos constantes de escala das variáveis consideradas BANKER CHARNES COOPER 1984 Permitir a comparação entre variáveis sem retornos constantes de escala ampliou a gama de aplicações e segundo LaPlante e Paradi 2015 tornou a BCC uma abordagem notável Além das DMUs outro componente para construção do modelo DEA são as variáveis de entradas e saídas que serão utilizadas como parâmetro de avaliação Conforme salientam Branco et al 2016 essas variáveis são escolhidas de acordo com o objetivo da análise A seleção desses itens entretanto não é uma tarefa trivial Périco Rebelatto e Santana 2008 verificaram que a maior parte dos estudos acerca do tema restringemse em afirmar que utilizaram as variáveis mais adequadas para avaliação do desempenho das DMUs avaliadas mas não detalham de maneira sistemática o método de seleção Existem contudo técnicas para realizar essa tarefa de escolha de variáveis Senra et al 2007 destacam alguns métodos como o de Norman e Stoker o qual alia análise de correção simples e DEA cita o método baseado em 3 estágios de Golany e Roll e menciona ainda o método de Lins e Moreira o qual seria um melhoramento do proposto por Norman e Stoker De fato independente do formato da escolha das variáveis é inequívoco que as entradas e saídas precisam ser escolhidas com cautela para não perder o aspecto discriminatório do modelo alertam Freaza Guedes e Gomes 2006 Escolhido o modelo e as variáveis a serem utilizadas ainda resta definir a orientação a ser avaliada input ou output Na orientação input buscase eficiência a partir da redução das entradas e manutenção da saída Já a orientação a output o objetivo é aumentar os resultados sem variar os insumos PÉRICO REBELATTO SANTANA 2008 A análise DEA apresenta como vantagem a dispensa de especificação funcional para a fronteira de produção portanto não requer que os inputs e outputs tenham uma relação determinada MACEDO SANTOS SILVA 2006 A ausência de especificação funcional como aspecto relevante que facilita e incentiva a utilização da DEA é reforçada por Gomes Oliveira e Matias 2017 p 657 a vantagem de ser computacionalmente simples uma vez que não requer a estimativa estocástica de uma função de produção Além disso a DEA tem como característica a possibilidade de utilizar múltiplas variáveis de entrada e saída sem exigência de qualquer suposição acerca da distribuição das variáveis segundo Freaza Guedes e Gomes 2006 7 24 ESTUDOS RELACIONADOS As pesquisas com instituições financeiras especificamente bancos dada a sua relevância na economia são vastas Da mesma forma o enfoque de análise de eficiência com a abordagem DEA também é frequente Assim na sequência são apresentados alguns estudos que evidenciam o interesse no assunto Analisar a eficiência de bancos que operam no Brasil utilizando DEA e Índice de Malmquist foi o que fizeram Martin et al 2011 Na amostra de oito bancos entre outros resultados verificaram que o retorno sobre o capital próprio é diretamente proporcional à eficiência dos bancos O estudo realizado por Pedrosa 2014 concentrou seus esforços para avaliar a eficiência de bancos que atuam no Brasil com controle nacional em relação aos bancos com controle estrangeiro A partir da análise DEA buscou hierarquizar os bancos por índice de produtividade Nesse apanhado concluiu que os bancos com controle nacional são mais eficientes que os de controle estrangeiro embora não haja um distanciamento substancial entre eles Yin Yu e Huang 2018 por seu turno tratam da eficiência de bancos chineses no período de 20052016 Além da abordagem DEA o estudo também utiliza o índice de Gini para realização das análises Os resultados do estudo sugeriram em especial que há heterogeneidades significantes na eficiência dos bancos analisados em todos os aspectos produtividade e lucratividade Outra pesquisa que trata de assuntos atinentes é a realizada por Henriques et al 2018 Novamente bancos brasileiros são alvo de análise de eficiência com a metodologia DEA Trouxeram um destaque interessante para identificar possibilidades de melhoria de eficiência a partir da escala Percebeuse que os maiores bancos possuem dificuldades de apresentar ganhos de eficiência de escala Por fim a avaliação de eficiência em relação as seções judiciárias do poder judiciário da 4ª região federal foi o que pesquisaram Venturini Souza e Bianchi 2018 As análises com ajuda do ferramental DEA percorreram o horizonte compreendido entre 2013 e 2017 e utilizaram as abordagens BCC e CCR nas perspectivas de input e output Dentre os resultados apontados verificouse que ter o maior número de processos julgados não bastava para ser a mais eficiente entre as seções O tema instiga o meio acadêmico e possui amplo espaço para novas pesquisas Além disso como o ambiente corporativo e a exigência do mercado vão sendo alterados ao longo do tempo tornase oportuno efetuar novos estudos direcionados ao assunto 3 PROCEDIMENTOS METODOLÓGICOS A fim de permitir a compreensão do modo como se elaborou a pesquisa apresentam se nesta seção a classificação do estudo e os procedimentos executados para coleta e seleção de dados Além disso constam os critérios de análises e a ferramenta utilizada 31 CLASSIFICAÇÃO DA PESQUISA A pesquisa realizada neste estudo é classificada quanto aos seguintes aspectos quantitativa pela forma de abordagem do problema descritiva de acordo com seus objetivos e documental com base nos procedimentos técnicos utilizados Uma análise quantitativa busca obter respostas através de números A característica dessa abordagem é ratificada por Richardson 1999 o qual afirma que a quantificação das informações coletadas e trabalhadas são a tônica de uma abordagem dessa natureza No estudo em questão a comparação de 8 eficiência entre bancos teve um componente matemático substancial visto que utilizou como ferramental a DEA Como fonte de dados foram utilizadas informações constantes em demonstrações contábeis publicadas por essas instituições bancárias caracterizando deste modo o estudo como documental A partir desses documentos buscouse avaliar as relações existentes entre os bancos avaliados em termos de eficiência operacional Essa forma de trabalhar está em sintonia com o que se entende por uma pesquisa descritiva com procedimento documental GIL 2008 32 AMOSTRA E COLETA DE DADOS Foram selecionados todos os bancos públicos que operam no varejo para realização do estudo em razão de atingirem uma gama maior de clientes em relação aos bancos de desenvolvimento Embora esses bancos possuam diferenças consideráveis de tamanho para fins de utilização da metodologia DEA o essencial é que as unidades sejam do mesmo ramo e que tenham características semelhantes Os bancos no Quadro 2 englobam a amostra escolhida e cumprem com os requisitos mencionados já que atuam no varejo com ampla gama de serviços e com diversas carteiras de clientes Quadro 2 Relação de bancos da amostra Instituição Financeira SiglaAcrônimo Banco da Amazônia SA BASA Banco do Brasil SA BB Banco do Estado de Sergipe SA BANESE Banco do Estado do Pará SA BANPARÁ Banco do Estado do Rio Grande do Sul SA BANRISUL Banco do Nordeste do Brasil SA BNB Banco do Estado do Espírito Santo SA BANESTES Banco de Brasília SA BRB Caixa Econômica Federal CEF Fonte elaborado a partir dos dados da pesquisa 2019 As instituições financeiras do Quadro 2 precisam fornecer suas informações contábeis periodicamente para o BACEN as quais são de acesso público para qualquer usuário Assim esse foi o canal utilizado para obtenção dos insumos que alimentaram as análises Foram capturadas as demonstrações do balanço patrimonial e do demonstrativo do resultado do exercício abrangendo o período de 9 anos 2010 2018 Embora existam estudos analisando a eficiência entre bancos em um único ano como o realizado por Branco et al 2016 optou se por avaliar um intervalo de tempo de forma semelhante ao que fizeram Gomes Oliveira e Matias 2017 O ano de 2018 entretanto foi avaliado sem considerar o BRB pois suas demonstrações não estavam publicadas até o final de abril de 2019 33 CRITÉRIOS DE ANÁLISE Para cálculo do IEO foram utilizados como numerador as despesas de pessoal e despesas administrativas Apesar de o BACEN não utilizar as despesas de pessoal nesse índice BACEN 2016 essa variável é considerada nos relatórios de demonstrações financeiras do BANRISUL BANRISUL 2019 e CEF CEF 2018 Já para o denominador foram utilizadas as receitas de intermediação financeira receita de prestação de serviços e 9 receita de tarifas bancárias Esse é um índice do tipo menormelhor e foi apurado para todos os anos do período analisado A utilização da DEA requer para a comparação de eficiência entre unidades tomadoras de decisão o estabelecimento de variáveis de entrada e saída PÉRICO REBELATTO SANTANA 2008 Não há na literatura uma uniformidade na seleção dessas variáveis pelo contrário existe uma diversidade muito grande nas escolhas desses elementos Contudo é possível observar variáveis já utilizadas em outras pesquisas e que coadunam com a proposta do estudo em questão e por isso servem de amparo para as escolhas realizadas As variáveis de entrada que foram trabalhadas são o ativo não circulante imobilizado as despesas administrativas e as despesas de pessoal Tanto as despesas administrativas quanto de pessoal já foram utilizadas nos trabalhos de Freaza Guedes e Gomes 2006 e Staub Souza e Tabak 2010 Já o ativo permanente também é utilizado de forma recorrente como variável de entrada e aparece nos trabalhos de Henriques et al 2018 e Baten Kasin e Rahman 2015 Com relação à variável de saída optouse por utilizar o resultado antes do imposto de renda utilizado nos trabalhos de Camargo Júnior e Matias 2005 e Kwon e Lee 2015 O Quadro 3 apresenta de forma organizada as variáveis escolhidas Quadro 3 Variáveis de estudo Entrada Input Saída Output Ativo Permanente AP Resultado Antes do Imposto de Renda RE Despesas de Pessoal DP Despesas Administrativas DA Fonte elaborado a partir dos dados da pesquisa 2019 O processamento dos dados coletados foi feito com o software SIAD apresentado por Meza Biondi Neto e Ribeiro 2005 Essa ferramenta permite a implementação de modelos DEA no ambiente computacional facilitando a obtenção dos resultados matemáticos Considerando que não há como garantir que as variáveis de entrada forneçam retornos constantes de escala no caso dos bancos para o estudo serão feitas avaliações na abordagem BCC semelhante ao que fizeram Périco Rebelatto e Santana 2008 Com relação à orientação optouse por avaliar pela ótica das entradas input semelhante ao que fizeram CARAFFINI SOUZA BEHR 2018 4 ANÁLISE DOS DADOS Nesta seção inicialmente é apresentada a evolução do IEO de cada banco avaliado A comparação entre os bancos com enfoque BCC é apresentada na sequência juntamente com a avaliação dos pesos atribuídos às variáveis eficientes e os decréscimos necessários nas variáveis dos bancos que apresentaram ineficiência Para finalizar é traçado um paralelo entre os resultados de eficiência observados pela análise DEA e o IEO 41 ANÁLISE DO IEO O IEO procura mostrar o esforço que o banco faz para gerar receita E isso depende da capacidade do banco de otimizar seus processos internos e também da capacidade de reduzir seus desperdícios o que envolve a eficiência técnica e alocativa MATTOS TERRA 2015 Logo redução do quadro de pessoal bem como redução do número de agências em geral contribuem para a melhoria desse índice A Tabela 1 apresenta os resultados do IEO para cada banco 10 Tabela 1 Índice de eficiência operacional no período de 2010 a 2018 BANCOANO 2010 2011 2012 2013 2014 2015 2016 2017 2018 BASA 3924 3975 4256 4783 4177 3766 3757 3847 3898 BB 2720 2370 2694 2794 2289 1841 2087 2157 2316 BANESE 3849 3169 3588 4025 4535 3779 3760 3800 4292 BANPARÁ 4627 4006 3723 3430 3332 3448 3751 3845 4297 BANRISUL 3013 2679 2825 3086 2844 2348 2593 3036 3317 BNB 4554 3934 4367 3919 3867 2983 3805 4064 4243 BANESTES 2542 2606 2915 2700 2343 2087 1817 2037 2391 BRB 3623 3585 4091 4541 4201 3907 4246 4459 CEF 3078 2777 2938 2717 2179 1862 1869 2079 2333 Fonte elaborada a partir dos dados da pesquisa 2019 Lembrando que se trata de um índice do tipo menormelhor o resultado mais expressivo no índice de eficiência operacional individual observado na Tabela 1 foi obtido pelo BANESTES no ano de 2016 chegando ao valor de 1817 Esse resultado é proveniente em especial do pico obtido nas receitas de intermediação financeira desse banco Na outra ponta ficou o BASA o qual atingiu o valor de 4783 no ano de 2013 e também teve maior influência das receitas do que das despesas Houve portanto uma amplitude de quase 30 pontos percentuais entre os resultados observados nos extremos A média do período por sua vez ficou em 3304 Em geral observase na Tabela 1 que os bancos apresentaram um comportamento de leve melhora de seus índices durante o período Apesar disso os últimos dois anos 2017 e 2018 sugerem um tendência de diminuição da eficiência logo após a fase de melhores resultados ocorrida anos de 2015 e 2016 O grupo formado pelos bancos CEF BANRISUL BB e BANESTES obtiveram os menores valores Ou seja apresentam um melhor índice de eficiência operacional Já o conjunto de bancos que está com esse indicador num patamar desfavorável é composto por BASA BENESE BANPARÁ BNB e BRB 42 ANÁLISE BCC A comparação em termos de eficiência dos 9 bancos públicos considerando os inputs e output escolhidos por meio da técnica DEA na abordagem BCC mostra os bancos que estão na fronteira de eficiência e os que precisam reduzir os seus inputs para atingir o mesmo patamar A Tabela 2 apresenta os resultados obtidos em cada ano do período compreendido entre 2010 e 2018 Tabela 2 Análise BCC BancoAno 2010 2011 2012 2013 2014 2015 2016 2017 2018 BASA 081 033 074 045 097 1 1 1 076 BB 1 1 1 1 1 1 1 1 1 BANESE 1 1 1 1 1 1 1 1 1 BANPARÁ 1 1 1 1 1 1 1 1 1 BANRISUL 1 1 1 1 074 1 1 1 098 BNB 1 1 035 1 1 045 1 1 1 BANESTES 1 055 052 058 084 071 081 084 089 BRB 075 033 061 045 044 031 046 064 CEF 056 1 1 086 1 002 093 1 1 Fonte elaborada a partir dos dados da pesquisa 2019 11 A análise BCC na Tabela 2 identificou como DMUs eficientes em todo o período avaliado os bancos BB BANESE e BANPARÁ Podese incluir ainda o BANRISUL o qual ficou com a máxima eficiência em 7 anos e bem próximo nos outros 2 anos avaliados Nesse resultado observase que os 3 bancos eficientes não são os 3 maiores BB CEF BANRISUL ou os 3 menores BRB BANPARÁ BANESE evidenciando a característica do modelo que permite comparações observando DMUs distintas em termos de tamanho conforme observado por Casado 2007 Já os bancos BASA BANESTES e BRB tiveram desempenho abaixo da fronteira de eficiência em mais da metade dos anos avaliados denotando um distanciamento contínuo Nesse grupo podese incluir também a CEF a qual ficou abaixo em 4 anos A média de eficiência do período atingiu o resultado de 8585 Esse número ficou consideravelmente abaixo do que observou Pedrosa 2014 o qual apurou o valor de 9877 no mesmo modeloorientação Cabe citar contudo que o intervalo de tempo e as variáveis utilizadas não são homogêneos e por isso as diferenças podem ser interpretadas como factíveis e naturais Outro aspecto interessante de se avaliar são os pesos que a DEA atribui às variáveis Na busca pelo melhor conjunto de pesos que resolvem o modelo matemático é possível encontrar soluções que zeram uma ou mais variáveis MELLO et al 2005 Com isso evita se que as variáveis sejam ponderadas de forma discricionária Na Tabela 3 são exibidos os pesos atribuídos pela ferramenta nas variáveis de entrada de cada banco durante o período avaliado 12 Tabela 3 Pesos variáveis de entrada Banco Inputs 2010 2011 2012 2013 2014 2015 2016 2017 2018 BASA AP 0 0 0 0 0 0 0 0 0 DP 0 000000302 0 0 0 000000066 000000062 000000034 0 DA 000000543 0 000000421 000000396 000000407 000000233 000000232 000000298 000000358 BB AP 0 0 0 0 0 0 0 0 0 DP 000000002 000000007 000000007 0 000000005 000000005 000000002 000000004 0 DA 000000006 0 0 000000007 000000001 0 000000003 000000001 000000007 BANESE AP 000000012 0 0 0 000001158 0 0 000000091 0 DP 000000184 000000889 0 0 0 0000001 00000013 0 000000573 DA 000001269 0 000001111 000001172 0 000000814 000000648 000000634 0 BANPARÁ AP 000001072 000000315 0 0 000000902 000000843 0 0 000000353 DP 0 000000679 000000697 000000596 0 0 00000033 000000304 0 DA 000000615 0 0 0 0 0 0 0 000000091 BANRISUL AP 000000004 000000003 0 000000017 0 0 000000023 0 00000001 DP 000000035 000000089 0 0 000000056 000000052 00000002 000000038 000000034 DA 000000093 0 000000128 00000009 000000015 000000012 0 000000019 0 BNB AP 000000204 000000199 000000496 000000424 00000009 000000426 000000466 000000144 000000027 DP 000000033 0 0 0 000000052 0 0 000000039 0 DA 000000042 000000081 0 0 0 0 0 0 000000076 BANESTES AP 000000011 0 0 0 0 0 0 0 0 DP 000000524 000000485 000000438 000000402 0 000000047 000000064 000000041 0 DA 0 0 0 0 000000475 000000382 00000032 00000034 000000389 BRB AP 000000362 000000256 0 0 0 0 0 0 DP 0 0 0 0 0 0 0 000000024 DA 000000176 000000206 00000039 00000033 000000292 000000296 000000255 0000002 CEF AP 000000016 000000013 000000012 00000001 000000002 0 000000007 000000001 0 DP 0 0 0 0 0 0 0 0 0 DA 0 000000001 0 0 000000008 00000001 0 000000008 000000009 Fonte elaborada a partir dos dados da pesquisa 2019 Exceto pelos bancos BANESE BANRISUL e BNB no ano de 2010 todas as ponderações calculadas e apresentadas na Tabela 3 eliminaram pelo menos uma das variáveis Ou seja foi atribuído peso 0 zero a essas variáveis para melhorar o resultado de eficiência do banco Freaza Guedes e Gomes 2006 chegaram a constatações análogas quando no seus estudos tanto os bancos eficientes quanto os bancos ineficientes desprezaram pelo menos uma variável na apuração de eficiência Verificase que a variável DA é o item que menos vezes é zerado pela ferramenta Ou seja na maior parte das vezes é uma variável que contribui para o resultado em termos de eficiência Entretanto todos os bancos apresentam um crescimento de suas despesas ao longo do período com exceção do BB que apresenta queda Já a variável DP é a segunda menos zerada E a melhoria desse item DP exige assim como o verificado por Venturini Souza e Bianchi 2018 um esforço adicional pela característica mais estável do empregado público dificultando a redução do quadro de pessoal 13 A última análise efetuada diz respeito ao distanciamento que as bancos ineficientes encontramse da fronteira de eficiência Os percentuais que cada banco ineficiente precisa diminuir em suas entradas são apresentados na Tabela 4 Tabela 4 Ineficiências Banco Inputs 2010 2011 2012 2013 2014 2015 2016 2017 2018 BASA AP 3346 6832 5059 6516 5472 6250 DP 2953 6603 3846 6522 3974 5098 DA 1903 6980 2554 5474 297 2369 BANRISUL AP 3269 273 DP 2553 273 DA 2553 1417 BNB AP 6516 5493 DP 8841 8598 DA 8269 8015 BANESTES AP 6021 6151 5883 5204 6248 5736 5106 5049 DP 4545 4793 4195 2990 2875 1869 1551 1829 DA 6319 5588 4442 1563 2875 1869 1551 1115 BRB AP 2503 6658 6671 7403 7908 8570 7059 5658 DP 3316 7114 6176 7438 7137 7699 5566 3624 DA 2503 6658 3837 5498 5576 6945 5351 3624 CEF AP 4442 1410 9032 715 DP 8056 7321 9182 6691 DA 7125 5603 9007 5560 Fonte elaborada a partir dos dados da pesquisa 2019 A apuração das unidades ineficientes na Tabela 4 apresenta necessidade de melhoria nas três variáveis de entrada A DA é a variável que menos precisa sofrer ajustes para atingir os mesmos resultados dos bancos eficientes Entretanto o BANRISUL foge desse padrão no ano de 2018 quando a DA apresentou necessidade elevada de melhoria em relação às demais variáveis O ganho de eficiência mediante a redução do AP por sua vez foi o que caracterizou os resultados do BASA BANESTES e BRB Por fim podese verificar que a necessidade de melhora na variável DP foi impactante no BNB e na CEF em especial no ano de 2015 para a CEF A necessidade de redução de AP e DP em bancos ineficientes do ponto de vista da DEA também é observada nos estudos de Henriques et al 2018 O menor impacto da variável DA não surpreende pois coaduna com o observado na Tabela 3 Já a ineficiência por conta da variável AP pode ser explicada em parte pela estratégia dos bancos em disponibilizar unidades em regiões com menor apelo comercial dada o seu caráter público ARAÚJO CINTRA 2011 Em relação à variável DP conforme observado por Venturini Souza e Bianchi 2018 há dificuldades para melhoria de eficiência pela estabilidade dos empregados públicos 43 COMPARAÇÃO DOS RESULTADOS NAS ANÁLISES PELO IEO E BCC A confrontação dos resultados obtidos em cada uma das ferramentas de análise permite ampliar as avaliações efetuadas Isso ocorre em parte pelo fato de se utilizar diferentes variáveis em cada abordagem Essa forma de estudo utilizando mais de uma 14 técnica para comparar ou complementar as análises de eficiência encontra paralelo na literatura nos trabalhos de Neves Júnior et al 2007 Martin et al 2011 e Yin Yu e Huang 2018 A Tabela 5 exibe os bancos eficientes E e ineficientes I por modeloano Tabela 5 Comparação de resultados Ano 2010 2011 2012 2013 2014 2015 2016 2017 2018 Modelo DEA IEO DEA IEO DEA IEO DEA IEO DEA IEO DEA IEO DEA IEO DEA IEO DEA IEO BASA I I I I I I I I I I E I E I E I I I BB E E E E E E E E E E E E E E E E E E BANESE E I E I E I E I E I E I E I E I E I BANPARÁ E I E I E I E I E I E I E I E I E I BANRISUL E E E E E E E E I E E E E E E E I E BNB E I E I I I E I E I I I E I E I E I BANESTES E E I E I E I E I E I E I E I E I E BRB I I I I I I I I I I I I I I I I CEF I E E E E E I E E E I E I E E E E E Fonte elaborada a partir dos dados da pesquisa 2019 Os resultados nas análises IEO e BCC apresentaram correspondências mas também apresentaram situações antagônicas conforme verificado na Tabela 5 Como similaridade podese destacar nos bancos eficientes os resultados do BB e BANRISUL O BB ficou no bloco de bancos eficientes em todo o período avaliado pelo IEO tendo o mesmo comportamento na avaliação DEA Já o BANRISUL apesar de ter apresentado leve piora no IEO nos últimos três anos permaneceu a maior parte do período como banco eficiente em ambas metodologias Nos bancos ineficientes também há bancos que têm ocorrências uniformes nas duas perspectivas Os bancos BRB e BASA compõem esse grupo pois mostram fraco desempenho no IEO e coincidiram com resultados de ineficiência na análise DEA no intervalo de estudo Assim como Martin et al 2011 o estudo identificou falta de convergência em algumas situações As divergências foram observadas por exemplo nos bancos BANESTES e BANPARÁ No IEO o BANESTES atingiu excelentes resultados chegando ao melhor índice entre todos os bancos no ano de 2016 Por outro lado quando a avaliação ocorreu pela DEA o banco computou sucessivos resultados de ineficiência A variável AP utilizada apenas na DEA não responde isoladamente o motivo desse distanciamento nos resultados mas é inegável que impactou negativamente tendo em vista que as ineficiências apontadas na Tabela 4 são mais representativas nessa variável O oposto ocorreu com o banco BANPARÁ cujo resultado no IEO foi fraco e ao mesmo tempo apresentou eficiência em todos os anos na análise DEA A variável que merece destaque nesse caso é a DA Essa variável é considerada nas duas medidas porém no cálculo da DEA ela é zerada em sete dos nove anos da avaliação para esse banco influenciando menos os resultados nessa abordagem Outra forma de avaliar a influência das variáveis na eficiência é através de cálculos estatísticos como fizeram Neves Júnior et al 2007 5 CONSIDERAÇÕES FINAIS O propósito do artigo foi analisar a eficiência operacional dos principais bancos públicos do Brasil BASA BB BANESE BANPARÁ BANRISUL BNB BANESTES BRB e CEF no período de 2010 a 2018 O alcance desse objetivo foi conduzido mediante 15 avaliações da performance dos bancos através de DEA e do IEO bem como relacionando os resultados dos dois instrumentos Após o cálculo do IEO foi possível verificar que os bancos estudados dividiamse em dois grupos O grupo composto por CEF BANRISUL BB e BANESTES apresentaram melhores resultados ou seja um IEO baixo chegando a valores inferiores a 19 O outro grupo BASA BENESE BANPARÁ BNB e BRB exibiu um nível inferior de eficiência ficando o destaque negativo com o banco BASA que atingiu 4783 em 2013 Ambos os grupos sinalizaram evolução no IEO em que pese a leve piora da eficiência no último biênio A análise pela DEA apontou os bancos BB BANESE e BANPARÁ como eficientes em todos os anos Entretanto mais do que fazer um ranqueamento entre os bancos a avaliação por DEA permitiu identificar as variáveis que tiveram maior influência nos resultados de eficiência apurados Nessa avaliação percebeuse que a DA é uma variável que no geral contribuiu positivamente para o resultado dos bancos Evidenciouse ainda o quanto cada banco ineficiente precisa melhorar em cada variável para chegar no nível dos bancos eficientes A DA em relação às demais variáveis ficou menos distante dos resultados dos bancos eficientes A relação entre os resultados das duas avaliações é outro aspecto interessante que foi avaliado Constatouse que nem sempre um banco considerado eficiente na avaliação DEA possui o mesmo comportamento no IEO Essas divergências foram ocasionadas tanto pela diferença de variáveis utilizadas em cada ferramenta como pela influência que variáveis específicas exerceram na apuração dos bancos individualmente Do ponto de vista da análise IEO verificase que há grande margem para melhoria dos bancos ineficientes Ao mesmo tempo para todos os bancos expõe a trajetória dos resultados obtidos possibilitando correções e melhorias Na ótica da DEA os bancos ineficientes têm o direcionamento de onde atuar variáveis para alcançar o resultado do benchmarking Quando há o confrontamento das análises surgem diferenças que também apontam ineficiências que subsidiam a gestão dos bancos envolvidos Esse estudo contribui para o incremento da eficiência operacional nos bancos públicos pois explora e evidencia as lacunas existentes nessas instituições Embora forneça informações concretas e úteis para a análise de eficiência não se pode ser definitivo em relação aos resultados produzidos Sabese que a escolha de outras variáveis pode modificar as posições encontradas na análise DEA e mesmo o IEO pode ser apurado de forma distinta influenciando os resultados Por isso a fim de expandir o conhecimento recomendase a aplicação das análises com outras variáveis ou incluir bancos privados em estudos futuros REFERÊNCIAS ALTUNBAS Y EVANS L MOLYNEUX P Bank ownership and efficiency Journal of Money Credit and Banking Columbus v 33 n 4 p 926954 2001 Disponível em httpswwwjstororgstable2673929 Acesso em 19 nov 2018 ARAÚJO V L CINTRA M A M O papel dos bancos públicos federais na economia brasileira Brasília IPEA Instituto de Pesquisa Econômica Aplicada abr 2011 Disponível em httpwwwipeagovbrportalimagesstoriesPDFsTDstd1604pdf Acesso em 27 out 2018 ARAÚJO M V M Investimento em tecnologia nas instituições financeiras e a influência das fintechs 2018 Dissertação Mestrado em Economia Escola de Economia Fundação Getúlio Vargas São Paulo 2018 Disponível em httpbibliotecadigitalfgvbrdspacehandle1043824740 Acesso em 20 ago 2018 16 ASSAF NETO A Estrutura e análise de balanços um enfoque econômicofinanceiro 11 ed São Paulo Atlas 2015 BANCO CENTRAL DO BRASIL Informações para análise econômicofinanceiro Brasília 2018 Disponível em httpswww3bcbgovbrifdata Acesso em 13 out 2018 BANCO CENTRAL DO BRASIL Relatório de estabilidade financeira Brasília 2016 Disponível em httpswwwbcbgovbrhtmsestabilidade201609refPubpdf Acesso em 15 out 2018 BANCO CENTRAL DO BRASIL Relatórios anuais Brasília 2017 Disponível em httpswwwbcbgovbridREVSFNano2017 Acesso em 21 nov 2018 BANCO CENTRAL DO BRASIL Sistema financeiro nacional Brasília 2019 Disponível em httpswwwbcbgovbrestabilidadefinanceirasfn Acesso em 14 abr 2019 BANKER A CHARNES A COOPER W Some models for estimating technical and scale inefficiencies in data envelopment analysis Management Science Columbus v 30 n 9 p 10781092 1984 Disponível em httpswwwjstororgstable2631725 Acesso em 22 mar 2019 BANCO DO ESTADO DO RIO GRANDE DO SUL Demonstrações financeiras Porto Alegre 2019 Disponível em httpribanrisulcombrbanrisulwebconteudoptaspidioma0tipo8223conta28id 257233 Acesso em 19 maio 2019 BATEN M KASIM M RAHMAN M Cost and profit efficiency of online banks do national commercial banks perform better than private banks The Journal of Internet Banking and Commerce Candler v 20 n 3 p 119 2015 Disponível em httpwwwicommercecentralcomopenaccesscostandprofitefficiencyofonlinebanks donationalcommercialbanksperformbetterthanprivatebanksphpaid64359 Acesso em19 abr 2019 BERGER A MESTER L Inside the black box what explains differences in the efficiencies of financial institutions Journal of Banking Finance Sl v 1 n 21 p 895947 1997 Disponível em httpswwwsciencedirectcomsciencearticlepiiS0378426697000101 Acesso em 15 mar 2019 BRANCO A et al Efficiency of the brazilian banking system an assessment using DEA under three approaches Journal of Applied Finance Banking London v 6 n 4 p 27 42 2016 Disponível em httpswwwscienpresscomjournalfocusaspmainid56SubidIVIssue1835 Acesso em 10 nov 2018 BRASIL Presidência da República Lei Complementar nº 4595 de 31 de dezembro de 1964 Dispõe sobre a política e as instituições monetárias bancárias e creditícias cria o conselho monetário nacional e dá outras providências Brasília 1964 Disponível em httpwwwplanaltogovbrccivil03LEISL4595htm Acesso em 14 abr 2019 17 BRASIL Presidência da República Lei Complementar nº 6385 de 7 de dezembro de 1976 Dispõe sobre o mercado de valores mobiliários e cria a Comissão de Valores Mobiliários Brasília 1976 Disponível em httpwwwplanaltogovbrccivil03LEISL6385htm Acesso em 14 abr 2019 BRASIL Presidência da República Lei Complementar nº 6404 de 15 de dezembro de 1976 Brasília 1976 Dispõe sobre as sociedades por ações Disponível em httpwwwplanaltogovbrccivil03leisl6404compiladahtm Acesso em 14 abr 2019 BRASIL Presidência da República Lei Complementar nº 10303 de 31 de outubro de 2001 Altera e acrescenta dispositivos na Lei nº 6404 de 15 de dezembro de 1976 que dispõe sobre as Sociedades por Ações e na Lei nº 6385 de 7 de dezembro de 1976 que dispõe sobre o mercado de valores mobiliários e cria a Comissão de Valores Mobiliários Brasília 2001 Disponível em httpwwwplanaltogovbrCCIVIL03LEISLEIS2001L10303htm Acesso em 14 abr 2019 CAIXA ECONÔMINA FEDERAL Análise de desempenho 2018 Brasília 2018 Disponível em httpwwwcaixagovbrDownloadscaixademonstrativo financeiroRelatorioAnaliseDesempenho2018pdf Acesso em 19 maio 2019 CAMARGO JÚNIOR A MATIAS A Eficiência operacional de bancos no Brasil América Latina e EUA In CONGRESSO INTERNACIONAL DE CUSTOS 9 2005 Florianópolis Anais Florianópolis CIC 2005 Disponível em httpsanaiscbcemnuvenscombranaisissueview9 Acesso em 21 mar 2019 CARAFFINI J SOUZA R BEHR A Transformação digital e desempenho no setor bancário In CONGRESSO TRANSFORMAÇÃO DIGITAL 1 2018 São Paulo Anais São Paulo CTD 2018 Disponível em httpbibliotecadigitalfgvbrocsindexphpctdctd2018paperview6965 Acesso em 20 maio 2019 CASADO F Análise envoltória de dados conceitos metodologia e estudo da arte na educação superior Revista Sociais e Humanas Santa Maria v 20 n 1 p 5971 2007 Disponível em httpsperiodicosufsmbrsociaisehumanasarticleview907 Acesso em 23 maio 2019 CHARNES A COOPER W RHODES E Measuring the efficiency of decision making units European Journal of Operational Research Sl v 2 p 429444 1978 Disponível em httpswwwsciencedirectcomsciencearticlepii0377221778901388 Acesso em 28 abr 2019 CLEMENTE A KÜHL M Intermediação financeira no Brasil influência da taxa de captação sobre a taxa de aplicação In CONGRESSO USP CONTROLADORIA E CONTABILIDADE 6 2006 São Paulo Anais São Paulo CUCC 2006 Disponível em httpscongressouspfipecafiorganaisartigos62006anresumoaspcodtrabalho236 Acesso em 25 maio 2019 DEPARTAMENTO INTERSINDICAL DE ESTATÍSTICAS E ESTUDOS SOCIOECONÔMICOS Desempenho dos bancos 2017 São Paulo 2017 Disponível em 18 httpswwwdieeseorgbrdesempenhodosbancos2018desemprenhoDosBancos2017html Acesso em 26 out 2018 FORTUNA E Mercado financeiro produtos e serviços 18 ed Rio de Janeiro Qualitymark 2011 FREAZA F GUEDES L GOMES L Análise de eficiência do mercado bancário brasileiro utilizando a metodologia da análise envoltória de dados In SIMPÓSIO BRASILEIRO DE PESQUISA OPERACIONAL 38 2006 Goiânia Anais Goiânia SBPO 2006 Disponível em httpwwwdinuembrsbposbpo2006trabalhoidx00htm Acesso em 07 nov 2018 GIL A Como elaborar projetos de pesquisa 4 ed São Paulo Atlas SA 2008 GOMES M OLIVEIRA S MATIAS A Eficiência do setor bancário brasileiro no período de 2006 a 2013 bancos domésticos x bancos estrangeiros Revista Nova Economia Belo Horizonte v 27 n 3 p 641670 2017 Disponível em httpsrevistasfaceufmgbrindexphpnovaeconomiaarticleview3057 Acesso em 11 dez 2018 HENRIQUES I et al E Efficiency in the Brazilian banking system using data envelopment analysis Future Business Journal Sl v 4 n 1 p 157178 2018 Disponível em httpswwwsciencedirectcomsciencearticlepiiS2314721017300208 Acesso em 30 out 2018 NEVES JÚNIOR I et al M Eficiência operacional uma análise exploratória dos 50 maiores bancos brasileiros pelo ranking BACEN In CONGRESSO USP DE CONTROLADORIA E CONTABILIDADE 7 2007 São Paulo Anais São Paulo CUCC 2007 Disponível em httpscongressouspfipecafiorganaisartigos72007anresumoasppagina9con1 Acesso em 25 mar 2019 KWON H LEE J Twostage production modeling of large US banks A DEAneural network approach Expert Systems with Applications Greenburgh v 1 n 42 p 67586766 2015 Disponível em httpsdlacmorgcitationcfmid2799079 Acesso em 03 maio 2019 LAPLANTE A PARADI J Evaluation of bank branch growth potencial using data envelopment analysis The International Journal of Management Science Omega Sl v 52 n 1 p 3341 2015 Disponível em httpswwwsciencedirectcomsciencearticlepiiS0305048314001303 Acesso em 18 maio 2019 MAÇADA A Impacto dos investimentos em tecnologia da informação nas variáveis estratégicas e na eficiência dos bancos brasileiros 2001 Tese Doutorado em Administração Escola de Administração Universidade Federal do Rio Grande do Sul Porto Alegre 2001 Disponível em httpslumeufrgsbrhandle101831845 Acesso em 05 jun 2019 MACEDO A SANTOS R SILVA F Desempenho organizacional no setor bancário brasileiro uma aplicação da análise envoltória de dados Revista de Administração 19 Mackenzie São Paulo v 7 n 1 p 1144 janmar 2006 Disponível em httpeditorarevistasmackenziebrindexphpRAMarticleview89 Acesso em 10 dez 2018 MARTIN D et al Determinantes da lucratividade de grandes bancos brasileiros uma abordagem em painel com DEA e MPI In ENCONTRO DA ANPAD 35 2011 Rio de Janeiro Anais Rio de Janeiro ENANPAD 2011 Disponível em httpwwwanpadorgbradminpdfFIN930pdf Acesso em 01 jun 2019 MATTOS E TERRA R Avaliação da qualidade do gasto público e mensuração da eficiência Brasília Secretaria do Tesouro Nacional p 211233 2015 Disponível em httpwwwtesourofazendagovbrtesouronacionallancalivrosobrequalidadedogasto Acesso em 22 ago 2018 MELLO J et al Curso de análise envoltória de dados In SIMPÓSIO BRASILEIRO DE PESQUISA OPERACIONAL 37 2005 Gramado Anais Gramado SBPO 2005 Disponível em httpwwwdinuembrsbposbpo2005trabalhoidx00htm Acesso em 17 set 2018 MEZA L BIONDI NETO L RIBEIRO P SIAD V20 Sistema integrado de apoio à decisão uma implementação de modelos de análise envoltória de dados e um método multicritério In SIMPÓSIO BRASILEIRO DE PESQUISA OPERACIONAL 37 2005 Gramado Anais Gramado SBPO 2005 Disponível em httpwwwdinuembrsbposbpo2005trabalhoidx00htm Acesso em18 set 2018 PEDROSA P A E Eficiência bancária no Brasil uma análise por envoltória de dados 2014 Dissertação Mestrado em Economia Faculdade de Economia Universidade de Brasília Brasília 2014 Disponível em httprepositoriounbbrhandle1048217349 Acesso em 02 dez 2018 PÉRICO A REBELATTO D SANTANA N Eficiência bancária os maiores bancos são os mais eficientes Uma análise por envoltória de dados Revista Gestão e Produção São Carlos v 159 n 2 p 421431 maiago 2008 Disponível em httpwwwscielobrscielophppidS0104 530X2008000200016scriptsciabstracttlngpt Acesso em 15 out 2018 PRIMO U et al L Determinantes da rentabilidade bancária no Brasil Revista de Administração e Contabilidade da Unisinos São Leopoldo v10 n 4 p 308323 outdez 2013 Disponível em httprevistasunisinosbrindexphpbaseissueview387showToc Acesso em 05 maio 2019 RICHARDSON R Pesquisa social métodos e técnicas 3 ed São Paulo Atlas 1999 ROSA M A O depósito compulsório e o mercado financeiro 2014 Dissertação Mestrado em Economia Faculdade de Ciências Econômicas Universidade Federal do Rio Grande do Sul Porto Alegre 2014 Disponível em httpslumeufrgsbrbitstreamhandle10183103964000921620pdfsequence1isAllowed y Acesso em 02 dez 2018 ROSSETTI J Introdução à economia 20 ed São Paulo Atlas 2003 20 SENRA L et al Estudo sobre métodos de seleção de variáveis em DEA Pesquisa Operacional Rio de Janeiro v 27 n 2 p 191207 2007 Disponível em httpwwwscielobrscielophpscriptsciarttextpidS010174382007000200001 Acesso em 19 mar 2019 SPONG K SULLIVAN R DEYOUNG R What makes a bank efficient a look at financial characteristics and bank management and ownership structure FRB of Kansas City Review Kansas p 119 Dec 1995 Disponível em httpswwwkansascityfedorgpublicatfipPRS951SpongDeYoungPDF Acesso em 10 mar 2019 STAUB R SOUZA G TABAK B Evolution of bank efficiency in Brazil a DEA approach European Journal of Operational Research Sl v 202 n 1 p 204213 2010 Disponível em httpswwwsciencedirectcomsciencearticlepiiS0377221709002951 Acesso em 10 nov 2018 VENNET R Cost and profit efficiency of financial conglomerates and universal banks in Europe Journal of Money Credit and Banking Columbus v 34 n 1 p 254282 2002 Disponível em httpswwwjstororgstable3270685seq1metadatainfotabcontents Acesso em 20 abr 2019 VENTURINI L SOUZA A BIANCHI M Eficiência na alocação de recursos públicos do poder judiciário um estudo na 4ª região federal In CONGRESSO BRASILEIRO DE CUSTOS 25 2018 Vitória Anais Vitória CBC 2018 Disponível em httpsanaiscbcemnuvenscombranaisissueview50 Acesso em 10 maio 2019 YIN Z YU Y HUANG J Evaluation and evolution of bank efficiency considering heterogeneity technology An empirical study from China PLoS ONE San Francisco v 14 n 4 2018 Disponível em httpsdoiorg101371journalpone0204559 Acesso em 10 mar 2019 1 IMPACTO DOS GASTOS COM TECNOLOGIA NA EFICIÊNCIA DOS PRINCIPAIS BANCOS BRASILEIROS NO PERÍODO DE 2012 A 2022 UMA AVALIAÇÃO USANDO FRONTEIRA ESTOCÁSTICA Filipe de Oliveira Gomes v3 n5 85 2 IMPACTO DOS GASTOS COM TECNOLOGIA NA EFICIÊNCIA DOS PRINCIPAIS BANCOS BRASILEIROS NO PERÍODO DE 2012 A 2022 UMA AVALIAÇÃO USANDO FRONTEIRA ESTOCÁSTICA Filipe de Oliveira Gomes1 1 Filipe de Oliveira Gomes é Mestre em Economia pelo Instituto Brasileiro de Ensino Desenvolvimento e Pesquisa IDP Email filgomeshotmailcom httpsorcidorg0009000424309093 3 O IDP é um centro de excelência no ensino na pesquisa e na extensão nas áreas da Administração Pública Direito e Economia O Instituto tem como um de seus objetivos centrais a profusão e difusão do conhecimento de assuntos estratégicos nas áreas em que atua constituindose um think tank independente que visa contribuir para as transformações sociais políticas e econômicas do Brasil DIREÇÃO E COORDENAÇÃO Diretor Geral Francisco Schertel Coordenador do Mestrado em Economia José Luiz Rossi CONSELHO EDITORIAL Coordenação Thiago Caldeira Renan Holtermann Milton Mendonça Supervisão e Revisão Luiz Augusto Magalhães Mathias Tessmann Apoio Técnico Igor Silva Projeto Gráfico e Diagramação Juliana Vasconcelos wwwidpedubr Revista Técnica voltada à divulgação de resultados preliminares de estudos e pesquisas aplicados em desenvolvimento por professores pesquisadores e estudantes de pósgraduação com o objetivo de estimular a produção e a discussão de conhecimentos técnicos relevantes na área de Economia Convidamos a comunidade acadêmica e profissional a enviar comentários e críticas aos autores visando o aprimoramento dos trabalhos para futura publicação Por seu propósito se concentrar na recepção de comentários e críticas a Revista Debates em Economia Aplicada não possui ISSN e não fere o ineditismo dos trabalhos divulgados As publicações da Revista estão disponíveis para acesso e download gratuito no formato PDF Acesse wwwidpedubr As opiniões emitidas nesta publicação são de exclusiva e inteira responsabilidade dos autores não exprimindo necessariamente o ponto de vista do IDP Qualquer citação aos trabalhos da Série só é permitida mediante autorização expressa dos autores 4 1 Introdução 6 2 Referencial Teórico 11 3 Metodologia 16 31 Estimação da Eficiência 16 32 Base de Dados 17 4 Resultados 23 41 Testes de Especificação 23 42 Resultado da Estimação 25 5 Conclusão 30 6 Referências 32 5 RESUMO Este estudo avalia a relação entre os gastos com tecnologia da informação TI e a eficiência dos principais bancos brasileiros A evolução tecnológica tem impactado fortemente o setor financeiro levando à disponibilização de plataformas digitais e canais virtuais de atendimento aos clientes O maior nível de tecnologia adotada pelos bancos pode contribuir para reduzir custos operacionais e elevar o nível de eficiência das instituições podendo resultar na melhoria dos serviços prestados aos clientes e maior facilidade de acesso ao crédito pela população A análise da relação entre os gastos com tecnologia e a eficiência foi realizada por meio da estimação de eficiência de Fronteira Estocástica SFA com base nas demonstrações financeiras dos cinco maiores bancos brasileiros no período de 2012 a 2022 que representam 76 dos ativos totais do setor Os resultados evidenciaram uma relação positiva entre o fluxo de gastos com TI e o nível de eficiência dos bancos o que significa maior nível de lucratividade Paralelamente o saldo de investimentos de TI não apresentou relevância estatística o que pode ser entendido como um indicativo de mudança no perfil de gastos com tecnologia dos bancos com a migração do modelo proprietário de negócios e soluções de TI para um modelo mais focado em serviços Além disso observouse que uma maior quantidade de pontos de atendimento especializados e menores gastos administrativos favorecem a eficiência dos bancos demonstrando a necessidade de reduzir despesas e alocar recursos da forma mais eficiente possível PALAVRASCHAVE Instituições Financeiras Eficiência Investimentos em Tecnologia da Informação Fronteira estocástica ABSTRACT This study evaluates the relationship between information technology IT expenditures and the efficiency of the main Brazilian banks Technological evolution has strongly impacted the financial sector leading to the availability of digital platforms and virtual customer service channels The higher level of technology adopted by banks can contribute to reducing operating costs and raising the level of efficiency of institutions which may result in improved services provided to customers and easier access to credit by the population The analysis of the relationship between spending on technology and efficiency was performed using the Stochastic Frontier Approach SFA based on the financial statements of the five largest Brazilian banks in the period from 2012 to 2022 which represent 76 of sector totals assets The results showed a positive relationship between the flow of IT expenses and the level of efficiency of the banks which means a higher level of profitability At the same time the balance of IT investments was not statistically significant which can be understood as an indication of a change in the profile of technology spending by banks with the migration from the IT proprietary model of business and solutions to a model more focused on services In addition it was observed that a greater number of specialized service points and lower administrative expenses favor the efficiency of banks demonstrating the need to reduce expenses and allocate resources in the most efficient way possible 6 KEYWORDS Financial Institutions Efficiency Investments in Information Technology Stochastic frontier CLASSIFICAÇÃO JEL C13 G21 O33 7 1 INTRODUÇÃO As Instituições Financeiras são fundamentais para a dinâmica da economia e desenvolvimento socioeconômico do País tendo em vista que os Bancos disponibilizam serviços financeiros acesso ao crédito e soluções de pagamentos dentre outros serviços presentes nas relações econômicas entre indivíduos e empresas Os bancos aparecem como um dos segmentos que mais têm investido em Tecnologia da Informação TI no Brasil e mundo Febraban 2023 tendo grande parte de seus produtos e serviços dependentes dessa tecnologia o que tem resultado na elevação do uso de soluções e compartilhamento de informações seguras e rápidas redução do custo de transações e geração de maior comodidade e satisfação aos usuários É razoável que os gastos em TI sejam indutores da transformação digital e que contribuam para a melhoria de produtos e serviços bem como para a automação e otimização de processos operacionais Sistemas de gerenciamento de riscos e compliance por exemplo podem reduzir o tempo necessário para cumprir regulamentações e aumentar a precisão das avaliações de risco Além disso sistemas de processamento de dados em tempo real permitem que as instituições financeiras tomem decisões mais fundamentadas e ágeis contribuindo para a melhoria da eficiência O conceito de eficiência se refere à capacidade de utilizar os recursos disponíveis da maneira mais produtiva possível para maximizar a produção de bens e serviços Farrel 1957 A eficiência é um aspecto fundamental na teoria econômica e na gestão da produção pois busca alcançar o equilíbrio entre o uso adequado dos recursos e a obtenção dos melhores resultados A teoria da produtividade total dos fatores PTF mede a eficiência com que os fatores de produção incluindo o capital o trabalho e a tecnologia são combinados para gerar a produção Solow 1957 Kendrick 1961 Em linha com esse raciocínio a PTF poderia ser utilizada para relacionar gastos com tecnologia da informação e a melhoria da eficiência de uma instituição financeira 8 Assim se uma instituição financeira aumenta seus gastos com tecnologia isso pode contribuir para melhorar sua PTF e consequentemente sua eficiência Isso ocorre porque a tecnologia pode ajudar a reduzir custos operacionais melhorar o processamento de dados automatizar processos e melhorar a tomada de decisões Contudo o fator determinante para a elevação do nível de eficiência da empresa medido pela sua lucratividade não é o volume dos gastos com tecnologia mas o retorno desses mesmos investimentos e seus impactos na geração de receitas na redução de custos e na mitigação de riscos A partir desse entendimento este estudo empírico pretende analisar se os gastos com tecnologia da informação nos principais bancos brasileiros possuem relação com seus índices de eficiência E se sim pontuar se de forma positiva ou negativa Para atingimento deste objetivo geral serão avaliados outros aspectos relacionados ao tema como i estimar a eficiência dos principais bancos brasileiros ii estabelecer um ranking dos bancos quanto à eficiência iii analisar como os investimentos em tecnologia impactam nos índices de eficiência e iv analisar se a eficiência dos bancos brasileiros é impactada pelos ciclos econômicos representados pela taxa real de juros da economia Na análise da relação entre os gastos com tecnologia da informação e os índices de eficiência dos bancos brasileiros será utilizado o método paramétrico de estimação de eficiência de Fronteira Estocástica SFA Stochastic Frontier Approach ampliando a literatura sobre o tema por considerar dados em painel no período de 2012 a 2022 e por analisar não somente a relação com o saldo estoque dos investimentos em TI mas também quanto aos fluxos correntes de despesas com tecnologia Conhecer os gastos em TI dos principais bancos brasileiros e o seu impacto na eficiência dessas empresas é uma questão essencial para o setor bancário que atua em um ambiente extremamente competitivo e cada vez com menos barreiras a entrantes sendo informação relevante para direcionar a atuação dos gestores o alinhamento dos negócios e o esforço organizacional 9 A elevação da eficiência dos Bancos se mostra relevante na medida em que pode auxiliar na melhoria dos serviços prestados e facilitar o acesso ao crédito à população Além disso a atuação eficiente dos grandes conglomerados financeiros é fator relevante para a estabilidade da economia elevação do grau de confiança do investidor e promoção do desenvolvimento econômico do país A Figura 1 nos mostra a evolução anual do lucro líquido dos cinco maiores bancos brasileiros nos últimos 11 anos 2012 a 2022 onde observamos elevação significativa em valores absolutos que passou do patamar de R 854 bilhões em 2012 para R 1062 bilhões em 2022 correspondendo a um aumento de lucratividade de 24 Figura 1 Lucro líquido anual dos principais bancos brasileiros Fonte Elaborado pelo autor Dados do BCB valores deflacionados trazidos à data base 122022 No mesmo período 2012 a 2022 observamos elevados gastos com tecnologia da informação TI conforme constante na Figura 2 cujo montante anual chegou a atingir em 2016 o equivalente a 38 sobre o lucro líquido anual das instituições o que corresponde ao valor de R 271 bilhões 854 883 890 996 713 902 971 1249 861 1112 1062 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 Lucro R bilhão Variação 10 Figura 2 Gastos com TI dos principais bancos brasileiros Fonte Elaborado pelo autor com base nas demonstrações financeiras dos bancos valores deflacionados trazidos à data base 122022 Cabe ressaltar que uma das principais fontes de receita dos bancos é a margem financeira proveniente da taxa de juros cobrada de seus clientes pelos empréstimos tomados estando esse patamar de juros relacionado à taxa de juros da economia Os juros são uma forma de remunerar o risco que os bancos assumem ao conceder empréstimos aos seus clientes e quanto maior o risco maior deve ser a taxa de juros cobrada Quando a taxa de juros da economia está alta os bancos tendem a aumentar as suas taxas de empréstimo e consequentemente aumentar a sua margem de lucro Por outro lado quando a taxa de juros da economia está baixa os bancos podem ter uma redução na sua margem de lucro já que as taxas de empréstimo também tendem a cair No entanto a baixa taxa de juros pode estimular a demanda por crédito e aumentar o volume de empréstimos o que pode compensar a redução na margem de lucro Além disso uma taxa de juros mais baixa também pode incentivar o consumo e investimento na economia o que pode gerar novas oportunidades de negócios e lucro para os bancos 289 286 292 272 271 274 275 268 256 238 239 34 32 33 27 38 30 28 21 30 21 22 0 10 20 30 40 50 60 70 0 5 10 15 20 25 30 35 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 Gastos TI R Bi Gastos TI Lucro 11 Para considerar os efeitos da inflação calculamos a taxa real de juros da economia ajustando a taxa SELIC ao índice de inflação tendo sido considerado o Índice Nacional de Preços ao Consumidor Amplo IPCAIBGE onde observamos no mesmo período uma variação na taxa real de juros entre o percentual negativo de 378 aa e o patamar máximo de juros positivos de 752 aa no fechamento de 2022 conforme Figura 3 Figura 3 Comparação entre a Taxa SELIC e a Taxa Real de Juros aa 2012 a 2022 Fonte Elaborado pelo autor Dados do BCB e IBGE 2 REFERENCIAL TEÓRICO A Produtividade Total dos Fatores PTF é uma importante forma de análise econômica que busca explicar a relação entre os insumos utilizados na produção e a produção final Essa teoria foi desenvolvida ao longo do tempo por diversos autores como Solow 1956 Jorgenson e Griliches 1967 e Mankiw Romer e Weil 1992 A PTF defende que a produção final de uma economia depende não somente dos insumos utilizados na produção mas também da maneira como esses insumos são combinados e utilizados Assim é possível que uma economia que utiliza uma quantidade 378 752 6 4 2 0 2 4 6 8 10 12 14 16 1T12 2T12 3T12 4T12 1T13 2T13 3T13 4T13 1T14 2T14 3T14 4T14 1T15 2T15 3T15 4T15 1T16 2T16 3T16 4T16 1T17 2T17 3T17 4T17 1T18 2T18 3T18 4T18 1T19 2T19 3T19 4T19 1T20 2T20 3T20 4T20 1T21 2T21 3T21 4T21 1T22 2T22 3T22 4T22 SELIC Taxa Real de Juros 12 menor de insumos possa produzir mais do que outra que utiliza uma quantidade maior de insumos simplesmente porque os insumos são utilizados de forma mais eficiente Além disso a elevação do nível de tecnologia tem importância fundamental na produtividade dos fatores pois permite a utilização mais eficiente dos insumos reduzindo o desperdício de recursos e aumentando a produção A introdução de novas tecnologias pode aumentar a produtividade dos trabalhadores reduzir o tempo de produção e aumentar a qualidade do produto final A importância da tecnologia na produtividade foi destacada por diversos autores a exemplo de Solow 1957 que mostrou que a maior parte do crescimento econômico nos Estados Unidos desde a Segunda Guerra Mundial não podia ser explicada pelo aumento da quantidade de capital e trabalho mas sim pela elevação do nível de tecnologia A PTF consiste portanto em uma importante forma de entender a relação entre os insumos utilizados na produção e a produção final sendo a elevação do nível de tecnologia um fator essencial para aumentar a eficiência na utilização dos insumos e consequentemente a produtividade dos fatores A transformação digital tem sido um dos principais fatores de mudança nos últimos anos no âmbito do setor financeiro não restando dúvidas de que os bancos estão se adaptando rapidamente a essa nova realidade Os bancos estão investindo em tecnologia para melhorar sua eficiência operacional oferecer serviços personalizados e elevar a satisfação dos clientes Por exemplo a implementação de chatbots e assistentes virtuais está melhorando a experiência dos clientes ao lidar com questões simples e rotineiras Além disso a tecnologia de análise de dados tem permitido aos bancos oferecer serviços mais personalizados aos seus clientes como ofertas de produtos financeiros específicos a cada perfil A transformação digital dos bancos é condição necessária para o setor acompanhar o rápido avanço de novas tecnologias do mundo corporativo cuja disseminação tem alterado as condições de mercado e facilitado a entrada de novos competidores sejam eles empresas 13 de tecnologia voltadas ao segmento financeiro as chamadas fintechs ou mesmo grandes empresas de tecnologia que começaram a disponibilizar serviços financeiros as denominadas big techs Essa transformação digital vai além de apenas disponibilizar produtos ou serviços que cada vez mais se tornam commodities financeiras mas sim altera a forma de atender e de se relacionar com os consumidores o que exige uma alteração de posicionamento e de estratégia por parte dos bancos sendo fundamental oferecer aos seus clientes uma melhor experiência com atendimento conveniente ágil e integral interligados em todos os canais físicos e digitais Os gastos e investimentos em TI se mostram relevantes nesse cenário contribuindo para a sustentabilidade dos negócios que cada vez mais se tornam fortemente dependentes de tecnologia Assim esperase observar uma relação positiva entre os gastos com tecnologia da informação e o nível de eficiência dos bancos demonstrando que tais gastos de fato contribuem para a elevação da eficiência O estudo de Feng e Wu 2018 analisou a relação entre o aumento dos investimentos em TI e a performance e valor de mercado de bancos americanos listados em bolsa no período de 2000 a 2017 utilizando modelo de efeitos fixos com análise de regressão linear Os resultados indicaram que investimentos em TI melhoram a eficiência dos bancos principalmente nos grandes bancos Por outro lado o estudo de Gupta et al 2018 utilizando a análise de fronteira estocástica SFA não confirmou a presença de uma relação positiva entre os gastos com TI e a eficiência do lucro no setor bancário indiano tendo como base os dados coletados dos relatórios anuais de dez bancos nacionalizados e três bancos do setor privado no período de 2006 a 2013 Zuo et al 2021 utilizandose da técnica nãoparamétrica de análise envoltória de dados DEA analisou o impacto da transformação digital em 20 bancos comerciais chineses de 2011 a 2019 tendo verificado que os investimentos em tecnologia contribuem para uma 14 melhoria substancial na eficiência produtiva dos bancos embora tenha constatado heterogeneidade entre os bancos Outro estudo de Farouk e Dandago 2015 investigou o impacto do investimento em TI no desempenho financeiro dentre 24 bancos nigerianos no período de 2006 a 2010 por meio do modelo de regressão de dados em painel tendo verificado que deter elevado nível de investimento em TI não aumenta a lucratividade dos bancos Os estudos em âmbito internacional indicam que a relação entre investimentos em TI e a eficiência dos bancos depende de vários fatores tais como o tamanho do banco composição do seu capital social e o nível de concorrência no mercado No âmbito dos estudos já realizados no setor bancário brasileiro Becker Lunardi e Maçada 2003 utilizando DEA para medir a eficiência técnica de 74 bancos brasileiros no ano de 2000 concluíram que os bancos que mais investem em TI apresentam melhores índices de eficiência Ruiz et al 2008 analisando a eficiência técnica das instituições financeiras no Brasil entre 1995 e 2005 por meio de fronteira estocástica de custo nos mostraram que o aumento da taxa de juros da economia representado pela taxa média do CBond implica no aumento de custos para as instituições financeiras e resulta em uma perda de eficiência Bombonatti Filho et al 2016 com base na análise qualitativa de dados dos cinco principais bancos brasileiros do período de 2009 a 2014 verificaram que os bancos que mais investem em tecnologia são os que mais crescem em termos de receitas e lucro líquido apesar das turbulências econômicas no Brasil e no mundo Mendonça et al 2018 utilizaram SFA para avaliar a eficiência de 47 Instituições financeiras no Brasil no período de 2008 a 2015 adotando a forma funcional translog para função de custo tendo identificado a existência de relação positiva entre a eficiência na gestão de custos e a rentabilidade das instituições financeiras Freitas e Kirch 2019 a partir de um modelo de regressão linear verificaram que o investimento em tecnologia da informação pelos bancos possui uma relação 15 significativamente positiva com as variáveis de rentabilidade tendo por base 19 bancos brasileiros listados em bolsa no período 2011 a 2017 Barreto et al 2021 por meio de análise de correlação e regressão de dados em painel identificaram uma relação positiva entre os investimentos em TI e o Retorno sobre os Ativos ROA nos cinco maiores bancos brasileiros no período de 2009 a 2018 Este estudo pretende contribuir com a avaliação da relação entre gastos com TI e a lucratividade no setor bancário brasileiro adicionando uma nova dimensão à literatura existente e fornecendo subsídios para futuras pesquisas nesta área como se destaca mais especificamente abaixo Este trabalho difere do realizado por Becker Lunardi e Maçada 2003 por adotar a metodologia SFA e considerar dados em painel para capturar a evolução de eficiência ao longo do período de 2012 a 2022 Diferente da abordagem de Bombonatti Filho 2016 esta pesquisa irá analisar a relação entre os gastos de TI e o índice de eficiência das instituições financeiras utilizando a abordagem de fronteira estocástica Adicionalmente à análise de Freitas e Kirch 2019 e de Barreto 2021 este trabalho irá verificar a relação do fluxo de gastos com tecnologia da informação sobre a lucratividade dos bancos além de também analisar o impacto do saldo estoque dos investimentos em TI Em complemento aos achados de Ruiz 2008 este trabalho irá verificar a relação entre a taxa real de juros da economia representada pela diferença entre a taxa de juros da economia taxa SELIC e o índice de inflação do período INPC e a eficiência das instituições financeiras 16 3 METODOLOGIA 31 Estimação da Eficiência As técnicas paramétricas e não paramétricas de medição da eficiência são amplamente utilizadas para avaliar a eficiência de unidades de decisão em diversas áreas como finanças saúde e educação A Análise Envoltória de Dados DEA e a Análise de Fronteira Estocástica SFA são duas técnicas populares de medição da eficiência que utilizam abordagens diferentes para avaliar a eficiência de unidades de decisão A DEA foi proposta por Charnes et al 1978 e é uma técnica não paramétrica que avalia a eficiência relativa de unidades de decisão que possuem múltiplas entradas e saídas A SFA foi proposta por Aigner et al 1977 e é uma técnica paramétrica que utiliza a especificação de uma função de produção para avaliar a eficiência de unidades de decisão Tanto a DEA quanto a SFA têm vantagens e limitações e a escolha da técnica mais adequada depende das características dos dados e dos objetivos da análise Schmidt e Sickles 1984 propuseram a SFA como uma alternativa à DEA para lidar com dados de produção e discutiram as principais diferenças entre as duas técnicas Coelli et al 2005 discutiram as principais vantagens e limitações da DEA e SFA e apresentaram estudos comparativos entre elas A DEA se mostra mais adequada para dados em que não há conhecimento prévio da distribuição e pode lidar com várias entradas e saídas A SFA por outro lado requer a especificação de uma função de produção e pressupõe uma distribuição probabilística específica para os erros mas pode ser mais eficiente na identificação de unidades ineficientes Como forma de avaliar a eficiência dos bancos brasileiros ao longo do tempo será utilizado neste trabalho o método paramétrico de estimação de eficiência de Fronteira Estocástica SFA tendo em vista este decompor a ineficiência em duas partes uma decorrente de efeitos aleatórios eou erros de mensuração e a outra representando a ineficiência resultante de fatores que estão sob influência da gestão 17 Modelo de Fronteira Estocástica SFA y f x vit uit 1 Relação de produção entre insumos x vetor e um produto y com vit efeitos aleatórios iid N 0 σv2 uit ineficiência produtiva iid NT mit σu2 Será estimada uma fronteira de produção que permite apenas um output baseado na forma funcional Translog e na distribuição de probabilidades normal truncada sendo considerados os efeitos variáveis no tempo de forma a capturar a evolução tecnológica do setor bancário no período em análise Tais definições são fundamentadas pelos testes de especificação realizados para definição do modelo mais adequado aos dados conforme se verá adiante na seção de resultados Também serão incluídas variáveis explicativas da ineficiência com objetivo de evidenciar a relação entre as variáveis de interesse e a eficiência das instituições 31 Estimação da Eficiência Na estimação da função de produção a especificação do output dos inputs e das variáveis explicativas será Função de produção Output Lucro Líquido LL Input 1 Ativo Total AT Input 2 Intermediação Financeira Despesa sobre Receita IF Input 3 1 Despesas Totais DT Variáveis explicativas da ineficiência Gastos com Tecnologia da Informação GTI 18 Saldo dos Investimentos em Tecnologia da Informação SITI Taxa Real de Juros da Economia JUROS Despesas Administrativas DA Despesas de Pessoal DP Número de Agências AG Número de Pontos de Atendimento PA Assim o modelo de estimação na forma funcional translog pode ser expresso por ln 𝐿𝐿𝑖𝑡 𝛽0 𝛽t t 𝛽AT ln 𝐴𝑇𝑖𝑡 𝛽IF ln 𝐼𝐹𝑖𝑡 𝛽DT ln 𝐷𝑇𝑖𝑡 1 2 𝛽tt 𝑡2 1 2 𝛽AT AT 𝑙𝑛 𝐴𝑇𝑖𝑡2 1 2 𝛽IF IF 𝑙𝑛 𝐼𝐹𝑖𝑡2 1 2 𝛽DT DT 𝑙𝑛 𝐷𝑇𝑖𝑡2 1 2 𝛽AT IF 𝑙𝑛 𝐴𝑇𝑖𝑡𝑙𝑛 𝐼𝐹𝑖𝑡 1 2 𝛽AT DT 𝑙𝑛 𝐴𝑇𝑖𝑡𝑙𝑛 𝐷𝑇𝑖𝑡 1 2 𝛽IF DT 𝑙𝑛 𝐼𝐹𝑖𝑡𝑙𝑛 𝐷𝑇𝑖𝑡 𝛽AT t 𝑙𝑛 𝐴𝑇𝑖𝑡 𝑡 𝛽IF t 𝑙𝑛 𝐼𝐹𝑖𝑡 𝑡 𝛽DT t 𝑙𝑛 𝐷𝑇𝑖𝑡 𝑡 𝑣it 𝑢it 2 Onde uit δ0 δ1 GTIit δ2 SITIit δ3 JUROSit δ4 DAit δ5 DPit δ6 AGit δ7 PAit 3 O principal indicador de resultado dos bancos considerado pelo mercado é o lucro líquido sendo este definido como output da função de produção Cabe destacar que os bancos se caracterizam como uma indústria financeira de serviços não produzindo unidades físicas e sim se baseando em unidades monetárias Os bancos fundamentam sua atividade principal pela intermediação financeira cujo valor agregado está representado pelo Ativo Total o que abrange dentre outros o saldo das operações de crédito títulos e valores mobiliários e aplicações interfinanceiras A eficiência se fundamenta na busca por melhorar a relação entre resultados e custos buscando maximizar resultados e minimizar custos Nesse sentido analisar a relação entre as receitas e despesas com a intermediação financeira é importante para avaliar a capacidade dos bancos em gerar resultado As despesas totais foram calculadas pela soma das despesas administrativas e as despesas de pessoal sendo definidas como inputs do modelo por estarem fortemente 19 relacionadas aos gastos necessários para suporte às operações dos bancos e ao funcionamento da infraestrutura de atendimento Os Gastos com Tecnologia da Informação GTI foram considerados como sendo a soma das despesas com Processamento de Dados despesas com Comunicações de dados despesas com a amortização de softwares adquiridos e desenvolvidos internamente ativo intangível e despesas com a depreciação de sistemas de processamento de dados ativo tangível Importante destacar que essas variáveis representam fluxos correntes de despesas com TI Por outro lado o saldo dos Investimentos em Tecnologia da Informação SITI será representado pelo saldo estoque do ativo intangível correspondente à aquisição ou desenvolvimento de softwares e pelo saldo estoque do ativo imobilizado de uso referente aos sistemas de processamentos de dados Cabe destacar que não existe uma nomenclatura padrão na divulgação dos investimentos em TI pelos bancos brasileiros Alves et al 2018 sendo que as definições acima permitem abranger de forma mais assertiva os gastos e investimentos com tecnologia da informação nas empresas sejam eles investimentos diretos CAPEX despesas de custeio OPEX ou mesmo desenvolvimentos internos ou contratados de softwares Este critério também contempla a recente mudança do perfil de gastos em tecnologia dos bancos com a migração das tecnologias que sustentam os negócios para serviços XaaS Everything as a Service a exemplo de infraestrutura software plataformas digitais etc Grande parte do custo da indústria financeira que é baseada na prestação de serviços reside na estrutura física de atendimento aos clientes representados pelo número de agências bancárias e de postos de atendimento sendo também selecionadas como variáveis explicativas A taxa real de juros será a taxa nominal de juros ajustada pela inflação calculada pela fórmula taxa real de juros 1 taxa nominal de juros 1 taxa de inflação 1 A taxa nominal de juros considerada será a SELIC que representa a taxa básica de juros da economia 20 e o índice de inflação a ser considerado será o Índice Nacional de Preços ao Consumidor Amplo IPCAIBGE Todas as variáveis monetárias são mensuradas em moeda nacional R obtidas com valores nominais e ajustadas para considerar os efeitos da inflação pelo IPCA trazendoas a valor presente na data base de dezembro2022 último trimestre em avaliação O escopo do estudo será limitado à análise dos cinco maiores bancos do Brasil representados pelo Banco do Brasil BB Bradesco Caixa Econômica Federal Caixa Itaú e Santander que representam 76 dos Ativos Totais das instituições classificadas na categoria B1 do Banco Central do Brasil BCB atualmente composta por 99 empresas do tipo Banco Comercial Banco Múltiplo com Carteira Comercial ou Caixas Econômicas A restrição aos cinco principais bancos se justifica na falta de padronização das informações dos gastos e investimentos em TI nos bancos brasileiros cuja estruturação de uma base de dados com alto nível de precisão e completude exige a análise detalhada das demonstrações financeiras das instituições e extração das informações contidas em suas Notas Explicativas A alternativa à análise das demonstrações financeiras seria acessar de forma padronizada o saldo das contas contábeis dos bancos junto ao Banco Central do Brasil BCB no Portal do Cosif Plano Contábil das Instituições do Sistema Financeiro Nacional Porém os saldos dessas contas não correspondem exclusivamente aos investimentos e gastos com TI mas sim aos valores por categoria do ativo ou do passivo sendo necessário portanto assumir valores por meio do estabelecimento de proxy o que poderia prejudicar as estimações do modelo Assim serão utilizados dados trimestrais dos cinco maiores bancos no período de 44 trimestres ou 11 anos de março2012 a dezembro2022 obtidos por meio do IFdata Base de dados selecionados de instituições financeiras brasileiras mantida pelo Banco Central do Brasil BCB Os demais dados financeiros relacionados aos gastos e investimentos de TI serão extraídos das demonstrações contábeis trimestrais dos bancos disponíveis nos respectivos 21 sites de Relações com Investidores RI o que implica na análise de 220 demonstrações financeiras O período selecionado está alinhado ao tempo máximo de um ciclo econômico definido por Juglar que possui duração de 7 a 11 anos Korotayev e Tsirel 2010 abrangendo os movimentos de expansão e contração da economia embora os ciclos econômicos possam ser afetados por fatores como mudanças nas políticas econômicas eventos geopolíticos e inovações tecnológicas conforme Burns 1934 Além disso o período de 11 anos corresponde a duas vezes o tempo de evolução das tecnologias digitais estabelecido por Brynjolfsson e Mcafee 2011 que ponderam que o prazo de 5 a 6 anos é suficiente para que uma nova tecnologia seja desenvolvida testada aprimorada e adotada em larga escala pela sociedade A tabela 1 a seguir apresenta a análise descritiva das variáveis estabelecidas no modelo de estimação de eficiência de Fronteira Estocástica SFA abrangendo o output e os inputs da função de produção bem como as variáveis explicativas da ineficiência cujos valores foram apurados com base trimestral e no período de 2012 a 2022 Tabela 1 Análise descritiva das variáveis Variáveis Média Mediana Desvio padrão Máximo Mínimo Lucro Líquido LL a 4769 4760 2258 13069 1839 Ativo Total AT a 1566337 1632403 388689 2359739 796640 Intermediação Financeira Despesa sobre Receita IF c 0791 0747 0222 2173 0410 Despesas Totais DT a 10402 10929 2512 15468 4415 Gastos com TI GTI a 1346 1178 510 2360 370 Saldo dos Investimentos em TI SITI a 5875 4727 2778 12742 1495 Taxa Real de Juros JUROS d 2686 3131 2827 7524 3784 Despesas Administrativas DA a 5080 5438 1550 8333 2012 Despesas de Pessoal DP a 5321 5654 1766 9198 1736 Número de Agências AG b 3746 3408 913 5556 2372 Número de Pontos de Atendimento PA b 4891 4519 1042 7048 2976 22 Dados com base em 220 observações a Em R Milhão em valores deflacionados trazidos à data base 122022 b Em unidades de pontos c Razão entre as Despesas de Intermediação Financeira e as Receitas de Intermediação Financeira d Em aa A visão da evolução temporal das variáveis explicativas da ineficiência a serem consideradas GTI Gastos com TI e SITI Saldo dos Investimentos em TI permite observar a variação em valores deflacionados do período em análise trazidos à data base 122022 conforme demonstrado nas Figuras 4 e 5 a seguir cuja relação com a eficiência das empresas se pretende verificar neste estudo Figura 4 Comparativo dos gastos com TI entre os bancos 20122022 Fonte Elaborado pelo autor com base nas demonstrações financeiras dos bancos Consideradas as despesas com Processamento de Dados Comunicações de Dados Amortização de Softwares e Depreciação de Sistemas de Processamento de Dados 2228 6551 3525 7292 4258 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 R MILHÃO BB Bradesco Caixa Itaú Santander 23 Figura 5 Comparativo do saldo de investimentos em TI entre os bancos 20122022 Fonte Elaborado pelo autor com base nas demonstrações financeiras dos bancos Considerados os saldos com Softwares Adquiridos e Desenvolvidos Internamente Ativo Intangível e com Sistema de Processamento de Dados Ativo Tangível 4 RESULTADOS 41 Testes de Especificação Com o objetivo de determinar a especificação ideal do modelo a ser analisado foi realizado teste de razão de verossimilhança unicaudal Teste LR para determinar se o modelo ajustado com base na forma funcional Translog supera o modelo CobbDouglas O teste LR rejeitou a hipótese nula referente à versão restrita do modelo Cobb Douglas em apoio à forma funcional Translog Os valores resultantes dos testes foram maiores do que os valores listados na Tabela de Kodde e Palm2 para os respectivos graus de 2 Os valores críticos para o Teste de Razão de Verossimilhança encontramse na Tabela 1 de David A Kodde e Franz C Palm 1986 sendo justificado o uso da tabela uma vez que a estatística apresenta uma distribuição qui quadrada mista 5493 10473 2193 12742 5055 0 2000 4000 6000 8000 10000 12000 14000 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 R MILHÃO BB Bradesco Caixa Itaú Santander 24 liberdade indicando que o modelo Translog onde todos os parâmetros foram considerados superou o modelo restrito conforme dados dispostos na tabela 2 Tabela 2 Teste LR de Especificação para Fronteira Estocástica Teste Unrestricted log likelihood function ULLF Restricted log likelihood function RLLF λ a X2 gl b Valores Críticos c Resultado Translog 57187 62044 9713 5 8574 Aceitase Translog a λ 2ULLF RLLF diferença entre o modelo completo de Translog e o modelo restrito Ho b Graus de Liberdade df c Valores Críticos constantes da Tabela de Kodde e Palm para 10 de significância Também foi realizado o teste de Hausman 1978 para comparar os efeitos fixos e efeitos aleatórios do modelo de fronteira estocástica e avaliar qual modelo é mais apropriado para os dados O teste de Hausman apresentou valorp de 00463 inferior ao nível de significância de 5 rejeitando a hipótese nula onde o modelo de efeitos fixos é mais apropriado e sugerindo que os efeitos aleatórios são consistentes e que não há correlação entre as variáveis explicativas e os efeitos não observados Quanto à definição do modelo de distribuição de probabilidades foram realizadas as estimações considerando os modelos Semi Normal e Normal Truncada A distribuição Semi Normal assume que os erros da fronteira estocástica são simétricos e permite que tenha uma maior variabilidade em relação à média Por outro lado a distribuição Normal Truncada assume que os erros são normalmente distribuídos mas com truncamento em uma determinada região Essa distribuição é mais restritiva pois impõe a suposição de normalidade nos erros o que pode ser apropriado em algumas situações em que a variabilidade dos erros é relativamente pequena e não há evidências de assimetria ou outliers 25 As estimações realizadas nos modelos de distribuição não apresentaram variações nos valores de loglikelihoods e resultaram nos mesmos valores no Critério de Informação de Akaike AIC e no Critério Bayesiano de Schwarz BIC indicando que ambos os modelos têm um ajuste similar em relação à complexidade Assim optouse pela definição da distribuição Normal Truncada por não haver evidências de assimetria ou de outliers na análise dos resíduos assumido que os erros na estimação da fronteira estocástica seguem uma distribuição normal e não apresentam variabilidade excessiva O modelo proposto possui 3 inputs na função de produção e 7 variáveis explicativas para uma base com 220 observações número amplo em relação ao número de preditores o que contribui para a sua robustez O valorp associado ao teste F resultou no valor 42e13 bem abaixo do nível de significância prédeterminado de 005 sugerindo que modelo de regressão como um todo é estatisticamente significativo e tem poder preditivo e explicativo sobre a variável dependente 42 Resultado da Estimação A estimação dos escores de eficiência dos principais bancos brasileiros resultou em valores de eficiência média calculada pela média aritmética das eficiências dispostas na Figura 6 onde verificamos leve tendência de elevação ao longo do período avaliado em que pese a ocorrência de oscilações em períodos específicos como o observado no último trimestre2022 que se explica pelo aumento na provisão para créditos de difícil liquidação eou efetivação da contabilização de perdas como o ocorrido no caso das Lojas Americanas 26 Figura 6 Eficiência média dos bancos 20122022 A evolução da eficiência de cada banco está representada na Figura 7 sendo que os índices devem assumir valores entre 0 e 1 e quanto mais próximo de 1 mais eficiente será o banco em sua capacidade de gerar lucro Figura 7 Evolução dos índices de eficiência dos bancos 515 25 35 45 55 65 75 85 95 1T12 2T12 3T12 4T12 1T13 2T13 3T13 4T13 1T14 2T14 3T14 4T14 1T15 2T15 3T15 4T15 1T16 2T16 3T16 4T16 1T17 2T17 3T17 4T17 1T18 2T18 3T18 4T18 1T19 2T19 3T19 4T19 1T20 2T20 3T20 4T20 1T21 2T21 3T21 4T21 1T22 2T22 3T22 4T22 ESCORES DE EFICIÊNCIA ÚLTIMOS 44 TRIMESTRES EFFMedia Linear EFFMedia 00 01 02 03 04 05 06 07 08 09 10 1T12 2T12 3T12 4T12 1T13 2T13 3T13 4T13 1T14 2T14 3T14 4T14 1T15 2T15 3T15 4T15 1T16 2T16 3T16 4T16 1T17 2T17 3T17 4T17 1T18 2T18 3T18 4T18 1T19 2T19 3T19 4T19 1T20 2T20 3T20 4T20 1T21 2T21 3T21 4T21 1T22 2T22 3T22 4T22 ESCORE DE EFICIÊNCIA BB Bradesco Caixa Itaú Santander 27 A tabela 3 apresenta o resultado da estimação com os coeficientes das variáveis de entrada do modelo inputs e das variáveis explicativas da ineficiência dispostas na primeira coluna Tabela 3 Resultado da estimação Coeficiente Estimativa Desvio Padrão valor z Prz Intercepto 𝛽0 32 e05 84 e04 38 00002 logAtivo Total 𝛽AT 58 e04 12 e04 50 00000 logIntermediação Financeira 𝛽IF 90 e03 56 e03 16 01068 log1Despesa Total 𝛽DT 20 e04 60 e03 33 00008 logAtivo Total2 𝛽AT AT 48 e03 99 e02 49 00000 logAtivo Total x logIntermediação Financeira 𝛽AT IF 10 e03 53 e02 20 00492 logAtivo Total x log1Despesa Total 𝛽AT DT 12 e03 74 e02 17 00975 logIntermediação Financeira2 𝛽IF IF 55 e02 37 e02 15 01349 logIntermediação Financeira x log1Despesa Total 𝛽IF DT 60 e02 38 e02 16 01102 log1Despesa Total2 𝛽DT DT 29 e02 97 e02 03 07643 ZGastos com TI GTI δ1 13 e00 45 e01 28 00046 ZSaldo dos Investimentos em TI SITI δ2 92 e02 57 e02 16 01077 ZTaxa Real de Juros JUROS δ3 96 e01 29 e01 33 00010 ZDespesas Administrativas DA δ4 12 e01 13 e01 09 03613 ZDespesas de Pessoal DP δ5 45 e01 84 e02 53 00000 ZNúmero de Agências AG δ6 26 e00 59 e01 44 00000 ZNúmero de Pontos de Atendimento PA δ7 17 e00 41 e01 42 00000 sigmaSq σ2 29 e02 62 e01 47 00000 gamma ƴ 91 e02 33 e01 278 2e16 não significativo significativo a 10 significativo a 5 significativo a 1 significativo a 01 Observamos que no resultado da função de produção se destacou a contribuição do Ativo Total sobre o índice de eficiência resultando no valor positivo para o coeficiente 𝛽AT e com alta significância estatística 01 o que demonstra que a capacidade de geração de resultados nos bancos está positivamente relacionada ao volume de ativos dessas 28 instituições cujo montante desses cinco maiores bancos brasileiros totalizou em dezembro2022 o valor de R 848 trilhões Verificamos que controlar as despesas totais eleva o lucro dos bancos de forma estatisticamente significativa corroborando os achados de Mendonça et al 2018 que pondera que quanto mais eficiente é a gestão de custos da empresa maior será sua rentabilidade A estimativa com relação à razão do volume das despesas sobre as receitas com intermediação financeira não apresentou significância estatística sugerindo que os bancos ponderam a realização de suas operações às condições de vantajosidade do negócio e de exposição ao risco refletindo as condições econômicas do momento cuja estratégia de atuação pode implicar na variação do índice de relação entre as despesas e receitas com intermediação financeira decorrente por exemplo de ações de reforço de provisões para crédito de difícil liquidação restrição à concessão de crédito ponderada pelo risco ou da manutenção do patamar de captação de recursos Em consonância ao trabalho de Becker Lunardi e Maçada 2003 verificamos que os gastos com Tecnologia da Informação contribuem de forma relevante para a eficiência dos bancos brasileiros demonstrando a necessidade de alocação de recursos em TI para a elevação do nível de tecnologia e aumento de seu patamar de eficiência Cabe ressaltar que os gastos com Tecnologia da Informação definidos no modelo compreendem tanto as despesas administrativas dispendidas com TI quanto também as despesas com a depreciação e amortização dos investimentos realizados com TI abrangendo dessa forma todas as iniciativas relacionadas aos gastos com tecnologia da informação Nesse sentido observase que o saldo de investimentos de TI não apresentou relevância estatística em relação ao nível de eficiência dos bancos Isso pode ser explicado em razão de que o saldo dos investimentos em TI representa um estoque dos valores dispendidos em Tecnologia da Informação relacionados especificamente à aquisição de softwares desenvolvimento interno de softwares ou compra de sistemas de processamento 29 de dados cujo período prazo de vida útil pode ter duração de até 120 meses dependendo da natureza do investimento Dada a rápida evolução tecnológica experimentada atualmente é possível que este estoque não mais represente adequadamente os impactos da tecnologia mais atual nos lucros correntes o que não acontece com a variável GTI fluxo Tal observação reforça a tendência de mudança do perfil de gastos em tecnologia dos bancos onde verificamos uma migração do modelo proprietário de negócios de TI para o modelo de serviços XaaS Everything as a Service onde as empresas deixam de adquirir os produtos e passam assinar os serviços Observamos que a elevação na Taxa Real de Juros da economia contribui para a ineficiência dos Bancos analisados em consonância aos achados de Ruiz et al 2008 que considerou como indicador macroprudencial a taxa média do CBond Verificamos ainda que reduzir despesas de pessoal contribuem para a elevação da eficiência dos bancos sugerindo que a evolução dos serviços bancários está vinculada à evolução tecnológica do setor que possui como característica ser menos dependente e até mesmo poupadora de mão de obra implicando que funcionários antes alocados em atividades que passaram a ser abrangidas por novas tecnologias são realocados para lidar com outras atividades complexas de negócios e de atendimento das necessidades dos clientes No entanto observamos não haver significância estatística na variação das despesas administrativas excetuadas as despesas de pessoal o que pode ser explicado pela necessidade da empresa em custear não apenas os gastos com TI mas todas as despesas não operacionais necessárias para dar suporte aos negócios dos bancos e ao funcionamento da infraestrutura de atendimento Também constatamos que ter uma maior quantidade de Pontos de Atendimento implica na perda de lucratividade para as instituições o que pode ser explicado pelos custos associados ao funcionamento da infraestrutura física de atendimento desses locais sendo considerado no total desses pontos a soma da quantidade de Agências bancárias e dos Postos 30 de Atendimento estes últimos de menor tamanho e de atendimento restrito seja em público ou em serviços especializados Por outro lado observamos que uma maior quantidade de agências bancárias caracterizadas como pontos de atendimento especializados favorece o lucro das instituições demonstrando a necessidade de se manter uma adequada e ampla rede de atendimento para o atendimento qualificado e segmentado aos clientes 5 CONCLUSÃO O objetivo deste trabalho foi analisar o impacto dos gastos com tecnologia da informação nos índices de eficiência do setor bancário brasileiro por meio da estimação de eficiência usando a metodologia de fronteira estocástica SFA Os resultados permitiram observar correlação positiva entre os fluxos dos gastos com tecnologia da informação e a melhoria dos índices de eficiência dos principais bancos brasileiros no período de 2012 a 2022 tendo por base os dados das demonstrações financeiras trimestrais das instituições Por outro lado o saldo de investimentos de TI estoque não apresentou relevância estatística em relação ao nível de eficiência dos bancos Isso pode ser interpretado como um indicativo de mudança do perfil de gastos em tecnologia dos bancos onde há migração do modelo proprietário de negócios de TI para o modelo de serviços onde as empresas deixam de adquirir os produtos e passam a contratar os serviços Nesse caso a rápida evolução tecnológica experimentada faz com que o estoque acumulado possivelmente defasado não mais represente adequadamente os impactos da tecnologia mais atual nos níveis correntes de lucratividade Não obstante os gastos com TI comporem as despesas administrativas das Instituições Financeiras verificouse que quanto mais controlados são os gastos com pessoal maior será o índice de eficiência média dos bancos demonstrando a necessidade permanente da administração em reduzir despesas ou alocar recursos da forma mais eficiente possível 31 Apesar da transformação digital dos Bancos nos últimos anos e a migração do atendimento para canais digitais a exemplo do Internet Banking e aplicativos em dispositivos móveis APPs observamos que ter uma maior quantidade de pontos de atendimento especializados agências bancárias para atendimento aos clientes e comercialização de produtos e serviços financeiros favorece o lucro dos bancos em que pese uma maior quantidade total de pontos de atendimento agências bancárias somadas aos postos de atendimento ser um fator que contribui para a ineficiência das instituições Constatamos ainda que a elevação na taxa real de juros da economia contribui para a ineficiência dos bancos possivelmente decorrente do aumento de custos para as instituições refletindo questões relacionadas à concorrência restrições e alterações das condições impostas tanto na ponta da captação quanto na da aplicação de recursos financeiros Ainda que este trabalho tenha sido baseado nos cinco maiores bancos brasileiros que representam 76 dos Ativos Totais das instituições classificadas na categoria B1 do Banco Central do Brasil a análise da relação entre a eficiência dos bancos e os seus gastos com TI poderia ser ampliada para todas as 99 empresas da referida categoria de forma a avaliar o impacto nos diversos tipos de segmentos foco de atuação e tamanho das instituições financeiras Também poderiam ser analisadas possíveis alterações no custo da tecnologia ao longo desse período e sua relação comparativa quanto ao volume de investimentos em TI realizados A relação entre o spread bancário e a elevação da eficiência dos bancos fundamentada pelos investimentos em TI é outro ponto possível de ser explorado Esperase por fim que este trabalho traga subsídios para a atuação dos gestores de bancos na tomada de decisão sobre gastos e investimentos em TI promovendo a melhoria de produtos e serviços bancários o aumento da eficiência das instituições financeiras e favorecendo o desenvolvimento econômico do país 32 6 REFERÊNCIAS AIGNER D LOVELL C K SCHMIDT P Formulation and estimation of stochastic frontier production function models Journal of econometrics v 6 n 1 p 2137 1977 ALVES A T A R B A GALEGALE N V SANTOS F A Inversiones en ti los bancos brasileños revelan estas inversiones Revista La Junta v 1 n 1 2018 BARRETO L S PEREIRA V S PENEDO A S T Impacto dos investimentos em tecnologia sobre a rentabilidade do setor bancário brasileiro Future Studies Research Journal Trends and Strategies v 13 n 1 p 94111 2021 BECKER J LUNARDI G MAÇADA A Análise de eficiência dos bancos brasileiros um enfoque nos investimentos realizados em Tecnologia de Informação TI Revista Produção 13 2 p 7081 2003 BOMBONATTI FILHO O GASPAR M A CARDOSO M V COSTA I Investimentos em TI e o acesso da população brasileira aos serviços e facilidades bancárias Revista Espacios v 37 n 26 p 23 2016 BRYNJOLFSSON E MCAFEE A Race against the machine How the digital revolution is accelerating innovation driving productivity and irreversibly transforming employment and the economy 2011 BURNS A F Production trends in the United States since 1870 National Bureau of Economic Research 1934 CHARNES A COOPER W W RHODES E Measuring the efficiency of decision making units European journal of operational research v 2 n 6 p 429444 1978 COELLI T J RAO D S P ODONNELL C J BATTESE G E An introduction to efficiency and productivity analysis Springer science business media 2005 33 FAROUK B K U DANDAGO K I Impact of investment in information technology on financial performance of Nigerian banks Is there a productivity paradox Journal of Internet Banking and Commerce v 20 n 1 p 122 2015 FARREL M J A measurement of productive efficiency Journal of The Royal Statistical Society v120 p254290 1957 Febraban Federação Brasileira de Bancos Pesquisa FEBRABAN de Tecnologia Bancária 2023 2023 FENG Z WU Z Technology investment firm performance and market value evidence from Banks 2018 FREITAS O D KIRCH G Performance dos bancos brasileiros no contexto de digitalização Revista Brasileira de Finanças v 17 n 2 p 3855 2019 GUPTA S D RAYCHAUDHURI A HALDAR S K Information technology and profitability evidence from Indian banking sector International Journal of Emerging Markets 2018 HAUSMAN J A Specification tests in econometrics Econometrica Journal of the econometric society p 12511271 1978 JORGENSON D W GRILICHES Z The explanation of productivity change The review of economic studies v 34 n 3 p 249283 1967 KENDRICK J W Productivity trends in the United States Productivity trends in the United States 1961 KODDE D PALM F Table 1 Upper and lower bounds for the critical value for jointly testing equality and inequality restrictions 1986 34 KOROTAYEV A V TSIREL S V A spectral analysis of world GDP dynamics Kondratieff waves Kuznets swings Juglar and Kitchin cycles in global economic development and the 20082009 economic crisis Structure and Dynamics v 4 n 1 2010 MANKIW N G ROMER D WEIL D N A contribution to the empirics of economic growth The quarterly journal of economics v 107 n 2 p 407437 1992 MENDONÇA D J SOUZA J A E CARVALHO F D M DE BENEDICTO G C Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil Revista Capital Científico v 16 n 3 2018 RUIZ C TABAK B M CAJUEIRO D O Mensuração da eficiência bancária no Brasil a inclusão de indicadores macroprudenciais Revista Brasileira de Finanças v 6 n 3 art 136 p 411436 2008 SCHMIDT P SICKLES R C Production frontiers and panel data Journal of Business Economic Statistics v 2 n 4 p 367374 1984 SOLOW R M A contribution to the theory of economic growth The quarterly journal of economics v 70 n 1 p 6594 1956 SOLOW R M Technical Change and the Aggregate Production Function Review of Economics and Statistics 39 3 August 1957 pp 312320 Real Business Cycles p 543551 2013 ZUO L STRAUSS J ZUO L The digitalization transformation of commercial banks and its impact on sustainable efficiency improvements through investment in science and technology Sustainability v 13 n 19 p 11028 2021 35 FUNDAÇÃO GETULIO VARGAS ESCOLA DE ECONOMIA DE SÃO PAULO LUIS FELIPE GAZQUEZ RUA IMPACTO DE FATORES MACROECONÔMICOS E DA SOLVÊNCIA SOBRE A EFICIÊNCIA BANCÁRIA DOS PRINCIPAIS BANCOS BRASILEIROS SÃO PAULO 2020 LUIS FELIPE GAZQUEZ RUA IMPACTO DE FATORES MACROECONÔMICOS E DA SOLVÊNCIA SOBRE A EFICIÊNCIA BANCÁRIA DOS PRINCIPAIS BANCOS BRASILEIROS Dissertação apresentada à Escola de Economia da Fundação Getulio Vargas como requisito para obtenção do título de Mestre em Economia Campo de Conhecimento Economia bancária Orientador Prof Dr Jéfferson A Colombo SÃO PAULO 2020 Rua Luis Felipe Gazquez Impacto de fatores macroeconômicos e da solvência sobre a eficiência bancária dos principais bancos brasileiros Luis Felipe Gazquez Rua 2020 112 f Orientador Jéfferson Augusto Colombo Dissertação mestrado profissional MPFE Fundação Getulio Vargas Escola de Economia de São Paulo 1 Bancos Brasil 2 Eficiência organizacional 3 Análise estocástica 4 Análise envoltória de dados I Colombo Jéfferson Augusto II Dissertação mestrado profissional MPFE Escola de Economia de São Paulo III Fundação Getulio Vargas IV Título CDU 3367181 Ficha Catalográfica elaborada por Raphael Figueiredo Xavier CRB SP009987O Biblioteca Karl A Boedecker da Fundação Getulio Vargas SP LUIS FELIPE GAZQUEZ RUA IMPACTO DE FATORES MACROECONÔMICOS E DA SOLVÊNCIA SOBRE A EFICIÊNCIA BANCÁRIA DOS PRINCIPAIS BANCOS BRASILEIROS Dissertação apresentada à Escola de Economia da Fundação Getulio Vargas como requisito para obtenção do título de Mestre em Economia Campo de Conhecimento Economia bancária Data de Aprovação 27102020 Banca Examinadora Prof Dr Jéfferson A Colombo orientador FGVEESP Prof Dr Emerson Fernandes Marçal FGVEESP Prof Dr Peter Fernandes Wanke UFRJCOPPEAD Dedico esse trabalho aos meus alicerces Aos meus pais pelo que sou A minha esposa que me faz melhor a cada dia Ao meu filho que nasceu durante o mestrado e fez essa jornada valer ainda mais AGRADECIMENTOS Quero agradecer a minha esposa Jéssica pelo apoio constante e incondicional as palavras de incentivo e a perseverança em acreditar em mim Eu não teria conseguido sem você Obrigado por ter segurado a barra sozinha enquanto eu focava nesse trabalho Ao meu filho que nasceu durante essa titulação Me perdoe pelas ocasiões em que não consegui estar mais próximo Aos meus pais e familiares que fizeram o possível e impossível para que eu tivesse a melhor educação e as melhores condições Tenho muito orgulho de vocês Obrigado pelo que me proporcionaram Obrigado pela família maravilhosa que são Aos amigos pessoais pelas ausências e aos amigos e líderes do trabalho que se dispuseram ao máximo em me ajudar Ao orientador Prof Dr Jéfferson A Colombo pela confiança suporte e prestatividade que conduziu todas as interações A Deus que me proporcionou participar desse desafio e conhecer todas essas pessoas O mundo não será salvo pelos caridosos mas pelos eficientes Roberto Campos RESUMO A literatura acerca da eficiência bancária cresceu muito nos últimos anos No Brasil por conta das transformações no cenário macro e microeconômico que tornaram o ambiente mais competitivo e dinâmico isso não foi diferente Nesse cenário esse estudo tem como objetivo estimar o impacto parcial de choques em variáveis macroeconômicas que representassem expectativas futuras medida de riscopaís atividade econômica e inflação e microeconômicas bancoespecífica que representasse a solvência na eficiência bancária das principais instituições financeiras do Brasil do primeiro trimestre de 2009 até o quarto trimestre de 2019 Os escores de eficiência são estimados através da análise de fronteira estocástica SFA e a análise de sua interrelação com as demais variáveis se dá por meio da aplicação de Modelos Autorregressivos Vetoriais VAR seguidos de suas análises estruturais Em linhas gerais os resultados contribuem para a literatura destacando que cada instituição financeira é própria e que não há uma convergência entra as amostras analisadas para o período destacado Palavraschave Eficiência bancária brasileira Análise de fronteira estocástica SFA Modelo Autorregressivo Vetorial VAR Análise Estrutural VAR ABSTRACT The literature on banking efficiency has grown a lot in recent years In Brazil due to the changes in the macro and microeconomic scenario which made the environment more competitive and dynamic this was no different In this scenario this study aims to estimate the partial impact of shocks on macroeconomic variables representing future expectations country risk measure economic activity and inflation and microeconomic variables bankspecific representing solvency on the banking efficiency of the main financial institutions in Brazil from the first quarter of 2009 to the fourth quarter of 2019 Efficiency scores are estimated through the stochastic frontier analysis SFA and the analysis of their interrelation with the other variables is done through the application of Vector Autoregression Models VAR followed by their structural analysis In general the results contribute to the literature highlighting that each financial institution is its own and that there is no convergence between the samples analyzed for the highlighted period Keywords Brazilian banking efficiency Stochastic frontier analysis SFA Vector autoregression model VAR Structural Analysis VAR LISTA DE FIGURAS Figura 1 Principais etapas da análise VAR 42 Figura 2 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Itaú 61 Figura 3 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Banco do Brasil 62 Figura 4 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Bradesco 63 Figura 5 Função Impulsoresposta das diversas variáveis ao escore de eficiência da Caixa 64 Figura 6 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Santander 65 Figura 7 Função Impulsoresposta das diversas variáveis ao escore de eficiência da média ponderada dos Privados 66 Figura 8 Função Impulsoresposta das diversas variáveis ao escore de eficiência da média ponderada dos Públicos 67 Figura 9 Função Impulsoresposta das diversas variáveis ao escore de eficiência da média ponderada dos bancos 68 Figura 10 Escores de eficiência de custos e demais variáveis no tempo 102 Figura 11 Correlograma cruzado do escore de eficiência de custos do Itaú para cada uma das variáveis do Modelo VAR 106 Figura 12 Correlograma cruzado do escore de eficiência de custos do Banco do Brasil para cada uma das variáveis do Modelo VAR 106 Figura 13 Correlograma cruzado do escore de eficiência de custos do Bradesco para cada uma das variáveis do Modelo VAR 107 Figura 14 Correlograma cruzado do escore de eficiência de custos do Caixa para cada uma das variáveis do Modelo VAR 107 Figura 15 Correlograma cruzado do escore de eficiência de custos do Santander para cada uma das variáveis do Modelo VAR 108 Figura 16 Correlograma cruzado do escore de eficiência de custos da média ponderada dos Privados para cada uma das variáveis do Modelo VAR 108 Figura 17 Correlograma cruzado do escore de eficiência de custos da média ponderada dos Públicos para cada uma das variáveis do Modelo VAR 109 Figura 18 Correlograma cruzado do escore de eficiência de custos da média ponderada Geral para cada uma das variáveis do Modelo VAR 109 Figura 19 Gráficos da Estabilidade dos Modelos VAR para todas as amostras 110 LISTA DE QUADROS Quadro 1 Síntese dos estudos considerados para a identificação das hipóteses 35 Quadro 2 Resultado dos testes DieboldMariano para os critérios MSE MAE e MAPE 76 Quadro 3 Síntese comparativa entre os sinais esperados e os resultados apurados 84 LISTA DE GRÁFICOS Gráfico 1 Abordagem de estimativas nos 100 principais artigos 30 LISTA DE TABELAS Tabela 1 Estatística descritiva das variáveis 56 Tabela 2 Critérios de seleção da ordem de defasagens 57 Tabela 3 Probabilidades do teste de causalidade de Granger para Eficiência de Custos 60 Tabela 4 Resumo das Funções Impulsoresposta das diversas variáveis ao escore de eficiência de todas as amostras 69 Tabela 5 Estimativas pontuais das decomposições de variância do erro de previsão do Itaú 70 Tabela 6 Estimativas pontuais das decomposições de variância do erro de previsão do Banco do Brasil 70 Tabela 7 Estimativas pontuais das decomposições de variância do erro de previsão do Bradesco 71 Tabela 8 Estimativas pontuais das decomposições de variância do erro de previsão da Caixa 72 Tabela 9 Estimativas pontuais das decomposições de variância do erro de previsão do Santander 73 Tabela 10 Estimativas pontuais das decomposições de variância do erro de previsão da média ponderada dos privados 73 Tabela 11 Estimativas pontuais das decomposições de variância do erro de previsão da média ponderada dos públicos 74 Tabela 12 Estimativas pontuais das decomposições de variância do erro de previsão da média ponderada dos bancos 75 Tabela 13 Efeito das Funções Impulsoresposta dos choques do INEC à eficiência de custos de todas as amostras 78 Tabela 14 Efeito das Funções Impulsoresposta dos choques do Risco Brasil à eficiência de custos de todas as amostras 79 Tabela 15 Efeito das Funções Impulsoresposta dos choques do PIB à eficiência de custos de todas as amostras 81 Tabela 16 Efeito das Funções Impulsoresposta dos choques do IGPM à eficiência de custos de todas as amostras 82 Tabela 17 Efeito das Funções Impulsoresposta dos choques de Basileia à eficiência de custos de todas as amostras 83 Tabela 18 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável INEC 86 Tabela 19 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável Risco Brasil 86 Tabela 20 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável PIB 86 Tabela 21 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável IGPM 87 Tabela 22 Estatística descritiva dos escores de eficiência de receita e lucro 88 Tabela 23 Probabilidades do teste de causalidade de Granger para Eficiência de Receita Lucro e Custos 89 Tabela 24 Matriz de correlação das variáveis dos modelos VAR 103 Tabela 25 Teste ADF para principais variáveis do estudo 104 Tabela 26 Teste LM de cada amostra 105 Tabela 27 Probabilidades do teste de causalidade de Granger para as demais variáveis 111 Tabela 28 Matriz de correlação dos escores de eficiência de custo lucro e receita 112 SUMÁRIO 1 INTRODUÇÃO 17 2 REVISÃO DA LITERATURA 21 21 Eficiência bancária definição 21 22 Técnicas de estimação de eficiência 26 23 Eficiência bancária e suas relações com outras variáveis 30 24 As hipóteses de pesquisa 34 3 METODOLOGIA 40 31 Análise de fronteira estocástica 40 32 Modelo Autorregressivo Vetorial 42 321 Especificação e estimação do modelo VAR 43 322 Checagem do Modelo 48 323 Análise Estrutural 48 3231 Causalidade de Granger 49 3232 Função ImpulsoResposta 50 3233 Análise da Decomposição da Variância dos Erros de Previsão 52 324 Previsão 52 33 Dados e amostra 54 4 RESULTADOS E DISCUSSÃO 57 41 Resultados Modelo VAR 57 42 Análises Estruturais 58 421 Análises de causalidade de Granger 58 422 Análises das Funções ImpulsoResposta 60 4221 Análise individual dos cinco grandes bancos 61 4222 Análise das médias ponderadas dos públicos privados e geral 65 423 Análises das Decomposições das Variâncias dos Erros de Previsão 69 43 Previsão 75 44 Discussão dos resultados à luz das hipóteses 77 45 Teste de Robustez 84 451 Utilização de variáveis alternativas 84 452 Variação na ordem das variáveis 85 453 Eficiência de receita e lucro 87 16 5 CONCLUSÃO 91 6 REFERÊNCIAS 93 APÊNDICE 101 17 1 INTRODUÇÃO O tema eficiência bancária cresceu muito nos últimos anos fato evidenciado pelo aumento exponencial de publicações sobre esse conteúdo Esse fator se deu por diversas razões dentre as quais destacamos os desafios de medir e analisar uma das indústrias de serviço com a estrutura de receitas mais complexa BERGER HUMPHREY 1992 a divergência metodológica acerca do melhor modo de mensuração da eficiência AIELLO BONANNO 2015 os interesses pelas consequências e impactos das crises financeiras AHMAD et al 2019 a desregulamentação e mudanças tecnológicas FIORDELISI MARQUES IBANEZ MOLYNEUX 2011 e até mesmo o alto grau de dispersão dos estudos sobre o tema evidenciando que apesar de tradicional a eficiência bancária ainda não é uma área de conhecimento bem estabelecida ABREU KIMURA SOBREIRO 2019 No Brasil um dos maiores e mais complexos sistemas bancários do mundo essa evolução não foi diferente e seu efeito é oriundo das diversas transformações como aumento da concorrência onda de concentração com fusões e aquisições surgimento de novos produtos financeiros e a revolução na tecnologia da informação NAKANE WEINTRAUB 2005 Tivemos também a recente entrada de bancos estrangeiros logo após a estabilização financeira trazida pelo Plano Real que contribuiu para esse ambiente mais competitiva e dinâmico GOMES OLIVEIRA MATIAS 2017 e uma nova realidade do cenário macroeconômico com um ambiente de inflação mais baixa que forçou os bancos a se adaptarem NAKANE WEINTRAUB 2005 Como consequência dessa imensa quantidade de estudos várias foram as análises das relações entre a eficiência bancária e outras variáveis Artigos focaram em variáveis socioculturais como a cultura do país ver por exemplo Barros et al 2018 ou macroeconômicas como inflação Athanasoglou Brissimis e Delis 2008 Primo et al 2013 AlGasaymeh 2016 Tan e Anchor 2017 e produção nacional Sufian 2009 Chen Higgins e Mason 2005 Chortareas Girardone e Ventouri 2012 Primo et al 2013 AlGasaymeh 2016 Tan e Anchor 2017 Chen et al 2018 Kallel Hamad e Triki 2019 Houveram 18 também estudos relacionandoa com variáveis do setorindústria como concentração de mercado Berger e Hannan 1998 Athanasoglou Brissimis e Delis 2008 concorrência Casu e Girardone 2009 e liberalização do sistema financeiro Aiello e Bonanno 2018 ou até mesmo variáveis bancoespecíficas como empréstimos vencidos Berger e DeYoung 1997 o tamanho e ramificação da instituição Elyasiani e Mehdian 1990 Athanasoglou Brissimis e Delis 2008 Henriques et al 2018 os riscos bancários Athanasoglou Brissimis e Delis 2008 Fiordelisi MarquesIbanez e Molyneux 2011 Tan e Anchor 2017 o nível de capital Fiordelisi MarquesIbanez e Molyneux 2011 Kallel Hamad e Triki 2019 o custo do crédito Shamshur e Weill 2019 e até a qualidade dos serviços Kallel Hamad e Triki 2019 O propósito deste estudo é estimar o impacto parcial de choques em variáveis macro e microeconômicas sobre flutuações futuras no grau de eficiência bancária nas principais instituições financeiras do Brasil Para isso são utilizadas variáveis macroeconômicas como proxy para expectativas futuras medida de riscopaís atividade econômica inflação e uma variável bancoespecífica relacionada a solvência além da própria variável de eficiência A estimação ocorre por meio de modelos de Vetores Autorregressivos VAR com posterior análise das funções de impulso resposta IRFs causalidade de Granger e decomposição da variância dos erros de previsão FEVD Embora haja uma vasta literatura sobre o tema nosso artigo traz algumas contribuições para a literatura sobre a relação entre a eficiência bancária e variáveis macroeconômicas e específicas Em primeiro lugar nossa amostra considera dados bancários trimestrais recentes do 1º trimestre de 2009 até o 4º trimestre de 2019 das cinco maiores instituições financeiras brasileiras e outras três séries de dados oriundas de uma média ponderada desses cinco bancos Em segundo lugar utilizamos a técnica paramétrica mais difundida para estimação da eficiência a análise de fronteira estocástica pois ela nos permitiria analisar tanto a eficiência de custos quanto outras eficiências como lucro e receita BERGER MESTER 1997 que poderiam evidenciar outras qualidades e características gerenciais FIORDELISI MARQUESIBANEZ MOLYNEUX 2011 Logo utilizamos a eficiência de custos como referência principal e as demais mensurações como teste de robustez Em terceiro lugar optamos por 19 um modelo de séries temporais o Modelo Autorregressivo Vetorial Vector Autoregressive VAR para analisar as interrelações entre a eficiência e as demais variáveis Essa escolha nos permitiu realizar uma análise estrutural para cada um dos bancos ou seja o kit de ferramentas padrão na análise VAR STOCK WATSON 2001 análise de causalidade de Granger as funções impulso resposta e a decomposição da variância dos erros de previsão e testar a acurácia preditiva dos modelos através do teste de DieboldMariano Surpreendentemente não encontramos outros estudos que adotaram essa metodologia com esse fim para esse tema Os resultados obtidos mostram que a expectativa futura a atividade econômica e a inflação Grangercausam a eficiência de custos para a maioria das amostras o que não se mostrou verdadeiro para o riscopaís e o indicador de solvência das instituições Além disso a eficiência de custos Grangercausa a expectativa futura a atividade econômica e o indicador de solvência na maioria das amostras As demais análises das funções impulsoreposta e das decomposições de variância do erro de previsão nos mostraram que os resultados de cada instituição financeira é próprio e que não há uma convergência entra as amostras analisadas para o período destacado evidenciando que as habilidades e características próprias de gerenciamento são fato determinante para o comportamento da eficiência de custos Como medida de impacto ou seja quantificação as funções impulsoreposta também se mostraram heterogêneas entre os bancos brasileiros Após oito trimestres um choque positivo de um desvio padrão nas expectativas futuras ceteris paribus afetou a eficiência de custos entre 200 Santander e 47 pontosbase ou bps Bradesco Para o riscopaís os efeitos variaram de 27 bps Banco do Brasil até 156 bps Santander para a atividade econômica entre 96 bps Bradesco e 155 bps Santander para a inflação entre 37 bps Itaú e 56 bps Santander e entre 150 bps Santander e 90 bps Caixa para o indicador de solvência Este estudo está organizado da seguinte maneira A seção 2 fornece uma breve revisão da literatura para a eficiência bancária e suas técnicas de apuração além das suas relações com outras variáveis e apresenta as principais hipóteses da pesquisa Na seção 3 são apresentadas as metodologias para definição dos 20 escores de eficiência e para estimação do modelo autorregressivo vetorial e uma breve explicação das variáveis e dados Os resultados empíricos as análises e discussões acerca dos modelos e os diversos testes de robustez são detalhados na seção 4 A seção 5 conclui o estudo com as suas contribuições para a literatura suas limitações e sugestões de estudos futuros 21 2 REVISÃO DA LITERATURA Nesta seção apresentaremos um resumo da literatura sobre a eficiência bancária A primeira subseção conceitua o assunto citando desde os primeiros estudos sobre eficiência na década de 50 até os estudos mais atuais já específicos do universo bancário A segunda subseção mergulha nas diferentes técnicas de estimação de eficiência e seus respectivos usos na literatura A terceira subseção apresenta alguns destaques da literatura nos estudos que analisaram as relações entre variáveis diversas e a eficiência bancária A quarta e última subseção descreve as hipóteses de pesquisa desse estudo e algumas das suas respectivas referências 21 Eficiência bancária definição Farrel 1957 foi um dos primeiros autores a estudar a eficiência e a produtividade Seus estudos deram origem tanto a definição de eficiência e produtividade quanto às suas estimações Kumbhakar e Lovell 2010 destacaram esse pioneirismo do autor na medição da eficiência técnica e econômica que apresentou o método de medição da eficiência técnica através dos insumos a eficiência alocativa desses insumos e a eficiência de custos Uma premissa fundamental do estudo de Farrel 1957 foi permitir alocações imperfeitas de inputs e outputs definindoas como a distância de uma empresa a uma função de fronteira de produção referência Ou seja se a produção de uma empresa estiver na fronteira essa é perfeitamente eficiente mas se estiver abaixo será ineficiente Adicionalmente a razão entre a produção realizada ou auferida e a produção potencial ou fronteira de produção é o nível de eficiência de cada empresa ou unidade tomadora de decisão decision making units DMU Ele ainda propõe uma decomposição da eficiência em eficiência técnica ET e eficiência alocativa EA sendo a sua soma a representação da medida de eficiência total A eficiência técnica é a capacidade de uma empresa ou DMU de minimizar os inputs e maximizar os outputs já a eficiência alocativa 22 pressupõe a capacidade de usar os insumos nas melhores proporções possíveis levando em conta questões como preço e tecnologia de produção Mais tarde Havrylchyk 2006 utilizando os estudos de Farrel 1957 como referência o complementou descrevendo a eficiência técnica como a capacidade de produzir o máximo de outputs em um determinado nível de inputs ou a capacidade de usar o nível mínimo de inputs em um determinado nível de outputs v 30 p 1982 tradução nossa e a eficiência alocativa como a capacidade de selecionar a combinação ideal de insumos à luz de determinados preços a fim de produzir um determinado nível de produtos v 30 p 1982 tradução nossa Leibenstein 1966 outro precursor do tema o primeiro a utilizar a expressão X ineficiência para descrever a ineficiência de custo empregou o termo como a ineficiência oriunda de razões organizacionais e não por razões tecnológicas como por exemplo quando os interesses dos líderes e dos liderados não estão perfeitamente alinhados com os dos acionistas proprietários Elyasiani e Mehdian 1990 diferindo da literatura até então que na sua maioria adotava uma estrutura de função de custo mas seguindo na linha dos estudos de Farrel 1957 adotaram uma abordagem de fronteira de produção que segundo os autores era mais apropriado para a mensuração de eficiência enquanto as funções de custo eram mais úteis para a mensuração de economias de escala e escopo Embora o tema seja muito relevante sua análise é muito complexa e desafiadora para o setor bancário Berger e Humphrey 1992 afirmaram que a indústria bancária é uma das indústrias de serviço mais difíceis de medir a eficiência por conta do desafio de escolher quais funções constituem os produtos importantes dado que muitas receitas são implícitas e combinadas Apesar dessas dificuldades sua análise é essencial Allen e Rai 1996 afirmaram que à medida que políticas regulatórias e as realidades do mercado aumentam a concorrência a sobrevivência dos economicamente mais aptos deve prevalecer Com a harmonização dos ambientes regulatórios os bancos mais eficientes terão uma vantagem competitiva ALLEN RAI 1996 v 20 p 656 tradução nossa 23 Carvallo e Kasman 2005 afirmaram em seu estudo com 481 bancos de 16 países da América Latina que a pressão competitiva do setor é particularmente importante para os países emergentes pois neles as instituições financeiras são os principais intermediários para viabilizar economias poupança e investimentos Fiorentino Karmann e Koetter 2006 contextualizaram a importância do tema em seu estudo sobre o mercado bancário alemão Eles pontuaram que em todo o mundo o mercado financeiro e consequentemente as instituições financeiras vivenciaram mudanças substancias tanto no ambiente externo quanto no interno Questões como o aumento da concorrência que foi fruto não só de outros bancos mas de novos entrantes não bancários como a desregulamentação ou assimetria regulatória fomentada pelos órgãos responsáveis custos de informação reduzidos e o progresso tecnológico Até por esses fatores eles ponderam seja pelos reguladores ou pelos profissionais do ramo a dependência cada vez maior da teoria econômica para mensuração da eficiência e sua análise Eles também afirmaram que os índices de eficiência conforme descrição acima de Farrel 1957 assumem um valor entre zero e um sendo que se o índice for igual a um a empresa ou DMU é totalmente eficiente Eles mencionam um exemplo de um escore de eficiência em uma fronteira de custos de 90 que significaria que a DMU conseguiria reduzir os custos em 10 sem alterar os outputs Além das necessidades de mercado a análise da eficiência bancária é relevante tanto pelo aspecto microeconômico quanto pelo aspecto macroeconômico No aspecto microeconômico a busca pela eficiência do sistema bancário é uma realidade para aumento da concorrência e para a melhoria da regulação e supervisão No aspecto macroeconômico a eficiência bancária auxilia a estabilidade de todo o sistema financeiro e impacta o custo de intermediação financeira Além disso conforme afirmou Andries e Cocris 2010 uma maior eficiência dos bancos poderá acarretar numa melhor alocação dos recursos financeiros e consequentemente em um aumento dos investimentos que favoreceriam o crescimento econômico Matthews 2010 complementa em seu estudo de eficiência bancária em mercados emergentes que vários estudos sobre o tema utilizaram 24 equivocadamente a expressão de Leibenstein Xineficiência como eficiência técnica apesar do conceito semelhante Ele afirma que o a eficiência técnica é oriunda da teoria da firma neoclássica e compreende a maximização do lucro Um banco poderia ser tecnicamente ineficiente por razões técnicas como falta de capacitação para seus recursos humanos ou a defasagem tecnológica Como o surgimento de novas tecnologias não é constante e nem todas são perenes e penetram de fato no mercado algum banco poderia não acompanhar um movimento necessário e com isso ficar para trás na comparação com seus concorrentes A Xineficiência não é fruto da variabilidade de habilidades ou da variação temporal da difusão de novas tecnologias mas pela forma como cada banco usa explora e se organiza com tais habilidades e tecnologias O estudo de Aiello e Bonanno 2015 que analisou mais de 120 trabalhos publicados entre os anos de 2000 e 2014 destacou que a crescente atratividade dos estudos de eficiência passa pela divergência metodológica mencionando o questionamento sobre qual seria a melhor fronteira bancária e como estimála adequadamente que segundo eles ainda não obteve uma resposta conclusiva Eles afirmam que isso acarretou numa proliferação de pesquisas empíricas e uma heterogeneidade relevante nos resultados v 30 p 113 tradução nossa Gomes Oliveira e Matias 2017 no seu estudo que comparou a eficiência dos bancos domésticos e estrangeiros no Brasil entendem que para o mercado financeiro brasileiro questões relacionadas à eficiência ganharam maior relevância após a entrada de bancos estrangeiros principalmente a partir da década de 1990 logo após o Plano Real que iniciou a estabilização financeira fazendo com que o setor fosse cada vez mais impactado pelas forças de mercado onde o ambiente é mais competitivo e dinâmico O estudo que difere bastante dos demais é o de Chen et al 2018 que apresenta um caminho contrário tentando propor uma combinação de fatores viáveis e seus ajustes necessários para alcançar a eficiência total que eles comprovam ser atingível Eles utilizaram uma amostra com os principais bancos de Taiwan durante os anos de 1999 e 2011 e concluíram que é possível que esses bancos economizem 103 dos custos totais no curto prazo e 88 no longo prazo 25 Um fato curioso destacado em outro recente estudo de Huljak Martin e Moccero 2019 sobre eficiência de custo e produtividade dos bancos na zona do euro é a crítica aos indicadores de eficiência Average Cost AC razão entre os custos totais e o total de ativos e o Cost to Income Ratio CIR razão entre os custos administrativos e a receita operacional que são os indicadores mais comumente utilizados analistas de mercado para medir eficiência Apesar de fáceis de apurar eles alegam que não capturam a eficiência do setor Sendo que o AC é dependente do modelo de negócio da instituição financeiro e do seu tamanho Por exemplo bancos de atacado tendem a investir menos em estrutura física presencial em agências do que bancos de varejo que necessitam de capilaridade Eles também alegam que o AC depende de fatores específicos do país e as vezes até da região como o custo de mãodeobra e as exigências regulatórias Logo o AC seria um índice dos preços médios dos insumos para a produção de alguma unidade bancária ativos serviços entre outros e não um indicador de eficiência Já o CIR além de também serem impactados por fatores regionais e locais específicos reúne aspectos diferentes do desempenho além da eficiência como produtividade por exemplo Outra desvantagem do CIR é que ele é impactado pelo risco de crédito o que distorce a estimativa de eficiência Em contrapartida os autores afirmam que uma análise de fronteira estocástica traria vantagens pois controla o fato de os bancos produzirem diferentes produtos e consumirem ou pagarem por diferentes insumos permitindo então a comparação entre bancos de diferentes modelos de negócio especialização tamanho e estrutura Curiosamente apesar de vários estudos mencionarem a quantidade considerável de papers publicados sobre o tema eficiência Abreu Kimura e Sobreiro 2019 ao analisarem os 87 artigos publicados mais bem avaliados segundo a Academic Journal Guide of Association of Business Schools entre janeiro de 2011 e julho de 2017 encontraram alto grau de dispersão dos estudos e um número relativamente limitado de autores dedicados ao tema Segundo os autores esse seria um indício de que apesar da tradição do tema a eficiência bancária ainda não é uma área de conhecimento bem definida 26 22 Técnicas de estimação de eficiência Há muitas técnicas métodos para estimar eficiência e consequentemente muita literatura a respeito Ferrier e Lovell 1990 por exemplo elaboraram um artigo comparando duas técnicas para estimar a eficiência bancária uma técnica de estimativa econométrica de fronteira de custos e uma de série de programas lineares de fronteira de produção Elas apresentaram resultados diferentes em relação à eficiência de custos Eles atribuem as correlações entre as duas técnicas não serem estatisticamente significativas a quatro fatores 1 A abordagem de programação não é estocástica e portanto interpreta o ruído como ineficiência 2 A abordagem econométrica impõe estrutura paramétrica à tecnologia e à distribuição da ineficiência e assim mistura erro de especificação com ineficiência 3 A eficiência alocativa está disponível apenas como um valor médio sobre a amostra para a abordagem econométrica de modo que os valores reais da eficiência alocativa e portanto técnica sejam sempre superestimados ou subestimados 4 Para evitar uma categorização excessiva dos bancos e consequentemente um número excessivo de bancos ostensivamente eficientes dentro de cada categoria a abordagem de programação implementada exclui as variáveis categóricas do tipo de instituição Duas dessas categorias têm um impacto estatisticamente significativo no custo na abordagem econométrica esse impacto é atribuído à ineficiência na abordagem de programação FERRIER LOVELL 1990 p 243 tradução nossa Berger e Mester 1997 constataram que as técnicas mais comuns de estimativa de eficiência são como técnicas não paramétricas a análise envoltória de dados data envelopment analysis DEA e a livre disposição de envoltória free disposable hull analysis FDH e como técnicas paramétricas análise de fronteira estocástica stochastic frontier analysis SFA a abordagem de fronteira espessa thick frontier approach TFA e a abordagem da distribuição livre distributionfree approach DFA Eles afirmaram que os métodos não paramétricos geralmente ignoram os preços e portanto podem ser responsáveis apenas pela ineficiência técnica no uso de muitos insumos ou na produção de poucos produtos v 21 p 905 tradução nossa Eles 27 complementam que eles não explicam a ineficiência alocativa na resposta as escolhas de produtos e insumos ou na comparação de empresas que se especializam em diferentes produtos e insumos dado que isso não é possível sem o benefício do preço Logo as técnicas não paramétricas geralmente se concentram na otimização tecnológica e não na otimização econômica e não correspondem aos conceitos de eficiência de custo e lucro Outra desvantagem das técnicas não paramétricas é que elas geralmente não permitem erros aleatórios nos dados assumindo o erro de medição e a sorte como fatores que afetam os resultados embora algum progresso esteja sendo feito nesse sentido De fato eles separam as diferenças de eficiência do erro aleatório assumindo que o erro aleatório é zero BERGER MESTER 1997 v 21 p 905 tradução nossa Berger e Humphrey 1997 complementam ao analisaram 130 estudos de análise de eficiência bancária em 21 países relatando uma distribuição muito similar entre os métodos paramétricos e não paramétricos Fiorentino Karmann e Koetter 2006 também compararam as duas técnicas afirmando encontrarem uma consistência muito baixa entre ambas Eles afirmaram que os métodos não paramétricos são mais sensíveis à heterogeneidade da amostra e que a correlação entre a ordem de classificação de eficiência entre as duas técnicas é positiva mas baixa Segundo os autores esse modelo dos artigos de 1977 é indicado como ln𝑦𝑗 ln 𝑥𝑗𝛽 𝑣𝑗 𝑢𝑗 1 onde 𝑥𝑗 corresponde a um input para a empresa j 𝑣𝑗 corresponde a um erro aleatório adicionado ao termo de ineficiência não negativo o 𝑢𝑗 Esse erro aleatório e efeitos combinados de variáveis de entrada não especificadas na função de produção que explicam o erro de medição e outros eventuais fatores aleatórios que podem afetar o valor do output Como o limite superior é determinado pela variável estocástica exp𝑥𝑗𝛽 𝑣𝑗 o modelo é estocástico sendo o erro aleatório podendo ser negativo ou positivo logo os resultados dessa fronteira variam em relação à parte determinística do modelo nesse caso 28 exp𝑥𝑗𝛽 Para a utilização da técnica de análise de fronteira estocástica é necessário assumir uma forma funcional para estimar o modelo A complexidade do setor bancário conforme descrito anteriormente como um setor de produção múltipla inviabiliza a especificação de uma função de produção logo se faz necessária a especificação de uma fronteira de custos A fronteira de custos é apresentada pelos autores como a forma geral de log ln 𝐶𝑗 𝑓 ln 𝑦𝑟𝑗 ln 𝑐𝑖𝑗 𝜀𝑗 onde 𝐶𝑗 é o custo total da empresa j 𝑦𝑟𝑗 mede o résimo output da empresa j o 𝑐𝑖𝑗 é o preço do iésimo input da empresa j e o termo do erro 𝜀𝑗 é o erro aleatório 𝑣𝑗 e o termo de ineficiência 𝑢𝑗 descrito na equação 1 acima O termo de erro aleatório 𝑣𝑗 é classificado como independente e identicamente distribuída iid como 𝑣𝑗 𝑁 0 𝜎𝑣 2 e independente das variáveis explicativas O termo de ineficiência 𝑢𝑗 também é iid com 𝑢𝑗 𝑁 0 𝜎𝑢 2 independente do termo de erro e é extraído de uma distribuição não negativa truncada em zero Sobre o surgimento da análise de fronteira estocástica Kumbhakar e Lovell 2010 detalharam no seu capítulo introdutório as suas origens contando que surgiu em dois trabalhos publicados quase que simultaneamente por duas equipes distintas Meeusen e van den Broek em junho de 1977 e Aigner Lovell e Schmidt em julho de 1977 sendo esse último uma versão que reunia e mesclava um artigo do Aigner e outro do Lovell e Schmidt Apesar das publicações terem ocorrido em continentes distintos com somente um mês de diferença entre elas os documentos eram muito semelhantes sendo que estavam em preparação há três anos Logo após um terceiro artigo também muito conhecido foi lançado por Battese e Corra também em 1977 sendo que esses autores já tinham servido de referência para o artigo do Aigner Lovell e Schmidt Um estudo mais recente de Aiello e Bonanno 2015 realizou uma Análise de MetaRegressão MetaRegressionAnalysis MRA examinando 1661 escores de eficiência tendo encontrado como um dos seus principais resultados que os métodos paramétricos produzem níveis mais baixos de eficiência que os métodos não paramétricos Eles destacam também algumas diferenças relevantes por conta da qualidade dos estudos e o número de observações e variáveis Em estudo posterior dos mesmos autores 2018 que utilizou a 29 mesma base para análise e teve como objetivo mensurar o impacto da seleção do método e fatores específicos do país ou região na variabilidade do escore de eficiência destacou que não há consenso sobre a superioridade de um método de estimativa sobre outro v 32 n 1 p 195 tradução nossa O estudo de Silva et al 2017 que comparou as eficiências dos bancos locais chineses contrapõe a afirmação sobre heterogeneidade de Fiorentino Karmann e Koetter 2006 acima ao fornecer testes de robustez com resultados semelhantes através da remoção da heterogeneidade da amostra ao estimar níveis de eficiência no DEA e utilização de efeitos fixos para absorver as heterogeneidades dos bancos no SFA Eles atribuem a diferença entre as duas técnicas à natureza diferente de fronteiras eficientes que cada método estima Outra contribuição importante desse estudo foi que apesar dos resultados não fornecerem classificações de eficiência semelhantes entre os dois métodos na análise individual eles forneceram resultados semelhantes em relação às pontuações médias para todo o mercado ao longo dos anos analisados Caminhando na mesma linha o recente estudo de Ahmad et al 2019 afirmou que entre os 100 principais mais citados artigos sobre desempenho do setor bancário conforme Gráfico 1 abaixo 26 utilizaram índices financeiros exemplos retorno sobre o patrimônio líquido ou return on equity ROE retorno sobre os ativos ou return on assets ROA margem de juros líquida ou net interest margin NIM crédito não produtivo ou nonperforming loan NPL e os outros 74 utilizaram análise de fronteira na qual destacamos que 34 utilizaram métodos paramétricos predominantemente SFA e análise fatorial explanatória ou exploratory factor analysis EFA além de outros já mencionados acima e 40 não paramétricos sendo majoritariamente a DEA destaque também para a DEA em redes ou network DEA 30 Gráfico 1 Abordagem de estimativas nos 100 principais artigos Fonte adaptado de AHMAD ET AL 2019 p 16 23 Eficiência bancária e suas relações com outras variáveis Dentre os estudos que analisam comparam e associam a eficiência com outras variáveis destacamos o estudo de Elyasiani e Mehdian 1990 que investigou o impacto do tamanho e a ramificação quantidade de agências ou distribuição da rede Eles concluíram que o tamanho nesse caso medido pelo total de ativos ou receita total tem um efeito positivo na eficiência ou seja bancos maiores são mais eficientes Os autores vincularam a maior parte desse aumento na eficiência à eficiência de escala afirmando que isso ocorre pois os bancos conseguem se aproximar do ponto mínimo da curva de custo médio de longo prazo Outro ponto interessante e até controverso com outros estudos futuros é que eles afirmam que uma parte pequena dessa maior eficiência dos bancos grandes é associada a diferenças na eficiência técnica Não paramétrica 40 Paramétrica 34 Indicadores Financeiros 26 DEA 33 DEA em redes 3 Outros 4 SFA 31 EFA 1 DFA 2 TFA 1 ROA 8 ROE 7 Outros 8 NIM 2 NPL 1 31 Berger e DeYoung 1997 abordaram a relação entre os empréstimos problemáticos vencidos e a eficiência bancária empregando causalidade de Granger para testar hipóteses sobre a relação entre a qualidade do empréstimo e a eficiência e concluíram que os empréstimos problemáticos precedem ineficiência e que a eficiência precede reduções nos empréstimos problemáticos Além disso eles analisaram se os estudos devem ou não considerar os empréstimos problemáticos nas estimativas de eficiência mas não foram conclusivos apresentando resultados ambíguos Vale ressaltar que segundo o estudo de Ahmad et al 2019 esse artigo do Berger e DeYoung 1997 foi o mais citado enaltecendo sua relevância Outro estudo de Berger e Hannan 1998 tentou relacionar a concentração de mercado com a eficiência operacional testando a hipótese de que pressões mais brandas para minimizar os custos por conta da concorrência podem resultar em menor eficiência operacional para empresas em mercados concentrados Eles encontraram fortes evidências de que essa hipótese é verdadeira Fiorentino Karmann e Koetter 2006 concluíram que as medidas de eficiência tanto a técnica paramétrica quanto a não paramétrica apresentam baixa correlação com os indicadores de desempenho mais comumente utilizados como custobenefício e taxas de retorno O estudo de Athanasoglou Brissimis e Delis 2008 investigou os efeitos de determinantes específicos do próprio banco eficiência operacional risco financeiro e tamanho específicos da indústria a concentração da indústria e a participação de bancos privados no setor e macroeconômicos comportamento cíclico ou relação com o ciclo de negócios e inflação esperada na lucratividade dos bancos gregos entre 1985 e 2001 Eles encontraram resultados empíricos que sugerem que os determinantes do próprio banco com exceção do tamanho afetam significativamente a lucratividade já os da indústria não são relevantes pelo menos no sistema bancário grego que segundo os autores evoluiu muito ao longo do período analisado Por último o ciclo de negócio afeta significativamente o lucro dos bancos assim como a inflação esperada que apresentou relação positiva 32 Casu e Girardone 2009 testaram através da causalidade Granger a relação entre concorrência e a eficiência bancária para cinco países da União Europeia E diferentemente do que pressupõe a teoria do monopólio elas não encontraram evidências de que o poder de mercado dos bancos se traduz em uma diminuição na eficiência de custos porém elas não puderam concluir que a rejeição da hipótese de relaxamento na gestão oriunda de um monopólio suporte o paradigma da estrutura eficiente Elas testaram também a causalidade reversa que seria da eficiência à concorrência mas não encontraram resultados que sugerissem que o aumento da eficiência promoveria um aumento no poder de mercado Já Fiordelisi MarquesIbanez e Molyneux 2011 publicaram um estudo onde examinaram duas questões do setor bancário europeu a relação entre 1 eficiência bancária e risco bancário e 2 o impacto do capital na troca entre risco e eficiência Na primeira relação analisaram se baixa eficiência ou ineficiência poderia acarretar numa tentativa de aumento dos retornos e consequentemente prejudicar os padrões operacionais como gestão do crédito risco e na outra direção se um aumento dos riscos poderia ser um predecessor de uma redução na eficiência Na segunda relação analisaram por exemplo se os baixos índices de capitalização são fruto da eficiência ou da ineficiência Para tal utilizaram medidas prospectivas e retrospectivas de risco capital e três medidas de eficiência bancária receita custo e lucro Esse olhar para outras medidas de eficiência é fundamental pois refletem habilidades gerenciais diferentes ou seja as habilidades para maximizar as receitas e minimizar os custos v 35 p 1316 tradução nossa por exemplo Seus resultados mostraram que uma menor eficiência tanto para a medida de custo quanto receita Grangercausa maior risco bancário confirmando o que eles chamaram de hipótese de má gestão que aumentos no capital precedem ganhos de eficiência de custos que a eficiência de custos Grangercausa aumentos de capital e níveis mais elevados de capital tendem a ter um efeito positivo nos níveis de eficiência Outro estudo que também analisou a relação da eficiência bancária com risco foi o de Tan e Anchor 2017 que testou o impacto do risco em nível micro e da concorrência na eficiência do setor bancário chinês no período de 2003 a 2013 Eles consideraram formas abrangentes de comportamento de risco incluindo 33 risco de crédito risco de liquidez e risco de insolvência Eles destacaram o risco de liquidez que afeta significativamente e negativamente a eficiência e que uma maior concorrência precederia um declínio na eficiência Eles mencionam também algumas relações com variáveis macroeconômicas onde os resultados sugerem que um ambiente de alta inflação ou de boom econômico taxa de crescimento do PIB a eficiência é maior No estudo de Aiello e Bonanno 2018 que teve como um dos seus objetivos mensurar o impacto de fatores específicos do país na variabilidade do escore de eficiência e para isso analisou 1661 escores de eficiência destacou que a liberalização do sistema financeiro também desempenha um papel positivo no escore mencionando que países e regiões com maior liberalização no setor como os Estados Unidos União Europeia Europa Oriental e Oceania possuíam pontuações mais altas Por outro lado países de regiões com regulamentação mais rígida como na América Latina e na África números mais baixos Outra análise muito interessante foi feita por Barros et al 2018 onde realizaram uma avaliação da eficiência bancária em um cenário cultural comum selecionando países em desenvolvimento de mesmo idioma e passado colonial Angola Brasil e Moçambique Para os autores dessa forma seria possível afirmar que os resultados estão puramente relacionados ao tamanho da escala e às questões políticas e econômicas de cada país Como os resultados encontrados revelaram que o Brasil possui o mais alto nível de eficiência e retorno de escala puderam afirmar que os países com mais alto grau de desenvolvimento são mais eficientes Assim como diversos outros estudos que relacionaram a eficiência com o tamanho da organização Henriques et al 2018 que avaliou a eficiência bancária dos 37 maiores bancos brasileiros entre 2012 e 2016 concluíram que os cinco maiores bancos brasileiros demonstraram uma ineficiência diretamente relacionada ao nível de escala Eles complementam que uma das possíveis causas para isso seria a necessidade que eles apresentam em atenderem um grande número de clientes em todo o país acarretando uma estrutura de ativos e despesas muito elevada em relação aos bancos menores 34 Kallel Hamad e Triki 2019 também contribuem com a análise das relações da eficiência bancária com outras variáveis Em sua comparação entre as eficiências dos bancos tunisianos e marroquinos entre 2005 e 2014 eles puderam concluir como já era a expectativa deles que bancos mais capitalizados têm um nível de custos mais elevado que segundo eles é necessário para cobrir os riscos assumidos Já ao contrário do que eles esperavam o que eles chamaram de a qualidade do serviço bancário controlado por eles pelo índice de outras receitas operacionais sobre ativos totais apresentou um efeito negativo no custo total ou seja os bancos com maior receita de serviços tendem a ser mais eficientes no gerenciamento de custos Eles interpretaram esse resultado com fruto do alto nível de inadimplência nos sistemas bancários analisados Outra interessante contribuição foi do estudo de Shamshur e Weill 2019 que investigou o impacto da eficiência de custo dos bancos no custo do crédito para as empresas testando a hipótese de que um banco mais eficiente poderia fornecer empréstimos mais baratos a seus clientes Eles encontraram evidências de que essa hipótese é verdadeira porém com algumas ressalvas A eficiência reduz o custo do crédito para as pequenas e médias empresas mas não exerceu influência significativa para as grandes ou microempresas 24 As hipóteses de pesquisa Antes de introduzirmos a seção sobre a metodologia definimos as principais hipóteses de pesquisa Para tal consideramos nosso propósito de identificar variáveis através de suas séries temporais que contenham informações sobre o desenvolvimento futuro da eficiência bancária das principais instituições financeiras do Brasil Formulamos portanto cinco hipóteses de pesquisa para os testes empíricos tendo por base a razoabilidade econômica e referências discutidas ao longo da revisão de literatura O Quadro 1 apresenta os estudos sintetizados desenvolvidos em diferentes ambientes econômicos 35 Quadro 1 Síntese dos estudos considerados para a identificação das hipóteses Fonte elaborado pelo próprio autor H1 Expectativas futuras Eficiência bancária A primeira hipótese H1 levantada é a de que indicadores prospectivos relacionados à previsão da atividade econômica e às expectativas futuras possuem uma relação positiva com a eficiência de custos das maiores instituições financeiras do Brasil Em Beccalli Casu e Girardone 2006 essa associação foi analisada através dos preços das ações bancárias e os resultados H Tema Descrição Referências Objeto amostra do estudo Beccalli Casu e Girardone 2006 concordam França Alemanha Itália Espanha e Reino Unido Fiordelisi MarquesIbanez e Molyneux 2011 concordam União Euroéia Chen Higgins e Mason 2005 concordam Estados Unidos Chen Higgins e Mason 2005 concordam Estados Unidos Tamadonnejad et al 2013 concordam Leste da Ásia Borges 2012 concorda União Euroéia AlGasaymeh 2016 concorda Conselho de Cooperação do Golfo Chen et al 2018 não encontrou resultados significativos Taiwan Chortareas Girardone e Ventouri 2012 discordam União Euroéia Sufian 2009 discorda Malásia Kallel Hamad e Triki 2019 discordam Tunísia e Marrocos Tan e Anchor 2017 concordam China AlGasaymeh 2016 concorda Conselho de Cooperação do Golfo Primo et al 2013 concordam Brasil Chen Higgins e Mason 2005 concordam Estados Unidos Tan e Anchor 2017 discordam China Primo et al 2013 concordam Brasil AlGasaymeh 2016 concordam Conselho de Cooperação do Golfo Chortareas Girardone e Ventouri 2012 concordam União Euroéia Sufian 2009 concorda Malásia Zhang et al 2013 concordam Brasil Rússia Índia e China BRIC Fiordelisi MarquesIbanez e Molyneux 2011 discordam União Euroéia Athanasoglou Brissimis e Delis 2008 discordam Grécia Tecles e Tabak 2010 discordam Brasil Mendonça 2018 discordam Brasil H3 Atividade econômica A eficiência de custos das maiores instituições financeiras do país é positivamente relacionada com a atividade econômica H4 Inflação A eficiência de custos das maiores instituições financeiras do país é negativamente relacionada com a inflação H2 Medida de riscopaís A eficiência de custos das maiores instituições financeiras do país é negativamente relacionada com a medida de riscopaís H5 Solvência A eficiência de custos das maiores instituições financeiras do país é negativamente relacionada com a solvência H1 Expectativas futuras A eficiência de custos das maiores instituições financeiras do país é positivamente relacionada com a expectativa futura de mercado 36 sugeriram que a relação é positiva Em Fiordelisi MarquesIbanez e Molyneux 2011 com o objetivo de analisar um indicador prospectivo de risco bancário que conseguisse capturar a probabilidade de inadimplência dos bancos ou seja que avalie diversos riscos bancários não só o risco de crédito na visão do mercado financeiro recorreram ao Expected Default Frequency EDF apurado pela Moodys KMV um ano à frente Os resultados encontrados mostraram que melhorias na eficiência de custo bancário Grangercausam uma menor probabilidade de inadimplência Também corroborando com a nossa hipótese Chen Higgins e Mason 2005 encontraram uma associação positiva entre um indicador de previsão da atividade econômica futura1 está associado a expectativa de aumento do crescimento econômico e a eficiência de custos Nesse mesmo estudo também como um indicador preditivo do desempenho econômico analisaram a relação entre a eficiência de custos e o prêmio dos títulos federais fed funds premium FFP apurado como a diferença entre a taxa de fundos federais e a taxa de três meses de letras do Tesouro Eles encontraram que um spread alto que indica baixo crescimento econômico está negativamente associada à eficiência de custos H2 Risco país Eficiência bancária A segunda hipótese H2 levantada é a de que a eficiência de custos das maiores instituições financeiras do país é negativamente relacionada com a medida de riscopaís Corroborando com essa afirmação Tamadonnejad et al 2013 apresentaram resultados que mostraram que a relação entre a eficiência bancária e o riscopaís são negativamente relacionadas Os autores argumentaram que elevações no riscopaís acarretaram em falta de confiança da população e corridas repentinas aos bancos que consequentemente geraram um aumento nos empréstimos mal executados e o risco de inadimplência Na mesma linha Borges 20122 afirmou que os bancos oriundos de países com classificações piores ratings que expressam uma opinião desfavorável em relação ao país e consequentemente não atrairiam investidores provavelmente sofrerão reduções nos escores de eficiência uma vez que o coeficiente é 1 Os autores utilizaram um indicador de Ciclo de Negócio apurado pelo The Conference Board e disponível no seu site 2 Nesse estudo os ratings não foram considerados estatisticamente significativos a um nível de significância de 5 ou 10 somente num nível de 15 37 negativo Também concordando com a afirmação AlGasaymeh 2016 encontrou evidências de que as variáveis de riscopaís desempenham um papel importante na influência da eficiência dos bancos nos países analisados Segundo o autor os riscos mais baixos criam um ambiente de trabalho que favorece a eficiência pois os bancos têm um desempenho melhor nesse cenário estável e equilibrado H3 Atividade econômica Eficiência bancária A terceira hipótese H3 levantada é a de que a eficiência de custos das maiores instituições financeiras do país é positivamente relacionada com a atividade econômica Concordando com essa afirmação Tan e Anchor 2017 encontraram resultados que mostram que a taxa de crescimento do PIB está significativa e positivamente relacionada à eficiência dos bancos analisados indicando que são mais eficientes durante os períodos de boom econômico Eles interpretam isso como resultado de uma demanda maior por serviços bancários uma melhora da qualidade dos tomadores de empréstimos que acarreta uma redução no custo do monitoramento de riscos e uma redução no custo de economias de escala e escopo que precedem melhorias na eficiência Na mesma linha AlGasaymeh 2016 encontrou resultados de que quanto maior o PIB per capita menor o custo dos bancos e consequentemente maior a eficiência Outro estudo que também encontrou evidências que corroboram com essa hipótese foi o de Primo et al 2013 que evidenciaram a relevância estatística do PIB dado que nos momentos de maior desenvolvimento econômico a rentabilidade dos bancos aumentou Eles explicam o fato pelo aumento na demanda de crédito a redução do nível de inadimplência e um aumento nas receitas de tarifas oriundo da inclusão financeira Da mesma forma Chen Higgins e Mason 2005 apresentaram resultados de que a variação percentual do PIB está positivamente associada à eficiência de custos Já Chen et al 2018 não encontraram um impacto significativo do PIB sobre a eficiência Do lado contrário à nossa hipótese Chortareas Girardone e Ventouri 2012 encontraram resultados de que o PIB per capita é significativo na eficiência bancária mas de forma negativa o que indicaria que bancos em expansão poderiam ser menos eficientes no controle de custos Da mesma forma Kallel Hamad e Triki 2019 verificaram que o coeficiente associado ao crescimento do 38 PIB é significativamente positivo com relação ao aumento de custos ou seja redução da eficiência Também contrário à nossa hipótese Sufian 2009 encontrou que as taxas de crescimento da renda nacional medido pelo PIB exibiram uma relação negativa e estatisticamente significativa com a eficiência bancária nos bancos analisados que segundo ele seriam fruto da sua relativa volatilidade no período analisado H4 Inflação Eficiência bancária A quarta hipótese H4 levantada é a de que a eficiência de custos das maiores instituições financeiras do país é negativamente relacionada com a inflação Tan e Anchor 2017 apresentaram resultados que contrariam essa afirmação evidenciando que um aumento da inflação geraria um aumento na eficiência Para eles no cenário analisado como medida de combate à inflação o governo e as autoridades regulatórias adotam medidas como aumento da taxa de juros eou aumento dos compulsórios que reduzem os volumes de empréstimos concedidos e por isso reduzem o custo de monitoramento e gerenciamento de riscos que aprimora a eficiência Em outro estudo mais alinhado com a concordância da hipótese Primo et al 2013 utilizaram o retorno sobre o patrimônio líquido return on equity ROE e o retorno sobre os ativos return on assets ROA para a eficiência em dois modelos distintos para cada variável sendo que encontrou resultados mistos entre eles e a inflação Todos foram significativos mas em três com sinal negativo e em um com sinal positivo Os autores afirmaram que a predominância da relação negativa indicaria que os efeitos relacionados à perda de valor sobre os ativos seriam superiores aos ganhos da redução de valor real dos passivos Já de forma menos ambígua Al Gasaymeh 2016 encontrou uma relação significativa e positiva que segundo ele indicaria que os bancos em países com inflação mais elevada teriam custos potenciais mais elevados H5 Solvência Eficiência bancária A quinta hipótese H5 levantada é a de que a eficiência de custos das maiores instituições financeiras do país é negativamente relacionada com a solvência Em linha com essa afirmação Chortareas Girardone e Ventouri 2012 concluíram que as exigências de capital que apresentaram um coeficiente 39 negativo e estatisticamente significativo em todos os modelos testados aumentam a eficiência do banco Do mesmo modo Sufian 2009 pontuou que sua variável relacionada ao capital do banco Patrimônio Líquido sobre o Total de Ativos exibiu uma relação negativa e estatisticamente significativa com a eficiência bancária Segundo o autor esses resultados sugeririam que os bancos mais eficientes operariam mais alavancados ou seja com menos patrimônio e que os bancos menos eficientes poderiam estar envolvidos em operações com maior risco que exigiriam maior colchão de capital Também alinhado com essa hipótese Zhang et al 2013 encontrou evidências de que um excesso de capital também medido como o Patrimônio Líquido sobre o Total de Ativos poderia induzir uma ineficiência restringindo o crescimento dos ativos mais rentáveis Já na contramão da nossa afirmação Fiordelisi MarquesIbanez e Molyneux 2011 encontraram que a razão de capital definida por eles da mesma forma que os autores anteriores Grangercausa positivamente a eficiência de custo estatisticamente significativa Ou seja um aumento na soma dos coeficientes do índice de capital defasado precede temporalmente os aumentos de eficiência de custos p 1319 tradução nossa Também contrariando nossa hipótese Athanasoglou Brissimis e Delis 2008 constataram que a variável que representa o capital também utilizaram o mesmo indicador é estatisticamente significativo e positivo em relação à eficiência Segundo os autores isso significaria que um banco com uma estrutura de capital sólida teria condições de perseguir e conquistar negócios mais eficientes e até mesmo mais tempo e flexibilidade para enfrentar desafios com perdas inesperadas Outro estudo que contraria essa hipótese é o de Tecles e Tabak 2010 que também encontrou um sinal negativo no coeficiente de patrimônio dos bancos na sua relação com a eficiência bancária Eles afirmaram que isso corroboraria com a teoria do risco moral dado que quando uma proporção maior de capital acionário está em risco a alta liderança da instituição financeira tem mais incentivos para monitorar a eficiência Na mesma linha Mendonça 2018 também encontrou significância estatística na variável de capital e coeficiente positivo que indicaria que quanto maior for a capitalização maior será a rentabilidade Ele alegou que isso é decorrente da sinalização ao mercado de uma menor probabilidade de falência que lhe permitiria reduzir seus custos de empréstimo e captação 40 3 METODOLOGIA 31 Análise de fronteira estocástica Para esse estudo assim como Williams 2004 Fiorentino Karmann e Koetter 2006 Fiordelisi MarquesIbanez e Molyneux 2011 e tantos outros a técnica paramétrica selecionada foi a análise de fronteira estocástica Além de bastante difundida essa técnica conforme afirmou Berger e Mester 1997 corresponde bem tanto aos conceitos de eficiência de custo quanto de lucro Conforme Fiordelisi MarquesIbanez e Molyneux 2011 ou Williams 2004 a eficiência de custos foi mensurada através da análise de fronteira estocástica Para esse estudo foi utilizado o modelo de fronteira estocástica de Battese e Coelli 1995 ln 𝑇𝐶𝑖𝑡 𝑥𝑖𝑡 𝛽 𝑉𝑖𝑡 𝑈𝑖𝑡 2 onde 𝑡 representa a dimensão de tempo 𝑇𝐶𝑖 é o logaritmo do custo de produção do iésimo banco 𝑥𝑖 é um vetor 1 𝑥 𝑘 de inputs e outputs de preços padronizados do iésimo banco 𝛽 é um 𝑘 𝑥 1 vetor de parâmetros desconhecidos 𝑉𝑖 é uma variável aleatória assumida iid distribuída como uma 𝑁0 𝜎𝑣 2 e independente de 𝑈𝑖 𝑈𝑖 é uma variável aleatória não negativa que representa a ineficiência de custo na produção assumida como iid de modo que 𝑈𝑖 é obtida por truncamento em zero da distribuição 𝑁𝜎 𝜎𝑈 2 sendo 𝜎 um parâmetro a ser estimado Para esse estudo foi utilizada a seguinte função de custo translog para estimar uma fronteira de referência comum ln 𝑇𝐶 𝛼0 𝛼𝑖 ln 𝑦𝑖 3 𝑖1 𝛽𝑖 ln 𝑤𝑖 3 𝑖1 𝜏1 ln 𝐸 𝑡1𝑇 1 2 𝛿𝑖𝑗 ln 𝑦𝑖 ln 𝑦𝑗 3 𝑗1 3 𝑖1 𝛾𝑖𝑗 ln 𝑤𝑖 ln 𝑤𝑖 ln 𝑤𝑗 3 𝑗1 3 𝑖1 𝜙1 ln 𝐸 ln 𝐸 𝑡1 𝑇2 𝜌𝑖𝑗 ln 𝑦𝑖 ln 𝑤𝑗 3 𝑗1 3 𝑖1 𝜓𝑖 ln 𝑦𝑖 ln 𝐸 3 𝑗1 𝜑𝑖𝑇 ln 𝑦𝑖 3 𝑗1 𝜃𝑖 ln 𝑤𝑖 ln 𝐸 3 𝑖1 𝜐𝑖𝑇 ln 𝑤𝑖 3 𝑖1 ln 𝑢𝑐 ln 𝜀𝑐 3 41 onde ln 𝑇𝐶 é o logaritmo do custo de produção 𝑦𝑖𝑖 123 são outputs de quantidade 𝑤𝑗𝑗 123 são inputs de preço ln 𝐸 é o logaritmo natural do capital total 𝑇 é uma tendência temporal 𝑢𝑐 são os componentes da ineficiência de custo E para garantir a homogeneidade linear nos preços dos fatores 𝛽𝑗 3 𝑗1 1 𝛾𝑖𝑗 3 𝑖1 0 𝜌𝑖𝑗 3 𝑗1 0 4 ainda é suficiente e necessário aplicar as seguintes restrições a simetria padrão que de acordo com esta restrição assumese que 𝛿𝑖𝑗 𝛿𝑗𝑖 e 𝛾𝑖𝑗 𝛾𝑗𝑖 Como restrição linear da função custo os inputs e outputs desse trabalho foram definidos também seguindo como referência o trabalho de Fiordelisi Marques Ibanez e Molyneux 2011 de acordo com a abordagem de valor agregado Consideramos as variáveis trabalho despesa de pessoal sobre ativo total manutenção despesas administrativas sobre ativo total e despesas de captação despesas de captação sobre o volume de captações como insumos ou seja inputs e saldo de captações saldo da carteira de crédito e saldo de títulos e valores mobiliários TVM e derivativos como produtos ou outputs Também estimamos utilizando o mesmo modelo funcional adotado para eficiência de custos a eficiência da receita e do lucro usando como variável dependente a receita total e o lucro total respectivamente A definição de inputs e outputs assim como no trabalho referência é a mesma adotada para estimar a eficiência de custos Conforme Belotti et al 2013 as medidas ou escores de eficiência foram estimadas via 𝐸𝑒𝑥𝑝𝑢𝜀 como o estimador criado por Battese e Coelli em seu estudo publicado em 1988 e assumem valores entre 0 e 1 sendo 1 a indicação de um banco totalmente eficiente Esse valor indica a porcentagem de custos observados que seriam suficientes para produzir o resultado observado se o banco fosse totalmente eficiente Por exemplo numa medida de eficiência de 080 poderíamos afirmar que com 80 dos custos totais seria possível produzir os mesmos resultados 42 32 Modelo Autorregressivo Vetorial Conforme a descrição de Lütkepohl 2005 de que se fosse possível obter uma série temporal de dados para uma variável de interesse e esses dados contiverem informações sobre o desenvolvimento futuro de uma variável seria plausível usar como previsão alguma função desses dados coletados optamos por um modelo de séries temporais o Modelo Autorregressivo Vetorial Vector Autoregressive VAR Cavalcanti 2010 destacou que o uso desse método se difundiu rapidamente e que hoje estão entre os instrumentos mais usados em investigações empíricas na área de macroeconomia v 14 n 2 p 251 Ele credita esse feito à percepção de que com ele é possível analisar as interrelações entre diversas variáveis através de um conjunto mínimo de restrições para identificar o componente exógeno de cada variável portanto é possível estimar o efeito de um choque nessa variável sobre as demais Christiano 2012 lembrou que o VAR sugerido por Sims 1980 teria como um dos seus propósitos e o que teve maior impacto projetar e avaliar modelos econômicos Segundo ele os economistas estão acostumados a pensar em modelos econômicos em termos de impulsos e mecanismos de propagação e os VARs são um dispositivo para organizar os dados precisamente nessas categorias v 114 p 1086 tradução nossa Isso reforça nossa escolha Lütkepohl 2005 apresentou as principais etapas de uma análise VAR Figura 1 abaixo que foi a referência seguida Figura 1 Principais etapas da análise VAR Fonte adaptado de LÜTKEPOHL 2005 p 6 43 Christiano 2012 apresenta o VAR como um vetor 𝑦𝑡N x 1 que denota o conjunto de variáveis selecionadas na análise A suposição de que 𝑦𝑡 segue um VAR de ordem 𝑝 significa que ele pode ser expresso como 𝑦𝑡 𝐵𝑜 𝐵1𝑦𝑡1 𝐵𝑝𝑦𝑡𝑝 𝑢𝑡 𝐸𝑢𝑡𝑢 𝑡 𝑉 5 Onde 𝑢𝑡 não está correlacionado com 𝑦𝑡1 𝑦𝑡𝑝 Supõese que 𝑝 tenha um valor suficientemente grande para que 𝑢𝑡 não seja correlacionado automaticamente ao longo do tempo Os distúrbios do VAR 𝑢𝑡 são considerados uma transformação linear dos choques economicamente fundamentais 𝜀𝑡 𝑢𝑡 𝐶𝜀𝑡 𝐶𝐶 𝑉 321 Especificação e estimação do modelo VAR Conforme descrito em Cavalcanti 2013 o primeiro passo é a estimação de um modelo VAR na forma reduzida incluindo como variáveis endógenas além da medida de eficiência outras variáveis domésticas e específicas das instituições financeiras que de acordo com a teoria econômica seriam possíveis candidatas a explicar a evolução da eficiência bancária Como a inclusão simultânea de todas as possíveis variáveis relevantes não é viável por conta dos graus de liberdade na estimação foram testados modelos com diferentes subconjuntos de variáveis Para esse exercício na especificação do modelo utilizamos dimensões econômicas complementares entre si tanto em nível macroeconômico quanto no nível das instituições financeiras Especificamente além da variável que mede o grau de eficiência bancária utilizamos as seguintes variáveis endógenas no VAR a Índice Nacional de Expectativa do Consumidor INEC um indicador criado pela Confederação Nacional da Indústria que sintetiza a opinião dos brasileiros capturada em pesquisa de opinião pública sobre alguns 44 aspectos capazes de afetar as suas decisões de consumo CNI CONFEDERAÇÃO NACIONAL DA INDÚSTRIA 2010 v 2 p 3 Seus resultados variam de 0 a 100 e quanto maior mais confiante é a opinião da população b Risco Brasil ou EMBI Brasil é um índice que avalia a diferença dos retornos financeiros obtidos a cada dia por uma carteira selecionada de títulos brasileiros e a taxa de retorno oferecida por títulos emitidos pelo Tesouro americano Essa diferença é chamada de spread ou spread soberano Quanto maior esse índice maior é a percepção de risco dos investidores O índice é apurado pelo JPMorgan IPEAData 2020 c Produto Interno Bruto PIB oscilação entre os trimestres dos dados do Boletim da Seção de Atividade Econômica estimado mensalmente pelo Banco Central IPEAData 2020 d Índice Geral de Preços do Mercado IGPM registra o ritmo evolutivo de preços como medida síntese da inflação nacional É composto pela média ponderada do Índice de Preços por Atacado IPA 60 Índice de Preços ao Consumidor IPCFGV 30 e Índice Nacional de Preços da Construção Civil INCC 10 e compreende o período entre os dias 20 do mês de referência e 21 do mês anterior É apurado pela área de Conjuntura Econômica da Fundação Getulio Vargas FGV Os dados foram acumulados para representar o trimestre IPEAData 2020 e Índice de Basileia IB índice que representa a relação entre o Patrimônio de Referência3 e os Ativos ponderados pelo risco4 Riskweighted asset RWA Esses dados são divulgados trimestralmente pelo Banco Central do Brasil BANCO CENTRAL DO BRASIL 2020 Além das variáveis domésticas acima descritas outras especificações alternativas foram testadas com variáveis como Concessão de Crédito Inadimplência População Residente Salário Mínimo Taxa de Câmbio Nominal 3 Capital regulatório formado pela soma das parcelas de Capital Nível I e Capital Nível II O Capital Nível I compreende a soma da parcela do capital de melhor qualidade e imediatamente disponível para absorver perdas também denominado Capital Principal e instrumentos de capital e dívida perpétuos elegíveis como patrimônio regulatório ou Capital Complementar O Capital Nível II compreende a parcela do capital composta por instrumentos subordinados elegíveis como patrimônio regulatório aptos a absorver perdas durante o funcionamento da instituição BANCO CENTRAL DO BRASIL 2020 4 Soma da exposição aos riscos de Crédito Mercado e Operacional BANCO CENTRAL DO BRASIL 2020 45 e Real e Taxa de Juros SELIC Foram também testadas variáveis específicas das instituições financeiras como Quantidade de Agências e Postos de Atendimento PA Índice de Imobilização e Retorno sobre o Capital Como próximo passo da identificação do modelo conforme descreveu Cavalcanti 2010 é necessário definir um método de identificação do fator exógeno de cada uma das variáveis sendo assim possível estimar o efeito de um choque em uma das variáveis sobre as demais Ele afirma que o método mais utilizado é a ortogonalização dos resíduos do VAR através da chamada decomposição de Cholesky que impõe uma estrutura recursiva à matriz de relações contemporâneas entre as variáveis do modelo de modo que a primeira variável não seja afetada contemporaneamente por nenhuma das demais a segunda seja afetada apenas pela primeira a terceira seja afetada pelas primeiras duas e assim por diante Ao pesquisador cabe apenas selecionar a ordenação causal adequada das variáveis sob análise CAVALCANTI 2010 v 14 n 2 p 252 Lütkepohl 2005 descreveu esse processo como ordenação causal instantânea das variáveis que por ter sida defendida por Herman Wold é também chamada de Causalidade de Wold Para ele estabelecer essa ordem pode ser um exercício bastante difícil na prática p 61 tradução nossa Cavalcanti 2013 endossa essa afirmação classificando essa etapa como uma das mais difíceis e controversas Na sua opinião a principal dificuldade desse método de identificação reside na apresentação de argumentos convincentes em favor de uma estrutura recursiva com certa ordenação causal específica n 18 p 80 De acordo com os argumentos a seguir a ordenação causal adotada para esse estudo de acordo com a razoabilidade econômica foi a INEC é a variável que ajuda a antecipar variações no ritmo da atividade econômica portanto tendem a ser um indicador antecedente das variações no consumo das famílias e por conseguinte do nível de atividade da economia CNI 2019 v20 Por essas características pode 46 ser classificada como uma variável de resposta rápida ou seja deve preceder as demais variáveis b Risco Brasil como os desempenhos registrados pelo índice ajudam os investidores a formar opinião e a planejar suas decisões futuras IPEAData 2020 ele também contribui para o nível futuro de atividade da economia Como também é resultado das percepções dos agentes assim como o INEC também pode ser classificado como uma variável de resposta rápida Contudo como ele mede apenas a oscilação de preços dos títulos praticados de um dia para o outro IPEAData 2020 como uma variável real deve vir após o INEC c PIB como é razoável supor que as variáveis INEC e Risco Brasil afetem as decisões de produção e que haja uma defasagem entre esse momento e o momento em que a produção efetivamente ocorra que é o que o PIB mede portanto ele deve suceder essas duas variáveis d IGPM como a inflação é uma variável que pode oscilar rapidamente em função de variações no ambiente econômico e é razoável supor que responda contemporaneamente aos choques nas demais variáveis logo deve ser a última variável econômica e Índice de Basileia Considerando a representação dos bancos na economia nacional apesar de um dos setores mais relevantes variáveis nacionais devem preceder as variáveis específicas pois não são por estas afetadas Conforme descrito anteriormente outras especificações alternativas foram testadas nesse caso invertendose a ordenação causal com o objetivo de verificar a robustez dos resultados obtidos Além das etapas descritas acima é fundamental testarmos a estacionariedade da série pois segundo Bueno 2011 esse é o principal conceito que se deve ter ao estimar uma série temporal Gujarati e Porter 2011 reforçam esse ponto ao afirmar que o trabalho empírico baseado nos dados de séries temporais supõe que a série temporal subjacente seja estacionária p 731 grifo do autor Utilizaremos para tal o teste de raiz unitária que segundo os dois últimos autores se tornou amplamente utilizado nos últimos anos Dentre diversas opções optamos por aquele que foi definido por Elder e Kennedy 2001 como 47 o mais popular de muitos testes concorrentes na literatura v 32 n 2 p 138 tradução nossa o teste de DickeyFuller aumentado augmented DickeyFuller ADF Conforme descrito por Gujarati e Porter 2011 nele é possível adotarmos três variedades 1 um passeio aleatório puro 2 um passeio aleatório com deslocamento e 3 um passeio aleatório com deslocamento e tendência p 753 Eles também afirmam que se a série temporal original for não estacionária muito frequentemente suas primeiras diferenças tornamse estacionárias p 445 o que nos foi útil para algumas variáveis A próxima etapa da estimação do modelo foi a seleção da ordem de defasagens das variáveis endógenas do VAR Lütkepohl 2005 afirma que escolher uma ordem de defasagem desnecessariamente grande reduzirá a precisão do modelo e consequentemente da estimativa das respostas ao impulso Para tal foram realizados os critérios mais utilizados nos trabalhos aplicados e selecionada a ordem que mais vezes resultou ou seja que teve maioria Conforme descrito em Ivanov e Kilian 2005 e Lütkepohl 2005 os critérios mais utilizados são o erro de predição final final prediction error FPE o critério de informação de Akaike Akaikes information criterion AIC o critério de informação de Hannan e Quinn Hannan and Quinn information criterion HQIC e o critério de informação bayesiano de Schwarz Schwarzs Bayesian information criterion SBIC Bueno 2011 menciona os três últimos AIC HQIC e SBIC como os principais critérios de informação e acrescenta que seu uso seria uma forma de encontrar o modelo com a quantidade ideal de parâmetros Ele afirma que como os critérios associam uma penalidade ao aumento no número de regressores que por sua vez diminuem a soma dos resíduos eles são a maneira para avaliar se o regressor adicional traz mais benefícios do que custos 48 322 Checagem do Modelo Para nos utilizarmos do que Lütkepohl 2005 descreveu como ferramentas estatísticas para verificar as propriedades de autocorrelação dos resíduos de um determinado modelo VAR p 157 tradução nossa realizaremos o teste de multiplicador de Lagrange Lagrange Multiplier LM ou também chamado teste de BreuschGodfrey para testar as hipóteses nulas de ausência de autocorrelação residual para as ordens de defasagem até a ordem selecionada nos testes mencionados acima Outra checagem relevante é a estabilidade do modelo VAR Segundo afirmou Baum 2013 só é possível analisar as funções impulso resposta e a decomposição da variância se o modelo estimado for estável ou seja não possuir comportamento explosivo Para tal analisamos o comportamento das raízes inversas do polinômio característico do sistema VAR Conforme descreveu Biage Correa e Neder 2008 se todas as raízes encontraremse dentro do círculo ou seja terem módulo menor do que 1 o sistema é estável se se alguma for maior do que 1 o sistema é instável ou seja comportamento explosivo e se for igual a 1 o sistema é nãoestacionário Enders 2014 descreveu essa análise de forma simplificada afirmando que uma maneira sucinta de caracterizar as condições de estabilidade é afirmar que as raízes características devem estar dentro do círculo unitário 4th ed p 29 tradução nossa 323 Análise Estrutural Conforme descrito por Lütkepohl 2005 os modelos VAR são muito utilizados para análise dos relacionamentos entre as variáveis de interesse pois representam as correlações entre um conjunto de variáveis Stock e Watson 2001 afirmaram que o kit de ferramentas padrão na análise VAR é a análise de causalidade de Granger as funções impulso resposta e a decomposição da 49 variância dos erros de previsão Outra abordagem interessante dada pelos autores é que por conta da complexa dinâmica dos modelos VAR esses itens de sua análise são mais informativos que os coeficientes de regressão estimados ou as estatísticas R² 3231 Causalidade de Granger Assim como vários autores como Berger e DeYoung 19975 que tiveram sua abordagem replicada por Williams 2004 e foram citados por Fiordelisi MarquesIbanez e Molyneux 2011 e tantos outros que abordaram o tema eficiência mencionados na seção 2 nós também utilizaremos as técnicas de causalidade de Granger Porém diferentemente desses nós utilizaremos a análise de Granger como uma das ferramentas analíticas no escopo do modelo VAR Conforme mencionado por Berger e DeYoung 1997 e Fiordelisi Marques Ibanez e Molyneux 2011 como essa técnica produz associações estatísticas grosseiras não geraremos prova de causalidade econômica mas como as hipóteses testadas são relações únicas ordenadas por tempo e firmadas entre pares de variáveis ela poderá indicar quais hipóteses são consistentes ou inconsistentes O segundo estudo ainda destaca que apesar desse método ter várias limitações tem sido amplamente utilizado na literatura econômica e nos estudos bancários mencionando os trabalhos acima citados Outros estudos mais recentes já citados aqui também adotaram a mesma técnica como Tan e Anchor 2017 Casu e Girardone 2009 por exemplo mencionaram que a extensão da metodologia original de Granger aos dados do painel permite o uso de séries temporais para testar qualquer relação causal entre duas variáveis v 105 p 5Os autores mencionaram que a aplicação da causalidade de Granger na microeconomia bancária era incomum até então porém não inédita mencionando um trabalho de BERGER Allen N The Relationship between Capital and Earnings in Banking Journal of Money Credit and Banking v 27 n 2 p 432456 1995 50 135 tradução nossa que é o que faremos Lütkepohl 2005 afirmou que em condições adequadas a combinação dessa técnica com modelos VAR é bastante fácil de lidar Berger e DeYoung 1997 descrevem a causalidade de Granger como uma técnica que tenta determinar se a variável 𝑌𝑖 Grangercausa a variável 𝑌𝑗 testando se o histórico passado de 𝑌𝑖 melhora a previsão de 𝑌𝑗 depois de ter considerado o histórico passado de todos os outros 𝑌𝑠 Eles especificam o modelo da seguinte forma 𝑌𝑖𝑡 𝑓1𝑌𝑖𝑙𝑎𝑔 𝑋𝑖𝑙𝑎𝑔 𝑍𝑖𝑙𝑎𝑔 𝑊𝑖𝑙𝑎𝑔 𝐾𝑖𝑙𝑎𝑔 𝑉𝑖𝑙𝑎𝑔 𝑢𝑖𝑡 6 Onde 𝑌𝑖𝑡 é a variável dependente no nosso caso o escore de eficiência sendo 𝑖 a instituição financeira e 𝑡 o trimestre As variáveis do lado direito da equação consideram valores defasados das variáveis dependentes 𝑌 𝑋 𝑍 𝑊 𝐾 e 𝑉 no nosso caso escore de eficiência INEC Risco Brasil PIB IGPM e Índice de Basileia como é o procedimento para análises de causalidade de Granger Lütkepohl 2005 descreve de forma muito similar afirmando que se 𝑌𝑗 pode ser previsto com mais eficiência se as informações no processo 𝑌𝑖 forem levadas em consideração além de todas as outras informações no universo então 𝑌𝑖 Grangercausa para 𝑌𝑗 Ele afirmou que como a causalidade de Granger pode não explicar as interações entre as variáveis de um sistema e que em qualquer estudo pode ser necessário conhecer a resposta de uma variável a um impulso em outra variável em um sistema que envolve várias variáveis podemos recorrer a Análise de Impulso Resposta 3232 Função ImpulsoResposta Segundo Lütkepohl 2005 pode ser útil conhecer o comportamento de uma variável a um impulso em outra variável em um sistema com várias variáveis Em outras palavras como uma variável se comporta a um impulso em outra variável Ele complementa que se houver uma reação a esse impulso podemos 51 supor que há um tipo de causalidade como um efeito de um choque ou inovação exógena em uma algumas ou até mesmo em todas as variáveis do modelo Para tal análise utilizaremos a Função ImpulsoResposta Impulse Response Function IRF que segundo Cavalcanti 2013 é a descrição dos efeitos de variações inesperadas nas variáveis do sistema sobre a trajetória de cada variável ao longo do tempo p 81 Baum 2013 a define de forma muito similar afirmando que elas medem os efeitos de um choque e esclarece que se o modelo VAR for estável subseção 322 podemos reescrever o modelo na forma de média móvel 𝑌𝑡 𝜇 𝐷𝑖𝑋𝑡𝑖 𝑖0 𝜙𝑖𝑈𝑡𝑖 𝑖0 7 Para ele essa equação acima é a representação da média móvel vetorial vector moving average VMA e demonstra que todos os valores passados de 𝑌𝑡 foram substituídos Ele denomina as matrizes 𝐷𝑖 como as funções multiplicadores dinâmicas ou funções de transferência6 e a sequência de coeficientes de média móvel 𝜙𝑖 são as funções de impulso resposta no horizonte 𝑖 Conforme descrito na subseção 321 deste estudo como os resíduos do modelo proposto foram ortogonalizados através da decomposição de Cholesky e de acordo com a proposição de Sims 1980 as IRFs submetidas a essa escolha são conhecidas como IRFs ortogonalizadas orthogonalized IRFs OIRFs Baum 2013 Como nosso intuito será analisalas de forma cumulativa faremos uso do conceito de OIRF cumulativas cumulative OIRF COIRF Vale destacar conforme o estudo de Kim 2019 no qual nos baseamos que há uma abordagem alternativa e que não estaria sujeita ao problema da ordenação das variáveis a função generalizada de resposta ao impulso Generalized Impulse Response Function GIRF Contudo apesar de ter sido empregada por muitas pesquisas o próprio autor conclui que lhe parece mais razoável a OIRF 6 Bueno 2011 denomina esses elementos da matriz como multiplicadores de impacto de um choque sobre as variáveis endógenas p 217 52 3233 Análise da Decomposição da Variância dos Erros de Previsão Outra importante análise é a decomposição de variância dos erros de previsão forecast error variance decomposition FEVD descrita por Cavalcanti 2013 como uma opção que permite quantificar a relevância que em média os choques em cada variável do sistema possuem na explicação das flutuações dessas variáveis em torno de seus valores previstos p81 Para Baum 2013 ela mensura o quanto percentualmente da variação do erro de previsão de uma variável no caso endógena pode ser atribuída através de choques ortogonais para outra variável endógena ou para ela mesma Adicionalmente conforme descrito no item acima sobre a Função ImpulsoResposta a FEVD também é sensível ao problema da ordenação Baum 2013 Enders 2014 afirmou que a análise de impulso e a decomposição de variância denominadas conjuntamente de contabilidade da inovação podem ser ferramentas úteis para examinar as relações entre variáveis econômicas p 302 tradução nossa Por isso optase por incluir a análise do FEVD neste estudo de maneira complementar às funções de impulsoresposta na medida em que quantificam a importância relativa de cada variável do sistema na explicação das flutuações futuras da variável de eficiência bancária 324 Previsão Conforme Figura 1 acima um modelo VAR tem como principais usos a análise estrutural subseção 323 e a previsão conforme descrito por Lütkepohl 2005 Bueno 2011 afirmou que quando se conhece o processo gerador de dados no nosso caso o modelo VAR estimado a previsão ℎ passos à frente é descrita como 𝐸𝑋𝑡ℎ𝐼𝑡 𝑋𝑡ℎ𝑡 𝜙1𝑋𝑡ℎ1𝑡 𝜙2𝑋𝑡ℎ2𝑡 𝜙𝑝𝑋𝑡ℎ𝑝𝑡 8 53 em que 𝑋𝑡𝑗𝑡 𝑋𝑡𝑗 para 𝑗 0 E como temos que 𝑋𝑡 é estacionário podemos transformálo em um modelo de médias móveis infinito 𝑋𝑡ℎ 𝐼 𝜙𝑗𝐿𝑗 𝑝 𝑗1 1𝑒𝑡ℎ 𝑒𝑡ℎ Ψ1𝑒𝑡ℎ1 Ψ𝑝𝑒𝑡ℎ2 9 Logo podemos reescrever a previsão correspondente como 𝑋𝑡ℎ𝑡 Ψ𝑗𝑒𝑡ℎ𝑗 𝑗ℎ 10 De acordo com Stock e Watson 2001 analisamos o desempenho do modelo VAR com pseudoprevisões fora da amostra ou seja estimamos o modelo até um período anterior aos últimos dados utilizados para que a previsão possa ser comparada com dados reais Adicionalmente assim como eles também rodamos pseudoprevisões para uma autorregressão AR com quatro defasagens ou seja uma regressão da variável de eficiência em defasagem de seus quatro últimos valores passados Conforme sugestão de Enders 2014 para mensurar a precisão da previsão e comparar os métodos VAR e AR utilizamos o teste de DieboldMariano7 Esse teste calcula uma medida de precisão preditiva proposta pelos autores de mesmo nome 1995 De forma sucinta conforme descreveu Baum 2003 dada uma série de dados real e duas previsões aplicase um critério de perda nós utilizaremos três conforme descrição abaixo e calculase medidas de precisão preditiva que permitem que a hipótese nula de igual precisão seja testada Isso se dá através do teste de que a diferença média entre os critérios de perda para as duas previsões é zero Os três critérios de perda utilizados foram erro médio absoluto mean absolute error MAE erro quadrático médio mean square error MSE e erro percentual médio absoluto mean absolute percentage error MAPE Como nenhum deles é perfeito e apresenta vantagens e desvantagens utilizaremos os três como referência de análise Kennedy 2008 os descreve da seguinte maneira 7 Diebold e Mariano 1995 complementam que seu teste de comparação da previsão seria um dos muitos diagnósticos que devem ser examinados ao comparar modelos p 262 tradução nossa 54 Erro médio absoluto também conhecido como desvio médio absoluto é a média dos valores absolutos dos erros de previsão e é uma medida sensível ao dimensionamento Erro quadrático médio é a raiz quadrada da média dos valores quadrados dos erros de previsão também é sensível ao dimensionamento e além disso é sensível à outliers É segundo o autor8 a medida mais usada Erro percentual médio absoluto é a média dos valores absolutos dos erros percentuais e diferentemente das demais não é sensível ao dimensionamento porém apresenta outras desvantagens como a base para medir a porcentagem pode ser zero ou muito grande tornandoa indefinida ou distorcida e penaliza mais os erros positivos do que os negativos 33 Dados e amostra Nosso exercício para levantamento dos dados pode ser dividido em duas frentes 1 os dados utilizados para a análise de fronteira estocástica específicos de cada instituição financeira que serviram como referência para apuração dos escores de eficiência e 2 os dados utilizados para investigação das inter relações entre eficiência e outras variáveis ambos disponíveis no Apêndice A Os dados utilizados na primeira frente os específicos dos bancos foram extraídos do IFData portal do Banco Central do Brasil que disponibiliza trimestralmente informações das instituições que autoriza a funcionar e que estejam em operação normal Para a segunda frente além dos dados já levantados para a primeira e os dados de eficiência resultantes desse exercício as variáveis macroeconômicas foram capturadas do Ipeadata a base de dados macroeconômicos financeiros e regionais do Brasil mantida pelo Instituto de Pesquisa Econômica Aplicada IPEA 8 Lütkepohl 2005 reforça essa afirmação dizendo que os preditores que minimizam os erros quadráticos médios de previsão são os mais utilizados p 33 tradução nossa 55 Utilizamos dados trimestrais recentes dos cinco maiores bancos brasileiros9 Itaú Banco do Brasil Caixa Bradesco e Santander para o período de 2009 e 2019 Além destes foram criadas três séries de dados uma média ponderada entre esses cinco bancos aqui denominada Geral outra entre os três bancos privados Itaú Unibanco Bradesco e Santander aqui denominada Privados e outra entre os dois bancos públicos Banco do Brasil e Caixa Econômica Federal aqui denominada Públicos O intuito foi possibilitar a realização de mais comparações além dos bancos entre si Estas séries foram elaboradas conforme critério utilizado pelo Banco Central do Brasil 2017 ao estabelecer a sua segmentação das instituições financeiras ou seja o conceito de Exposição total10 Tentando não perpetuar o problema abordado por DeYoung 1997 de que os estudos existentes incluem poucas justificativas estatísticas para suas escolhas de muitas ou poucas observações de séries temporais nossa decisão por partir de 2009 não foi arbitrária Nós não objetivávamos avaliar nesse estudo a crise de 2008 a qual imaginávamos que afetaria nossa análise Os dados da Matriz de correlação das variáveis do modelo VAR disponível no Apêndice B se olharmos somente as correlações de cada uma das oito amostras ou seja sem analisar a correlação entre as variáveis de diferentes instituições financeiras a correlação mais alta foi de 056 e a mais baixa de 040 indicando que as variáveis em questão não apresentaram alta correlação sugerindo que não há problema de colinearidade De acordo com Kennedy 2008 a multicolinearidade é um problema quando a correlação está acima de 080 9 De acordo com a segmentação do Banco Central BANCO CENTRAL DO BRASIL 2017 os bancos classificados como S1 são aqueles que i tenham porte ExposiçãoProduto Interno Bruto superior a 10 ou ii exerçam atividade internacional relevante ativos no exterior superiores a US 10 bilhões Para efeito desse estudo como buscamos analisar e comparar a eficiência bancária Banco BTG Pactual foi a única instituição desconsiderada dentre os classificados como S1 data base Dez19 Além de não atender o item i da classificação ele não possui uma rede de agências ampla o que inviabilizaria as comparações 10 Disponíveis no IFData sem ponderação de risco conforme definido na Circular 3748 de 27 de fevereiro de 2015 56 As estimativas da primeira e da segunda frente a modelagem e todos os testes foram feitos através do software Stata 142 e a estatística descritiva das variáveis selecionadas está na Tabela 1 abaixo Tabela 1 Estatística descritiva das variáveis Fonte Elaborado pelo próprio autor Nota O prefixo dif diz respeito ao operador de primeira diferença Por conta dos resultados dos testes onde algumas variáveis apresentaram não estacionaridade conforme Bueno 2011 descreveu bastou a primeira diferença para estacionarizálas Isso ocorreu para os Índices de Basileia e o Risco Brasil Os resultados desses testes estão disponíveis no Apêndice C Tipo Indicação Média Desvio Padrão Mínimo Máximo Itaú ITAUcosteffic 0715 0016 0678 0744 BB BBcosteffic 0735 0022 0685 0781 Bradesco BRADESCOcosteffic 0724 0020 0682 0770 Caixa CAIXAcosteffic 0725 0052 0548 0785 Santander SANTANDERcosteffic 0724 0022 0669 0767 Privados PRIVcosteffic 0720 0016 0684 0751 Públicos PUBcosteffic 0729 0036 0611 0783 Geral GERALcosteffic 0724 0022 0654 0763 Itaú difITAUbasileia 0000 0009 0022 0018 BB difBBbasileia 0001 0007 0022 0019 Bradesco difBRADESCObasileia 0000 0010 0031 0023 Caixa difCAIXAbasileia 0000 0011 0027 0024 Santander difSANTANDERbasileia 0002 0014 0030 0051 Privados difPRIVbasileia 0001 0008 0015 0015 Públicos difPUBbasileia 0000 0008 0025 0020 Geral difGERALbasileia 0000 0006 0012 0012 INEC 0610 0056 0504 0723 difRiscoBrasil 0000 0006 0015 0013 PIB 0020 0037 0073 0079 IGPM 0014 0015 0027 0039 Específicas Domésticas Variável Escore de Eficiência de custo Índice de Basileia Índice Nacional de Expectativa do Consumidor Risco Brasil Produto Interno Bruto Índice Geral de Preços do Mercado 57 4 RESULTADOS E DISCUSSÃO 41 Resultados Modelo VAR Começamos essa seção apresentando os resultados do Modelo VAR para as oito séries de dados cinco bancos e as três séries criadas Conforme descrito na seção anterior como etapa inicial nesse processo foram testadas a estacionaridade das séries através do teste de ADF Os resultados nas três variedades disponíveis no Apêndice C apresentaram resultado de que as séries das variáveis selecionadas são estacionárias com deslocamento Seguindo as etapas descritas na Metodologia o próximo passo foi a seleção da ordem de defasagens das variáveis endógenas do VAR Para todos os modelos na maioria dos critérios quando não na totalidade a quarta ordem de defasagem foi a que apresentou os melhores resultados o que seria intuitivo e razoável para dados trimestrais O estudo de Ivanov e Kilian 2005 reforça esse resultado pois a maioria dos estudos empíricos com levantamento de dados trimestrais a ordem de defasagens encontrada foi de quatro períodos Esses resultados podem ser observados na Tabela 2 abaixo Tabela 2 Critérios de seleção da ordem de defasagens Fonte Elaborado pelo próprio autor Nota Os testes consideraram o nível de significância de 5 Após a especificação e estimação dos modelos foi realizada sua checagem através do teste LM Lagrange Multiplier para verificar a ausência de Estatística Lag Estatística Lag Estatística Lag Estatística Lag Itaú 36E24 4 385202 4 362203 4 327502 0 Banco do Brasil 14E24 4 394435 4 394435 4 336314 1 Bradesco 22E23 4 367112 4 344113 4 318718 1 Caixa 14E23 4 372002 4 349003 4 316686 1 Santander 16E23 4 370092 4 347093 4 319436 0 Geral 69E25 4 400266 4 377310 4 336556 1 Privados 42E25 4 405372 4 382415 4 341388 4 Públicos 24E24 4 387812 4 364856 4 333312 1 Série FPE AIC HQIC SBIC 58 autocorrelação residual Embora tenha rejeitado a hipótese nula em alguns lags o modelo não apresenta fortes evidências de presença de autocorrelação serial Os resultados estão disponíveis no Apêndice D Os gráficos de correlograma cruzado Apêndice E corroboram com essa afirmação Outra checagem realizada foi a estabilidade dos modelos VAR checando se o modelo não possuía comportamento explosivo ou seja se as raízes características estão dentro do círculo unitário Foi encontrada estabilidade para os oito modelos e os resultados estão disponíveis no Apêndice F 42 Análises Estruturais 421 Análises de causalidade de Granger Demos sequência aos modelos VAR através das Análises Estruturais mais especificamente na análise de causalidade de Granger nas funções impulso resposta e na decomposição da variância dos erros de previsão Vale lembrarmos que os testes de causalidade de Granger testarão a hipótese nula de que os coeficientes estimados dos valores defasados das variáveis do modelo para os dados analisados são conjuntamente zero Ou seja se não rejeitarmos a hipótese nula não rejeitaremos a hipótese de que a variável excluída representada na tabela como b não Grangercausa a variável da equação representada na tabela como a De modo a facilitar a interpretação e visualização dos resultados as probabilidades da Tabela 3 abaixo destacadas em verde expressam os casos em que rejeitamos a hipótese nula ou seja há evidências estatísticas de que as variáveis da coluna b Grangercausa as variáveis da coluna a De forma análoga as destacadas em vermelho não Grangercausam Os resultados revelaram que o INEC Grangercausa a Eficiência de Custo para a maioria dos bancos sendo o Banco do Brasil e a média ponderada dos bancos públicos as exceções O efeito descrito por Gujarati e Porter 2011 como 59 causalidade bilateral ou feedback também é percebido nessa relação ou seja a Eficiência de Custos também Grangercausa o INEC para a maioria dos bancos sendo o Bradesco e novamente os bancos públicos as exceções Os dados também revelaram que o Risco Brasil Grangercausa a Eficiência de Custos para o Itaú Bradesco Caixa e a média ponderada dos bancos privados mas não Grangercausa para Banco do Brasil Santander a média ponderada dos públicos e a média ponderada geral E o efeito feedback foi descartado nas cinco amostras Foi possível também evidenciar que o PIB Grangercausa a Eficiência de Custos e que a Eficiência de Custos Grangercausa o PIB com exceção do Banco do Brasil e da média ponderada geral dos bancos O IGPM apresentou evidências de que Grangercausa a Eficiência de Custos para Itaú Bradesco Santander e as três médias ponderadas da nossa análise mas não Granger causa para os dois bancos públicos o Banco do Brasil e a Caixa O efeito feedback nesse caso também evidenciou para a maioria dos bancos que a Eficiência de Custos não Grangercausa o IGPM com exceção da Caixa e da média ponderada dos bancos públicos O Índice de Basileia não Grangercausa a Eficiência de Custos para Itaú Banco do Brasil Bradesco e as médias ponderadas dos privados e a geral mas Grangercausa para Caixa Santander e a média ponderada dos públicos já o efeito feedback nesse caso se mostrou verdadeiro para as oito amostras ou seja a Eficiência de Custos Grangercausa o Índice de Basileia Os demais resultados dos testes de causalidade de Granger como não foram foco desse estudo estão disponíveis no Apêndice G 60 Tabela 3 Probabilidades do teste de causalidade de Granger para Eficiência de Custos Fonte Elaborado pelo próprio autor Notas Os números expressos na tabela representam a probabilidade de aceitar a hipótese nula sendo que a fonte verde representa os casos em que rejeitamos e em vermelho os casos em que aceitamos Consideramos o nível de significância de 5 422 Análises das Funções ImpulsoResposta Seguindo a análise dos modelos VAR estimamos as funções impulsoresposta a partir dos coeficientes estimados Como descrito acima nosso foco está na resposta do escore de eficiência de cada Instituição Financeira aos impulsos também chamados de choques nas demais variáveis do modelo INEC Risco Brasil PIB IGPM e Basileia isoladamente mantendo todas as demais constantes As Figuras 2 3 4 5 e 6 respectivamente Itaú Banco do Brasil Bradesco Caixa Santander na subseção Análise Individual dos cinco grandes bancos e 7 8 e 9 respectivamente Privados Públicos e Geral na subseção Análise das médias ponderadas mostram essas funções ortogonalizadas cumulativas ao longo do período considerando oito trimestres após o choque11 11 Foi considerada significância de 68 indicada na legenda do gráfico como Intervalo de Confiança IC para todas as demonstrações das Funções ImpulsoResposta Equação a Excluído b Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral Eficiência de Custo INEC 0005 0487 0000 0000 0000 0000 0104 0000 INEC Eficiência de Custo 0003 0032 0285 0017 0000 0000 0146 0003 Eficiência de Custo Risco Brasil 0014 0876 0000 0041 0500 0001 0561 0053 Risco Brasil Eficiência de Custo 0671 0105 0078 0288 0060 0118 0538 0782 Eficiência de Custo PIB 0000 0000 0000 0000 0000 0000 0000 0000 PIB Eficiência de Custo 0000 0539 0000 0004 0006 0002 0007 0128 Eficiência de Custo IGPM 0013 0916 0009 0091 0000 0001 0023 0000 IGPM Eficiência de Custo 0570 0353 0905 0000 0976 0561 0000 0108 Eficiência de Custo Basileia 0522 0660 0287 0000 0001 0193 0006 0714 Basileia Eficiência de Custo 0000 0000 0002 0046 0005 0000 0002 0011 Eficiência de Custo Todas 0000 0000 0000 0000 0000 0000 0000 0000 61 De forma sucinta o eixo das abcissas mostra a quantidade de trimestres após o choque e o eixo das ordenadas a resposta cumulativa em pontos percentuais do choque de um desviopadrão 4221 Análise individual dos cinco grandes bancos Figura 2 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Itaú Fonte elaborado pelo próprio autor Dos resultados do Itaú inferese pela Figura 2 acima que o INEC e o próprio escore de eficiência afetam positivamente os movimentos futuros do escore de eficiência mantendo tudo o mais constante Já as respostas do escore de eficiência aos choques sobre o IGPM PIB e Basileia também foram estatisticamente diferentes de zero mas negativas nos três casos exceção feita ao efeito no último trimestre no caso 8º para o PIB e no 1º trimestre para Basileia onde os efeitos não foram estatisticamente diferentes de zero Para o 62 Risco Brasil com exceção do primeiro trimestre onde a resposta do escore de eficiência foi negativa para os demais trimestres não apresentou relação estatisticamente diferente de zero Figura 3 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Banco do Brasil Fonte elaborado pelo próprio autor Para os resultados do Banco do Brasil podemos inferir pela Figura 3 acima que o Risco Brasil o PIB o IGPM e Basileia não apresentaram relação estatisticamente diferente de zero com o escore de eficiência A reposta do escore de eficiência ao choque no INEC só foi estatisticamente diferente de zero a partir do 7º trimestre sendo negativa nesse caso Diferentemente das demais variáveis choque do próprio escore de eficiência afeta positivamente os seus movimentos futuros mantendo tudo o mais constante da mesma forma que o resultado encontrado nessa relação para o Itaú 63 Figura 4 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Bradesco Fonte elaborado pelo próprio autor No caso do Bradesco na Figura 4 acima os resultados mostram que o Risco Brasil e Basileia assim como no Banco do Brasil não apresentaram relação estatisticamente diferente de zero com o escore de eficiência O mesmo vale para o IGPM com exceção do 1º trimestre que apresentou relação negativa Já o choque no PIB com exceção do 1º trimestre onde o resultado não foi estatisticamente diferente de zero afetou negativamente os movimentos futuros do escore de eficiência Para o INEC e o próprio escore de eficiência os choques afetaram positivamente sendo que no caso do INEC esse efeito só foi estatisticamente diferente de zero até o 4º trimestre 64 Figura 5 Função Impulsoresposta das diversas variáveis ao escore de eficiência da Caixa Fonte elaborado pelo próprio autor Já os resultados da Caixa conforme Figura 5 acima mostram que o PIB assim como no Banco do Brasil e o IGPM não apresentaram relação estatisticamente diferente de zero com o escore de eficiência com exceção do 1º trimestre para o IGPM onde o efeito foi positivo No caso do Risco Brasil o choque nos movimentos futuros do escore de eficiência só foi estatisticamente diferente de zero nos 2º e 3º trimestres tendo efeito positivo Para o próprio escore de eficiência e para Basileia os efeitos dos choques no escore de eficiência foram positivos em todos os trimestres No efeito do INEC nos movimentos futuros do escore de eficiência houve uma oscilação nos três primeiros trimestres os efeitos foram positivos e no 8º foi negativo Nos demais trimestres os efeitos não foram estatisticamente diferentes de zero 65 Figura 6 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Santander Fonte elaborado pelo próprio autor Como último banco dessa etapa da análise conforme Figura 6 acima os resultados do Santander mostram que o os choques no INEC e na Basileia afetaram negativamente em todos os trimestres os movimentos futuros do escore de eficiência Já os choques no Risco Brasil e no PIB afetaram positivamente o escore de eficiência sendo que os resultados não foram estatisticamente significativos somente para o 1º trimestre no caso do Risco Brasil e somente no 3º trimestre no caso do PIB Para o IGPM com exceção do 1º trimestre que apresentou resultado negativo no escore de eficiência os demais trimestres não foram estatisticamente diferentes de zero Sobre o efeito do choque do próprio escore de eficiência sobre ele mesmo assim como todos os bancos foi positivo 4222 Análise das médias ponderadas dos públicos privados e geral 66 Figura 7 Função Impulsoresposta das diversas variáveis ao escore de eficiência da média ponderada dos Privados Fonte elaborado pelo próprio autor Ao compararmos os resultados dessa média ponderada dos bancos privados com os resultados individuais das instituições financeiras que o compõe podemos destacar que um choque no INEC afetou os movimentos futuros da média ponderada do escore de eficiência positivamente nos primeiros trimestres como no Itaú e Bradesco e negativamente no último como no Santander No caso do Risco Brasil o efeito foi negativo no início como no Itaú e positivo no 8º trimestre como no Santander Para o PIB o início foi positivo como no Santander e negativo no 3º e 7º trimestres como no Itaú e Bradesco No caso do IGPM o choque foi estatisticamente diferente de zero somente para o 2º e 5º trimestres com efeito negativo assim como o Itaú que contudo teve efeito em todos O choque em Basileia afetou o escore de eficiência de forma negativa similar ao Itaú e Santander 67 Figura 8 Função Impulsoresposta das diversas variáveis ao escore de eficiência da média ponderada dos Públicos Fonte elaborado pelo próprio autor Já na comparação dos resultados da média ponderada dos bancos públicos com os resultados do Banco do Brasil e da Caixa notamos que um choque no INEC afetou os movimentos futuros da média ponderada do escore de eficiência positivamente nos primeiros trimestres como na Caixa No Risco Brasil o efeito foi positivo no 1º trimestre e estatisticamente não diferente de zero nos demais parecido com o movimento da Caixa Para o PIB diferentemente do Banco do Brasil e da Caixa onde ambos apresentaram resultados estatisticamente não diferentes de zero o efeito foi positivo a partir do 6º trimestre De similar o IGPM só apresentou o resultado do 1º trimestre positivo como na Caixa E por último o choque em Basileia afetou o escore de eficiência de forma positiva como na Caixa em todos os trimestres Em linhas gerais não houve muita semelhança entre os resultados da média ponderada dos bancos públicos e os resultados do Banco do Brasil 68 Figura 9 Função Impulsoresposta das diversas variáveis ao escore de eficiência da média ponderada dos bancos Fonte elaborado pelo próprio autor Na última análise dos gráficos das Funções ImpulsoResposta Ortogonalizadas Cumulativas ao analisarmos a média ponderada das cinco instituições financeiras notamos uma semelhança dos resultados do efeito do choque no INEC nos movimentos futuros do escore de eficiência ao resultado da Caixa e dos Bancos Privados com um efeito positivo nos primeiros trimestres e negativo nos últimos No Risco Brasil o efeito foi positivo somente no 8º trimestre que foi o que observamos somente na média ponderada dos privados No PIB o efeito foi estatisticamente não diferente de zero para todos os trimestres o que ocorreu na Caixa no Banco do Brasil e em alguns poucos trimestres do Itaú Bradesco e Santander Para o IGPM o efeito foi negativo no 4º e 5º trimestres e positivo no 8º Esse efeito positivo no final só foi encontrado nessa situação Basileia apresentou um resultado estatisticamente não diferente de zero como encontrado no Banco do Brasil e no Bradesco Por último como todos os gráficos apresentados nessa etapa da análise em todas as instituições e nas três médias ponderadas o efeito do choque do próprio escore de eficiência sobre os movimentos futuros dele mesmo foram positivos 69 Além das análises e interpretações dos gráficos de função impulsoresposta realizadas acima para cada uma das oito amostras cinco instituições financeiras e as três médias a Tabela 4 abaixo representa a consolidação das funções impulsoresposta das diversas variáveis ao escore de eficiência de todas as amostras Tabela 4 Resumo das Funções Impulsoresposta das diversas variáveis ao escore de eficiência de todas as amostras Fonte elaborado pelo próprio autor Nota Os casos estatisticamente indistinguíveis de zero foram identificados como indist e os números entre parênteses indicam os trimestres após o choque 423 Análises das Decomposições das Variâncias dos Erros de Previsão A nossa última Análise Estrutural dos modelos VAR é a decomposição da variância dos erros de previsão conforme descrito na seção 3 Ela nos permitirá avaliar conforme descrito por Stock e Watson 2001 o quanto em percentual da variância do erro cometido na previsão do escore de eficiência é devido a um choque em uma das variáveis do modelo no caso o INEC o Risco Brasil o PIB o IGPM o Basileia e o próprio escore de eficiência As tabelas 5 6 7 8 9 10 11 e 12 abaixo apresentarão a fração da variância do erro de previsão do escore de eficiência que pode ser atribuída a cada variável do modelo em cada um dos oito trimestres após o choque Variável Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral INEC Positivo Indist 16 e negativo 78 Positivo 14 e indist 58 Positivo 13 indist 47 e negativo 8 Negativo Positivo 12 indist 37 e negativo 8 Positivo 13 e indist 48 Positivo 12 indist 34 e negativo 58 Risco Brasil Negativo 1 e indist 28 Indist Indist Positivo 23 e indist 14 8 Indist 1 e positivo 28 Negativo 1 indist 27 e positivo 8 Positivo 1 e indist28 Indist 17 e positivo 8 PIB Negativo e indist 8 Indist Negativo 28 e indist 1 Indist Positivo e indist 3 Positivo 1 indist 28 e negativo 37 Indist 15 e positivo 68 Indist IGPM Negativo Indist Negativo 1 e indist 28 Positivo 1 e indist 28 Negativo 1 e indist 28 Indist 168 e negativo 2 5 Positivo 1 indist 2 68 e negativo 3 5 Indist 13 6 7 negativo 4 5 e positivo 8 Basileia Negativo e indist 1 Indist Indist Positivo Negativo Negativo Positivo Indist Eficiência de Custo Positivo Positivo Positivo Positivo Positivo Positivo Positivo Positivo 70 Tabela 5 Estimativas pontuais das decomposições de variância do erro de previsão do Itaú Fonte elaborado pelo próprio autor A tabela 5 acima mostra as estimativas da decomposição de variância do erro de previsão para o escore de eficiência do Itaú Podemos notar que os choques no escore de eficiência são responsáveis por cerca de 55 das oscilações futuras do próprio índice revelando uma persistência das estruturas que o compõe Por outro lado outros fatores representam os 45 restantes com destaque para o Risco Brasil cerca de 15 Podemos também observar o aumento da importância do Risco Brasil na explicação dos movimentos do escore de eficiência do Itaú começando a partir do 4º trimestre e saltando novamente depois do 7º atingindo cerca de 15 conforme mencionado As demais variáveis oscilaram entre cerca de 8 e 6 tendo destaque o salto de relevância a partir do 3º trimestre Tabela 6 Estimativas pontuais das decomposições de variância do erro de previsão do Banco do Brasil Fonte elaborado pelo próprio autor Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 894 21 35 21 27 02 3 789 36 42 41 62 31 4 742 39 71 50 59 40 5 706 37 73 70 76 38 6 652 40 74 76 78 79 7 568 63 157 65 70 78 8 555 75 154 64 68 84 Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 973 03 07 10 00 07 3 955 03 15 20 02 06 4 887 55 24 26 02 08 5 883 64 29 18 01 05 6 884 72 23 15 01 04 7 873 80 22 19 02 05 8 847 106 22 18 02 05 71 No caso do Banco do Brasil de acordo com a tabela 6 acima notamos que os choques no escore de eficiência são responsáveis por cerca de 85 das oscilações futuras do próprio índice revelando uma persistência das estruturas que o compõe muito superior à do Itaú Já os outros fatores que representam os 15 restantes somente o INEC merece destaque com cerca de 11 Cabe destacar o aumento da importância dele na explicação dos movimentos do escore de eficiência do Banco do Brasil a partir do 4º trimestre e novamente a partir do 8º Vale lembrarmos que um choque positivo no INEC leva a uma resposta negativa no escore de eficiência do Banco do Brasil As demais variáveis oscilaram entre cerca de 0 e 2 Tabela 7 Estimativas pontuais das decomposições de variância do erro de previsão do Bradesco Fonte elaborado pelo próprio autor Na análise do Bradesco de acordo com a tabela 7 acima podemos observar que os choques no escore de eficiência são responsáveis por cerca de 61 das oscilações futuras do próprio índice revelando uma persistência das estruturas que o compõe mais próxima as do Itaú e menos relevantes que a do Banco do Brasil Merece menção o PIB com cerca de 17 que saltou de cerca de 0 para cerca de 8 a partir do 3º trimestre que conforme vimos na análise das funções de impulsoresposta afeta negativamente o escore de eficiência deste banco Outra variável muito relevante para explicação das oscilações futuras é o INEC assim como foi com o Banco do Brasil nesse caso com cerca de 13 muito similar ao valor observado logo no 2º trimestre Para essa variável nesse banco Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 827 128 14 00 26 06 3 743 129 12 81 23 12 4 700 117 40 111 21 12 5 703 112 39 108 26 11 6 694 106 43 122 24 11 7 643 96 51 172 24 14 8 609 134 51 169 24 14 72 choques positivos levaram a uma resposta também positivo no escore de eficiência Tabela 8 Estimativas pontuais das decomposições de variância do erro de previsão da Caixa Fonte elaborado pelo próprio autor Para a Caixa de acordo com a tabela 8 acima os choques no escore de eficiência são responsáveis por cerca de 43 das oscilações futuras do próprio índice revelando que outras variáveis contribuem significativamente para os movimentos futuros do seu escore de eficiência O principal destaque diferentemente dos demais bancos analisados é o Basileia que representou cerca de 33 tendo saltando do 2º para o 3º trimestre de cerca de 5 para cerca de 41 lembrando que choques positivos em Basileia afetavam positivamente o escore de eficiência da Caixa Outra variável muito relevante é o INEC assim como foi para o Banco do Brasil e Bradesco que nesse caso representou cerca de 13 das oscilações futuras Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 910 25 00 05 09 51 3 458 47 61 22 06 407 4 445 45 61 25 09 415 5 455 50 58 36 14 387 6 474 114 56 30 11 315 7 466 118 60 31 14 311 8 430 129 63 37 15 328 73 Tabela 9 Estimativas pontuais das decomposições de variância do erro de previsão do Santander Fonte elaborado pelo próprio autor Já para o Santander de acordo com a tabela 9 acima podemos observar que os choques no escore de eficiência são responsáveis por cerca de 33 das oscilações futuras do próprio índice revelando que as demais variáveis somadas explicam melhor as oscilações futuras do Índice do que ele próprio Com exceção do IGPM com cerca de 7 as demais variáveis representam mais de 10 das oscilações futuras do Índice do Santander A mais relevante é o INEC como foi para o Bradesco e o Banco do Brasil com cerca de 27 tendo atingido essa relevância logo no 4º trimestre da análise As demais variáveis na ordem decrescente de relevância foram Basileia com cerca de 13 PIB com cerca de 11 e Risco Brasil com cerca de 10 Tabela 10 Estimativas pontuais das decomposições de variância do erro de previsão da média ponderada dos privados Fonte elaborado pelo próprio autor Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 798 38 11 34 17 102 3 586 114 73 32 39 156 4 468 264 54 25 36 153 5 468 257 47 50 34 145 6 391 220 97 88 68 136 7 362 246 90 97 69 136 8 326 268 101 113 67 126 Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 861 60 34 22 04 20 3 804 81 32 46 18 19 4 754 104 39 64 17 22 5 780 75 46 46 37 17 6 764 75 46 54 45 17 7 679 65 112 48 80 16 8 630 114 114 44 82 15 74 A tabela 10 acima mostra as estimativas da decomposição de variância do erro de previsão para o escore de eficiência da média ponderada dos três bancos privados da nossa amostra Podemos notar que os choques no escore de eficiência são responsáveis por cerca de 63 das oscilações futuras do próprio índice Dentre os demais fatores destacamos o INEC e o Risco Brasil ambos com cerca de 11 sendo que o segundo atingiu essa relevância somente a partir do 7º trimestre Outra variável que merece destaque é o IGPM que representou cerca de 8 da oscilação futura Tabela 11 Estimativas pontuais das decomposições de variância do erro de previsão da média ponderada dos públicos Fonte elaborado pelo próprio autor A tabela 11 acima mostra as estimativas da decomposição de variância do erro de previsão para o escore de eficiência da média ponderada dos dois bancos públicos da nossa amostra onde podemos notar que os choques no escore de eficiência são responsáveis por cerca de 72 das oscilações futuras do próprio índice evidenciando uma persistência maior que a média dos bancos privados Para as demais variáveis merecem destaque o INEC e o PIB com cerca de 10 e 7 respectivamente lembrando que ambos apresentaram resultados que evidenciaram efeito positivo no escore de eficiência após choques positivos em seus resultados O INEC logo no 2º trimestre apresentou cerca de 5 seguido por cerca de 9 no 3º Já o PIB saltou de cerca de 2 para 5 entre o 4º e 5º trimestres As demais variáveis oscilaram entre cerca de 3 e 4 tendo atingido em alguns momentos uma relevância próxima aos 6 Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 897 49 11 00 13 30 3 820 86 13 13 34 34 4 757 75 31 15 65 58 5 760 63 38 45 51 44 6 756 94 34 37 42 38 7 732 91 33 63 45 36 8 718 96 33 71 44 37 75 Tabela 12 Estimativas pontuais das decomposições de variância do erro de previsão da média ponderada dos bancos Fonte elaborado pelo próprio autor Por último nessa etapa da análise a tabela 12 acima mostra as estimativas da decomposição de variância do erro de previsão para o escore de eficiência da média ponderada dos cinco bancos desse estudo É possível notar uma persistência do próprio escore de eficiência na explicação das suas oscilações futuras sendo responsável por cerca de 57 delas Dentre as demais variáveis merece destacarmos o INEC com cerca de 23 com saltos expressivos no 4º 7º e 8º trimestre Lembramos que choques positivos nessa variável afetaram positivamente o escore de eficiência nos primeiros trimestres mas negativamente nos últimos As outras variáveis oscilaram entre cerca de 3 e 6 43 Previsão Conforme descrito acima na seção 3 após as Análises Estruturais utilizamos o modelo VAR especificado rodando somente até o 4º trimestre de 2018 para prever os dados do último ano os quatro trimestres de 2019 ou seja uma pseudoprevisão fora da amostra e comparamos com uma autorregressão AR com quatro defasagens para os mesmos períodos Os testes foram realizados para as oito amostras nos três critérios especificados e conforme Quadro 2 abaixo Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 920 72 00 01 05 01 3 872 71 20 02 13 22 4 772 122 24 41 22 19 5 765 133 17 35 34 16 6 735 147 28 31 31 28 7 649 186 46 28 47 44 8 565 234 62 31 54 54 76 Quadro 2 Resultado dos testes DieboldMariano para os critérios MSE MAE e MAPE Fonte elaborado pelo próprio autor Nota os resultados em verde identificam a melhor previsão Os resultados acima mostram que os modelos AR apresentam menor erro de previsão para as amostras do Itaú Banco do Brasil Bradesco Caixa e Santander Para a média ponderada dos bancos privados somente no critério MSE o VAR apresentou melhor resultado Já nas médias ponderadas dos bancos públicos e na média ponderada geral o VAR apresentou melhores resultados nos três critérios Apesar de esperarmos que os resultados do modelo VAR previssem melhor os escores de eficiência como o resultado das previsões são extrapolados para um único ano ou seja quatro trimestres que teve suas diversas idiossincrasias com os parâmetros do modelo mudando constantemente regulação ciclo econômico choques políticos etc é natural termos resultados diferentes das expectativas Portanto conforme Clements e Hendry 2002 afirmaram os principais problemas que afligem as previsões econômicas surgem das coisas que não sabemos que não sabemos e dessas as mudanças em termos determinísticos podem ser as mais perniciosas p 16 Critério Modelo Itaú Banco do Brasil Bradesco Caixa VAR 000026 000172 000212 000050 AR 000011 000021 000041 000033 Diferença 000015 000151 000171 000017 VAR 001271 003275 003862 001918 AR 000940 001087 001811 001447 Diferença 000331 002188 002051 000471 VAR 001793 004613 005157 002576 AR 001312 001530 002415 001950 Diferença 000481 003084 002742 000626 Critério Modelo Santander Privados Públicos Geral VAR 000140 000010 000009 000004 AR 000022 000013 000020 000005 Diferença 000118 000003 000011 000000 VAR 002966 000815 000617 000462 AR 001365 000814 000930 000612 Diferença 001601 000000 000313 000150 VAR 004099 001119 000831 000639 AR 001885 001103 001282 000840 Diferença 002214 000016 000451 000201 MSE MAE MAPE MSE MAE MAPE 77 tradução nossa ou seja previsões foradajanela devem ser analisados com cautela para avaliar se um modelo é melhor ou pior 44 Discussão dos resultados à luz das hipóteses Para uma discussão mais abrangente e profunda na sequência compararemos cada hipótese da pesquisa com os resultados encontrados nas Análises Estruturais causalidade de Granger nas funções impulsoresposta e na decomposição da variância dos erros de previsão dos modelos VAR das oito amostras abordadas H1 Expectativas futuras eficiência bancária A hipótese H1 pressupunha que indicadores prospectivos que pudessem de alguma forma indicar expectativas futuras representado nesse estudo pelo INEC possuem uma relação positiva com a eficiência de custos das instituições financeiras Pela análise da causalidade de Granger pudemos constatar que na maioria das amostras o INEC Grangercausa a Eficiência de Custos ou seja os valores passados do INEC ajudam a prever os valores da Eficiência de Custos Outra evidência encontrada para a maioria dos casos foi o efeito feedback ou seja a Eficiência de Custos também Grangercausa o INEC Esses resultados corroboram com os estudos de Beccalli Casu e Girardone 2006 Fiordelisi MarquesIbanez e Molyneux 2011 e Chen Higgins e Mason 2005 Já pela análise das funções impulsoresposta foi possível observar que o INEC afeta os movimentos futuros dos escores de eficiência porém observamos que esse efeito é diferente entre as instituições financeiras De forma sucinta o efeito é positivo para Itaú com efeito subindo gradativamente após o choque até seu ápice no 4º trimestre de 40 bps até 30 bps no 8º e para o Bradesco com maior efeito no 2º e 3º trimestres na casa de 50 bps depois decrescendo Esses efeitos ratificam os três estudos citados Contudo encontramos resultados negativos para Banco do Brasil com efeito decrescente até atingir o 8º trimestre com 110 bps Santander também com aumento gradativo do efeito 78 negativo até o último trimestre com 200 bps e a média ponderada geral que até começou com efeito positivo mas foi decrescendo ao longo dos trimestres até atingir 50 bps no 8º trimestre Para Caixa e as médias ponderadas dos privados e públicos encontramos resultados positivos nos primeiros trimestres e negativos nos demais os quais não podemos discordar das referências pois eles utilizaram metodologias diferentes que não nos permitem essa comparação dos efeitos de um choque ao longo de um período subsequente Vale destacar a Caixa que apresentou um efeito que é decrescente chegando até 50 bps no último trimestre A tabela 13 abaixo mostra esses efeitos trimestre a trimestre em bps por amostra uma releitura dos gráficos de função impulsoresposta apresentados acima mas que facilitarão a visualização dos efeitos Vale destacar como efeito de conceito e interpretação dos dados que um efeito de 100 bps significa que um choque positivo de 1 desvio padrão na variável mencionada melhora o escore de eficiência de custos em 1 ponto percentual pp ou seja seria possível produzir os mesmos resultados com 1 menos custos totais Tabela 13 Efeito das Funções Impulsoresposta dos choques do INEC à eficiência de custos de todas as amostras Fonte elaborado pelo próprio autor Nota os números apresentados estão em bps De forma complementar a análise da decomposição da variância dos erros de previsão mostrou que choques no INEC são os principais responsáveis pela variância do erro de previsão dos escores de eficiência de cada banco com exceção é claro do próprio escore de eficiência reforçando a relevância do mesmo Destaque para o Santander e a média ponderada geral dos bancos que tiveram cerca de 25 de representatividade na variância Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 8 5 28 9 15 9 13 8 2 16 4 46 25 42 16 26 9 3 20 17 47 24 88 8 30 1 4 21 38 44 15 117 7 21 8 5 26 58 41 12 136 5 2 16 6 38 73 46 21 167 5 0 27 7 29 99 22 35 200 8 9 41 8 33 114 2 52 200 21 11 47 79 H2 Riscopaís eficiência bancária A hipótese H2 era de que a medida de riscopaís representada nesse estudo pelo Risco Brasil possuem uma relação negativa com a eficiência de custos das instituições financeiras analisadas A análise da causalidade de Granger nos mostrou que não há um consenso entre as diferentes amostras analisadas sendo que para Itaú Bradesco Caixa e a média ponderada dos privados o Risco Brasil Grangercausa a eficiência de custos mas para as outras quatro amostras essa afirmação é falsa Contudo de forma unânime a Eficiência de Custos não Grangercausa o Risco Brasil Na interpretação das análises das funções impulsoresposta também não há um resultado similar nas oito amostras De modo geral com exceção de um ou outro trimestre e do Santander os resultados não foram estatisticamente diferentes de zero O caso do Santander cujo efeito nos choques do Risco Brasil causaram um efeito positivo na Eficiência de Custos com efeito crescente chegando até 156 bps no último trimestre contraria as afirmações de Tamadonnejad et al 2013 Borges 2012 e AlGasaymeh 2016 A tabela 14 abaixo facilita a interpretação dos dados e ressalta a discrepância dos efeitos no Santander em relação às demais amostras Tabela 14 Efeito das Funções Impulsoresposta dos choques do Risco Brasil à eficiência de custos de todas as amostras Fonte elaborado pelo próprio autor Nota os números apresentados estão em bps A análise da decomposição da variância dos erros de previsão mostrou que choques no Risco Brasil são outros importantes responsáveis pela variância do erro de previsão dos escores de eficiência de cada banco Destacamos o Itaú o Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 10 7 9 1 8 7 6 0 2 3 1 5 20 32 6 10 4 3 7 11 12 25 34 2 0 7 4 12 25 14 19 44 5 11 5 5 7 27 22 10 78 3 15 0 6 15 25 10 2 83 11 13 7 7 12 18 2 11 105 5 10 1 8 6 15 19 24 156 25 8 19 80 Santander e a média ponderada dos bancos privados todos com mais de 10 de representatividade na variância H3 Atividade econômica eficiência bancária A hipótese H3 era de que a atividade econômica representada no nosso estudo pelo PIB afeta positivamente a eficiência de custos das amostras analisadas Para essa hipótese diferentemente das H1 e H2 não havia consenso entre as referências bibliográficas citadas A análise da causalidade de Granger como era esperado nos mostrou que para todas as amostras o PIB Grangercausa a eficiência de custos e que o efeito feedback é verdadeiro na maioria dos casos exceção ao Banco do Brasil e à média ponderada geral o que independe do sinal desse efeito corrobora com a opinião da maioria dos estudos mencionados nessa hipótese e contraria o estudo de Chen et al 2018 que não encontrou um impacto significativo do PIB sobre a eficiência Já a interpretação das análises das funções impulsoresposta nos mostrou uma divisão clara entre os resultados Para o Itaú o Bradesco e para a média ponderada dos privados em linhas gerais os choques no PIB causaram um efeito negativo na eficiência de custos como posto nos estudos de Chortareas Girardone e Ventouri 2012 Sufian 2009 e Kallel Hamad e Triki 2019 Os resultados do Itaú apesar de estatisticamente diferentes de zero e negativos representaram um efeito absoluto no escore de eficiência pouco significativo Já para o Bradesco nos últimos trimestres o efeito se aproximou de 100 bps No caso dos dois bancos públicos e sua média ponderada e para o Santander apesar de alguns trimestres onde os resultados foram indistinguíveis de zero notamos indicações de um efeito positivo na eficiência de custos Esses resultados ratificam as considerações de Tan e Anchor 2017 AlGasaymeh 2016 e Primo et al 2013 que afirmaram que os bancos são mais eficientes durante os períodos de crescimento econômico O efeito no escore de eficiência do Santander merece ser destacado pois chegou a 155 bps no 8º trimestre E como era esperado depois desse resultado divergente a média ponderada geral foi indistinguível de zero em todos os trimestres A tabela 15 abaixo transcreve visualmente esses efeitos Apesar de termos encontrado evidências que 81 ratificam todos os estudos abordados efeito positivo negativo e sem resultado significativo boa parte das razões desses efeitos descritas nesses estudos não nos parecem adequadas pois seriam efeitos gerais que afetariam todo mercado independentemente de ser um banco público ou privado como melhora na qualidade dos tomadores e aumento na demanda de crédito Com certeza existem fatores e ações específicas itens de gestão e governança adotadas por cada uma dessas Organizações que causaram esses resultados Tabela 15 Efeito das Funções Impulsoresposta dos choques do PIB à eficiência de custos de todas as amostras Fonte elaborado pelo próprio autor Nota os números apresentados estão em bps Como último elemento de análise dessa hipótese a decomposição da variância dos erros de previsão mostrou que choques no PIB foram responsáveis em seis amostras por menos de 10 da variância do erro de previsão dos escores de eficiência sendo o Bradesco e o Santander com cerca de 17 e 11 respectivamente as exceções H4 Inflação eficiência bancária A penúltima hipótese a H4 era de que a inflação representada pelo IGPM afeta negativamente a eficiência de custos das instituições e médias analisadas Da mesma forma que a anterior para essa hipótese também não havia consenso entre as referências A primeira análise da causalidade de Granger mostrou que para os três bancos privados e para as três médias ponderadas o IGPM Grangercausa a eficiência de custos mas a eficiência de custos não Grangercausa o IGPM exceção feita à Caixa e à média ponderada dos públicos Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 8 8 0 4 14 6 1 1 2 17 1 27 8 23 1 7 0 3 23 9 46 3 19 7 2 6 4 14 3 41 14 41 8 18 2 5 20 1 56 10 71 13 20 4 6 19 8 83 15 90 15 36 2 7 20 9 96 24 115 15 45 6 8 5 29 80 44 155 11 64 3 82 A interpretação das análises das funções impulsoresposta nos mostrou que com exceção do Itaú os choques no IGPM na maioria dos trimestres causaram efeitos indistinguíveis de zero No caso do Itaú os choques causam um efeito negativo na eficiência de custos como era nossa hipótese inicial alinhado com o constatado por Primo et al 2013 e AlGasaymeh 2016 porém com pouco efeito absoluto no escore chegando no ápice no 5º trimestre com 37 bps conforme tabela 16 abaixo Tabela 16 Efeito das Funções Impulsoresposta dos choques do IGPM à eficiência de custos de todas as amostras Fonte elaborado pelo próprio autor Nota os números apresentados estão em bps Na decomposição da variância dos erros de previsão a variável IGPM foi a que mostrou menos relevância na variância do erro de previsão da eficiência de custos sendo menor que 9 em todas as amostras H5 Solvência eficiência bancária A quinta e última hipótese a H5 pressupunha que a solvência que no nosso estudo foi abordada como o índice de Basileia era relacionada negativamente com a eficiência de custos e assim como as duas últimas hipóteses não havia um consenso na literatura considerada com estudos e indicações confirmando ou negando essa afirmação As análises da causalidade de Granger mostraram que somente para Caixa Santander e média ponderada dos públicos Basileia Grangercausa eficiência de custos e nas demais amostras a maioria não Grangercausa Já no efeito contrário de forma unânime a eficiência de custos Grangercausa Basileia Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 9 2 13 6 10 2 7 2 2 20 2 7 2 5 7 3 1 3 23 2 6 3 4 9 16 4 4 31 5 14 10 14 17 22 10 5 37 8 15 10 14 12 20 8 6 33 5 10 4 27 1 14 1 7 34 1 4 0 37 5 11 5 8 31 2 8 4 56 12 14 10 83 Esses resultados contrariam as evidências apresentadas nos estudos relacionados a essa hipótese Nas análises das funções impulsoresposta pudemos observar que os choques no índice de basileia apresentaram um efeito negativo na eficiência de custos para o Itaú Santander e média ponderada dos privados que foi o que pressupunha a nossa hipótese e que afirmou os estudos de Chortareas Girardone e Ventouri 2012 Sufian 2009 e Zhang et al 2013 No caso do Santander os efeitos absolutos relativos foram bem representativos com 150 bps no último trimestre Já para a Caixa e a média ponderada dos bancos públicos o efeito foi positivo alinhado às conclusões de Fiordelisi Marques Ibanez e Molyneux 2011 Athanasoglou Brissimis e Delis 2008 Tecles e Tabak 2010 e Mendonça 2018 No caso da Caixa os últimos trimestres também tiveram efeitos de destaque chegando a 90 bps nos últimos 2 trimestres Para as outras três amostras os resultados foram indistinguíveis de zero Tabela 17 Efeito das Funções Impulsoresposta dos choques de Basileia à eficiência de custos de todas as amostras Fonte elaborado pelo próprio autor Nota os números apresentados estão em bps Na decomposição da variância dos erros de previsão a variável Basileia foi extremamente relevante para a Caixa com cerca de 32 da variância do erro de previsão da eficiência de custos mas para as demais amostras teve valores bem mais inferiores com destaque para o Santander com cerca de 13 e o Itaú com cerca de 8 Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 3 7 6 13 25 5 10 1 2 13 8 2 65 51 6 16 5 3 19 4 2 78 73 9 28 5 4 19 4 1 65 93 10 33 2 5 33 7 1 64 114 9 37 3 6 40 11 6 68 130 11 37 4 7 33 12 3 90 139 11 41 10 8 26 14 3 88 150 10 53 7 84 Depois de apresentado e os resultados as análises e as interpretações das hipóteses de pesquisa apresentamos um resumo sintético comparando a relação esperada e os resultados obtidos para cada hipótese no Quadro 2 abaixo Quadro 3 Síntese comparativa entre os sinais esperados e os resultados apurados Fonte elaborado pelo próprio autor Nota os campos indicados com os dois sinais significam que ambos os efeitos ocorreram ao longo dos trimestres de acordo com a respectiva ordem Os campos identificados como indist significam que os resultados foram predominantemente indistinguíveis de zero 45 Teste de Robustez 451 Utilização de variáveis alternativas Uma série de verificações de robustez para confirmar a validade dos modelos e resultados empíricos relatados acima foram feitas Itaú Banco do Brasil Bradesco Caixa H1 Expectativas futuras INEC e H2 Medida de riscopaís Risco Brasil indist indist H3 Atividade econômica PIB indist indist H4 Inflação IGPM indist indist indist H5 Solvência Basileia indist indist Santander Privados Públicos Geral H1 Expectativas futuras INEC e e H2 Medida de riscopaís Risco Brasil e indist indist H3 Atividade econômica PIB indist H4 Inflação IGPM indist e H5 Solvência Basileia indist Hipótese Hipótese Tema Variável Sinal Esperado Sinal Apurado nos Testes Empíricos Sinal Apurado nos Testes Empíricos Sinal Esperado Variável Tema 85 Primeiramente outras variáveis foram testadas tanto no aspecto de mercado quanto no específico das instituições financeiras Seguindo a metodologia descrito na seção 3 testamos a estacionaridade de todas as variáveis levantadas e não foi possível estacionarizar nem por primeira diferença por aplicação de logaritmo ou por ambos somente a quantidade de agências e postos de atendimento de todas as amostras levantadas logo essa variável foi descartada As demais variáveis testadas e não utilizadas foram descartadas na checagem do modelo com destaque para a estabilidade do modelo VAR que nos fez descartar por conta de resultados desfavoráveis em pelo menos uma das amostras modelos com a variável de primeira diferença do logaritmo da quantidade de agências12 primeira diferença do Retorno sobre o Capital ambos casos de variáveis específicas ou Taxa de Juros variável doméstica 452 Variação na ordem das variáveis Numa segunda etapa também descrita na mesma seção testamos especificações alternativas com inversão ou trocas na ordenação causal dado que essa é uma etapa muito relevante e crítica para nossa estrutura do estudo conforme descrito na seção 3 Como quanto mais correlacionado forem os resíduos mais sensível poderão ser as respostas a diferentes ordenações das variáveis do sistema COLOMBO LONCAN e CALDEIRA 2018 e até mesmo como forma de testar a sensibilidade dos resultados a essas alterações decidimos supor que ao invés das variáveis que descrevemos como preditoras de expectativas futuras e indicadores de riscopaís afetaram as decisões de produção as expectativas futuras e o riscopaís seriam afetados pelos resultados do produto interno Testamos portanto a ordenação IGPM PIB Risco Brasil e INEC mantendo obviamente a eficiência de custos como variável dependente e Basileia variável específica de cada instituição como última Comparamos os resultados das análises das funções impulsoresposta deste exercício de robustez com os resultados do modelo final escolhido no mesmo 12 A variável quantidade de agências e postos de atendimento mencionada acima corresponde a soma dessas duas informações disponíveis na fonte de dados mencionada na seção específica 86 formato apresentado acima conforme tabelas 18 19 20 e 21 abaixo para cada variável que teve sua posição alterada Obviamente as variáveis eficiência de custos e Basileia por não sofrerem alteração na posição não tiveram diferença nos seus resultados Tabela 18 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável INEC Fonte elaborado pelo próprio autor Nota os números apresentados estão em pp Tabela 19 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável Risco Brasil Fonte elaborado pelo próprio autor Nota os números apresentados estão em pp Tabela 20 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável PIB Fonte elaborado pelo próprio autor Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 01 01 00 00 00 00 01 00 2 01 00 01 01 00 00 01 00 3 01 01 01 01 01 00 00 00 4 01 02 01 01 01 00 01 00 5 02 02 01 00 00 00 01 00 6 02 02 01 01 01 00 01 01 7 02 02 01 00 01 01 01 00 8 01 02 01 00 01 01 01 00 Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 01 01 01 00 01 00 00 00 2 01 00 02 00 04 01 00 00 3 01 01 03 00 05 01 01 00 4 01 01 03 00 07 01 01 00 5 01 01 03 01 12 01 01 00 6 01 00 05 01 15 01 00 00 7 01 01 05 02 18 01 00 01 8 00 01 03 04 23 01 01 02 Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 00 01 00 00 01 01 00 00 2 00 00 00 00 01 01 00 00 3 01 01 01 00 03 00 00 00 4 01 02 01 00 03 01 01 01 5 01 02 02 01 03 01 01 01 6 00 02 01 00 03 00 00 01 7 01 02 01 00 03 00 00 01 8 00 01 02 01 02 02 00 01 87 Nota os números apresentados estão em pp Tabela 21 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável IGPM Fonte elaborado pelo próprio autor Nota os números apresentados estão em pp A respeito das Tabelas acima podemos concluir que não há grandes oscilações que mudem o sentido da análise ou seja que indicavam que choques geravam efeito positivo e passaram a gerar negativo ou viceversa A única exceção a essa afirmação foi a interpretação do choque da variável Risco Brasil na amostra do Santander que no modelo selecionada apresentava indícios de que o efeito era positivo mas nesse teste de robustez apresentou evidências de que o efeito seria negativo Contudo interpretamos que esse foi um caso isolado de uma das amostras portanto além da razoabilidade econômica ser mais factível no modelo escolhido esses resultados corroboram para a escolha do modelo e evidenciam que para os dados analisados em linhas gerais nossos resultados são qualitativa semelhantes independentemente de como as variáveis são ordenadas na decomposição de Cholesky da matriz de covariância dos resíduos 453 Eficiência de receita e lucro Na terceira e última etapa reestimamos os mesmos modelos aprovados porém com foco nos escores de eficiência de receita e lucro muito similar ao exercício feito por Fiordelisi MarquesIbanez e Molyneux 2011 A estatística descritiva dos escores de eficiência de receita e lucro estão na Tabela 22 abaixo Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 00 01 02 00 01 00 01 00 2 02 00 02 02 01 00 02 01 3 02 02 02 02 01 00 02 00 4 04 03 02 02 00 01 02 00 5 04 04 02 00 01 00 01 01 6 03 04 01 00 00 01 01 02 7 03 05 01 00 01 01 01 02 8 04 05 02 00 04 00 02 02 88 Tabela 22 Estatística descritiva dos escores de eficiência de receita e lucro Fonte elaborado pelo próprio autor Como primeiro desafio por conta da não estacionaridade disponível no Apêndice C ao invés de usar diretamente o escore de receita e lucro extraímos sua primeira diferença conforme mencionado na subseção 33 Os dados da Matriz de correlação das variáveis de escore de eficiência de custo lucro e receita para cada uma das oito amostras disponível no Apêndice H apresentaram uma média de 017 sendo que a correlação mais alta foi de 070 e a mais baixa de 012 ou seja sugerindo que não há problema de colinearidade abaixo de 08 conforme descreveu Kennedy 2008 Desse modo foi possível rodar os Modelos VAR para as oito amostras considerando um cenário de escore de receita e outro cenário de escore de lucro Para testar a estabilidade desses Modelos VAR para o escore de eficiência de lucro no caso das amostras da Caixa média ponderada do públicos e média ponderada geral o modelo se mostrou instável O mesmo ocorreu também para a Caixa mas nesse caso para o escore de receita Essa instabilidade do modelo conforme mencionado anteriormente nos inviabiliza de analisar os resultados das funções impulsoresposta Em todo caso como exercício de reflexão e Média Desvio Padrão Mínimo Máximo Itaú 0997224 0000044 0997118 0997326 BB 0997299 0000040 0997182 0997367 Bradesco 0997282 0000049 0997167 0997378 Caixa 0997268 0000045 0997123 0997336 Santander 0997283 0000076 0997149 0997534 Privados 0997257 0000042 0997161 0997359 Públicos 0997282 0000038 0997181 0997351 Geral 0997267 0000024 0997209 0997313 Itaú 0995559 0000029 0995506 0995607 BB 0995592 0000038 0995521 0995676 Bradesco 0995585 0000041 0995496 0995693 Caixa 0995589 0000053 0995459 0995736 Santander 0995597 0000062 0995489 0995709 Privados 0995576 0000026 0995527 0995628 Públicos 0995591 0000034 0995525 0995686 Geral 0995582 0000023 0995544 0995635 Variável Escore de Eficiência de receita Escore de Eficiência de lucro 89 alinhamento quanto aos resultados do escore de eficiência de custos comparamos na Tabela 23 os resultados dos testes de causalidade de Granger Tabela 23 Probabilidades do teste de causalidade de Granger para Eficiência de Receita Lucro e Custos continua Equação a Excluído b Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral Eficiência de Receita INEC 0020 0307 0732 0019 0051 0694 0206 0110 Eficiência de Lucro INEC 0263 0107 0006 0014 0263 0835 0003 0019 Eficiência de Custo INEC 0005 0487 0000 0000 0000 0000 0104 0000 INEC Eficiência de Receita 0119 0046 0765 0050 0002 0393 0038 0210 INEC Eficiência de Lucro 0000 0001 0002 0052 0497 0050 0022 0013 INEC Eficiência de Custo 0003 0032 0285 0017 0000 0000 0146 0003 Eficiência de Receita Risco Brasil 0000 0575 0033 0798 0000 0235 0898 0008 Eficiência de Lucro Risco Brasil 0597 0030 0012 0019 0434 0004 0029 0060 Eficiência de Custo Risco Brasil 0014 0876 0000 0041 0500 0001 0561 0053 Risco Brasil Eficiência de Receita 0000 0000 0161 0000 0000 0000 0000 0000 Risco Brasil Eficiência de Lucro 0272 0000 0281 0374 0000 0001 0074 0001 Risco Brasil Eficiência de Custo 0671 0105 0078 0288 0060 0118 0538 0782 Eficiência de Receita PIB 0000 0025 0010 0014 0024 0270 0263 0000 Eficiência de Lucro PIB 0020 0007 0000 0546 0144 0020 0449 0149 Eficiência de Custo PIB 0000 0000 0000 0000 0000 0000 0000 0000 PIB Eficiência de Receita 0992 0020 0109 0000 0055 0241 0037 0277 PIB Eficiência de Lucro 0650 0000 0000 0372 0000 0025 0000 0006 PIB Eficiência de Custo 0000 0539 0000 0004 0006 0002 0007 0128 Eficiência de Receita IGPM 0001 0274 0754 0922 0051 0502 0743 0558 Eficiência de Lucro IGPM 0572 0296 0000 0000 0274 0640 0014 0010 Eficiência de Custo IGPM 0013 0916 0009 0091 0000 0001 0023 0000 IGPM Eficiência de Receita 0000 0000 0487 0115 0001 0007 0000 0124 IGPM Eficiência de Lucro 0057 0000 0815 0000 0000 0850 0003 0343 IGPM Eficiência de Custo 0570 0353 0905 0000 0976 0561 0000 0108 90 Tabela 23 Probabilidades do teste de causalidade de Granger para Eficiência de Receita Lucro e Custos conclusão Fonte Elaborado pelo próprio autor Nota Os números expressos na tabela representam a probabilidade de aceitar a hipótese nula sendo que a fonte verde representa os casos em que rejeitamos e em vermelho os casos em que aceitamos Consideramos o nível de significância de 5 Em linhas gerais apesar de resultados diferentes para cada uma das oito amostras e para os três tipos de escore de eficiência O que temos de grande divergência entre os resultados dos três escores são o PIB Grangercausa a eficiência de custos que foi unânime para todas as amostras mas não se mostrou verdadeiro para todas no caso dos escores de lucro e receita e para a Eficiência de Custos que Grangercausava Basileia mas que não foi verdade para eficiência de receita e de lucro Equação a Excluído b Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral Eficiência de Receita Basileia 0000 0000 0276 0001 0000 0091 0299 0001 Eficiência de Lucro Basileia 0022 0056 0000 0397 0559 0000 0195 0088 Eficiência de Custo Basileia 0522 0660 0287 0000 0001 0193 0006 0714 Basileia Eficiência de Receita 0005 0985 0000 0000 0000 0152 0013 0067 Basileia Eficiência de Lucro 0001 0001 0249 0531 0002 0009 0501 0025 Basileia Eficiência de Custo 0000 0000 0002 0046 0005 0000 0002 0011 Eficiência de Receita Todas 0000 0000 0000 0000 0000 0000 0025 0000 Eficiência de Lucro Todas 0002 0002 0000 0000 0085 0000 0000 0000 Eficiência de Custo Todas 0000 0000 0000 0000 0000 0000 0000 0000 91 5 CONCLUSÃO Este estudo teve por objetivo quantificar através de suas séries temporais as variáveis que continham informações sobre as flutuações futuras da eficiência bancária das principais instituições financeiras do Brasil Primeiramente foi definida uma metodologia para apuração dos escores de eficiência a análise de fronteira estocástica Para representar esses escores de eficiência bancária optamos por utilizar a eficiência de custos sendo que realizamos testes de robustez com os demais escores lucro e receita e não obtivemos conclusões gerais distintas Na próxima etapa da metodologia optamos pelo Modelo VAR para analisar as interrelações entre diversas variáveis e com isso estimar o efeito de um choque nessa variável sobre a eficiência bancária Foram realizados alguns testes e concluise que o modelo apresentado era consistente e as hipóteses que se buscavam responder seriam endereçadas Para esse exercício foram selecionadas algumas variáveis específicas de cada instituição que serviriam para a definição dos escores de eficiência e outras que serviriam como variáveis endógenas e que juntamente com algumas variáveis macroeconômicas seriam possíveis candidatas a explicar a evolução da eficiência bancária Os dados capturados são trimestrais entre 2009 e 2019 e além dos cinco maiores bancos atuando no Brasil Itaú Banco do Brasil Bradesco Caixa e Santander foram definidas três médias ponderadas com base nessas cinco instituições bancos privados bancos públicos e geral totalizando oito amostras que permearam todo estudo Os resultados apurados através na análise estrutural do Modelo VAR evidenciaram que as expectativas futuras o PIB e a inflação Grangercausam a eficiência de custos para a maioria das amostras como afirma a maioria da literatura e nossas hipóteses de pesquisa Porém a análises não se mostrou verdadeira para solvência e para o riscopaís contrariando nossas hipóteses e as referências bibliográficas citadas Já o efeito contrário da eficiência de custos Grangercausar as variáveis se mostrou verdadeiro na maioria das amostras para as expectativas futuras o PIB e a solvência 92 De forma complementar a análise das funções impulsoreposta e as decomposições de variância do erro de previsão nos mostraram que os resultados entre as instituições financeiras são divergentes sendo que não há um padrão entre os resultados de cada variável para cada amostra ou seja o escore de eficiência de custos de cada instituição financeira se comporta de maneira distinta a cada choque nas variáveis analisadas As constatações das análises estruturais do Modelo VAR descritas acima complementam a literatura abordando a reflexão de que não há um padrão único entre as variáveis analisadas e seus efeitos na eficiência bancária dada as características específicas de cada uma das instituições financeiras estudadas Adicionalmente o modelo também foi testado com pseudoprevisões fora da amostra que não evidenciaram melhores critérios de previsão do que uma autorregressão simples Sobre as limitações do estudo podemos citar o período relativamente curto para a metodologia selecionada 44 trimestres e o fato de que o estudo só abrangeu as cinco maiores instituições financeiras do Brasil o que inviabilizaria a ampliação das evidências empíricas às instituições medianas pequenas ou de nichos de mercado Como sugestões para estudos futuros podemos imaginar além dos pontos mencionados nas limitações desse estudo a utilização de amostras de bancos de outros países que poderiam trazer uma diversidade relevante às conclusões e a eventual utilização de outras metodologias para definição dos escores de eficiência bancário como forma de comparação dos resultados das análises de interrelação das variáveis do modelo 93 6 REFERÊNCIAS ABREU Emmanuel Sousa de KIMURA Herbert SOBREIRO Vinicius Amorim What Is Going On with Studies on Banking Efficiency Research in International Business and Finance v 47 p 195219 2019 AHMAD Nisar et al Banking Sector Performance Profitability and Efficiency a citationbased systematic literature review Journal of Economic Surveys v 0 n 0 p 134 2019 AIELLO Francesco BONANNO Graziella Efficiency in banking a meta regression analysis International Review of Applied Economics v 30 p 112 149 2015 AIELLO Francesco BONANNO Graziella On the sources of heterogeneity in banking efficiency literature Journal of Economic Surveys v 31 n 1 p 194 225 2018 ALGASAYMEH Anwar Bank efficiency determinant Evidence from the gulf cooperation council countries Research in International Business and Finance v 38 p 214223 2016 ALLEN Linda RAI Anoop Operational efficiency in banking An international comparison Journal of Banking Finance v 20 p 655672 1996 ANDRIES Alin Marius COCRIS Vasile A Comparative Analysis of the Efficiency of Romanian Banks Romanian Journal of Economic Forecasting v 13 p 5475 2010 ATHANASOGLOU Panayiotis P BRISSIMIS Sophocles N DELIS Matthaios D Bankspecific industryspecific and macroeconomic determinants of bank profitability Journal of International Financial Markets Institutions Money v 18 p 121136 2008 BANCO CENTRAL DO BRASIL Resolução nº 4553 de 30 de janeiro de 2017 Estabelece a segmentação do conjunto das instituições financeiras e demais instituições autorizadas a funcionar pelo Banco Central do Brasil para fins de aplicação proporcional da regulação prudencial Diário Oficial da União Brasília 31 jan 2017 Seção 1 p 25 94 BANCO CENTRAL DO BRASIL IFdata Disponível em httpswww3bcbgovbrifdata Acesso em 04 abr 2020 BARROS C et al Efficiency in banking of developing countries with the same cultural background A novel distance function frontier model Journal of Economic Studies v 45 n 3 p 638659 2018 BATTESE G E COELLI T J A Model for Technical Inefficiency Effects in a Stochastic Frontier Production Function for Panel Data Empirical Economics v 20 p 325332 1995 BAUM Christopher F VAR SVAR and VECM models Applied Econometrics Notes Boston College 2013 BAUM Christopher F Statistical Software Components S433001 DMARIANO Stata module to calculate DieboldMariano comparison of forecast accuracy Boston College Department of Economics 2003 revisado em 25 abr 2011 Disponível em httpfmwwwbcedurepecbocodeddmarianohlp Acesso em 19 set 2020 BECCALLI Elena CASU Barbara GIRARDONE Claudia Efficiency and Stock Performance in European Banking Journal of Business Finance Accounting v 33 n 1 p 245262 2006 BELOTTI Federico et al Stochastic frontier analysis using Stata The Stata Journal v 13 n 4 p 719758 2013 BERGER Allen N DEYOUNG Robert Problem Loans and Cost Efficiency in Commercial Banks Journal of Banking Finance v 21 p 849870 1997 BERGER Allen N HANNAN Timothy H The Efficiency Cost of Market Power in the Banking Industry A Test of the Quiet Life and Related Hypotheses The Review of Economics and Statistics v 80 n 3 p 454465 1998 BERGER Allen N HUMPHREY David B Measurement and Efficiency Issues in Commercial Banking In GRILICHES Zvi Ed Output Measurement in the Service Sectors Chicago University of Chicago Press National Bureau of Economic Research 1992 cap 7 p 245300 95 BERGER Allen N HUMPHREY David B Efficiency of financial institutions International survey and directions for future research European Journal of Operational Research v 98 p 175212 1997 BERGER Allen N MESTER Loretta J Inside the black box What explains differences in the efficiencies of financial institutions Journal of Banking Finance v 21 p 895947 1997 BIAGE Milton CORREA Vanessa Petrelli NEDER Henrique Dandas Risco país fluxos de capitais e determinação da taxa de juros no Brasil Uma análise de impactos por meio da metodologia VEC Revista Economia v 9 n1 p 63 113 2008 BORGES Pedro Raposo Bank performance during the European Debt Crisis a DEA approach 2012 52 f Dissertação Mestrado em Finanças Instituto Superior de Economia e Gestão da Universidade Técnica de Lisboa Lisboa 2012 BUENO R L S Econometria de séries temporais 2 ed São Paulo Cengage Learning 2011 325 p CARVALLO Oscar KASMAN Adnan Cost efficiency in the Latin American and Caribbean banking systems Journal of International Financial Markets Institutions Money v 15 p 5572 2005 CASU Barbara GIRARDONE Claudia Testing the relationship between competition and efficiency in banking A panel data analysis Economics Letters v 105 p 134137 2009 CAVALCANTI Marco A F H Identificação de modelos VAR e causalidade Granger uma nota de advertência Economia Aplicada v 14 n 2 p 251260 2010 CAVALCANTI Marco A F H Uma análise econométrica da evolução da indústria de transformação brasileira no período 20022012 IPEA 2013 Nota Técnica Carta de Conjuntura IPEA N18 CHEN KuHsieh et al Is full efficiency achievable An empirical study of the banking industry in Taiwan Contemporary Economic Policy v 37 n 1 p 195 209 2018 96 CHEN YiKai HIGGINS Eric J MASON Joseph R Is Bank Efficiency Cyclical The Relationship Between Economic and Financial Market Conditions and Bank Performance Drexel University working paper 2005 CHORTAREAS Georgios E GIRARDONE Claudia VENTOURI Alexia Bank supervision regulation and efficiency Evidence from the European Union Journal of Financial Stability v 8 p 292302 2012 CHRISTIANO Lawrence J Christopher A Sims and Vector Autoregressions The Scandinavian Journal of Economics v 114 p 10821104 2012 CLEMENTS Michael P HENDRY David F A Companion to Economic Forecasting Malden MA Blackwell Publishing Ltd 2002 597 p CNI CONFEDERAÇÃO NACIONAL DA INDÚSTRIA Metodologia do Índice Nacional de Expectativa do Consumidor INEC Versão 20 Brasília CNI 2019 COELLI Tim RAO D S Prasada BATTESE George E An Introduction to Efficiency and Productivity Analysis 1st ed New York NY Kluwer Academic Publishers 1998 275 p COLOMBO J A LONCAN TR CALDEIRA JF Do foreign portfolio capital flows affect domestic investment Evidence from Brazil International Journal of Finance Economics v 24 p 129 2018 DEYOUNG Robert A diagnostic test for the distributionfree efficiency estimator An example using US commercial bank data European Journal of Operational Research v 98 p 243249 1997 DIEBOLD Francis X MARIANO Roberto S Comparing Predictive Accuracy Journal of Business Economic Statistics v 13 n 3 p 253263 ELDER John KENNEDY Peter E Testing for unit roots what should students be taught Journal of Economic Education v 32 n 2 p 137146 2001 ELYASIANI Elyas MEHDIAN Seyed Efficiency in the commercial banking industry a production frontier approach Applied Economics v 22 p 539551 1990 97 ENDERS Walter Applied Econometrics Time Series 4th ed New York John Wiley and Sons Press 2014 485 p FARREL M J The Measurement of Productive Efficiency Journal of the Royal Statistical Society v 120 n 3 p 253290 1957 FERRIER Garry D LOVELL C A Knox Measuring cost efficiency in banking Econometric and Linear Programming Evidence Journal of Econometrics v 46 p 229245 1990 FIORDELISI Franco MARQUESIBANEZ David MOLYNEUX Efficiency and risk in European banking Journal of Banking Finance v 35 p 13151326 2011 FIORENTINO Elisabetta KARMANN Alexander KOETTER Michael The cost efficiency of German banks a comparison of SFA and DEA Deutsche Bundesbank Series 2 Banking and Financial Studies v 10 2006 Discussion Paper GOMES Matheus da Costa OLIVEIRA Sonia Valle Walter Borges de MATIAS Alberto Borges Eficiência do setor bancário brasileiro no período de 2006 a 2013 bancos domésticos x bancos estrangeiros Nova Economia v 27 n 3 p 641670 2017 GUJARATI D N PORTER D C Econometria básica 5 ed Porto Alegre AMGH 2011 924 p HAVRYLCHYK O Efficiency of the Polish banking industry foreign versus domestic banks Journal of Banking and Finance v 30 n 7 p 19751996 2006 HENRIQUES Iago Cotrim et al Efficiency in the Brazilian banking system using data envelopment analysis Future Business Journal v 4 p 157178 2018 HULJAK Ivan MARTIN Reiner MOCCERO Diego The CostEfficiency and Productivity Growth of Euro Area Banks European Central Bank Working Paper No 2305 2019 98 IPEAData EMBI Risco Basil Disponível em httpwwwipeadatagovbrExibeSerieaspxserid40940moduleM Acesso em 13 jun 2020 IVANOV Ventzislav KILIAN Lutz A Practitioners Guide to Lag Order Selection For VAR Impulse Response Analysis Studies in Nonlinear Dynamics Econometrics v9 n 1 p 136 2005 KALLEL Hella HAMAD Salah Ben TRIKI Mohamed Modeling the efficiency of Tunisian and Moroccan banks using the SFA approach International Journal of Productivity and Performance Management v 68 n 5 p 879908 2019 KENNEDY Peter A Guide to Econometrics 6th ed Malden Blackwell Publishing 2008 585 p KIM Hyeongwoo Generalized Impulse Response Analysis General or Extreme EconoQuantum v 10 n2 p135141 2013 KUMBHAKAR S C LOVELL C A K Stochastic Frontier Analysis Cambridge Cambridge University Press 2010 LEIBENSTEIN H Allocative Efficiency vs XEfficiency The American Economic Review v 56 n 3 p 392415 1966 LÜTKEPOHL Helmut New Introduction to Multiple Time Series Analysis Berlim SpringerVerlag 2005 MATTHEWS Kent Banking efficiency in emerging market economies Cardiff Economics Working Papers No E201012 2010 MENDONÇA Douglas José et al Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil Revista Capital Científico Eletrônica RCCe v 16 n 3 p 521 2018 NAKANE Márcio I WEINTRAUB Daniela B Bank privatization and productivity Evidence for Brazil Journal of Banking Finance v 29 p 2259 2259 2005 99 PRIMO Uverlan Rodrigues et al Determinantes da Rentabilidade Bancária no Brasil BASE Revista de Administração e Contabilidade da Unisinos v 10 n 4 p 308323 2013 SHAMSHUR Anastasiya WEILL Laurent Does bank efficiency influence the cost of credit Journal of Banking and Finance v 105 p 6273 2019 SILVA Thiago Christiano et al A comparison of DEA and SFA using micro and macrolevel perspectives Efficiency of Chinese local banks Physica A Statistical Mechanics and its Applications v 469 p 216223 2017 SIMS C A Macroeconomics and Reality Econometrica v 48 n 1 p 148 1980 STOCK James H WATSON Mark W Vector Autoregressions Journal of Economic Perspectives v 15 n 4 p 102115 2001 SUFIAN Fadzlan Determinants of bank efficiency during unstable macroeconomic environment Empirical evidence from Malaysia Research in International Business and Finance v 23 p 5477 2009 TAMADONNEJAD A et al Political Stability Country Risk and Bank Efficiency in East Asian Countries Prosiding Perkem VIII v 2 p 616628 2013 TAN Yong ANCHOR John The impacts of risktaking behaviour and competition on technical efficiency Evidence from the Chinese banking industry Research in International Business and Finance v 41 p 90104 2017 TECLES Patricia Langsch TABAK Benjamin M Determinants of bank efficiency The case of Brazil European Journal of Operational Research v 207 p 15871598 2010 WILLIAMS Jonathan Determining management behaviour in European banking Journal of Banking Finance v 28 p 24272460 2004 WOOLDRIDGE Jeffrey M Introdução à Econometria uma abordagem moderna Tradução Priscilla Rodrigues da Silva Lopes e Livia Marina Koeppl Revisão técnica Heloisa Pinna Bernardo 6a ed São Paulo SP Cengage Learning 2016 848 p 100 ZHANG Jianhua et al Market concentration risktaking and bank performance Evidence from emerging economies International Review of Financial Analysis v 30 p 149157 2013 101 APÊNDICE 102 APÊNDICE A ESCORES DE EFICIÊNCIA DE CUSTO E DEMAIS VARIÁVEIS Figura 10 Escores de eficiência de custos e demais variáveis no tempo Fonte Elaborado pelo próprio autor 103 APÊNDICE B MATRIZ DE CORRELAÇÃO Tabela 24 Matriz de correlação das variáveis dos modelos VAR Fonte Elaborado pelo próprio autor Itaú BB Bradesco Caixa Santander Privados Públicos Geral Itaú 1000 BB 0671 1000 Bradesco 0708 0600 1000 Caixa 0465 0612 0432 1000 Santander 0616 0418 0432 0230 1000 Privados 0921 0673 0865 0455 0761 1000 Públicos 0582 0806 0532 0962 0317 0574 1000 Geral 0824 0841 0761 0830 0578 0857 0914 1000 0028 0358 0043 0419 0404 0097 0438 0228 1000 0188 0366 0254 0364 0109 0223 0400 0362 0309 1000 0557 0334 0426 0043 0392 0542 0084 0322 0192 0050 0063 0021 0036 0054 0095 0012 0048 0036 0168 0342 Itaú 0059 0128 0106 0050 0143 0116 0007 0052 0024 0133 BB 0063 0167 0069 0073 0110 0038 0112 0053 0045 0146 Bradesco 0136 0021 0029 0019 0110 0104 0007 0056 0160 0139 Caixa 0036 0170 0112 0240 0000 0063 0239 0120 0112 0132 Santander 0185 0211 0065 0304 0212 0173 0301 0276 0104 0442 Privados 0149 0123 0063 0020 0089 0118 0058 0095 0068 0139 Públicos 0153 0144 0090 0228 0148 0072 0221 0176 0097 0185 Geral 0190 0165 0001 0133 0143 0125 0157 0162 0100 0196 Itaú BB Bradesco Caixa Santander Privados Públicos Geral 1000 0108 1000 Itaú 0172 0115 1000 BB 0037 0069 0383 1000 Bradesco 0151 0144 0348 0016 1000 Caixa 0125 0129 0183 0474 0078 1000 Santander 0032 0186 0295 0110 0153 0144 1000 Privados 0270 0160 0235 0192 0103 0325 0077 1000 Públicos 0057 0193 0154 0064 0060 0038 0179 0269 1000 Geral 0165 0216 0089 0172 0104 0218 0041 0861 0721 1000 PIB IGPM Índice de Basileia PIB IGPM Índice de Basileia Índice de Basileia Variáveis Variáveis IGPM PIB difRiscoBrasil INEC Escore de Eficiência Escore de Eficiência difRisco Brasil INEC 104 APÊNDICE C TESTES ADF Tabela 25 Teste ADF para principais variáveis do estudo Fonte Elaborado pelo próprio autor Notas dif é o indicativo das variáveis em que foi tirada a primeira diferença e ln é o indicativo de que foi tirado o logaritmo natural Os testes consideraram o nível de significância de 10 Absoluto Drift Trend Absoluto Drift Trend 0361 0038 0841 Não Estacionária Estacionária Não Estacionária 0449 0053 0686 Não Estacionária Estacionária Não Estacionária 0012 0001 0077 Estacionária Estacionária Estacionária 0276 0026 0895 Não Estacionária Estacionária Não Estacionária 0002 0000 0010 Estacionária Estacionária Estacionária Itaú 0009 0001 0039 Estacionária Estacionária Estacionária BB 0582 0085 0906 Não Estacionária Estacionária Não Estacionária Bradesco 0393 0043 0715 Não Estacionária Estacionária Não Estacionária Caixa 0001 0000 0018 Estacionária Estacionária Estacionária Santander 0298 0028 0529 Não Estacionária Estacionária Não Estacionária Privados 0345 0035 0666 Não Estacionária Estacionária Não Estacionária Públicos 0089 0007 0311 Estacionária Estacionária Não Estacionária Geral 0092 0007 0299 Estacionária Estacionária Não Estacionária Itaú 0080 0006 0088 Estacionária Estacionária Estacionária BB 0022 0344 0243 Estacionária Não Estacionária Não Estacionária Bradesco 0292 0028 0088 Não Estacionária Estacionária Estacionária Caixa 0979 0631 0745 Não Estacionária Não Estacionária Não Estacionária Santander 0452 0053 0568 Não Estacionária Estacionária Não Estacionária Privados 0057 0004 0267 Estacionária Estacionária Não Estacionária Públicos 0856 0256 0622 Não Estacionária Não Estacionária Não Estacionária Geral 0136 0011 0407 Não Estacionária Estacionária Não Estacionária Itaú 0042 0003 0051 Estacionária Estacionária Estacionária BB 0029 0243 0303 Estacionária Não Estacionária Não Estacionária Bradesco 0194 0016 0069 Não Estacionária Estacionária Estacionária Caixa 0371 0039 0645 Não Estacionária Estacionária Não Estacionária Santander 0891 0308 0810 Não Estacionária Não Estacionária Não Estacionária Privados 0006 0001 0050 Estacionária Estacionária Estacionária Públicos 0147 0012 0545 Não Estacionária Estacionária Não Estacionária Geral 0079 0006 0464 Estacionária Estacionária Não Estacionária Itaú 0208 0018 0392 Não Estacionária Estacionária Não Estacionária BB 0912 0350 0547 Não Estacionária Não Estacionária Não Estacionária Bradesco 0367 0038 0163 Não Estacionária Estacionária Não Estacionária Caixa 0344 0035 0438 Não Estacionária Estacionária Não Estacionária Santander 0013 0001 0902 Estacionária Estacionária Não Estacionária Privados 0872 0278 0557 Não Estacionária Não Estacionária Não Estacionária Públicos 0064 0005 0229 Estacionária Estacionária Não Estacionária Geral 0138 0011 0340 Não Estacionária Estacionária Não Estacionária Itaú 0155 0013 0390 Não Estacionária Estacionária Não Estacionária BB 0001 0000 0005 Estacionária Estacionária Estacionária Bradesco 0013 0001 0007 Estacionária Estacionária Estacionária Caixa 0352 0036 0532 Não Estacionária Estacionária Não Estacionária Santander 0064 0005 0000 Estacionária Estacionária Estacionária Privados 0101 0008 0303 Não Estacionária Estacionária Não Estacionária Públicos 0015 0001 0055 Estacionária Estacionária Estacionária Geral 0077 0006 0159 Estacionária Estacionária Não Estacionária Itaú 0009 0001 0053 Estacionária Estacionária Estacionária BB 0772 0175 0342 Não Estacionária Não Estacionária Não Estacionária Bradesco 0056 0004 0058 Estacionária Estacionária Estacionária Caixa 0609 0095 0381 Não Estacionária Estacionária Não Estacionária Santander 0639 0105 0905 Não Estacionária Não Estacionária Não Estacionária Privados 0874 0281 0573 Não Estacionária Não Estacionária Não Estacionária Públicos 0008 0001 0008 Estacionária Estacionária Estacionária Geral 0314 0031 0064 Não Estacionária Estacionária Estacionária dif ln Agências PAs Resultado Escore de Eficiência de Receita Escore de Eficiência de Lucro Risco Brasil Pvalor Índice de Basileia difRiscoBrasil PIB IGPM dif Índice de Basileia Variáveis Escore de Eficiência de Custos INEC 105 APÊNDICE D TESTE LM Tabela 26 Teste LM de cada amostra Fonte Elaborado pelo próprio autor Amostra Lag chi2 df Prob chi2 Amostra Lag chi2 df Prob chi2 1 3329 36 0598 1 5781 36 0012 2 3964 36 0311 2 2310 36 0953 3 2808 36 0824 3 4026 36 0287 4 4050 36 0278 4 4392 36 0171 1 5305 36 0033 1 4036 36 0284 2 3964 36 0311 2 3589 36 0474 3 4645 36 0114 3 3059 36 0723 4 4664 36 0110 4 2365 36 0944 1 4049 36 0279 1 3469 36 0531 2 4869 36 0077 2 4485 36 0148 3 3119 36 0697 3 4867 36 0077 4 3367 36 0580 4 3981 36 0304 1 5159 36 0045 1 4066 36 0273 2 2932 36 0777 2 5830 36 0011 3 7117 36 0000 3 5950 36 0008 4 3656 36 0443 4 5022 36 0058 Geral Públicos Privados Santander Caixa Bradesco BB Itaú 106 APÊNDICE E CORRELOGRAMA CRUZADO Figura 11 Correlograma cruzado do escore de eficiência de custos do Itaú para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor Figura 12 Correlograma cruzado do escore de eficiência de custos do Banco do Brasil para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of ITAUcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of ITAUcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of ITAUcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of ITAUcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of ITAUcosteffic and difITAUbasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BBcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BBcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BBcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BBcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BBcosteffic and difBBbasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 107 Figura 13 Correlograma cruzado do escore de eficiência de custos do Bradesco para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor Figura 14 Correlograma cruzado do escore de eficiência de custos do Caixa para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BRADESCOcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BRADESCOcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BRADESCOcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BRADESCOcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BRADESCOcosteffic and difBRADESCObasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of CAIXAcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of CAIXAcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of CAIXAcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of CAIXAcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of CAIXAcosteffic and difCAIXAbasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 108 Figura 15 Correlograma cruzado do escore de eficiência de custos do Santander para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor Figura 16 Correlograma cruzado do escore de eficiência de custos da média ponderada dos Privados para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of SANTANDERcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of SANTANDERcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of SANTANDERcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of SANTANDERcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of SANTANDERcosteffic and difSANTANDERbasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PRIVcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PRIVcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PRIVcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PRIVcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PRIVcosteffic and difPRIVbasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 109 Figura 17 Correlograma cruzado do escore de eficiência de custos da média ponderada dos Públicos para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor Figura 18 Correlograma cruzado do escore de eficiência de custos da média ponderada Geral para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PUBcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PUBcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PUBcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PUBcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PUBcosteffic and difPUBbasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of GERALcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of GERALcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of GERALcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of GERALcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of GERALcosteffic and difGERALbasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 110 APÊNDICE F ESTABILIDADE DO MODELO VAR Figura 19 Gráficos da Estabilidade dos Modelos VAR para todas as amostras Fonte Elaborado pelo próprio autor 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix Santander Itaú BB Privados 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix Bradesco Caixa Públicos Geral 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix Santander Privados 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix Públicos Geral 111 APÊNDICE G TESTES DE CAUSALIDADE DE GRANGER Tabela 27 Probabilidades do teste de causalidade de Granger para as demais variáveis Fonte Elaborado pelo próprio autor Notas Os números expressos na tabela representam a probabilidade de aceitar a hipótese nula sendo que a fonte verde representa os casos em que rejeitamos e em vermelho os casos em que aceitamos Consideramos o nível de significância de 5 Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral Risco Brasil 0001 0009 0332 0001 0000 0000 0001 0000 PIB 0102 0057 0080 0321 0000 0002 0135 0001 IGPM 0000 0000 0004 0000 0000 0000 0000 0000 Basileia 0000 0136 0908 0042 0004 0000 0077 0002 Todas 0000 0000 0000 0000 0000 0000 0000 0000 INEC 0004 0010 0078 0001 0003 0101 0040 0326 PIB 0294 0016 0056 0047 0066 0280 0166 0615 IGPM 0261 0039 0405 0044 0221 0093 0193 0205 Basileia 0847 0069 0602 0006 0020 0001 0502 0039 Todas 0017 0000 0000 0000 0000 0000 0025 0001 INEC 0047 0221 0005 0001 0228 0083 0130 0323 Risco Brasil 0002 0830 0115 0158 0271 0042 0417 0167 IGPM 0003 0100 0814 0003 0231 0020 0029 0053 Basileia 0179 0118 0042 0429 0479 0465 0012 0234 Todas 0000 0035 0000 0000 0002 0000 0000 0001 INEC 0862 0873 0060 0000 0802 0834 0068 0571 Risco Brasil 0052 0005 0014 0000 0310 0068 0000 0001 PIB 0629 0732 0543 0000 0711 0502 0061 0474 Basileia 0004 0098 0046 0000 0039 0005 0000 0001 Todas 0010 0080 0055 0000 0007 0021 0000 0002 INEC 0032 0009 0000 0021 0154 0180 0000 0000 Risco Brasil 0001 0000 0317 0149 0266 0428 0005 0000 PIB 0004 0000 0000 0166 0000 0078 0006 0091 IGPM 0707 0000 0091 0007 0339 0652 0003 0002 Todas 0000 0000 0000 0001 0002 0000 0000 0000 Risco Brasil PIB IGPM Basileia Equação Excluído Probabilidade INEC 112 APÊNDICE H MATRIZ DE CORRELAÇÃO DOS ESCORES DE EFICIÊNCIA Tabela 28 Matriz de correlação dos escores de eficiência de custo lucro e receita Fonte Elaborado pelo próprio autor Custo Lucro Receita Custo 1000 Lucro 0152 1000 Receita 0023 0239 1000 Custo 1000 Lucro 0073 1000 Receita 0003 0259 1000 Custo 1000 Lucro 0128 1000 Receita 0060 0703 1000 Custo 1000 Lucro 0091 1000 Receita 0124 0504 1000 Custo 1000 Lucro 0105 1000 Receita 0114 0198 1000 Custo 1000 Lucro 0179 1000 Receita 0063 0381 1000 Custo 1000 Lucro 0079 1000 Receita 0035 0553 1000 Custo 1000 Lucro 0083 1000 Receita 0076 0465 1000 Privados Públicos Geral Bradesco Caixa Santander Escore de Eficiência Amostra Escore de Eficiência Itaú Banco do Brasil Research in International Business and Finance 23 2009 5477 Contents lists available at ScienceDirect Research in International Business and Finance journal homepage wwwelseviercomlocateribaf Determinants of bank efficiency during unstable macroeconomic environment Empirical evidence from Malaysia Fadzlan Sufian ab a Planning and Research Department CIMB Bank Berhad Malaysia b Department of Economics Faculty of Economics and Management Universiti Putra Malaysia UPM Malaysia a r t i c l e i n f o Article history Received 5 November 2007 Received in revised form 7 July 2008 Accepted 9 July 2008 Available online 17 July 2008 JEL classification G21 G28 Keywords Financial crisis Bank efficiency Malaysia a b s t r a c t The present study investigates for the first time the efficiency of Malaysian banking sector around the Asian financial crisis 1997 The efficiency estimates of individual banks are evaluated by using the Data Envelopment Analysis DEA approach To examine the robustness of the estimated efficiency scores under various alter natives and to differentiate how efficiency scores vary with changes in inputs and outputs the present study focuses on three major approaches viz intermediation approach value added approach and operating approach The analysis further links the variation in calculated efficiencies to a set of explanatory variables ie bank size profitability and ownership The empirical findings clearly bring forth the high degree of inefficiency in the Malaysian bank ing sector particularly a year after the East Asian crisis The results suggest that the decline in technical efficiency is more abrupt under the intermediation approach relative to the value added approach and operating approach The regression results focusing on bank efficiency and other bank specific traits suggest that efficiency is negatively related to expense preference behavior and economic conditions while bank efficiency is positively related to loans inten sity 2008 Elsevier BV All rights reserved Correspondence address 21st Floor 6 Jalan Tun Perak 50050 Kuala Lumpur Malaysia Tel 60 3 2693 1722x6455 fax 60 3 2691 4415 Email addresses fadzlansufiancimbcom fadzlan14gmailcom 02755319 see front matter 2008 Elsevier BV All rights reserved doi101016jribaf200807002 F Sufian Research in International Business and Finance 23 2009 5477 55 1 Introduction The world was stunned when East Asia the highest growth region during the 1990s was hit by a banking crisis in 1997 The crisis which started in Thailand and spread rapidly throughout the region drew widespread attention from economists and financial analysts It was characterized by skyrock eting interest rates 34 in Korea 13 in Indonesia a dramatic drop in stock price indexes 55 in Thailand 52 in Malaysia real exchange rate depreciation 97 in Korea 87 in Thailand a decline in net capital flow to the region 20 billion and a drop in the regions gross domestic product 481 billion In order to understand the nature of the East Asian crisis as well as similar crises around the world economists have used a variety of methodologies while focusing primarily on macrolevel vari ables Papers in the macrotradition include Frankel and Rose 1996 Sachs et al 1996 Kaminsky et al 1998 Corsetti et al 1998 Eichengreen and Rose 1998 Chinn 1998 Kaminsky and Reinhart 1998 Krugman 1999 Berg and Pattillo 1999 Alba et al 1999 and Tanner 2000 among oth ers In addition to these studies a number of papers have adopted a microperspective deriving from banks balance sheet and income statements Generally the variables examined in these papers fol low the wellknown CAMEL framework1 These include Sinkey 1975 Barth et al 1985 Lane et al 1986 Thomson 1991 and GonzalezHermosillo et al 1997 There were also papers which com bined macro and microvariables including those of Goldstein and Turner 1996 Honohan 1997 and GonzalezHermosillo 1999 The characteristics of banking systems that are more prone to experience banking and financial crisis and its impact has also drawn academic interests in recent years Barth et al 2000 suggests that greater regulatory restrictions on bank activities are associated with higher probability of suffering a major banking crisis Beck et al 2006 find that the likelihood of financial crises is lower in more concentrated banking systems yet higher in less competitive and countries with less developed legal systems Daniel and Jones 2007 suggests that even if a banking system is well designed countries will enjoy an initial period of rapid low risk growth before entering a period with an elevated risk of banking crisis By using aggregate and bank level data for 35 developed and developing countries Dermiguc Kunt et al 2006 find that depositors leave weaker banks for stronger ones following a banking crisis Kroszner et al 2007 find that sectors that are highly dependent on external finance tend to experience greater contraction during a banking crisis in countries with deeper financial systems than in countries with shallower financial systems The literature examining the efficiency of financial institutions with parametric andor non parametric frontier techniques has expanded rapidly in recent times While a large body of literature spanning a halfcentury exists on banking efficiency in the United States see surveys in Berger et al 1993 Berger and Humphrey 1997 Berger 2007 and references therein more recent studies examine several other countries such as India Ataullah and Le 2006 Hong Kong Drake et al 2006 Greece Pasiouras 2008b Singapore Sufian 2007 and Ukraine Kyj and Isik 2008 Apart from focusing on various countries these studies also examine several other issues of bank efficiency ie the impact of risk on bank efficiency eg Drake and Hall 2003 the impact of offbalance sheet activities on bank efficiency eg LozanoVivas and Pasiouras 2008 the relationship between bank efficiency and share prices eg Pasiouras et al 2008 the impact of mergers on bank efficiency eg AlSharkas et al 2008 The comparison of efficiency between foreign and domestic banks has also been studied extensively eg Bhattacharya et al 1997 Isik and Hassan 2002 Ataullah and Le 2006 Despite its severity and deep influence on both the real and financial sectors the impact of the East Asian crisis of 1997 on the efficiency of the financial industry has not been critically examined yet Fukuyama 1995 Humphrey and Pulley 1997 Leightner and Lovell 1998 and Isik and Hassan 2003a have suggested that frontier techniques can be used to assess the impact of major economic events such as economic crisis or financial liberalization on the performance of banking firms How 1 The CAMEL framework is used by financial regulators to rate the health of financial institutions The framework criteria are associated with Capital Adequacy C Asset Quality A Management M Earnings E and Liquidity L 56 F Sufian Research in International Business and Finance 23 2009 5477 Table 1 Shortterm claims on the East Asian economies change YoY Year Hong Kong Indonesia Korea Malaysia Philippines Singapore Taiwan Thailand 1990 744 6032 3101 1103 1210 1774 2602 1706 1991 538 1228 1725 4496 028 270 8487 3923 1992 389 1387 1149 3690 230 959 2469 346 1993 2239 938 1017 8149 2487 911 2413 1871 1994 2009 1327 3679 1102 3442 1783 4921 2011 1995 133 2953 3520 2000 2826 2546 4053 316 1996 1811 2419 2438 4158 9024 1406 2537 397 1997 169 250 1290 2899 5381 1134 475 1228 1998 4419 3248 4949 3543 2718 4013 3960 2224 1999 2149 1969 1811 1677 1199 1814 3085 710 Source Bank of International Settlements Units are in millions of US Dollars ever to the best of our knowledge Isik and Hassans 2003a study is the only empirical research performed to directly examine the impact of financial disruptions on bank efficiency by using the frontier technique By employing a unique data set of all banks operating in Malaysia during the period of 19951999 this study contributes to the existing literature by providing new empirical evidence on the impact of the East Asian crisis on the Malaysian banking sectors efficiency Given that banks are the dominant financial institution in Malaysia their health is very critical to the health of the general economy at large as demonstrated during the East Asian crisis which left many financial institutions in distress However at the present time this type of analysis is completely missing in the literature To do so the nonparametric frontierbased Data Envelopment Analysis DEA method is used to estimate the technical pure technical and scale efficiency To examine the robustness of the estimated efficiency scores under various alternatives the present study focuses on three major approaches namely intermediation approach value added approach and operating approach Finally we analyze how ownership structure and different bank characteristics such as capitalization problem loans ratio and size influences the efficiency estimates The following section identifies those factors that characterize the Asian financial crisis Section 3 describes the data sources and model specifications which are employed in the study Section 4 presents the results of the analysis of Malaysian banking sector using DEA and the panel regression techniques Finally we conclude in Section 5 2 The Asian financial crisis in perspective Over the last quarter of the 20th century both developed and developing countries have experi enced severe banking crisis Chile Argentina and Mexico 1980s Sweden 1990s Thailand Malaysia Korea Philippines and Indonesia 1997 Paraguay 19951998 Russia 1998 Turkey 1994 2000 and 2001 Argentina 2001 The main causes of these crises are poor banking practices and lack of revenue diversification inadequate capital shortcomings in the assessment of credit risk lending to connected enterprises excessive maturity or currency mismatches and rapid rise of nonperforming loans Due to less than fully developed stock markets in the East Asian region equity financing was not an important source of financing Rather banks borrowing from both offshore and onshore became an important focus of financing in East Asia Suetorsak 2006 A huge accumulation of foreign debt financing relative to equity increasingly raised banks leverage risk Interest rate risk and maturity risk were also a concern of the East Asian banks Interest rate risk reflects a maturity mismatch between banks assets and liabilities For example East Asian banks often borrowed abroad with a shortterm maturity and lent domestically with a longterm maturity for domestic projects The total shortterm maturity claims of the East Asian region are shown in Table 1 Banks in East Asia also assumed exceptional liquidity risk From Table 2 it is clear that East Asian banks had a high proportion of shortterm debts relative to longterm debts They exposed themselves to liquidity risk due to problems with shortterm debt renewal It is observed from Table 2 that prior to F Sufian Research in International Business and Finance 23 2009 5477 57 Table 2 Percentage of shortterm maturity claims to longterm maturity claims on banks East Asian economies Shortterm maturity claims on banks Longterm maturity claims on banks Percentage of shortterm claims to longterm maturity claims on banks 1995 Indonesia 5516 3384 163 Korea 35016 14984 2337 Malaysia 2115 2385 887 Philippines 1078 1122 961 Thailand 8073 17727 455 Hong Kong 7845 19819 396 Singapore 1251 7171 174 1996 Indonesia 7282 4518 1612 Korea 44531 21369 2084 Malaysia 3273 3227 1014 Philippines 3048 2152 1416 Thailand 6989 19011 368 Hong Kong 14186 23917 593 Singapore 1978 7879 251 1997 Indonesia 7428 4972 1494 Korea 38268 29032 1318 Malaysia 5257 5243 1003 Philippines 4609 891 5173 Thailand 7968 18132 439 Hong Kong 10305 26031 396 Singapore 2765 9747 284 Source Bank of International Settlements Units are in millions of US Dollars the crisis in 1997 among the crisis affected countries Korea had the highest percentage of shortterm debts relative to longterm debts followed by Indonesia while Thailand and Malaysia the least On the other hand Singapore and Hong Kong which were relatively unscathed by the East Asian crisis had relatively low percentage of shortterm debts relative to longterm debts Another risk resulting from excessive external debt accumulations was exchange rate risk After increasing their offshore borrowing in foreign currencies East Asian banks lent out money domestically with little restraint in local currencies This lending boom led to excessive domestic credit growth Frankel and Rose 1996 Nevertheless banks were also exposed to foreign exchange losses risk in the event of domestic currency depreciation Chang and Velasco 1999 This was problematic since the domestic credit expansion was inconsistent with the East Asian quasifixed exchange rate regimes causing a depletion of foreign exchange reserves and eventually causing a collapse of the fixed exchange rate regime Krugman 1999 This suggests that the internal imbalance of the banking industry triggered an unsustainable external balance In other words microdecisions of banks caused macroimbalances in these countries Regardless of the cause of the East Asian crisis all economists have agreed that the crisis has been severe Although initially only financial in nature the crisis has led to significant real economic losses in the once fast growing developing economies The Malaysian situation possesses some important characteristics Basic macroeconomic fun damentals were not signaling imminent danger Low inflation and falling unemployment rate characterized the country over the decade preceding the East Asian crisis Despite the sound macroeco nomic fundamentals a few other major indicators pointed towards serious problems by the mid1990s further compounded by the financial systems weakness The financial institutions were weakened by largescale exposure to the property sector high nonperforming loans and shortterm loans that were unhedged against currency movements Furthermore the credit expansion especially to the private sector was pronounced during 19951997 particularly towards the real estate and property sectors Prior to the crisis the banking system had relatively high exposures to the property and equity sectors which amounted to approximately 426 of total loans compared to only 21 exposure towards the manufacturing sector Once devaluation and drops in real estate and stock markets got underway the crisis took on a life of its own The drop in property and equity prices caused large losses and outright defaults by many property companies as well as those who participate in the equity market by margin accounts resulting in a dramatic increase in the amount of nonperforming loans NPLs In sum the domestic banks were hit simultaneously on many fronts namely the withdrawal of foreign funds foreign exchange losses a sharp rise in NPLs and losses on equity holdings The erosion of the banking systems capital base due to all these factors had severely constrained their ability to lend to even solvent companies in the midst of the crisis due to the need to comply with international capital adequacy rules 3 Methodology and data 31 Data Envelopment Analysis The present study employs the nonparametric frontier DEA approach with variable returns to scale VRS assumption to measure inputoriented technical efficiency of Malaysian banks DEA involves constructing a nonparametric production frontier based on the actual inputoutput observations in the sample relative to which efficiency of each firm in the sample is measured Coelli 1996 Let us give a short description of the DEA 2 Assume that there is data on K inputs and M outputs for each N bank For the ith bank these are represented by the vectors xi and yi respectively Let us call the K N input matrix X and the M N output matrix Y To measure the efficiency for each bank we calculate a ratio of all inputs such as uyivxi where u is an M 1 vector of output weights and v is a K 1 vector of input weights To select optimal weights we specify the following mathematical programming problem min uy uyiv xi uyivxi 1 j 1 2 n u v 0 1 The above formulation has a problem of infinite solutions and therefore we impose the constraint vxi 1 which leads to min μ φ μy1φxi 1 μyi φkj 0 j 1 2 N μ φ 0 2 where we change notation from u and v to μ and φ respectively in order to reflect transformations Using the duality in linear programming an equivalent envelopment form of this problem can be derived min θ λ yi Yλ 0 θxi Xλ 0 λ 0 3 where θ is a scalar representing the value of the efficiency score for the ith bank which will range between 0 and 1 λ is a vector of N 1 constants The linear programming has to be solved N times once for each bank in the sample In order to calculate efficiency under the assumption of VRS the convexity constraint N1λ 1 will be added to ensure that an inefficient bank is only compared against banks of similar size and therefore provides the basis for measuring economies of scale within the DEA concept The convexity constraint determines how closely the production frontier envelops the observed inputoutput combinations and is not imposed in the constant returns to scale CRS case 32 Multivariate Tobit regression analysis Coelli et al 1998 suggested several ways in which environmental variables can be accommodated in a DEA analysis The term environmental variables is usually used to describe factors which could influence the efficiency of a firm In this case such factors are not traditional inputs and are assumed to be outside the control of the manager As defined in Eq 3 the DEA score falls between the interval 0 2 Good reference books on efficiency measures are Coelli et al 1998 Thanassoulis 2001 and Cooper et al 2000 and 1 0 θ 1 making the dependent variable a limited dependent variable A commonly held view in previous studies is that the use of the Tobit model can handle the characteristics of the distribution of in efficiency measures and thus provide results that can provide important policy guidelines to improve performance Accordingly DEA efficiency scores obtained in the first stage is used as a dependent variable in the second stage and are regressed against a set of bank characteristics and other environmental variables The standard Tobit model can be defined as follows for observation bank i yi β x ε y y if y 0 and y 0 otherwise 4 where εi N0 σ2 xi and β are vectors of explanatory variables and unknown parameters respectively while y is a latent variable and yi is the DEA efficiency score 33 Specification of bank inputs outputs and data It is commonly acknowledged that the choice of variables in efficiency studies significantly affects the results The problem is compounded by the fact that variable selection is often constrained by the paucity of data on relevant variables The cost and output measurements in banking are especially difficult because many of the financial services are jointly produced and prices are typically assigned to a bundle of financial services The role of commercial banks is generally defined as collecting the savings of households and other agents to finance the investment needs of firms and consumption needs of individuals Three approaches dominate the literature the production approach the intermediation approach and more recently the revenue or value added approach The operating approach of defining inputs and outputs of banks has also been popular Jemric and Vujcic 2002 The first two approaches apply the traditional microeconomic theory of the firm to banking and differ only in the specification of banking activities The third and fourth approach goes a step further and incorporates some specific activities of banking into the classical theory and thereby modifies it Under the production approach pioneered by Benston 1965 a financial institution is defined as a producer of services for account holders that is they perform transactions on deposit accounts and process documents such as loans Hence according to this approach the number of accounts or its related transactions is the best measure for output while the number of employees and physical capital are considered as inputs This approach has primarily been employed in studying the efficiency of bank branches Berger and Humphrey 1992 The intermediation approach on the other hand assumes that financial firms act as an intermediary between savers and borrowers and posits total loans and securities as outputs whereas deposits along with labour and physical capital are defined as inputs The operating approach or income based approach views banks as business units with the final objective of generating revenue from the total cost incurred for running the business Leightner and Lovell 1998 Accordingly it defines banks output as total revenue interest and noninterest income and inputs as the total expenses interest and noninterest expenses More recently Drake et al 2006 proposed the revenue approach or value added approach in DEA The value added approach identifies those balance sheet categories as outputs that contribute to the bank value added In general under this approach deposits and loans are viewed as outputs because they are responsible for the significant proportion of value added The appropriateness of each approach varies according to the circumstances It is apparent that banks undertake simultaneous functions However based on practical considerations and to examine the robustness of the estimated efficiency scores under various alternatives the present study focuses on three major approaches intermediation approach value added approach and operating approach Under the intermediation approach we assume deposits x2 labour x3 and capital x1 as inputs 3 likelihood function L is maximized to solve β and σ based on 171 observations banks of yi and xi is L yi 0 1 F yi 0 12Πσ2 12 e12σ2yβxi2 where Fi βxiσ 12Π12et22 dt The first product is over the observations for which the banks are 100 efficient y 0 and the second product is over the observations for which banks are inefficient y 0 Fi is the distribution function of the standard normal evaluated at βxiσ 60 F Sufian Research in International Business and Finance 23 2009 5477 Table 3 Descriptive statistics for inputs and outputs 1995 RMm 1996 RMm 1997 RMm 1998 RMm 1999 RMm Outputs Total loans Mean 528906408 679483803 835088961 969727785 972539809 SD 872785964 1011235357 1192196075 1437264431 1443451230 Investments Mean 161352108 205769856 253581888 289988042 293979955 SD 333493592 336822895 422983641 465112056 523261154 Interest income Mean 57731719 78759181 104331055 147455503 105609500 SD 95501644 122835793 156100693 222718785 172386525 Noninterest income Mean 8887419 10628472 12633661 14038148 13928582 SD 15051184 17125609 19600668 19773007 21481235 Inputs Capital Mean 9910225 11936536 13206267 16308094 17344830 SD 18322180 20036486 20568534 22365871 24843507 Total deposits Mean 718373928 873139222 1114000858 1215176988 1289761248 SD 1210740871 1354051762 1664963238 1759391510 1882915776 Labour Mean 7598469 8612244 10318052 11536691 10278524 SD 12388690 13770891 16303227 16699356 13970262 Interest expense Mean 34141642 47306778 64503800 98911603 66229870 SD 55924949 74647921 94199438 147673993 110340995 Noninterest expense Mean 6758075 8061311 9271452 11818221 10339167 SD 11035623 12738836 13915652 16604321 13341085 Source Individual Banks Annual Reports The table presents summary statistics of the variables used to construct the efficiency frontier for the DEA intermediation operating and value added approaches over the period 19951999 for producing loans y1 and investments y2 Under the value added approach labour x3 capital x1 and interest expenses x4 are used as inputs producing outputs like deposits x2 loans y1 and investments y2 Under the operating approach three types of inputs are considered namely interest expenses x4 labour x3 and other operating expenses excluding employee expenses x5 The relevant outputs are interest income y3 and noninterest income y4 emanating mostly from commission exchange brokerage etc We use annual bank level and macroeconomic data of all Malaysian commercial banks over the period 19951999 The variables are obtained from published balance sheet information in annual reports of each individual institution while the macroeconomic variable is sourced from various issues of Bank Negara Malaysias annual reports The dataset covers the whole gamut of the industrys total assets The final sample consists of 36 banks in 1995 and 1996 and 33 banks in 1997 1998 and 1999 yielding a total of 171 bankyear observations Table 3 presents summary statistics of the output and input variables used to construct the DEA model Several bank and industry specific attributes may influence a particular banks efficiency level Some of these factors may be neither inputs nor outputs in the production process but rather circumstances faced by a particular bank We use an array of bank specific variables to control banks production technologies the input and product market share they are facing and other factors that might confound the empirical relationship between bank characteristics and efficiency The independent variables used to explain Malaysian banks efficiency are grouped under two main characteristics The first represent bankspecific attributes while the second encompass economic conditions during the period examined The bankspecific variables included in the regressions are LNDEPO ROA LOANSTA LNTA LLPTL NIITA NIETA and EQASS To measure the relationship between economic conditions and bank efficiency LNGDP is used Agency issues associated with different types of firm ownership are an area of concern in many banking systems To capture the effects of organizational forms and governance on Malaysian bank efficiency binary dummy variables DUMFORB DUMPUBL and DUMGOVT are included in the regression models Finally to explore the home field and global advantage hypotheses suggested by Berger et al 2000 we have further classified foreign banks operating in Malaysia into three major groups according to their origins namely DUMAMER DUMEURO and DUMASIA During the period of investigation there were four banks headquartered in the North America and Canada four banks headquartered in the European countries and six foreign banks from other Asian countries operating in Malaysia The independent variables their hypothesized relationship with efficiency and notations are detailed in Table 4 To test the relationship between Malaysian bank efficiency and the bank specific and macroeconomic determinants described above the following regression model is estimated θjt β0β1 Characteristics β2 Ecom β3 Ownership β4 Origins εjt 5 where θjt is the technical efficiency of the jth bank in period t obtained from DEA intermediation operating and value added approaches Characteristics is a set of bank characteristics Econ is a vector of economic conditions Ownership is an array of bank ownership characteristics and finally Origins is a set of bank origins variables 4 Results and discussion In this section we will discuss the technical efficiency TE change of the Malaysian banking sector measured by the DEA method and its decomposition into pure technical efficiency PTE and scale efficiency SE components The efficiency of Malaysian commercial banks is first examined by employing the DEA method for each year under investigation The results are classified into two broad heads First we describe the estimates of technical efficiency during the East Asian crisis under three alternative approaches Second to substantiate the results under the DEA approach a multivariate regression framework is employed to relate bank efficiency level to a set of bank specific traits and other macroeconomic determinants 41 Efficiency of the Malaysian banking sector The summary results of technical pure technical and scale efficiency estimates under the three approaches are presented in Tables 57 respectively The average technical efficiency estimate M represents the average of all optimal values obtained from CRS model for each commercial bank Table 5 4 The empirical results suggest a large asymmetry between banks regarding their technical efficiency scores In particular the different approaches of classifying inputs and outputs of banks produced divergent sets of efficiency estimates The estimates of technical efficiency were observed to be consistently higher under operating approach visàvis the intermediation and value added approaches On the other hand under the intermediation approach banks are characterized by relatively low level 4 The CRS assumption is only justifiable when all decisionmaking units DMUs are operating at an optimal scale However firms or DMUs in practice might face either economies or diseconomies of scale Thus if one makes the CRS assumption when not all DMUs are operating at the optimal scale the computed measures of technical inefficiency will be contaminated with scale inefficiency Banker et al 1984 extended the CCR model by relaxing the CRS assumption The resulting BCC model was used to assess the efficiency of DMUs characterized by VRS The VRS assumption provides the measurement of PTE which is the measurement of technical efficiency devoid of the SE effects If there appears to be a difference between the TE and PTE scores of a particular DMU then it indicates the existence of scale inefficiency 62 F Sufian Research in International Business and Finance 23 2009 5477 Table 4 Descriptive of the variables used in the regression models Variable Description Hypothesized relationship with efficiency Remarks Bank characteristics ROA Return on assets Is used as a proxy of bank profitability A positive relationship with bank efficiency is expected LNDEPO Natural logarithm of total deposits Is used as a proxy of market share We do not have a priori expectation on the variable sign LOANSTA Total loans over total assets Is used as an indicator for bank liquidity position Liquidity is an indication of banks ability to meet its customers daytoday cash needs and respond to sudden cash withdrawals The variable is expected to enter the regression model positively LNTA Natural logarithm of total assets Is used as a proxy of bank size to capture the possible cost advantages associated with size economies of scale The variable is expected to take a positive sign LLPTL Loan loss provisions over total loans Is used as a proxy for asset quality It is expected that higher levels of nonperforming assets will result in lower efficiency levels NIITA Noninterest income over total assets Is used as a proxy for banks diversification strategy into nontraditional activities We do not have a priori expectation on the variable sign NIETA Noninterest expense over total assets Is used as a proxy for management quality An inverse relationship is expected between this variable and bank efficiency level EQASS Total book value of shareholders equity over total assets Is used as a proxy of capital adequacy A positive sign is expected Regulators view a higher level of equity as a cushion of future losses Economic conditions LNGDP Natural logarithm of gross domestic products Is used as a proxy for economic conditions We do not have a priori expectation on the variable sign Ownership DUMFORB Dummy variable that takes a value of 1 for foreign banks 0 otherwise Is used to examine the relationship between foreign bank ownership with efficiency The variable is expected to take a positive sign DUMPUBL Dummy variable that takes a value of 1 for publicly listed banks 0 otherwise Is used to examine the relationship between publicly listed banks with efficiency We do not have a priori expectation on the variable sign DUMGOVT Dummy variable that takes a value of 1 for government links banks 0 otherwise Is used to examine the relationship between government ownership and efficiency The variable is expected to take a negative sign Origins DUMAMER Dummy variable that takes a value of 1 for American banks 0 otherwise Is used to test the home field advantage versus global advantage and liability of unfamiliarness hypotheses DUMEURO Dummy variable that takes a value of 1 for European banks 0 otherwise Is used to test the home field advantage versus global advantage and liability of unfamiliarness hypotheses DUMASIA Dummy variable that takes a value of 1 for Asian banks 0 otherwise Is used to test the home field advantage versus global advantage and liability of unfamiliarness hypotheses Table 5 Average technical efficiency of Malaysian banks Year of banks of efficient banks Average efficiency M Average inefficiency 1 MM Standard deviation σ Minimum Interval I M σ I M σ of banks in I Intermediation approach 1995 36 3 0519 0927 0220 0135 0299 0739 750 1996 36 3 0515 0942 0200 0307 0315 0715 806 1997 33 3 0570 0754 0179 0371 0391 0749 727 1998 33 2 0555 0802 0195 0339 0360 0750 758 1999 33 1 0327 2058 0179 0107 0148 0506 818 Value added approach 1995 36 3 0551 0815 0200 0294 0351 0751 806 1996 36 6 0817 0224 0138 0404 0679 0955 667 1997 33 6 0852 0174 0132 0518 0720 0984 697 1998 33 11 0898 0114 0114 0562 0784 1012 848 1999 33 5 0702 0425 0173 0323 0529 0875 727 Operating approach 1995 36 7 0889 0125 0095 0584 0794 0984 667 1996 36 5 0865 0156 0079 0750 0786 0944 639 1997 33 5 0872 0147 0093 0684 0779 0965 606 1998 33 4 0773 0294 0115 0594 0658 0888 727 1999 33 4 0752 0330 0129 0540 0623 0881 727 Table 6 Average pure technical efficiency of Malaysian banks Year of banks of efficient banks Average efficiency M Average inefficiency 1 MM Standard deviation σ Minimum Interval I M σ I M σ of banks in I Intermediation approach 1995 36 7 0784 0276 0210 0141 0574 0994 667 1996 36 7 0840 0190 0140 0502 0700 0980 583 1997 33 10 0884 0131 0114 0643 0770 0998 515 1998 33 10 0838 0193 0163 0421 0675 1001 879 1999 33 8 0759 0318 0229 0245 0530 0988 576 Value added approach 1995 36 11 0834 0199 0163 0428 0671 0997 556 1996 36 16 0912 0096 0124 0463 0788 1036 833 1997 33 18 0930 0075 0106 0633 0824 1036 758 1998 33 19 0944 0059 0091 0691 0853 1035 879 1999 33 10 0790 0266 0188 0411 0602 0978 545 Operating approach 1995 36 16 0930 0075 0080 0719 0850 1010 806 1996 36 15 0926 0080 0080 0770 0846 1006 778 1997 33 12 0936 0068 0084 0731 0852 1020 848 1998 33 10 0880 0136 0109 0654 0771 0989 485 1999 33 13 0878 0139 0140 0600 0738 1018 788 Table 7 Average scale efficiency of Malaysian banks Year of banks of efficient banks Average efficiency M Average inefficiency 1 MM Standard deviation σ Minimum Interval I M σ I M σ of banks in I Intermediation approach 1995 36 3 0665 0504 0187 0395 0478 0852 639 1996 36 3 0616 0623 0207 0388 0409 0823 722 1997 33 3 0644 0553 0170 0402 0474 0814 636 1998 33 2 0663 0508 0172 0432 0491 0835 667 1999 33 1 0445 1247 0196 0141 0249 0641 758 Value added approach 1995 36 3 0662 0511 0187 0404 0475 0849 611 1996 36 6 0896 0116 0090 0637 0806 0986 722 1997 33 6 0917 0091 0099 0559 0818 1016 909 1998 33 11 0950 0053 0063 0803 0887 1013 818 1999 33 5 0902 0109 0140 0504 0762 1042 879 Operating approach 1995 36 7 0957 0045 0075 0584 0882 1032 972 1996 36 5 0936 0068 0065 0760 0871 1001 833 1997 33 5 0934 0071 0070 0767 0864 1004 818 1998 33 4 0882 0134 0098 0666 0784 0980 576 1999 33 4 0863 0159 0103 0626 0760 0966 667 66 F Sufian Research in International Business and Finance 23 2009 5477 of technical efficiency Illustratively in year 1999 ie a year after the East Asian crisis only one bank 3 was found to be efficient and the average technical efficiency for all banks stood at 327 under the intermediation approach The number of efficient banks during the sample period ranged from three banks in 19951997 to only one bank in 1999 under the intermediation approach and seven banks in 1995 to four banks in 1998 and 1999 under the operating approach It is interesting to note that the number of efficient banks was the highest under the value added approach ranging from 3 banks in 1995 to 11 banks in 1998 In sum in and around the East Asian crisis period there was no perceptible change in number of efficient banks under the intermediation and operating approaches although a noticeable increase was observed under the value added approach Under the operating approach the dispersion of technical efficiency scores as measured by its standard deviation depicts an increasing trend during the years 19971999 On the other hand the percentage of banks wherein technical efficiency lies within the interval of 1 SD around the mean hovered around 61 in 1997 to 73 in 1998 and 1999 under the operating approach and 67 in 1996 to 85 in 1998 under the value added approach These numbers were higher under the intermedi ation approach As the technical efficiency estimates itself is time varying these proportions do not necessarily corroborate the degree of in efficiency of the banking system For example under the intermediation approach around 75 in 1995 and around 82 in 1999 of banks recorded technical efficiency within the interval of 1 SD around the mean Yet banks were much more efficient in 1995 than in 1999 As against the changing benchmark of comparison these proportions quantify the number of banks that are close to the average over time and thus merely capture the kurtosis of the efficiency distribu tion depending on the approach For instance under the intermediation approach the efficiency scores displays a leptokurtic distribution ie the efficiency scores has a high peak with a small variance sug gesting that a lot of scores fall in the center of the distribution On the other hand under the operating approach the efficiency of the Malaysian banking sector seem to follow a mesokurtic distribution ie the efficiency scores displays a moderate peak with gradual curves suggesting a normal number of scores in the middle range of the distribution Overall the empirical findings presented in Table 5 clearly bring forth the high degree of inefficiency of Malaysian banking sector around the East Asian crisis period particularly a year after the East Asian crisis Most of the inefficiency stemmed from the under utilization of resources inputs Finally considering the evolution of efficiency over time a clear temporal pattern does not emerge from these different approaches However under the intermediation approach inefficiency exists in the production of banking services and appears to be an important determinant of banks costs It is also observed from Table 5 that although in general technical efficiency level seems to deteriorate abruptly a year after the East Asian crisis under all approaches the deterioration seems to be more pronounced under the intermediation approach model The set of inputoutput is run under both CRS and VRS assumptions If the efficiency score of each bank produced by these models varies then the bank is said to experience VRS Avkiran 1999 Once pure technical efficiency is estimated using VRS scale efficiency is derived by dividing the technical efficiency by pure technical efficiency Table 6 presents the PTE estimates while SE estimates under all three approaches are presented in Table 7 It is observed that over the sample period both pure technical and scale efficiency measures especially under intermediation approach display significant variation and the Malaysian banking sector did not achieve sustained efficiency gains Estimates of pure technical efficiency under the intermediation approach vary from a low of 76 in 1999 to a high of 88 in 1997 In most of the years about 20 of banks were found to be pure technically efficient under the intermediation approach Interestingly the percentage of banks whose pure technical efficiency falls within the interval of 1 SD around the mean displayed a large asymmetry particularly during the period 19971999 under all approaches It is observed from Table 6 that under the intermediation approach the percentage stood at around 52 in 1997 to 88 in 1998 while under value added approach and operating approach the figures stood at around 55 in 1999 to 88 in 1998 and 49 in 1998 to 85 in 1997 respectively It is interesting to note that the number of efficient banks under CRS technical efficiency assump tion and VRS pure technical efficiency assumption differs markedly irrespective of the choice of various inputs and outputs This clearly demonstrates the existence of sizable scale inefficiency among Table 8 Multivariate Tobit regression analysis Explanatory variables Model 1 Intermediation approach operating approach Value added approach CONSTANT 7500 2103 4462 1149 3383 1132 Bank characteristics LNDEPO 0231 0541 0026 0028 0043 0028 LOANSTA 0541 0107 0194 0058 0159 0058 LNTA 0208 0056 0015 0031 0029 0030 LLPTL 0056 0291 0292 0159 0162 0157 NIITA 0448 2008 3208 1097 3308 1080 NIETA 16007 2269 10413 1239 8475 1221 EQASS 0542 0449 0282 0245 0552 0242 ROA 0005 0009 0033 0005 0033 0005 Economic conditions LNGDP 0626 0194 0322 0106 0213 0105 No of observations 171 171 171 Log likelihood 81395 184796 187381 R2 0495 0500 0460 Adj R2 0464 0469 0426 θjt α β1 LNDEPO β2 LOANSTA β3 LNTA β4 LLPTL β5 NIETA β6 EQUITYTA β7 ROA β8 LOGGDP εj The dependent variable is banks efficiency scores derived from DEA intermediation operating and value added approaches LNDEPO is a measure of banks market share calculated as a natural logarithm of total bank deposits LOANSTA is a measure of banks loans intensity calculated as the ratio of total loans to bank total assets LNTA is the size of the banks total asset measured as the natural logarithm of total bank assets LLPTL is a measure of banks risk calculated as the ratio of total loan loss provisions divided by total loans NIETA is a measure of bank management quality calculated as total noninterest expenses divided by total assets NIITA is a measure of banks diversification towards non interest income calculated as total noninterest income divided by total assets EQUITYTA is a measure of banks leverage intensity measured by banks total shareholders equity divided by total assets ROA is a proxy measure for bank profitability calculated as bank profit after tax divided by total assets and LNGDP is natural logarithm of gross domestic product Values in parentheses are standard errors and indicate significance at 1 5 and 10 levels respectively 68 F Sufian Research in International Business and Finance 23 2009 5477 others Hauner 2005 and Pasiouras 2008ab QML HuberWhite standard errors and covariates are calculated Several general comments regarding the test results are warranted The model performs reasonably well in at least two respects For one results for most variables remain stable across the various regressions tested The R2s are also reasonably high ranging from 43 to 47 The findings sug gest that all explanatory variables have the expected signs and in most cases are statistically different from zero In models 27 regressions when we add the other group of variables to the baseline specification that include the bank specific attribute variables the coefficients of the baseline variables stay mostly the same they keep the same sign the same order of magnitude they remain significant as they were so in model 1 regressions albeit sometimes at different levels and with few exceptions do not become significant if they were not in model 1 regressions Therefore for models 27 regressions we will only discuss the results of the new variables added to the baseline specification From Table 8 it is observed that LNDEPO reveals a negative relationship and is statistically significant in intermediation approach regression model suggesting that the more efficient banks are associated to the banks with lower market share thus diminishing market leadership argument The results imply that banks with small market share like the foreign banks can be at least as efficient as market dominant banks in their intermediation function because maintaining or expanding market share might involve extra costs and inputs and thus exacerbating inefficiency LOANSTA reveals positive relationship and is statistically significant in all regression models The findings imply that banks with higher loanstoasset ratios tend to have higher efficiency scores Thus bank loans seem to be more highly valued than alternative bank outputs ie investments and securities The positive relationship found between technical efficiency and LOANSTA may be supporting the efficient market hypothesis Market power in loan markets may be the result of efficient operations Due to their ability to manage operations more productively relatively efficient banks might have lower production costs which enable them to offer more reasonable loan terms and ultimately gaining larger market shares over inefficient banks Likewise LNTA shows positive coefficients and is statistically significant in the intermediation approach regression model suggesting that the larger the bank the more efficient the bank will be purely because of the economies of scale arguments Hauner 2005 offers two potential explanations for which size could have a positive impact on bank efficiency First if it relates to market power large banks should pay less for their inputs Second there may be increasing returns to scale through the allocation of fixed costs eg research or risk management over a higher volume of services or from efficiency gains from a specialized workforce Thus assuming that the average cost curve for Malaysian banks is Ushaped the recent growth policies of the small and medium Malaysian banks seem to be consistent with cost minimization As expected LLPTL shows negative relationship with banks technical efficiency and is statisti cally significant in the operating approach regression model The finding is consistent with earlier findings by among others Kwan and Eisenbeis 1995 Resti 1997 and Barr et al 2002 Further more most research conducted on explaining the causes of bank or thrift industry failures have found that failing institutions carried a large proportion of nonperforming loans in their books prior to failure DermigucKunt 1989 Whalen 1991 Barr and Siems 1994 Berger and Humphrey 1992 Barr and Siems 1994 and Wheelock and Wilson 1995 have also found that banks approaching failure tend to have low cost efficiency and experience high ratios of problem loans and that failing banks tend to be located far from the best practice frontiers The results imply that Malaysian banks should focus more on credit risk management which has been proven to be problematic in the recent past Serious banking problems have arisen from the failure of banks to recognize impaired assets and create reserves for writing off these assets An immense help towards smoothing these anoma lies would be provided by improving the transparency of the financial systems which in turn will assist banks to evaluate credit risk more effectively and avoid problems associated with hazardous exposure The empirical findings seem to suggest that NIITA consistently possesses positive relationship with banks technical efficiency and are statistically significant in all regression models The findings also suggest that the elasticity of technical efficiency with respect to NIITA is quite high particularly in the operating approach and value added approach regression models The results imply that banks tend F Sufian Research in International Business and Finance 23 2009 5477 69 to become more managerially efficient as they increase their income emanating from noninterest sources The empirical finding is consistent with earlier findings by among others Jeon and Miller 2005 The findings seem to suggest that NIETA consistently exhibit negative relationship with bank effi ciency and are statistically significant in the operating approach and value added approach regression models The finding is in consonance with the bad management hypothesis of Berger and DeYoung 1997 Low measure of technical efficiency is a signal of poor senior management practices which apply to inputusage daytoday operations and managing the loan portfolio Sub par managers do not sufficiently monitor and control their operating expenses Managers in these banks might not practice adequate loan underwriting monitoring and control Clearly efficient cost management is a prerequisite for the improved efficiency of the Malaysian banking system ie the high elasticity of efficiency to this variable denotes that banks have much to gain if they improve their managerial prac tices Furthermore the Malaysian banking system has not reached the maturity level required to link quality effects pending from increased spending to higher bank efficiency From Table 8 it is clear that EQASS exhibits negative relationship with bank efficiency and is sta tistically significant in the value added approach regression model which is in line with the findings of Akhigbe and McNulty 2005 The findings imply that the more efficient banks ceteris paribus use more leverage less equity compared to their peers The results seems to suggest that the less efficient banks could have been involved in riskier operations and in the process tend to hold more equity vol untarily or involuntarily ie the reason might be banks deliberate efforts to increase safety cushions and in turn decrease the cost of funds or perhaps regulatory pressures that mandate riskier banks to carry more equity It is observed that ROA exhibits positive relationship and is statistically significant in the operat ing and value added approaches regression models These findings indicate that the more profitable banks tend to exhibit lower inefficiency which corroborates similar findings of some previous studies Isik and Hassan 2002 Hasan and Marton 2003 Miller and Noulas 1996 Banks reporting higher profitability ratios are usually preferred by clients and therefore attract the biggest share of deposits as well as the best potential creditworthy borrowers Such conditions create a favourable environment for the profitable banks to be more efficient from the point of view of intermediation activities Another factor which could explain Malaysian banks inefficiency is the relatively volatile rates of national income growth during the period of analysis measured by LNGDP which exhibit negative and statistically significant relationship in all regression models Demand for financial services tends to grow as economies expand and societies become wealthier However during the period of study the Malaysian economy had experienced a volatile economic growth ie from a robust 995 average growth in 1995 and 1996 declining to record 730 growth in 1997 falling into a recession in 1998 before recovering to register 615 growth in 1999 The volatile economic growth could have resulted in banks to suffer from lower demand for their financial services increased loan defaults and thus lower output 43 Bank ownership and efficiency Banks of different ownership forms may react differently to the same efficiency determinants Thus in the preceding analysis we repeat the regression models above to examine the factors that influence the efficiency of the foreign banks publicly listed banks and government linked banks The regression results focusing on the relationship between bank ownership and efficiency are presented in Table 9 As expected DUMFORB entered the regression models positively and is statistically significant in the value added approach regression models The findings imply that banks with controlling share of foreign ownership are likely to be more efficient compared to their domestically owned counter parts This should come as no surprise because of the ability of foreign owned banks to capitalize on their access to better risk management and operational techniques which is usually made available through their parent banks abroad In addition since foreign ownership is likely to be concentrated foreign owned banks are less prone to typical corporate governance conflicts dispersed owners and the management The evidence seems to suggest that foreign owned banks are more likely to cherry pick the best borrowers available in the market especially those from their own countries of origin 70 F Sufian Research in International Business and Finance 23 2009 5477 Table 9 Multivariate Tobit regression analysis Explanatory variables Model 2 Model 3 Intermediation approach Operating approach Value added approach Intermediation approach Operating approach Value added approach CONSTANT 7511 2102 4484 1142 3427 1104 7362 3943 4555 3683 8068 3414 Bank characteristics LNDEPO 0228 0052 0020 0028 0032 0027 0237 5169 0022 0946 0043 0900 LOANSTA 0533 0109 0179 0059 0130 0057 0537 5029 0197 2820 0483 4035 LNTA 0206 0056 0012 0031 0022 0029 0220 4155 0007 0287 0022 0397 LLPTL 0060 0291 0284 0158 0146 0153 0086 0593 0313 0936 0369 1270 NIITA 0020 2267 2396 1232 1676 1191 0039 0016 3485 2852 0579 0243 NIETA 15800 2324 10022 1263 7688 1220 14930 4829 11142 8674 8475 2681 EQASS 0537 0449 0273 0244 0534 0236 0467 0887 0333 1037 1650 2735 ROA 0005 0009 0033 0005 0032 0005 0005 0648 0034 4631 0008 0669 Economic conditions LNGDP 0629 0194 0327 0106 0023 0102 0624 3536 0324 2826 0763 3413 Bank ownership DUMFORB 0012 0030 0023 0016 0047 0016 DUMPUBL 0039 1512 0027 1371 0061 1449 DUMGOVT No of observations 171 171 171 171 171 171 Log likelihood 81478 185791 191615 81970 185678 63024 R2 0496 0506 0486 0498 0505 0287 Adj R2 0461 0472 0451 0464 0471 0237 Explanatory variables Model 4 Intermediation approach Operating approach Value added approach CONSTANT 7584 4057 4504 3628 7667 3256 Bank characteristics LNDEPO 0232 5108 0026 1096 0050 1057 LOANSTA 0539 5087 0193 2718 0485 4083 LNTA 0212 4114 0017 0665 0032 0604 LLPTL 0039 0270 0300 0897 0455 1715 NIITA 0293 0122 3130 2553 0291 0125 NIETA 16143 5576 10481 8440 10450 3769 EQASS 0535 1029 0278 0867 1548 2675 ROA 0005 0583 0033 4448 0009 0739 Explanatory variables Model 4 Intermediation approach Operating approach Value added approach Economic conditions LNGDP 0638 3600 0328 2852 0734 3307 Bank ownership DUMFORB DUMPUBL DUMGOVT 0023 0950 0012 0620 0052 1537 No of observations 171 171 171 Log likelihood 81661 185017 62979 R2 0497 0501 0286 Adj R2 0462 0467 0267 θjt α β1 LNDEPO β2 LOANSTA β3 LNTA β4 LLPTL β5 NIETA β6 EQUITYTA β7 ROA β8 LOGGDP β9 DUFORB β10 DUPUBL β11 DUMGOVT εj The dependent variable is banks efficiency scores derived from DEA intermediation operating and value added approaches LNDEPO is a measure of banks market share calculated as a natural logarithm of total bank deposits LOANSTA is a measure of banks loans intensity calculated as the ratio of total loans to bank total assets LNTA is the size of the banks total asset measured as the natural logarithm of total bank assets LLPTL is a measure of banks risk calculated as the ratio of total loan loss provisions divided by total loans NIETA is a measure of bank management quality calculated as total noninterest expenses divided by total assets NIITA is a measure of banks diversification towards non interest income calculated as total noninterest income divided by total assets EQUITYTA is a measure of banks leverage intensity measured by banks total shareholders equity divided by total assets ROA is a proxy measure for bank profitability calculated as bank profit after tax divided by total assets LNGDP is natural logarithm of gross domestic product DUMFORB DUMPUBL and DUMGOVT are dummy variables that take a value of 1 for foreign banks publicly listed banks and government related banks respectively 0 otherwise Values in parentheses are standard errors and indicate significance at 1 5 and 10 levels respectively 72 F Sufian Research in International Business and Finance 23 2009 5477 thereby improving the quality of their loan portfolios The empirical observation that foreign banks perform better than domestic banks in developing countries also implies the technical savvy of banks from developed countries generally overcomes the home field advantage in developing countries especially when the domestic economy has relatively unsophisticated financial markets and institu tions Jeon and Miller 2005 The results are in accordance with earlier findings by Isik and Hassan 2003b on Turkish banks Hasan and Marton 2003 on Hungarian banks and Sathye 2003 on Indian banks To investigate the relationship between publicly traded and Malaysian bank efficiency a binary dummy variable DUMPUBL is introduced as an explanatory variable in model 2 regressions It is observed from columns 46 of Table 9 that the variable entered the regression models with a negative sign but is never significant in all the regression models Although the market discipline hypothesis suggests that banks whose shares are publicly traded should exhibit higher efficiency the empirical findings seem to suggest that the Malaysian capital market exerts no discipline over bank management The results are in accord with Chu and Lim 1998 who suggests that stock markets respond more strongly to profit rather than cost efficiency And finally to examine the relationship between government linked banks and efficiency DUM GOVT is included in model 3 regressions The results are presented in columns 79 of Table 9 The estimated coefficients entered the regression models with a negative sign but are never significant in any of the regression models 44 Bank origins and efficiency As pointed out by Berger et al 2000 the methodologies used in the previous studies may not be able to distinguish properly the home field and global advantage hypotheses Following the procedures set by Berger et al 2000 we address this drawback by distinguishing among nations of origin of foreign banks to test for the limited form of the global advantage hypothesis Accordingly we repeat Eq 5 by further classifying foreign banks operating in Malaysia into three major groups according to their parents continents In each set these regressions are performed by considering each bank origin at a time The regression results are presented in Table 10 Under the intermediation approach the empirical findings seem to suggest that foreign banks from the Americas and Europe are relatively more efficient compared to their domestic bank counterparts On the other hand efficiency is negatively related to foreign banks from Asian countries suggesting that foreign banks from other Asian countries were the least efficient banking group during the period of study Unlike the intermediation approach it is observed from Table 10 that foreign banks from the Asian countries were the most efficient banking group under the operating approach and is statistically significant at the 1 level of significance This is followed by banks from the Americas although is not statistically significant at any conventional levels It is clear from column 5 of Table 10 that efficiency is negatively related to banks from other European countries statistically significant at the 1 level implying that banks from the European countries were the least efficient banking group under the operating approach Columns 79 of Table 10 presents the regression results for the value added approach Similar to the operating approach the empirical findings seem to suggest that banks from other Asian countries to be the most efficient banking group statistically significant at the 1 level followed by banks from the Americas Again it is clear from column 8 of Table 10 that efficiency is negatively related to banks from other European countries under the value added approach It is worth highlighting several important points from the findings Firstly the results highlight the importance to differentiate foreign banks according to their origins The empirical findings clearly sug gest that generalizations made in regard to the higher efficiency levels of the foreign banks without tak ing into account their nations or regions of origins could lead to bias conclusions And secondly differ ent inputs and outputs may result in different conclusions The empirical findings from this study sug gest that foreign banks from European countries were found to be relatively more efficient than their domestic bank counterparts under the intermediation approach but were relatively inefficient under the value added and operating approaches Likewise although foreign banks from Asian countries were found to be relatively efficient compared to their domestic bank peers under the value added and operating approaches they were found to be relatively inefficient under the intermediation approach F Sufian Research in International Business and Finance 23 2009 5477 73 Table 10 Multivariate Tobit regression analysis Explanatory variables Model 5 Model 6 Intermediation approach Operating approach 1 2 3 1 2 3 CONSTANT 7562 2105 7520 2105 7504 2107 4464 1151 4376 1139 4199 1108 Bank characteristics LNDEPO 0207 0070 0232 0051 0231 0056 0025 0038 0025 0028 0067 0029 LOANSTA 0551 0108 0541 0107 0542 0113 0194 0059 0194 0058 0121 0060 LNTA 0184 0073 0208 0056 0207 0061 0014 0040 0014 0303 0062 0032 LLPTL 0050 0291 0050 0292 0054 0295 0292 0159 0266 0158 0198 0155 NIITA 0342 2016 0314 2103 0470 2105 3204 1102 3793 1139 1979 1107 NIETA 16133 2280 16113 2321 16051 2630 10417 1246 9954 1258 7836 1382 EQASS 0540 0449 0523 0458 0539 0455 0282 0245 0365 0248 0429 0239 ROA 0005 0009 0005 0009 0005 0009 0033 0005 0033 0005 0032 0005 Economic conditions LNGDP 0632 0194 0628 0194 0626 0194 0322 0106 0314 0105 0306 0102 Bank nationality DUMAMER 0031 0059 0001 0038 DUMEURO 0009 0040 0038 0022 DUMASIA 0001 0040 0078 0021 No of observations 171 171 171 171 171 171 Log likelihood 81532 81418 81396 184796 186272 191337 R2 0496 0495 0495 0500 0509 0537 Adj R2 0461 0460 0460 0465 0475 0505 Explanatory variables Model 7 Value added approach 1 2 3 CONSTANT 3446 1130 3363 1132 3125 1091 Bank characteristics LNDEPO 0018 0021 0043 0028 0083 0029 LOANSTA 0170 0058 0160 0057 0088 0059 LNTA 0004 0023 0028 0030 0075 0032 LLPTL 0167 0156 0156 0157 0069 0153 NIITA 3202 1083 3446 1131 2098 1090 Table 10 Continued Explanatory variables Model 7 Value added approach 1 2 3 NIETA 8603 1224 8367 1248 5938 1362 EQASS 0551 0181 0572 0246 0697 0236 ROA 0032 0005 0032 0005 0031 0005 Economic conditions LNGDP 0219 0104 0211 0105 0197 0101 Bank nationality DUMAMER 0031 0032 DUMEURO 0009 0022 DUMASIA 0077 0021 No of observations 171 171 171 Log likelihood 187862 187465 193914 R2 0463 0461 0500 Adj R2 0426 0423 0465 θjt α β1 LNDEPO β2 LOANSTA β3 LNTA β4 LLPTL β5 NIETA β6 EQUITYTA β7 ROA β8 LOGGDP β9 DUMAMER β10 DUMEURO β11 DUMASIA εj The dependent variable is banks efficiency scores derived from DEA intermediation operating and value added approaches LNDEPO is a measure of banks market share calculated as a natural logarithm of total bank deposits LOANSTA is a measure of banks loans intensity calculated as the ratio of total loans to bank total assets LNTA is the size of the bank total asset measured as the natural logarithm of total bank assets LLPTL is a measure of banks risk calculated as the ratio of total loan loss provisions divided by total loans NIETA is a measure of bank management quality calculated as total noninterest expenses divided by total assets EQUITYTA is a measure of banks leverage intensity measured by banks total shareholders equity divided by total assets ROA is a proxy measure for bank profitability calculated as bank profit after tax divided by total assets LNGDP is natural logarithm of gross domestic product DUMAMER DUMEURO and DUMASIA are dummy variables that take a value of 1 for foreign banks originated from the Americas European countries and Asia respectively 0 otherwise Values in parentheses are standard errors and indicate significance at 1 5 and 10 levels respectively F Sufian Research in International Business and Finance 23 2009 5477 75 In essence the findings which suggest that the American banks were relatively more efficient compared to other foreign banks and their domestic bank peers do not lend support to the limited form of global advantage hypothesis Berger et al 2000 argued that under the limited form of global advantage hypothesis only the efficient institutions in one or a limited number of nations with specific favourable market or regulatory conditions in their home countries can operate more efficiently than domestic banks in other nations 5 Concluding remarks The present study investigates the efficiency of the Malaysian banking sector around the Asian financial crisis 1997 The efficiency estimates of individual banks are evaluated by using the DEA approach Three different approaches viz intermediation approach value added approach and oper ating approach have been employed to differentiate how efficiency scores vary with changes in inputs and outputs A multivariate Tobit model is employed to examine the relationship between efficiency scores derived from the DEA to a set of explanatory variables ie bank size profitability and ownership The estimates of technical efficiency were observed to be consistently higher under operating approach visàvis the intermediation and value added approaches On the other hand under the intermediation approach banks are characterized by relatively low level of technical efficiency The empirical findings clearly bring forth the high degree of inefficiency of the Malaysian banking sector around the East Asian crisis period particularly a year after the East Asian crisis The results suggest that although in general technical efficiency level seems to deteriorate abruptly a year after the East Asian crisis under all approaches the deterioration seems to be more pronounced under the intermediation approach model The empirical findings suggest that the number of efficient banks under CRS technical efficiency technology and VRS technology pure technical efficiency differs markedly irrespective of the choice of various inputs and outputs suggesting the existence of sizable scale inefficiency among Malaysian commercial banks during the East Asian crisis period During the preand postcrisis periods scale inefficiency seems to outweigh pure technical inefficiency in determining the total technical ineffi ciency of the Malaysian banking sector under the intermediation approach On the other hand under the value added approach while scale inefficiency outweighs pure technical inefficiency in determin ing the total technical inefficiency of the Malaysian banking sector during the precrisis period the empirical findings seem to suggest that pure technical inefficiency outweigh scale inefficiency during the postcrisis period Finally under the operating approach although pure technical efficiency is gen erally lower during the precrisis period the trend is less clear during the postcrisis period Thus with respect to their scale of operations Malaysian banks are likely to lose sizeable output particularly in the case of the intermediation approach although at a lesser degree under the value added approach The results from the multivariate regression analysis suggest that technical efficiency is positively and significantly associated with loans intensity suggesting that banks with higher loanstoasset ratios exhibits higher efficiency scores On the other hand it appears that the expense preference behaviour not to be holding in the Malaysian banking sector thus supporting Berger and DeYoungs 1997 bad management hypothesis The findings clearly suggest that efficient cost management to be a prerequisite for the improved efficiency of the Malaysian banking system and that banks have much to gain if they improve their managerial practices Furthermore the Malaysian banking system has not reached the maturity level required to link quality effects pending from increased spending to higher bank efficiency Similarly LNGDP is also negatively related to Malaysian banks efficiency lev els This could be explained by the volatile economic growth which could have resulted in banks to suffer from lower demand for their financial services increases loan defaults and thus lower output We find that the foreign banks have succeeded in capitalizing on their advantages and exhibit a higher level of efficiency compared to their domestic bank peers However we do not find evidence of higher efficiency levels of the publicly listed banks Similarly efficiency is not significantly related to the government link banks The empirical findings do not support for the limited form of global advantage hypothesis and rejects the home field advantage hypothesis as the results suggest that the American banks are relatively more efficient compared to their domestic bank counterparts 76 F Sufian Research in International Business and Finance 23 2009 5477 Due to its limitations the paper could be extended in a variety of ways Firstly the scope of this study can be extended to examine the efficiency of the conventional banks compared to the Islamic banks Secondly the present study can also be extended to estimate the determinants of pure technical and scale efficiency along with the technical efficiency estimates Finally investigation of changes in productivity over time as a result of technical change or technological progress or regress by employing the Malmquist Productivity Index MPI could also be another extension to the present paper Acknowledgements We would like to thank an anonymous reviewer for valuable comments that helped improve an earlier version of the manuscript Any remaining errors are of course our own References Akhigbe A McNulty JE 2005 Profit efficiency sources and differences among small and large US commercial banks J Econ Finan 29 3 289299 Alba P Hernandez L Klingebiel D 1999 Financial liberalization and the capital account Thailand 198897 The World Bank Working Paper AlSharkas AA Hassan MK Lawrence S 2008 The impact of mergers and acquisitions on the efficiency of the US banking industry further evidence J Business Finan Acc 35 5070 Ataullah A Le H 2006 Economic reforms and bank efficiency in developing countries the case of the Indian banking industry Appl Finan Econ 16 653663 Avkiran NK 1999 The evidence on efficiency gains the role of mergers and the benefits to the public J Bank Finan 23 9911013 Banker RD Charnes A Cooper WW 1984 Some models for estimating technical and scale inefficiencies in data envelopment analysis Manage Sci 30 9 10781092 Barr R Siems T 1994 Predicting bank failure using DEA to quantify management quality Working Paper Federal Reserve Bank of Dallas Barr RS Killgo KA Siems TF Zimmel SA 2002 Evaluating the productive efficiency and performance of US commercial banks Manage Finan 28 8 325 Barth JR Brumbaugh DR Suaerhaft D Wang G 1985 Thrift Institution Failures Causes and Policy Issues Bank Structure and Competition Federal Reserve Bank of Chicago Barth JR Caprio J Levine R 2000 Banking Systems around the globe do regulation and ownership affect performance and stability In Mishkin F Ed Prudential Regulation and Supervision What Works and What Doesnt National Bureau of Economic Research Washington DC Beck TA DermigucKunt A Levine R 2006 Bank concentration competition and crises first results J Bank Finan 30 15811603 Benston GJ 1965 Branch banking and economies of scale J Finan 20 2 312331 Berg A Pattillo C 1999 Predicting currency crises the indicators and an alternative J Int Money Finan 18 561586 Berger AN DeYoung 1997 Problem loans and cost efficiency in commercial banks J Bank Finan 21 6 849870 Berger AN Humphrey DB 1992 Measurement and efficiency issues in commercial banking In Griliches Z Ed Measure ment Issues in the Service Sectors National Bureau of Economic Research University of Chicago Press pp 245279 Berger AN Humphrey DB 1997 Efficiency of financial institutions international survey and directions for future research Eur J Oper Res 98 175212 Berger AN DeYoung R Genay H Udell GF 2000 Globalization of financial institutions evidence from cross border banking performance BrookingsWharton Paper Finan Serv 3 23158 Berger AN Hunter WC Timme SG 1993 The efficiency of financial institutions a review and preview of research past present and future J Bank Finan 17 221249 Berger AN 2007 International comparisons of banking efficiency Finan Mark Instit Instrum 16 119144 Bhattacharya A Lovell CAK Sahay P 1997 The impact of liberalization on the productive efficiency of Indian commercial banks Eur J Oper Res 98 332345 Chang R Velasco A 1999 Liquidity crises in emerging markets theory and policy NBER Working Paper Chinn MD 1998 Before the fall were East Asian currencies overvalued NBER Working Paper Chu SF Lim GH 1998 Share performance and profit efficiency of banks in an oligopolistic market evidence from Singapore J Multinatl Finan Manage 8 155168 Coelli T 1996 A Guide to DEAP Version 21 A Data Envelopment Analysis Computer Program Working Paper CEPA University of New England Armidale Coelli T PrasadaRao DS Battese GE 1998 An Introduction to Efficiency and Productivity Analysis Kluwer Academic Publishers Boston Cooper WW Seiford LM Tone K 2000 Data Envelopment Analysis Kluwer Academic Publishers Boston Corsetti G Pesenti P Roubini N 1998 What caused the Asian currency and financial crisis NBER Working Paper Daniel BC Jones JB 2007 Financial liberalization and banking crises in emerging economies J Int Econ 72 1 202221 DermigucKunt A Detragiache E Gupta A 2006 Inside the crisis an empirical analysis of banking systems in distress J Int Money Finan 25 5 702718 DermigucKunt A 1989 Deposit institutions failure a review of the empirical literature Federal Reserve Bank Cleveland Econ Rev 25 4 218 F Sufian Research in International Business and Finance 23 2009 5477 77 Drake L Hall MJB 2003 Efficiency in Japanese banking an empirical analysis J Bank Finan 27 891917 Drake L Hall MJB Simper R 2006 The impact of macroeconomic and regulatory factors on bank efficiency a nonparametric analysis of Hong Kongs banking system J Bank Finan 30 5 14431466 Eichengreen B Rose AK 1998 Staying afloat when the wind shifts external factors and emerging market banking crises NBER Working Paper Frankel JA Rose AK 1996 Currency crashes in emerging markets empirical indicators NBER Working Paper Fukuyama H 1995 Measuring efficiency and productivity growth in Japanese banking a nonparametric approach Appl Finan Econ 5 2 95107 Goldstein M Turner P 1996 Banking Crises in Emerging Economies Origins and Policy Options BIS Economic Papers GonzalezHermosillo B Pazarbasioglu C Billings R 1997 Determinants of banking system fragility a case study of Mexico Int Monet Fund 44 295314 GonzalezHermosillo B 1999 Determinants of exante banking system distress a macro micro empirical exploration of some research episodes IMF Working Paper Hasan I Marton K 2003 Development and efficiency of the banking sector in a transitional economy a Hungarian experience J Bank Finan 27 12 22492271 Hauner D 2005 Explaining efficiency differences among large German and Austrian banks Appl Econ 37 9 969980 Honohan P 1997 Banking system failures in developing and transition countries diagnosis and prediction BIS Working Paper Humphrey DB Pulley LB 1997 Banks responses to deregulation profits technology and efficiency J Money Credit Bank 29 7393 Isik I Hassan MK 2003a Financial disruption and bank productivity the 1994 experience of Turkish banks Q Rev Econ Finan 43 291320 Isik I Hassan MK 2003b Efficiency ownership and market structure corporate control and governance in the Turkish banking industry J Business Finan Acc 30 910 13631421 Isik I Hassan MK 2002 Technical scale and allocative efficiencies of Turkish banking industry J Bank Finan 26 4 719766 Jemric I Vujcic B 2002 Efficiency of banks in Croatia a DEA approach Comp Econ Stud 44 169193 Jeon Y Miller SM 2005 Performance of domestic and foreign banks the case of Korea and the Asian financial crisis Global Econ Rev 34 2 145165 Kaminsky GL Reinhart CM 1998 Financial crises in Asia and Latin America then and now Am Econ Rev 88 2 444448 Kaminsky GL Lizondo S Reinhart CM 1998 Leading indicators of currency crises IMF Staff Papers 45 1 148 Kroszner RS Laeven L Klingebiel D 2007 Banking crises financial dependence and growth J Finan Econ 84 1 187228 Krugman P 1999 Currency crises NBER Conference Report Series Kwan SH Eisenbeis R 1995 An analysis of inefficiencies in banking J Bank Finan 19 34 733734 Kyj L Isik I 2008 Bank Xefficiency in Ukraine an analysis of service characteristics and ownership J Econ Business 60 369393 Lane WR Looney SW Wansley JW 1986 An application of the Cox proportional hazards and international overborrowing J Bank Finan 10 4 511531 Leightner JE Lovell CAK 1998 The impact of financial liberalization on the performance of Thai banks J Econ Business 50 115131 LozanoVivas A Pasiouras F 2008 The impact of nontraditional activities on the estimation of bank efficiency international evidence Working Paper University of Bath Miller SM Noulas AG 1996 The technical efficiency of large bank production J Bank Finan 20 3 495509 Pasiouras F 2008a International evidence on the impact of regulations and supervision on banks efficiency an application of twostage data envelopment analysis Rev Quant Finan Account 30 2 187223 Pasiouras F 2008b Estimating the technical and scale efficiency of Greek commercial banks the impact of credit risk off balance sheet activities and international operations Res Int Business Finan 22 3 301318 Pasiouras F Liadaki A Zopounidis C 2008 Bank efficiency and share performance evidence from Greece Appl Finan Econ 18 14 11211130 Resti A 1997 Evaluating the cost efficiency of the Italian banking system what can be learned from the joint application of parametric and nonparametric techniques J Bank Finan 21 2 221250 Sachs JA Tornell A Velaso A 1996 Financial crises in emerging markets the lessons from 1995 NBER Working Paper Sathye M 2003 Efficiency of banks in a developing economy the case of India Eur J Oper Res 148 3 662671 Saxonhouse GR 1976 Estimated parameters as dependent variables Am Econ Rev 66 1 178183 Sinkey JF 1975 A multivariate statistical analysis of the characteristics of problem banks J Finan 30 1 2136 Suetorsak R 2006 Banking crisis in East Asia a micromacro perspective Rev Quant Finan Account 26 219248 Sufian F 2007 Trends in the efficiency of Singapores commercial banking groups a nonstochastic frontier DEA window analysis approach Int J Product Perform Manage 56 99136 Tanner E 2000 Exchange market pressure and monetary policy Asia and Latin America in the 1990s IMF Staff Papers 47 3 311333 Thanassoulis E 2001 Introduction to the Theory and Application of Data Envelopment Analysis A Foundation Text with Integrated Software Kluwer Academic Publishers Boston Thomson JB 1991 Predicting Bank Failures in the 1980s Federal Reserve Bank Cleveland Econ Rev 27 1 920 Whalen G 1991 A proportional hazards model of bank failure an examination of its usefulness as an early warning tool Federal Reserve Bank Cleveland Econ Rev 27 1 2131 Wheelock DC Wilson PW 1995 Explaining bank failures deposit insurance regulation and efficiency Rev Econ Stat 77 4 689700 Revista Capital Científico Eletrônica RCCe ISSN 21774153 Disponível em revistasunicentrobrindexphpcapitalcientificoindex www3unicentrobrppgadm www3unicentrobr Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil Relationship between efficiency in cost management measured by the method of Stochastic Frontier Analy sis SFA and the profitability of financial institutions in Brazil Douglas José Mendonça1 Júlia Alves e Souza2 Francisval De Melo Carvalho3 e Gideon Carvalho de Benedicto4 Recebido em 17122017 Revisado em 10022018 Aprovado em 28032018 Disponível em 01072018 1 Universidade Federal de Lavras Brasil Doutorando em Administração email mendoncadouglasyahoocombr 2 Universidade Federal de Lavras Brasil Doutoranda em Administração email juliajasouzagmailcom 3 Universidade Federal de Lavras Brasil Doutor em Administração email francarvdaeuflabr 4 Universidade Federal de Lavras Brasil Doutor em Contabilidade e Controladoria email gideonbenedictogmailcom Resumo Este estudo tem por objetivo investigar a relação entre a eficiência na gestão de custos e a rentabilidade das instituições financeiras bancárias no Brasil Utilizouse a abordagem de intermediação para a estimação da eficiência na gestão de custos das instituições financeiras aplicando o método da Análise da Fronteira Estocástica SFA modelado com indicadores econômicofinanceiros Para investigar a relação entre eficiência e a rentabilidade foi ajustado um modelo de regressão linear múltipla com dados em painel A amostra da pesquisa é composta por 47 instituições financeiras classificadas como banco múltiplo comercial e Caixa Econômica e o período analisado compreende os anos de 2008 a 2015 Diante da aplicação do modelo econométrico foi identificada a existência da relação direta e estatisticamente significativa entre a eficiência na gestão de custos e a rentabilidade das instituições financeiras indicando que maiores níveis de eficiência na gestão de custos estão relacionados a maiores níveis de rentabilidade Os sinais obtidos para os coeficientes da variável de investigação e das variáveis de controle corroboraram as expectativas baseadas na literatura científica Assim os achados deste estudo indicam que a eficiência na gestão de custos é um fator determinante para melhorar a rentabilidade do sistema bancário brasileiro Palavraschave Eficiência Rentabilidade Instituições Financeiras Gestão de Custos Análise da Fronteira Estocástica Abstract This study aims to investigate the relationship between efficiency in cost management and the profitability of banking financial institutions in Brazil The intermediation approach was used to estimate the efficiency in cost management of financial institutions applying the Stochastic Frontier Analysis SFA method modeled with economicfinancial indicators To investigate the relationship between efficiency and profitability a multiple linear regression Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 6 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 model with panel data was fitted The research sample is made up of 47 financial institutions classified as multiple commercial and Caixa Econômica and the period analyzed covers the years from 2008 to 2015 In view of the application of the econometric model the existence of a direct and statistically significant relationship was identified between efficiency in the management of costs and the profitability of financial institutions indicating that higher levels of efficiency in the management of costs are related to higher levels of profitability The signals obtained for the coefficients of the research variable and the control variables corroborated the expectations based on the scientific literature Thus the findings indicate that efficiency in cost management is a determining factor to improve the profitability of the Brazilian banking system Keywords Efficiency Profitability Financial Institutions Cost Management Stochastic Frontier Analysis 1 INTRODUÇÃO As instituições financeiras bancárias são importantes para o desenvolvimento econômico de um país Assaf Neto 2012 afirma que os bancos são protagonistas na prestação de serviços de intermediação financeira e por esse motivo auxiliam no financiamento da economia A intermediação financeira é o processo que utiliza os recursos financeiros acumulados dos agentes poupadores e os coloca novamente na economia em forma de empréstimos para os agentes tomadores Como fruto dessa operação as instituições bancárias recebem o spread já que captam recursos a uma determinada taxa de juros e os emprestam a uma outra taxa que é superior à de captação PHILIPPON 2015 Devido ao processo de evolução do mercado financeiro a intermediação financeira deixou de ser a única fonte de receita de tais instituições SCHOLTENS WENSVEEN 2000 Nesse sentido Brighi e Venturelli 2015 afirmam que as evoluções nesse mercado resultaram em uma expansão do escopo das atividades bancárias que passaram a ir além da fonte de renda tradicional ganhos de spread para incorporar também a utilização de fontes nãotradicionais de renda tais como seguros e previdência privada As implicações de tais mudanças nas fontes de receitas dos bancos têm sido amplamente abordadas na literatura embora ainda não exista um consenso sobre os efeitos gerados na rentabilidade das instituições financeiras BRIGHI VENTURELLI 2015 Nota se que essa rentabilidade é alvo de muitas discussões no mercado financeiro e na literatura científica Conforme Doumpos e Cohen 2014 a literatura científica na área de gestão tem explorado a relação entre eficiência e rentabilidade bancária pois a eficiência é um potencial fator determinante do desempenho financeiro das instituições Segundo Dietrich e Wanzenried 2011 a rentabilidade é um critério importante para analisar o resultado das instituições financeiras pois permite identificar o sucesso das estratégias competitivas das instituições que operam em ambientes similares Já a eficiência é um conceito que tem relação com o modo de se fazer as coisas representando a relação entre a quantidade produzida e os recursos consumidos dessa forma a eficiência compara o que foi produzido dado os recursos disponíveis com o que poderia ter sido produzido com os mesmos recursos MELLO et al 2005 Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 7 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 A eficiência na gestão de custos é mensurada por meio da estimação de uma fronteira ótima de custo Uma vez que essa fronteira é estimada ela se torna o padrão em relação ao qual é medida a eficiência das empresas A fronteira de custo indica o custo mínimo necessário para produzir um conjunto de produtos ou serviços dados os preços dos insumos o nível de produção e a tecnologia existente NGUYEN et al 2016 Assim as empresas consideradas mais eficientes na gestão de custos são aquelas que incorrem nos menores custos para produzirem uma determinada quantidade de produtos ou serviços Nesse sentido Philippon 2015 afirma que a eficiência na gestão dos custos oriundos da operação de intermediação financeira impacta diretamente o desenvolvimento da economia no longo prazo O custo de intermediação financeira afeta o custo do utilizador do financiamento externo DIAMOND 1984 Uma gestão eficiente dos custos das instituições financeiras pode representar taxas menores de repasse afetando o fluxo monetário e contribuindo para a redução do custo de capital dos agentes tomadores de empréstimos O método da Análise da Fronteira Estocástica SFA é utilizado na literatura para mensurar a eficiência na gestão de custos consistindo em uma metodologia paramétrica e estocástica projetada para avaliar eficiência AIGNER LOVELL SCHMIDT 1977 O que a SFA faz é estimar a função da fronteira ótima de custo utilizandose de regressões com múltiplas variáveis Para isso tal método trata os insumos como variáveis independentes e o produto ou indicador que sintetize vários produtos como variável dependente e faz uma comparação da eficiência das empresas analisadas em relação aos padrões técnicos ou comportamentais das próprias empresas SILVA 2014 Diante desse contexto o presente artigo tem como objetivo investigar a relação entre a eficiência na gestão de custos mensurada pela SFA e a rentabilidade das instituições financeiras bancárias no Brasil Com base na literatura e nas evidências de estudos anteriores KIRKWOOD NAHM 2006 DOLIVEIRA 2014 MENDONÇA et al 2017 é formulada a hipótese de que instituições mais eficientes na gestão de custos tendem a ter maior rentabilidade Nessa perspectiva pretendese responder à seguinte questão qual a relação existente entre a eficiência na gestão de custos mensurada pela SFA e a rentabilidade das instituições financeiras bancárias no Brasil Devido à importância das instituições financeiras bancárias para a economia brasileira tornase relevante compreender se a eficiência está relacionada à rentabilidade Muitos países que implementaram novas políticas de regulamentação do setor estão vivenciando um cenário com instituições mais competitivas Este também é o caso do Brasil no período entre 2001 e 2010 o setor bancário brasileiro apresentou evolução no lucro líquido e uma expansão no volume de operações de crédito demonstrando um forte crescimento MODRO SANTOS 2015 A importância do setor bancário brasileiro está relacionada com a significativa participação do setor no Produto Interno Bruto PIB do país e os altos lucros que os bancos têm apresentado na casa dos bilhões de reais dado que as instituições financeiras bancárias estão entre as maiores e mais lucrativas empresas no Brasil MANTOVANI SANTOS 2015 Conforme Nguyen et al 2016 um setor bancário eficiente pode refletir uma economia eficiente uma vez que as instituições financeiras bancárias são as principais financiadoras da economia Os resultados desse estudo poderão contribuir com os gestores investidores e analistas financeiros gerando subsídios para tomada de decisão uma vez que oferece informações a respeito da estrutura do setor financeiro e sobre a importância da eficiência Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 8 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 para a rentabilidade O estudo também tem o potencial de contribuir para a ampliação das discussões acadêmicas a respeito dessa temática A estrutura do presente trabalho é composta por cinco seções sendo esta introdução a primeira delas A plataforma teórica que norteia o estudo é apresentada na segunda seção e na terceira são descritos os procedimentos metodológicos adotados Na quarta seção são apresentados os resultados obtidos pela pesquisa Já as considerações finais constam na quinta e última seção 2 REFERENCIAL TEÓRICO 21 Teoria da intermediação financeira A intermediação financeira é uma atividade produtiva em que as instituições financeiras captam recursos com o propósito de repassálos por meio de empréstimos Nesse sentido os intermediários financeiros unem os ofertantes com os tomadores de recursos financeiros facilitando assim a realocação de tais recursos na economia DIAMOND 1984 O estudo das instituições financeiras pode ser feito por meio da teoria da intermediação financeira Conforme Allen e Santomero 1998 essa teoria trata de modelos de alocação de recursos e dos custos decorrentes da operação de intermediação de recursos financeiros Os intermediários financeiros são agentes aos quais é delegada a autoridade para investir em ativos financeiros Tais intermediários podem ser divididos em dois grupos os intermediários financeiros bancários e os intermediários financeiros não bancários DIAMOND 1984 Os intermediários bancários efetuam diretamente a intermediação de dinheiro em espécie no mercado financeiro por exemplo os bancos múltiplos e bancos comerciais entre outros Já os intermediários não bancários efetuam a intermediação de títulos de valores mobiliários com o objetivo de gerar liquidez a estes títulos dessa forma podem atuar também no mercado de capitais por exemplo as sociedades de leasing as companhias de seguros e as corretoras PHILIPPON 2015 Scholtens e Wensveen 2000 afirmam que no processo de evolução do mercado financeiro os bancos comerciais se tornaram empresas globais desenvolvendo suas atividades em várias partes do mundo e passando a obter outras fontes de receitas para além daquelas geradas pela atividade tradicional de intermediação financeira Essas mudanças acarretaram em uma evolução na teoria da intermediação para captar as mudanças ocorridas no setor financeiro mundial Visando captar tais mudanças a abordagem do gerenciamento de risco surge como uma crítica à teoria tradicional da intermediação financeira Essa abordagem parte do pressuposto de que os intermediários financeiros não têm como finalidade apenas a correção das imperfeições de mercado mas também absorvem o risco que seria do poupador no processo de transferência dos recursos financeiros para o tomador PHILIPPON 2015 Os intermediários financeiros são compensados pelos serviços de intermediação prestados O ganho pela intermediação financeira é conhecido como spread que é a diferença entre a taxa de captação e a de repasse ou seja as instituições intermediadoras captam recursos a uma determinada taxa de juros e os emprestam a uma outra taxa que é superior à de captação Os resultados da intermediação são mensurados por meio do total da receita de intermediação menos os custos agregados da operação de intermediação financeira custo Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 9 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 de captação no mercado custo da operação provisão para perdas e despesas de pessoal dentre outros O resultado da intermediação é a soma de todos os spreads oriundos das operações de crédito ZAERNJUK et al 2014 Uma ilustração simplificada do processo de intermediação financeira bancária com a representação do spread e do custo é demonstrada na Figura 01 Nessa figura o valor intermediado é de R 10000 O banco capta o recurso do poupador e paga uma taxa de R 500 ou 5 O valor é repassado para o tomador a uma taxa de R 1200 ou 12 Figura 01 Ilustração do processo de intermediação financeira bancária Fonte Adaptada de Philippon 2015 O spread dessa operação é de R 700 ou 7 que é a diferença entre a taxa de captação e a de repasse Como o custo agregado da operação foi de R 300 valor dado no exemplo o resultado da intermediação financeira é de R 400 R 700 R 300 Vale destacar que o custo de intermediação financeira afeta diretamente o custo dos utilizadores do financiamento externo ou seja das empresas ou pessoas que utilizam os serviços dos intermediadores DIAMOND 1984 Philippon 2015 afirma que a eficiência na gestão dos custos oriundos da operação de intermediação financeira impacta diretamente o desenvolvimento da economia no longo prazo 22 A reestruturação do negócio bancário Os bancos desempenham um papel vital no desenvolvimento econômico de um país desenvolvendo três funções principais facilitação dos pagamentos mobilização de poupança e alocação de fundos de empréstimo KHALIL MEHMOOD AHMAD 2015 O negócio bancário está mudando rapidamente e está se tornando mais competitivo Ao redor do mundo as empresas desse setor passaram por mudanças drásticas em termos de globalização financeira e competição tecnológica Ao mesmo tempo os bancos também experimentaram o crescimento dos custos como consequência de fatores como o aumento do rigor nos requisitos regulatórios as inovações nos serviços financeiros o desenvolvimento tecnológico e os desafios das crises financeiras NGUYEN et al 2016 De acordo com Brighi e Venturelli 2015 as mudanças no ambiente regulatório a abertura dos mercados e os desenvolvimentos tecnológicos vêm criando um novo cenário competitivo no setor bancário A reestruturação no setor alterou as fontes de receitas e o modelo de gestão As instituições que tinham alta dependência dos ganhos de spread perderam muito desse ganho com a regulação do setor e a padronização das taxas Diante disso para compensar a perda de receitas com o spread as chamadas receitas de serviço abertura e manutenção de contas tarifas em geral administração de fundos dentre outros passaram a ter maior representatividade no total das receitas bancárias PHILIPPON 2015 Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 10 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Visando o aumento das receitas os bancos se utilizaram dos avanços tecnológicos e passaram a investir em meios alternativos de serviços com destaque para o internet banking pessoa física e o office banking pessoa jurídica além destes podese citar o aumento na rede de terminais de autoatendimento caixas eletrônicos e o investimento em correspondentes bancários que são empresas terceirizadas e remuneradas pelos bancos para prestar alguns serviços que antes eram prestados dentro das agências por exemplo o pagamento de títulos SOUZA MACEDO 2009 As reestruturações do setor no Brasil começaram a ocorrer após a implantação do Plano Real e seguiram a tendência internacional Conforme Nguyen et al 2016 as instituições financeiras passaram a ter duas fontes principais de receitas as oriundas da intermediação financeira e as da prestação de serviços Esse processo forçou uma alteraçção estrutural dos bancos que resultou não apenas em uma mudança do foco sobre as atividades rentáveis mas também em um novo desenho do mercado BARROS WANKE 2014 Giambiagi e Villela 2005 dividem o período de reestruturação no sistema bancário brasileiro em três etapas A primeira etapa compreende o período entre os anos de 1991 a 1994 marcada pela hiperinflação e pelo binômio privatização do setor ou abertura para a participação estrangeira essa etapa forçou o setor a se modernizar para encarar o novo ambiente A segunda etapa compreende o período entre os anos de 1995 a 1998 quando o Plano Real estabilizou a economia e marcou uma revolução comportamental no setor privado Já a terceira etapa compreende o período entre os anos de 1999 a 2002 e foi marcada por uma tríplice mudança de regime cambial monetário e fiscal Diante das mudanças ocorridas Philippon 2015 explica que as fontes de receita bancária e a eficiência na gestão dos custos oriundos dessas operações passaram a ter uma importância ainda maior visto que os bancos possuem produtos e tecnologias muito similares Assim a manutenção da rentabilidade das empresas do setor está relacionada com a capacidade de gerenciar os custos do negócio bancário 23 Evidências empíricas sobre a eficiência e a rentabilidade de instituições bancárias Conforme Gitman 2010 a eficiência na gestão empresarial cria valor e melhora o desempenho organizacional minimizando o uso dos recursos Observase que o resultado financeiro é fruto de decisões tomadas nos processos empresariais Diante disso o autor argumenta que a utilização eficiente dos recursos leva à redução dos custos e ao aumento no nível de serviço agregando valor e podendo representar um aumento da rentabilidade das empresas Nesse sentido podese identificar alguns trabalhos empíricos que buscaram obter evidências sobre a relação entre a eficiência e a rentabilidade das instituições financeiras Por exemplo o estudo de Kirkwood e Nahm 2006 utilizou a Análise Envoltória de Dados DEA para mensurar a eficiência e analisou a relação entre eficiência e a rentabilidade dos dez maiores bancos da Austrália entre os anos de 1995 a 2002 Os resultados evidenciaram que as grandes instituições financeiras bancárias apresentaram maior eficiência e revelaram a existência de uma relação significativa entre eficiência e rentabilidade em que os bancos maiores e mais eficientes apresentaram maior rentabilidade Já o estudo de Mendonça et al 2017 investigou a relação entre a eficiência econômico financeira mensurada pela DEA e a lucratividade no setor bancário brasileiro no período Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 11 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 de 2011 a 2015 As evidências obtidas indicaram que quanto maior a eficiência econômico financeira da instituição bancária maior sua lucratividade Por fim notase que o estudo de DOliveira 2014 teve uma abordagem diferenciada utilizando o método de Análise da Fronteira Estocástica SFA para mensurar a eficiência na gestão de custos O trabalho examinou quais variáveis específicas dos bancos quais específicas da indústria e quais variáreis macroeconômicas determinaram a rentabilidade dos bancos brasileiros no período de 1995 a 2013 Dentre outros aspectos os resultados demonstraram uma relação significativa e positiva entre a eficiência no custo e a rentabilidade dos bancos evidenciando que a eficiência é um fator que influencia o resultado das empresas do setor Dessa forma a literatura fornece indícios de que a eficiência pode ser considerada um fator que influencia a rentabilidade bancária uma vez que a gestão mais eficiente pode contribuir para o melhor desempenho econômicofinanceiro Entretanto não está claro qual a magnitude do impacto que essa eficiência tem sobre a rentabilidade o que representa uma oportunidade de pesquisa que pode contribuir para o avanço do conhecimento na área 3 METODOLOGIA 31 Tipo de pesquisa amostra e coleta de dados Este trabalho é caracterizado como um estudo quantitativo e descritivo Uma pesquisa quantitativa segundo Gil 2010 é aquela que utiliza uma análise estatística e matemática considerando aquilo que pode ser quantificável A pesquisa descritiva para Martins 2002 tem como objetivo descrever as características de uma determinada população ou fenômeno estabelecendo relações entre variáveis e fatos A pesquisa utilizou dados secundários obtidos por meio do relatório anual do Banco Central do Brasil BCB intitulado 50 Maiores Bancos e o Consolidado do Sistema Financeiro Nacional disponível no website do BCB 2016 Esse relatório contém os dados consolidados referentes a todas as instituições que compõem o Sistema Financeiro brasileiro Ressaltase que o modelo de demonstrações adotado segue o padrão do Plano Contábil das Instituições do Sistema Financeiro Nacional COSIF conforme a Circular n 1273 emitida pelo BCB BCB 1987 Foram coletados os dados contidos nas demonstrações contábeis anuais das instituições englobando o período de 2008 a 2015 A amostra do estudo foi formada por todas as instituições bancárias cujos dados estavam disponíveis para todos os anos em análise Dessa forma a amostra final da pesquisa é composta por 47 instituições financeiras bancárias as quais são listadas no Quadro 01 Observase que a amostra é diversificada e que contém não apenas instituições financeiras nacionais mas também filiais internacionais que atuam no Brasil Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 12 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Quadro 01 Instituições financeiras que compõem a amostra INSTITUIÇÃO FINANCEIRA TIPO DE BANCO INSTITUIÇÃO FINANCEIRA TIPO DE BANCO Banco ABC Brasil SA Múltiplo BPN Brasil Banco Múltiplo SA Múltiplo Banco Alfa SA Comercial Banco Bradesco SA Múltiplo Banco Cooperativo do Brasil SA Múltiplo BRB Banco De Brasília SA Múltiplo Banestes SA Banco do Est do ES Múltiplo Banco BTG Pactual SA Múltiplo Banco Indo Funchal Brasil SA Múltiplo Caixa Econômica Federal Caixa Econômica Banco do Est do RS SA Múltiplo Citibank NA Com Estrangeiro Banco do Brasil SA Múltiplo Banco Credit Suisse Brasil SA Múltiplo Banco BBM SA Múltiplo Deutsche Bank SA Múltiplo Banco AJ Renner SA Múltiplo Banco Fator SA Múltiplo Banco Capital SA Múltiplo HSBC Bank Brasil SA Múltiplo Banco Cargill SA Múltiplo Banco CNH Ind Capital SA Múltiplo Banco Cédula SA Múltiplo Banco Indusval SA Múltiplo Banco Fibra SA Múltiplo ING Bank NV Com Estrangeiro Banco Guanabara SA Múltiplo Itaú Unibanco SA Múltiplo Banco Keb Hana do Brasil SA Comercial Banco John Deere SA Múltiplo Banco de la Provde Buenos Aires Com Estrangeiro Banco Mercantil do Brasil SA Múltiplo Banco Luso Brasileiro SA Múltiplo Banco Pine SA Múltiplo Banco Pottencial SA Comercial Banco Rendimento SA Comercial Banco Rabobank Inte Brasil SA Múltiplo Banco J Safra SA Múltiplo Banco Ribeirão Preto SA Múltiplo Banco Santander SA Múltiplo Banco Triangulo SA Múltiplo Banco Paulista SA Múltiplo Banco BMG SA Múltiplo Banco Sofisa SA Múltiplo Banco BNP Paribas Brasil SA Múltiplo Banco Votorantim SA Múltiplo Bank of Ame Merrill Lynch SA Múltiplo Fonte elaborado pelos autores com base em BCB 2016 A amostra possui instituições financeiras classificadas como bancos múltiplos bancos comerciais e Caixa Econômica Assaf Neto 2003 explica os aspectos referentes a cada um desses tipos de instituição Conforme o autor os bancos comerciais são constituídos obrigatoriamente na forma de sociedades anônimas Eles têm a finalidade de executar operações de crédito de curto e de médio prazo atendendo tanto às necessidades de recursos para capital de giro de pessoas jurídicas como também às pessoas físicas e terceiros em geral Já os bancos múltiplos são organizados sob forma de sociedade anônima e em sua denominação social deve constar a expressão banco Podem ser privados ou públicos e realizam operações ativas passivas e acessórias normalmente realizadas por distintos tipos de instituições financeiras E a Caixa Econômica é o principal agente das políticas públicas do Governo Federal e executa atividades características dos bancos comerciais e múltiplos Ela é responsável pela operacionalização das políticas públicas do governo federal para a habitação popular e saneamento básico utilizando os recursos de cadernetas de poupança e depósitos judiciais e pela gestão dos recursos do Fundo de Garantia por Tempo de Serviço FGTS ASSAF NETO 2003 32 Cálculo da eficiência utilizando o método SFA De acordo com Jiang Yao e Feng 2013 a abordagem de intermediação é considerada a mais adequada para mensurar a eficiência dos bancos e é amplamente utilizada na literatura internacional GAGANIS PASIOURAS 2013 DUYGUN SENA SHABAN 2013 KHALIL MEHMOOD AHMAD 2015 GHROUBI ABAOUB 2016 Considerando esse aspecto o presente estudo emprega a abordagem de intermediação com o intuito de mensurar a eficiência na gestão de custos das instituições financeiras bancárias brasileiras Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 13 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Sob a ótica dessa abordagem as despesas incorridas na captação de recursos são consideradas como inputs assim como as despesas operacionais e de pessoal Já os recursos emprestados e gerados por meio de investimentos são tratados como outputs divididos em dois resultados distintos o resultado da própria intermediação e o resultado operacional que engloba os lucros das demais atividades bancárias GHROUBI ABAOUB 2016 Na Figura 02 são apresentados os inputs e outputs utilizados para estimar a eficiência na gestão de custos das instituições financeiras Figura 02 Indicadores de input e output utilizados para mensurar a eficiência de custo Fonte Adaptada de Khalil Mehmood e Ahmad 2015 Conforme representado na Figura 02 o termo eficiência na gestão de custos pode ser orientado de duas maneiras para a maximização do resultado outputs ou para a minimização do custo inputs Especificamente na mensuração da eficiência na gestão de custos por meio do método da SFA é necessária a estimação de uma função custo Essa função estima a fronteira do custo mínimo necessário para produzir um dado produto a partir dos insumos inputs disponíveis Assim para ser eficiente uma instituição deve continuar produzindo o mesmo volume de produtos outputs utilizando um volume menor de inputs Assim o método SFA é orientado para minimizar o nível de inputs e estimar os níveis de eficiência na gestão de custos das instituições financeiras KHALIL MEHMOOD AHMAD 2015 Os escores estimados pelo modelo para cada empresa variam entre 0 zero e 1 um Quanto maior o escore maior a eficiência na gestão de custos o escore igual a 1 um indicaria uma instituição totalmente eficiente enquanto que o escore igual a 0 zero indicaria uma instituição totalmente ineficiente Os inputs e outputs do modelo foram operacionalizados utilizando indicadores econômicofinanceiros que representam as características em análise A fórmula para o cálculo e as respectivas definições dos indicadores são apresentadas no Quadro 02 Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 14 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Quadro 02 Indicadores de input e output utilizados no modelo SFA INPUTS NOME SIGLA FÓRMULA DEFINIÇÃO Indicador de Despesa de Pessoal IDP Identifica o custo com pessoal necessário para o banco exercer a atividade bancária seja de intermediação ou prestação de serviço Indicador de Despesa Operacional IDO Mede o custo operacional da instituição que seria a despesa que a instituição possui para prestar os demais serviços que não são de intermediação financeira Indicador de Despesa com Intermediação Financeira IDIF Mensura o custo da instituição bancária para prestar o serviço de intermediação financeira OUTPUTS Indicador de Resultado da Intermediação Financeira IRIF Indica o quanto a instituição financeira conseguiu apurar de resultado das operações de intermediação financeira Indicador de Resultado Operacional IRO Representa o quanto a instituição financeira conseguiu apurar de resultado operacional ou seja de receitas não oriundas da intermediação financeira Fonte Elaborado pelos autores Para analisar a eficiência da relação entre inputs e outputs a SFA requer uma especificação explícita da forma desta relação ou seja há a necessidade da estimação de uma função de custos Nesse estudo foi utilizada a função translog transcendental logarítmica que consiste em uma aproximação de segunda ordem de qualquer forma funcional BATTESE COELLI 1995 A função utilizada é representada pela Equação 01 Ln CTit β0 β1 LnIRIFit β2 LnIROit β3 LnIDPit β4 LnIDOit β5 LnIDIFit β6 LnIRIFit2 β7 LnIROit2 β8 LnIDPit2 β9 LnIDOit2 β10 LnIDIFit2 β11 LnIRIFit LnIROit β12 LnIRIFit LnIDPit β13 LnIRIFit LnIDOit β14 LnIRIFit LnIDIFit β15 LnIROit LnIDPit β16 LnIROit LnIDOit β17 LnIROit LnIDIFit υit νit Onde i indica a observação da iésima instituição financeira na amostra t indica o tésimo período Ln CTit é o logaritmo natural do custo total IRIFit é o Indicador de Resultado da Intermediação Financeira IROit é o Indicador de Resultado Operacional IDPit é o Indicador de Despesa de Pessoal IDOit é o Indicador de Despesa Operacional IDIFit é o Indicador de Despesa com Intermediação Financeira e o υit e νit são os termos de erro Já o Custo Total CTit utilizado na estimação da função de custo é o somatório das despesas de intermediação financeira despesas de pessoal despesas administrativas despesas tributárias e outras despesas operacionais dividido pelo ativo total ou seja calculado como percentual do ativo Para a realização dos cálculos da Fronteira Estocástica foi utilizado o software Frontier 41 Dessa forma foram mensurados os índices de eficiência para cada uma das instituições financeiras bancárias em cada ano e esses índices compõem a variável eficiência na gestão de custos EFI 33 Aplicação do modelo econométrico com dados em painel Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 15 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Na primeira etapa da pesquisa foram calculados os escores de eficiência utilizando se do método da Análise da Fronteira Estocástica SFA Já na segunda fase o estudo investigou por meio de um modelo de regressão com dados em painel se a eficiência foi fator determinante para a rentabilidade das instituições financeiras bancárias Para investigar a relação entre a eficiência mensurada pela SFA e a rentabilidade bancária foi utilizado um modelo econométrico de regressão linear múltipla com dados em painel A variável dependente é o Retorno sobre Ativos ROA que representa a rentabilidade das instituições bancárias e é calculado pela divisão do lucro líquido pelo ativo total O ROA mede a capacidade de gestão dos ativos ou seja indica o lucro líquido obtido para cada R 100 em ativos A variável de investigação desse estudo é a EFI eficiência na gestão de custos composta pelos indicadores de eficiência calculados para cada banco conforme descrito no tópico 32 Adicionalmente foram incluídas variáveis de controle no modelo As variáveis de controle são aquelas que poderiam afetar a variável dependente ROA conforme indicado pela literatura e por evidências empíricas anteriores e são utilizadas com o intuito de isolar o seu efeito sobre a variável dependente Para este estudo foram utilizadas as variáveis tamanho dos bancos capitalização grau de intensidade de empréstimo indicador de despesas operacionais em relação ao ativo total e grau de diversificação Tais variáveis juntamente às suas definições e aos sinais esperados são apresentadas no Quadro 03 Quadro 03 Variáveis utilizadas no modelo de regressão com dados em painel e sinais esperados NOME SIGLA DEFINIÇÃO E CÁLCULO SINAL ESPERADO Eficiência EFI Mensurada pela SFA esse indicador pondera as receitas de intermediação e operacional com as despesas incorridas dessas operações visando refletir a dinâmica da atividade bancária Tamanho dos Bancos TAM Calculado pelo o logaritmo dos ativos totais É utilizado para medir o ganho em escala ao maximizar a utilização de seus fatores de produção Capitalização CAP É calculada pelo logaritmo da razão entre patrimônio líquido e total de ativos Mede o grau de capitalização ou seja é a capacidade que uma instituição possui de captar depósito à vista Grau de Intensidade de Empréstimo GIE É calculado pelo logaritmo da razão entre empréstimos e total de ativos Referese indiretamente ao nível de liquidez Assim volume de empréstimo maior implica em baixa liquidez Indicador de Despesas Operacionais em Relação ao Ativo DOA É calculado pelo logaritmo da razão entre despesas operacionais e total de ativos Referese aos gastos operacionais relativamente ao porte do banco Grau de Diversificação GDI É calculado pelo logaritmo da razão entre receitas operacionais e total de ativos É referente às fontes alternativas de receita como por exemplo receitas com serviços e taxas Fonte Elaborado pelos autores Assim o modelo é composto de cinco variáveis de controle além da variável de investigação EFI e da dependente ROA A estrutura utilizada no modelo operacionalizado nesta pesquisa é representada pela Equação 02 ROAit β0 β1 EFIit β2 TAMit β3 CAPit β4 GIEit β5 DOAit β6 GDIit ε Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 16 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Onde ROAii é a rentabilidade Retorno sobre Ativos da instituição financeira bancária i no período t β0 é o intercepto da reta β1 a β6 são os coeficientes angulares EFIit TAMit CAPit GIEit DOAit são as variáveis independentes para a instituição financeira bancária i no período t e ε é o termo de erro da regressão O coeficiente β1 indicará a relação entre a eficiência na gestão de custos e a rentabilidade ROA das instituições financeiras Esperase que esse coeficiente seja positivo e estatisticamente significativo o que indicaria uma relação direta entre tais variáveis 4 RESULTADOS E DISCUSSÕES Inicialmente são apresentados os resultados obtidos pela mensuração da eficiência pelo método SFA Na Tabela 01 constam as estatísticas descritivas dos escores de eficiência na gestão de custos mensurados pela SFA para cada ano Tabela 01 Estatísticas descritivas da eficiência mensurada no período Ano Média Desviopadrão Mínimo Máximo 2008 08233 00894 05943 09365 2009 09221 00073 09035 09482 2010 08662 00595 06981 09503 2011 08590 00999 05666 09735 2012 07625 01730 03580 09859 2013 08613 00800 05683 09671 2014 08927 00551 07243 09636 2015 08789 00562 07206 09577 Fonte Elaborada pelos autores A eficiência média na gestão de custos variou entre 7625 e 9221 durante o período analisado O ano com a menor média foi o de 2012 já o de maior média foi 2009 Com relação aos valores mínimos o ano de 2012 apresentou o menor valor 03580 e o valor mínimo mais elevado foi apresentado no ano de 2009 09035 Já sobre os valores máximos do período o ano de 2012 apresentou o maior valor 09859 enquanto que o valor máximo menos elevado 09365 foi obtido no ano de 2008 Podese observar que o desviopadrão apresentou valores entre 00551 a 01730 no período O ano de 2012 teve a maior variabilidade nos escores de eficiência na gestão de custos das instituições financeiras analisadas desviopadrão de 01730 apresentando tanto o menor valor mínimo do período quanto o maior valor máximo Com base nos valores máximos estimados observase que nenhuma instituição foi totalmente eficiente na gestão de custos nenhuma delas atingiu o escore igual a 1 ou 100 Já com base nos valores mínimos é possível notar também que nenhuma instituição foi totalmente ineficiente na gestão de custos não houve instituições com escore igual a 0 Após a análise das estatísticas descritivas o método de regressão linear múltipla com dados em painel foi utilizado para investigar a relação entre a eficiência na gestão de custos e a rentabilidade das instituições do setor bancário brasileiro Para a aplicação do modelo de regressão inicialmente foram verificadas as correlações entre as variáveis do estudo A estatística VIF fatores de inflação de variância confirmou a ausência de multicolinearidade sendo que o indicador variou entre 1105 e 2200 Para que a regressão Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 17 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 seja aceitável é necessário que o VIF seja menor que cinco conforme Gujarati 2006 Dessa forma não houve problemas com relação à multicolinearidade das variáveis Com relação à normalidade dos resíduos o estudo pautouse no proposto no Teorema do Limite Central o qual afirma que para amostras com um número superior a 30 observações presumese que a distribuição das médias é aproximadamente normal WOOLDRIDGE 2012 Considerando que foram utilizadas 376 observações podese assumir a distribuição normal para as variáveis e os resíduos da regressão Após aplicar o modelo por meio do método de Mínimos Quadrados Ordinários MQO foram efetuados os testes para a identificação da abordagem de dados em painel que melhor se ajustaria aos dados O resultado do teste de BreuschPagan foi igual a 64373 com pvalor de 0000 validando que a abordagem de efeitos aleatórios é preferível em relação à POLS Procedendo ao teste de Hausman obtevese o resultado igual a 15806 com pvalor de 00148 indicando que o modelo efeitos aleatórios é preferível em relação ao modelo de efeitos fixos Assim concluiuse que a abordagem de painel que melhor se ajusta aos dados é a de efeitos aleatórios Para verificar a homocedasticidade dos resíduos utilizouse o teste de White Foi detectada a heterocedasticidade e esta foi corrigida por meio da matriz de covariância de White errospadrão robustos Os resultados finais obtidos a partir da aplicação do modelo são apresentados na Tabela 02 Tabela 02 Resultados da regressão pelo modelo de efeitos aleatórios com errospadrão robustos Variável dependente ROA Retorno sobre Ativos Rentabilidade Variável Coeficiente Erropadrão robusto Estatística t EFI 00122 00059 2070 TAM 00011 00006 1812 CAP 00086 00019 4468 GIE 00001 00009 0122 DOA 00035 00014 2513 GDI 00017 00011 1580 Const 00104 00123 0842 Wald chi25 3658 Prob chi2 00099 Nº de observações 376 R2 total 01359 Fonte Elaborado pelos autores Observase que o modelo é significativo a 1 uma vez que o pvalor do teste qui quadrado é inferior a 001 e que a estimativa por efeitos aleatórios é consistente Como resultado mais relevante destacase que a variável de investigação eficiência EFI mensurada pela SFA foi significativa a 5 95 de confiança apresentando relação positiva O resultado para essa relação corrobora o encontrado por DOliveira 2014 que também indica que maiores níveis de eficiência na gestão de custos estão relacionados a maiores níveis de rentabilidade das instituições financeiras Observase que o coeficiente da variável EFI apresentou um valor de 00122 o que indica que o aumento de uma unidade na eficiência geria um aumento de 00122 unidade na rentabilidade Diante disso é possível inferir que o fato de a instituição financeira obter máxima eficiência escore de eficiência igual a 1 estaria associado a um ROA cerca de 122 superior ao de uma instituição totalmente ineficiente escore de eficiência igual a 0 Também foram obtidos resultados estatisticamente significativos para as variáveis de controle TAM CAP e DOA A variável TAM foi significativa a 10 e esse achado evidencia que o tamanho da instituição financeira pode influenciar na rentabilidade DOliveira 2014 Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 18 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 e Mendonça et al 2017 também encontraram resultados semelhantes para essa variável A relação encontrada foi positiva o que significa que quanto maior for o tamanho da instituição maior será a sua rentabilidade uma vez que as instituições maiores reduzem seus riscos ao diversificarem as carteiras e com isso diminuem o custo de captação A variável CAP foi significativa a 1 Seu coeficiente positivo indica que quanto maior for a capitalização maior será a rentabilidade pois bancos bem capitalizados reduzem seus custos de financiamento e captação ao sinalizarem uma menor possibilidade de falência Os estudos de Kirkwood e Nahm 2006 DOliveira 2014 e Mendonça et al 2017 encontraram resultados análogos a este para essa variável Já a variável DOA foi significativa a 5 Os estudos de Kirkwood e Nahm 2006 e Mendonça et al 2017 encontraram resultados semelhantes A relação encontrada foi indireta sinal negativo o que significa que quanto maiores forem os gastos operacionais relativamente ao porte do banco menor será a rentabilidade Diante disso evidenciase a relevância de que as instituições financeiras controlem os gastos operacionais para terem uma maior rentabilidade Por fim vale destacar que os sinais esperados dos coeficientes que seriam obtidos para a variável de investigação e as variáveis de controle embasados na literatura científica foram todos confirmados para os coeficientes encontrados no presente estudo 5 CONSIDERAÇÕES FINAIS Esta pesquisa objetivou investigar a relação entre eficiência na gestão de custos e a rentabilidade das instituições financeira bancárias no Brasil Diante da aplicação do modelo econométrico foi identificada a existência da relação estatisticamente significativa entre essas variáveis Dessa forma a hipótese teórica testada foi confirmada Considerando o aspecto gerencial os resultados obtidos evidenciam que a eficiência na gestão de custos se relaciona de forma direta com a rentabilidade demonstrando que a eficiência exerce um impacto significativo no retorno sobre o total de ativos aplicados ROA Assim os achados deste estudo indicam que a eficiência na gestão de custos é um fator determinante para melhorar a rentabilidade do sistema bancário brasileiro Destacase que foi possível identificar a magnitude dessa relação o fato de a instituição financeira obter máxima eficiência escore de eficiência igual a 1 estaria associado a um ROA cerca de 122 superior ao de uma instituição totalmente ineficiente escore de eficiência igual a 0 Do ponto de vista econômico é importante ressaltar que a eficiência na operação de intermediação financeira não afeta apenas a rentabilidade destas instituições mas pode impactar o desenvolvimento da economia no longo prazo gerando taxas menores de repasse e influenciando o fluxo monetário Assim estudar os aspectos referentes à eficiência das instituições bancárias assume crescente relevância no atual cenário econômico brasileiro de instabilidade e de altas taxas de juros praticadas pelos bancos Vale mencionar também que as evidências encontradas e discutidas nesta pesquisa devem ser consideradas levando em conta os critérios de seleção da amostra as 47 instituições financeiras no período de 2008 a 2015 e as limitações da metodologia adotada Existe ainda a limitação das variáveis utilizadas nos modelos econométricos uma vez que existem outras variáveis que podem influenciar a rentabilidade e que não foram consideradas nesse estudo Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 19 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Sugerese para futuros estudos investigar quais variáveis são determinantes da eficiência bancária bem como identificar outros fatores que possam influenciar a rentabilidade dessas instituições Por fim esperase que as evidências apresentadas neste trabalho contribuam para aumentar o conhecimento sobre a importância da eficiência na gestão de custos para a rentabilidade das instituições financeiras bancárias brasileiras fomentando a discussão sobre esta temática REFERÊNCIAS AIGNER D LOVELL C A K SCHMIDT P Formulation and estimation of stochastic frontier production function models Journal of Econometrics v 6 n 1 p 2137 1977 ALLEN F SANTOMERO A M The theory of financial intermediation Journal of Banking Finance v 21 n 11 p 14611485 1998 ALLEN F SANTOMERO A M What do financial intermediaries do Journal of Banking Finance v 25 n 2 p 271294 2001 ASSAF NETO A Estrutura e análise de balanços 10 ed São Paulo Atlas 2012 ASSAF NETO A Mercado Financeiro 5 ed São Paulo Atlas 2003 BARROS C P WANKE P Banking efficiency in Brazil Journal of International Financial Markets Institutions and Money v 28 p 5465 2014 BATTESE G E COELLI T J A model for technical inefficiency effects in a stochastic frontier production function for panel data Empirical Economics v 20 n 2 p 325332 1995 BCB BANCO CENTRAL DO BRASIL 50 maiores bancos e o consolidado do Sistema Financeiro Nacional 2016 Disponível em httpwww4bcbgovbrtop50porttop50 asp Acesso em 24 mai 2017 BCB BANCO CENTRAL DO BRASIL Plano Contábil das Instituições do Sistema Financeiro Nacional COSIF 1987 Disponível em httpwwwbcbgovbrCOSIF Acesso em 23 mai 2017 BRIGHI P VENTURELLI V How functional and geographic diversification affect bank profitability during the crisis Finance Research Letters v16 p 110 2015 DOLIVEIRA E H Determinantes da lucratividade bancária no Brasil 2014 51 f Dissertação Mestrado em Economia Programa de PósGraduação em Economia Universidade de Brasília Brasília 2014 DIAMOND D W Financial intermediation and delegated monitoring The Review of Economic Studies v 51 n 3 p 393414 1984 Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 20 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 DIETRICH A WANZENRIED G Determinants of bank profitability before and during the crisis Journal of International Financial Markets Institutions Money v 21 p 307327 2011 DOUMPOS M COHEN S Applying data envelopment analysis on accounting data to assess and optimize the efficiency of Greek local governments Omega International Journal of Management Science v 46 p 7485 2014 DUYGUN M SENA V SHABAN M Schumpeterian competition and efficiency among commercial banks Journal of Banking Finance v 37 n 12 p 51765185 2013 GAGANIS C PASIOURAS F Financial supervision regimes and bank efficiency International evidence Journal of Banking Finance v 37 n 12 p 54635475 2013 GHROUBI M ABAOUB E A metafrontier function for the estimation of Islamic and conventional banks cost and revenue efficiency the case of Malaysia from 2006 to 2012 International Journal of Business and Management v 11 n 5 p 254 2016 GIAMBIAGI F VILLELA A A Economia brasileira contemporânea Rio de Janeiro Elsevier 2005 GIL A C Como elaborar projetos de pesquisa 5 ed São Paulo Atlas 2010 GITMAN L J Princípios de administração financeira 12 ed São Paulo Harbra 2010 GUJARATI D N Econometria básica 4 ed Rio de Janeiro Elsevier 2006 JIANG C YAO S FENG G Bank ownership privatization and performance Evidence from a transition country Journal of Banking Finance v 37 n 9 p 33643372 2013 KHALIL S MEHMOOD B AHMAD N Cost efficiency of Pakistani banking sector a Stochastic Frontier Analysis The Journal of Commerce v 7 n 3 p110126 2015 KIRKWOOD J NAHM D Australian banking efficiency and its relation to stock returns The Economic Record v 82 n 258 p 253267 2006 MANTOVANI M H C SANTOS J O dos Análise da relação entre alavancagem e rentabilidade dos bancos brasileiros listados da Bolsa de Valores de São Paulo do período de 2001 a 2010 REGE Revista de Gestão v 22 n 4 p 509524 2015 MARTINS G de A Manual para elaboração de monografias e dissertações São Paulo Atlas 2002 MELLO J C C B S MEZA L A GOMES E G NETO L B Curso de análise de envoltória de dados In Simpósio Brasileiro de Pesquisa Operacional 37 2005 Gramado Anais Rio de Janeiro SOBRAPO 2005 Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 21 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 MENDONÇA D J SOUZA J A e BENEDICTO G C de SILVA S N A da Relação entre eficiência econômicofinanceira e lucratividade em instituições bancárias brasileiras Revista de Finanças e Contabilidade da Unimep v 4 n 1 p 2037 2017 MODRO W M SANTOS J O dos A relação entre o retorno das ações ordinárias métricas de desempenho e fatores econômicos Revista Administração em Diálogo v 17 n 3 2015 NGUYEN T P T NGHIEM S H ROCA E SHARMA P Bank reforms and efficiency in Vietnamese banks Applied Economics v 48 n 30 p 28222835 2016 PHILIPPON T Has the US finance industry become less efficient On the theory and measurement of financial intermediation The American Economic Review v 105 n 4 p 14081438 2015 SCHOLTENS B WENSVEEN D V A critique on the theory of financial intermediation Journal of Banking Finance v 24 n 8 p 12431251 2000 SILVA M R M Medindo a eficiência dos municípios brasileiros na provisão de políticas trabalhistas 2014 69 f Dissertação Mestrado em Economia Programa de PósGraduação em Economia Universidade de Brasília Brasília 2014 SOUZA M F A S MACEDO M A S Análise de desempenho contábilfinanceiro no setor bancário brasileiro por meio da aplicação da análise envoltória de dados DEA Revista BASE v 6 n 2 p 81100 2009 WOOLDRIDGE J M Introductory econometrics a modern approach 5 ed Mason Cengage Learning 2012 ZAERNJUK V M KRYUKOVA E M BOKAREVA E V CHERNIKOVA L I A study of the theoretical approaches to the banking financial intermediation and its development trends World Applied Sciences Journal v 30 n 12 p 17231725 2014 See discussions stats and author profiles for this publication at httpswwwresearchgatenetpublication256033901 Stochastic Frontier Analysis Using Stata Article in The Stata Journal Promoting communications on statistics and Stata September 2012 DOI 102139ssrn2145803 CITATIONS 240 READS 49275 4 authors Federico Belotti University of Rome Tor Vergata 60 PUBLICATIONS 2355 CITATIONS SEE PROFILE Silvio Daidone Food and Agriculture Organization of the United Nations 77 PUBLICATIONS 1777 CITATIONS SEE PROFILE Giuseppe Ilardi Banca dItalia 17 PUBLICATIONS 745 CITATIONS SEE PROFILE Vincenzo Atella University of Rome Tor Vergata 137 PUBLICATIONS 2370 CITATIONS SEE PROFILE All content following this page was uploaded by Federico Belotti on 07 October 2017 The user has requested enhancement of the downloaded file The Stata Journal Editors H Joseph Newton Department of Statistics Texas AM University College Station Texas editorsstatajournalcom Nicholas J Cox Department of Geography Durham University Durham UK editorsstatajournalcom Associate Editors Christopher F Baum Boston College Nathaniel Beck New York University Rino Bellocco Karolinska Institutet Sweden and University of MilanoBicocca Italy Maarten L Buis WZB Germany A Colin Cameron University of CaliforniaDavis Mario A Cleves University of Arkansas for Medical Sciences William D Dupont Vanderbilt University Philip Ender University of CaliforniaLos Angeles David Epstein Columbia University Allan Gregory Queens University James Hardin University of South Carolina Ben Jann University of Bern Switzerland Stephen Jenkins London School of Economics and Political Science Ulrich Kohler University of Potsdam Germany Frauke Kreuter Univ of MarylandCollege Park Peter A Lachenbruch Oregon State University Jens Lauritsen Odense University Hospital Stanley Lemeshow Ohio State University J Scott Long Indiana University Roger Newson Imperial College London Austin Nichols Urban Institute Washington DC Marcello Pagano Harvard School of Public Health Sophia RabeHesketh Univ of CaliforniaBerkeley J Patrick Royston MRC Clinical Trials Unit London Philip Ryan University of Adelaide Mark E Schaffer HeriotWatt Univ Edinburgh Jeroen Weesie Utrecht University Ian White MRC Biostatistics Unit Cambridge Nicholas J G Winter University of Virginia Jeffrey Wooldridge Michigan State University Stata Press Editorial Manager Lisa Gilmore Stata Press Copy Editors David Culwell and Deirdre Skaggs The Stata Journal publishes reviewed papers together with shorter notes or comments regular columns book reviews and other material of interest to Stata users Examples of the types of papers include 1 expository papers that link the use of Stata commands or programs to associated principles such as those that will serve as tutorials for users first encountering a new field of statistics or a major new technique 2 papers that go beyond the Stata manual in explaining key features or uses of Stata that are of interest to intermediate or advanced users of Stata 3 papers that discuss new commands or Stata programs of interest either to a wide spectrum of users eg in data management or graphics or to some large segment of Stata users eg in survey statistics survival analysis panel analysis or limited dependent variable modeling 4 papers analyzing the statistical properties of new or existing estimators and tests in Stata 5 papers that could be of interest or usefulness to researchers especially in fields that are of practical importance but are not often included in texts or other journals such as the use of Stata in managing datasets especially large datasets with advice from hardwon experience and 6 papers of interest to those who teach including Stata with topics such as extended examples of techniques and interpretation of results simulations of statistical concepts and overviews of subject areas The Stata Journal is indexed and abstracted by CompuMath Citation Index Current ContentsSocial and Behav ioral Sciences RePEc Research Papers in Economics Science Citation Index Expanded also known as SciSearch Scopus and Social Sciences Citation Index For more information on the Stata Journal including information for authors see the webpage httpwwwstatajournalcom Subscriptions are available from StataCorp 4905 Lakeway Drive College Station Texas 77845 telephone 9796964600 or 800STATAPC fax 9796964601 or online at httpwwwstatacombookstoresjhtml Subscription rates listed below include both a printed and an electronic copy unless otherwise mentioned US and Canada Elsewhere Printed electronic Printed electronic 1year subscription 98 1year subscription 138 2year subscription 165 2year subscription 245 3year subscription 225 3year subscription 345 1year student subscription 75 1year student subscription 99 1year institutional subscription 245 1year institutional subscription 285 2year institutional subscription 445 2year institutional subscription 525 3year institutional subscription 645 3year institutional subscription 765 Electronic only Electronic only 1year subscription 75 1year subscription 75 2year subscription 125 2year subscription 125 3year subscription 165 3year subscription 165 1year student subscription 45 1year student subscription 45 Back issues of the Stata Journal may be ordered online at httpwwwstatacombookstoresjjhtml Individual articles three or more years old may be accessed online without charge More recent articles may be ordered online httpwwwstatajournalcomarchiveshtml The Stata Journal is published quarterly by the Stata Press College Station Texas USA Address changes should be sent to the Stata Journal StataCorp 4905 Lakeway Drive College Station TX 77845 USA or emailed to sjstatacom Copyright c 2013 by StataCorp LP Copyright Statement The Stata Journal and the contents of the supporting files programs datasets and help files are copyright c by StataCorp LP The contents of the supporting files programs datasets and help files may be copied or reproduced by any means whatsoever in whole or in part as long as any copy or reproduction includes attribution to both 1 the author and 2 the Stata Journal The articles appearing in the Stata Journal may be copied or reproduced as printed copies in whole or in part as long as any copy or reproduction includes attribution to both 1 the author and 2 the Stata Journal Written permission must be obtained from StataCorp if you wish to make electronic copies of the insertions This precludes placing electronic copies of the Stata Journal in whole or in part on publicly accessible websites fileservers or other locations where the copy may be accessed by anyone other than the subscriber Users of any of the software ideas data or other materials published in the Stata Journal or the supporting files understand that such use is made without warranty of any kind by either the Stata Journal the author or StataCorp In particular there is no warranty of fitness of purpose or merchantability nor for special incidental or consequential damages such as loss of profits The purpose of the Stata Journal is to promote free communication among Stata users The Stata Journal ISSN 1536867X is a publication of Stata Press Stata Stata Press Mata and NetCourse are registered trademarks of StataCorp LP The Stata Journal 2013 13 Number 4 pp 719758 Stochastic frontier analysis using Stata Federico Belotti Centre for Economic and International Studies University of Rome Tor Vergata Rome Italy federicobelottiuniroma2it Silvio Daidone Centre for Health Economics University of York York UK silviodaidoneyorkacuk Giuseppe Ilardi Economic and Financial Statistics Department Bank of Italy Rome Italy giuseppeilardibancaditaliait Vincenzo Atella Centre for Economic and International Studies University of Rome Tor Vergata Rome Italy atellauniroma2it Abstract This article describes sfcross and sfpanel two new Stata commands for the estimation of crosssectional and paneldata stochastic frontier models sfcross extends the capabilities of the frontier command by including addi tional models Greene 2003 Journal of Productivity Analysis 19 179190 Wang 2002 Journal of Productivity Analysis 18 241253 and command functional ities such as the possibility of managing complex survey data characteristics Similarly sfpanel allows one to fit a much wider range of timevarying ineffi ciency models compared with the xtfrontier command including the model of Cornwell Schmidt and Sickles 1990 Journal of Econometrics 46 185200 the model of Lee and Schmidt 1993 in The Measurement of Productive Efficiency Techniques and Applications a production frontier model with flexible temporal variation in technical efficiency the flexible model of Kumbhakar 1990 Journal of Econometrics 46 201211 the inefficiency effects model of Battese and Coelli 1995 Empirical Economics 20 325332 and the true fixed and randomeffects models of Greene 2005a Journal of Econometrics 126 269303 A brief overview of the stochastic frontier literature a description of the two commands and their options and examples using simulated and real data are provided Keywords st0315 sfcross sfpanel stochastic frontier analysis production frontier cost frontier crosssectional panel data 1 Introduction This article describes sfcross and sfpanel two new Stata commands for the estima tion of parametric stochastic frontier SF models using crosssectional and panel data c 2013 StataCorp LP st0315 720 Stochastic frontier using Stata Since the publication of the seminal articles by Meeusen and van den Broeck 1977 and Aigner Lovell and Schmidt 1977 this class of models has become a popular tool for efficiency analysis a stream of research has produced many reformulations and exten sions of the original statistical models generating a flourishing industry of empirical studies An extended review of these models can be found in the recent survey by Greene 2012 The SF model is motivated by the theoretical idea that no economic agent can exceed the ideal frontier and deviations from this extreme represent individual inefficiencies From the statistical point of view this idea has been implemented by specifying a regression model characterized by a composite error term in which the classical idiosyn cratic disturbance aiming at capturing measurement error and any other classical noise is included with a onesided disturbance that represents inefficiency1 Whether cross sectional or panel data production or cost frontier time invariant or varying inefficiency parametric SF models are usually estimated by likelihoodbased methods and the main interest is on making inference about both frontier parameters and inefficiency The estimation of SF models is already possible using official Stata routines How ever the available commands cover a restricted range of models especially in the panel data case The sfcross command mirrors the frontier commands functionality but adds new features such as i the estimation of normalgamma models via simulated maximum likelihood SML Greene 2003 ii the estimation of the normaltruncated normal model proposed by Wang 2002 in which both the location and the scale parameters of the inefficiency distribution can be expressed as a function of exogenous covariates and iii the opportunity to manage complex survey data characteristics via the svyset command As far as paneldata analysis is concerned the Stata xtfrontier command allows the estimation of a normaltruncated normal model with timeinvariant inefficiency Battese and Coelli 1988 and a timevarying version named the time decay model Battese and Coelli 1992 Our sfpanel command allows one to fit a wider range of timevarying inefficiency models including the model of Cornwell Schmidt and Sick les 1990 that of Lee and Schmidt 1993 the flexible model of Kumbhakar 1990 the time decay and the inefficiency effects models of Battese and Coelli 1992 1995 and the true fixed TFE and randomeffects TRE models of Greene 2005a For the last two models the command allows different distributional assumptions provid ing the modeling of both inefficiency location and scale parameters Furthermore the command allows the estimation of the randomeffects timeinvariant inefficiency mod els of Pitt and Lee 1981 and Battese and Coelli 1988 as well as the fixedeffects version of the model of Schmidt and Sickles 1984 which is characterized by no distri butional assumptions on the inefficiency term In addition because the main objective 1 The literature distinguishes between production and cost frontiers The former represents the maximum amount of output that can be obtained from a given level of inputs while the latter characterizes the minimum expenditure required to produce a bundle of outputs given the prices of the inputs used in its production F Belotti S Daidone G Ilardi and V Atella 721 of SF analysis is the estimation of inefficiency we provide postestimation routines to compute both inefficiency and efficiency scores as well as their confidence intervals Jondrow et al 1982 Battese and Coelli 1988 Horrace and Schmidt 1996 Finally sfcross and sfpanel also allow the simultaneous modeling of heteroskedasticity in the idiosyncratic error term In the development of these new commands we make extensive use of Mata to speed up the estimation process We allow for the use of Stata factor variables weighted esti mation constrained estimation resamplingbased variance estimation and clustering Moreover by using Mata structures and libraries we provide a very readable code that can be easily developed further by other Stata users All of these features make the commands simple to use extremely flexible and fast they also ensure the opportunity to fit stateoftheart SF models Finally we would like to emphasize that sfpanel offers the possibility of performing a constrained fixedeffects estimation which is not yet available with xtreg Moreover the models of Cornwell Schmidt and Sickles 1990 and Lee and Schmidt 1993 al though proposed in the SF literature are linear paneldata models with timevarying fixed effects and thus potentially very useful in other contexts The article is organized as follows In section 2 we present a brief review of the SF approach evolution focusing on the models that can be estimated using the proposed commands Sections 3 and 4 describe the syntax of sfcross and sfpanel focusing on the main options Sections 5 and 6 illustrate the two commands using simulated data and two empirical applications from the SF literature Finally section 7 offers some conclusions 2 A review of stochastic frontier models We begin our discussion with a general formulation of the SF crosssectional model and then review extensions and improvements that have been proposed in the literature focusing on those models that can be estimated using sfcross and sfpanel Given the large number of estimators allowed by the two commands we deliberately do not discuss the derivation of the corresponding criterion functions We refer the reader to the cited works for details on the estimation of each model A summary of all estimable models and their features is reported in table 1 21 Crosssectional models Consider the following SF model yi α x iβ εi i 1 N 1 εi vi ui 2 vi N0 σ2 v 3 ui F 4 where yi represents the logarithm of the output or cost of the ith productive unit xi is a vector of inputs input prices and quantities in the case of a cost frontier and β is the vector of technology parameters The composed error term εi is the sum or the difference of a normally distributed disturbance vi representing measurement and specification error and a onesided disturbance ui representing inefficiency 2 Moreover ui and vi are assumed to be independent of each other and independent and identically distributed across observations The last assumption about the distribution F of the inefficiency term is needed to make the model estimable Aigner Lovell and Schmidt 1977 assumed a halfnormal distribution that is ui N0σu2 while Meeusen and van den Broeck 1977 opted for an exponential one ui E σu Other commonly adopted distributions are the truncated normal Stevenson 1980 and the gamma distributions Greene 1980ab 2003 The distributional assumption required for the identification of the inefficiency term implies that this model is usually fit by maximum likelihood ML even if modified ordinary leastsquares or generalized method of moments estimators are possible often inefficient alternatives3 In general SF analysis is based on two sequential steps in the first estimates of the model parameters θ are obtained by maximizing the loglikelihood function ℓθ where θ αβσu2σv2 4 In the second step point estimates of inefficiency can be obtained through the mean or the mode of the conditional distribution fui εi where εi yi α xi β The derivation of the likelihood function is based on the independence assumption between ui and vi Because the composite model error εi is defined as εi vi ui its probability density function is the convolution of the two component densities as fεεi 0 to fuui fv εi ui dui 5 Hence the loglikelihood function for a sample of n productive units is ℓθ i1 to n log fεεi θ The marginalization of ui in 5 leads to a convenient closedform expression only for the normalhalf normal normalexponential and normaltruncated normal models In all other cases for example the normalgamma model numerical or simulationbased techniques are necessary to approximate the integral in 5 The second estimation step is necessary because the estimates of the model parameters allow for the computation of residuals ε but not for the inefficiency estimates 2 In this section we consider only production functions However the sign of the ui term in 2 is positive or negative depending on whether the frontier describes a cost or a production function respectively 3 Notice that when a distributional assumption on u is made sfcross and sfpanel estimate model parameters by likelihoodbased techniques 4 Different model parameterizations are used in the SF literature as for example θ αβσ2λ where σ2 σu2 σv2 and λ σu σv Because the main objective of SF analysis is the estimation of technical or cost efficiency a strategy for disentangling this unobserved component from the compounded error is required As mentioned before the most wellknown solutions to this problem proposed by Jondrow et al 1982 and Battese and Coelli 1988 exploit the conditional distribution of u given ε Thus a point estimate of the inefficiencies can be obtained using the mean Eu ε or the mode Mu ε of this conditional distribution Once point estimates of u are obtained estimates of the technical cost efficiency can be derived as Eff exp u where u is either Eu ε or Mu ε5 22 Paneldata models The availability of a richer set of information in panel data allows one to relax some of the assumptions previously imposed and to consider a more realistic characterization of the inefficiencies Pitt and Lee 1981 were the first to extend the model 14 to longitudinal data They proposed the ML estimation of the following normalhalf normal SF model yit α xit β εit i 1N t 1Ti 6 εit vit ui vit N0σv2 ui N0σu2 The generalization of this model to the normaltruncated normal case has been proposed by Battese and Coelli 19886 As pointed out by Schmidt and Sickles 1984 the estimation of an SF model with timeinvariant inefficiency can also be performed by adapting conventional fixedeffects estimation techniques thereby allowing inefficiency to be correlated with the frontier regressors and avoiding distributional assumptions about ui However the timeinvariant nature of the inefficiency term has been questioned especially in the presence of empirical applications based on longpanel datasets To relax this restriction Cornwell Schmidt and Sickles 1990 have approached the problem by proposing the following SF model with individualspecific slope parameters yit α xit β vit uit i 1N t 1Ti u it ωi ωi1 t ωi2 t2 7 in which the model parameters are estimated by extending the conventional fixed and randomeffects paneldata estimators This quadratic specification allows for a unitspecific temporal pattern of inefficiency but requires the estimation of a large number of parameters N 3 5 A general presentation of the postestimation procedures implemented in the sfcross and sfpanel routines is given by Kumbhakar and Lovell 2000 and Greene 2012 to whom we refer the reader for further details 6 The normalexponential model is another straightforward extension allowed by sfpanel Following a slightly different estimation strategy Lee and Schmidt 1993 proposed an alternative specification in which the uit is specified as uit gt ui where gt is represented by a set of time dummy variables This specification is more parsimonious than 7 and it does not impose any parametric form but it is less flexible because it restricts the temporal pattern of uit to be the same for all productive units7 Kumbhakar 1990 was the first to propose the ML estimation of a timevarying SF model in which gt is specified as gt 1 expγt δt21 This model contains only two additional parameters to be estimated γ and δ and the hypothesis of timeinvariant technical efficiency can be easily tested by setting γ δ 0 A similar model called time decay has been proposed by Battese and Coelli 1992 in which gt expη t Ti The common feature of all of these timevarying SF models is that the intercept α is the same across productive units thus generating a misspecification bias in the presence of timeinvariant unobservable factors which are unrelated with the production process but affect the output Therefore the effect of these factors may be captured by the inefficiency term producing biased results Greene 2005a approached this issue through a timevarying SF normalhalf normal model with unitspecific intercepts obtained by replacing 6 by the following specification yit αi xit β εit 8 Compared with previous models this specification allows one to disentangle timevarying inefficiency from unitspecific timeinvariant unobserved heterogeneity For this reason Greene 2005a termed these models TFE or TRE according to the assumptions on the unobserved unitspecific heterogeneity While the estimation of the TRE specification can be easily performed using simulationbased techniques the ML estimation of the TFE variant requires the solution of two major issues related to the estimation of nonlinear paneldata models The first is purely computational because of the large dimension of the parameter space Nevertheless Greene 2005ab showed that a maximumlikelihood dummy variable MLDV approach is computationally feasible also in the presence of a large number of nuisance parameters αi N 1000 The second the socalled incidental parameters problem is an inferential issue that arises 7 Han Orea and Schmidt 2005 and Ahn Hoon Lee and Schmidt 2001 propose estimating the models of Cornwell Schmidt and Sickles 1990 and Lee and Schmidt 1993 respectively through a generalized method of moments GMM approach They show that GMM is preferable because it is asymptotically efficient Currently sfpanel allows for the estimation of the models of Cornwell Schmidt and Sickles 1990 and Lee and Schmidt 1993 through modified leastsquares dummy variables and iterative leastsquares approaches respectively We leave for future updates the implementation of the GMM estimator F Belotti S Daidone G Ilardi and V Atella 725 when the number of units is relatively large compared with the length of the panel In these cases the unitspecific intercepts are inconsistently estimated as N with fixed T because only Ti observations are used to estimate each unitspecific parame ter Neyman and Scott 1948 Lancaster 2002 As shown in Belotti and Ilardi 2012 because this inconsistency contaminates the variance parameters which represent the key ingredients in the postestimation of inefficiencies the MLDV approach appears to be appropriate only when the length of the panel is large enough T 108 Although model 8 may appear to be the most flexible and parsimonious choice among the several existing timevarying specifications one can argue that a portion of the timeinvariant unobserved heterogeneity does belong to inefficiency or these two components should not be disentangled at all The sfpanel command provides options for the estimation of these two extremes for the models of Schmidt and Sickles 1984 Pitt and Lee 1981 and Battese and Coelli 1988 in which all timeinvariant unob served heterogeneity is considered as inefficiency and for the two true specifications in which all timeinvariant unobserved heterogeneity is ruled out from the inefficiency component As pointed out by Greene 2005b neither formulation is completely satis factory a priori and the choice should be driven by the features of the data at hand9 Despite the usefulness of SF models in many contexts a practical disclaimer is in order in both crosssectional and paneldata models the identification through dis tributional assumptions of the two components u and v heavily depends on how the shape of their distributions is involved in defining the shape of the ε distribution Iden tification problems may arise when either the shapes are very similar as pointed out by Ritter and Simar 1997 in the case of small samples for the normalgamma cross sectional model or one of the two components is responsible for most of the shape of the ε distribution The latter is the case where the ratio between the inefficiency and measurement error variability the socalled signaltonoise ratio σuσv is very small or very large In these cases the profile of the log likelihood becomes quite flat producing nontrivial numerical maximization problems 23 Exogenous inefficiency determinants and heteroskedasticity A very important issue in SF analysis is the inclusion in the model of exogenous vari ables that are supposed to affect the distribution of inefficiency These variables which usually are neither the inputs nor the outputs of the production process but nonetheless 8 A common approach to solve this problem is based on the elimination of the αi through a data transformation The consistent estimation of the fixedeffects variant of Greenes 2005a model is still an open research issue in SF literature Promising solutions have been pro posed by Chen Schmidt and Wang 2011 for a homoskedastic normalhalf normal model and Belotti and Ilardi 2012 for a more flexible heteroskedastic specification in normalhalf normal and normalexponential models We are currently working to update the sfpanel command along these directions 9 A way to disentangle unobserved heterogeneity from inefficiency is to include explanatory vari ables that are correlated with inefficiency but not with the remaining heterogeneity The use of untestable exclusion restrictions is a quite standard econometric technique to deal with identifi cation issues affect the productive unit performance have been incorporated in a variety of ways i they may shift the frontier function and the inefficiency distribution ii they may scale the frontier function and the inefficiency distribution and iii they may shift and scale the frontier function and the inefficiency distribution Moreover Kumbhakar and Lovell 2000 stress that in contrast with the linear regression model in which the misspecification of the second moment of the errors distribution determines only efficiency losses the presence of uncontrolled observable heterogeneity in uᵢ and vᵢ may affect the inference in SF models Indeed while neglected heteroskedasticity in vᵢ does not produce any bias for the frontiers parameter estimates it leads to biased inefficiency estimates as we show in section 53 In this section we present the approaches that introduce heterogeneity in the location parameter of the inefficiency distribution and heteroskedasticity of the inefficiency as well as in the parameter of the idiosyncratic error term for the models implemented in the sfcross and sfpanel commands Because these approaches can be easily extended to the paneldata context we deliberately confine the review to the crosssectional framework As pointed out by Greene 2012 researchers have often incorporated exogenous effects using a twostep approach In the first step estimates of inefficiency are obtained without controlling for these factors while in the second the estimated inefficiency scores are regressed or otherwise associated with them Wang and Schmidt 2002 show that this approach leads to severely biased results thus we shall focus only on model extensions based on simultaneous estimation A natural starting point for introducing exogenous variables in the inefficiency model is in the location of the distribution The most wellknown approaches are those suggested by Kumbhakar Ghosh and McGuckin 1991 and Huang and Liu 1994 They proposed to parameterize the mean of the pretruncated inefficiency distribution Basically model 13 can be completed with uᵢ N μᵢ σᵤ² μᵢ zᵢψ where uᵢ is a realization from a truncated normal random variable zᵢ is a vector of exogenous variables including a constant term and ψ is the vector of unknown parameters to be estimated the socalled inefficiency effects One interesting feature of this approach is that the vector zᵢ may include interactions with input variables thus allowing one to test the hypothesis that inefficiency is neutral with respect to its impact on input usage¹⁰ ¹⁰ Battese and Coelli 1995 proposed a similar specification for panel data An alternative approach to analyzing the effect of exogenous determinants on inefficiency is obtained by scaling its distribution Then a model that allows heteroskedasticity in uᵢ and vᵢ becomes a straightforward extension For example Caudill and Ford 1993 Caudill Ford and Gropper 1995 and Hadri 1999 proposed to parameterize the variance of the pretruncated inefficiency distribution in the following way uᵢ N 0 σᵤᵢ² 10 σᵤᵢ² exp zᵢψ 11 Hadri 1999 extends this last specification by allowing the variance of the idiosyncratic term to be heteroskedastic so that 3 can be rewritten as vᵢ N 0 σᵥᵢ² σᵥᵢ² exp hᵢφ 12 where the variables in hᵢ do not necessarily appear in zᵢ As in Wang 2002 both sfcross and sfpanel allow one to combine 9 and 10 for the normaltruncated normal model In postestimation it is possible to compute nonmonotonic effects of the exogenous factors zᵢ on uᵢ A different specification has been suggested by Wang and Schmidt 2002 in which both the location and variance parameters are scaled by the same positive monotonic function hzᵢ ψ Their model uᵢ hzᵢ ψuᵢ with uᵢ N μ σ² is equivalent to the assumption that uᵢ Nμhzᵢ ψ σ²hzᵢ ψ² in which the zᵢ vector does not include a constant term¹¹ ¹¹ We are currently working to extend the sfcross command to normaltruncated normal models with scaling property Wang and Schmidt 2002 728 Stochastic frontier using Stata Table 1 A summary of sfcross and sfpanel estimation and postestimation capabilities Reference Arguments of F 2 Est Loc eff Heter Heter JLMS4 BC5 CI6 d and m1 method3 in u in u in v Crosssectional models Aigner Lovell and Schmidt 1977 hnormal HN ML x x x x x Meeusen and van den Broeck 1977 exponential E ML x x x x x x Stevenson 1980 tnormal TN ML x x x x x x Greene 2003 gamma G SML x x x x Paneldata models Schmidt and Sickles 1984 fe W x Schmidt and Sickles 1984 regls GLS x Pitt and Lee 1981 pl81 HN ML x x x Battese and Coelli 1988 bc88 TN ML x x x Cornwell Schmidt and Sickles 1990 fecss MW x Lee and Schmidt 1993 fels ILS x Kumbhakar 1990 kumb90 HN ML x x x Battese and Coelli 1992 bc92 TN ML x x x Battese and Coelli 1995 bc95 TN ML x x x x x x Greene 2005a tfe E MLDV x x x x x x Greene 2005a tfe HN MLDV x x x x x Greene 2005a tfe TN MLDV x x x x x x Greene 2005a tre E SML x x x x x x Greene 2005a tre HN SML x x x x x Greene 2005a tre TN SML x x x x x x 1 The model option is available in sfpanel The distribution option is available in sfcross 2 Distribution F of u HN half normal E exponential TN truncated normal G gamma 3 Estimation method ML maximum likelihood SML simulated maximum likelihood GLS generalized least squares W within group MW modified within group ILS iterative least squares MLDV maximumlikelihood dummy variable 4 Inefficiency and efficiency estimation via the approach of Jondrow et al 1982 5 Efficiency estimation via the approach of Battese and Coelli 1988 6 Confidence interval for inefficiencies via the approach of Horrace and Schmidt 1996 3 The sfcross command The new Stata command sfcross provides parametric ML estimators of SF models where the default is represented by production The general syntax of this command is as follows sfcross depvar indepvars if in weight distributiondistname emeanvarlistm noconstant usigmavarlistu noconstant vsigmavarlistv noconstant svfrontierlistname svemeanlistname svusigmalistname svvsigmalistname cost simtypesimtype nsimulations base postscore posthessian This command and its panel analog sfpanel are written using the moptimize suite of functions the optimization engine used by ml and share the same features of all Stata estimation commands including access to the estimation results and options for the maximization process see help maximize Stata 112 is the earliest version that can be used to run the commands aweight fweight iweight and pweight are allowed see help weight sfcross supports the svy prefix see help survey The default is the normalexponential model Most options are similar to those of other Stata estimation commands A full description of all available options is provided in the sfcross help file 31 Main options for sfcross distributiondistname specifies the distribution for the inefficiency term as half normal hnormal truncated normal tnormal exponential exponential or gamma gamma The default is distributionexponential emeanvarlistm noconstant may be used only with distributiontnormal With this option sfcross specifies the mean of the truncated normal distribution in terms of a linear function of the covariates defined in varlistm Specifying noconstant suppresses the constant in this function usigmavarlistu noconstant specifies that the technical inefficiency component is heteroskedastic with the variance expressed as a function of the covariates defined in varlistu Specifying noconstant suppresses the constant in this function vsigmavarlistv noconstant specifies that the idiosyncratic error component is heteroskedastic with the variance expressed as a function of the covariates defined in varlistv Specifying noconstant suppresses the constant in this function svfrontierlistname specifies a 1 k vector of initial values for the coefficients of the frontier The vector must have the same length of the parameter vector to be estimated svemeanlistname specifies a 1 km vector of initial values for the coefficients of the conditional mean model This can be specified only with distributiontnormal svusigmalistname specifies a 1 ku vector of initial values for the coefficients of the technical inefficiency variance function svvsigmalistname specifies a 1 kv vector of initial values for the coefficients of the idiosyncratic error variance function cost specifies that sfcross fit a cost frontier model simtypesimtype specifies the method for random draws when distributiongamma is specified runiform generates uniformly distributed random variates halton and genhalton create respectively Halton sequences and generalized Halton sequences where the base is expressed by the prime number in base The default is simtyperuniform See help mata halton for more details on generating Halton sequences nsimulations specifies the number of draws used when distributiongamma is specified The default is nsimulations250 base specifies the number preferably a prime used as a base for the generation of Halton sequences and generalized Halton sequences when distributiongamma is specified The default is base7 Note that Halton sequences based on large primes 10 can be highly correlated and their coverage may be worse than that of the pseudorandom uniform sequences postscore saves an observationbyobservation matrix of scores in the estimation results list Scores are defined as the derivative of the objective function with respect to the parameters This option is not allowed when the size of the scores matrix is greater than the Stata matrix limit see R limits posthessian saves the Hessian matrix corresponding to the full set of coefficients in the estimation results list 32 Postestimation command after sfcross After the estimation with sfcross the predict command can be used to compute linear predictions inefficiency and score variables Moreover the sfcross postestimation command allows one to compute the inefficiency confidence interval through the option ci as well as nonmonotonic marginal effects in the manner of Wang 2002 using when appropriate the option marginal The syntax of the command is the following predict type newvar if in statistic predict type stubnewvarxb newvarv newvaru if in scores where statistic includes xb stdp u m jlms bc ci and marginal xb the default calculates the linear prediction stdp calculates the standard error of the linear prediction u produces estimates of technical or cost inefficiency via Euε using the estimator of Jondrow et al 1982 m produces estimates of technical or cost inefficiency via Muε the mode of the conditional distribution of uε This option is not allowed when the estimation is performed with the distributiongamma option jlms produces estimates of technical or cost efficiency via expEuε bc produces estimates of technical or cost efficiency via Eexpuε the estimator of Battese and Coelli 1988 This option is not allowed when the estimation is performed with the distributiongamma option ci computes the confidence interval using the approach proposed by Horrace and Schmidt 1996 It can be used only when u or bc is specified The default is level95 or a 95 confidence interval If the option level is used in the previous estimation command the confidence interval will be computed using the level This option creates two additional variables newvarLBcilevel and newvarUBcilevel the lower and the upper bound respectively This option is not allowed when the estimation is performed with the distributiongamma option marginal calculates the marginal effects of the exogenous determinants on Eu and Varu The marginal effects are observation specific and are saved in the new variables varnamemM and varnameuV the marginal effects on the mean and the variance of the inefficiency respectively varnamem and varnameu are the names of each exogenous determinant specified in options emeanvarlistm noconstant and usigmavarlistu noconstant marginal can be used only if the estimation is performed with the distributiontnormal option When they are both specified varlistm and varlistu must contain the same variables in the same order This option can be specified in two ways i together with u m jlms or bc and ii alone without specifying newvar scores calculates score variables When the argument of the option distribution is hnormal tnormal or exponential score variables are generated as the derivative of the objective function with respect to the parameters When the argument of the option distribution is gamma they are generated as the derivative of the objective function with respect to the coefficients This difference is due to the different moptmize evaluator type used to implement the estimators see help mata moptimize 4 The sfpanel command sfpanel allows for the estimation of SF paneldata models through ML and leastsquares techniques The general sfpanel syntax is the following sfpanel depvar indepvars if in weight modelmodeltype options As for its crosssectional counterpart version 112 is the earliest version of Stata that can be used to run sfpanel Similarly all types of weights are allowed but the declared weight variable must be constant within each unit of the panel Moreover the command does not support the svy prefix The default model is the timedecay model of Battese and Coelli 1992 A description of the main commandspecific estimation and postestimation options is provided below A full description of all available options is provided in the sfpanel help file 41 Main options for sfpanel True fixed and randomeffects models Greene 2005ab distributiondistname specifies the distribution for the inefficiency term as half normal hnormal truncated normal tnormal or exponential exponential The default is distributionexponential emeanvarlistm noconstant may be used only with distributiontnormal With this option sfpanel specifies the mean of the truncated normal distribution in terms of a linear function of the covariates defined in varlistm Specifying noconstant suppresses the constant in this function usigmavarlistu noconstant specifies that the technical inefficiency component is heteroskedastic with the variance expressed as a function of the covariates defined in varlistu Specifying noconstant suppresses the constant in this function vsigmavarlistv noconstant specifies that the idiosyncratic error component is heteroskedastic with the variance expressed as a function of the covariates defined in varlistv Specifying noconstant suppresses the constant in this function feshow allows the user to display estimates of individual fixed effects along with structural parameters Only for modeltfe simtypesimtype specifies the method to generate random draws for the unitspecific random effects runiform generates uniformly distributed random variates halton and genhalton create respectively Halton sequences and generalized Halton sequences where the base is expressed by the prime number in base The default is simtyperuniform See help mata halton for more details on generating Halton sequences Only for modeltre nsimulations specifies the number of draws used in the simulation The default is nsimulations250 Only for modeltre base specifies the number preferably a prime used as a base for the generation of Halton sequences and generalized Halton sequences The default is base7 Note that Halton sequences based on large primes 10 can be highly correlated and their coverage may be worse than that of the pseudorandom uniform sequences Only for modeltre ML randomeffects timevarying inefficiency effects model Battese and Coelli 1995 emeanvarlistm noconstant fits the Battese and Coelli 1995 conditional mean model in which the mean of the truncated normal distribution is expressed as a linear function of the covariates specified in varlistm Specifying noconstant suppresses the constant in this function usigmavarlistu noconstant specifies that the technical inefficiency component is heteroskedastic with the variance expressed as a function of the covariates defined in varlistu Specifying noconstant suppresses the constant in this function vsigmavarlistv noconstant specifies that the idiosyncratic error component is heteroskedastic with the variance expressed as a function of the covariates defined in varlistv Specifying noconstant suppresses the constant in this function ML randomeffects flexible timevarying efficiency model Kumbhakar 1990 btvarlistbt noconstant fits a model that allows a flexible specification of technical inefficiency handling different types of time behavior using the formulation uit ui1 expvarlistbt1 Typically explanatory variables in varlistbt are represented by a polynomial in time Specifying noconstant suppresses the constant in the function The default includes a linear and a quadratic term in time without constant as in Kumbhakar 1990 42 Postestimation command after sfpanel After the estimation with sfpanel the predict command can be used to compute linear predictions inefficiency and score variables Moreover the sfpanel postestimation command allows one to compute the inefficiency confidence interval through the option ci as well as nonmonotonic marginal effects in the manner of Wang 2002 using when appropriate the option marginal The syntax of the command is the following predict type newvar if in statistic predict type stubnewvarxb newvarv newvaru if in scores where statistic includes xb stdp u u0 m jlms bc ci marginal and trunctlevel xb the default calculates the linear prediction stdp calculates the standard error of the linear prediction u produces estimates of technical or cost inefficiency via Euε using the estimator of Jondrow et al 1982 u0 produces estimates of technical or cost inefficiency via Euε using the estimator of Jondrow et al 1982 when the random effect is zero This statistic can only be specified when the estimation is performed with the modeltre option m produces estimates of technical or cost inefficiency via Muε the mode of the conditional distribution of uε This statistic is not allowed when the estimation is performed with the option modelfecss modelfels modelfe or modelregls jlms produces estimates of technical or cost efficiency via expEuε bc produces estimates of technical or cost efficiency via Eexpuε estimator of the Battese and Coelli 1988 This statistic is not allowed when the estimation is performed with the option modelfecss modelfels modelfe or modelregls ci computes the confidence interval using the approach of Horrace and Schmidt 1996 This option can only be used with the u jlms and bc statistics but not when the estimation is performed with the option modelfels modelbc92 modelkumb90 modelfecss modelfe or modelregls The default is level95 or a 95 confidence interval If the option level is used in the previous estimation command the confidence interval will be computed using the level This option creates two additional variables newvarLBci level and newvarUBci level the lower and the upper bound respectively marginal calculates the marginal effects of the exogenous determinants on Eu and Varu The marginal effects are observation specific and are saved in the new variables varnamemM and varnameuV the marginal effects on the unconditional mean and the variance of inefficiency respectively varnamem and varnameu are the names of each exogenous determinant specified in options emeanvarlistm noconstant and usigmavarlistu noconstant marginal can only be used if estimation is performed with the modelbc95 option or if the inefficiency in modeltfe or modeltre is distributiontnormal When they are both specified varlistm and varlistu must contain the same variables in the same order This option can be specified in two ways i together with u m jlms or bc and ii alone without specifying newvar trunctlevel excludes from the inefficiency estimation the units whose effects are at least at one time period in the upper and bottom tlevel range trunc can only be used if the estimation is performed with modelfe modelregls modelfecss and modelfels scores calculates score variables This option is not allowed when the estimation is performed with the option modelfecss modelfels modelfe or modelregls When the argument of the option model is tfe or bc95 score variables are generated as the derivative of the objective function with respect to the parameters F Belotti S Daidone G Ilardi and V Atella 735 When the argument of the option model is tre bc88 bc92 kumb90 or pl81 they are generated as the derivative of the objective function with respect to the coefficients This difference is due to the different moptimize evaluator type used to implement the estimators see help mata moptimize 5 Examples with simulated data In this section we use simulated data to illustrate sfcross and sfpanel estimation capabilities focusing on some of the models that cannot be estimated using official Stata routines12 51 The normalgamma SF production model There is a large debate in the SF literature about the nonidentifiability of the normal gamma crosssectional model Ritter and Simar 1997 pointed out that this model is difficult to distinguish from the normalexponential one and that the estimation of the shape parameter of the gamma distribution may require large sample sizes up to several thousand observations On the other hand Greene 2003 argued that their result was a matter of degree not a definitive result and that the nonidentifiability of the true value of the shape parameter remains an empirical question In this section we illustrate the sfcross command by fitting a normalgamma SF production model We consider the following datagenerating process DGP yi 1 03x1i 07x2i vi ui i 1 N vi N0 1 ui Γ2 2 where the inefficiency is gamma distributed with shape and scale parameters equal to 2 the idiosyncratic error is N0 1 and the two regressors x1i and x2i are normally distributed with 0 means and variances equal to 1 and 4 respectively The sample size is set to 1000 observations a large size as noted by Ritter and Simar 1997 but in general not so large given the current availability of microdata Let us begin by fitting the normalexponential model using the following syntax 12 We report the Mata code used for the DGP and the models estimation syntax for each example in the sj examples simdatado ancillary file 736 Stochastic frontier using Stata sfcross y x1 x2 distributionexp nolog Stoc frontier normalexponential model Number of obs 1000 Wald chi22 41988 Prob chi2 00000 Log likelihood 24230869 y Coef Std Err z Pz 95 Conf Interval Frontier x1 3709605 068792 539 0000 2361306 5057904 x2 6810641 0339945 2003 0000 6144361 747692 cons 1474677 1131198 130 0192 3691784 0742431 Usigma cons 2173649 0957468 2270 0000 1985989 2361309 Vsigma cons 3827463 1498911 255 0011 0889652 6765274 sigmau 2964844 1419372 2089 0000 2699305 3256505 sigmav 1210911 0907524 1334 0000 1045487 140251 lambda 2448441 2058941 1189 0000 2044895 2851986 estimates store exp predict uhatexp u Note that the normalexponential model is the sfcross default so we might omit the option distributionexponential13 As can be seen although there is only one equation to be estimated in the model the command fits three of Matas moptimize equations see M5 moptimize Indeed given that sfcross allows both the inef ficiency and the idiosyncratic error to be heteroskedastic see table 1 the output also reports variance parameters estimated in a transformed metric according to 11 and 12 respectively In this example the inefficiency is assumed to be homoskedastic so sfcross estimates the coefficient of the constant term in 11 rather than directly esti mating σu To make the output easily interpretable sfcross also displays the variance parameters in their natural metric As expected the normalexponential model produces biased results especially for the frontiers constant term and the inefficiency scale parameter σu We also run the predict command using the u option In this way inefficiency estimates are obtained through the approach of Jondrow et al 1982 Because the inefficiencies are drawn from a gamma distribution a better fit can be obtained using the following command 13 The option nolog allows one to omit the display of the criterion function iteration log sfcross and sfpanel allow one to use all maximize options available for ml estimation commands see help maximize and the additional options postscore and posthessian which report the score and the Hessian as an e vector and matrix respectively F Belotti S Daidone G Ilardi and V Atella 737 sfcross y x1 x2 distributiongamma nsim50 simtypegenha base7 nolog Stoc frontier normalgamma model Number of obs 1000 Wald chi22 43802 Prob chi2 00000 Log simulatedlikelihood 24190008 Number of Randomized Halton Sequences 50 Base for Randomized Halton Sequences 7 y Coef Std Err z Pz 95 Conf Interval Frontier x1 3809769 0670487 568 0000 2495639 5123899 x2 6877634 0336088 2046 0000 6218914 7536354 cons 9361791 4121864 227 0023 1283087 174405 Usigma cons 153519 226486 678 0000 1091286 1979094 Vsigma cons 2734356 333033 082 0412 9261682 379297 sigmau 2154578 2439909 883 0000 1725717 2690016 sigmav 8722163 1452384 601 0000 6293397 1208825 lambda 2470234 1969744 1254 0000 2084171 2856297 gshape 1879186 3845502 489 0000 1125482 2632891 estimates store gamma predict uhatgamma u In the normalgamma crosssectional model the parameters are estimated using sim ulated maximum likelihood SML A better approximation of the loglikelihood function requires the right choice about the number of draws and the way they are created In this example we use generalized Halton sequences simtypegenhalton with base equal to 7 base7 and only 50 draws nsim50 Indeed a Halton sequence gen erally has a more uniform coverage than a sequence generated from pseudouniform random numbers Moreover as noted by Greene 2003 the computational efficiency when compared with that of pseudouniform random draws appears to be at least 10 to 1 Thus in our example the same results can be approximately obtained using 500 pseudouniform draws see help mata halton14 14 For all models fit using SML the default option of sfcross and sfpanel is simtypeuniform with nsim250 In our opinion small values for example 50 for Halton sequences and 250 for pseu douniform random draws are sufficient for exploratory work On the other hand larger values in the order of several hundreds are advisable for more precise results We suggest using Halton se quences rather than pseudorandom random draws However as pointed out by Drukker and Gates 2006 Halton sequences based on large primes d 10 can be highly correlated and their coverage can be worse than that of the pseudorandom uniform sequences 738 Stochastic frontier using Stata As expected in this example the parameters of the normalgamma model are prop erly estimated Furthermore this model is preferable to the normalexponential one as corroborated by the following likelihoodratio test15 lrtest exp gamma Likelihoodratio test LR chi21 817 Assumption exp nested in gamma Prob chi2 00043 Similar conclusions may be drawn by comparing the estimated mean inefficiencies with the true simulated one even if the Spearman rank correlation with the latter is high and very similar for both uhat gamma and uhat exp16 summarize u uhatgamma uhatexp Variable Obs Mean Std Dev Min Max u 1000 4097398 291035 0259262 1990251 uhatgamma 1000 4048885 2839368 4752663 2027557 uhatexp 1000 2964844 264064 363516 1895619 spearman u uhatgamma uhatexp obs1000 u uhatga uhatexp u 10000 uhatgamma 09141 10000 uhatexp 09145 09998 10000 52 Paneldata timevarying inefficiency models Cornwell Schmidt and Sickles 1990 and Lee and Schmidt 1993 provide a fixed effects treatment of models like those proposed by Kumbhakar 1990 and Battese and Coelli 1992 Currently sfpanel allows for the estimation of the models of Corn well Schmidt and Sickles 1990 and Lee and Schmidt 1993 by means of modified leastsquares dummy variables and iterative least squares ILS respectively An in teresting aspect of these models is that although they have been proposed in the SF literature they are actually linear paneldata models with timevarying fixed effects and thus potentially very useful in other contexts However their consistency requires white noise errors and they are less efficient than the GMM estimator proposed by Ahn Hoon Lee and Schmidt 2001 and Han Orea and Schmidt 2005 15 Notice that exp and gamma are the names of the exponential and gamma models estimation results saved with the estimates store command 16 In line with Ritter and Simar 1997 our simulation results indicate that in the normalgamma model a relatively large sample is needed to achieve a reasonable degree of precision in the estimates of inefficiency distribution parameters F Belotti S Daidone G Ilardi and V Atella 739 In this section we report the main syntax to fit such models We start by specifying the following stochastic production frontier translog model yit uit 02x1it 06x2it 06x3it 02x2 1it 01x2 2it 02x2 3it 015x1itx2it 03x1itx3it 03x2itx3it vit vit N0 025 i 1 n t 1 T As already mentioned the main feature of these models is the absence of any distri butional assumption about inefficiency In this example the DGP follows the Lee and Schmidt 1993 model where ui δiξ For each unit the parameter δi is drawn from a uniform distribution in 0 12τ 1 1 with τ 08 The elements of the vector ξ ξ1 ξT are equally spaced between 2 and 2 This setup implies a standard deviation of the inefficiency term σu 183 Once the sample is declared to be a panel see help xtset the models of Lee and Schmidt 1993 and Cornwell Schmidt and Sickles 1990 can be estimated using the following syntaxes sfpanel y x1 x2 x3 x1sq x2sq x3sq x1x2 x1x3 x2x3 modelfels output omitted estimates store fels predict uhatfels u sfpanel y x1 x2 x3 x1sq x2sq x3sq x1x2 x1x3 x2x3 modelfecss output omitted estimates store fecss predict uhatfecss u Notice that we use the predict command with the u option to postestimate inef ficiency As an additional source of comparison we use the same simulated data to assess the behavior of the Schmidt and Sickles 1984 timeinvariant inefficiency model The fixedeffects version of this model can be fit using sfpanel and the official xtreg command However when the estimation is performed using sfpanel the predict command with the option u can be used to obtain inefficiency estimates17 sfpanel y x1 x2 x3 x1sq x2sq x3sq x1x2 x1x3 x2x3 modelfe output omitted estimates store fesssf predict uhatfess u xtreg y x1 x2 x3 x1sq x2sq x3sq x1x2 x1x3 x2x3 fe output omitted estimates store fessxt Table 2 reports the estimation results from the three models Unsurprisingly both the frontier and variance parameters are well estimated in the ls93 and css90 models This result shows that when the DGP follows the model by Lee and Schmidt 1993 the 17 Both xtreg and sfpanel also allow for the estimation of the randomeffects version of this model through the feasible generalized leastsquares approach estimator by Cornwell Schmidt and Sickles 1990 provides reliable results On the other hand being the data generated from a timevarying model variance estimates from the ss84 model show a substantial bias Table 2 Schmidt and Sickles ss84 Cornwell Schmidt and Sickles css90 and Lee and Schmidt ls93 estimation results ss84 css90 ls93 x1 0254 0185 0171 00695 00237 00230 x2 0626 0619 0611 00354 00121 00117 x3 0602 0591 0596 00220 00073 00075 x1sq 0193 0204 0209 00234 00078 00076 x2sq 0099 0103 0101 00080 00027 00026 x3sq 0198 0201 0201 00036 00012 00012 x1x2 0149 0142 0145 00198 00066 00064 x1x3 0293 0295 0295 00130 00043 00043 x2x3 0306 0300 0301 00076 00026 00025 cons 0050 00866 σu 0223 1859 1832 σv 2096 0499 0497 We do not expect large differences with regard to inefficiency scores given the similarities in terms of variance estimates between css90 and ls93 Note that for these models including ss84 inefficiency scores are retrieved in postestimation with the assumption that the best decisionmaking unit is fully efficient18 As seen in the following summarize command both css90 and ls93 average inefficiencies are close to the true values while the Spearman rank correlations are almost equal to 1 As expected the ss84 estimated inefficiencies are highly biased and the corresponding units ranking is completely unreliable 18 This assumption involves calculating ûi α αi with α maxi1nαi in the case of timeinvariant inefficiency models and uit αt αit with αt maxi1nαit in the case of timevarying inefficiency models F Belotti S Daidone G Ilardi and V Atella 741 summarize u uhatfels uhatfecss uhatfess Variable Obs Mean Std Dev Min Max u 2500 1345894 1265155 0 4486315 uhatfels 2500 1669869 1419268 0 5550043 uhatfecss 2500 2214224 1314245 0 6500714 uhatfess 2500 645184 2232496 0 127254 spearman u uhatfels uhatfecss uhatfess obs2500 u uhatls uhatcss uhatess u 10000 uhatfels 09794 10000 uhatfecss 08974 09129 10000 uhatfess 00005 00092 01991 10000 Finally we show additional features of sfpanel i the possibility of computing elasticities via the official lincom command and ii the possibility of performing a constrained fixedeffects estimation which is not yet available with xtreg With respect to the former point it is well known that parameters in a translog pro duction frontier do not represent output elasticities In particular a linear combination of frontier parameters is needed for computing such elasticities Moreover to calculate output elasticities at means we first need to compute and store the mean for each input variable using the following syntax quietly summarize x1 scalar x1m rmean quietly summarize x2 scalar x2m rmean quietly summarize x3 scalar x3m rmean Then the lincom command can be used to combine estimated frontier parameters using the following standard syntax lincom x1 x1sq x1m x1x2x2m x1x3x3m 1 x1 1108946x1sq 1074533x1x2 105167x1x3 0 y Coef Std Err t Pt 95 Conf Interval 1 3203578 05348 599 0000 2154752 4252405 lincom x2 x2sq x2m x1x2x1m x2x3x3m 1 x2 1074533x2sq 1108946x1x2 105167x2x3 0 y Coef Std Err t Pt 95 Conf Interval 1 5751999 0254143 2263 0000 5253585 6250413 742 Stochastic frontier using Stata lincom x3 x3sq x3m x1x3x1m x2x3x2m 1 x3 105167x3sq 1108946x1x3 1074533x2x3 0 y Coef Std Err t Pt 95 Conf Interval 1 156379 0158945 984 0000 1252075 1875505 Finally the constant return to scale CRS hypothesis can be trivially tested by using the following syntax lincom x1 x1sq x1m x1x2x2m x1x3x3m x2 x2sq x2m x1x2x1m x2x3x3m x3 x3sq x3m x1x3x1m x2x3x2m 1 1 x1 x2 x3 1108946x1sq 1074533x2sq 105167x3sq 218348x1x2 2160617x1x3 2126204x2x3 1 y Coef Std Err t Pt 95 Conf Interval 1 0519367 0609852 085 0395 0676648 1715383 In this example the CRS hypothesis cannot be rejected To run a constrained fixed effects estimation we can define the required set of constraints to impose CRS through the official Stata command constraint using the following syntax Constraints definition constraint define 1 x1 x2 x3 1 constraint define 2 x1sq x1x2 x1x3 0 constraint define 3 x2sq x1x2 x2x3 0 constraint define 4 x3sq x1x3 x2x3 0 Then the constrained model can be estimated using sfpanel with the modelfe and constraints1 2 3 4 options F Belotti S Daidone G Ilardi and V Atella 743 sfpanel y x1 x2 x3 x1sq x2sq x3sq x1x2 x1x3 x2x3 modelfe constraints1 2 3 4 Timeinvariant fixedeffects model LSDV Number of obs 2500 Group variable id Number of groups 500 Time variable time Obs per group min 5 avg 50 max 5 1 x1 x2 x3 1 2 x1sq x1x2 x1x3 0 3 x2sq x1x2 x2x3 0 4 x3sq x1x3 x2x3 0 y Coef Std Err z Pz 95 Conf Interval x1 3530365 0851901 414 0000 1860671 520006 x2 5092917 0434568 1172 0000 4241179 5944655 x3 1376718 0270375 509 0000 0846792 1906644 x1sq 0343576 0287476 120 0232 0907019 0219868 x2sq 1282553 0098209 1306 0000 1090067 1475039 x3sq 21594 004442 4861 0000 2072339 2246461 x1x2 0610211 0242651 251 0012 0134624 1085799 x1x3 0266635 0159577 167 0095 0579401 0046131 x2x3 1892764 0092834 2039 0000 2074716 1710813 cons 2326412 1062126 219 0029 0244682 4408141 sigmau 7140381 sigmav 25700643 The constrained frontier estimates are more biased than the unconstrained ones but are still not too far from the true values This is an artifact of our DGP because the scale elasticity has been simulated without imposing CRS 53 True fixed and randomeffects models As already discussed in section 22 the true fixed and randomeffects models allow one to disentangle timeinvariant heterogeneity from timevarying inefficiency In this section we present the main syntax and some of the options to fit such models We start by specifying the following normalexponential stochastic production frontier model yit 1 αi 03x1it 07x2it vit uit vit N0 1 13 uit E 2 i 1 n t 1 T 14 where the nuisance parameters αi i 1 n are drawn from a N0 θ2 with θ 15 In the fixedeffects design TFEDGP the two regressors x1it and x2it are distributed for each unit according to a normal distribution centered in the corresponding uniteffect αi with variances equal to 1 and 4 respectively This design ensures correlation between regressors and individual effects a typical scenario in which the fixedeffects specification represents the consistent choice19 19 Notice that higher values of θ correspond to higher correlations between the regressors and the unitspecific effects 744 Stochastic frontier using Stata As far as the randomeffects design is concerned TREDGP x1it and x2it are not correlated with the unitspecific effects and are distributed according to a normal dis tribution with 0 mean and variances equal to 1 and 4 respectively The generated sample consists of a balanced panel of 1000 units observed for 10 periods for a total of 10000 observations Once the sample is declared as a panel we fit the following models i a normalexponential TFE model on TFEDGP data tfe120 sfpanel yf x1c x2c modeltfe distributionexp rescale output omitted estimates store tfec predict utfec u ii a normalexponential TRE model on TFEDGP data tre1 sfpanel yf x1c x2c modeltre distributionexp nsim50 simtypegenhalton base7 rescale output omitted estimates store trec predict utrec u iii a normalexponential TRE model on TREDGP data tre2 sfpanel yr x1nc x2nc modeltre distributionexp nsim100 simtypegenhalton base7 rescale output omitted estimates store trenc predict utrenc u predict u0trenc u0 As shown in the first column of table 3 when the model is correctly specified the frontier parameters are properly estimated However in this example the MLDV estimator of σv is slightly biased by the incidental parameter problem even if the length of the panel is quite large21 This problem does not seem to affect variance estimates in the tre1 model In this case the parameters are estimated using the SML technique assuming that the unobserved heterogeneity is distributed as N0 θ2 where θ represents the standard deviation of the unobserved heterogeneity and that Eαix1it x2it 0 Thus because the estimates are obtained using the TFEDGP data the frontier and θ parameter estimates are biased 20 Note that yf x1 c and x2 c are the variables from the TFEDGP while yr x1 nc and x2 nc are from the TREDGP 21 See section 22 for a discussion of the MLDV estimator problems in the TFE model F Belotti S Daidone G Ilardi and V Atella 745 Table 3 TFE and TRE estimation results tfe1 tre1 tre2 x1 c 0304 0776 00164 00198 x2 c 0700 0811 00081 00094 x1 nc 0295 00176 x2 nc 0706 00089 cons 1062 1090 00342 00540 σu 2075 2035 2023 σv 0770 1095 0973 θ 0602 1542 On the contrary by fitting a correctly specified TRE model on TREDGP data column tre2 in table 3 all parameters including the frontier ones are accurately estimated After each estimation we use the predict command to obtain inefficiency esti mates As already mentioned option u instructs the postestimation routine to com pute inefficiencies through the estimator of Jondrow et al 1982 see help sfpanel postestimation In the case of the TRE model the predict command also allows for the option u0 to estimate inefficiencies assuming the random effects are zero At this point we can summarize the estimated inefficiencies to compare them with the actual values summarize u utfec utrec utrenc u0trenc Variable Obs Mean Std Dev Min Max u 10000 2004997 200852 0003777 2083139 utfec 10000 2075017 1948148 2008319 2042197 utrec 10000 2034946 1818154 2430926 1876244 utrenc 10000 2025002 1831147 2656734 1998998 u0trenc 10000 2200728 2086419 1338385 1947738 spearman u utfec utrec utrenc u0trenc obs10000 u utfec utrec utrenc u0trec u 10000 utfec 07654 10000 utrec 07541 09291 10000 utrenc 07700 09925 09464 10000 u0trenc 06297 07313 08168 07965 10000 746 Stochastic frontier using Stata All the estimates of Jondrow et al 1982 are very close to the true simulated ones u Actually the estimated average inefficiency after a correctly specified TRE model shows a lower bias than the estimated average inefficiency after a correctly specified TFE model This is due to the incidental parameters problem Also note the good performances of the TRE model when it is fit on the TFEDGP data u tre c Introducing heteroskedasticity Finally we deal with the problem of heteroskedasticity a very important issue for applied research For both TFE and TRE models we compare the estimates obtained from a model that neglects heteroskedasticity with those obtained from a heteroskedastic one To introduce heteroskedasticity we replace equations 1314 with the following vit N0 σvit uit E σuit σvit exp 051 5 zvit σuit exp 052 1 zuit where both inefficiency and idiosyncraticerror scale parameters are now a function of a constant term and of an exogenous covariate zuit and zvit drawn from a standard normal random variable Note that because of the introduction of heteroskedasticity we will deal with average σu and σv which in our simulated sample are approximately 31 and 17 respectively In this case each observation has a different signaltonoise ratio which implies an average of about 19 We estimate four different models i a homoskedastic TFE model on heteroskedastic TFEDGP data tfe1 sfpanel yf x1c x2c modeltfe distributionexp rescale output omitted estimates store tfehom predict utfehom u ii a heteroskedastic TFE model on heteroskedastic TFEDGP data tfe2 sfpanel yf x1c x2c modeltfe distributionexp usigmazu vsigmazv output omitted estimates store tfehet predict utfehet u iii a homoskedastic TRE model on heteroskedastic TREDGP data tre1 sfpanel yr x1nc x2nc modeltre distributionexp nsim50 simtypegenhalton base7 rescale output omitted estimates store trehom predict utrehom u F Belotti S Daidone G Ilardi and V Atella 747 iv a heteroskedastic TRE model on heteroskedastic TREDGP data tre2 sfpanel yr x1nc x2nc modeltre distributionexp usigmazu vsigmazv nsim50 simtypegenhalton base7 rescale output omitted estimates store trehet predict utrehet u predict u0trehet u0 Estimation results are reported in table 4 As expected tfe1 variance estimates are biased by both the incidental parameters problem and the neglected heteroskedasticity in u and v These estimates can be significantly improved by considering both sources of heteroskedasticity using the options usigmavarlist u and vsigmavarlist v tfe2 Exactly the same argument applies in the TRE case tre1 versus tre2 but without the incidental parameters problem Table 4 TFE and TRE estimation results homoskedasticity versus heteroskedasticity tfe1 tfe2 tre1 tre2 x1 c 0324 0295 00271 00245 x2 c 0723 0732 00134 00121 x1 nc 0316 0310 00290 00264 x2 nc 0681 0689 00147 00135 cons 1576 1113 00637 00652 σu 3717 3264 3642 3168 σv 1185 1402 1526 1693 θ 1579 1565 As we mentioned in section 23 neglecting heteroskedasticity in u and v leads to biased inefficiency estimates This conclusion is confirmed by the summarize command summarize u utfehom utfehet utrehom utrehet u0trehet Variable Obs Mean Std Dev Min Max u 10000 3091925 3915396 000169 5220689 utfehom 10000 3717061 3941147 3442658 5154804 utfehet 10000 3271297 3828366 2642199 5206564 utrehom 10000 3641955 3788298 3739219 5176109 utrehet 10000 3173224 3709123 3241621 5183721 u0trehet 10000 32855 3844297 1828969 542632 748 Stochastic frontier using Stata The average inefficiency is upward biased by about 15 for both TFE and TRE models in which heteroskedasticity has been neglected A slightly better result is also obtained in terms of Spearmans rank correlation spearman u utfehom utfehet utrehom utrehet u0trehet obs10000 u utfem utfet utrem utret u0tret u 10000 utfehom 07287 10000 utfehet 07536 09589 10000 utrehom 07380 09830 09531 10000 utrehet 07623 09461 09835 09642 10000 u0trehet 07039 08173 08455 08944 09121 10000 6 Empirical applications In this section we illustrate sfcross and sfpanel capabilities through two empirical applications from the SF literature The first analyzes the cost inefficiency of Swiss railways using data from the Swiss Federal Office of Statistics on public transport companies the second focuses on the technical inefficiency of Spanish dairy farms using data from a voluntary recordkeeping program22 61 Swiss railways This application is based on an unbalanced panel of 50 railway companies over the period 19851997 which resulted in 605 observations We think that this application is interesting for at least two reasons i cost frontiers are much less diffuse in the literature compared with production frontiers given the lack of reliable cost and price data and ii the length of the panel makes this database quite unusual in the SF literature A detailed description of the Swiss railway transport system and complete information on the variables used are available in Farsi Filippini and Greene 2005 To estimate a CobbDouglas cost frontier we impose linear homogeneity by normalizing total costs and prices through the price of energy Therefore the model can be written as lnTCit PEit β0 βY ln Yit βQ ln Qit βN ln Nit βpk ln Pkit Peit βPl ln Plit Peit t19861997 βt dyeart uit vit 15 where i and t are the subscripts denoting the railway company and year respectively As is common uit is interpreted as a measure of cost inefficiency Two output measures are included in the cost function passenger output and freight output Length of network 22 Both datasets are freely available from the webpage of professor William Greene httppeoplesternnyueduwgreene F Belotti S Daidone G Ilardi and V Atella 749 is included as an output characteristic Further we have price data for three inputs capital labor and energy All monetary values including total costs are in 1997 Swiss Francs CHF We have also included a set of time dummies dyeart to control for unobserved timedependent variation in costs We consider three timevarying inefficiency specificationsthe Kumbhakar 1990 model kumb90 the Battese and Coelli 1992 model bc92 and the Greene 2005a randomeffects model treand three timeinvariant models With respect to the latter group we estimate the fixedeffects version of the Schmidt and Sickles 1984 model ss84 the Pitt and Lee 1981 pl81 model and the Battese and Coelli 1988 bc88 model All models are fit assuming that the inefficiency is halfnormally distributed that is all except bc88 and bc92 in which u Nμσu2 and ss84 in which no distributional assumption is made The choice of also including Greenes specification is driven by the multioutput technology that characterizes a railway company for which unmeasured quality captured by the random effects may play an important role in the production process Finally as a benchmark we fit a pooled crosssectional model pcs Table 5 shows the results Coefficient estimates of input prices and outputs are all significant across the seven models and have the expected signs positive marginal costs and positive ownprice elasticities Looking at table 6 we further observe that the three timeinvariant specifications provide inefficiency estimates that are highly correlated Perhaps the most interesting result is that inefficiency scores obtained from kumb90 and bc92 models are also highly correlated with those coming from timeinvariant models table 6 and figure 1 This is not surprising because the two timeinvariance hypotheses H0 t t2 0 in the kumb90 model and H0 η 0 in the bc92 specification cannot be rejected at the 5 level Hence we may conclude that there is evidence of timeinvariant technical inefficiency in the Swiss railway transport system at least for the study period Consistently with this result we also find that the tre model provides inefficiency estimates that have no link with those obtained from any of the other models Moreover because of a very low estimate of the inefficiency variance the estimated signaltonoise ratio λ is the lowest one In our opinion these results are driven from the peculiar timevarying inefficiency specification of this model Indeed when the inefficiency term is constant over time the tre specification does not allow one to disentangle timeinvariant unobserved heterogeneity from inefficiency This interpretation is supported by the fact that the estimated standard deviation of the random effects θ dominates the inefficiency one σu 750 Stochastic frontier using Stata Table 5 Swiss railways estimation results 50 firms for a total of 605 observations pcs ss pl81 bc88 kumb90 bc92 tre bse bse bse bse bse bse bse lnY 0492 0114 0200 0199 0193 0199 0201 0015 0032 0034 0033 0033 0033 0026 lnQ 0030 0014 0021 0021 0020 0020 0028 0006 0006 0006 0006 0006 0006 0005 lnN 0393 0448 0485 0503 0477 0499 0583 0027 0051 0045 0047 0044 0047 0034 lnpk 0171 0318 0310 0311 0311 0313 0311 0032 0017 0017 0017 0017 0017 0017 lnpl 0592 0546 0548 0546 0538 0543 0560 0074 0037 0037 0037 0037 0037 0037 dyear1986 0009 0010 0009 0009 0015 0008 0015 0056 0015 0015 0015 0015 0015 0015 dyear1987 0003 0020 0012 0012 0023 0009 0018 0056 0015 0015 0015 0017 0015 0015 dyear1988 0010 0039 0028 0027 0043 0023 0034 0057 0015 0015 0015 0019 0016 0016 dyear1989 0036 0065 0052 0052 0070 0046 0058 0057 0016 0016 0016 0021 0016 0016 dyear1990 0024 0084 0068 0068 0086 0060 0074 0058 0016 0016 0016 0022 0017 0016 dyear1991 0030 0098 0078 0078 0096 0069 0086 0058 0017 0018 0017 0024 0019 0018 dyear1992 0046 0111 0094 0094 0109 0083 0101 0058 0017 0017 0017 0023 0019 0017 dyear1993 0015 0100 0081 0081 0092 0069 0089 0057 0017 0017 0017 0023 0020 0017 dyear1994 0001 0082 0063 0063 0069 0049 0070 0056 0017 0017 0017 0022 0020 0017 dyear1995 0019 0059 0048 0047 0045 0031 0050 0057 0016 0016 0016 0022 0021 0016 dyear1996 0027 0037 0028 0027 0018 0010 0017 0057 0017 0016 0016 0022 0022 0017 dyear1997 0019 0038 0030 0029 0009 0009 0029 0060 0018 0017 0017 0023 0024 0017 Constant 8310 2682 4895 4929 4626 4871 4894 0976 0652 0643 0634 0637 0637 0531 Continued on next page F Belotti S Daidone G Ilardi and V Atella 751 pcs ss pl81 bc88 kumb90 bc92 tre bse bse bse bse bse bse bse t 0023 0015 t2 0002 0001 η 0002 0002 λ 2882 7803 11366 7716 23930 7887 1634 σ 0464 0560 0807 0551 1682 0562 0098 σu 0438 0555 0804 0546 1681 0557 0083 σv 0152 0071 0071 0071 0070 0071 0051 θ 0347 Estimated cost inefficiencies ûit Mean 0350 0807 0663 0679 0687 0682 0091 SD 0233 0550 0429 0425 0445 0428 0076 Min 0060 0000 0015 0020 0015 0019 0018 Max 1134 2507 2006 1991 2124 2031 0629 Log likelihood 116572 595159 596523 597649 597285 595516 Notes Standard errors for ancillary parameters are not reported Table 6 Swiss railways correlation of inefficiency estimates Variables pcs ss84 pl81 bc88 kumb90 bc92 tre pcs 1000 ss84 0439 1000 pl81 0595 0975 1000 bc88 0608 0971 0991 1000 kumb90 0573 0984 0990 0998 1000 bc92 0603 0974 0991 1000 0998 1000 tre 0140 0378 0405 0407 0400 0406 1000 Figure 1 Swiss railways inefficiency scatterplots and output volumes heterogeneity of output and less than ideal proxies for inputs The output variable is given by the liters of milk produced per year This measure explains only partially the final output of this industry milk can also be considered an intermediate input to produce dairy products Furthermore variables such as slaughtered animals should also be considered part of the final output The functional form employed in the empirical analysis is the following translog production function with time dummy variables to control for neutral technical change ln yit β0 sumj14 βj ln xjit 12 sumj14 sumk14 βjk ln xjit ln xkit sumt19931998 βt dyeart uit vit 16 where j and t are the subscripts denoting farm and year respectively Four inputs have been employed in the production frontier number of milking cows x1 number of manequivalent units x2 hectares of land devoted to pasture and crops x3 and kilograms of feedstuffs fed to the dairy cows x4 More details on these variables are available in Cuesta 2000 and Alvarez and Arias 2004 We have fit three models with timevarying inefficiency the normalhalf normal Kumbhakar 1990 model kumb90 a randomeffects model fit through the feasible generalized leastsquares method the Cornwell Schmidt and Sickles 1990 model css90 fit through the modified leastsquares dummy variable technique and finally the Lee and Schmidt 1993 model ls93 fit through ILS Note that the last two models are fit using approaches that do not allow intercept β0 and time dummies dyeart F Belotti S Daidone G Ilardi and V Atella 753 to be simultaneously included in the frontier equation Finally we also considered two models with timeinvariant inefficiency that is the uit term reduced to ui in 16 the first was proposed by Schmidt and Sickles 1984 and estimated without any distri butional assumption through the leastsquares dummy variable approach ss84 the second was proposed by Pitt and Lee 1981 and estimated through ML assuming a halfnormal inefficiency pl81 Table 7 reports the results of our exercise There is a certain degree of similarity between the different models because both parameter significance and magnitudes are comparable For the ss84 css90 and ls93 models the most efficient firm in the sample for each period is considered as fully efficient thus the smallest value of inefficiency is 0 On average and as expected the css90 model shows a higher level of inefficiency and its distribution also has more variability while the other models seem to behave very similarly in this application Finally as we can see in table 8 linear correlations between inefficiencies are very high This does not come as a surprise given the similarity of the estimated frontier parameters and it looks like an indication that in medium short panels and in certain economic sectors and contexts a timeinvariant inefficiency specification is a valid solution 754 Stochastic frontier using Stata Table 7 Spanish dairy farms estimation results 247 firms for a total of 1482 obs ss84 css90 ls93 kumb90 pl81 bse bse bse bse bse x1 0642 0527 0641 0661 0660 0036 0065 0036 0028 0028 x2 0037 0043 0037 0038 0041 0017 0027 0017 0015 0015 x3 0011 0079 0010 0050 0049 0025 0063 0025 0018 0018 x4 0308 0226 0307 0351 0356 0020 0035 0020 0018 0017 x11 0135 0187 0133 0308 0314 0157 0192 0155 0171 0178 x22 0002 0060 0001 0111 0112 0069 0111 0068 0064 0067 x33 0242 0168 0243 0129 0131 0188 0317 0187 0119 0115 x44 0105 0125 0105 0112 0118 0050 0084 0050 0048 0049 x12 0010 0059 0009 0060 0064 0073 0100 0072 0077 0081 x13 0084 0114 0085 0088 0091 0102 0158 0101 0090 0090 x14 0075 0142 0074 0140 0146 0083 0132 0082 0084 0088 x23 0001 0067 0002 0020 0011 0050 0107 0050 0049 0050 x24 0011 0062 0011 0025 0025 0041 0060 0041 0039 0040 x34 0012 0110 0013 0015 0017 0046 0085 0046 0041 0041 dyear1994 0035 0042 0027 0007 0010 0007 dyear1995 0062 0072 0048 0009 0014 0008 dyear1996 0072 0078 0052 0010 0016 0009 dyear1997 0075 0074 0051 0010 0017 0009 dyear1998 0092 0077 0064 0012 0018 0010 Constant 11512 11695 11711 0016 0019 0016 Continued on next page ss84 css90 ls93 kumb90 p181 bse bse bse bse bse t 0347 0212 t2 0045 0028 λ 1948 3711 2010 4485 2775 σ 0168 0237 0171 0356 0230 σu 0149 0229 0153 0348 0216 σv 0077 0062 0076 0077 0078 Estimated technical inefficiencies uit Mean 0315 0531 0316 0182 0179 SD 0149 0227 0150 0119 0117 Min 0000 0000 0000 0008 0009 Max 0873 1412 0879 0667 0623 Log likelihood 1355248 1351826 Notes Clusterrobust standard errors are in parentheses Standard errors for ancillary parameters are not reported Table 8 Spanish dairy farms correlation of inefficiency estimates Variables ss84 css90 ls93 kumb90 pl81 ss84 1000 css90 0868 1000 ls93 1000 0871 1000 kumb90 0938 0726 0936 1000 pl81 0931 0709 0929 0995 1000 7 Concluding remarks In this article we introduced the new Stata commands sfcross and sfpanel which implement an extensive array of SF models for crosssectional and panel data With respect to the available official Stata commands frontier and xtfrontier we add multiple features for estimating frontier parameters and for postestimating unit inefficiency and efficiency In the development of the commands we widely exploit Mata potentiality By using Mata structures we provide a very readable code that can be easily developed further by other Stata users We illustrated the commands estimation capabilities through simulated data focusing on some of the models that cannot be estimated using official Stata commands 756 Stochastic frontier using Stata Finally we illustrated the proposed routines using real datasets under different possible empirical scenarios short versus long panels cost versus production frontiers homoge neous versus heterogeneous outputs 8 Acknowledgments We are grateful to David Drukker and all participants at the 2009 Italian Stata Users Group meeting for useful comments We thank William Greene for valuable advice and discussions and for maintaining an excellent webpage and making available several databases two of which we have extracted and used in the empirical applications 9 References Ahn S C Y Hoon Lee and P Schmidt 2001 GMM estimation of linear panel data models with timevarying individual effects Journal of Econometrics 101 219255 Aigner D J C A K Lovell and P Schmidt 1977 Formulation and estimation of stochastic frontier production function models Journal of Econometrics 6 2137 Alvarez A and C Arias 2004 Technical efficiency and farm size A conditional analysis Agricultural Economics 30 241250 Battese G E and T J Coelli 1988 Prediction of firmlevel technical efficiencies with a generalized frontier production function and panel data Journal of Econometrics 38 387399 1992 Frontier production functions technical efficiency and panel data With application to paddy farmers in India Journal of Productivity Analysis 3 153169 1995 A model for technical inefficiency effects in a stochastic frontier production function for panel data Empirical Economics 20 325332 Belotti F and G Ilardi 2012 Consistent estimation of the true fixedeffects stochas tic frontier model CEIS Tor Vergata Research Paper Series Caudill S B and J M Ford 1993 Biases in frontier estimation due to heteroscedas ticity Economics Letters 41 1720 Caudill S B J M Ford and D M Gropper 1995 Frontier estimation and firm specific inefficiency measures in the presence of heteroscedasticity Journal of Business and Economic Statistics 13 105111 Chen YY P Schmidt and HJ Wang 2011 Consistent estimation of the fixed effects stochastic frontier model httpwwweconomicsuwocanewslettermisc2011schmidt nov11pdf Cornwell C P Schmidt and R C Sickles 1990 Production frontiers with cross sectional and timeseries variation in efficiency levels Journal of Econometrics 46 185200 F Belotti S Daidone G Ilardi and V Atella 757 Cuesta R A 2000 A production model with firmspecific temporal variation in tech nical inefficiency With application to Spanish dairy farms Journal of Productivity Analysis 13 139158 Drukker D M and R Gates 2006 Generating Halton sequences using Mata Stata Journal 6 214228 Farsi M M Filippini and W Greene 2005 Efficiency measurement in network indus tries Application to the Swiss railway companies Journal of Regulatory Economics 28 6990 Greene W 2005a Reconsidering heterogeneity in panel data estimators of the stochas tic frontier model Journal of Econometrics 126 269303 2005b Fixed and random effects in stochastic frontier models Journal of Productivity Analysis 23 732 Greene W H 1980a Maximum likelihood estimation of econometric frontier functions Journal of Econometrics 13 2756 1980b On the estimation of a flexible frontier production model Journal of Econometrics 13 101115 2003 Simulated likelihood estimation of the normalgamma stochastic frontier function Journal of Productivity Analysis 19 179190 2012 Econometric Analysis 7th ed Upper Saddle River NJ Prentice Hall Hadri K 1999 Estimation of a doubly heteroscedastic stochastic frontier cost function Journal of Business and Economic Statistics 17 359363 Han C L Orea and P Schmidt 2005 Estimation of a panel data model with para metric temporal variation in individual effects Journal of Econometrics 126 241267 Horrace W C and P Schmidt 1996 Confidence statements for efficiency estimates from stochastic frontier models Journal of Productivity Analysis 7 257282 Huang C J and JT Liu 1994 Estimation of a nonneutral stochastic frontier production function Journal of Productivity Analysis 5 171180 Jondrow J C A K Lovell I S Materov and P Schmidt 1982 On the estimation of technical inefficiency in the stochastic frontier production function model Journal of Econometrics 19 233238 Kumbhakar S C 1990 Production frontiers panel data and timevarying technical inefficiency Journal of Econometrics 46 201211 Kumbhakar S C S Ghosh and J T McGuckin 1991 A generalized production fron tier approach for estimating determinants of inefficiency in US dairy farms Journal of Business and Economic Statistics 9 279286 758 Stochastic frontier using Stata Kumbhakar S C and C A K Lovell 2000 Stochastic Frontier Analysis Cambridge Cambridge University Press Lancaster T 2002 The incidental parameters problem since 1948 Journal of Econo metrics 95 391414 Lee Y H and P Schmidt 1993 A production frontier model with flexible tempo ral variation in technical efficiency In The Measurement of Productive Efficiency Techniques and Applications ed H O Fried C A Knox Lovell and S S Schmidt 237255 New York Oxford University Press Meeusen W and J van den Broeck 1977 Efficiency estimation from CobbDouglas production functions with composed error International Economic Review 18 435 444 Neyman J and E Scott 1948 Consistent estimates based on partially consistent observations Econometrica 16 132 Pitt M M and LF Lee 1981 The measurement and sources of technical inefficiency in the Indonesian weaving industry Journal of Development Economics 9 4364 Ritter C and L Simar 1997 Pitfalls of normalgamma stochastic frontier models Journal of Productivity Analysis 8 167182 Schmidt P and R C Sickles 1984 Production frontiers and panel data Journal of Business and Economic Statistics 2 367374 Stevenson R E 1980 Likelihood functions for generalized stochastic frontier estima tion Journal of Econometrics 13 5766 Wang HJ 2002 Heteroscedasticity and nonmonotonic efficiency effects of a stochastic frontier model Journal of Productivity Analysis 18 241253 Wang HJ and P Schmidt 2002 Onestep and twostep estimation of the effects of exogenous variables on technical efficiency levels Journal of Productivity Analysis 18 129144 About the authors Federico Belotti is a researcher at the Centre for Economics and International Studies CEIS of the University of Rome Tor Vergata Silvio Daidone is a research associate at the Centre for Health Economics CHE of the Uni versity of York and an economist at the Agricultural Development Economics Division of the Food and Agriculture Organization of the United Nations Giuseppe Ilardi is a researcher at the Economic and Financial Statistics Department of the Bank of Italy Vincenzo Atella is an associate professor at the University of Rome Tor Vergata View publication stats CHAPTER 3 Theory of Bank Efficiency and Bank Competition 31 INTRODUCTION Efficiency is a commonly used measure in the production process There are inputs and outputs in production Higher efficiency mainly means any unitcompany has the ability to use the same amount of input to generate a higher volume of output whereas on the other hand higher efficiency can also be interpreted as any unitcompany has the ability to minimize the input to produce the same volume of output This interpretation of effi ciency purely considers the volumes of input and output whereas in the production process not only is the volume of output and input being considered by the managers and regulators but more attention has also been given to minimize cost as well as maximize profit and revenue Thus it is important to analyse efficiency in technical terms as well as evaluate efficiency from a cost profit and revenue perspective More specifically managers and regulators are greatly concerned about how to use a certain amount of input to minimize the cost and maximize the profit and revenue The chapter mainly explains the theories of bank efficiency as well as bank competition Specifically this chapter is organized as follows Section 32 discusses the framework of efficiency and is divided into four subsections which explain cost efficiency technical efficiency profit efficiency and revenue efficiency on a onebyone basis With regards to bank competition four indicators to measure bank competition will be explained which include Concentration ratio PanzareRosse Hstatistic Lerner index and Boone indicator 32 THE FRAMEWORK OF EFFICIENCY 321 Technical Efficiency A simple measure of firm efficiency is defined by Farrell 1957 whose work is derived from Debreu 1951 and Koopmans 1951 He argued that technical efficiency reflects a firms ability to obtain maximal output from a Efficiency and Competition in Chinese Banking ISBN 9780081000748 httpdxdoiorg101016B9780081000748000030 2016 Elsevier Ltd All rights reserved 45 given set of inputs Farrell explained his idea by making the assumption that firms use two inputs X1 and X2 to produce one output y and the production is under the assumption of Constant Return to Scale CRS see Fig 31 In other words an increase decrease in the inputs leads to the same proportional increase decrease of the output The unit isoquant SS0 describes the technological set to produce the certain amount of output using the combination of the inputs X1 and X2 In other words SS0 shows the minimum number of inputs needed to produce one unit of output All the production along this curve SS0 is supposed to be perfectly efficient whereas any other points above or located at the right of the curve such as the point P are regarded as inefficient production due to the fact that the number of inputs used in the production to produce one unit of output is more than the efficient production The distance QP represents the technical inefficiency of the firm It also represents the number of inputs which can be reduced without any in fluence on the output production In other words it represents the number of inputs which can be reduced without any decrease in the output The percentage of the input reductiontechnical inefficiency level for the point P can be represented by the ratio QPOP whereas the technical efficiency of a firm can be measured by the ratio OQOP The value of technical efficiency ranges from 0 to 1 The value of 1 means that the firm is fully technically efficient The condition for the assumption of constant return to scale CRS is that all the firms operate at the optimal scale which is impossible sometimes because of imperfect competition The variable return to scale VRS Q R S P A O X1Y X2Y A S Q Figure 31 The measurement of technical efficiency Coelli T 1996 A Guide to DEAP Version 21 A Data Envelopment Analysis Computer Program CEPA working paper 199608 46 Efficiency and Competition in Chinese Banking further decomposes the technical efficiency into pure technical efficiency PTE and scale efficiency SE In Fig 32 we make another assumption which is that firms use one input X to product one output Y The constant return to scale frontier is represented by CRS firms production can be laying on the curve or located to the right side the technical inefficiency for the firm P is PPc over AP whereas the technical efficiency can be repre sented by the ratio APcAP The SE considers the possibility that firms do not operate at the optimal size The CRS assumption is replaced by the VRS assumption to measure the SE The variable return to scale frontier is represented by VRS under which the technical inefficiency for firm P is PPv over AP the PTE for the firm P can be represented by the ratio APvAP whereas the SE can be represented by the ratio APcAPv If the firm operates under CRS the value of scale efficiency equals 1 whereas the value of SE will be less than 1 in a VRS situation In other words firms have scale inefficiency 322 Cost Efficiency Rather than focussing on manipulating the inputs and outputs in banking operations which is reflected by technical efficiency more practically bank managers and banking regulatory authorities emphasize minimizing the cost of banking operations one of the indicator used to measure this perfor mance is cost efficiency It is also called economic efficiency It can be Y CRS VRS NIRS A P Q Pv Pc O X Figure 32 The measurement of pure technical efficiency and scale efficiency CRS Constant Return to Scale NIRS nonincreasing returns to scale Coelli T 1996 A Guide to DEAP Version 21 A Data Envelopment Analysis Computer Program CEPA working paper 199608 Theory of Bank Efficiency and Bank Competition 47 represented by the ratio of minimal cost achieved by the bestpractice bank to the actual cost incurred by a specific bank The banks with higher cost efficiency normally have higher ability to choose their inputs and mixes according to their prices the purpose of which is to minimize the cost whereas there are two sources of cost inefficiency One is attributed to the technology in other words one part of cost inefficiency is derived from technical inefficiency whereas the other part of cost inefficiency is due to the suboptimal allocation of resources that is part of cost inefficiency is related to allocative inefficiency The overall cost efficiency can be calcu lated by multiplying the technical efficiency by allocative efficiency 323 Revenue Efficiency Berger and Mester 1997 argue that revenue efficiency measures the change in a banks revenue adjusted for a random error relative to the estimated revenue obtained from producing an output bundle as efficiently as the best practice bank Revenue efficiency emphasizes the fact that the banking operation aims to maximize revenue and revenue efficiency is measured by the ratio of actual revenue to maximum revenue There are two sources of revenue inefficiency More specifically revenue inefficiency can be attrib uted to outputoriented technical efficiency or outputoriented allocative inefficiency The technical inefficiency can be interpreted as producing too few outputs given the input quantities whereas the allocative inefficiency means nonoptimal combination of outputs are produced given their prices The following figure Fig 33 explains revenue efficiency in the frontier methodological framework We assume that the bank uses input X to produce two outputs y1 and y2 TT0 represents the production possibility curve Under the outputoriented assumption a bank can be regarded as technically efficient if the operation point is located on TT0 Thus we can say banks B C D and E are technically efficient under the outputoriented assumption Due to the fact that Bank A locates below the frontier it is regarded as inefficient In other words compared to revenueefficient banks the bank which is located at point A has the potential to increase the production levels of both outputs to point A0 located on the production possibility frontier Therefore for Bank A the distance AA0 represents the outputoriented technical inefficiency The outputoriented technical ef ficiency TE is measured by the ratio of actual output to maximum output of the bank Using the above diagram the outputoriented TE can be represented by the ratio of OAOA0 It is assumed that the isorevenue line is represented by PP0 Bank D is regarded as revenue efficient Because the 48 Efficiency and Competition in Chinese Banking Figure 33 Revenue efficiency Kumar S Gulati R 2013 Regulation and Efficiency of Indian Banks Springer India cost at point D equals the revenue at point F the outputoriented allocative efficiency can be measured by the ratio of OAOF If it is assumed that the production is located at the point A the increase in revenue can be represented by the distance AF Finally the revenue efficiency of the bank can be represented by the ratio of OAOF Due to the fact that revenue efficiency includes both outputoriented TE and allocative efficiency it is considered as a composite efficiency measure The revenue inefficiency for Bank A can be represented by the distance AF The relationship between revenue efficiency TE and allocative efficiency can be expressed as Revenue efficiency technical efficiency allocative efficiency 324 Profit Efficiency Profit efficiency is a more comprehensive type of efficiency concept which considers both cost and revenue performance The profit efficiency assumes that both revenue and cost are controlled by managers In empirical research the analysis of profit efficiency is preferred to cost efficiency due to the fact the operating revenues and loan losses are ignored by cost efficiency The profit efficiency can be defined as how close a bank is able to produce the maximum possible profit given the levels of input and output process and other exogenous conditions Under the given conditions profit efficiency can be improved if the banks profits move closer to the profit of a benchmark bank bestpractice bank with highest profit The profit efficiency can be measured by the ratio of actual profit over the maximum profit The score of profit efficiency is bounded above and has a maximum value of 1 In other words the value of profit efficiency ranges from minus infinity to 1 The value of profit efficiency for the bestpractice bank is 1 Negative values of profit efficiency are possible due to the fact that banks can throw away more than 100 of the potential profits The following figure is used to better illustrate the concept of profit efficiency In the following figure Fig 34 the production frontier is represented by the curve OQ The actual inputoutput combination of bank A which is represented by point A is xAyA Thus Bank A profit can be calculated as qAyA pAxA All the production points located on the line CD yield the same amount of profit as the point A The objective of bank A is to reach highest isoprofit line parallel to CD that can be attained at any point on or below the curve OQ Point B will be the highest point which is also the tangency of EF and OQ If it is assumed that the inputoutput combination for point B is x and y line OE represents the maximum profit achieved and Bank A is able to achieve the maximum profit as long as it reaches one point on Line EF Say point A The calculation of maximum profit can be expressed as qAyA pAxA this also equals qByB pBxB The profit efficiency of bank A can be calculated by the ratio of actual profit divided by the maximum profit Figure 34 Profit efficiency Kumar S Gulati R 2013 Regulation and Efficiency of Indian Banks Springer India 33 THE MEASUREMENT OF EFFICIENCY The methods used to measure efficiency can be mainly divided into two streams one is called the parametric approach and the other one is called the nonparametric approach The former can be more specifically divided into three analyses which include stochastic frontier analysis SFA distri bution free approach DFA and think frontier approach TFA The non parametric approach can be further divided into two analyses which are data envelopment analysis DEA and free disposal hull FDH analysis With regards to the parametric approach a prespecified functional form for the bestpractice frontier is required The efficiency of a firm is compared to the benchmark firm firm with best practice A firm is inefficient if the cost is higher than the best practice from or the profit is lower than the benchmark firm The nonparametric approach is a mathematical programming technique to estimate efficiency and Stephards distance function methodology is implemented The main argument for the DEA over a parametric technique such as SFA lies to the fact that it works particularly well with small samples Furthermore it is able to handle multiple inputs and outputs stated in different measurement units and does not necessitate knowledge of any functional form of the frontier see Charnes et al 1995 Most empirical papers show that using DEA to estimate the efficiency frontier can yield robust results see Seiford and Thrall 1990 DEA has a number of draw backs first the efficiency scores derived from DEA are sensitive to the selection of inputs and outputs Secondly the number of efficiency firms on the frontier tends to increase with the number of input and output vari ables Thirdly DEA assumes that it has no statistical noise and it is sensitive to extreme observations and measurement errors This section will be arranged as follows Section 331 will discuss the parametric approach As outlined previously three specific analyses are focussed upon which include SFA DFA as well as TFA In Section 332 two nonparametric approaches will be explained which are DEA and FDH 331 Parametric Approach 3311 Stochastic Frontier Approach One of the most widely used parametric technique to measure bank effi ciency is SFA which was proposed by Aigner et al 1997 Battese and Corra 1997 and Meeusen and Van Den Broeck 1997 The assumption of this method is that the decisionmaking unit DMU does not totally Theory of Bank Efficiency and Bank Competition 51 control the deviation from the frontier Random errors which are linked together with the choice of the functional form are allowed by SFA The SFA model is also called the composed error model in which two terms are included and considered the statistical noise term follows a symmetric distribution whereas another term which represents inefficiency follows a onesided distribution To make clearer with regards to the formation of the SFA the following stochastic cost function model is used ln Ci ¼ ln Cðyi wiÞ þ εi ¼ ln Cðyi wiÞ þ vi þ mi In the above function C represents the total cost i represents the specific bank y stands for a vector of output and w is the input price In the banking operation normally three input prices are considered which are price of funds price of labour and price of capital n represents the statistical noise and u stands for the nonnegative disturbance term mi measures the distance of a specific individual firm from the efficient cost frontier and this term represents the technical and allocative efficiency of the firm The deterministic cost frontier is represented by Cyiwi Aigner et al 1977 and Mester 1993 argue that vi follows the normal distribution and mi follows the halfnormal or exponential distribution Berger 1993 argues that the halfnormal or exponential distribution term on inefficiency assumes that most firms nearly reach the full efficiency and are clustered together The possibility of higher inefficiency is relatively low However this is not always the case due to the fact that some specific DMU would experience a higher degree of inefficiency Rather than using the halfnormal distribution for Ui Stevenson 1980 and Berger and DeYoung 1997 suggested that alternative distribution for inefficiency such as a truncated normal distribution is more appropriate Besides the truncated normal distribution the normalgamma distribution is considered by Greene 1990 One of the obvious disadvantages of the truncated normal distribution as well as normalgamma distribution lies in the fact that the inefficiency is difficult to separate from random error Berger and Humphrey 1997 argue that random error of these two distributions is close to a symmetric normal distribution There are mainly two methods being used to estimate the parameters of the frontier model and the error terms which are the maximum likelihood estimation or the corrected ordinary least square OLS see Battese and Coelli 1995 Coelli 1995 Fried et al 1993 In comparison the maximum 52 Efficiency and Competition in Chinese Banking likelihood estimation is preferred to the corrected OLS estimator due to the fact the maximum likelihood estimation produces more accurate results in large samples Coelli 1995 Olesen et al 1980 Jondrow et al 1982 argue that the distribution of the inefficiency term conditional on the estimate of the two error terms will be used to obtain estimates of inefficiency 3312 Distribution Free Approach DFA another parametric method was proposed by Schmidt and Sickles 1984 and Berger 1993 It is a panel estimation method The distributional assumptions on the error component were not imposed by this parametric method The similarity between DFA and SFA lies in the fact that the former specifies a functional form for the efficiency frontier The main difference between these two parametric methods is that the inefficiencies from the residual are separated in a different way DFA assumes that inefficiencies are relatively stable and should persist over time which disentangles them from random errors The random errors in DFA have the following characteristics 1 they are momentary and 2 the random errors tend to cancel one another through averaging DFA estimates the cost or profit function for each period of a panel data set There are two parts included in the residual which are inefficiency terms and random error term Due to the assumption that the error term tends to average out over time the inefficiency of the bank can be estimated by the average of a banks residuals Due to the fact that the distribution of inefficiency or error terms do not have any restrictive assumptions so compared to the SFA the DFA is easier to implement The reason lies in the fact that the maximum likelihood methods are not required by DFA to estimate the cost or profit function There are mainly two methods used to estimate the cost or profit function which is generalized least square GLS reflected by the work of Schmidt and Sickles 1984 The second method is OLS which is represented by the work of Berger 1993 The difference between average residual and average residual of the firm on the frontier gives the estimate of inefficiency There are some disadvantages to DFA Misleading results can be obtained if the period under examination is too long the results will also be biased if the inefficiency component of the error term is not constant The accuracy of the results will also be affected if the random error term is averaged out by an insufficient number of available data years The appropriate number of years enough to address all the above issues is six as argued by DeYoung 1997 Theory of Bank Efficiency and Bank Competition 53 3313 Thick Frontier Approach The last parametric method used to measure efficiency is the TFA which was developed by Berger and Humphrey 1991 1992 Similar to other parametric frontier approaches a functional form is specified for the frontier cost function in TFA The main difference between this method and other parametric methods lies to the fact that a thick frontier is estimated by this approach whereas with regards to other parametric techniques a frontier edge is used to measure the efficiency Furthermore the distributional assumption for crosssectional data is also avoided by this method This method is implemented by estimating the cost function Both the lowest averagecost quartile and the Highest average cost quartile of the firm are estimated Two groups of efficiencies are represented by firms in these two different quartiles More specifically the assumption is that firms in the lowest averagecost quartile have efficiency greater than the average level and they form the thick frontier On the other hand firms in the highest averagecost quartile have efficiency lower than the average level The difference between the two groups of firms with higher than average ef ficiency and lower than average efficiency represent the inefficiency as well as the exogenous difference in the regression Random error and luck are assumed to be represented by the error term within each of the frontiers The difference in the parameters between the upper cost frontier and lower cost frontier is used by TFA to measure cost efficiency whereas with regards to the exogenous factors they are estimated by calculating the difference between the lowest cost unction and highest cost function One disadvantage of TFA lies in the fact that the overall level of efficiency rather than the efficiency of the individual DMU is estimated TFA may not generate a robust estimate of the overall level of efficiency because the assumption made by this approach does not usually hold 332 NonParametric Approach 3321 Data Envelopment Analysis The idea of the nonparametric efficiency approach was first originated by Farrell 1957 in his seminal paper One of the disadvantages of his work lies to the fact that singleoutput case is considered whereas it cannot applied to large dataset with multiple inputs and outputs His work has been signifi cantly extended by Charnes et al 1978 in which a mathematical pro gramming algorithm was proposed namely DEA This method was used to measure the efficiency of a set of DMUs from which a significant 54 Efficiency and Competition in Chinese Banking improvement of their work compared to Farrell 1957 lies in the fact that this method can be applied to multiple inputs and outputs and until now many research studies use this method to measure efficiency in particular in the banking sector Two different DEA models are available with regards to the input or output orientation Inputoriented DEA aims to minimize the amount of input to produce a certain level of output On the other hand output oriented DEA focusses on keeping the same amount of input to maxi mize the output production With regards to the choice of orientation it depends over which aspect the managers have more control More spe cifically Coelli 1998 argued that inputoriented is more appropriate if producers can freely adjust the input usage and are required to meet the market demand whereas if only a certain amount of resources is available for the firm to maximize the output production the outputoriented model would be more appropriate Coelli and Perelman 1999 suggest that there is very minor impact of the choice of input or outputoriented model on the estimation of the efficiency score 3322 Free Disposal Hull One nonparametric technique used to measure efficiency is called FDH analysis The FDH model was firstly proposed by Deprins 1984 and then the model was extended by Tulkens 1993 The FDH analysis relaxes the convexity assumption of basic DEA models Both DEA and FDH are consistent estimators if the true production set is convex However due to the fact that FDH requires fewer assumptions thus a lower rate of conver gency is shown by FDH compared to DEA On the other hand if the true production set is not convex FDH has the ability to estimate the efficiency in a more precise way in comparison to DEA Because Both DEA and FDH are nonparametric methods compared to the parametric method such as SFA no particular functional form for the boundary is assumed and the measurement error is ignored by the nonparametric method The boundary of a reconstructed production possibility subset is regarded as the bestpractice technology which can be represented by enveloping a set of observations Berger and Humphrey 1997 argue that the set of production possibilities comprises only the DEA vertices and the FDH points interior to these vertices One of the assumptions made by the DEA is that linear substitution between observed input combinations on an isoquant is possible while the observations in Theory of Bank Efficiency and Bank Competition 55 piecewise linear form generate the isoquant In comparison one of the assumptions made by the FDH is that linear substitution between observed input combinations is impossible Therefore the isoquant is more like a step function formed by the intersection of lines drawn from observed Leontieftype input combinations Thus compared to DEA frontier the data are enveloped more tightly in the FDH which has a staircase shape The relationship between DEA and FDH frontiers can be described as two possible situations first the FDH frontier could be congruent with the DEA frontier The second situation is that the FDH frontier could be interior to the DEA frontier Tulkens 1993 argues that larger estimates of average efficiency are typically generated by FDH compared to DEA 34 THEORY OF BANK COMPETITION Similar to other industries the types of competition in the banking sector can be mainly classified into the following categories perfect competition monopolistic competition oligopoly and monopoly The perfect competitive environment in the banking sector has the following characteristics see Mankiw and Taylor 2006 1 there are a large number of banks in the market 2 banks offer a homogenous product with regards to the cost and attribute of the product 3 the cost for new banks to enter the market is very low In a perfectly competitive envi ronment banks are pricetakers rather than pricemakers The price of a product offered by banks will be determined by the industry supply or demand whereas they have no ability to influence the volumes of demand and supply in the market Another type of competition is called monop olistic competition Under this competitive environment lots of banks are in the market but unlike perfectly competitive markets these banks offer differentiated products to customers The cost of entry to and exit from the banking market is low Banks have a degree of control over the price of the product offered In other words they are price makers rather than price takers to some extent The third type of competition is called oligopoly Under this competitive environment there are a small number of banks in the markets whereas all of them provide either homogenous or hetero geneous products in the market the entry or exit to the market is quite expensive The banks operating in the oligopoly market have power to set the price in the market The degree of price control by banks in oligopoly is higher than in monopolistic competition One special characteristic of 56 Efficiency and Competition in Chinese Banking oligopoly over monopolistic competition and perfect competition is the interdependence among banking firms The market in the condition of oligopoly is made up of a few large banks because the size of the banks is very large their actions will affect the market conditions due to the fact that in the oligopoly market few banks are operating and each bank is large enough that its actions will affect market conditions Thus other banks will be aware of one banks action and respond appropriately to keep their competitive position in the market Finally the market in the con dition of monopoly has only one bank because this bank is the only firm that provides financial services to the market there is no competition and this bank has absolute power in setting the price in the market Further more other potential banks are unable to enter the market They also have the ability to charge a different price to different markets ie they may charge a lower price in a very elastic market to increase the quantity sold whereas a higher price would be charged to the consumers in a relatively inelastic market to maximize the profit 341 StructureeConductePerformance SCP Theory The SCP paradigm states that market structure would determine firm conduct which would determine performance Market structure can be measured by a number of factors such as the number of competitors in an industry the heterogeneity of product and the cost of entry and exit Conduct refers to a number of specific actions taken by a firm which include price taking product differentiation tacit collusion and exploitation of market power The performance of the firm can be measured from a number of indicators such as productive efficiency allocative efficiency and profitability The range of options and constraints facing a firm is defined by the attributes of the industry within which a firm operates In some industries with higher competition very few options are available to the firms and the firms have lots of constraints Firms in these industries generate maximized social welfare and in the long run the returns earned by the firms can only cover the cost of capital In summary the industry structure determines the firms conduct and longrun firm performance On the other hand the firms operating in a lower competitive industry environment have a greater range of conduct options and the number of constraints faced by the firms is limited Firms can make use of the available options to obtain the competitive advantage For instance the firms in these industries can use market power to set prices that generate significant economic value Theory of Bank Efficiency and Bank Competition 57 Structure Number of competitors Heterogeneity of product Cost of entry and exit Conduct Price taking Product differentiation Tacit collusion Exploitation of market power Performance Productive efficiency Allocative efficiency Profitability Figure 35 The structureconductperformance SCP model Own illustration However the sustainability of their advantages is determined by one of the attributes of industrystructure barriers to entry If there are no barriers to entry the competitive advantages of the firms in the industry will disappear when new competitors enter the market Therefore industry structure has an important effect on firm conduct and performance even though firms in these industries can sometimes have competitive advantages Barney and Clark 2007 The SCP model is summarized in Fig 35 342 Lerner Index The Lerner index was developed in 1934 by the American economist Abba Lerner The Lerner index is defined as the difference between price and marginal cost divided by price and it can be specified as LIit Pit MCitPit in which P is the price of banking outputs MC represents the margin cost whereas i and t represent the specific bank at specific year The value of Lerner index ranges from a minimum of zero to a maximum of one When P MC the Lerner index is zero which indicates that the firm has no pricing power As the value of Lerner index increases the difference between price and marginal cost becomes bigger which indicates that banks have higher market power Ariss 2010 In other words LI 0 indicates that there is perfect competition whereas LI 1 means the market is in a condition of monopoly Casu and Girardone 2009 argue that the Lerner index of monopoly power measures the degree of market power very well and is a good and widely used indicator in measuring competition in banking literature It represents the extent to which banks have the market power to set their price above the marginal cost Berger et al 2009 A similar opinion is held by DemirgucKunt and Peria 2010 who suggest that computing direct measures of market power is an alternative way to examine the competitive condition in the banking industry and the Lerner index defined as the difference between price and marginal cost relative to price is frequently used in the banking sector Lerner 1934 argues that market power is determined by demand elasticity and the Lerner index provides a number which links with the demandeprice elasticity inverse relationship Rojas 2011 suggests that the Lerner index is popular due to the fact that it shows the firms market power location between perfect competition and mo nopoly it also reflects the role that demand elasticity plays in determining a firms markup Compared to the traditional concentration ratio the Lerner index provides a more accurate measurement of bank market power The advantage of the Lerner index over the PanzareRosse Hstatistic which will be discussed later lies in the fact that the former is not a longrun equilibrium measure of competition and can be calculated at each point in time DemirgucKunt and Peria 2010 There are also some arguments relating to the disadvantages of the Lerner index Fernandez de Guevara et al 2005 argue that there are several problems with regards to the estimation of the Lerner index First the value of the Lerner index changes according to different revenues used by the study Frequently only interest revenue and costs are considered whereas other noninterest revenues and expenses are omitted The consideration of traditional loanedeposit services as the revenue ignores the banking activities of providing other services which have grown substan tially during recent years This will lead to an inaccurate result regarding the competitive condition in the banking sector Second the cost of risk which is very important in the profit and loss account of a banking system is not considered in general practice The ignorance of the cost of risk can be attributed to reasons such as insufficient data and calculation difficulties If the cost of risk in not included in the estimation of cost function it will lead to a wrong interpretation of the Lerner index because the margin is over estimated Third Bikker et al 2007 argue that the weakness of the Lerner index is attributed to the fact that the prices and costs required to calculate the index are not clearly identified by the available bank balancesheet data so that prices and costs can be proxied by many debatable choices 343 Panzar and Rosse 1987 HStatistic An empirical test is developed by John C Panzar and James N Rosse to discriminate between monopoly oligopolistic monopolistically competi tive and perfectly competitive markets They have provided a concise in dicator the socalled Hstatistic which is based on the static properties of reducedform revenue equation The Hstatistic can be interpreted as a continuous and increasing measure of the overall level of competition in a Theory of Bank Efficiency and Bank Competition 59 specific market This method makes the assumption that different pricing strategies will be employed in reaction to changes in input costs In other words market power is measured by the extent to which changes in factor prices unit price of funds capital and labour are reflected in revenue The Hstatistic ranges from minus infinity to unity see Table 31 A negative H arises when the competitive structure is a monopoly a perfect colluding oligopoly or a conjectural variation of shortrun oligopoly In all the cases an increase in input prices will translate into higher marginal costs a reduction of equilibrium output and subsequently a fall in total revenue Vesala 1995 If H lies between zero and unity the market structure is characterized by monopolistic competition Under monopolistic competition potential entry leads to contestable market equilibrium and revenue increases less than proportionally to the changes in input prices as the demand facing individual banks for banking products is inelastic Triole 1987 Under perfect competition the Hstatistic equals unity In this particular situation a proportional increase in factor input prices raises both marginal and average costs and induces an equiproportional change in revenues without distorting the optimal output of any firm The Hstatistic would also equal to one if a contestable market exists Baumol et al 1982 who put forward the contestability theory argue that under very restrictive circumstances such as free entry to and exit from the market and higher price elastic demand for industrys output the competition in a highly concentrated market still exists Due to the existence of these features larger firms in the market need to take competitive measures to price their outputs One assumption of PR Hstatistic is that the test can only be applied to firms which produce a single output Therefore banks are regarded as providers of traditional loandeposit services as well as nontraditional Table 31 Discriminatory power of Hstatistic Estimated value of H Competitive environment Market equilibrium H 0 Monopolistic market behaviour Conjectural variation shortrun oligopoly H 0 Equilibrium H 0 Disequilibrium 0 H 1 Monopolistic competition Natural monopoly in a perfectly contestable market Perfect competition H 1 activities using factor inputs such as labour funds and capital In other words all the activities should be considered to estimate the competitive condition of the specific market Nevertheless supposing the above assumption is omitted the competitive conditions of separate segments cannot be analysed due to the fact that there are no detailed data available for the estimation of the reducedform revenue equations The advantages of using PR Hstatistic to measure the competition are 1 It works well with firmspecific data on revenues and factor prices 2 it does not require information about equilibrium output prices and quantities for the firm andor industry 3 this method can generate concise competitive condition if the sample size is small Matthews et al 2007 Due to the fact that Hstatistic was developed on the basis of a static model there are no predictions on the Hvalue which is one of the weaknesses of this test In addition the overall market equilibrium required by the test cannot be fulfilled because of market entry and exit which leads to further limits on the interpretation of such analysis Due to the fact that the Hstatistic is a static approach one characteristic of the implementation is that the test needs to be undertaken on obser vations that are in longrun equilibrium In other words in the equilibrium test the dependent variable will be replaced by the profitability measure such as return on assets ROA or return on equity ROE rather than the revenues In equilibrium the resulting Hstatistic will be significantly equal to zero However the significant and negative Hstatistic indicates that the test is not in longrun equilibrium The purpose of the test is to justify on the ground that competitive markets will equalize riskadjusted return across firms such that in equilibrium rate of return should not be correlated with factor input prices 344 Boone Indicator Nowadays a growing number of studies in the literature use the Boone 2008 indicator to measure bankingsector competition This indicator is based on the efficiency hypothesis which was proposed by Demsetz 1973 The efficiency hypothesis argues that the growth of efficient firms leads to improvement in industry performance In other words the hypothesis emphasizes the relationship between efficiency and performance This indicator is based on the assumption that efficiency leads to improvement in performance This hypothesis mainly argues that banks with higher effi ciency have higher levels of performance in the form of higher profit and then have higher ability to attract a larger number of market shares The Theory of Bank Efficiency and Bank Competition 61 theoryassumption of the Boone indicator is in line with the studies by Stiorh 2000 and Stiroh and Straham 2003 who suggest that a large proportion of assets will be transformed from banks with lower profit to the ones with higher profit due to the increase in competition The basic idea under the Boone indicator argues that stronger competition leads to the reallocation of assets A number of reasons explain the increase in competition such as increase in the number of banks providing banking services or a higher degree of aggressive interaction between banks under the condition that individual banks have a certain level of efficiency The industry with a higher degree of competition will make the profit of the higherefficiency bank increase at the expense of banks with lower levels of efficiency Schaeck and Cihak 2010 argue that there are a number of advantages of the Boone indicator compared to other competition indicators such as Panzar and Rosse 1987 HStatistic and the Lerner index More specifically compared to the PR Hstatistic which requires one to identify a situation from the statistic the Boone indicator is monotonically related to competition Xu et al 2013 argue that the Boone indicator will provide robust results under binding interestrate regulations Finally the Boone indicator has an additional advantage for studies on developing countries This indicator only requires information on profit and cost and the computation of the Boone indicator does not require information on prices Leon 2014 A number of studies use the Boone indicator to measure competition in banking markets van Leuvensteijn et al 2007 Maslovych 2009 Schaeck and Cihak 2010 among others However Leon 2015 argues that the Boone indicator is still a young tool and therefore has not been thoroughly scrutinized by the literature 35 SUMMARY AND CONCLUSION Efficiency as one important measure of performance has received greater attention from government companies as well as academic researchers Both the government and company management try to use different ways to improve efficiency and further promote performance Efficiency can be measured from different perspectives More specifically technical efficiency mainly focusses on volume of production investment and output In other words technical efficiency mainly concentrates on using a certain amount of input to maximize output production or producing a certain volume of 62 Efficiency and Competition in Chinese Banking output using fewer numbers of inputs In firm operation in daily life not only does the volume of investment and output concern the managers but more importantly is how efficiently the firm controls the cost In other words how to use a certain amount of inputs to minimize cost is the goal which needs to be achieved within the firm It is well known that con trolling cost as efficiently as possible is not the final step for the company the company must aim to maximize the revenue and more importantly the profit Thus how efficiently a company uses input resources to generate profit attracts the greatest attention from the public This chapter explains the theories from technical cost revenue and profit perspectives thus systematically discussing the theories of technical cost revenueand profit efficiencies Not only the theories of efficiency being discussed this chapter also explains two mainly streams of methods used to measure efficiency which are parametric method as well as nonparametric method To be more specific three parametric approaches are discussed in the chapter which include SFA DFA as well as TFA on the other hand there are mainly two nonparametric approaches being discussed in the chapter which are DEA and FDH Not only are specific methods explained in this chapter but also the different methods are compared After the discussion about the theory and measures of efficiency this chapter systematically explains the theories of bank competition Starting from the traditional structureeconducteperformance hypothesis which mainly uses the concentration ratio or HerfindaleHirschman index as a competition indicator After the discussion of the concentration ratio the chapter further discusses the Lerner index competition indicator The ad vantages of this indicator lie in the fact that it not only measures competitive conditions but more importantly it reflects the market power of banks Then the chapter discusses the PanzarRosse H statistic Finally the chapter discusses the Boone competition indicator One characteristic of Boone compared to the other competition indicators lies in the fact that this in dicator can measure the competition of different banking markets ie loan market deposit market and noninterest income market This indicator is the newest competition indicator used in academic research today Through the systematic explanation and discussion of the theories of efficiency and competition not only can we have a clear and better understanding of efficiency and competition but provide a basis for our analysis and discussion in future chapters Theory of Bank Efficiency and Bank Competition 63 REFERENCES Aigner D Lovell CAK Schmidt P 1977 Formation and estimation of stochastic frontier production function models Journal of Econometrics 6 1 21e37 Ariss RT 2010 On the implications of market power in banking evidence from developing countries Journal of Banking and Finance 34 765e775 Barney JB Clark DN 2007 ResourceBased Theory Creating and Sustaining Competitive Advantage Oxford University Press Oxford Battese GE Coelli TJ 1995 A model for technical inefficiency effects in a stochastic frontiers production function for panel data Empirical Economics 20 2 325e332 Battese GE Corra GS 1977 Estimation of a production frontier model with appli cations to the pastoral zone of eastern Australia Australian Journal of Agricultural Economics 21 3 169e179 Baumol WJ Panzar JC Willig RD 1982 Contestable Markets and the Theory of Industry Structure Harcourt Brace Jovanovich San Diego CA Berger AN DeYoung R 1977 Problem loans and cost efficiency in commercial banks Journal of Banking and Finance 21 6 849e870 Berger AN Humphrey DB 1991 The dominance of inefficiencies over scale and product mix economies in banking Journal of Monetary Economics 28 1 117e148 Berger AN Humphrey DB 1992 Measurement and efficiency issues in commercial banking In Griliches Z Ed Output Measurement in the Service Sectors University of Chicago Press Chicago and London pp 245e279 Berger AN Humphrey DB 1997 Efficiency of financial institutions international survey and directions for future research European Journal of Operational Research 98 175e212 Berger AN Klapper LF TurkAriss R 2009 Bank competition and financial stability Journal of Financial Services Research 35 99e118 Berger AN Mester LJ 1997 Inside the black box what explains differences in the efficiencies of financial institutions Journal of Banking and Finance 21 895e947 Berger AN 1993 Distribution free estimates of efficiency of the US banking industry and tests of the standard distributional assumptions Journal of Productivity Analysis 4 3 261e292 Bikker JA Spierdijk L Finnie P 2007 The Impact of Market Structure Contestability and Institutional Environment on Bank Competition DNB working paper No 156 Boone J 2008 A new way to measure competition Economic Journal 118 1245e1261 Casu B Girardone C 2009 Testing the relationship between competition and efficiency in banking a panel data analysis Economics Letters 105 134e137 Charnes A Cooper WW Rhodes E 1978 Measuring the efficiency of decision making units European Journal of Operational Research 2 6 429e444 Charnes A Cooper WW Lewin AY Seiford LM 1995 Data Envelopment Anal ysis Theory Methodology and Applications Kluwer Boston Coelli T Perelman S 1999 A comparison of parametric and nonparametric distance functions with application to European Railways European Journal of Operational Research 117 2 326e339 Coelli TJ 1995 Estimators and hypothesis tests for a stochastic frontier function a Monte Carlo analysis Journal of Productivity Analysis 6 3 247e268 Coelli T 1996 A Guide to DEAP Version 21 A Data Envelopment Analysis Computer Program CEPA working paper 199608 Coelli T 1998 A multistage methodology for the solution of oriented DEA models Operations Research Letters 23 3e5 143e149 Debreu G 1951 The coefficient of resource utilization Econometrica 19 273e292 64 Efficiency and Competition in Chinese Banking DemirgucKunt A Peria MSM 2010 A Framework for Analysing Competition in the Banking Sector An Application to the Case of Jordan Policy Research working paper No 5499 1e24 Demsetz H 1973 Industry structure market rivalry and public policy Journal of Law and Economics 16 1e9 Deprins D Simar L Tulkens H 1984 Measuring labour efficiency in post offices In Marchand M Pestieau P Tulkens H Eds The Performance of Public Enterprises Concepts and Measurement NorthHolland Amsterdam DeYoung R 1997 A diagnostic test for the distribution fee efficiency estimator an example using US commercial bank data European Journal of Operational Research 98 2 243e249 Farrell MJ 1957 The measurement of productive efficiency Journal of the Royal Statistical Society 120 253e290 Fernandez de Guevara J Maudos J Perez F 2005 Market power in European banking Journal of Financial Services Research 27 109e137 Fried HO Lovell CAK Schmidt SS 1993 The Measurement of Productive Efficiency Techniques and Application Oxford University Press New York Greene WH 1990 A gammadistributional stochastic frontier model Journal of Econometrics 46 1e2 141e163 Jondrow J Knox Lovell CA Materor IS Schmidt P 1982 On the estimation of technical inefficiency in the stochastic frontier production function model Journal of Econometrics 19 2e3 233e238 Koopmans TC 1951 An analysis of production as an efficient combination of activities In Koopmans TC Ed Activity Analysis of Production and Allocation Wiley New York Kumar S Gulati R 2013 Regulation and Efficiency of Indian Banks Springer India Leon F 2014 Measuring competition in banking a critical review of methods Series etudes et documents du CERDI 12 1e14 Leon F 2015 Measuring Competition in Banking A Critical Review of Methods Working paper Halshs No 01015794 Lerner AP 1934 The concept of monopoly and the measurement of monopoly power The Review of Economic Studies 1 157e175 Mankiw NG Taylor M 2006 Economics Cengage Learning EMEA London Maslovych M 2009 The Boone Indicator as a Measure of Competition in the Banking Sector The Case of Ukraine Kier School of Economics MA thesis Matthews K Murinde V Zhao T 2007 Competitive conditions among the major British Banks Journal of Banking and Finance 31 2025e2042 Meeusen W Van Den Broeck J 1977 Efficiency estimation from CobbDouglas pro duction functions with composed error International Economic Review 18 2 435e444 Mester LJ 1993 Efficiency in the savings and loan industry Journal of Banking and Finance 17 2e3 267e286 Olesen JA Schmidt P Waldman DM 1980 A monte carlo study of estimators of stochastic frontier production function Journal of Econometrics 13 1 67e82 Panzar JC Rosse JN 1987 Testing for monopoly equilibrium Journal of Industrial Economics 35 443e456 Rojas C 2011 Market power and the lerner index a classroom experiment Journal of Industrial Organization Education 5 1e19 Schaeck K Cihak M 2010 Competition Efficiency and Soundness in Banking An Industrial Organization Perspective European Banking Center discussion paper No 2010e20 Schimidt P Sickles RC 1984 Production frontiers and panel data Journal of Business and Economic Statistics 2 4 367e374 Theory of Bank Efficiency and Bank Competition 65 Seiford LM Thrall RM 1990 Recent development in DEA the mathematical programming approach to frontier analysis Journal of Econometrics 46 7e38 Stevenson RE 1980 Likelihood functions for generalised stochastic frontier estimation Journal of Econometrics 13 1 57e66 Stiroh KJ Strahan PE 2003 Competitive dynamics of deregulation evidence from US banking Journal of Money Credit and Banking 35 801e828 Stiroh KJ 2000 Compositional Dynamics and the Performance of the US Banking Industry Federal Reserve Bank of New York staff report No 98 Triole J 1987 The Theory of Industrial Organization MIT press Cambridge Tulkens H 1993 On FDH efficiency analsysis some methodological issues and applica tions to retail banking courts and urban transit Journal of Productivity Analysis 4 183e210 Van Leuvensteijn M Bikker JA van Rixtel AARJM Kok Sorensen C 2007 A New Approach to Measuring Competition in the Loan Markets of the Euro Area European Central Bank working paper No 768 Vesala J 1995 Testing for Competition in Banking Behavioural Evidence From Finland Bank of Finland Studies p E1 Xu B van Rixtel A van Leuvensteijn M 2013 Measuring Bank Competition in China A Comparison of New Versus Conventional Approaches Applied to Loan Markets BIS working paper 422 pp 1e68 66 Efficiency and Competition in Chinese Banking Universidade do Estado do Rio de Janeiro Centro de Ciências Sociais Faculdade de Ciências Econômicas Giovanni Barboza Xavier Avaliação de desempenho no setor bancário uma comparação entre bancos digitais e tradicionais via Análise de Fronteira de Eficiência e Índice de Malmquist Rio de Janeiro 2023 Giovanni Barboza Xavier Avaliação de desempenho no setor bancário uma comparação entre bancos digitais e tradicionais via Análise de Fronteira de Eficiência e Índice de Malmquist Dissertação apresentada como requisito parcial para obtenção do título de Mestre ao Programa de PósGraduação em Ciências Econômicas da Universidade do Estado do Rio de Janeiro Área de concentração II Economia Aplicada Orientador Prof Dr Alexandre Marinho Coorientador Prof Dr Luiz Fernando Rodrigues de Paula Rio de Janeiro 2023 CATALOGAÇÃO NA FONTE UERJREDE SIRIUSBIBLIOTECA CCSB Bibliotecário Fabiano Salgueiro CRB76974 Autorizo apenas para fins acadêmicos e científicos a reprodução total ou parcial desta dissertação desde que citada a fonte Assinatura Data X3 Xavier Giovanni Barboza Avaliação de desempenho no setor bancário uma comparação entre bancos digitais e tradicionais via Análise de Fronteira de Eficiência e Índice de Malmquist Giovanni Barboza Xavier 2023 73 f Orientador Prof Dr Alexandre Marinho Coorientador Prof Dr Luiz Fernando Rodrigues de Paula Dissertação mestrado Universidade do Estado do Rio de Janeiro Faculdade de Ciências Econômicas Bibliografia f 6873 1 Análise de envoltório de dados Teses 2 Bancos Brasil Teses I Marinho Alexandre II Paula Luiz Fernando Rodrigues de III Universidade do Estado do Rio de Janeiro Faculdade de Ciências Econômicas IV Título CDU 33681005 Giovanni Barboza Xavier Avaliação de desempenho no setor bancário uma comparação entre bancos digitais e tradicionais via Análise de Fronteira de Eficiência e Índice de Malmquist Dissertação apresentada como requisito parcial para obtenção do título de Mestre ao Programa de PósGraduação em Ciências Econômicas da Universidade do Estado do Rio de Janeiro Área de concentração II Economia Aplicada Aprovada em 10 de março de 2023 Banca Examinadora Prof Dr Alexandre Marinho Orientador FCEUERJ Prof Dr Luiz Fernando Rodrigues de Paula Coorientador IEUFRJ e FCEUERJ Prof Dr Pedro James Frias Hemsley IEUFRJ Rio de Janeiro 2023 AGRADECIMENTOS Aos meus pais Janice e Carlos por sempre estarem ao meu lado me apoiando nos momentos difíceis Ao meu orientador Alexandre Marinho e ao meu coorientador Luiz Fernando de Paula pelos ensinamentos apoio e colaboração que me permitiram estar concluindo este trabalho Ao meu amigo Renan pela parceria durante todo o curso e aos meus amigos Natalia Henry e Catharine por todo companheirismo ao longo do período em que me dediquei a este trabalho Aos meus professores e colegas do Programa de PósGraduação em Ciências Econômicas PPGCEUERJ com quem convivi ao longo desses anos de curso pelo aprendizado e pelas experiências que certamente enriqueceram a minha formação acadêmica Agradeço à Coordenação de Aperfeiçoamento de Pessoal de Nível Superior Brasil CAPES Código de Financiamento 001 pelo apoio financeiro para o presente trabalho RESUMO XAVIER Giovanni Barboza Avaliação de desempenho no setor bancário uma comparação entre bancos digitais e tradicionais via Análise de Fronteira de Eficiência e Índice de Malmquist 2023 73 f Dissertação Mestrado em Ciências Econômicas Faculdade de Ciências Econômicas Universidade do Estado do Rio de Janeiro Rio de Janeiro 2023 O setor bancário tem fundamental importância para o desenvolvimento econômico de um país Através da intermediação financeira função primária dos bancos os recursos chegam aos agentes deficitários permitindo que os investimentos desejados sejam realizados Com os avanços tecnológicos e a sua incorporação na indústria financeira uma nova modalidade de bancos surgiu os bancos digitais Este trabalho tem como objetivo realizar uma análise comparativa da eficiência dos bancos digitais e dos bancos de varejo tradicionais no mercado de crédito brasileiro Para tanto foi utilizada a técnica nãoparamétrica DEA corrigida pelo método de Bootstrap com orientação para inputs para mensurar a eficiência técnica dos bancos a partir da abordagem da intermediação financeira A amostra ficou constituída de 73 bancos comerciais em que por meio de testes estatísticos adequados a combinação de inputs e outputs definida se mostrou convexa e com retornos variáveis de escala o que justificou a aplicação do modelo DEABCC Os resultados sugerem de modo geral que os bancos comerciais brasileiros apresentaram nível elevado de ineficiência nas atividades de intermediação financeira Foi observado que o grupo dos bancos digitais apresenta desempenho inferior quando comparado com o grupo de todos os bancos de varejo tradicionais Entretanto foi visto que quando comparados apenas com os bancos de varejo de mesmo porte a performance observada foi semelhante Por fim com a aplicação do Índice de Malmquist para o triênio de 20192021 foi percebida um aumento de produtividade dos bancos comerciais brasileiros atribuída a uma melhora de gestão dos insumos pelos bancos digitais e tradicionais Por outro lado não foi possível observar indícios de progresso tecnológico durante o período Palavraschave Análise Envoltória de Dados Bancos Digitais Intermediação Financeira ABSTRACT XAVIER Giovanni Barboza Banking benchmarking a comparison between digital and traditional banks with Efficiency Frontier Analysis and Malmquist Index 2023 73 f Dissertação Mestrado em Ciências Econômicas Faculdade de Ciências Econômicas Universidade do Estado do Rio de Janeiro Rio de Janeiro 2023 The banking sector is of fundamental importance for the economic development of a country Through financial intermediation the primary function of banks resources reach the deficit agents allowing the desired investments to be carried out With technological advances and their incorporation into the financial industry a new type of banking has emerged digital banks This work aims to carry out a comparative analysis of the efficiency of digital banks and traditional retail banks in the Brazilian credit market For this purpose the DEA nonparametric technique corrected by the Bootstrap method with input orientation was used to measure the technical efficiency of banks based on the financial intermediation approach The sample consisted of 73 commercial banks in which through adequate statistical tests the defined combination of inputs and outputs proved to be convex and with variable returns to scale which justified the application of the DEABCC model The results suggest in general that Brazilian commercial banks showed a high level of inefficiency in financial intermediation activities It was observed that the group of digital banks performs poorly when compared to the group of all traditional retail banks However it was seen that when compared only with retail banks of the same size the performance observed was similar Finally with the application of the Malmquist Index for the triennium 20192021 an increase in the productivity of Brazilian commercial banks was noticed attributed to an improvement in the management of inputs by digital and traditional banks On the other hand it was not possible to observe signs of technological progress during the period Keywords Data Envelopment Analysis Digital Banks Financial Intermediation LISTA DE ILUSTRAÇÕES Figura 1 Produtividade e eficiência 22 Figura 2 Modelo de Farrel para eficiência técnica e alocativa 24 Figura 3 Comparação das fronteiras de eficiência dos modelos CCR e BCC 30 Quadro 1 Principais características da DEA e da SFA 32 LISTA DE TABELAS Tabela 1 Abordagens para avaliar a eficiência bancária presentes na literatura 12 Tabela 2 Abordagens utilizadas na literatura nacional 19 Tabela 3 Inputs e output do modelo proposto 42 Tabela 4 Estatísticas descritivas dos inputs e output 20192021 44 Tabela 5 Testes de convexidade da amostra 45 Tabela 6 Testes de retornos de escala da amostra 46 Tabela 7 Escores de eficiência dos bancos no 3º trimestre de 2021 47 Tabela 8 Sumário estatístico das eficiências no 3º trimestre de 2021 49 Tabela 9 Eficiência dos bancos digitais no 3º trimestre de 2021 50 Tabela 10 Segmentação das instituições financeiras conforme o BCB 51 Tabela 11 Eficiência dos bancos de varejo no 3º trimestre de 2021 51 Tabela 12 Sumário da eficiência dos bancos no triênio 20192021 52 Tabela 13 Eficiência dos bancos digitais no triênio 20192021 53 Tabela 14 Eficiência dos bancos de varejo no triênio 20192021 53 Tabela 15 Teste 1 de equivalência de médias de Kneip et al 2016 54 Tabela 16 Teste 2 de equivalência de médias de Kneip et al 2016 55 Tabela 17 Eficiência de escala dos bancos digitais no 3º trimestre de 2021 57 Tabela 18 Eficiência de escala dos bancos de varejo no 3º trimestre de 2021 57 Tabela 19 Índice de produtividade de Malmquist para os bancos digitais 59 Tabela 20 Índice de alterações de eficiência para os bancos digitais 60 Tabela 21 Índice de alterações de tecnologia para os bancos digitais 61 Tabela 22 Índice de produtividade de Malmquist para os bancos de varejo 61 Tabela 23 Índice de alteração de eficiência para os bancos de varejo 62 Tabela 24 Índice de alterações de tecnologia para os bancos de varejo 63 LISTA DE ABREVIATURAS E SIGLAS AA AE AT BB BRB BCB BCC CCR CEF Cosif CRS DEA DMU FDH FGV IA IGP IM PA SFA TFA VAA VRS Asset Approach Alteração de Eficiência Alteração Tecnológica Banco do Brasil Banco Regional de Brasília Banco Central do Brasil Banker Charnes e Cooper Charnes Cooper e Rhodes Caixa Econômica Federal Padrão Contábil das Instituições Reguladas pelo Banco Central do Brasil Constant Returns to Scale Data Envelopment Analysis Decision Making Unit Free Disposal Hull Fundação Getúlio Vargas Intermediation Approach Índice Geral de Preços Índice de Malmquist Production Approach Stochastic Frontier Analysis Thick Frontier Analysis Value Added Approach Variable Returns to Scale SUMÁRIO INTRODUÇÃO 1 Objetivos 3 Estrutura da dissertação 4 1 REVISÃO DA LITERATURA SOBRE EFICIÊNCIA BANCÁRIA 6 11 Literatura internacional sobre eficiência no setor bancário 6 12 As principais abordagens de análise 11 13 A literatura nacional sobre eficiência bancária 14 2 METODOLOGIA 21 21 Produtividade e eficiência 21 22 Eficiência técnica e eficiência alocativa 23 23 A Análise Envoltória de Dados DEA 25 231 Modelo CCR 26 232 Modelo BCC 29 233 Comparação entre os modelos CCR e BCC 30 24 Características da técnica DEA e sua comparação com a SFA 31 25 O viés da DEA e a técnica de Bootstrap 33 26 A convexidade na DEA 35 27 Testes para os retornos de escala 37 28 Comparação de médias 38 29 Índice de Malmquist 39 3 RESULTADOS 41 31 Modelo proposto 41 32 A amostra 43 33 Convexidade e retornos de escala 45 34 Desempenho do setor bancário brasileiro em 2021 47 34 Evolução da eficiência dos bancos digitais e bancos de varejo entre 20192021 52 35 Comparação entre bancos digitais e bancos de varejo tradicionais 54 36 Análise da eficiência de escala 56 37 Análise temporal do setor bancário brasileiro 59 CONCLUSÕES 64 REFERÊNCIAS 68 1 INTRODUÇÃO O setor financeiro e principalmente o setor bancário tem fundamental importância para o desenvolvimento econômico de um país Este setor tem como função principal exercer a atividade de intermediação de recursos entre poupadores e tomadores fazendo com que valores poupados dos agentes superavitários cheguem aos agentes que demandam destes recursos para desempenhar suas atividades Instituições financeiras sólidas e eficientes na intermediação financeira permitem que os investimentos desejados sejam realizados o que é condição essencial para o desenvolvimento econômico de um país Cava et al 2016 explicam que o setor bancário brasileiro desempenha um papel importante na economia em que os mercados de capitais e de seguros ainda se encontram subdesenvolvidos o que destaca a participação dos bancos nos financiamentos de investimentos que contribuem para o desenvolvimento do país Sobre este aspecto Staub et al 2010 destacaram que o desenvolvimento do sistema bancário e aumento da sua eficiência estão relacionados com maiores crescimentos econômicos Conforme Périco et al 2016 ao longo do tempo praticamente em todos os lugares do mundo as mudanças vistas nas funções dos bancos são notáveis com o surgimento de novos produtos e com o aumento da disponibilidade dos serviços para os clientes entretanto o papel de intermediação financeira ainda prevalece Segundo os autores todos os outros recursos incorporados nas funções de um banco ainda são resultados de sua função primária a de intermediação financeira Boot et al 2020 explicam que a última década apresentou enormes desafios para as instituições financeiras incluindo crise financeira reforma regulatória baixa lucratividade e deterioração da confiança do público mas contra este cenário as mudanças tecnológicas na indústria financeira estão aumentando impulsionadas agora pela recente crise da COVID19 que tem aumentado a demanda por serviços digitais acelerando este processo Com os avanços tecnológicos e a sua incorporação na indústria financeira uma nova modalidade de banco emergiu os bancos digitais em que através da tecnologia determinadas funções são oferecidas ao público por meio de plataformas digitais sem o contato físico com o cliente De tal modo os serviços realizados em espaços específicos passam a ser realizados em ambientes virtuais onde tais bancos operam com poucas ou nenhumas agências e em sua maioria oferecem serviços atrativos para os clientes com isenção de taxas e tarifas bancárias 2 De acordo com Philippon 2019 Fintechs um termo para empresas que empregam tecnologia nos serviços financeiros abrangem inovações digitais e inovações nos modelos de negócios com tecnologia no setor financeiro e podem romper as estruturas recentes das empresas e a forma como estas criam e entregam produtos e também democratizar o acesso a serviços financeiros Segundo a OECD 2020 a disrupção digital tem potencial para prover uma maior eficiência no setor bancário por meio da inovação pois esta tendência tem como principal característica a desmaterialização dos processos trazendo benefícios como a redução de recursos e despesas o que pode promover a redução no valor de serviços para os clientes Ainda conforme a OECD 2020 este processo pode ser capaz de gerar um impacto na competitividade do setor em que os bancos de varejo tradicionais poderão ter que se mover para este modelo baseado em plataforma com atendimento centrado no cliente Dado a importância de um setor bancário eficiente para o desenvolvimento econômico pelo desempenho de sua função principal de intermediação financeira e a recente disrupção digital no setor com o surgimento dos bancos digitais surge o questionamento se estes bancos pelo seu modo de operação são mais ou menos eficientes que os bancos tradicionais Esta é a pergunta que este trabalho se propõe a responder É pretensão deste trabalho avaliar a eficiência do sistema bancário brasileiro no período recente de 2019 a 2021 período em que bancos digitais passaram a atuar mais intensamente no mercado bancário brasileiro Buscase com isso contribuir com a literatura nacional de eficiência bancária investigando se esta nova modalidade de instituição financeira os bancos digitais são mais ou menos eficientes que os bancos de varejo tradicionais Pretendese também examinar se a entrada destas instituições promoveu um aumento de eficiência no mercado de crédito brasileiro A Análise Envoltória de Dados DEA desenvolvida por Charnes Cooper e Rhodes 1978 é uma técnica utilizada para mensurar a eficiência de unidades produtivas que os autores denominam de Decision Making Units DMUs Essas unidades produtivas são caracterizadas por empregarem o uso dos mesmos insumos inputs para a produção de determinados produtos outputs A análise de eficiência na DEA tem como objetivo medir o desempenho da DMU em converter os insumos utilizados em produtos Esta técnica emprega a programação matemática para a construção da fronteira eficiente A DEA é uma técnica não paramétrica ou seja não necessita que seja especificada uma forma para a fronteira eficiente Existem outras técnicas para mensurar a eficiência de unidades produtivas como a Free Disposal Hull FDH e a Stochastic Frontier Analysis SFA Essas técnicas têm sido largamente utilizada pelos pesquisadores para investigar a eficiência 3 dos bancos Neste estudo é utilizada a técnica DEA que é mais aprofundada ao longo do capítulo 2 que é destinado a metodologia do trabalho Bancos podem ser avaliados por diferentes abordagens Neste trabalho será procurado avaliar a eficiência dos bancos brasileiros e realizar a comparação entre bancos digitais e bancos de varejo tradicionais no desempenho de sua principal função a intermediação financeira Pretendese fazer assim pois a concessão de crédito é a função dos bancos que mais tem impacto no desenvolvimento da economia de um país e permitirá investigar como os bancos digitais desempenham esta atividade Por fim é utilizado o Índice de Malmquist para o triênio de 2019 a 2021 com o objetivo de analisar a evolução do desempenho do setor bancário brasileiro ao longo deste período Este índice permite realizar uma análise completa da evolução da produtividade ao longo de um período determinado e sua decomposição permite obter informações sobre as variações de eficiências dos bancos e sobre as variações da tecnologia de produção Objetivos Fundamentandose na contextualização e justificativa do problema de pesquisa realizadas acima o presente estudo possui como objeto geral realizar a análise comparativa do desempenho dos bancos digitais com os bancos de varejo tradicionais atuantes no mercado de crédito brasileiro Ainda este trabalho pretende alcançar os seguintes objetivos específicos a Resumir as principais abordagens de avaliação dos bancos presentes na literatura sobre eficiência bancária b Definir o modelo com os inputs e outputs a serem utilizados de acordo com o que for mais indicado para avaliar os bancos na atividade de concessão de crédito c Avaliar a convexidade e os retornos de escala do conjunto de produção representado na amostra a ser selecionada d Mensurar a eficiência técnica dos bancos pela abordagem da intermediação financeira e Comparar os escores de eficiência médios dos bancos digitais com os bancos de varejo tradicionais 4 f Calcular e decompor o Índice de Produtividade de Malmquist para investigar as variações de eficiência e avanços tecnológicos no triênio de 2019 a 2021 g Contribuir com a literatura nacional sobre eficiência bancária investigando como os bancos digitais desempenham em comparação com os bancos de varejo e como impactam no mercado de crédito brasileiro Estrutura da dissertação Com a finalidade de alcançar o objetivo geral e os objetivos específicos propostos esta dissertação está dividida em 3 capítulos além desta introdução e das conclusões No capítulo 1 é feita uma revisão da literatura do setor bancário São apresentados os primeiros trabalhos que se propuseram a avaliar a eficiência de instituições financeiras discussão acerca das abordagens de análise que podem ser adotadas apresentada primeiramente por Favero e Papi 1995 é discorrida Ao final é feita uma contextualização das principais pesquisas aplicadas ao mercado de crédito brasileiro No capítulo 2 é apresentada toda a metodologia utilizada por este trabalho Inicialmente são explanadas as contribuições dos trabalhos da década de 50 considerados como o alicerce das técnicas utilizadas atualmente para se avaliar a eficiência Em seguida são apresentados modelos da Análise Envoltória de Dados DEA desenvolvidos por Charnes Cooper e Rhodes 1978 e Banker Charnes e Cooper 1984 Ainda são expostos os testes utilizados para se testar a convexidade do conjunto de produção e a presença de retornos constantes e variáveis de escala O teste de Kneip et al 2016 para se realizar a comparação da média das eficiências de grupos de produtores distintos utilizado para se alcançar o objetivo principal deste trabalho é explicado em sequência Por fim o Índice de Malmquist e a sua decomposição nos índices de alterações de eficiência e de alterações tecnológicas são expostos O capítulo 3 se inicia com a explicação do modelo proposto por este trabalho para analisar a eficiência bancária brasileira recente São discutidos os resultados obtidos descrevendo e analisando os desempenhos mensurados pelos bancos As comparações entre bancos digitais e de varejo tradicionais realizadas pelos testes estatísticos propostos são discorridas em seguir Por fim é realizada uma análise temporal do mercado de crédito 5 brasileiro por meio do Índice de Malmquist avaliando as alterações de eficiência técnica e as alterações da tecnologia ao longo do triênio de 2019 a 2021 Nas conclusões deste trabalho são apresentadas as principais conclusões obtidas no capítulo anterior e feita um diagnóstico geral do desempenho do mercado de crédito brasileiro Por fim são feitas sugestões para trabalhos futuros que abordem a mesma temática deste trabalho 6 1 REVISÃO DA LITERATURA SOBRE EFICIÊNCIA BANCÁRIA Este capítulo tem como objetivo fazer uma resenha da literatura sobre eficiência no setor bancário apresentando os mais relevantes trabalhos presentes na literatura internacional e nacional Na primeira seção serão apresentados os trabalhos considerados pioneiros que introduziram as técnicas de medição de eficiência para avaliação dos bancos e alguns dos principais trabalhos da literatura internacional Em seguida na segunda seção são discutidas as principais abordagens utilizadas para se avaliar as instituições financeiras apresentando o objetivo que cada uma delas pretende alcançar suas principais escolhas de inputs e outputs e os resultados alcançados Por fim são discutidas as pesquisas que buscaram avaliar a performance de bancos nacionais discorrendo sobre aspectos como a escolha dos insumos e produtos a formação da amostra e as conclusões alcançadas por estes estudos 11 Literatura internacional sobre eficiência no setor bancário Os primeiros autores a utilizar a técnica DEA para avaliar o desempenho do setor bancário foram Sherman e Gold 1985 O desempenho operacional de um branco era frequentemente medido usando índices contábeis como retorno sobre ativos e retorno sobre investimentos Sherman e Gold 1985 Segundo os autores a utilização destes índices apresentava limitações pois estes eram elaborados por meio de agregações extensas que não refletiam a eficiência operacional podendo levar a conclusões falsas Com o objetivo de aplicar uma técnica que permita superar as fraquezas que a utilização de índices contábeis apresenta Sherman e Gold 1985 aplicaram a DEA para identificar e desenvolver maneiras para melhorar a eficiência de agências bancárias O modelo CCR foi aplicado em 14 agências de um banco que desejava substituir os índices que utilizava por uma técnica que fosse superior e levando a melhores avaliações Os autores utilizaram como inputs uma medida de utilização de tempo de empregados o aluguel pago por cada agência como um representativo do capital utilizado e por fim o custo total dos suprimentos utilizados Como output foram utilizados o número de transações 7 realizadas pela agência como abertura de contas pedidos de empréstimos seguros de vida vendidos etc Os resultados apresentados pelos autores mostram que das agências analisadas oito apresentaram o escore de eficiência máximo enquanto seis agências apresentaram algum nível de ineficiência Nenhuma das cinco maiores agências foi identificada como ineficiente e duas das ineficientes eram as menores em termos de transações totais Os resultados foram validados pelos gerentes bancários que apontaram como possível explicação o fato de que as agências menores eram administradas por gerentes menos experientes Enquanto que Sherman e Gold 1985 estavam interessados em analisar as agências de um banco em particular Rangan et al 1988 foram os primeiros a avaliar o mercado bancário a partir de uma visão do setor como um todo Esses autores selecionaram uma amostra com 215 bancos em operação nos Estados Unidos e utilizaram da técnica da DEA para mensurar a eficiência técnica destas instituições Rangan et al 1988 tinham a intenção de analisar o desempenho das instituições na atividade de intermediação financeira no ano de 1986 Para isso utilizaram como inputs o número de trabalhadores de tempo integral ativos fixos e fundos arrecadados Como outputs foram utilizados os empréstimos imobiliários os empréstimos comerciais os empréstimos industriais os depósitos à vista e por fim os depósitos a prazo e de poupança A partir da eficiência média da amostra obtida através do modelo DEA Rangan et al 1988 concluíram que os bancos norteamericanos poderiam produzir o mesmo nível de outputs com 70 dos inputs que usavam Ferrier e Lovell 1990 compararam duas técnicas para avaliar o desempenho de uma amostra de 575 bancos americanos em atividade no ano de 1984 A primeira é uma abordagem econométrica de função translog e a segunda é a Análise Envoltória de Dados Os autores utilizaram como inputs o número total de empregados custos de aluguel e despesas com mobília e equipamentos Como outputs foram utilizados os números de contas de depósitos à vista depósitos a prazo empréstimos imobiliários e o número de empréstimos comerciais Os autores afirmaram que as técnicas estavam de substancial acordo em vários aspectos importantes Ferrier e Lovell 1990 concluíram que as duas técnicas estavam correlacionadas positivamente Berger e Humphrey 1991 propuseram uma nova técnica para mensurar e analisar as eficiências de todos os bancos dos Estados Unidos em 1984 o método paramétrico denominado Análise de Fronteira Espessa Thick Frontier Analysis TFA Os resultados do estudo sugerem 8 que as ineficiências dos bancos derivam dos inputs de capital físico e mãodeobra o que os autores chamaram de ineficiências operacionais Do ponto de vista político Berger e Humphrey 1991 indicam que a concorrência adicional promovida por uma desregulamentação do setor bancário pressionaria muitos dos bancos considerados ineficientes a cortar custos fundirse com instituições que apresentam eficiências operacionais ou deixar o mercado No estudo de Berg et al 1993 foram estudados os desempenhos do setor bancário de três países nórdicos A amostra foi composta com 503 bancos finlandeses 150 bancos noruegueses e 126 bancos suecos em atuação no ano de 1990 Os autores constataram que a amostra representava praticamente toda a indústria bancária destes países O objetivo era avaliar como bem preparados estariam os bancos de diferentes tamanhos destes países para encarar a competição mais intensa do Mercado Comum Europeu A técnica DEA foi utilizada por Berg et al 1993 considerando tanto o modelo com retornos constantes de escala quanto o modelo que permite retornos variáveis Os mercados bancários dos países foram analisados individualmente e em seguida o Índice de Malmquist foi aplicado para realizar uma comparação entre os países O modelo deste trabalho tem os empréstimos e os depósitos como outputs O número de agências é definido como output por representar a disponibilidade dos serviços bancários Os inputs considerados são o trabalho medido em horas de trabalho e capital medido pelos valores contábeis de máquinas e equipamentos Os resultados encontrados por Berg et al 1993 mostraram que o mercado de crédito sueco era o mais eficiente com os escores médios de 078 considerando retornos variáveis de escala e 069 considerando retornos constantes Com o objetivo de realizar uma comparação entre os países foi construído um Índice de Malmquist onde os países da Suécia foram considerados como base tecnológica por ser o país que se mostrou mais eficiente quando avaliado individualmente Os bancos suecos se mostraram 5263 mais produtivos que os bancos finlandeses com a maior parte da diferença tendo como origem o componente de eficiência A diferença tecnológica entre esses países era de aproximadamente 10 Quando comparados os bancos suecos com os bancos noruegueses a diferença de produtividade foi de 4046 com a maior parte da diferença se mantendo no componente de eficiência Os bancos da Noruega foram 10 mais produtivos que os bancos da Finlândia com o componente tecnológico apontando para esta direção nos modelos CCR e BCC Berg et al 1993 concluem que a maior parcela dos bancos que se encontram na fronteira nórdica de eficiência são suecos indicando que os bancos deste país estão na melhor posição para expansão em um futuro mercado bancário comum nórdico 9 A estimação da fronteira eficiente nos estudos de performance bancária é dividida entre os métodos paramétricos e não paramétricos Entre os métodos paramétricos a técnica mais utilizada é a Análise da Fronteira Estocástica Stochastic Frontier Analysis SFA que especifica uma forma funcional para a fronteira eficiente presente nos trabalhos de Ferrier e Lovell 1990 Berger e DeYoung 1997 Bonin et al 2005 e Berger et al 2009 Bonin et al 2005 usando dados do período de 1996 a 2000 avaliaram os efeitos do tipo de controle na eficiência bancária em onze países emergentes europeus Os autores utilizaram o método paramétrico SFA para mensurar os escores de eficiência destes bancos As variáveis de saída foram o total de depósitos o total de empréstimos e o total de ativos líquidos e investimentos Como entradas os autores selecionaram variáveis que representassem os preços do capital e os preços dos fundos arrecadados Os resultados mostraram que os bancos públicos não eram significativamente menos eficientes que os bancos privados sugerindo que a privatização não seria por si só suficiente para aumentar a eficiência do mercado de crédito destes países Bancos estrangeiros foram considerados mais eficientes em custos do que outros tipos de controle A técnica DEA foi amplamente utilizada para avaliar a eficiência de setores bancários de diversos países Barros et al 2012 aplicaram a DEA para analisar a eficiência técnica de bancos japoneses de 2000 a 2007 Azad et al 2019 avaliaram a eficiência do setor bancário de Bangladesh durante o ano de 2017 com a técnica DEA Ariff e Can 2008 avaliaram a eficiência de custos e de lucros de 28 bancos comerciais chineses com essa técnica não paramétrica Da mesma forma Kao e Liu 2013 examinaram a eficiência dos bancos taiwaneses A DEA é uma técnica não paramétrica de construção de fronteira eficiente enquanto que a SFA especifica uma forma funcional para a fronteira Vários estudos foram propostos na literatura para comparar essas técnicas Kuchler 2013 compararam a DEA e a SFA na estimação da eficiência dos bancos dinamarqueses durante a crise dos anos 2008 a 2010 O autor encontrou uma correlação considerável mas não perfeita entre os rankings de eficiência estimados pela DEA e SFA No mesmo sentido Tabak et al 2014 compararam as duas técnicas para estimar a eficiência de uma amostra de bancos chineses Os autores indicaram que a DEA e a SFA se mostraram consistentes para construir rankings de eficiência ao longo dos anos para o setor bancário como um todo porém indicaram que as correlações apontaram para divergências nos diagnósticos das performances individuais dos bancos Paradi et al 2011 construíram um modelo DEA de dois estágios com o objetivo de avaliar a performance de 816 agências de um dos cinco maiores bancos canadenses por três 10 abordagens diferentes produção rentabilidade e intermediação As eficiências obtidas nos três modelos são os outputs do segundo estágio onde uma variável dummy com o valor 1 é empregado como input para todas as agências Segundo os autores esta avaliação de múltiplas abordagens permitiria que os gestores através da aplicação DEA aprimorassem suas gestões em negócios reais O mercado bancário italiano foi estudado por Favero e Papi 1995 onde foram mensuradas as eficiências técnicas e de escala de uma amostra de 174 bancos coletada no ano de 1991 Os autores explicam que a indústria bancária estava sofrendo mudanças na legislação e na regulação devido ao processo de unificação do mercado europeu com aumento da competição e integração e portanto seria importante avaliar o impacto destas mudanças no mercado de crédito Favero e Papi 1995 avaliaram o mercado italiano sob duas abordagens diferentes a abordagem da intermediação financeira e a abordagem dos ativos Os autores aplicaram os modelos DEACCR e DEABCC e obtiveram como escores de eficiência média para a abordagem da intermediação os valores de 088 e 091 respectivamente Enquanto que na abordagem de ativos foram observados os valores médios de 079 e 084 Foi aplicado um teste de Wald para testar a hipótese nula de igualdade entre as eficiências medidas dos dois modelos que foi rejeitada para as duas abordagens de estudo Assim os autores afirmaram que parte da ineficiência do setor bancário italiano é derivada da existência de ineficiências de escala Favero e Papi 1995 aplicaram uma regressão para investigar os determinantes das ineficiências no mercado bancário italiano usando o modelo BCC devido ao resultado do teste aplicado Eles avaliaram o impacto do tamanho dos bancos da especialização produtiva do tipo de controle da estrutura de mercado e da localização Os autores indicaram que as ineficiências eram melhor explicadas pela especialização produtiva tamanho e localização dos bancos Uma importante contribuição de Favero e Papi 1995 foi a realização de uma discussão sobre as abordagens que podem ser adotadas para se avaliar a eficiência de um setor bancário Foram identificadas pelos autores cinco abordagens presentes na literatura bancária Foram discutidos os principais aspectos a atividade bancária que se pretende avaliar e como é feita a escolha de inputs e outputs em cada uma delas Na seção a seguir essa discussão será apresentada 11 12 As principais abordagens de análise Uma importante decisão que deve ser feita em relação a avaliação de desempenho do setor bancário é a escolha de inputs e outputs Ahmad et al 2020 afirmam que diferentes abordagens têm sido utilizadas pelos pesquisadores para realizar esta definição e que ainda não existe consenso sobre o assunto Fethi e Pasiouras 2010 no mesmo sentido afirmaram que existe uma discussão em curso na literatura sobre eficiência bancária sobre as definições de inputs e outputs mais apropriadas Esses autores destacaram a seguinte frase de Bergendahl 1998 Tem havido quase tantas suposições de entradas e saídas como tem havido aplicações de DEA Bhatia et al 2018 no mesmo sentido afirmam que devido a complexidade das transações financeiras existem diferentes escolas de pensamento para definir inputs e outputs para a eficiência bancária Favero e Papi 1995 discutiram as especificações de inputs e outputs dos trabalhos que buscaram avaliar o desempenho das instituições financeiras até aquele momento Os autores afirmaram que economistas estavam divididos sobre a definição correta de entradas e saídas e que uma definição da atividade bancária é essencial para que se possa avaliar a eficiência do setor bancário Foram identificadas por estes autores quatro abordagens para a especificação de inputs e outputs a abordagem da produção Production Approach PA a abordagem da intermediação Intermediation Approach IA a abordagem dos ativos Asset Approach AA e por fim a abordagem do valor agregado Value Added Approach VAA Alguns dos principais trabalhos sobre eficiência bancária foram classificados por Favero e Papi 1995 de acordo com a abordagem utilizada apresentando os inputs e outputs escolhidos Essa classificação é apresentada na tabela 1 12 Tabela 1 Abordagens para avaliar a eficiência bancária presentes na literatura Autores Inputs Outputs Abordagem Sherman e Gold 1985 Trabalho Capital Custo de suprimentos Número de transações PA Rangan et al 1988 Trabalho Capital Fundos arrecadados Empréstimos Depósitos IA Ferrier e Lovell 1990 Trabalho Gasto com materiais Aluguel e mobília Número de contas de depósitos Número de empréstimos PA Berger e Humphrey 1991 Trabalho Capital Fundos arrecadados Depósitos Empréstimos IA Berg et al 1993 Trabalho Capital Empréstimos Depósitos Número de agências VAA English et al 1993 Trabalho Fundos arrecadados Empréstimos Investimentos AA Fonte FAVERO PAPI 1995 p389 Segundo Favero e Papi 1995 na abordagem da intermediação bancos são vistos como intermediários em que são transformados e transferidos os recursos financeiros de unidades superavitárias para unidades deficitárias Ahmad et al 2020 afirmaram que esta é claramente a abordagem mais popular na amostra de estudos que estes autores selecionaram que consiste nos 100 artigos mais citados da literatura Bhatia 2018 afirmam que a abordagem da intermediação foi proposta pela primeira vez por Sealey e Lindley 1977 Favero e Papi 1995 definem a abordagem de ativos como uma variante da abordagem da intermediação em que são considerados como outputs outros ativos dos bancos como investimentos além dos empréstimos que a instituição fornece Enquanto que a abordagem da 13 intermediação foca na principal atividade desenvolvida pelos bancos a concessão de crédito a abordagem dos ativos tem uma visão mais ampla da instituição considerando outros ativos produzidos A principal deficiência destas duas abordagens é que elas levam em consideração a maioria dos serviços prestados pelos bancos Favero e Papi 1995 Bhatia et al 2018 explicam que outra variação da abordagem da intermediação é a abordagem do valor adicionado em que as atividades dos bancos capazes de gerar valor adicional são consideradas como outputs Nesta linha o estudo de Berg et al 1993 considera como produto o número de agências pois representa a disponibilidade dos serviços bancários De acordo com Bhatia et al 2018 a abordagem da intermediação é a mais popular entre os pesquisadores e eles concordam que os bancos são intermediários de serviços financeiros Apesar de depósitos serem considerados como outputs em alguns trabalhos que tinham como objetivo avaliar a eficiência do mercado bancário pela abordagem da intermediação Bhatia et al 2018 afirmam que é mais lógico consideralos como inputs pois bancos usam depósitos para praticar atividades capazes de gerar receitas como empréstimos e pagar juros aos depositantes Ainda sobre o papel dos depósitos no modelo de eficiência bancária Fethi e Pasiouras 2010 explicam que existe uma controvérsia na literatura sobre este assunto Estes autores afirmam que alguns estudos selecionam como outputs apenas aqueles de acordo com o proposto por Sealey e Lindley 1977 enquanto que outros utilizam os depósitos como um output adicional o que os deixa mais próximo da abordagem do valor adicionado do que da abordagem da intermediação A abordagem da produção considera o banco como um fornecedor de serviços bancários De acordo com Favero e Papi 1995 nesta abordagem o banco é considerado como produtor de contas de depósitos e serviços de empréstimos onde outputs são mensurados pelo número de contas ou transações processadas A abordagem da produção está presente no trabalho de Ferrier e Lovell 1990 Alguns trabalhos procuram avaliar os bancos a partir de uma abordagem de rentabilidade O estudo de Faria et al 2006 avaliou os bancos atuantes no mercado de crédito do Brasil pela sua rentabilidade usando com inputs despesas com juros e despesas não relacionadas a juros e como outputs receitas da intermediação financeira receitas de prestação de serviços e outras receitas operacionais Bhatia et al 2018 explicam que muitos pesquisadores usam mais de uma abordagem no mesmo estudo e outros modificam as especificações de inputs e outputs de forma que eles 14 não se enquadram em nenhuma abordagem mencionada para se adequar melhor ao objetivo pretendido pelos estudos Este presente trabalho propõe avaliar o setor bancário brasileiro de acordo com o modelo tradicional de intermediação financeira proposto por Sealey e Lindley 1977 em que os bancos são vistos como intermediários entre agentes poupadores e agentes tomadores de empréstimos A orientação do modelo aplicado será voltada para insumos que como Fethi e Pasiouras 2010 explicam a orientação para inputs é de longe a mais utilizada para estimar a eficiência dos bancos pois é assumido que os gestores tem maior controle sobre recursos empregados no processo do que sobre os produtos 13 A literatura nacional sobre eficiência bancária Vários estudos foram propostos para avaliar a eficiência no mercado de crédito brasileiro Ghilardi 2006 avaliaram o desempenho dos 50 maiores bancos que atuavam no Brasil por meio da técnica não paramétrica DEA Este trabalho avaliou o desempenho dos bancos considerando que o grau de eficiência pode ser medido pela capacidade de geração de receita utilizandose do patrimônio líquido e o número de funcionários os dois inputs do modelo Como outputs forma considerados a receita de serviços e outras receitas operacionais e o somatório das receitas de operações com arrendamento mercantil e as receitas de operações com títulos e valores mobiliários Ghilardi 2006 realizou uma comparação entre os modelos DEACCR e DEABCC com um índice de análise financeira tradicional o índice ROI Return On Investiment ROI Como resultado do estudo não foram observadas mudanças significativas entre os modelos DEA porém quando comparados com o índice ROI o autor afirma que parece que não se trata do mesmo assunto ou de que cada análise leva em conta princípios antagônicos dadas as contradições observadas na comparação Faria et al 2006 estavam interessados em avaliar o desempenho do setor bancário após um processo de 10 anos de mudanças profundas por uma onda de fusões e aquisições bancárias além da entrada de novas instituições estrangeiras no mercado brasileiro São avaliados o mercado bancário como um todo e individualmente os bancos envolvidos nestas operações por meio da técnica DEA A amostra foi composta de seis bancos varejistas que participaram 15 de fusões e aquisições no período analisado de 1995 à 2005 três bancos atacadistas e três bancos públicos Dois modelos foram construídos por Faria et al 2006 O primeiro modelo para a atividade de intermediação financeira tinha como inputs as seguintes rubricas contábeis captações totais despesas de pessoal e outras despesas administrativas e ativo permanente e imobilizado de arrendamento Como outputs foram utilizadas as operações de crédito e arrendamento mercantil e o somatório dos títulos e valores mobiliários e aplicações financeiras O modelo de resultados para avaliar a eficiência do banco em gerar renda tinha como inputs as despesas de intermediação financeira as despesas de pessoal e outras despesas As receitas de intermediação financeira e as receitas de prestações de serviços e outras receitas operacionais foram consideradas como as saídas deste modelo Os resultados apresentados por Faria et al 2006 indicam que o período de fusões e aquisições levou a um maior desempenho dos bancos nas atividades de intermediação financeira Os autores afirmaram que as instituições foram aumentando de tamanho e melhorando sua performance sendo possível atestar que o mercado de crédito brasileiro apresentou economias de escala na concessão de crédito Por outro lado os autores concluíram que no modelo de resultados o aumento do porte das instituições gerou impacto negativo na rentabilidade das mesmas Périco et al 2008 selecionaram os 12 maiores bancos comerciais para formarem a sua amostra Estes autores justificam a escolha pelo fato de não terem como objetivo explicar quais bancos são eficientes e quais não são mas sim de demonstrar de forma sucinta que nem sempre os maiores bancos são considerados os mais eficientes O modelo elaborado pelos autores teve como inputs o ativo total as operações de crédito os depósitos e o patrimônio líquido Receita financeira resultado financeiro resultado operacional e resultado líquido foram as quatro saídas do modelo O modelo escolhido foi o DEABCC com orientação para outputs Na amostra selecionada os cinco maiores bancos foram os únicos a apresentarem o escore de eficiência máximo Todos os outros bancos operavam com algum nível de ineficiência Périco et al 2008 destacaram que os 12 maiores bancos operavam com algum nível de ineficiência dado ao escore médio de 79 Os autores também afirmaram que a pergunta proposta inicialmente sobre a possibilidade de o maior banco ser o mais eficiente foi respondida positivamente sendo possível desde que ele otimize a utilização de seus recursos Outro ponto fundamental deste trabalho é a relação entre porte da instituição e eficiência que fica evidenciada no fato de que os cinco maiores bancos foram eficientes e os escores 16 diminuíam de acordo com o tamanho da instituição apontando para a presença de economias de escala no mercado de crédito brasileiro Paula e Faria 2008 avaliaram a evolução da eficiência técnica e de escala do setor bancário brasileiro no período de 2000 a 2006 Foram selecionadas para a amostra 38 instituições bancárias onde oito eram grandes bancos varejistas oito bancos de varejo regionais três bancos varejistas para alta renda doze bancos atacadistas e sete bancos especializados em crédito Os autores justificaram a segmentação pelo interesse em avaliar a eficiência bancária considerando uma dimensão ainda pouco estudada no mercado de crédito brasileiro o tipo e a natureza da instituição e seu nicho de mercado Para mensurar a eficiência técnica Paula e Faria 2008 optaram pelo modelo DEA BCC pois estavam interessados em avaliar a eficiência sem a influência da escala de operação do banco e este modelo permite retornos variáveis de escala A eficiência de escala foi calculada apenas para os bancos varejistas Os autores justificaram essa escolha pelo fato de que os bancos atacadistas possuem características semelhantes a um banco de investimento não tendo a concessão de crédito como sua atividade principal e por sua vez os bancos especializados em crédito possuem um foco de negócios bem específico para poderem ser comparados com bancos varejistas Nos resultados Paula e Faria 2008 destacam que a evolução da eficiência técnica no período de análise não apresentou melhorias acentuadas nem perdas Os autores apontaram que uma melhoria recente na eficiência técnica no modelo dos resultados coincidiu com o boom de crédito ocorrido no país Foram identificados níveis maiores de eficiência técnica nos grandes bancos varejistas e bancos de varejo para alta renda que os autores atribuíram a um possível aprimoramento operacional dos bancos e cortes nos custos administrativos Sobre a eficiência de escala Paula e Faria 2008 mostram que o grupo dos grandes bancos varejistas apresentaram o segundo melhor desempenho entre os segmentos selecionados enquanto que os bancos varejistas regionais apresentaram o pior desempenho da amostra Os autores indicam que isto parecia evidenciar que o tamanho da instituição afeta o seu desempenho Os resultados da eficiência de escala no modelo de intermediação mostram uma pequena perda de escala Os autores indicam que a opção de ser grande no mercado varejista parece ser vantajosa em função do maior potencial de geração de receitas e lucros decorrentes da ampliação no volume dos serviços ofertados e do potencial de vendas cruzadas que um banco pode oferecer Staub et al 2010 avaliaram o mercado de crédito brasileiro no período de 2000 a 2007 Foram estimadas as eficiências de custos técnica e alocativa aplicando a técnica DEA Os 17 bancos foram avaliados a partir da abordagem da intermediação em que os inputs representavam os usos de capital trabalho e fundos arrecadados Como outputs os autores utilizaram os empréstimos investimentos e depósitos A inclusão dos depósitos como outputs foi justificada pelo fato de que eles são proporcionais aos serviços de depósitos providos pelos bancos o que aproxima o modelo construído pelos autores da abordagem do valor adicionado variante do modelo da intermediação Os resultados de Staub et al 2010 indicam que bancos estrangeiros e bancos com participações estrangeiras são os menos eficientes comparados com outros tipos de controle sugerindo que a hipótese de vantagem global não estava prevalecendo no Brasil Os autores indicaram que o tamanho não era um importante fator determinante de eficiência Também não foram verificadas diferenças substanciais nos desempenhos dos bancos de diferentes áreas de atuação Por fim os autores afirmaram que os altos níveis de ineficiência técnica sugerem que ainda existe um largo caminho para melhor a eficiência bancária no Brasil A eficiência dos bancos brasileiros foi avaliada durante o longo período de 1998 a 2010 por Wanke e Barros 2014a com a utilização de um modelo paramétrico estocástico As variáveis foram escolhidas de acordo com a abordagem de intermediação financeira proposta por Sealey e Lindley 1977 onde o banco é visto como uma empresa financeira que emprega depósitos empregados e ativos totais para gerar empréstimos e investimentos em títulos Os autores indicaram que através da aplicação deste modelo foi verificado uma melhora no desempenho do setor bancário brasileiro ao longo do tempo Wanke e Barros 2014b aplicaram um modelo DEA de dois estágios para investigar a eficiência dos bancos brasileiros no ano de 2012 O número de agências e o número de funcionários foram os inputs do output intermediário de despesas administrativas e despesas de pessoal que posteriormente se tornaram inputs do segundo estágio que tinha como outputs capital próprio e ativos permanentes Os autores indicaram estarem interessados em olhar para os bancos como uma sequência de um processo produtivo que transforma recursos humanos e físicos em termos financeiros Os resultados apresentados por Wanke e Barros 2014b indicam que apenas um banco da amostra de 40 instituições se mostrou eficiente no primeiro estágio Os autores afirmaram que os bancos brasileiros não eram eficientes em usar os recursos humanos e físicos No segundo estágio nenhum banco se mostrou eficiente As escores médios de eficiência do primeiro e segundo estágio foram respectivamente 043 e 086 indicando que os bancos brasileiros são mais eficientes em converter despesas administrativas e de pessoal em ativos permanentes e capital próprio do que em gerir seus recursos humanos e físicos 18 Cava et al 2016 promoveram uma análise diferente do setor bancário brasileiro Estes autores se propuseram a avaliar a eficiência pela abordagem da produção No modelo elaborado pelos autores o banco é visto como um fornecedor de serviços depósitos representados pelo output de contas de depósitos e como um fornecedor de outros serviços representados pelas receitas não relacionadas à juros Como inputs foram utilizados o número de empregados as despesas operacionais líquidas de despesas da intermediação financeira e os ativos fixos Os bancos foram avaliados no ano de 2013 utilizando o modelo DEABCC pois os autores afirmaram que os bancos apresentam ganhos relativos ao tamanho dos serviços totais produzidos Os resultados obtidos por Cava et al 2016 indicam que os maiores bancos são mais eficientes visto que apresentam o maior escore médio de eficiência Os autores apontam que uma possível explicação seria as economias de escala alcançadas pelos maiores bancos Considerando o tipo de controle os bancos públicos foram considerados mais eficientes que os bancos privados e estrangeiros indicando que estes bancos são mais eficientes em produzir serviços Ainda bancos estrangeiros se mostraram mais eficientes que os bancos privados Bancos com risco classificado com AAA obtiveram os maiores escores de eficiência médio indicando que bancos com melhores serviços apresentam melhores resultados ao lado de melhor classificação de risco Carneiro et al 2016 pela abordagem da intermediação financeira avaliaram a eficiência de 99 instituições financeiras que atuavam no mercado brasileiro em 2013 Foram considerados como inputs o total de depósitos o número de trabalhadores e as despesas com juros enquanto que como outputs foram considerados os empréstimos as receitas de intermediação financeira e os investimentos e aplicações dos bancos O modelo escolhido foi o DEABCC pois os autores consideraram que as instituições financeiras atuam em ambientes onde há ganhos e perdas de escala No conjunto de instituições que apresentaram o nível máximo de eficiência no estudo de Carneiro et al 2016 destacamse as instituições financeiras privadas com participação estrangeira e as públicas com escores médios respectivos de 88 e 79 Os resultados vão de encontro com os obtidos por Staub et al 2010 onde as instituições financeiras públicas são mais eficientes que as privadas nacionais Por outro lado no trabalho de Staub et al 2010 as instituições privadas com participação estrangeira se mostraram economicamente ineficientes O tamanho foi identificado como determinante para o escore de eficiência com os bancos de grande porte se mostrando mais eficientes Também foi identificado uma forte relação entre risco de crédito com a eficiência 19 Mendonça 2017 investigou a relação entre a eficiência mensurada pela técnica DEA com a da SFA para uma amostra de 47 bancos em atuação no mercado brasileiro nos anos de 2008 a 2015 Os resultados deixam evidentes que cada método tem objetivos distintos que se refletem na forma como estimam a eficiência das instituições analisadas A DEA apresentou um maior número de instituições eficientes o que os autores atribuíram ao método de fazer um ranqueamento das instituições eficientes O método SFA apresentou uma menor dispersão nos escores de eficiência o que o autor explicaram pela atribuição de pesos iguais para as variáveis no momento de estimar a fronteira eficiente A tabela a seguir apresenta os inputs e outputs utilizados nos trabalhos apresentados nesta seção que avaliaram a eficiência do setor bancário brasileiro O resumo a seguir contém apenas os trabalhos que são passíveis de ser enquadrados na classificação das abordagens de análise apresentada por Favero e Papi 1995 Tabela 2 Abordagens utilizadas na literatura nacional continua Autores Inputs Outputs Abordagem Faria et al 2006 Captações Despesas de pessoal Ativo permanente e imobilizado de arrendamento Operações de crédito Arrendamento mercantil IA Paula e Faria 2008 Número de funcionários Depósitos totais Permanente e imobilizado de arrendamento Capital Operações de crédito e arrendamento mercantil Investimentos AA Staub et al 2010 Uso de capital Trabalho Fundos arrecadados Empréstimos Investimentos Depósitos AA 20 Tabela 2 Abordagens utilizadas na literatura nacional conclusão Wanke e Barros 2014a Cava et al 2016 Carneiro et al 2016 Depósitos Empregados Ativos totais Número de empregados Despesas operacionais líquidas Despesas da intermediação financeira Ativos fixos Total de depósitos Número de trabalhadores Despesas com juros Empréstimos Investimentos Contas de depósito Receitas não relacionadas à juros Empréstimos Receitas de intermediação financeira Investimentos e aplicações AA PA AA Fonte O autor 2022 21 2 METODOLOGIA Este capítulo apresenta a metodologia utilizada para a análise que se pretende Os conceitos de produtividade e eficiência são diferenciados na primeira seção Em seguida são apresentadas as contribuições dos trabalhos de Koopmans 1951 Debreu 1951 e Farrell 1957 que são considerados como o alicerce das técnicas utilizadas para se avaliar a eficiência atualmente Nas seções seguintes são apresentados a Análise Envoltória de Dados DEA seus principais conceitos e os modelos de Charnes Cooper e Rhodes 1978 conhecido como modelo CCR e Banker Charnes e Cooper 1984 conhecido como modelo BCC A explicação dos métodos utilizados neste trabalho continua com a apresentação das principais características da técnica nãoparamétrica DEA e sua comparação com a técnica paramétrica de Análise da Fronteira Estocástica SFA O viés característico da DEA é discutido e o método de Bootstrap introduzido no âmbito da DEA pelo trabalho de Simar e Wilson 1998 é exposto em sequência Em seguida são apresentados os testes utilizados para se testar a convexidade do conjunto de produção e a presença de retornos constantes de escala contra retornos variáveis O teste de Kneip et al 2016 utilizado para se alcançar o objetivo principal deste trabalho que permite realizar a comparação da média das eficiências de grupos de produtores distintos é explicado em sequência Por fim o Índice de Produtividade de Malmquist e a sua decomposição nos índices de alterações de eficiência e de alterações tecnológicas são expostos na última seção 21 Produtividade e eficiência Existem algumas expressões que são bastante utilizadas quando estamos nos referindo ao desempenho produtividade eficiência Apesar de tratarem de um assunto semelhante possuem conceitos diferentes que não se confundem A produtividade é definida como a razão entre o output produzido e o input que foi utilizado 22 𝒑𝒓𝒐𝒅𝒖𝒕𝒊𝒗𝒊𝒅𝒂𝒅𝒆 𝒐𝒖𝒕𝒑𝒖𝒕 𝒊𝒏𝒑𝒖𝒕 1 Coelli et al 1998 explicam que o cálculo da produtividade é simples quando o processo produtivo envolve um único input e um único output porém quando existirem mais de um algum método que realize a agregação desses inputs ou outputs precisa ser aplicado para se obter uma medida da produtividade Para ilustrar a diferença entre produtividade e eficiência Coelli et al 1998 apresentaram uma fronteira de produção que representa o máximo de um output y que pode ser obtido dado a quantidade do input x utilizado Figura 1 Produtividade e eficiência Fonte COELLI et al1998 p4 As unidades C e B encontramse posicionados na fronteira de produção F representando pontos eficientes A é ineficiente pois poderia aumentar o nível do output produzido dado a quantidade de input que utiliza Para ser eficiente A deveria produzir a mesma quantidade de output produzido pela unidade B Considerando um raio que passa pela origem e tem como inclinação yx uma medida de produtividade para as unidades consideradas caso A mudasse para o ponto de B teria sua produtividade aumentada pois a inclinação do raio seria maior No entanto caso A mudasse para o ponto C o raio da reta que passa pela origem é tangente à fronteira eficiente definindo o ponto de máxima produtividade possível 23 Coelli et al 1998 concluem que um ponto pode ser tecnicamente eficiente pois se encontra na fronteira de eficiência mas não ser a mais produtiva podendo aumentar a sua produtividade explorando economias de escala 22 Eficiência técnica e eficiência alocativa Os trabalhos de Koopmans 1951 Debreu 1951 e Farrell 1957 foram os alicerces para o desenvolvimento das técnicas que utilizadas atualmente para avaliar se um plano de produção é eficiente Casado e Souza 2015 explicam que os estudos dedicados a análise de eficiência da década de 1950 vieram a servir como base para o posterior desenvolvimento da técnica não paramétrica conhecida como Análise Envoltória de Dados Data Envelopment Analysis DEA por Charnes Cooper e Rhodes 1978 Koopmans 1951 definiu um conceito de eficiência conhecido como modelo de Pareto Koopmans Nesta definição a eficiência é alcançada se nenhum dos produtos pode ser aumentado sem que algum outro produto seja reduzido ou algum insumo seja aumentado e nenhum dos insumos possa ser reduzido sem que algum outro insumo seja aumentado ou algum produto seja reduzido Assim dentro da tecnologia que a unidade de produção tem disponível a unidade produtiva eficiente não consegue aumentar um produto mantendo constantes os níveis de insumos utilizados e o nível dos outros outputs produzidos para tanto precisa realocar algum destes recursos para alcançar o aumento no produto que deseja Debreu 1951 determinou em seus estudos o coeficiente de utilização de recursos onde a eficiência é alcançada pela redução dos insumos utilizados Casado e Souza 2015 explicam que Debreu 1951 estabeleceu o primeiro indicador de eficiência produtiva conhecido Este coeficiente é baseado na minimização dos insumos utilizados Consiste na redução radial equiproporcional de todos os recursos mantendo o nível atual de produção que tornaria a unidade produtiva eficiente Portanto segundo Debreu uma unidade é eficiente quando não é possível gerar a mesma produção com uma quantidade equiproporcionalmente menor dos insumos utilizados Ramos 2007 explica que Farrell 1957 desenvolveu seu conceito de eficiência técnica inspirado na análise da atividade de Koopmans 1951 e no conceito de coeficiente de utilização de recursos de Debreu 1951 Segundo Casado e Souza 2015 Farrel estendeu o 24 trabalho iniciado por Debreu para desenvolver um método para calcular o indicador de eficiência produtiva Conforme explicado por Ramos 2007 Farrell apresenta um modelo conforme a figura 2 a seguir com dois insumos X1 e X2 e um produto SS denota a isoquanta e AA a isocusto Figura 2 Modelo de Farrell para eficiência técnica e alocativa Fonte RAMOS 2007 p3 O ponto P representa uma firma que usa mais insumos que a firma Q na isoquanta considerada eficiente Farrell então considera que a razão OQOP poderia representar uma medida radial equiproporcional dos insumos X1 e X2 da eficiência técnica que como explica Ramos 2007 está presente em praticamente todos os modelos e aplicações de DEA usadas atualmente A eficiência técnica será igual a unidade se a firma no ponto P puder alcançar a isoquanta e se aproximará de zero quando a distância entre P e Q aumentar Podese perceber também que Q e Q são unidades de produção consideradas eficientes tecnicamente pois produzem o mesmo produto com combinações ótimas dos insumos X1 e X2 Uma firma que opera em um ponto da isoquanta QQ é tecnicamente eficiente mas isso não significa que estará operando com a combinação mais lucrativa dos fatores de produção dados os preços relativos Considerando a relação de preços dos insumos representada pela isocusto AA que tangencia a isoquanta SS a unidade Q é mais eficiente em termos de preço ou seja é mais eficiente alocativamente Assim Q e Q possuem a mesma eficiência técnica produtividade mas diferente eficiência alocativa custos Enquanto que para calcular a eficiência técnica precisamos apenas dos insumos e produtos para a eficiência alocativa é preciso conhecer o custo dos insumos seus preços Pela característica dos inputs utilizados neste trabalho que serão apresentados no próximo capítulo 25 por não apresentarem custos não é possível avaliar a eficiência dos bancos no emprego dos insumos em proporções adequadas dados os seus respectivos preços 23 A Análise Envoltória de Dados DEA A Análise Envoltória de Dados DEA foi apresentada primeiramente por Charnes Cooper e Rhodes 1978 A DEA é utilizada para a mensurar a eficiência de unidades produtivas homogêneas Estas unidades produtivas foram chamadas por estes autores de Decision Making Units DMUs e são caracterizadas por empregarem o uso dos mesmos inputs para a produção de determinados outputs A análise de eficiência tem como objetivo medir o desempenho da DMU em converter os insumos utilizados em produtos A DMU tecnicamente eficiente é aquela que opera na fronteira das possibilidades de produção Casado e Souza 2015 explicam que Charnes Cooper e Rhodes 1978 generalizaram os estudos de Farrel para trabalhar com múltiplos insumos e múltiplos outputs Essa generalização deu origem a DEA técnica que permite a construção de fronteiras de produção e indicadores de eficiência produtiva A DEA é uma técnica nãoparamétrica assim não necessita que seja especificada uma forma para a função que representa a fronteira de produção Esta técnica emprega programação matemática para a construção das fronteiras Os modelos básicos de DEA são o modelo originalmente proposto por Charnes Cooper e Rhodes 1978 conhecido como modelo CCR e o modelo proposto por Banker Charnes e Cooper 1984 conhecido como modelo BCC O primeiro modelo considera que a tecnologia de produção apresenta apenas retornos constantes de escala enquanto que o modelo BCC considera que as tecnologias podem apresentar retornos variáveis Os modelos DEA em sua aplicação podem ser divididos em modelos orientados para inputs e modelos orientados para outputs No primeiro buscase obter um menor uso de inputs dado o nível de outputs produzidos O pesquisador que utiliza a orientação para outputs busca obter o maior nível de outputs possível mantendo os inputs fixos A seguir serão apresentados os principais modelos DEA existentes CCR orientado para inputs CCR orientado para outputs BCC orientado para inputs e BCC orientado para outputs 26 231 Modelo CCR O modelo apresentado originalmente por Charnes Cooper e Rhodes 1978 trabalha com retornos constantes de escala Constant Returns to Scale CRS e foi inicialmente proposto como um modelo orientado para inputs Souza e Meza 2006 explicam que este modelo determina a eficiência pela otimização da divisão entre a soma ponderada dos outputs chamada de output virtual e a soma ponderada dos inputs chamada de input virtual A eficiência técnica de qualquer DMU no modelo CCR CRS original é obtida através de um Problema de Programação NãoLinear nas seguintes condições 𝒎á𝒙 𝒉𝟎 𝒖𝒋𝒚𝒋𝟎 𝑺 𝒋𝟏 𝒗𝒊𝒙𝒊𝟎 𝑹 𝒊𝟏 1 sa 𝒖𝒋𝒚𝒋𝒌 𝑺 𝒋𝟏 𝒗𝒊𝒙𝒊𝒌 𝑹 𝒊𝟏 𝟏 𝒌 𝟏 𝟐 𝒏 2 𝒖𝒋 𝒗𝒊 𝟎 𝒋 𝟏 𝑺 𝒊 𝟏 𝑹 3 Onde h0 Eficiência da DMU que está sendo avaliada yjk Quantidade do output yj para a DMU k xik Quantidade do input xi para a DMU k R Quantidade total de inputs S Quantidade total de outputs n Quantidade total de DMUs uj Peso referente ao output j vi Peso referente ao input i yj0 Quantidade do output yj para a DMU que está sendo avaliada xi0 Quantidade do input xi para a DMU que está sendo avaliada 27 A solução deste problema de maximização leva a determinação dos pesos u e v que maximizam a eficiência da DMU que está sendo avaliada A primeira condição do problema indica que as razões das demais DMUs deve ser menor ou igual à unidade enquanto que a segunda condição determina que os pesos devem ser maiores ou iguais a zero A maximização confere então para esta DMU em análise a ponderação mais favorável que as restrições permitem O modelo proposto por Charnes et al 1978 em uma forma fracionária pode ser formulado como um Problema de Programação Linear conhecido como modelo dos multiplicadores O modelo CCR input orientado na forma do modelo dos multiplicadores tem a seguinte formulação 𝒎á𝒙 𝒉𝟎 𝒖𝒋𝒚𝒋𝟎 𝑺 𝒋𝟏 4 sa 𝒗𝒊𝒙𝒊𝟎 𝑹 𝒊𝟏 𝟏 5 𝒖𝒋𝒚𝒋𝒌 𝑺 𝒋𝟏 𝒗𝒊𝒙𝒊𝒌 𝑹 𝒊𝟏 𝟎 𝒌 𝟏 𝒏 6 𝒖𝒋 𝒗𝒊 𝟎 𝒋 𝟏 𝑺 𝒊 𝟏 𝑹 7 O modelo dos multiplicadores pode ser derivado em um modelo dual que como explicam Souza e Meza 2006 apresentará uma menor quantidade de restrições e por ter solução computacional mais simples tem preferência sobre o modelo dos multiplicadores Esta forma dual conhecida como modelo da envoltória ou do envelope é apresentado a seguir 𝒎í𝒏 𝜽 8 sa 28 𝒚𝒋𝟎 𝒚𝒋𝒌𝝀𝒌 𝒏 𝒌𝟏 𝟎 𝒋 𝟏 𝑺 9 𝜽𝒙𝒊𝟎 𝒙𝒊𝒌𝝀𝒌 𝒏 𝒌𝟏 𝟎 𝒊 𝟏 𝑹 10 𝝀𝒌 𝟎 𝒌 𝟏 𝒏 11 Onde 𝜆𝑘 Peso da combinação convexa da DMU k No modelo da envoltória a DMU é eficiente quando 𝜃 1 caso for menor que a unidade reflete o valor que devemos multiplicar todos os inputs para produzir o nível de outputs atual de maneira eficiente Conforme Souza e Meza 2006 no modelo DEA é conveniente que o número de DMUs seja maior que o número de variáveis inputs e outputs visto que se n S R o modelo pode eventualmente não discriminar os escores das DMUs Logo quando o número de DMUs for maior que o número de variáveis ou sejam n S R o modelo do envelope possui menos restrições S R que o modelo dos multiplicadores n1 O modelo original CCR e os modelos dos multiplicadores e do envelope foram expostos para a orientação de inputs O modelo CCR output orientado na forma da envoltória tem a seguinte formulação 𝒎á𝒙 𝜽 12 sa 𝒙𝒊𝟎 𝒙𝒊𝒌𝝀𝒌 𝒏 𝒌𝟏 𝟎 𝒊 𝟏 𝑹 13 𝜽𝒚𝒋𝟎 𝒚𝒋𝒌𝝀𝒌 𝒏 𝒌𝟏 𝟎 𝒋 𝟏 𝑺 14 𝝀𝒌 𝟎 𝒌 𝟏 𝒏 15 29 No modelo CCR output orientado 𝜃 é o valor que devemos multiplicar todos os outputs para que a DMU em análise seja eficiente Desta forma 𝜃 é maior ou igual a 1 Logo devemos fazer 1 𝜃 se quisermos encontrar as eficiências no intervalo de 0 a 1 como no modelo CCR orientado para inputs 232 Modelo BCC Banker Charnes e Cooper 1984 desenvolveram um modelo que utiliza retornos variáveis de escala Variable Returns to Scale VRS assim de acordo com Casado e Souza 2015 evita problemas existentes em situações de competição imperfeita O modelo CCR é limitado no sentido de considerar que a tecnologia pode apresentar apenas retornos constantes de escala O Modelo BCC como é conhecido permite que a tecnologia apresente também retornos crescentes ou decrescentes de escala Na forma dos multiplicadores essa flexibilidade é acrescentada pela inclusão de um intercepto na função objetivo como pode ser visto nas equações a seguir que apresenta o modelo orientado para inputs 𝒎á𝒙 𝒉𝟎 𝒖𝒋𝒚𝒋𝟎 𝑺 𝒋𝟏 𝒖 16 sa 𝒗𝒊𝒙𝒊𝟎 𝑹 𝒊𝟏 𝟏 17 𝒖𝒋𝒚𝒋𝒌 𝑺 𝒋𝟏 𝒗𝒊𝒙𝒊𝒌 𝑹 𝒊𝟏 𝒖 𝟎 𝒌 𝟏 𝒏 18 𝒖𝒋 𝒗𝒊 𝟎 𝒋 𝟏 𝑺 𝒊 𝟏 𝑹 19 A fronteira é construída a partir dos pontos eficientes Como explicam Banker et al 1984 em um ponto eficiente retornos crescentes decrescentes ou constantes de escala são 30 verificados a depender do valor do intercepto Assim temos em um modelo orientado para inputs Para 𝑢 0 Retornos crescentes de escala Para 𝑢 0 Retornos constantes de escala Para 𝑢 0 Retornos decrescentes de escala 233 Comparação entre os modelos CCR e BCC Enquanto que no modelo CCR os pontos eficientes se encontram posicionados em uma reta que sai da origem no modelo BCC a fronteira é formada por segmentos de reta com inclinações diferentes devido ao intercepto de cada DMU eficiente que denota os retornos de escala presentes naquele ponto A Figura 3 a seguir ilustra essa diferença entre os modelos CCR e BCC Figura 3 Comparação das fronteiras eficientes dos modelos CCR e BCC Fonte BANKER et al 1984 p1089 31 Conforme explicam Tupy e Yamaguchi 1998 a eficiência produtiva tem dois componentes o puramente técnico físico e o alocativo preço o primeiro referese à habilidade de evitar perdas produzindo tantos produtos quanto os insumos utilizados permitem ou utilizando o mínimo de insumo possível no processo de produção o segundo componente referese à habilidade de combinar insumos e produtos em proporções ótimas dados os seus preços Para os autores uma firma além de ser eficiente na utilização dos insumos eficiência técnica e alocativa deverá também ser eficiente no produto eficiência de escala A eficiência de escala é obtida pela razão entre as eficiências dos modelos CCR e BCC ou seja entre as eficiências técnicas de longoprazo e de curtoprazo Uma firma eficiente no produto eficiência de escala opera onde uma mudança no nível dos outputs resultada em uma mudança proporcional nos inputs A ineficiência de escala é a ineficiência que advém de uma unidade produtiva estar operando fora do nível de produto com retornos constantes de escala A distância do modelo BCC para o modelo CCR representa a ineficiência de escala Na figura 3 podese observar que a DMU no ponto A é tecnicamente ineficiente enquanto que a DMU no ponto B representa uma firma que opera com o mesmo nível de produto mas com menos insumos sendo considerada tecnicamente eficiente A DMU do ponto E opera no ponto em que as curvas dos modelos CCR e BCC são iguais sendo além de eficiente o ponto mais produtivo O ponto E é eficiente em escala tendo produtividade máxima representando o tamanho ótimo de uma firma neste conjunto de produção 24 Características da técnica DEA e sua comparação com a SFA A principal característica da técnica de estimação DEA é ser baseada na programação linear não exigindo que se estabeleça especificações sobre a função de produção do processo que se está analisando Falcão e Correia 2012 destacam como uma importante característica desse método a sua flexibilidade que permite trabalhar com sistemas com diversos insumos e produtos De acordo com Falcão e Correa 2012 na DEA é analisada somente uma unidade por vez medindo o desempenho relativo dela o que gera um resultado para cada DMU que é utilizado para fazer a comparação e posteriormente ordenálos 32 Marinho 2001 elencou algumas das características mais interessantes da DEA que descrevem como essa técnica se configura a Caracteriza cada DMU como eficiente ou ineficiente através de uma única medida resumo de eficiência b Não faz julgamentos a priori sobre os valores das ponderações de inputs e outputs que levariam as DMUs ao melhor nível de eficiência possível c Pode prescindir mas não rejeita de sistemas de preços d Dispensa mas pode acatar préespecificações de funções de produção subjacentes e Pode considerar sistemas de preferências de avaliadores e de gestores f Baseiase em observações individuais e não em valores médios g Permite a incorporação na análise de insumos e de produtos avaliados em unidades de medidas diferentes h Possibilita a verificação de valores ótimos de produção e de consumo respeitando restrições de factibilidade i Permite a observação de unidades eficientes de referência para aquelas que forem assinaladas como ineficientes e j Produz resultados alocativos eficientes no sentido de Pareto Por sua vez a Análise da Fronteira Estocástica Stochastic Frontier Analysis SFA é um método paramétrico e estocástico Silva 2014 explica que a SFA estima a função fronteira ideal de produção ou de custo utilizandose regressão de múltiplas variáveis a partir de um conjunto de dados das DMUs na qual os insumos são colocados como variáveis independentes e o produto ou indicador que sintetiza vários produtos é colocado como variável independente Quadro 1 Principais características da DEA e da SFA DEA SFA Não paramétrica Não paramétrica Determinística Estocástica Não realiza suposições na distribuição do termo da ineficiência Realiza suposições na distribuição do termo da ineficiência Não inclui o erro como termo Inclui um termo composto do erro Não exige a especificação de uma função Exige a especificação de uma função Método Programação Linear Método Econométrico Fonte FALCÃO CORREIA 2012 p133 33 Coelli et al 2003 explicam que a seleção da melhor metodologia a ser utilizada em uma determinada aplicação pode ser influenciado por uma série de fatores como a disponibilidade de dados a importância do ruído dos dados o uso pretendido para os resultados A SFA tenta explicar o ruído enquanto a DEA assume que ele não existe Segundo Coelli et al 2003 se o pesquisador suspeitar que seus dados são de baixa qualidade poderá decidir se afastar da técnica DEA Segundo Coelli et al 2003 o método paramétrico da SFA reside na possibilidade de realizar inferências nas variáveis introduzidas tendo como objetivo estimar uma função para a fronteira eficiente O método DEA é uma técnica que avalia o posicionamento competitivo de organizações pois cria um ranking entre as empresas buscando identificar entre elas quais são eficientes e quais se mostram ineficientes Na DEA de acordo com Falcão e Correia 2012 quanto menor a amostra melhores serão as interpretações As duas técnicas apresentadas são metodologias distintas para construir fronteiras de eficiência uma considera técnicas de programação matemática e a outra usa técnicas econométricas A escolha entre a DEA e a SFA deverá ser feita pelo pesquisador considerando os dados disponíveis e os objetivos pretendidos por ele 25 O viés da DEA e a técnica de Bootstrap A DEA utiliza uma abordagem de Programação Linear para a estimação da fronteira Essa técnica não inclui o ruído ou erro aleatório pois não tem uma abordagem estatística Assaf e Matawie 2014 explicam que a DEA não é capaz de determinar a precisão das estimativas de eficiência ou fornecer uma base estatística para a fronteira estimada O principal problema dos estimadores DEA é que em tais estatísticas os estimadores de eficiência são avaliados em pontos de dados aleatórios retirados de uma população não observada em que não conhecemos o processo de geração de dados Conforme demonstrado por Wilson e Simar 1998 os teoremas usuais do limite central não são aplicáveis a menos que a dimensão do problema número de entradas e saídas seja pequeno Ou seja o problema vem do viés das estimativas da eficiência individual e o fato de que esse viés não desaparece exceto em casos envolvendo pequenas dimensões 34 A ferramenta de bootstrap foi introduzida na técnica DEA pelo trabalho de Simar e Wilson 1998 que forneceu uma metodologia geral de bootstrap em modelos de fronteiras não estocásticas Essa ferramenta foi introduzida por Efron 1979 e fornece base para a analisar a sensibilidade dos escores de eficiência à variação da amostragem Conforme a explicação de Bogetoft e Otto 2011 a ideia básica do bootstrap é por meio de reamostragem simular repetidamente o processo de geração de dados criando amostras do mesmo tamanho que a amostra original e a partir daí calcular as estatísticas necessárias chamadas de réplicas Esse processo gera uma amostra de réplicas que pode ser utilizada para tirar conclusões sobre a distribuição estatística em que se está interessado Neste caso estamos interessados na distribuição estatística dos estimadores de eficiência DEA O propósito desta abordagem segundo Assaf e Matawie 2014 é estimar a distribuição populacional dos escores de eficiência DEA fazendo ser possível realizar testes de hipóteses nas pontuações de eficiência Tziogkidis 2012 explicam que o estimador DEA possui um viés de amostragem que está associado ao fato de que a amostra observada é extraída de uma população subjacente não observada e que as pontuações de eficiência das DMUs na amostra dependem das DMUs que definem a fronteira Assim as pontuações de eficiência da DEA são superestimadas em comparação com a fronteira verdadeira da população com a exceção de que as DMUs que definem a fronteira populacional estejam todas incluídas na amostra Com base na demonstração realizada por Bogetoft e Otto 2011 se não houver erros de medição então todas as observações na amostra são da tecnologia 𝑇 𝑇 Temos que a estimativa DEA é tendenciosa para cima a eficiência estimada é maior ou igual a eficiência real pois estamos otimizando numa amostra menor que a população Para eliminar o viés ele precisa ser primeiramente estimado para logo obtermos uma estimativa corrigida pelo viés O viés pode ser estimado como demostrado a seguir 𝒗𝒊é𝒔𝒌 𝑬𝜽𝒌 𝜽𝒌 20 Onde 𝜃𝑘 A eficiência real baseada no conjunto de tecnologia não conhecido T 𝜃𝑘 A eficiência DEA estimada baseada na amostra 𝑇 35 Como não conhecemos a distribuição de 𝜃𝑘 não é possível calcular o valor esperado de 𝜃𝑘 E como dito por Bogetoft e Otto 2011 é aqui que o bootstrap entra Considerando que 𝜃𝑘𝑏 é uma estimativa réplica bootstrap de 𝜃𝑘 a estimativa bootstrap do viés é dada por 𝒗𝒊é𝒔𝒌 𝟏 𝑩 𝜽𝒌𝒃 𝑩 𝒃𝟏 𝜽𝒌 𝜽𝒌 𝜽𝒌 21 Logo o estimador corrigido do viés para 𝜃𝑘 é expresso na equação a seguir 𝜽𝒌 𝜽𝒌 𝒗𝒊é𝒔𝒌 𝜽𝒌 𝜽𝒌 𝜽𝒌 𝟐𝜽𝒌 𝜽𝒌 22 A precisão das estimativas pode ser determinada com base na variância da estimativa de bootstrap 𝝈𝟐 𝟏 𝑩 𝜽𝒌𝒃 𝑩 𝒃𝟏 𝜽𝒌² 23 26 A convexidade na DEA A DEA é caracterizada por gerar um ranking de eficiência para todas as unidades produtivas tornando possível comparar uma DMU com todas as outras que compõem a amostra A DEA também permite que uma DMU eficiente localizada na fronteira sirva como benchmark para uma outra unidade ineficiente que produza a mesma quantidade de produtos com a utilização de uma quantidade maior de insumos em comparação com a unidade eficiente ou ainda que produza uma quantidade de produtos inferior com a utilização da mesma quantidade de insumos da DMU eficiente A convexidade do conjunto de produção é fundamental para que a DEA possa cumprir todas as comparações e projeções que um ranking deve permitir A fronteira eficiente é formada por segmentos de reta que conectam todas as DMUs eficientes assim as combinações 36 convexas dos inputs e outputs de uma DMU eficiente será capaz de gerar uma outro ponto igualmente eficiente localizado na fronteira Desta forma a convexidade é necessária para que um ponto ineficiente tenha um ponto virtual na fronteira eficiente que ela seja capaz de alcançar promovendo a redução de seus inputs mantendo os outputs constantes Logo cada DMU tem uma referência de projeção que pode ser uma outra DMU eficiente ou um ponto virtual constituído por uma combinação convexa de outro ponto eficiente Dois testes desenvolvidos na segunda metade da última década apresentamse como opções para se testar a hipótese de convexidade do conjunto de produção Kneip et al 2015 mostram que os teoremas do limite central tradicionais não se aplicam para a realização de inferência de valores médios das técnicas não paramétricas DEA e FDH Free Disposal Hull FDH devido ao fato de que o viés dos escores individuais é de ordem superior à variância ou covariância Os autores desenvolveram novos teoremas do limite central para permitir a realização de inferências na DEA e na FDH O primeiro teste proposto por Kneip et al 2016 baseado nos resultados obtidos em Kneip et al 2015 testa a hipótese nula de convexidade contra a não convexidade do conjunto de produção O teste divide a amostra em duas subamostras aleatórias onde na primeira a convexidade é imposta e os retornos variáveis de escala são permitidos e na segunda subamostra a não convexidade da fronteira é permitida São comparadas a média da eficiência DEA da primeira subamostra com a média FDH da segunda Segundo Kneip et al 2016 na maioria dos trabalhos aplicados a escolha entre técnicas DEA e FDH é feita de forma arbitrária O teste permite que o investigador possa escolher entre a primeira que impõe convexidade no conjunto de produção e a segunda que não faz esta imposição O segundo teste para avaliar a convexidade do conjunto de produção é o teste de Simar e Wilson 2020 que utiliza o método de bootstrap para realizar múltiplas divisões de amostras e eliminar a sensibilidade que ocorre quando é realizada apenas uma divisão A maioria dos trabalhos empíricos que avaliam a eficiência do setor bancário brasileiro utilizam a técnica não paramétrica DEA Entretanto poucos realizam uma análise acerca da convexidade ou não do conjunto de produção Kneip et al 2016 afirmam que a habilidade de testar se o conjunto de produção é convexo ou não é crucial para realizar aplicações A flexibilidade da FDH de não impor qualquer restrição sobre o conjunto de produção gera regiões onde muitas firmas são comparáveis apenas com elas mesmas o que gera unidades que não são comparáveis com nenhuma outra mas mesmo assim são consideradas eficientes A técnica DEA é um estimador mais eficiente que o FDH mas somente se a tecnologia for convexa 37 Logo a importância de se realizar os testes está em avaliar o conjunto de produção a disposição para saber qual técnica poderá ser aplicada para avaliar a eficiência dos bancos brasileiros na atividade de intermediação financeira dado os inputs e outputs escolhidos neste trabalho 27 Testes para os retornos de escala O modelo CCR é limitado no sentido de permitir apenas retornos constantes de escala enquanto que o modelo BCC permite que sejam observados retornos crescentes e retornos decrescentes na tecnologia de produção Com o objetivo de utilizar o modelo que melhor represente o conjunto de produção os testes de Kneip et al 2016 e Simar e Wilson 2020 podem ser aplicados para testar a hipótese nula de retornos constantes de escala contra a hipótese de retornos variáveis de escala Os testes envolvem a divisão da amostra em duas subamostras onde na primeira retornos constantes de escala são impostos e na segunda retornos variáveis de escala são permitidos São comparadas as médias das eficiências DEA das duas subamostras que caso iguais apontam para a aplicação do modelo CCR e se diferentes apontam para a aplicação do modelo que permite retornos variáveis de escala o modelo BCC Os testes de retorno de escala têm o mesmo plano de fundo que os testes para a convexidade da tecnologia O teste de Kneip et al 2016 é baseado nos teoremas de limite central desenvolvidos em Kneip et al 2015 E o teste de Simar e Wilson 2020 utiliza o método de bootstrap para realizar múltiplas divisões da amostra É importante testar se os bancos na atividade de intermediação financeira possuem retornos variáveis ou constantes de escala Na década de 90 e no início do século XXI ocorreu uma forte tendência mundial no mercado financeiro de concentração das operações em um pequeno número de conglomerados financeiros em decorrência de operações de aquisições e fusões nestes mercados Faria et al 2006 Com isso é importante avaliar se estes movimentos e a concentração do setor proporciona economias de escala por essas instituições A aplicação dos testes neste trabalho permitirá averiguar se o conjunto de produção se apresenta como o esperado com retornos variáveis de escala validando estatisticamente a aplicação do modelo BCC ou se apresenta retornos constantes de escala diferentemente do esperável para a tecnologia do setor apontando para a aplicação do modelo CCR 38 28 Comparação de médias O objetivo principal deste trabalho envolve a comparação da média de grupos de bancos distintos Para alcançar este objetivo é necessário aplicar inferência estatística para realizar a comparação entre bancos tradicionais e bancos digitais Um teste de hipóteses pode nos dizer se a média entre os dois grupos de bancos são distintas estatisticamente permitindo averiguar se um grupo possui melhor desempenho que o outro Entretanto realizar inferência estatística sobre médias populacionais dos modelos DEA onde os estimadores são viesados não é uma tarefa simples É preciso lidar com os problemas derivados do viés descobertos por Kneip et al 2015 Os autores explicam que os estimadores de eficiência não paramétricos como a DEA têm viés que desaparece assintoticamente à medida que o tamanho da amostra aumenta mas apenas na taxa em que o estimador particular converge Conforme explicado por Kneip et al 2015 a menos que o número de entradas e saídas do modelo seja muito pequeno os resultados dos Teoremas do Limite Central padrões não valem para eficiência média estimados pelos estimadores não paramétricos Kneip et al 2015 para superar o problema gerado pelo viés dos estimadores individuais usaram uma média calculada sobre uma subamostra de dados O tamanho da subamostra é determinado para ajustar o viés e a variância com o objetivo de se obter uma distribuição estável e não degenerada Kneip et al 2015 Os autores desenvolveram uma Teorema do Limite Central priorizando o problema do viés da DEA que é a base dos testes amostrais desenvolvidos nos trabalhos de Kneip et al 2016 e Simar e Wilson 2020 Para realizar esta inferência o teste desenvolvido por Kneip et al 2016 será aplicado Este teste compara a eficiência média de dois grupos de DMUs distintos para testar a hipótese nula de equivalência de médias dos grupos 1 e 2 Este teste assim como os demais testes desenvolvidos em Kneip et al 2016 usa o teorema do limite central derivado de Kneip et al 2015 Existem hoje no mercado bancário brasileiro bancos de varejo de diferentes tamanhos Enquanto que ainda não existem bancos digitais de grande porte conforme a segmentação definida pelo BCB explicada no capítulo a seguir Este trabalho pretende comparar estes bancos digitais com dois grupos distintos de bancos de varejo O primeiro grupo inclui a 39 totalidade dos grupos de varejo enquanto que o segundo somente inclui as instituições com o mesmo porte dos bancos digitais Com isso pretendemos avaliar como estes novos bancos operam na atividade de intermediação financeira em comparação com a totalidade do mercado de varejo e verificar como o seu desempenho se posiciona em relação aos tradicionais de porte semelhante 29 Índice de Malmquist O Índice de Malmquist proposto originalmente por Malmquist 1953 e Caves et al 1982 permite a análise comparativa dos desempenhos de uma amostra de DMUs durante um período de tempo a partir da elaboração de um número índice Caves et al 1982 explicam que o Índice de Malmquist tem a capacidade de medir mudanças temporais de produtividade total dos fatores de produção considerando as alterações de eficiência e as mudanças tecnológicas relativas à amostra considerada Considerando que temos um período inicial 0 e um período t E0 t é a projeção dos insumos e produtos correspondentes ao período 0 na fronteira de eficiência construída no período t A alteração de eficiência AE medida a partir da modificação de desempenho no horizonte temporal considerado é dado pela equação a seguir 𝑨𝑬 𝑬𝒕 𝒕 𝑬𝟎 𝟎 24 A alteração tecnológica AT é obtida tal como expresso 𝑨𝑻 𝑬𝒕 𝟎𝑬𝟎 𝟎 𝑬𝒕 𝒕𝑬𝟎 𝒕 25 O índice Malmquist que representa a alteração da produtividade total dos fatores de produção em um determinado período de tempo 0 t é obtido pela multiplicação de AE e AT como expresso a seguir 40 𝑰𝑴 𝑨𝑻 𝑨𝑬 𝑬𝒕 𝟎𝑬𝟎 𝟎 𝑬𝒕 𝒕𝑬𝟎 𝒕 𝑿 𝑬𝒕 𝒕 𝑬𝟎 𝟎 26 𝑰𝑴 𝑬𝒕 𝟎𝑬𝟎 𝟎 𝑬𝒕 𝒕𝑬𝟎 𝒕 𝑿 𝑬𝒕 𝒕 𝑬𝟎 𝟎 𝟐 𝑬𝒕 𝟎𝑬𝒕 𝒕 𝑬𝟎 𝟎𝑬𝟎 𝒕 27 Para calcular as projeções dos períodos 0 t 0 0 t 0 t t utilizadas para se obter os índices de alteração de eficiência alteração de tecnologia e o Índice de Malmquist é utilizado o seguinte problema de programação linear onde E é o inverso de 𝑑0 𝒅𝟎 𝒑𝒙𝒒 𝒚𝒒𝟏 𝒎á𝒙𝝓𝝀 𝝓 28 sa 𝝓𝒚𝒊𝒒 𝒀𝒑𝝀𝒊 𝟎 𝒙𝒊𝒒 𝑿𝒑𝝀𝒊 𝟎 𝝀𝟏 𝝀𝒏 𝟎 29 Em que p q 𝜖 t t t1 t1 t t1 t1 t Em um modelo orientado para inputs o Índice de Malmquist entre os períodos t e t1 pode ser M 1 A produtividade aumentou M 1 A produtividade ficou constante M 1 A produtividade diminuiu Em um modelo orientado para outputs ocorre o inverso Quando o Índice é menor que a unidade produtividade diminuiu e quando o Índice é maior que unidade significa que a produtividade aumentou durante o período analisado Esta pesquisa irá utilizar as fronteiras de eficiência construídas nos extremos dos períodos analisados para mensurar as alterações de produtividade total dos fatores de produção incorridas pelos bancos no decorrer do período estudado Também serão descritas as fontes dessas alterações sendo alterações de eficiência ou alterações tecnológicas 41 3 RESULTADOS 31 Modelo proposto A eficiência das instituições bancárias será avaliada com enfoque na principal atividade exercida por elas Berger e Humphrey 1997 explicaram que os bancos exercem o papel de instituição organizadora de fundos entre poupadores e investidores fazendo com que valores poupados dos agentes superavitários cheguem aos agentes que demandam destes recursos para desenvolver seus projetos A abordagem da intermediação financeira é a mais adotada para a avaliação de desempenho dessas instituições pois é a que melhor representa as atividades que os bancos realizam De acordo com Mccaing e Stengos 2005 esta abordagem capta as atividades financeiras que possuem impacto nas economias nacionais O modelo de eficiência que será implementado neste trabalho vai de acordo com o modelo desenvolvido por Sealey e Lindley 1977 em que a instituição financeira é uma Decision Making Unit DMU que utiliza dos fundos captados e dos usos de capital e trabalho para produzir ativos remunerados Estes autores concluem que o output apropriado do ponto de vista do processo de decisão da firma financeira é o serviço provido para os agentes deficitários A partir da abordagem da intermediação financeira o modelo deste trabalho foi construído utilizando como primeiro input as captações de recursos realizadas pelos bancos que são compostas de depósitos totais depósitos à vista depósitos de poupança depósitos interfinanceiros depósitos a prazo conta de pagamento prépaga e outros depósitos recursos de aceites e emissões de títulos e por fim obrigações por empréstimos e repasses e por operações compromissadas As captações foram obtidas da conta de passivo do balanço dos bancos Para representar o capital empregado neste processo foi utilizado como input o ativo permanente ajustado que é o ativo permanente dos bancos menos o ativo imobilizado destinado para operações de arrendamento mercantil As despesas de pessoal constituem o último input do modelo obtida da demonstração de resultados dos bancos e representam o emprego de trabalho que é utilizado neste processo de transformação das captações em crédito para os agentes deficitários 42 As operações de crédito constituem o output único do modelo representando o resultado do processo descrito por Sealey e Lindley 1977 sendo o serviço fornecido pelas instituições financeiras para os agentes que demandam recursos Foram utilizadas as operações líquidas de provisões retirando assim as estimativas de perdas prováveis que o banco terá com devedores inadimplentes Tabela 3 Inputs e outputs do modelo de proposto Inputs I Captações Depósitos totais recursos de aceites e emissões de títulos obrigações por empréstimos e repasses obrigações por operações compromissadas II Trabalho Despesas de pessoal III Capital Ativo permanente imobilizado de arrendamento Output I Crédito Operações de crédito líquidas de provisão Fonte O autor 2023 A definição dos inputs nos modelos de intermediação financeira presentes na literatura pouco sofre variação São escolhidas variáveis que representem os usos de capital trabalho e depósitos arrecadados em quase todos os trabalhos A escolha realizada neste trabalho de utilizar o modelo de intermediação financeira puro sem a inclusão de outros ativos dos bancos como aplicações interfinanceiras de liquidez e títulos e valores mobiliários por exemplo se deve ao fato de que queremos avaliar os bancos no desempenho da atividade de concessão de crédito exclusivamente Isso deve ao fato de que esta atividade como foi visto é a principal exercida pelos bancos e todas as demais derivam dela e é a atividade com a maior capacidade de ter impactos na economia 43 32 A amostra A coleta de dados deste trabalho baseouse nas informações divulgadas pelo Banco Central do Brasil BCB das instituições financeiras que autoriza a funcionar e que estejam em operação O BCB disponibiliza trimestralmente relatórios constituídos de informações contábeis selecionadas destas instituições de acordo com o Padrão Contábil das Instituições Reguladas pelo Banco Central do Brasil Cosif Os relatórios podem ser obtidos através do portal IFData Apesar de as Fintechs apresentaremse no mercado brasileiro sob diversas formas de constituições instituições de pagamento sociedades de crédito direto etc neste trabalho buscouse trabalhar apenas com bancos comerciais bancos múltiplos que contenham a carteira comercial e caixas econômicas Tal escolhe se deve ao fato de que apenas estas instituições autorizadas a receber depósitos à vista do público são capazes de criar moeda escritural e possuem uma dinâmica particular de difícil comparação com outras instituições financeiras Para a construção da amostra foram selecionadas as instituições de maneira consolidada onde as instituições pertencentes a um mesmo conglomerado financeiro são apresentadas como se representassem uma única entidade Foram excluídos da amostra os bancos sem dados disponíveis bancos que não possuíam operações de crédito e por fim os bancos que não estavam em operação durante o intervalo de 3 anos definidos para a análise que compreende os anos de 2019 2020 e 2021 A amostra ficou constituída de 73 instituições bancárias sendo 9 bancos de controle público BB BRB Banrisul Banestes Banpará Banese Banco da Amazônia Banco do Nordeste do Brasil CEF 34 de controle privado nacional e 30 instituições de controle privado estrangeiro Para atingir os objetivos propostos neste trabalho os bancos foram divididos conforme a sua área de atuação podendo ser identificados 14 bancos de varejo e 5 bancos digitais Agibank C6 Bank BS2 Inter e Original Foram elaboradas matrizes de inputs e output com dados trimestrais 1º e 3º trimestres para o período compreendido entre o 1º trimestre de 2019 e 3º trimestre de 2021 último relatório trimestral disponível a época de elaboração deste trabalho A fim de tornar os dados comparáveis ao longo do período os valores foram atualizados para setembro de 2021 corrigidos pelo índice IGPDI da Fundação Getúlio Vargas FGV 44 Tabela 4 Estatísticas descritivas dos inputs e output 20192021 2019 2020 2021 1º T 3º T 1º T 3º T 1º T 3º T Captações Mín 2255 5501 8356 25012 10262 12584 Média 112122161 110902324 116304954 117124465 103370298 101347113 Median a 7164467 7513996 7617505 8614391 8010199 7767950 DP 361819196 352708625 364496190 362904728 316415706 308689288 Máx 179021948 3 170035563 6 172512485 0 170000797 6 151986689 9 155274888 6 Despesas de Pessoal Mín 546 585 608 278 250 227 Média 526739 561758 506637 456285 403756 386009 Median a 38629 41321 39697 32564 33578 30942 DP 1680637 1788337 1594519 1428698 1274718 1174353 Máx 8468064 8299113 7732324 6829318 6507198 6000818 Permanente Mín 425 167 159 141 122 114 Média 5865851 5975737 5764752 5219187 4519680 4221997 Median a 122026 127217 123876 132887 171427 179970 DP 24211560 24785450 23458973 21109324 17844795 16169585 Máx 156314940 158669017 155346711 131604007 113043317 99067117 Operações de Crédito Mín 36 6095 8472 4513 536 104 Média 51163803 51758849 53243754 50817065 46787852 46821505 Median a 2329947 2360252 2724170 2923333 2764280 3234986 DP 173005129 172699425 174591301 165276376 151067118 149474464 Máx 955053080 935523344 914826941 882294254 806209307 795546705 Fonte O autor 2023 Em R mil 45 33 Convexidade e retornos de escala A fronteira eficiente na DEA é formada por segmentos de reta que conectam todas as DMUs eficientes assim as combinações convexas desses pontos também estarão na fronteira Logo combinações convexas dos inputs e outputs de uma DMU eficiente será capaz de gerar uma outra unidade que estará na fronteira sendo igualmente eficiente A convexidade do conjunto de produção é necessária para que uma DMU ineficiente que promove a redução de seus inputs mantendo os outputs constantes seja capaz de alcançar um ponto na fronteira eficiente Com o objetivo de testar a hipótese de convexidade do conjunto de produção foram utilizados os testes de Kneip et al 2016 e Simar e Wilson 2020 implementados através do Software R Os testes realizam uma divisão aleatória da amostra em duas subamostras independentes em que são comparadas a média das estimativas de eficiência DEA da primeira subamostra onde a convexidade é imposta e retornos variáveis de escala permitidos com a média das eficiências Free Disposal Hull FDH da segunda subamostra permitindo a não convexidade da fronteira O teste de Kneip et al 2016 trabalha com apenas uma divisão aleatória da amostra enquanto que o teste de Simar e Wilson 2020 aplica o método de bootstrap permitindo o uso de múltiplas divisões de amostra Os testes foram aplicados para o conjunto de produção da amostra em todos os seus recortes dentro do período de análise Em todos os trimestres analisados a hipótese nula não foi rejeitada ao nível de 5 de significância apontando para a convexidade da amostra Tabela 5 Testes de convexidade da amostra 2019 2020 2021 1º T 3º T 1º T 3º T 1º T 3º T Kneip et al 2016 Estatística 1001 0976 0382 0149 0395 0363 pvalor 0839 0882 0595 0235 0736 0163 Simar e Wilson 2020 Estatística 0330 0347 0188 0253 0190 0271 pvalor 0301 0196 0751 0452 0712 0398 Fonte O autor 2023 46 Um modelo DEA pode ser construído considerando retornos constantes ou retornos variáveis de escala O modelo CCR desenvolvido por Charnes Cooper e Rhodes 1978 é utilizado quando a tecnologia de produção apresenta retornos constantes de escala enquanto que o modelo BCC desenvolvido por Banker Charnes e Cooper 1984 é adequado para estimar eficiências em tecnologias que apresentam retornos variáveis de escala Os testes para avaliar os retornos de escala de uma tecnologia são similares aos testes de convexidade Para testar a hipótese nula de retornos constantes de escala a amostra é subdividida em duas subamostras independentes onde é feita a comparação entre a média das eficiências DEA da primeira subamostra onde retornos constantes são impostos contra a média das eficiências DEA da segunda subamostra onde são permitidos retornos variáveis de escala O teste de Kneip et al 2016 realiza apenas uma divisão da amostra enquanto que o teste de Simar e Wilson 2020 permite múltiplas divisões da amostra utilizando o método de Bootstrap Tabela 6 Testes de retornos de escala da amostra 2019 2020 2021 1º T 3º T 1º T 3º T 1º T 3º T Kneip et al 2016 Estatística 1448 1605 1752 4055 2143 1144 pvalor 0046 0036 0033 0006 0011 0039 Simar e Wilson 2020 Estatística 0865 0650 0683 1001 0761 0715 pvalor 0012 0037 0035 0005 0009 0017 Fonte O autor 2023 Ao nível de 5 de significância em todos os trimestres do período selecionado a hipótese nula de retornos constantes de escala foi rejeita o que indica que os bancos possuem em relação a atividade de intermediação financeira uma tecnologia que apresenta retornos variáveis de escala 47 34 Desempenho do setor bancário brasileiro em 2021 Para avaliar a eficiência dos bancos comerciais brasileiros e alcançar o objetivo proposto por este trabalho de comparar o desempenho dos bancos comerciais digitais com os bancos comerciais do segmento de varejo a técnica DEA foi aplicada para medir o desempenho das organizações As estimativas foram obtidas pelo programa R com uso do pacote Benchmarking mesmo programa utilizado para realizar os testes de convexidade e retornos de escala com o uso do pacote FEAR Foram calculadas as eficiências para todo o período de análise com a técnica DEA utilizandose do modelo BCC orientado para inputs que permite retornos variáveis de escala o que vai de acordo com os resultados obtidos nos testes realizados A técnica de Bootstrap foi aplicada para estimar as eficiências sem o viés existente no estimador de eficiência da DEA Os resultados para as instituições bancárias da amostra no 3º trimestre de 2021 são apresentadas na Tabela 7 a seguir Tabela 7 Escores de eficiência dos bancos no 3º trimestre de 2021 continua Bancos 5 Eficiência 95 ABCBRASIL 04705 05499 07178 AGIBANK 04558 05322 06511 ALFA 04252 04896 05732 ANDBANK 00798 00959 01354 BANCO ARBI 01158 01380 01757 BANCO CAPITAL 05223 06318 09248 BANCO CARGILL 05165 06253 09287 BANCO CEDULA 05360 06384 09474 BANCO COOPERATIVO SICREDI 05708 06581 09332 BANCO DA AMAZONIA 07022 08028 09404 BANCO DA CHINA BRASIL 02786 03234 04221 BANCO DO ESTADO DE SERGIPE 02193 02473 02931 BANCO DO ESTADO DO PARÁ 04310 05053 05967 BANCO DO NORDESTE DO BRASIL 05228 06186 07546 BANCO FIBRA 02149 02525 03140 BANCO INBURSA 03629 04164 05260 BANCO JOHN DEERE 05427 06436 09317 BANCO KEB HANA DO BRASIL 02798 03266 04230 BANCO LUSO BRASILEIRO 04863 05560 06343 BANCO MIZUHO DO BRASIL 00809 00946 01267 BANCO MUFG BRASIL 01814 02126 02497 BANCO RABOBANK 05194 06303 09166 48 Tabela 7 Escores de eficiência dos bancos no 3º trimestre de 2021 continuação BANCO RIBEIRAO PRETO 03026 03489 04239 BANCO RODOBENS 03996 04618 05601 BANCO SEMEAR 04549 05328 06472 BANCO SUMITOMO MITSUI 02345 02838 03625 BANCO TOPÁZIO 00681 00809 01039 BANCO TRIANGULO 01615 01876 02424 BANCOOB 01327 01598 02186 BANESTES 01015 01176 01474 BANRISUL 03852 04606 05576 BB 04529 05646 07724 BMG 04217 04912 05986 BNP PARIBAS 01912 02263 02747 BOCOM 03961 04610 05457 BOFA MERRILL LYNCH 00370 00441 00533 BRADESCO 04033 04920 06452 BRB 06750 07965 09447 BTG PACTUAL 04268 05090 06681 C6 BANK 03835 04458 05159 CEF 05150 06280 09209 CAIXA GERAL 01490 01819 02287 CCB 03148 03884 05284 CITIBANK 01161 01387 01857 COMMERZBANK BRASIL 04137 04796 05596 CREDIT AGRICOLE 01190 01409 01738 CREDIT SUISSE 03106 03566 04077 CREFISA 05422 06290 09534 DAYCOVAL 05017 05885 07392 DEUTSCHE BANK 01057 01250 01517 FATOR 00446 00547 00735 GRUPO BONSUCESSO BS2 00444 00509 00599 ICBC DO BRASIL 02950 03514 04420 INDUSTRIAL DO BRASIL 03477 04012 04683 ING 00484 00579 00741 INTER 03140 03613 04162 INTESA SANPAOLO BRASIL 03468 04132 05362 ITAU 05600 07043 09502 JP MORGAN CHASE 00285 00344 00449 MERCANTIL DO BRASIL 05075 05921 06803 MODAL 00654 00747 00942 NOVO BANCO CONTINENTAL 05829 06861 08223 OMNI 04768 05593 06573 ORIGINAL 02961 03442 03922 PARANÁ BANCO 04618 05393 06533 PINE 01561 01809 0236 RENDIMENTO 01423 01603 01920 49 Tabela 7 Escores de eficiência dos bancos no 3º trimestre de 2021 conclusão SAFRA 03903 04675 05915 SANTANDER 05706 06775 09527 SCOTIABANK 01729 02049 02440 SOCIETE GENERALE 01297 01527 01858 SOFISA 03934 04636 05673 VOTORANTIM 05680 06839 08668 Fonte O autor 2023 As eficiências foram calculadas com para inputs pois o objetivo é verificar a performance dos bancos no ajuste de captações despesas de pessoal e ativo permanente para obter maior eficiência na intermediação financeira Segundo Fethi e Pasiouras 2010 a orientação para inputs é de longe a mais utilizada para estimar a eficiência dos bancos na abordagem da intermediação financeira pois é assumido que os gestores tem maior controle sobre os recursos empregados no processo do que sobre os produtos Os dados estão expostos na medida de DebreuFarrel para inputs onde o valor estimado para a eficiência é o valor pelo qual podemos multiplicar todos os insumos para produzir os mesmos valores de output crédito de maneira eficiente O escore de eficiência corrigido pelo método bootstrap é contínuo assim a probabilidade de uma DMU ter escore de eficiência igual a unidade é nula Os três bancos que se mostraram mais eficientes foram o Banco da Amazônia o Banco de Brasília BRB e o Itaú com os respectivos escores de eficiência 08028 07965 e 07043 Tabela 8 Sumário estatístico das eficiências no 3º trimestre de 2021 Eficiência Mín 00344 Média 03907 Mediana 04164 DP 02145 Máx 08028 Fonte O autor 2023 O banco com a menor eficiência apurada no período o banco JP Morgan Chase com escore de eficiência de 00344 é um banco que atua nas áreas de investment bank corporate bank asset management private banking e treasury and securities services A instituição é 50 focada em oferecer serviços financeiros para corporações governos e instituições não tendo como a concessão de crédito para seus clientes Assim o JP Morgan Chase na intermediação financeira é o banco mais ineficiente atuando no mercado brasileiro atualmente Bancos tem diferentes áreas de atuação Analisar a eficiência de todos os bancos juntos pode ser um erro porque estariam sendo agrupados juntos bancos com tipos de operações diferentes Além disso o foco principal deste trabalho é realizar a análise comparativa dos bancos digitais com os bancos de varejo tendo em vista que ambos os tipos de bancos buscam oferecer serviços financeiros para pessoas físicas em geral Tabela 9 Eficiência dos bancos digitais no 3º trimestre de 2021 Porte 5 Eficiência 95 Agibank S4 04558 05322 07178 C6 Bank S4 03835 04458 05159 BS2 S3 00444 00509 00599 Inter S3 03140 03613 04162 Original S3 02961 03442 03922 Média 03469 DP 01816 Fonte O autor 2023 Os bancos digitais como um todo operaram de maneira ineficiente durante o terceiro trimestre de 2021 com destaque para o banco BS2 que para se tornar eficiente deveria produzir a mesma quantidade de crédito com aproximadamente 5 dos insumos que utiliza atualmente A média dos bancos digitais ficou em 03469 indicando a ineficiência elevada dos bancos digitais na intermediação financeira O BCB segmenta as instituições financeiras que autoriza a funcionar conforme o porte que cada uma apresenta Como medida do porte das instituições a autoridade monetária considera a razão entre o valor da exposição total da instituição e o valor do PIB do Brasil O Segmento 1 S1 é composto pelos bancos que tenham porte igual ou superior a 10 do Produto Interno Bruto PIB O Segmento 2 S2 compreende as instituições com porte inferior a 10 e igual ou superior a 1 do PIB As instituições com porte inferior a 1 do PIB e igual ou superior a 01 do PIB integram o Segmento 3 S3 Por fim o Segmento 4 S4 é composto pelas instituições de porte inferior a 01 51 Tabela 10 Segmentação das instituições financeiras conforme o BCB Segmento Porte S1 10 do PIB S2 1 e 10 do PIB S3 01 e 1 do PIB S4 01 do PIB Fonte O autor 2023 Os bancos que apresentaram os maiores escores de eficiência entre os bancos digitais foram os bancos Agibank e C6bank bancos atualmente enquadrados segundo seu porte no segmento S4 Os bancos digitais que possuem porte um segmento acima dos dois primeiros Inter BS2 Original apresentam escores de eficiência menores Tabela 11 Eficiência dos bancos de varejo no 3º trimestre de 2021 Bancos Segmento Controle 5 Eficiência 95 BB S1 Público 04529 05646 07724 Bradesco S1 Privado Nacional 04033 04920 06452 CEF S1 Público 05150 06280 09209 Itaú S1 Privado Nacional 05600 07043 09502 Santander S1 Privado Estrangeiro 05706 06775 09527 Banco do Nordeste S2 Público 05228 06186 07546 Banrisul S2 Público 03852 04606 05576 Safra S2 Privado Nacional 03903 04675 05915 Banco da Amazônia S3 Público 07022 08028 09404 Banpará S3 Público 04310 05053 05967 Banestes S3 Público 01015 01176 01474 BRB S3 Público 06750 07965 09447 Mercantil do Brasil S3 Privado Nacional 05075 05921 06803 Banese S4 Público 02193 02473 02931 Média 05482 DP 01844 Fonte O autor 2023 Os três bancos mais eficientes de toda a amostra são bancos de varejo Banco da Amazônia BRB e Itaú Os dois primeiros são bancos de varejo regionais de controle público enquanto o Itaú é um banco de varejo nacional com controle privado nacional Os bancos de 52 varejo apresentaram média de 05482 de eficiência o que indica que para se tornarem eficientes deveriam utilizar em média 5482 dos insumos que utilizam hoje para produzir o mesmo nível de crédito Quando se considera apenas os bancos dos segmentos S1 e S2 o valor da média dos escores dos bancos de varejo aumenta para 05766 enquanto que considerando apenas os segmentos S3 e S4 a média das eficiências é de 05102 puxada para baixo pelo Banese e pelo Banestes que apresentaram as duas menores eficiências dos bancos de varejo respectivamente 02473 e 01176 34 Evolução da eficiência dos bancos digitais e bancos de varejo entre 20192021 Esta seção tem por objetivo demonstrar o que aconteceu com as posições relativas dos bancos digitais e dos bancos de varejo ao longo do período selecionado Os escores individuais obtidos pelos modelos DEA de cada ano não são passíveis de serem comparados entre si Entretanto é interessante ver o que aconteceu com a posição relativa de cada instituição do setor ao longo do triênio A tabela a seguir apresenta um sumário da eficiência dos bancos entre os anos de 2019 e 2021 Tabela 12 Sumário da eficiência dos bancos no triênio 20192021 1T 2019 3T 2019 1T 2020 3T 2020 1T 2021 3T 2021 Mín 00213 00114 00151 00128 00241 00343 Média 03142 03181 03192 03415 03657 03901 Mediana 02721 02836 02847 03006 03768 04194 DP 01959 01984 01990 01984 02025 02140 Máx 07013 06958 06809 07581 07033 07991 CV 6234 6237 6234 5809 5537 5485 Fonte O autor 2023 A partir da Tabela 12 é possível observar que o coeficiente de variação CV dos escores de eficiência dos bancos diminuiu gradativamente ao longo do período analisado O coeficiente de variação do primeiro trimestre de 2019 foi de 6234 e ao fim intervalo no terceiro trimestre de 2021 o coeficiente de variação era de 5485 apresentando uma redução de aproximadamente 8 53 Tabela 13 Eficiência dos bancos digitais no triênio 20192021 1T 2019 3T 2019 1T 2020 3T 2020 1T 2021 3T 2021 Agibank 01979 02590 02401 03415 04027 05325 C6 Bank 00213 00114 00151 01134 04898 04426 BS2 00544 0078 00540 00526 00435 00506 Inter 03309 02596 03035 02828 03484 03588 Original 01973 02163 02847 02725 03294 03412 Média 01603 01649 01795 02126 03227 03451 CV 7788 6883 7520 5787 5206 5253 Fonte O autor 2023 A partir da tabela 13 observase que o coeficiente de variação reduziu ao longo dos anos mais de 20 demonstrando que houve uma aproximação do desempenho dos bancos digitais ao longo do período Um ponto relevante é que os escores de eficiência de todos os bancos calculados pelo modelo DEABCC com orientação para inputs corrigido por bootstrap apresentaram valores dentro do intervalo de confiança de 95 em todos os períodos da amostra Tabela 14 Eficiência dos bancos de varejo no triênio 20192021 1T 2019 3T 2019 1T 2020 3T 2020 1T 2021 3 T 2021 Santander 0679286 0663284 0652354 0658722 0670965 06775 BB 06157 0506604 0497904 0502561 0460595 05646 CEF 0607291 0594171 0594228 0604319 0623342 0628 Itaú 0557846 0485505 0581137 0585735 0565432 07043 BRB 0504778 0536556 0680954 075814 0703383 07965 Safra 0468496 0414131 0485886 0558196 0511322 04675 Banrisul 0465373 0462167 0513722 0503145 0471492 04606 Bradesco 0446083 0483121 0463643 0463077 0478732 0492 Mercantil do Brasil 0397718 0327467 0373468 0431105 0595035 05921 Banpará 0370831 0357785 046786 0424338 0494943 05053 Banco do Nordeste 0338431 0349436 0395456 0360905 0446095 06186 Banese 0193963 0184841 0208032 0206879 0226816 02473 Banco da Amazônia 0182649 0208964 0281441 0397964 0568823 08028 Banestes 0074688 0091347 0080148 0078686 0107175 01176 Média 0421652 040467 0448302 0466698 0494582 0548193 CV 4049077 3866078 3640876 3635215 3142862 3364399 Fonte O autor 2023 54 35 Comparação entre bancos digitais e bancos de varejo tradicionais Com o objetivo de comparar as eficiências médias dos grupos de bancos digitais e bancos de varejo será implementado o teste desenvolvido por Kneip et al 2016 Os autores explicam que o teste para a eficiência média de dois grupos foi sugerido no trabalho de Charnes et al 1981 onde foi considerado dois grupos de escolas um recebendo um tratamento específico enquanto o outro grupo não recebia Utilizandose dos resultados obtidos em Kneip et al 2015 onde foi desenvolvido um teorema do limite central específico para realizar inferências sobre eficiências médias Kneip et al 2016 apresentaram um teste para dois grupos distintos de DMUs onde a hipótese nula de médias iguais é avaliada Foram realizados os testes considerando apenas os grupos dos bancos digitais e dos bancos de varejos Os bancos digitais que se encontram em atividade no mercado brasileiro estão classificados nos segmentos S3 e S4 enquanto que existem bancos de varejo em todos os segmentos Devido a isso dois testes serão realizados O primeiro teste considerará todos os bancos de varejo em atuação comparando a média de eficiências da totalidade dos bancos digitais com a média de eficiências da totalidade dos bancos de varejo Um segundo teste será realizado em que os bancos digitais serão comparados apenas com os bancos de varejo em atuação no mesmo segmento S3 e S4 Portanto foram feitos dois testes comparando os bancos digitais com dois grupos diferentes dos bancos de varejo com e sem os maiores bancos do setor O primeiro teste de comparação dos escores médios de eficiência dos bancos digitais com os escores médios de todos os bancos de varejo foi realizado considerando retornos variáveis de escala e também que a fronteira é a mesma para os dois grupos Os resultados obtidos neste teste estão resumidos na tabela a seguir Tabela 15 Teste 1 de equivalência de médias de Kneip et al 2016 Grupo 1 bancos digitais Grupo 2 bancos de varejo Eficiência média 0797 0872 Variância 0037 0024 Viés 0617 0229 Estatística 2343 pvalor 0019 bilateral 0010 unilateral 1 0990 unilateral 2 Fonte O autor 2023 55 Notase que as eficiências médias calculadas no teste diferem das eficiências médias calculadas e apresentadas anteriormente para este mesmo período 3º T de 2021 Isso se deve ao fato de que o teste considera apenas os dois grupos que foram selecionados da amostra primária para a estimação das eficiências enquanto que o modelo deste trabalho emprega a técnica DEA para uma amostra de 73 bancos comerciais A eficiência dos bancos de varejo apresentada é superior ao do grupo dos bancos digitais enquanto a variância é um pouco menor São apresentados pvalores distintos para os 3 testes que são realizados O primeiro teste é bilateral que tem como hipótese nula a igualdade das médias dos dois grupos Essa hipótese foi rejeita ao nível de significância de 5 indicando a diferença entre as médias de eficiência O teste seguinte é unilateral e tem como hipótese nula que a média do primeiro grupo é superior à média do segundo Esta hipótese foi rejeitada O segundo teste unilateral avalia a hipótese nula de a média do grupo 2 ser maior que a média do grupo 1 Esta hipótese foi aceita apontando para um desempenho médio superior dos bancos de varejo sobre os bancos digitais na intermediação financeira no terceiro trimestre de 2021 Tabela 16 Teste 2 de equivalência de médias de Kneip et al 2016 Grupo 1 bancos digitais Grupo 2 bancos de varejo S3 e S4 Eficiência média 0870 0906 Variância 0021 0019 Viés 0390 0282 Estatística 0398 pvalor 0690 bilateral 0345 unilateral 1 0655 unilateral 2 Fonte O autor 2023 Considerando que os bancos digitais que atuam no Brasil atualmente possuem porte menor do que 1 do PIB enquanto existem cinco bancos de varejo com porte maior ou igual a 10 foi realizada uma comparação dos escores médios de eficiência dos grupos considerando apenas aquelas instituições que se encontram nos segmentos S3 e S4 definidos pelo BCB A eficiência média estimada para o grupo 1 foi de 0870 próxima a média grupo 2 de 0906 O teste bilateral não rejeita a hipótese de igualdade dos escores médios Os resultados contrários dos testes unilaterais apontam também para essa igualdade Logo bancos digitais e bancos de varejo de médio e pequeno porte não apresentaram diferenças estatisticamente relevantes no seu desempenho médio durante o 3º trimestre de 2021 56 A segmentação do BCB utilizada para definir quais bancos são grandes e quais são pequenos é exógena ao modelo Para fazer essa divisão deveríamos olhar para a eficiência de escala dos bancos observando em qual parte da fronteira eficiente eles se situam Ou seja se apresentam retornos crescentes de escala retornos constantes ou retornos não crescentes segmentando os bancos de acordo com os dados calculados Porém uma limitação verificada neste trabalho é de que não temos disponível um pacote para o programa R que possibilite realizar testes de hipótese para testar a eficiência de escala dos grupos de bancos Portanto a segmentação do BCB de acordo com o porte das instituições foi uma alternativa de fora do modelo para classificar os bancos pelo seu tamanho A partir dos resultados apresentados depreendese que os bancos digitais em média apresentam resultados inferiores em termos de eficiência quando comparados com os bancos de varejo considerando tanto os grandes bancos quanto os bancos regionais Porém quando são considerados apenas os bancos de varejo de porte similar aos bancos digitais hoje em atuação os resultados não apontam diferenças entre o desempenho destes grupos 36 Análise da eficiência de escala As tabelas 17 e 18 mostram os resultados da eficiência de escala para os bancos digitais e para os bancos varejistas tradicionais respectivamente A eficiência de escala é obtida pela razão entre as eficiências dos modelos CCR e BCC ou seja entre as eficiências técnicas de longoprazo e de curtoprazo A ineficiência de escala é a ineficiência que advém de uma unidade produtiva estar operando fora do nível de produto com retornos constantes de escala 57 Tabela 17 Eficiência de escala dos bancos digitais no 3º trimestre de 2021 Segmento Eficiência de escala Agibank S4 07056 C6 Bank S4 06581 BS2 S3 06800 Inter S3 05644 Original S3 05029 Média 06222 DP 00853 Fonte O autor 2023 Entre os bancos digitais o que apresentou o maior nível de eficiência de escala foi o banco Agibank sendo este o que opera mais próximo da região que apresenta retornos constantes de escala Em média os bancos digitais apresentam uma eficiência de escala de 06222 Tabela 18 Eficiência de escala dos bancos de varejo no 3º trimestre de 2021 Segmento Eficiência de escala BB S1 03403 Bradesco S1 03665 CEF S1 05509 Itaú S1 03098 Santander S1 03711 Banco do Nordeste S2 06379 Banrisul S2 03729 Safra S2 03846 Banco da Amazônia S3 06617 Banpará S3 07433 Banestes S3 08064 BRB S3 04849 Mercantil do Brasil S3 05269 Banese S4 08502 Média 05291 DP 01835 Fonte O autor 2023 58 Pela Tabela 18 é possível observar que os bancos de varejo dos segmentos S3 e S4 são os que mais estão próximos de retornos constantes de escala com destaque para os bancos Banese e Banestes com eficiências de escala de 08502 e 08064 Já entre os grandes bancos varejistas presentes nos segmentos S1 e S2 observase altos níveis de ineficiência de escala com destaque para os bancos BB Bradesco Itaú e Santander em que os escores de eficiência de escala não ultrapassam o valor de 04 Os bancos varejistas médios são os que em média 06768 apresentam eficiência de escala maiores estando mais próximos de retornos constantes de escala Os bancos digitais operam em um nível abaixo de retornos constantes com eficiência de escala em média de 06222 o que significa que estes bancos podem se tornar mais eficientes em escala e se aproximar dos bancos de varejo médios Por outro lado os grandes bancos do setor apresentam em média 04167 a maior distância para os retornos constantes de escala e por serem os pontos mais afastados em relação a origem possuem retornos não crescentes de escala Os grandes bancos em comparação com os bancos digitais os mais próximos da origem possuem inclinação inferior Em resumo bancos digitais são pequenos demais e precisariam crescer para se tornarem mais eficientes em escala aproximandoos dos bancos de varejo médios A DEA resolve o problema e calcula as eficiências considerando que as DMUs são livres para escolher o seu nível de produção sem competirem entre si ou seja não leva em consideração a organização do mercado É preciso saber se existem condições de mercado para que os bancos digitais cresçam A experiência simplificada e de baixo custo permitiu aos bancos digitais crescerem de forma acelerada nos primeiros anos de vida Bancos digitais operam com uma estrutura menor pois não contam com as estruturas físicas das agências operam com um número menor de funcionários o que permite atender os clientes a um custo mais baixo No entanto pode ser que ainda falte aos bancos digitais a escala necessária para diluir sua estrutura de custo tornando se assim mais eficientes A ascensão dos bancos digitais no mercado de crédito é viabilizada pela capacidade que estes têm de reduzir as barreiras de entrada através da inovação A prática dos grandes bancos de preços elevados e rentabilidade elevada também é um incentivo a participação dos bancos digitais no mercado de crédito Não é possível prever qual será o destino da competição entre bancos tradicionais e bancos digitais no Brasil Um cenário possível é que o mercado seja quebrado em nichos atendidos por bancos tradicionais e digitais Em outro caminho os bancos de tradicionais podem não sobreviver e serem substituídos pelos novos bancos Outro cenário 59 é de os bancos tradicionais digitalizarem suas operações mantendo sua base consumidora e a liderança do mercado 37 Análise temporal do setor bancário brasileiro Com o objetivo de analisar a evolução do desempenho dos bancos comerciais atuantes no mercado brasileiro ao longo do triênio de 2019 a 2021 foi calculado o Índice de Malmquist entre os extremos do período analisado o primeiro trimestre de 2019 e o terceiro trimestre de 2021 O índice permite realizar uma análise completa da evolução da produtividade de uma DMU ao longo de um período determinado Seus componentes permitem obter informações sobre as variações de eficiências e sobre as variações da tecnologia de produção O índice foi implementado através do pacote FEAR Frontier Efficiency Analysis with R de Paul W Wilson no programa R Os intervalos de confiança foram construídos com base nos intervalos percentuais de Hall com base nas diferenças O índice foi calculado com orientação para inputs portanto resultados menores que a unidade apresentam melhoras na produtividade enquanto que valores superiores indicam pioras O intervalo de confiança não pode incluir a unidade para que se possa afirmar estatisticamente que o movimento de alta ou baixa na produtividade ocorreu Ou seja o intervalo de confiança deve estar totalmente à esquerda da unidade para que haja uma melhora na produtividade ou totalmente à direita para que haja uma queda na produtividade durante o período selecionado Tabela 19 Índice de produtividade de Malmquist para os bancos digitais Bancos 25 Malm 975 Agibank 0383 0573 0673 C6 Bank 0024 0007 0006 BS2 1028 1206 1329 Inter 1963 1902 2340 Original 0539 0713 0750 Média conclusivos 0880 Fonte O autor 2023 60 Na tabela 19 estão expressos os índices de produtividade de Malmquist para o período do primeiro trimestre de 2019 para o terceiro trimestre de 2021 No intervalo considerado apenas os bancos BS2 e Inter tiveram redução na produtividade dos seus fatores de produção entre os bancos digitais O banco C6 bank teve o melhor desempenho do período apresentando um aumento da sua produtividade na ordem de mais de 99 segundo o índice de Malmquist Os bancos digitais como um todo conforme a média do índice apresentam um avanço entre 2019 e 2021 de aproximadamente 12 Tabela 20 Índice de alterações de eficiência para os bancos digitais Bancos 25 Eff 975 Agibank 0051 0442 0561 C6 Bank 0017 0002 0002 BS2 0285 0950 1204 Inter 0893 1425 1879 Original 0125 0553 0701 Média conclusivos 0332 Fonte O autor 2023 Os bancos digitais apresentaram uma melhora relevante na eficiência técnica obtida no intervalo de análise O banco C6 bank foi o que apresentou a melhora mais acentuada na gestão de seus inputs Em relação à média considerando todos os bancos digitais nada se pode afirmar pois no índice de alterações de eficiência alguns bancos apresentam intervalos de confiança inconclusivos pois incluem a unidade Para os bancos BS2 e Inter não se podem fazer afirmações quanto a alterações na eficiência observada pois a unidade encontrase dentro dos intervalos de confiança estimados A média foi calculada tendo como base apenas os bancos com intervalos de confiança conclusivos apontando em média para uma melhora na eficiência técnica de aproximadamente 66 61 Tabela 21 Índice de alterações de tecnologia para os bancos digitais Bancos 25 Tech 975 Agibank 0911 1297 1828 C6 Bank 1052 3067 4822 BS2 0853 1270 1760 Inter 0975 1335 1833 Original 0681 1289 1798 Média conclusivos 3067 Fonte O autor 2023 A mudança da tecnologia de produção não pode ser considerada determinante para a produtividade positiva observada para os bancos digitais pelo índice de Malmquist Para quatro dos cincos bancos o intervalo de confiança inclui a unidade Para o banco C6 Bank o índice aponta que não houve progresso tecnológico no período Tabela 22 Índice de produtividade de Malmquist para os bancos de varejo Bancos 25 Malm 975 Banco da Amazônia 0205 0311 0310 Banese 1097 1094 1233 Banpará 0634 0783 0800 Banco do Nordeste 0708 0733 0753 Banestes 0835 0824 0953 Banrisul 1001 1053 1139 BB 0979 1018 1102 Bradesco 0765 0832 0863 BRB 0621 0728 0737 CEF 0963 0980 1067 Itaú 0726 0797 0879 Mercantil do Brasil 0884 0883 0927 Safra 1004 1015 1038 Santander 0761 0811 0972 Média conclusivos 0822 Fonte O autor 2023 Conforme os valores expressos na Tabela 22 percebese que os bancos de varejo tiveram um aumento na sua produtividade de em média 17 considerando apenas os bancos com intervalos de confiança conclusivos Apenas os bancos Safra Banese e Banrisul 62 apresentaram redução no seu desempenho segundo o índice de Malmquist O Banco da Amazônia que possui o maior escore de eficiência no terceiro trimestre de 2021 apresentou o maior aumento de produtividade do período de aproximadamente 70 Para BB e CEF nada se pode afirmar pois apresentam intervalos de confiança inconclusivos Tabela 23 Índice de alterações de eficiência para os bancos de varejo Bancos 25 Eff 975 Banco da Amazônia 0004 0204 0238 Banese 0393 0798 1011 Banpará 0184 0576 0678 Banco do Nordeste 0300 0583 0723 Banestes 0277 0678 0898 Banrisul 0173 0722 0886 BB 0304 0830 1075 Bradesco 0238 0682 0874 BRB 0106 0494 0586 CEF 0394 0783 1003 Itaú 0207 0670 0876 Mercantil do Brasil 0183 0578 0726 Safra 0359 0842 1085 Santander 0265 0703 0938 Média conclusivos 0589 Fonte O autor 2023 63 Tabela 24 Índice de alterações de tecnologia para os bancos de varejo Bancos 25 Tech 975 Banco da Amazônia 1031 1526 2155 Banese 0983 1371 1886 Banpará 0966 1359 1841 Banco do Nordeste 0862 1258 1667 Banestes 0803 1215 1722 Banrisul 1066 1459 2097 BB 0796 1226 1709 Bradesco 0712 1220 1689 BRB 1039 1474 2062 CEF 0857 1252 1722 Itaú 0725 1190 1674 Mercantil do Brasil 1032 1527 2167 Safra 0728 1205 1656 Santander 0714 1155 1616 Média conclusivos 1412 Fonte O autor 2023 O desmembramento do índice de Malmquist para os bancos de varejo apresenta resultados parecidos com os observados para os bancos digitais Esses bancos obtiveram um aumento relevante na eficiência técnica de em média 41 Para algumas instituições os intervalos de confiança apontam um retrocesso na tecnologia de produção Podemos afirmar que não ouve um progresso tecnológico positivo no período analisado Os resultados apresentados pelo Índice de Malmquist indicam que entre o primeiro trimestre de 2019 e o terceiro trimestre de 2021 os bancos comerciais como um todo se tornaram mais produtivos Bancos digitais e bancos tradicionais de varejo apresentaram melhoras relevantes na gestão de seus inputs aproximandose assim da fronteira eficiente Por outro lado não foram observadas mudanças tecnológicas positivas não existindo indícios de avanços tecnológicos no período Logo o aumento de produtividade observado se deve integralmente ao comportamento dos bancos no emprego de seus insumos no processo de intermediação financeira 64 CONCLUSÕES Este trabalho realizou através da técnica DEA a comparação entre os desempenhos de cinco bancos digitais Agibank C6 Bank BS2 Inter e Original com os bancos de varejo tradicionais Ainda foi realizada uma análise temporal do setor bancário brasileiro através da construção do Índice de Malmquist para o triênio de 20192021 A abordagem escolhida para avaliar os bancos foi a da intermediação financeira onde foi possível mensurar o desempenho das instituições no gerenciamento de seus inputs para a concessão de crédito Como foi visto a literatura sobre eficiência bancária internacional e nacional é vasta Muitos trabalhos avaliaram bancos de diversos países com várias abordagens diferentes Neste trabalho o modelo proposto por Sealey e Lindley onde o banco é visto como uma DMU que utiliza os fundos captados e o uso de capital e trabalho para produzir ativos remunerados neste caso operações de crédito Esta abordagem como foi percebido é a mais presente nos estudos sobre eficiência bancária pois avalia os bancos na atividade que mais pode ter impacto nas economias nacionais A DEA se mostrou como a melhor opção para o trabalho devido ao fato de não exigir que se especifique uma forma funcional para a fronteira eficiente como acontece com a SFA A utilização da técnica não paramétrica frequentemente utilizada nos trabalhos sobre eficiência bancária se mostra como a melhor opção para representar a atividade bancária visto que a literatura diverge sobre como seria a melhor forma de especificar uma função para representar a intermediação financeira pelos bancos A convexidade do conjunto de produção averiguada pelos testes de hipótese realizados validou a aplicação da DEA em comparação com outras técnicas não paramétricas Foi visto que a orientação para inputs é a mais adequada para a estimar a eficiência dos bancos pois é assumido que os gestores tem maior controle sobre recursos empregados no processo produtivo do que sobre os produtos Assim as eficiências foram calculadas com orientação para inputs sendo eles as captações dos bancos as despesas de pessoal variável proxy para o trabalho empregado e os ativos permanentes líquidos de imobilizado de arrendamento mercantil Foi utilizado o output de operações de crédito líquidas de provisão representando o produto produzido pelos bancos na atividade de intermediação financeira As eficiências foram calculadas com a correção dos escores de eficiência pelo método de bootstrap Foi apresentado que as estimativas da DEA são tendenciosas para cima apresentando valores superiores ao da eficiência real Este método permite estimar as 65 eficiências DEA sem esse viés O método de bootstrap também é empregado nos testes realizados neste trabalho sendo eles os testes de convexidade os testes de retornos de escala e os testes de comparação de médias aplicados Foi visto que a convexidade do conjunto de produção é fundamental para que a DEA possa cumprir todas as comparações e projeções que se deve esperar de um ranking de eficiência Foram aplicados os testes de Kneip et al 2015 e Simar e Wilson 2020 para testar a convexidade da amostra selecionada Os testes foram aplicados para o conjunto de produção da amostra em todos os seus recortes dentro do período de análise sendo constatado que em todos eles a amostra se apresentou convexa permitindo que a técnica DEA fosse aplicada O modelo DEA pode ser construído considerando que o conjunto de produção apresenta retornos constantes de escala com o modelo CCR ou retornos variáveis de escala com o modelo BCC Com o objetivo de testar qual modelo seria mais apropriado foram aplicados testes nos primeiros e terceiros trimestres dos anos de 2019 a 2021 A rejeição da hipótese de retornos constantes de escala em todo o período de análise indicou fortemente que os bancos brasileiros em relação a atividade de intermediação financeira possuem uma tecnologia que apresenta retornos variáveis de escala Foi observado que o setor bancário brasileiro apresenta níveis significativos de ineficiência visto que o escore médio obtido foi de 039 aproximadamente O Banco da Amazônia banco de varejo regional apresentou a maior desempenho nesta análise atemporal para o terceiro trimestre de 2021 com o escore de eficiência próximo de 080 O Banco de Brasília outro banco de varejo regional apresentou desempenho muito próximo com escore de eficiência de 079 aproximadamente O Itaú banco varejista de âmbito nacional apresentou o terceiro melhor desempenho com o escore de 070 Foi visto que entre os bancos digitais os que apresentaram os maiores escores de eficiência foram os bancos Agibank e C6bank 053 e 045 bancos enquadrados pelo BCB segundo seu porte no segmento S4 um segmento abaixo dos demais bancos digitais que apresentam escores de eficiência menores O banco BS2 apresentou escore de eficiência no valor de 005 sendo mais ineficiente deste grupo O banco Inter e o banco Original apresentaram os escores de 036 e 034 respectivamente Com o objetivo de realizar a objetivo principal deste trabalho de comparar os bancos digitais com os bancos de varejo na atividade de intermediação financeira foi aplicado o teste estatístico de Kneip et al 2016 para comparar a eficiência média de grupos produtivos distintos A primeira comparação realizada foi a do grupo dos bancos digitais com o grupo da totalidade dos bancos de varejo em atuação no setor bancário brasileiro O teste realiza o cálculo 66 de eficiência dos grupos desconsiderando o restante da amostra para depois comparálos O grupo dos bancos digitais apresentou eficiência média de 0797 enquanto que o grupo dos bancos de varejo apresentou eficiência média de 0872 O resultado do teste apontou para a diferença estatística entre os valores de eficiência com os bancos de varejo sendo considerados mais eficientes que os bancos digitais Dado que no mercado brasileiro bancos digitais são recentes e não é possível constatar um banco de grande porte entre eles diferentemente do que acontece com os bancos de varejo onde é possível encontrar bancos nos maiores segmentos S1 e S2 foi proposto a realização de uma outra comparação agora entre os bancos digitais e os bancos de varejo que se encontram no mesmo segmento em que são observados bancos digitais os segmentos S3 e S4 Neste caso os bancos digitais apresentaram eficiência média de 087 enquanto que os bancos de varejo ficaram por volta de 09 Os pvalores do teste não rejeitaram a hipótese de igualdade dos escores médios destes dois grupos Logo bancos digitais e bancos de varejo com o mesmo porte não apresentaram diferenças estatisticamente relevantes no seu desempenho médio no 3º trimestre de 2021 A partir dos resultados apresentados depreendese que os bancos digitais em média apresentam resultados inferiores em termos de eficiência quando comparados com os bancos de varejo considerando tanto os grandes bancos quanto os bancos regionais Porém quando são considerados apenas os bancos de varejo de porte similar aos bancos digitais hoje em atuação os resultados não apontam diferenças entre o desempenho destes grupos na abordagem da intermediação financeira Foi calculado o Índice de Malmquist para o triênio de 2019 a 2021 com o objetivo de analisar a evolução do desempenho dos bancos comerciais do mercado brasileiro O índice foi calculado com orientação para inputs onde resultados menores que 1 apresentam melhoras na produtividade enquanto que valores superiores a 1 indicam pioras Foi percebido que os bancos digitais como um todo conforme a média obtida pelo índice apresentam um avanço entre 2019 e 2021 de aproximadamente 12 em sua produtividade Os bancos de varejo apresentaram um aumento de produtividade de em média 18 Em relação a fonte do aumento de produtividade dos bancos digitais no período por meio da decomposição do Índice de Malmquist foi verificado que não houve ganhos de produtividade pelo progresso tecnológico no período para estes bancos enquanto que esses bancos tiveram um aumento médio de 76 em sua eficiência técnica Resultados similares foram vistos para os bancos de varejo onde não foi possível constatar um progresso tecnológico 67 no período analisado mas por outro lado foi verificado um aumento de 42 na eficiência técnica deste grupo Foi visto que os bancos comerciais como um todo se tornaram mais produtivos pela análise temporal realizada com o Índice de Malmquist Bancos digitais e bancos de varejo tradicionais apresentaram melhoras na gestão de seus inputs aproximandose assim da fronteira eficiente Por outro lado não foi possível observar indícios de progresso tecnológico durante o período Para concluir é sugerido para trabalhos futuros buscar avaliar comparativamente os bancos digitais com outros grupos de bancos com outras abordagens diferentes da abordagem da intermediação financeira utilizada neste trabalho A abordagem da produção de serviços por exemplo parece ser capaz de apresentar resultados interessantes para futuras análises porém exigirá dos pesquisadores encontrar dados que permitam realizar este estudo visto que os dados que são disponibilizados publicamente pelo BCB se referem basicamente a rubricas contábeis que podem ser difíceis de representar os serviços bancários 68 REFERÊNCIAS AHMAD N NAVEED A AHMAD S BUTT I Banking sector performance profitability and efficiency a citationbased systematic literature review Journal of Economic Surveys v 34 n 1 p 185218 2020 ARIFF M CAN L Cost and profit efficiency of Chinese banks a nonparametric analysis China Economic Review v 19 p 260273 2008 ASSAF A MATAWIE K M Improving the accuracy of DEA efficiency analysis a bootstrap application to the health care food service industry Applied Economics v 42 p 35473558 2010 AZAD A K WANKER P RAIHAN M Z ANWAR R MUSTAFA R Bank efficiency in Bangladesh revisited a slackbased network DEA approach Journal of Economic Studies v 46 2019 BANKER R D CHARNES A COOPER W W Some models for estimating technical and scale inefficiencies in data envelopment analysis Management Science v 30 p 10781092 1984 BANKER R D CUMMINS J D Performance measurement in the financial services sector frontier efficiency methodologies and other innovative techniques Journal of banking and finance v 34 p 14131416 2010 BERGENDAHL G DEA and benchmarks An application to nordic banks Annals of Operations Research v 82 p 233249 1998 BERG S A FORSUND F R HJALMARSSON L SOUMINEN M Banking efficiency in the Nordic countries Journal of Banking and Finance v 17 p 37188 1993 BERGER A N HUMPHREY D B The dominance of inefficiencies over scale and product mix economies in banking Journal of Monetary Economics v 28 p 11748 1991 BERGER A N DEYOUNG R Problem loans and cost efficiency in commercial banks Journal of Banking Finance v 21 p 849870 1997 69 BERGER A M HASAN I ZHOU M Bank ownership and efficiency in China what will happen in the worlds largest nation Journal of Banking and Finance v 33 p 113130 2009 BHATIA V BASU S MITRA S K DASH P A review of bank efficiency and productivity OPSEARCH v 55 p 557600 2018 BOGETOFT P OTTO L Benchmarking with DEA SFA and R International Series in Operations Research and Management Science v 157 2011 BONIN J P HASAN I WACHTEL P Bank performance efficiency and ownership in transition countries Journal of banking finance v 29 p 3153 2005 BOOT A HOFFMANN P LAEVEN L RATNOVSKI L Fintech whats old whats new Journal of Financial Stability forthcoming 2020 CARNEIRO M SALGADO JUNIOR A P MACORIS L S Avaliação da eficiência bancária por meio da abordagem de intermediação uma análise comparativa de instituições financeiras brasileiras Revista Eletrônica de Administração Porto Alegre v 85 p 336359 2016 Disponível em httpswwwscielobrjreadafDVKjqJyjMCSsDyGrXKjpVklang pt Acesso em 10 jun 2022 CASADO F L SOUZA A M Análise Envoltória de Dados conceitos metodologia e estudo da arte na educação superior UFSM 2015 CAVA P B SALGADO JUNIOR A P BRANCO A M F Evaluation of bank efficiency in Brazil a DEA approach Revista de Administração Mackenzie v 17 n 4 p 6284 2016 CAVES D W CHRISTENSEN L R DIEWERT W E The economic theory of index numbers of the measurement of input output and productivity Econometrica v 50 n 6 p 13931414 1982 COELLI T RAO D S P BATTESE G E An introduction to efficiency and productivity analysis Kluwer Academic Publishers 1998 COELLI T ESTACHE A PERELMAN S TRUJILLO L A primer of efficiency measurement for utilities and transport regulators World Bank Institute Development Studies 26062 2003 70 CHARNES A COOPER W W RHODES E Measuring efficiency of decision making units European Journal of Operational Research v 2 n 6 p 429444 1978 DEBREU G The coeficiente of resource utilization Econometrica v 19 p 273292 1951 EFRON B Bootstrap methods another look at the jackknife Annals of Statistics v 9 p 1 26 1979 ENGLISH M GROSSKOPF S HAYES K YAISAWARNG S Output allocative and technical efficiency of banks Journal of Banking and Finance v 17 34966 1993 FALCÃO V A CORREIA A R Eficiência portuária análise das principais metodologias para o caso dos portos brasileiros Journal of Transport Literature v 6 p 133146 2012 FARE R GROSSKOPF S LOVELL C A K Production frontiers Cambridge University Press 1994 FARIA J A de Eficiência no setor bancário brasileiro a experiência recente das fusões e aquisições 2006 126 f Dissertação Mestrado em Ciências Econômicas Faculdade de Ciências Econômicas Universidade Estadual do Rio de Janeiro Rio de Janeiro 2006 FARIA J A de PAULA L F de MARINHO A Fusões e aquisições bancárias no Brasil uma avaliação da eficiência técnica e de escala Texto para Discussão IPEA v 1233 p 134 2006 FARREL M J The measurement of productive efficiency Journal of the Royal Statistical Society v 120 p 253281 1957 FAVERO C A PAPI L Technical efficiency and scale efficiency in Italian banking sector a nonparametric approach Appl Econ v 27 p 385395 1995 FERRIER G D LOVELL C A K Measuring cost efficiency in banking econometric and linear programming evidence Journal of Econometrics v 46 p 22945 1990 FETHI M PASIOURAS F Assessing bank efficiency and performance with operational research and artificial intelligence techniques a survey European Journal of Operational Research v 204 n 2 189198 2010 71 FUKUYAMA H MATOUSEK R Modelling bank performance a network DEA approach European Journal of Operational Research v 259 p 721732 2017 KAO C LIU S T Multiperiod efficiency measurement in data envelopment analysis the case of Taiwanese commercial banks Omega v 47 p 9098 2014 KNEIP A SIMAR L WILSON P W When bias kills the variance central limit theorems for DEA and FDH efficiency scores Econometric Theory v 31 p 394422 2015 KNEIP A SIMAR L WILSON P W Testing hypotheses in nonparametric models of production Journal of Business and Economic Statistics v 34 n 3 p 435456 2016 KOOPMANS T C Activity analysis of production and allocation New York Wiley 1951 MCCAIG B STENGOS T Financial intermediation and growth some robustness results Economics Letters v 88 n 3 p 306312 2005 MALMQUIST S Index numbers and indifference surfaces Trabajos de Estatística v 4 p 209242 1953 MARINHO A Estudo de eficiência em alguns hospitais públicos e privados com a geração de rankings Texto para Discussão IPEA n 794 2001 MENDONÇA D J Relação entre eficiência e rentabilidade em instituições financeiras bancárias no Brasil 2017 95 f Dissertação Mestrado em Administração Programa de PósGraduação em Administração Universidade Federal de Lavras Minas Gerais 2017 OECD Digital disruption in banking and its impact on competition OECD Competition Committee p 150 2020 Disponível em httpswwwoecdorgcompetitiondigital disruptioninbankinganditsimpactoncompetition2020pdf Acesso em 10 jun 2022 PARADI J C HAIYAN ZHU H A survey on bank branch efficiency and performance Research with data envelopment analysis Omega v 41 p 6179 2013 PARADI J C ROUATT S ZHU H Twostage evaluation of bank branch efficiency using data envelopment analysis Omega v 39 p 99109 2011 72 PAULA L F de FARIA J A de Eficiência do setor bancário brasileiro por segmento de mercado uma avaliação recente Mesa 3 ANPEC Macroeconomia e Finanças 2008 PÉRICO A E SANTANA N B REBELATTO D A N Eficiência bancária os maiores bancos são os mais eficientes Uma análise por envoltória de dados Gest Prod São Carlos v 15 n 2 p 421431 2008 PÉRICO A E SANTANA N B REBELATTO D A N Estimating the efficiency from Brazilian banks a bootstrapped Data Envelopment Analysis DEA Production v 26 n 3 p 551561 2016 PHILIPPON T On fintech and financial inclusion BIS Working Papers v 841 2020 RAMOS R E B Michael J Farrell e a medição da eficiência técnica In ENCONTRO NACIONAL DE ENGENHARIA DE PRODUÇÃO XXVII 2007 Foz do Iguaçu Anais Foz do Iguaçu 2007 p 110 RANGAN N GRABOWSKY R ALY H Y PASURKA C The technical efficiency of US banks Economics Letters v 28 p 169175 1988 SEALEY C LINDLEY J Inputs outputs and a theory of production and cost at depository financial institutions Journal of Finance v 32 p 12511266 1977 SOUZA P O MEZA L A Uma pesquisa sobre índices de eficiência não radiais na análise envoltória de dados In SIMPÓSIO DE PESQUISA OPERACIONAL E LOGÍSTICA DA MARINHA 2006 Rio de Janeiro Anais Rio de Janeiro 2006 p 2741 SHERMAN H D GOLD F Bank branch operating efficiency evaluation with data envelopment analysis J Bank Finance v 9 n 2 p 297316 1985 SIMAR L WILSON P W Sensitivity analysis of efficiency scores how to bootstrap in nonparametric Frontier models Management Science v 44 n 1 p 4961 2020 SIMAR L WILSON P W Hyphotesis testing in nonparametric models of production using multiple sample splits Journal of Productivity Analysis v 53 p 287303 2020 STAUB R B SOUZA G S TABAK B M Evolution of bank efficiency in Brazil a DEA approach European Journal of Operation Research v 202 p 204213 2010 73 TABAK B M CAJUEIRO D O DIAS M V B The efficiency of chinese local banks a comparison of DEA and SFA BCB working papers v 346 p 115 2014 TZIOGKIDIS P Bootstrap DEA and hypothesis testing Cardiff Economics Working Papers v 18 p 121 2012 WANKE P BARROS C P Banking efficiency in Brasil Journal of International Financial Markets Institutions and Money v 28 p 5465 2014a WANKE P BARROS C P Twostage DEA an application to major Brazilian banks Expert Systems with Applications v 41 p 23372344 2014b ARTICLE IN PRESS JID EOR m5G November 29 20161015 European Journal of Operational Research 0 0 0 2016 111 Contents lists available at ScienceDirect European Journal of Operational Research journal homepage wwwelseviercomlocateejor Innovative Applications of OR Comparison of variable selection techniques for data envelopment analysis in a retail bank Juha Eskelinen Aalto University School of Business Department of Information and Service Economy PO Box 21220 0 0 076 Aalto Helsinki Finland a r t i c l e i n f o Article history Received 8 January 2015 Accepted 2 November 2016 Available online xxx Keywords Data envelopment analysis OR in banking Variable selection Productive efficiency analysis a b s t r a c t If there are too few units compared to inputs and outputs the efficiency evaluation based upon the data envelopment analysis suffers from a lack of discrimination The literature has proposed various statistical techniques when the value judgments do not guide the selection of the inputsoutputs Two techniques the variable reduction procedure of Jenkins and Anderson 2003 and the approach based upon the ef ficiency contribution measure of Pastor Ruiz and Sirvent 2002 were compared in an empirical retail bank context The objective was to select a representative set of outputs from the services the bank provides As the techniques take different approaches to selecting influential variables the output sets proposed by the techniques diverged This created some significant differences in the efficiency evalua tions of the bank branches The bank management gave feedback on the techniques and the results from a practical perspective The techniques led to different managerial interpretations of the performance complementing each other Thus the techniques can be utilized to evaluate the units from multiple per spectives 2016 Elsevier BV All rights reserved 1 Introduction Data envelopment analysis DEA is a method used to evaluate the efficiencies of comparable units with multiple inputs andor outputs It takes an optimistic perspective by choosing for each decisionmaking unit DMU the most beneficial nonnegative ie those weights will maximize its efficiency score This virtue has a consequence Having a low number of units under assessment compared to inputs and outputs leads to finding a large portion of the units efficient or nearly efficient The identification of many units falsely efficient and too optimistic efficiency scores is called a lack of discrimination see eg Podinovski Thanassoulis 2007 Such an efficiency estimation may not serve the purpose of the evaluation Banks consume several resources inputs to produce several services or other outcomes outputs Even though there are large nationwide and international banks the majority of the branch networks are small with a relatively low number of branches 1 In Corresponding author Email addresses juhapeskelinenaaltofi juhaeskelinenmelkiorfi 1 In the United States the commercial banks had on average 132 branches in 2011 data source US Federal Deposit Insurance Corporation In the United King dom the monetary financial institutions have 287 in Germany 189 in France 318 and in Finland 38 branches data source European Central Bank 2011 UK data from 2010 The exceptionally low figure in Finland is explained by the high proportion of small local cooperative and savings banks addition in the large banks the branches are often clustered into smaller homogeneous subsets Thus the lack of discrimination can be an issue in bank branch efficiency evaluations see eg the dis cussion of Paradi Zhu 2013 This essay focuses on the variable side ie inputs and outputs of the discrimination problem The number of variables can be re duced by selecting the most important outputsinputs to depict the activity Numerous statistical techniques have been proposed for situations where judgmental knowledge is not available or not sufficient enough for selecting the variables The following review illustrates the variety of techniques 2 Some selective techniques omit variables prior to DEA Lewin Morey and Cook 1982 among others used correlation and re gression analysis to test redundant variables to be omitted Jenkins and Anderson 2003 proposed a variable reduction procedure based on partial covariance abbreviated as VR in this paper where a combination of variables is selected based upon the proportion of the total variance retained GonzalezBravo 2007 2 Besides selective techniques that are the focus of this paper there is another branch of statistical techniques that deals with the lack of discrimination by deter mining statistically the weights for summing up the data in a reduced set of vari ables Various techniques ending up in aggregation have been proposed by Sengupta 1990 SinuanyStern and Friedman 1998 Ueda and Hoshiai 1997 Adler and Golany 20 01 20 02 Bian 2012 and Amirteimoori Despotis and Kordrostami 2014 httpdxdoiorg101016jejor201611009 03772217 2016 Elsevier BV All rights reserved Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 2 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 ARTICLE IN PRESS JID EOR m5G November 29 20161015 proposed a procedure to help model specification by using an out putinput ratiobased analysis prior to DEA Another subbranch of selective techniques tests the impact of the retained variables on DEA results Banker 1996 proposed the use of statistical tests for different inefficiency distribution assumptions to indicate the significance of a variable in the production process ie efficiency estimates Simar and Wilson 2001 recommended bootstrapping to test whether the input or output variables are relevant Pastor Ruiz and Sirvent 2002 in troduced an efficiency contribution measure ECM to test for the significance of a variable in efficiency estimation Fanchon 2003 Ruggiero 2005 and Sharma and Yu 2015 regressed the effi ciency scores against the variables and tested the regression coeffi cient in dropping the insignificant variables Wagner and Shimshak 2007 proposed a stepwise procedure that evaluated the impact of the variable set on the average efficiency scores There are a few published studies comparing some of the sta tistical variable selection techniques see Nataraja Johnson 2011 Sirvent Ruiz Borras Pastor 2005 These comparisons are based upon Monte Carlo simulations and the efficiency estimates are compared to the known true frontier We take another approach in this paper The main interest is in the implications that the use of variable selection could have in the interpretation and acceptance of the DEA results for performance management We use data that have been gathered during the re search collaboration with a Finnish retail bank The variable selec tion has a specific contextual focus to specify a model for evaluat ing the operational efficiencies of the branches during 20072010 using DEA The techniques and their outcomes are reviewed by the branch network management We focus on two techniques VR and ECM Our prior assump tion is that both techniques could be helpful in our empirical case where the output variables represent the sales quantities of bank ing services that are highly positively correlated Both techniques arise from the critique against intuitive variable exclusion based upon high correlations between variables and present statistical approaches for variable selection see Jenkins Anderson 2003 Pastor et al 2002 3 Both techniques aim to minimize information loss when omit ting variables but they use very different definitions of informa tion loss While the objective of VR is to minimize the reduction in data variability ECM focuses on minimizing distortion in the dis tribution of efficiency scores This paper describes what kinds of consequences the different definitions can have in the case con text The techniques should be considered as tools to help the man agement when deciding on inputs and outputs for efficiency eval uation If the purpose is solely the correct classification of efficient and inefficient DMUs in a certain context and the purpose is not variable selection there may be other choices to reduce dimen sionality 4 Even though the industry is less relevant in variable selection for DEA it is worth noting that VR has been associated as an option to improve discrimination in bank and bank branch effi ciency evaluation Paradi Zhu 2013 Paradi Yang Zhu 2011 The reader may also be interested in the application of ECM in 3 Even though we focus on the outputs both techniques can be applied to inputs or both Jenkins and Anderson 2003 present an example where they apply their approach to the entire set when there are high correlations between some of the inputs and outputs 4 ECM performed quite well in most tested scenarios when compared with a number of selection and aggregationbased approaches Nataraja Johnson 2011 VR has been compared with a technique that combines principal component anal ysis PCA and DEA Adler Yazhemsky 2010 The latter outperformed VR in the simulations Even though the combination PCA and DEA reduces dimensionality it is an aggregation approach rather than a tool for guiding the selection of variables the evaluation of a Spanish bank branch network Pastor Lovell Tulkens 2006 The rest of this paper is organized as follows Section 2 presents the two selected techniques used in this study Section 3 intro duces the empirical case including the data used and the ini tial estimation without variable reduction Section 4 describes how the techniques were applied to construct alternative specifications for efficiency estimation Section 5 discusses the key findings Section 6 summarizes the management feedback to the compari son Section 7 makes concluding remarks 2 Variable selection techniques First we briefly introduce a DEA estimator that provides the baseline for this study and then we explain both techniques ap plied for variable selection in this comparison 21 DEA estimator The efficiency of a DMU can be evaluated using the generalized DEA estimator presented by Banker Charnes and Cooper 1984 We use the outputoriented multiplier form of DEA here Assume there are n DMUs each consuming m inputs and producing p out puts The inputs are denoted by x ℜ m and outputs by y ℜ p Let X ℜ m n and Y ℜ pn be the matrices of observed inputs and outputs for the DMUs DEA finds the most favorable input weights ν and output weights μ for the DMU under assessment denoted by superscript r Scalar u o is associated with the returns to scale RTS assumption If constant returns to scale CRS are assumed u o equals zero and the estimator is called CCR Charnes Cooper Rhodes 1978 If variable returns to scale VRS are assumed u o is free and the estimator is called BCC Banker et al 1984 The generalized DEA estimator is the following min νT x r u o 1 st μT y r 1 μT Y νT X 1 T u o 0 T μ ν 0 The objective function provides an efficiency score for the DMU the smaller the better because of the output orientation If the score is one and all variable weights ν and μ are strictly positive the DMU is efficient The first constraint normalizes the sum of weighted outputs to one This normalization has a convenient sideproduct because the weighted value of the output μi y r i depicts directly the importance of the particular variable in the efficiency evaluation of the DMU This weighted value is called the virtual output and it is an impor tant concept in this comparison The second constraint limits the weights in such a way that the performance of the unit is scaled to the best performance among all units 22 Variable reduction based on partial covariance As variables in the efficiency evaluation are often highly cor related an intuitive approach is to omit some of the highly cor related variables without a significant loss of information Jenkins and Anderson 2003 criticized this approach because interrela tions between variables are less obvious and cannot be determined directly from the correlation matrix They wanted to introduce a systematic statistical procedure to help make decisions for variable selection The authors have not given any specific name or abbre viation to their procedure We call it variable reduction based on partial covariance similarly to Adler and Yazhemsky 2010 VR is a multivariate statistical procedure used to find a reduced set of initial input or output variables to be included in DEA The Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 3 ARTICLE IN PRESS JID EOR m5G November 29 20161015 aim is to minimize the information loss of original data when re moving variables The indicator of the information contained by a variable is the variance around the mean The following summary of the procedure is based on Jenkins and Anderson 2003 The conditional variance of a remaining i variable when the ef fect of variable i is removed is denoted by σ ii If the variables are perfectly correlated then the variance remaining is σ ii 0 and i can be removed without any loss of information This conditioning can be extended to the sets of retained and omitted variables To illustrate the procedure we apply it to the output variables The data are normalized to zero mean and unit variance to treat all outputs equally Then the sum of variances of the normalized outputs is equal to the number of outputs p The outputs are divided into the set of variables omitted i 1 k and variables retained i k 1 p where the variables are in an arbitrary order The variance covariance matrix of outputs denoted by V can be partitioned as variancecovariance matrices of omitted variables denoted by V 11 and retained variables V 22 covariance of omitted and retained variables V 12 and vice versa V 21 The partial variancecovariance matrix of omitted variables given the retained variables is computed V 11 2 V 11 V 12 V 1 22 V 21 The trace of V 112 represents the size of the remaining variance of omitted variables after conditioning the retained variables The sum of trace measures the information loss The proportion of in formation retained can be obtained by comparing the information loss to the original sum of variance p The test of the partial covariance is repeated for all combina tions of retained and remaining variables The information losses of the combinations are monitored A reduced variable set with minimal loss of information is chosen and this variable set is used with the appropriate DEA estimator Jenkins and Anderson 2003 do not give exact criteria for choosing the set of initial variables retained The number of vari ables is externally defined to avoid the lack of discrimination Jenk ins and Anderson state that the ultimate decision to omit particu lar variables depends upon the importance of the variables to the management and that the VR procedure plays a supportive role in this decision 23 Variable selection based upon the efficiency contribution measure Even though the ECM approach is also a selective technique it differs from VR in the role of DEA VR makes the variable selection using a statistical test without any use of DEA The ECM approach uses the appropriate DEA estimator to test the marginal contribu tion of each variable to the efficiency estimation Following Pastor et al 2002 the test procedure is presented for an outputoriented radial efficiency estimation with efficiency scores 1 where 1 is efficient The ECM approach is based on two variable specifications that are called total model and reduced model The total model con tains a set of variables In the reduced model one of the input or output variables called the candidate variable denoted by z will be removed As the first step the data used in the total model is adjusted in such a way that all the DMUs become efficient Then the candidate variable is removed and DEA is executed with the re duced model DMU r s efficiency score denoted by ϕ r represents the change in the efficiency scores between the total model and the reduced model ϕ r is called the efficiency contribution measure of variable z to DMU r If ϕ r 1 then we can conclude that the candidate variable z has no influence on the efficiency evaluation of the DMU Analogously ϕ r 11 indicates that the removal of the candidate variable has a 10 impact on the efficiency score of the DMU These scores are calculated for all DMUs The relevance of the candidate variable z for the total model can be tested statistically using a binomial test The null hypothesis is that the efficiency score distributions of the original model and the reduced model do not differ significantly and the candidate variable could be eliminated The test has two parameters The first one is a parameter for tolerated change between the models de noted by ϕ Then T is the number of DMUs that exceed this toler ance The second parameter is the probability level denoted by p 0 for the maximum allowed proportion of DMUs exceeding the tol erance Sirvent et al 2005 suggest that ϕ 11 and p 0 015 are a suitable conservative choice for practical purposes These parame ters mean that the proportion of DMUs affected by more than 10 should not exceed 15 at the given significance level Otherwise the variable cannot be eliminated The use of the ECM is a stepwise process Pastor et al 2002 propose both a backward procedure variable elimination and a forward procedure variable incorporation As the backward procedure is relevant for variable reduction it will be our focus First all variables are tested The candidate variable with the low est value of T is eliminated Then a new original model with the remaining variables is built The process continues until no vari ables can be eliminated 3 The empirical case The case is an application to evaluate the operational efficien cies of the bank branches in a Finnish retail bank Helsinki OP Bank Plc 5 This is a retail bank operating in the Helsinki metropolitan area It belongs to OP Group which is a leading financial service provider in Finland The variable selection techniques are applied to a set of out puts services for operational efficiency estimation when assum ing that all outputs are initially equal Two alternative output spec ifications are defined one using the VR technique and one using the ECM technique The impacts of the use of these techniques are assessed by the following questions How discriminative are the alternative output specifications in relation to the distribution of efficiency scores and the identifi cation of efficient units What are the implications for the ranking of the units based upon the efficiency scores of the alternative output specifica tions What is the importance of each main group of services to the efficiency evaluation when alternative output specifications are used Management feedback on the techniques and the results will be also considered 31 Outputs and inputs The bank branches provide more than 50 banking products that are tracked in the banks sales system Many of them are only vari ants such as different types of credit cards The output set can be reduced to ten services by aggregating different product vari ants These services denoted by i are presented in Table 1 The unit of measurement related to each service is the number of new agreements such as new housing loans granted mutual fund in vestments sold or current accounts opened during the period The services comprise three main groups financing invest ment and daily banking In addition to housing loans and consumer loans financing services include payment protection in surance The investment services include savings in an account or investment funds either directly or as insurance savings accounts Daily banking services include services that the customers need for 5 As of April 1 2016 the banks name has been Helsinki Area Cooperative Bank Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 4 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Table 1 Initial outputs for the bank branch efficiency evaluation Notation i Service Service group 1 Housing loans Financing 2 Consumer loans Financing 3 Payment protection insurance Financing 4 Mutual funds Investment 5 Insurance savings Investment 6 Savings accounts Investment 7 Critical illness cover Investment 8 Current accounts Daily banking 9 Online banking agreements Daily banking 10 Cards Daily banking payments and other daily affairs Thus new agreements for daily services reflect well the new customer base Critical illness cover is actually insurance but sold by banking professionals It is a rel evant output of the bank branch activity but it is not significant enough to be a fourth group in this analysis The bank manage ment decided to locate it under investment services in the effi ciency evaluation The initial inputs represent the working time of employees fulltime equivalent FTE in five different job roles in the sales activity The bank had standard unit costs for different job roles These unit costs were used to aggregate the cost of the workforce used as single input The unit of measurement presented in this paper is working year FTE standardized with these unit costs 32 Data This study is based on annual data of the 25 bank branches from the years 20072010 As some of the branches were founded during the time frame there are 95 observations There are several temporal options for DEA when panel data is used see Tulkens Vanden Eeckaut 1995 We use an intertemporal approach in the variable selection meaning that all observations specify the model used in the efficiency estimation If we applied a contemporaneous or sequential approach each period would have a different specifi cation which would complicate the comparison of the techniques 6 In this study DMUs are the observations of branchyear combina tions As the variable selection techniques are based upon statistical analysis the distributions and correlations are relevant when ex plaining both the specifications and outcomes of the efficiency es timations The cost of the work force is denoted by c The out puts ie services are denoted by y i i 1 10 The descriptive statistics are presented in Table 2 The distributions of the variables are positively skewed There are more small and mediumsize branches than large branches in the network The skewness differs by service One reason for these differences is that in the panel data the distribution is affected not only by the branch size but also by the yearly fluctuation of the service The correlations between the variables are presented in Table 3 They are relatively high between the outputs and between the out puts and the input There are however some lower correlations even between services within the same group The lowest corre lation is between y 4 mutual funds and y 6 savings accounts The reason for this is in the customers behavior during the turbu lent times of 20072010 During the bull market before the finan cial crisis customers sought profits from mutual funds but as the 6 Even though we select a single set of variables for all periods changes over time can also be allowed Chen and Johnson 2010 demonstrate how a variable selection technique in their case ECM can be used to study the change of the set of relevant variables and bring new insights to the dynamics of the performance share prices dropped customers changed to less risky instruments such as fixedterm savings accounts 33 Initial sales efficiency estimation The baseline estimation utilized a sales efficiency model that consists of all ten outputs We do not utilize the logical group ing of services The branches intent was to maximize the sales of banking products with their resources rather than minimize the resources to a given demand Thus the outputoriented measure ment of efficiency will be used We apply a radial outputoriented DEA estimator 1 Additive DEA estimators based on the difference between weighted outputs and weighted inputs are sometimes used in bank branch efficiency analysis They allow negative inputoutput values which we do not have 7 The radial DEA is more appropriate for our purposes to rank the units The efficiencies will be estimated using both CRS and VRS assumptions and a test will be used to determine the appro priateness of the CRS assumption When applying the initial sales efficiency estimation with ten outputs 17 units 18 of all units are found to be efficient when CRS is assumed and 27 units 28 when VRS is assumed The av erage efficiency scores are 082 CRS and 087 VRS As this is a real application we cannot obviously say which of the units are actually correctly or incorrectly identified as efficient or inefficient As the purpose is to give help to managers in specifying the output set we look at the initial estimation from their perspec tive If they are satisfied with the outcome of the initial estimation they would accept a high proportion of efficient units compared to a reduced variable set When used in targetsetting less improve ment would be required from the branch network Some of the improvement potential may be lost and there would be a risk of picking poor benchmark DMUs from which the others have little to learn They would also accept that a unit can be determined to be efficient based upon high performance in one output even if the larger number of other outputs would have significant space for improvement If the discrimination is increased the benchmarks are fewer and they represent more likely good practices but on the other hand some relevant benchmarks may be ignored The targets based upon an efficient frontier tend to be more of a stretch If the lo cal management and employees consider the targets impossible to achieve these targets may be potentially demotivating In our case the management considered the variable selection to be justified 4 Applying the variable selection techniques This section presents how the reduced output specifications were constructed using the variable selection techniques 41 Applying VR In this study VR was applied to select one output from each service category Thus the number of retained outputs was three Appendix A presents the ranking of the 36 combinations meeting this criterion We computed the matrices of partial variancecovariance for each possible combination of retainedomitted variables and sorted the combinations according to the variance retained as proposed by Jenkins and Anderson 2003 7 See for example Portela and Thanassoulis 2007 where the outputs include the change in the number of clients and differences in the monetary values of var ious financial services These changes can obviously also be negative Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 5 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Table 2 Descriptive statistics of variables Mean Median Std Dev Skewness Kurtosis Min Max c Cost of sales force 12 0 103 6 3 0 73 0 61 42 274 y 1 Housing loans 392 5 353 227 4 0 82 0 14 92 1074 y 2 Consumer loans 286 0 244 179 7 0 95 0 25 52 802 y 3 Payment protection 157 4 153 95 9 0 73 0 01 33 432 y 4 Mutual funds 684 1 459 568 6 1 67 2 66 100 2781 y 5 Insurance savings 131 9 101 88 6 1 01 0 83 16 458 y 6 Savings accounts 918 4 836 670 7 1 26 1 26 153 3003 y 7 Critical illness cover 91 4 72 71 5 1 07 0 29 12 297 y 8 Current accounts 849 2 728 471 5 0 81 0 01 277 2291 y 19 Online banking agreements 622 7 585 339 0 0 67 0 35 172 1619 y 10 Cards 896 7 756 528 8 0 69 0 57 185 2206 n 95 For c the unit of measurement is the working year FTE Outputs y 1 y 10 represent quantities of transactions Table 3 Correlation matrix between variables C y 1 y 2 y 3 y 4 y 5 y 6 y 7 y 8 y 9 y 10 c 1 y 1 0941 1 y 2 0781 0836 1 y 3 0743 0799 0686 1 y 4 0842 0777 0720 0620 1 y 5 0833 0834 0834 0710 0771 1 y 6 0734 0742 0685 0588 0433 0711 1 y 7 0812 0761 0695 0812 0682 0707 0678 1 y 8 0846 0850 0725 0825 0593 0707 0799 0853 1 y 9 0790 0811 0780 0784 0528 0724 0852 0825 0958 1 y 10 0766 0786 0885 0733 0577 0758 0814 0788 0892 0954 1 All correlations are significant at the 0001 level The first ranked combination y 3 Payment protection insurance y 4 Mutual funds and y 9 Online banking agreements was selected for use in the comparison We call it the VR specification in this comparison The differences between the topranked combinations were rather small We will discuss the sensitivity of this selection later in 53 42 Applying the ECM We used the backward procedure of the ECM to delete superflu ous variables from the initial variable set The DEA efficiency scores used as the efficiency contribution measure are potentially affected by the RTS specification As we do not have a predefined RTS speci fication the procedure was executed both for CCR and BCC estima tors separately Thus it could be possible that the procedure ends up with a different reduced variable set As Sirvent et al 2005 suggest parameters ϕ 11 and p 0 015 were used when eliminating outputs in the stepwise process Binomial distribution B94 085 was used with a 5 level of sta tistical significance to determine if the output can be eliminated 8 However in some steps two or more variables were tied for the highest effect with the initial tolerance level In such cases the tol erance was decreased ϕ 105 ϕ 103 etc until the variable to be removed was identified The stepwise variable elimination process is presented in Appendix B for both CRS and VRS assumptions Even though there were some differences between the intermediate steps both as sumptions ended up with the same three variables after which no further variable eliminations were possible When following the ECM procedure the ideal combination for evaluating efficiency is variables y 1 Housing loans y 4 Mutual funds and y 8 Current ac counts The rest of the outputs can be considered superfluous We 8 Binomial distribution B94 085 was used because there are only n 1 DMUs that can be affected At least one of the DMUs remains efficient in the reduced model call the combination of these three variables the ECM specification in this comparison This combination of y 1 y 4 and y 8 also meets the criterion to select one variable from each service group similarly to our appli cation of the VR procedure It retains 8353 of the total variance of the initial output It is however only the 21th of the variable alternatives according to the VR procedure see Appendix A At this stage we can conclude that y 4 Mutual funds is a statis tically highly influential service from the performance perspective as it is included in the reduced variable set in both the VR and ECM variable sets 5 Findings The efficiencies of the DMUs were estimated by employing both CRS and VRS assumptions The descriptive statistics of the effi ciency scores and the number of efficient units are presented in Table 4 The scores are presented in a form where scores below one represent inefficiency High average median and minimum ef ficiency scores and low standard deviation are also indicators of possible overestimation of efficiencies The initial DEA estimation which contains all ten variables is by default the least discriminative specification If VRS was as sumed 28 of the DMUs would be considered efficient The dis crimination can clearly be improved by adopting either of the se lective techniques The scores of the VR specification are on aver age lower than the scores of the ECM specification On the other hand the ECM specification finds less DMUs efficient than the VR specification In order to examine which RTS assumption should be used in the model the nonparametric KolmogorovSmirnov KS test Banker 1993 Banker Natarajan 2011 is used The test com pares CCR and BCC efficiency score distributions The null hypoth esis is that the frontier exhibits CRS and the alternative hypoth esis is VRS The results of the KS test are presented in Table 5 The initial estimation retains the null hypothesis of CRS if the 5 Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 6 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Table 4 Descriptive statistics of efficiency scores and number of efficient DMUs under CRS and VRS assumptions Number of efficient DMUs Efficiency scores Mean Median StdDev Min Constant returns to scale CCR Initial estimation 17 082 082 013 057 VR specification 4 069 068 015 041 ECM specification 3 076 075 012 056 Variable returns to scale BCC Initial estimation 27 087 091 012 065 VR specification 16 077 076 016 042 ECM specification 13 082 082 012 057 n 95 Table 5 KolmogorovSmirnov test of the constant returns to scale specification Most extreme differences D pvalue Initial estimation 0179 0095 VR 0242 0008 ECM 0221 0019 Critical D 5 sign is 0197 Fig 1 Rankings of the DMUs by VR and ECM specifications statistical significance level is required Both the VR and ECM spec ifications reject CRS for VRS The statistical techniques do not only increase discrimination but they may increase the difference be tween the scores of RTS assumptions as less information is used Based upon the RTS test we assume VRS in the remaining part of this paper 51 Ranking differences Some significant differences occur in the rankings of individ ual DMUs between the output specifications selected by the tech niques Fig 1 plots the ranking of DMUs The figure illustrates that the evaluations based upon the two variable sets identified by dif ferent techniques are related The point in the lower left corner ranking 1 with both the ECM and VR specifications includes 11 observations that are identified as efficient using both techniques There are however many significant differences The biggest difference in rankings was related to branch B23 which according to the ECM specification is efficient but the VR Table 6 Rank correlations between DMUs VRS Spearmans rank correlation coefficient Initial estimation VR ECM Initial estimation 1 VR specification 0796 1 ECM specification 0830 0765 1 All correlations are significant at the 0001 level specification ranks it as 54th with an efficiency score 075 This is illustrated in Fig 1 The reason for the high ECMbased score was the high housing loan sales when the market peaked in 2008 However the branch was not able to utilize that opportunity in the sales of payment protection insurance to the housing loan cus tomers which would have provided high scores when the VR spec ification was used too There are several similar examples where the excellent performance of a branch becomes mediocre using the other technique or a mediocre performance becomes poor The rank correlations of the DMUs are presented in Table 6 The rank correlation also indicates that even though the rankings based upon the two techniques are significantly correlated they draw a different picture of the relative performance The differences between the ECM and VR emerge from the dif ferent aims in the variable selection VR focuses on the variance between the variables and thus emphasizes the differences in out puts between the units On the other hand the ECM tends to retain variables that cause significant changes in the efficiency scores of the units These variables that the ECM retains are not necessarily the ones with big differences between the DMUs For example y 1 Housing loans is a core service in which most branches perform relatively well thus a large number of DMUs operate close to the frontier and y 1 cannot be deleted without a significant change in the efficiency scores of a large proportion of DMUs As different techniques take different approaches to dealing with the dimensionality issue some variations in the results are natural 52 Importance of outputs in efficiency evaluations The analysis of virtual outputs proposed by Boussofiane Dyson and Thanassoulis 1991 can be insightful Virtual outputs depict the importance of a particular output for the unit under assess ment for the maximum efficiency rating Virtual outputs can be utilized to identify units with good practices that could be dissem inated Table 7 presents the descriptive statistics of virtual outputs in the initial estimation and the two alternative specifications Because a lot of attention is paid to the service groups in the case bank a question arises about how important each main ser vice group is to the efficiency evaluation We use virtual outputs to provide management with information on this matter We com pare the expected virtual outputs and the mean of the actual Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 7 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Table 7 Descriptive statistics of virtual outputs all DMUs μ1 y 1 μ2 y 2 μ3 y 3 μ4 y 4 μ5 y 4 μ6 y 6 μ7 y 6 μ8 y 8 μ9 y 9 μ10 y 10 Initial estimation Mean 0221 00533 0104 0158 00455 00811 01122 0128 00588 00400 StdDev 0284 0128 0225 0199 0091 0187 0193 0245 0156 0113 Min 0 0 0 0 0 0 0 0 0 0 Max 1 0762 0937 0922 04 4 4 0921 0839 0833 0834 0674 VR specification Mean 0197 0278 0525 StdDev 0258 0258 0287 Min 0 0 0 Max 1 1 1 ECM specification Mean 0370 0268 0362 StdDev 0379 0247 0371 Min 0 0 0 Max 1 1 1 The weights of the efficient DMUs are not unique in DEA and may vary by solver used Table 8 Mean virtual outputs by service group all DMUs Financing services Investment services Daily banking services t 2 test statistics pvalue Initial estimation 38 40 23 7 511 0028 VR specification 20 28 53 46 021 00 0 0 ECM specification 37 27 36 6 706 0041 The test is Hotellings tsquared statistics 8 The test is indicative only virtual outputs on the service group level in the set of all DMUs Even though the weights are free for an individual unit in a ser vice business like retail banking it may not be desirable for the overall efficiency evaluation to be too strongly ruled by one group of services Table 8 presents the mean virtual outputs of each service group by alternative There are clear differences between alternatives in how the service groups are weighted The virtual outputs of the ECM specification are most equally balanced between the service groups If balanced weighting of the service groups is wanted the initial estimation underweights daily banking services The VR specification underweights financing services and overweights daily services The importance of the variables among all DMUs depends on how far the observations generally are from the frontier The dis tributional anomalies affect the mean virtual outputs For exam ple positive skewness indicates that the majority of the variables observations are far away from the tails defining the frontier As the skewness differs by variable it is generally more favorable to weight variables whose observations tend to be closer to the fron tier The VR procedure is especially sensitive to the distributional differences as it is based upon the few retained variables selected prior to DEA 9 The low sum of the virtual output of financing ser vices indicates that a high performance in the representative vari able payment protection insurance is rare In addition the ECM procedure is potentially affected by extreme performances in sin gle service However the ECM procedure tends to delete outputs that have the vast majority of observations far away from the fron tier as it is favorable for the DMUs not to weight them 8 The null hypothesis is that the vector of mean virtual outputs of the groups does not deviate from the expected mean vector The expected mean is approxi mately 33 13 of each service group when VR and ECM specifications are used When an initial specification is used the expected mean for financing services and daily services is 30 310 and 40 410 for investment services These tests are only very indicative because the virtual outputs of service groups depend on each other 9 VR assumes implicitly that the variable distributions are normal when using partial covariance to select variables The issue of normality requirement for VR in empirical cases has been brought up previously by Adler and Yazhemsky 2010 An interesting comparison from the managerial perspective is how the virtual outputs reflect the cost of the sales force As the initial input data contained the work of the sales force by job role and the standard unit costs we can make an indicative split of the costs of the sales force to the service groups The cost of the team leaders is allocated to the service groups by the FTEs of the direct sales force An indicative split of the cost of the sales force is fi nancing services 41 investment services 30 and daily banking services 28 The VR specification violates this split clearly while the ECM specification weights are quite well aligned with the re source allocation of the branch network The results show that the apparent equality of the variables may lead in practice to unexpected differences in the importance of the variables and their logical groups in the efficiency evalua tion This may be against the implicit knowledge of management Thus it is worth studying the virtual outputs and inputs gen erated through the variable selection and discussing the findings with management 53 Sensitivity of the variable selection Jenkins and Anderson 2003 have pointed out that the re sults of omitting or including variables in DEA are difficult to pre dict even if the variables are highly correlated The examples that the authors present show that one should be cautious and study the differences between the candidate variable sets We checked the extent to which the top four ranked output combinations see Appendix A provide different results These combinations are compared in Table 9 Interestingly the 1st output combination that we have used as the VR specification in the comparison does not provide the high est discrimination It identifies more efficient DMUs than the other three selections The 4th output combination has both lower aver age scores and higher variance between the scores However the rank correlations indicate that the four options give fairly similar rankings to DMUs The rank correlations of the 1st 2nd and 4th output combinations with the ECM specification are very similar rank correlations 07090764 but the rankings of the 3rd com bination are closer to the ECM rankings rank correlation 0877 Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 8 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Table 9 Comparison of four top output combinations according to the VR procedure Output combination Number of efficient DMUs Average Median Std Dev Min Spearman s rank correlation with 1st selection 1st y 3 y 4 y 9 16 077 076 016 042 1 2nd y 2 y 4 y 9 12 075 075 015 042 0889 3rd y 2 y 4 y 8 13 079 077 013 046 0850 4th y 3 y 4 y 10 13 074 072 018 038 0926 All rank correlations are significant at the 0001 level Thus we can conclude that the 2nd and 4th output combinations take a fairly similar perspective of efficiency to the 1st output combination and the change of one variable does not have a sig nificant impact in our case As described earlier in 42 the results of the ECM technique could potentially be impacted by the RTS assumption However both RTS assumptions converged to the same final variable sets The forward selection procedure of the ECM was also tested The single variable y 4 Mutual funds was considered to be essential and other variables were tested and added to the model in a step wise process following the steps described by Pastor et al 2002 The forward procedure also converged to the same the variable set y 1 y 4 and y 8 Thus in our case the ECM technique ends up with a single variable set regardless of the RTS assumption or the form of procedure 6 Management feedback The management of the branch network reviewed the main results of the initial efficiency estimation and the alternative se lection techniques The managers were provided with a graphi cal presentation of the efficiency scores of branches Because each branch had several yearly observations from 20072010 averages of efficiency scores were used The branchlevel results were more convenient for the managers This however reduced the visibil ity of some extreme observed performances and yearly perfor mance fluctuation The management also reviewed the virtual out puts of the service groups and compared them to the costs of different job roles that were attached to these groups In addi tion the fundamentals of each technique were discussed with the management The management stated that even though it preferred the ap proach of evaluating efficiency using an aggregation of variables based upon the revenue value of services an operational efficiency evaluation could be a supplementary analysis 10 If a wide data set is gathered then it would attempt to use all that information However the management found it hard to accept that a branch could claim itself to be efficient only on the basis of any individ ual service that the bank provides as would be possible with the initial variable set Thus variable selection is justified The managers commented that the principle of using the out put set with the largest overall differences between the observa tions was reasonable ie variance retained In addition the set of variables selected by VR was justified as the management had put emphasis on these services during the years under evaluation However the disadvantage in this case was that the results were impacted by the strongly overweighted daily services The principle of eliminating the least significant outputs accord ing to their contribution to efficiency was considered attractive by the management The output specification obtained from the ECM approach contained the core services with which the branches are constantly engaged The virtual outputs of the services group were 10 Eskelinen and Kuosmanen 2013 and Eskelinen Halme and Kallio 2014 present applications using the revenue valuebased variable aggregation for DEA at the case bank also balanced and aligned with the costs Thus the management favored the ECM over VR if only one reduced set of variables is to be selected 7 Concluding remarks It has been well known that the DEA results are sensitive to variable selection Even in a case where the variables are highly correlated the choice of the variables can affect the results of the efficiency evaluation significantly Dyson et al 2001 This study has demonstrated that the implications may concern not only the efficiency scores and rankings of individual DMUs but also the im portance that the variables and their logical groups have in the ef ficiency evaluation Thus it may affect the conclusions drawn from the results regarding the entire branch network On the other hand this case study also suggests that some con tradictions are natural The techniques compared represented two strategies of variable selection VR emphasized the combination of services with the largest performance differences between the units while the ECM ended up with a set of core outputs in which a high number of units perform relatively well The management found that the techniques resulted in different interpretations of the efficiency measurement The fundamental differences between the selection approaches can benefit decisionmaking in companies and public organiza tions Cooper Seiford and Tone 2007 p 166 state that If we cannot identify the characteristics of the production frontiers by preliminary surveys it may be risky to rely on only one particu lar model When the results from different selection techniques are uniform they confirm the basis for decisions Contradictory re sults regarding an important issue indicate uncertainty and further study may be needed before the decision Both techniques have benefits and drawbacks VR has been criticized because variance retained has little value to managers Pastor et al 2006 In our case the management appreciated the principle of VR and the reduced variable set made sense to them Thus it might be insightful to consider which set of outputs con tains the biggest differences and to discuss whether these differ ences are relevant from the performance improvement point of view The unpredictability of VR has been highlighted by Jenkins and Anderson 2003 The results are sensitive to extreme per formances skewness and also noise which Adler and Yazhemsky 2010 have recognized in their experiments In our case the sen sitivity to anomalies caused an unbalanced weighting of services contradicting the prior knowledge of management The ECMs criterion for variable selection contribution to effi ciency measurement was straightforward to the management and thus it was easily acceptable in our case As the technique avoids elimination with a large overall impact on the efficiency scores it seems to emphasize existing strengths in the network rather than areas where the units have the biggest opportunities to stretch their capabilities This study complements the previous simulationbased compar ative studies of variable selection by employing an empirical case The author hopes that this article can also be helpful to the prac titioners of DEAbased performance management Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 9 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Table A1 Proportion of variance retained three output combinations Ranking of combination Combinations of three output variables Proportion of variance retained 1 y 3 y 4 y 9 87 10 2 y 2 y 4 y 9 86 98 3 y 2 y 4 y 8 86 92 4 y 3 y 4 y 10 86 35 5 y 3 y 5 y 9 85 83 6 y 1 y 4 y 9 85 78 7 y 1 y 4 y 10 85 14 8 y 2 y 7 y 9 84 92 9 y 1 y 5 y 9 84 88 10 y 1 y 7 y 9 84 83 11 y 3 y 5 y 10 84 70 12 y 2 y 5 y 8 84 61 13 y 2 y 6 y 8 84 53 14 y 1 y 7 y 10 84 46 15 y 2 y 5 y 9 84 31 16 y 3 y 5 y 8 84 24 17 y 2 y 7 y 8 83 90 18 y 1 y 6 y 9 83 87 19 y 1 y 6 y 10 83 77 20 y 3 y 4 y 8 83 70 21 y 1 y 4 y 8 83 53 22 y 1 y 5 y 10 83 52 23 y 1 y 5 y 8 83 43 24 y 2 y 7 y 10 83 43 25 y 2 y 4 y 10 83 38 26 y 1 y 6 y 8 83 16 27 y 2 y 6 y 9 82 54 28 y 1 y 7 y 8 82 39 29 y 3 y 6 y 10 81 85 30 y 2 y 5 y 10 81 49 31 y 3 y 7 y 10 80 96 32 y 3 y 7 y 9 80 20 33 y 3 y 6 y 9 80 04 34 y 2 y 6 y 10 79 68 35 y 3 y 6 y 8 79 40 36 y 3 y 7 y 8 77 29 Acknowledgments The author would like to thank Jussi Huttunen and Marko He lin from Helsinki Area Cooperative Bank for the cooperation that has made this paper possible He extends his thanks to Pekka Korhonen Aalto University C A Knox Lovell University of Queensland Maria C A S Portela Portuguese Catholic Univer sity Andy L Johnson Texas AM and two anonymous referees for their suggestions for improving this paper Appendix A Proportion of variance retained according to VR Table A1 presents the three output combinations between ser vice groups The combinations have been ranked by the proportion of variance retained according to the variable reduction procedure of Jenkins and Anderson 2003 Appendix B ECMbased backward elimination procedure by step Table B1 presents the steps of the ECMbased backward elim ination of variables for both the constant and variable returns to scale assumptions The critical value for binomial test B94085 at the 5 statistical significance level is 19 The null hypothesis states that the distributions of the efficiency scores in the original model ie the model from the previous step and the reduced model are statistically indistinguishable Thus if T 19 at tolerance level ϕ 11 the null hypothesis is rejected and the output cannot be elimi nated If there are several outputs that could be removed the out put with the lowest T is eliminated If necessary the tolerance level is reduced in order to identify the variable to be eliminated Table B1 ECMbased backward elimination procedure by step T Constant returns to scale T Variable returns to scale ϕ 11 ϕ 105 ϕ 11 ϕ 105 Step 1 All 10 outputs delete y 1 18 28 delete y 1 10 18 delete y 2 0 0 delete y 2 0 1 delete y 3 1 6 delete y 3 0 6 delete y 4 11 26 delete y 4 14 27 delete y 5 0 2 delete y 5 0 3 delete y 6 1 2 delete y 6 0 2 delete y 7 6 18 delete y 7 4 13 delete y 8 7 12 delete y 8 3 9 delete y 9 0 1 delete y 9 0 1 delete y 10 0 2 delete y 10 0 0 Reduce to 9 outputs eliminate y 2 Reduce to 9 outputs eliminate y 10 ϕ 11 ϕ 105 ϕ 11 ϕ 105 Step 2 9 outputs delete y 1 20 30 delete y 1 10 18 delete y 3 1 7 delete y 2 0 3 delete y 4 13 28 delete y 3 0 6 delete y 5 1 2 delete y 4 16 27 delete y 6 1 4 delete y 5 0 3 delete y 7 6 18 delete y 6 0 4 delete y 8 7 12 delete y 7 5 14 delete y 9 0 1 delete y 8 3 9 delete y 10 0 4 delete y 9 0 1 Reduce to 8 outputs eliminate y 9 Reduce to 8 outputs eliminate y 9 continued on next page Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 10 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Table B1 continued T Constant returns to scale T Variable returns to scale ϕ 11 ϕ 105 ϕ 11 ϕ 105 ϕ 103 ϕ 101 Step 3 8 outputs delete y 1 20 31 delete y 1 10 18 25 35 delete y 3 1 5 delete y 2 0 3 3 8 delete y 4 13 28 delete y 3 0 6 8 13 delete y 5 1 2 delete y 4 16 27 34 42 delete y 6 1 4 delete y 5 0 3 3 6 delete y 7 6 17 delete y 6 3 7 9 14 delete y 8 9 14 delete y 7 6 14 17 23 delete y 9 2 1 delete y 8 8 13 15 19 Reduce to 7 outputs eliminate y 5 Reduce to 7 outputs eliminate y 5 ϕ 11 ϕ 11 ϕ 105 Step 4 7 outputs delete y 1 19 delete y 1 11 20 delete y 3 3 delete y 2 1 4 delete y 4 24 delete y 3 1 6 delete y 6 4 delete y 4 17 30 delete y 7 6 delete y 6 6 9 delete y 8 10 delete y 7 7 16 delete y 9 1 delete y 8 9 13 Reduce to 6 outputs eliminate y 9 Reduce to 6 outputs eliminate y 2 ϕ 11 ϕ 11 Step 5 6 outputs delete y 1 22 delete y 1 15 delete y 3 2 delete y 3 2 delete y 4 32 delete y 4 24 delete y 6 6 delete y 6 6 delete y 7 6 delete y 7 6 delete y 8 15 delete y 8 10 Reduce to 5 outputs eliminate y 3 Reduce to 5 outputs eliminate y 3 ϕ 11 ϕ 11 Step 6 5 outputs delete y 1 21 delete y 1 22 delete y 4 36 delete y 4 24 delete y 6 6 delete y 6 5 delete y 7 5 delete y 7 7 delete y 8 18 delete y 8 12 Reduce to 4 outputs eliminate y 7 Reduce to 4 outputs eliminate y 6 ϕ 11 ϕ 11 Step 7 4 outputs delete y 1 27 delete y 1 21 delete y 4 31 delete y 4 23 delete y 6 9 delete y 7 10 delete y 8 26 delete y 8 16 Reduce to 3 outputs eliminate y 6 Reduce to 3 outputs eliminate y 7 ϕ 11 ϕ 11 Step 8 3 outputs delete y 1 25 delete y 1 25 delete y 4 45 delete y 4 38 delete y 8 36 delete y 8 26 End up with 3 outputs End up with 3 outputs References Adler N Golany B 2001 Evaluation of deregulated airline networks using data envelopment analysis combined with principal component analysis with an application to Western Europe European Journal of Operational Research 132 260273 Adler N Golany B 2002 Including principal component weights to improve discrimination in data envelopment analysis The Journal of the Operational Re search Society 53 985991 Adler N Yazhemsky E 2010 Improving discrimination in data envelopment analysis PCADEA or variable reduction European Journal of Operational Re search 202 273284 Amirteimoori A Despotis D K Kordrostami S 2014 Variable reduction in data envelopment analysis Optimization 63 735745 Banker R D 1993 Maximum likelihood consistency and data envelopment anal ysis A statistical foundation Management Science 39 12651273 Banker R D 1996 Hypothesis test using data envelopment analysis Journal of Productivity Analysis 7 139159 Banker R D Charnes A Cooper W W 1984 Some models for estimating tech nical and scale inefficiencies in data envelopment analysis Management Science 30 10781092 Banker R D Natarajan R 2011 Statistical tests based on DEA efficiency scores In W W Cooper L Seiford J Zhu Eds Handbook on data envelopment analy sis 2nd ed pp 273295 New York Dordrecht Heidelberg London Springer Bian Y 2012 A GramSchmidt process based approach for improving DEA dis crimination in the presence of large dimensionality of data set Expert Systems with Applications 39 685705 Boussofiane A Dyson R G Thanassoulis E 1991 Applied data envelopment analysis European Journal of Operational Research 52 115 Charnes A Cooper W W Rhodes E 1978 Measuring the efficiency of decision making units European Journal of Operational Research 2 4294 4 4 Chen WC Johnson A L 2010 The Dynamics of performance space of ma jor league baseball pitchers 18762006 Annals of Operational Research 181 287302 Cooper W W Seiford L M Tone K 2007 Data envelopment analysis A com prehensive text with models applications references and DEASolver Software 2nd ed New York Springer Dyson R G Allen R Camanho A S Podinovski V V Sarrico C S Shale E A 2001 Pitfalls and protocols in DEA European Journal of Operational Research 132 245259 Eskelinen J Halme M Kallio M 2014 Bank branch sales evaluation using ex tended value efficiency analysis European Journal of Operational Research 232 654663 Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 11 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Eskelinen J Kuosmanen T 2013 Intertemporal efficiency analysis of sales teams of a bank Stochastic seminonparametric approach Journal of Banking Finance 37 51635175 Fanchon P 2003 Variable selection for dynamic measures efficiency in the com puter industry International Advances in Economic Research 9 175188 GonzalezBravo M I 2007 PriorRatioAnalysis procedure to improve data envel opment analysis for performance measurement Journal of the Operational Re search Society 58 12141222 Jenkins L Anderson M 2003 A multivariate statistical approach to reducing the number of variables in data envelopment analysis European Journal of Op erational Research 147 5161 Lewin A Y Morey R C Cook T J 1982 Evaluating the administrative effi ciency of courts Omega 10 401411 Nataraja N R Johnson A L 2011 Guidelines for using variable selection tech niques in data envelopment analysis European Journal of Operational Research 215 662669 Paradi J C Yang Z Zhu H 2011 Assessing bank and bank branch performance Modeling considerations and approaches In W W Cooper L M Seiford J Zhu Eds Handbook of data envelopment analysis 2nd ed pp 315362 New York Dordrecht Heidelberg London Springer Paradi J C Zhu H 2013 A survey on bank branch efficiency and performance research with data envelopment analysis Omega 41 6179 Pastor J T Lovell C A K Tulkens H 2006 Evaluating the financial perfor mance of bank branches Annals of Operations Research 145 321337 Pastor J T Ruiz J L Sirvent I 2002 A statistical test for nested radial DEA models Operations Research 50 728735 Podinovski V V Thanassoulis E 2007 Improving discrimination in data envel opment analysis Some practical suggestions Journal of Productivity Analysis 28 117126 Portela M C A S Thanassoulis E 2007 Comparative efficiency analysis of Portuguese bank branches European Journal of Operational Research 177 12751288 Ruggiero J 2005 Impact assessment of input omission on DEA International Jour nal of Information Technology Decision Making 4 359368 Sengupta J K 1990 Test of Efficiency in data envelopment analysis Computers and Operations Research 17 2 121132 Sharma M J Yu S J 2015 Stepwise regression data envelopment analysis for variable reduction Applied Mathematics and Computation 253 126134 Simar L Wilson P W 2001 Testing restrictions in nonparametric efficiency models Communications in Statistics Simulation and Computation 30 159184 SinuanyStern Z Friedman L 1998 DEA and the discriminant analysis of ratios for ranking units European Journal of Operational Research 111 470478 Sirvent I Ruiz J L Borras F Pastor J T 2005 A Monte Carlo evaluation of several tests for the selection of variables in DEA models International Journal of Information Technology Decision Making 4 325343 Tulkens H Vanden Eeckaut P 1995 Nonparametric efficiency progress and regress measures for panel data Methodological aspects European Journal of Operational Research 80 474499 Ueda T Hoshiai Y 1997 Application of principal component analysis for parsi monious summarization of DEA inputs andor outputs Journal of the Operations Research Society of Japan 40 466478 Wagner J M Shimshak D G 2007 Stepwise selection of variables in data en velopment analysis Procedures and managerial perspectives European Journal of Operational Research 180 5767 Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 The efficiency of Chinese local banks a comparison of DEA and SFA Benjamin Miranda Tabak Daniel Oliveira Cajueiro e Marina V B Dias January 2014 346 ISSN 15183548 CGC 00038166000105 Working Paper Series Brasília n 346 January 2014 p 115 Working Paper Series Edited by Research Department Depep Email workingpaperbcbgovbr Editor Benjamin Miranda Tabak Email benjamintabakbcbgovbr Editorial Assistant Jane Sofia Moita Email janesofiabcbgovbr Head of Research Department Eduardo José Araújo Lima Email eduardolimabcbgovbr The Banco Central do Brasil Working Papers are all evaluated in double blind referee process Reproduction is permitted only if source is stated as follows Working Paper n 346 Authorized by Carlos Hamilton Vasconcelos Araújo Deputy Governor for Economic Policy General Control of Publications Banco Central do Brasil ComunDipivCoivi SBS Quadra 3 Bloco B EdifícioSede 14º andar Caixa Postal 8670 70074900 Brasília DF Brazil Phones 55 61 34143710 and 34143565 Fax 55 61 34141898 Email editorbcbgovbr The views expressed in this work are those of the authors and do not necessarily reflect those of the Banco Central or its members Although these Working Papers often represent preliminary work citation of source is required when used or reproduced As opiniões expressas neste trabalho são exclusivamente dos autores e não refletem necessariamente a visão do Banco Central do Brasil Ainda que este artigo represente trabalho preliminar é requerida a citação da fonte mesmo quando reproduzido parcialmente Citizen Service Division Banco Central do Brasil DeatiDiate SBS Quadra 3 Bloco B EdifícioSede 2º subsolo 70074900 Brasília DF Brazil Toll Free 0800 9792345 Fax 55 61 34142553 Internet httpwwwbcbgovbrCONTACTUS The efficiency of Chinese local banks a comparison of DEA and SFA Benjamin Miranda Tabak Daniel Oliveira Cajueiro Marina V B Dias The Working Papers should not be reported as representing the views of Banco Central do Brasil The views expressed in the papers are those of the authors and do not necessarily reect those of Banco Central do Brasil Abstract This study investigates to which extent results produced by a single frontier model are reliable based on the application of data envelopment analysis and stochastic frontier approach to a sample of Chinese local banks Our findings show they do produce a consistent trend on efficiency scores over the years However rank correlations indicate they diverge with respect to individual performance di agnosis This shows that these models provide steady information on the efficiency of the banking system as a whole but they become inconsistent at individual level Keywords Bank efficiency Stochastic frontier approach Data envelopment analysis JEL Classification Numbers G21 G28 Universidade Catlica de Brasilia Departamento de Estudos e Pesquisas Banco Cen tral do Brasil Departamento de Economia Universidade de Braslia Universidade de Braslia 3 1 Introduction This paper applies both data envelopment analysis DEA and stochastic frontier approach SFA to investigate the efficiency of Chinese local banks We analyze some features related to the efficiency scores of both approaches We determine whether the methodology applied is relevant to the determination of best and worst performers and to the conclusions drawn regarding the whole market These methods are not consistent in the determination of individual performance Nevertheless they do produce a similar trend on efficiency over the years Efficiency of financial institutions can be measured by parametric or nonparametric models Both approaches have their features but few studies apply both types of method ologies to the same sample to investigate the consistency between their results Some of these studies consider the conditions Bauer et al 1998 propose to investigate this issue Fiorentino et al 2006 There are mixed results regarding the similarity of these models Some authors find that the consistency between DEA and SFA in ranking banks accord ing to their performance is low or even statistically insignificant Fiorentino et al 2006 Ferrier and Lovell 1990 On the other hand comparison of their results at industry level can point out the similarities between them Resti 1997 There are two main points we make that differ from previous literature on the com parison of DEA and SFA First we do not rule the methodologies as either consistent or inconsistent We analyze at which level their results are more reliable produce similar information and to which extent they must be considered with care Second we argue that the comparison of the behavior of efficiency along the sample period is much more relevant to determine the consistency between both models than the average levels of efficiency Parametric and nonparametric methodologies are extensively applied to measure the efficiency of financial institutions In many cases the purpose of the existing studies is to assist policy makers with relevant information such as the impact of regulation measures over the performance of these firms Lee and Chih 2013 Barth et al 2013 Gaganis and Pasiouras 2013 However it is common to apply only one frontier model either parametric or nonparametric to measure efficiency Investigating if these models are consistent has the purpose of pointing out to which extent results and conclusions obtained from the application of only one frontier method are reliable especially to assist policy decisions Studies of bank efficiency in China are mostly related to the effects of deregulation re forms over the performance of financial institutions Berger et al 2009 Sun et al 2013 Some of them as Yin et al 2013 conclude that these reforms led to an improvement on bank efficiency Our study finds stable efficiency scores over the period 20012012 which is not in line with the idea that deregulation reforms still have a positive impact on the efficiency of Chinese financial institutions 4 2 Methodology and Data Bank efficiency can be measured through basically two different types of models parametric or nonparametric The most applied parametric model is the stochastic frontier approach and the most applied nonparametric model is data envelopment analysis In this paper we apply data envelopment analysis and stochastic frontier approach to measure the efficiency of Chinese commercial banks The advantages of using DEA is that it does not require prior knowledge of either the distributional form of the inefficiency term or the production technology used in the industry The main feature of the SFA is that it accounts for random shocks Berger and Humphrey 1997 Seiford and Thrall 1990 Mester 1996 21 Data Envelopment Analysis Data envelopment analysis was first proposed by Charnes et al 1978 and was developed based on the seminal work of Farrell 1957 It uses linear programming techniques to build a nonparametric efficiency frontier of the data sample Therefore the frontier is constructed by the practices combination of inputs and outputs of the most efficient firms in the sample The linear programming problem consists of the maximization of the firms weighted outputinput ratio In this study we apply DEA to measure inputoriented outputoriented economic and allocative efficiencies We estimate the variable returns to scale model proposed by Banker et al 1984 The inputoriented model consists of the following linear programming problem minϕz ϕ sa qi Qλj 0 ϕxi Xλj 0 i1n λi λi 0 1 in which ϕ is technical inefficiency qi are the outputs for firm i Q and X are matrices of outputs and inputs for all firms respectively xi is the vector of inputs for the ith firm and λi is a vector of weights Allocative efficiency measures the quality of the combination of inputs to produce certain outputs considering input prices Sengupta 1999 Economic efficiency analyzes both technical and allocative efficiencies Bauer et al 1998 They can be measured by a cost minimization problem that determines the optimal levels of inputs to be used given their prices wi minwiχ wiχi sa j λj yj yi 0 χi λj xj 0 λj 0 2 i λi 1 3 Economic efficiency EE will be determined by the following ratio EEi fracwi χiwi xi 4 Allocative efficiency will be measured by the ratio cost efficiency over inputoriented technical efficiency TEinputs AEi fracEEiTEinputsi 5 22 Stochastic Frontier Approach The stochastic frontier approach was simultaneously developed by Meeusen and Van den Broek 1977 and Aigner et al 1977 and it estimates a parametric frontier of the best possible practices given a standard cost or profit function Both cost and profit functions are comparable to DEA economic efficiency measures since they consider the same efficiency concept Bauer et al 1998 We estimate a cost function because it is more commonly applied in the literature Berger and Mester 1997 Since SFA builds a parametric frontier it is necessary to specify the production technology and the distribution of the inefficiency term We use a translog form of the cost function since it is a flexible functional form Berger et al 2009 LozanoVivas and Pasiouras 2010 Also the inefficiency term v has a halfnormal distribution while the random error ν is normally distributed The translog cost function is written as follows lnCTw2 δ0 j δ1 lnyjit frac12 j k δjk lnyjit lnykit β1 lnw1w2it frac12 β11 lnw1w2it lnw1w2it j θj lnyjit lnw1w2it ln vit ln νit 6 in which CT is the firms total costs i and t stand for bank and time respectively In the true fixed effects model the inefficiency term vit is composed by a set of dummy variables which determine its behavior across time This study considers three outputs and two inputs Thus w1 and w2 are the two inputs used to produce the four outputs yj The normalization by the price of the last input w2 guarantees price homogeneity In this study we apply two different specifications of the stochastic frontier model the true fixed effects model from Greene 2005 and the model proposed by Battese and Coelli 1995 The true fixed effects model fits the sample better so we refer to this model when considering the parametric efficiency results1 We obtain efficiency scores from the Jondrow et al 1982 estimator 23 Data Our sample comprises an unbalanced panel with 461 yearly observations from 65 Chinese local commercial banks for the period 20012012 The data source is Bankscope Our definition of local banks includes city commercial banks and rural commercial banks Berger et al 2009 To analyze the efficiency trend over the years we create subsamples for each year and apply the DEA model for all of them For the stochastic frontier model we use the specification of panel data This paper uses the intermediation approach to define inputs and outputs for the empirical application which states that banks capture borrowed funds and use capital and labor to turn them into loans and other assets Sealey and Lindley 1977 Therefore we specify the outputs as deposits loans and liquid assets The input quantities are total interest expenses and total noninterest expenses Input prices are the ratios total interest expenses over deposits and noninterest expenses over fixed assets Table 1 shows descriptive statistics on these variables Table 1 about here 3 Empirical Results Results regarding the individual performance of firms should be looked into with care while general results related to the performance of the whole market are consistent Both data envelopment analysis and stochastic frontier approach results show that efficiency scores are roughly stable over the years which we report on Figure 1 This finding does not confirm the upward trend Yin et al 2013 observe for bank efficiency in China after 2001 From the results we report on tables 2 and 3 our scores 1The loglikelihood for the true fixed effects model is 249051 while for the Battese and Coelli 1995 model it is 6179308 7 do seem to vary over the years at first sight DEA scores seem to show a drawback in the period 20012007 with posterior efficiency gains SFA scores seem to indicate a slight improvement on efficiency over the sample period However a closer look into mean efficiency scores and their volatility which we report on Figure 1 leads to the conclusion that the performance of Chinese local banks does not show a statistically significant change over the years A possible reason for divergent results is that our sample comprises only Chinese local banks which includes only city commercial banks and rural commercial banks while Yin et al 2013 use a sample with different types of Chinese commercial banks Table 2 about here Table 3 about here From Figure 1 we observe that efficiency levels are in some periods higher according to results from the parametric methodology Nevertheless we point out that this difference is not relevant if both models provide the same conclusion over the banking systems performance Efficiency levels themselves are not enough to rule for the inconsistency between different methodologies In this case both models indicate that Chinese local banks have not experienced efficiency improvements over the period 20012012 This shows that for this empirical study DEA and SFA do provide similar conclusions with respect to the performance of the industry of Chinese local banks as a whole Figure 1 about here To analyze the consistency between individual efficiency scores produced by the two methodologies we compute Spearmans rank correlation between DEA economic effi ciency and SFA cost efficiency since they use the same efficiency concept Bauer et al 1998 We find a rank correlation of 26 which is not statistically significant at 1 level and is consistent with the findings of Ferrier and Lovell 1990 who compute the correlation of 14 2 Fiorentino et al 2006 also find a positive but low correlation between data envelopment analysis and the stochastic frontier approach between 44 and 58 One possible explanation for the difference between both models is that DEA is more sensitive to sample heterogeneity than the SFA Fiorentino et al 2006 To check if this argument is plausible we run a robustness check in which we use the DEA results of efficiency to remove efficient banks from the original sample Sample A leaving us with a sample of banks that are originally economically inefficient Sample B and a sample of banks that are originally technically inefficient Sample C We find that efficiency scores 2We compute the spearman rank correlation between scores relative to the year 2009 which is the year which comprises the greater number of observations 8 do not differ among these samples and thus heterogeneity is not the reason why DEA and SFA do not produce similar results These results are available from the authors upon request We do not repeat the same to the SFA estimation because we use the results of the true fixed effects model which already accounts for heterogeneity Greene 2005 It should be expected that DEA and SFA might produce contradictory results in some empirical applications Berger and Humphrey 1997 point out the conflict be tween parametric and nonparametric methodologies based on the fact that they have different degrees of dispersion and rank banks differently This latter conflict is exactly the inconsistency we report here From a theoretical perspective DEA and SFA are very different Data envelopment analysis considers a deterministic frontier while the stochastic frontier model considers a parametric one which incorporates states of nature For this reason Bauer et al 1998 propose some consistency conditions these models should meet in order to be sure they are providing policy makers with reliable infor mation The fact that DEA does not account for random shocks may be the source of the inconsistency between both models Fiorentino et al 2006 The implication of this inconsistency in the framework analyzed here is that individual performance should be analyzed with caution The application of only one methodology to determine best and worst performers may lead to wrong conclusions especially when rank correlation among different models are not statistically significant 4 Conclusions We apply both data envelopment analysis and the stochastic frontier model to a sample of Chinese local banks We analyze the consistency between these measures from micro and macro perspectives The majority of studies on Chinese bank efficiency apply either a parametric or a nonparametric methodology to investigate the performance of the countrys financial institutions Our findings show that these models are not consistent in the individual analysis of efficiency and results obtained at a micro level should be dealt with care However they do provide similar results regarding the behavior of average efficiency scores for the whole market over the years We argue that this behavior is more important to determine the consistency between both models than the average efficiency levels themselves We also conclude that Chinese local banks do not show improvement on performance over the period 20012012 References Aigner D Lovell C and Schmidt P 1977 Formulation and estimation of stochastic frontier production function models Journal of Econometrics 62137 9 Banker R Charnes A and Cooper W 1984 Some models for estimating technical scale inefficiencies in data envelopment analysis Management Science 3010781092 Barth J R Lin C Ma Y Seade J and Song F M 2013 Do bank regulation supervision and monitoring enhance or impede bank efficiency Journal of Banking Finance 3782879 2892 Battese G and Coelli T 1995 A model for technical inefficiency effects in a stochastic frontier production for panel data Empirical Economics 20325332 Bauer P W Berger A N Ferrier G D and Humphrey D B 1998 Consistency conditions for regulatory analysis of financial institutions A comparison of frontier efficiency methods Journal of Economics and Business 5085114 Berger A Hasan I and Zhou M 2009 Bank ownership and efficiency in china What will happen in the worlds largest nation Journal of Banking Finance 33113130 Berger A and Mester L 1997 Inside the black box what explains differences in the efficiencies of financial institutions Journal of Banking Finance 21895947 Berger A N and Humphrey D B 1997 Efficiency of financial institutions Inter national survey and directions for future research European Journal of Operational Research 98175212 Charnes A Cooper W and Rhodes E 1978 Measuring the efficiency of decision making units European Journal of Operational Research 3429444 Farrell M 1957 The measurement of productive efficiency Journal of the Royal Statistical Society 120A253281 Ferrier G and Lovell C 1990 Measuring cost efficiency in banking Econometric and linear programming evidence Journal of Econometrics 46229245 Fiorentino E Karmann A and Koetter M 2006 The cost efficiency of german banks a comparison of sfa and dea Discussion Paper 102006 Deutsche Bundesbank Gaganis C and Pasiouras F 2013 Financial supervision regimes and bank efciency International evidence Journal of Banking Finance 3754635475 Greene W 2005 Reconsidering heterogeneity in panel data estimators of the stochastic frontier model Journal of Econometrics 26269303 Jondrow J CAK L IS M and Schmidt P 1982 On the estimation of technical inefficiency in the stochastic frontier production function model Journal of Economet rics 19233238 Lee TH and Chih SH 2013 Does financial regulation affect the profit efficiency and risk of banks evidence from chinas commercial banks The North American Journal of Economics and Finance 26705 724 10 LozanoVivas A and Pasiouras F 2010 The impact of nontraditional activities on the estimation of bank efficiency International evidence Journal of Banking Finance 3414361449 Meeusen W and Van den Broek K 1977 Efficiency estimation from cobbdouglas production functions with composed error International Economic Review 18435 444 Mester L 1996 A study of bank efficiency taking into account riskpreferences Journal of Banking Finance 2010251045 Resti A 1997 Evaluating the costefficiency of the italian banking system What can be learned from the joint application of parametric and nonparametric techniques Journal of Banking Finance 21221250 Sealey C and Lindley J 1977 Inputs outputs and a theory of production and cost of depository financial institutions Journal of Finance 3212511266 Seiford L and Thrall R 1990 Recent developments in dea The mathematical pro gramming approach to frontier analysis Journal of Econometrics 46738 Sengupta J K 1999 A dynamic efficiency model using data envelopment analysis International Journal of Production Economics 62209218 Sun J Harimaya K and Yamori N 2013 Regional economic development strate gic investors and efficiency of chinese city commercial banks Journal of Banking Finance 3716021611 Yin H Yang J and Mehran J 2013 An empirical study of bank efficiency in china after WTO accession Global Finance Journal 11 Figure 1 Comparison of efficiency levels from DEA and SFA over the years 12 Table 1 Variables used on efficiency estimation Year Total Expenses Loans Liquid Assets Deposits Total interest expenses Total noninterest expenses w1 w2 2001 Mean 612951 8602172 3543838 1253685 3599656 2529854 00309 00145 SD 2319506 2854423 4474053 4435583 196382 1201174 00149 00023 N 11 11 11 11 11 11 11 11 2002 Mean 6517423 1087913 5200032 1552868 3827805 2689618 00255 00125 SD 3359987 5313906 7924992 8472038 2365301 135607 00136 00043 N 18 18 18 18 18 18 18 18 2003 Mean 7869842 1330581 5680322 1954243 4762005 3107837 00252 00116 SD 3995986 7105158 7198977 1060500 2836826 1646898 00129 00024 N 19 19 19 19 19 19 19 19 2004 Mean 781695 1439702 5217968 2179900 4509097 3307853 00205 00121 SD 487802 7771709 6370221 1289392 3250524 2147764 00111 00032 N 22 22 22 22 22 22 22 22 2005 Mean 1185633 2178598 8771696 3418808 7011715 4844617 00235 00119 SD 101892 2021322 1167075 3289473 6299166 4699437 00147 00029 N 33 33 33 33 33 33 33 33 2006 Mean 1299445 2449558 8431636 3727860 7578404 5416049 00227 00119 SD 1224961 2367041 1045025 3798744 7248159 5711323 00122 00030 N 45 45 45 45 45 45 45 45 2007 Mean 1695035 3093186 1274486 4811405 1063485 6315506 00234 00114 SD 162724 2988969 1247269 4916175 1056608 6397875 00098 00039 N 51 51 51 51 51 51 51 51 2008 Mean 2698787 4219558 2038509 6917867 1765891 9328953 00265 00126 SD 2876577 4839686 2912078 8074015 2022744 9439254 00097 00047 N 60 60 60 60 60 60 60 60 2009 Mean 2703437 5809675 2693320 9468915 1564529 1138907 00169 00108 SD 2860058 6701814 3393745 111E07 1772829 1140117 00055 00028 N 61 61 61 61 61 61 61 61 2010 Mean 360655 7451999 5140042 128E07 2107509 1499042 00161 000981 SD 3914733 8544235 6012537 143E07 2423446 1555168 00054 00026 N 59 59 59 59 59 59 59 59 2011 Mean 6445085 9653769 745E06 162E07 4311683 2133402 00264 001068 SD 6906545 110E07 894E06 173E07 5023891 206458 00111 00027 N 57 57 57 57 57 57 57 57 2012 Mean 1399510 190E07 148E07 304E07 1008609 3909015 00326 00094 SD 1220445 169E07 128E07 255E07 9501927 2967428 00075 00018 N 25 25 25 25 25 25 25 25 Ratio total interest expenses over deposits Ratio total noninterest expenses over total assets 13 Table 2 DEA economic efficiency scores Year Mean Std Dev Min Max 2001 084 024 036 1 2002 076 025 033 1 2003 073 026 033 1 2004 070 024 036 1 2005 074 026 019 1 2006 073 024 028 1 2007 056 024 022 1 2008 071 017 035 1 2009 073 014 048 1 2010 078 014 048 1 2011 079 017 031 1 2012 087 014 060 1 14 Table 3 SFA efficiency scores Cost efficiency Year Mean Std Dev Min Max 2001 089 004 081 093 2002 088 004 075 093 2003 088 005 076 093 2004 091 004 082 096 2005 092 003 085 096 2006 092 003 083 095 2007 092 002 080 095 2008 093 002 079 096 2009 092 002 087 096 2010 092 002 087 096 2011 092 002 086 095 2012 093 002 085 095 15 Journal of Econometrics 6 1977 2137 0 NorthHolland Publishing Company FORMULATION AND ESTIMATION OF STOCHASTIC FRONTIER PRODUCTION FUNCTION MODELS Dennis AIGNER University of Wisconsin Madison WI 53706 USA University of Southern Califbrnia Los Angeles CA 90007 USA CA Knox LOVELL University of North Carolina Chapel Hill NC 27514 USA Peter SCHMIDT University of North Carolina Chapel Hill NC 27514 USA Received October 1975 final version received October 1976 Previous studies of the socalled frontier production function have not utilized an adequate characterization of the disturbance term for such a model In this paper we provide an ap propriate specification by defining the disturbance term as the sum of symmetric normal and negative halfnormal random variables Various aspects of maximumlikelihood estimation for the coefficients of a production function with an additive disturbance term of this sort are then considered 1 Introduction The theoretical definition of a production function expressing the maximum amount of output obtainable from given input bundles with fixed technology has been accepted for many decades And for almost as long econometricians have been estimating average production functions It has only been since the pioneering work of Farrell 1957 that serious consideration has been given to the possibility of estimating socalled frontier production functions in an effort to bridge the gap between theory and empirical work For a variety of reasons these efforts have not been completely successful In this paper we suggest a new approach to the estimation of frontier production functions This involves the specification of the error term as being made up of two components one normal and the other from a onesided distribution This approach enables us to overcome some of the major shortcomings of previous work in the area This work was supported in part by NSF Grant GS39995 DJA We are indebted to Takeshi Amemiya Forrest Nelson and Dale J Poirier for helpful advice at various stages of the research and to Chinbang Chung for research assistance None of them should be held responsible for any errors that remain 22 D Aigner et al Stochastic frontier production function models The outline of the paper is as follows In section 2 we review previous app roaches to the estimation of parametric frontier production functions and in section 3 we propose a new approach The statistical properties of our model are discussed in section 4 Some Monte Carlo experiments are reported in section 5 and empirical examples are given in section 6 Section 7 concludes 2 Previous parametric frontier models Previous work on the estimation of parametric frontier production functions as characterized by the work of Aigner and Chu 1968 Afriat 1972 and Richmond 1974 begins by assuming a function giving maximum possible output as a function of certain inputs For a given firm say the ith we write Yi Axi i PI 1 Here yi is the maximum output obtainable from xi a vector of nonstochastic inputs and 1 is an unknown parameter vector to be estimated Aigner and Chu 1968 suggest the estimation of I by mathematical program ming methods based on a crosssection of N firms within a given industry Specifically they suggest minimization of subject to yi 6 fxi p which is a linear programming problem if fxi p is linear in 1 Alternatively they suggest minimization of subject to the same constraint which is a quadratic programming problem if fXi fl is linear Obviously something magical has happened in moving from 1 to either of these estimation methods In order to characterize differences in output among firms with identical input vectors or to explain how a given firms output lies below the frontier fni fi a disturbance term has been implicitly assumed One problem with these approaches is extreme sensitivity to outliers This has led to the development of socalled probabilistic frontiers Timmer 1971 Dugger 1974 which are estimated by the same types of mathematical pro gramming techniques discussed above except that some specified proportion of the observations is allowed to lie above the frontier The selection of this proportion is essentially arbitrary lacking explicit economic or statistical justifi D Aigner et al Stochastic frontier production function models 23 cation Another problem involves reconciling the observations above the frontier with the concept of the frontier as maximum possible output Typically this is accomplished by appealing to measurement error in the extreme observations However it seems preferable to incorporate the possibility of measurement error and of other unobservable shocks in a less arbitrary fashion As they have been applied previously therefore the mathematical program ming techniques do not lead to estimates with known statistical properties In an attempt to give them a statistical basis Schmidt 1976 explicitly added a onesided disturbance to 1 above which yields the model Yi fXxi Bi i lN where si 5 0 Given a distribution assumption for the disturbance term the model can then be estimated by maximumlikelihood techniques In particular the assumption that si has an exponential distribution leads to the linear programming technique while the assumption that si has a halfnormal distri bution leads to the quadratic programming technique Therefore Aigner and Chus estimates can be viewed as maximumlikelihood estimates under particular error specifications Unfortunately the observation that the model can be estimated by maximum likelihood techniques and that under appropriate assumptions linear and quadratic programming are maximumlikelihood techniques is of little practical value This is so because the usual regularity conditions for the application of maximum likelihood are violated In particular since yi 5 fxi p the range of the random variable y depends on the parameters to be estimated Therefore the usual theorems cannot be invoked to determine the asymptotic distributions of parameter estimates Under these circumstances it is not clear just how much we know about the frontier after having estimated it In another recent paper Aigner Amemiya and Poirier 1976 construct a more reasonable error structure than a purely onesided one Specifically they assume Ei i VT1 a if E 0 i 1 N Tlze if ET 5 0 3 where the errors 7 are independent normally distributed random variables with zero means and variance c2 for 0 8 1 otherwise E has either the negative or positive truncated normal distribution when 0 1 or 0 0 respectively Their justification for this error specification is that firms are presumed to differ in their production of y for a given set of values for the inputs according to random variation in 1 their ability to utilize best practice technology 24 D Aigner et al Stochastic frontier production function mode a source of error that is onesided si 5 0 andor 2 an input quantity or measurement error in y a symmetric error The parameter f3 is interpreted as the measure of relative variability in these two error sources its values circum scribing the full frontier function 0 l the average function 0 and intermediate cases of some interest A primary contribution of this error structure to the literature is that it allows the placement of the fitted function to be estimated along with the usual para meters of interest through the parameter 8 Thus the criticism levied at the average function by proponents of the frontier eg Aigner and Chu 1968 and criticisms that accompany strict use of the frontier or envelope function as the appropriate industry production function cf Timmer 1971 are ameliorated by this more accommodating specification Nevertheless the interpretation of 6 as a measure of the relative variability of error sources is only implicit in the Aigner Amemiya Poirier formulation A more direct approach is to specifically model the error process implied by the behavioral considerations mentioned above 3 A stochastic frontier We now return to the model as given in eq 2 but under the error structure Ei UiUi i 1 N 4 The error component z represents the symmetric disturbance the pi are assumed to be independently and identically distributed as N0 0 The error component U is assumed to be distributed independently of vi and to satisfy ui 5 0 We will be particularly concerned with the case in which ui is derived from a N0 rr distribution truncated above at zero However other onesided distributions are tenable and we will also briefly consider the case in which ui has an exponential distribution This model collapses to a deterministic frontier model when 01 0 and it collapses to the Zellner Kmenta and Dreze 1966 stochastic production func tion model when 0 0 Note that yi 5 fXi p vi so that the frontier itself is now clearly stochastic The economic logic behind this specification is that the production process is subject to two economically distinguishable random disturbances with different characteristics We believe that there is ample precedent in the literature for IAs 0 f 1 the positive error component has a large variance hence small influence in the likelihood function and the negative error dominates This gives rise to the full frontier as the limiting case 0 1 A similar interpretation follows for the case of 0 P 0 athough a behavioral explanation for this situation is lacking When 0 3 the likelihood function has the form of a mixture of two halfnormals each with equal influence D Aigner et al Stochastic frontier production function models 25 such a view although our interpretation is clearly new2 And from a practical standpoint such a distinction greatly facilitates the estimation and interpreta tion of a frontier The nonpositive disturbance ui reflects the fact that each firms output must lie on or below its frontier fxi p vi Any such deviation is the result of factors under the firms control such as technical and economic inefficiency the will and effort of the producer and his employees and perhaps such factors as defective and damaged product But the frontier itself can vary randomly across firms or over time for the same firm On this interpretation the frontier is stochastic with random disturbance vi 0 being the result of favorable as well as unfavorable external events such as luck climate topo graphy and machine performance Errors of observation and measurement on y constitute another source of vi 5 0 One interesting byproduct of this approach is that we can estimate the variances of Vi and Ui so as to get evidence on their relative sizes Another implication of this approach is that productive efficiency should in principle be measured by the ratio YilLfCxi B 21i13 5 rather than by the ratio This simply distinguishes productive inefficiency from other sources of disturb ance that are beyond the firms control For example the farmer whose crop is decimated by drought or storm is unlucky on our measure 5 but inefficient by the usual measure 63 Our discussion of estimation will be simplified somewhat if we consider a linear model We therefore write in obvious matrix form y xgs 7 in place of 2 where now a vu Marschak and Andrews 1944 suggest that the sum vi ui reflects the technical efficiency and the will effort and luck of a producer Zellner Kmenta and Dreze 1966 suggest that it reflects factors such as weather unpredictable variations in machine or labor performance and so on and they were perhaps the first to propose a stochastic production function although they clearly did not have a frontier in mind Other characterizations of the error term exist Aigner and Chu 1968 explain it by reason of technical and economic inefficiencies as well as by pure random shocks in the production process that might be due to careless handling and defective or damaged output Timmer 1971 cites technical and economic inefficiency as well as definitional and measurement problems in the variables And agri cultural economists frequently cite variation across farms in such environmental conditions as climate topography and soil type as indicative of a random production function 3As defined 5 is a strange efficiency measure since it is stochastic and depends on an unobservable vi It is offered here only to support the argument 26 D Aigner et al Stochastic frontier production function modeis 4 Estimation of the stochastic frontier model The distribution function of the sum of a symmetric normal random variable and a truncated normal random variable was apparently first derived by MA Weinstein 1964 The derivation of the density function of E is straight forward so we shall not include it here The result is 2 fE f E lFdC 0 0 cosEfCo Cr where IS 0 og 1 cc and f and F are the standard normal density and distribution functions respectively This density is asymmetric around zero with its mean and variance given by 42 EE Eu JJ b VE Vu Vu n2 6 n D U as can be easily ascertained from elementary considerations and calculation of the moments of U The particular parameterization in 8 is convenient because A is thereby interpreted to be an indicator of the relative variability of the two sources of random error that distinguish firms from one another4 A2 0 implies 03 co andor 0 f 0 ie that the symmetric error dominates in the determination of E Eq 8 then becomes the density of a N0 02 random variable as can be seen by inspection Similarly when C 0 the onesided error becomes the domi nant source of random variation in the model and 5 takes on the form of a negative halfnormal 5 We prefer to use this interpretation of I even though aU2 is not the variance of UT 2n UUZ is Another useful parameterization is to use oz along with c ajo For oVz 0 and thus 1 03 5 becomes J2 f 4 no exp 1 E2 2a2 1 for E j 0 zz 0 otherwise D Aigner et al Stochastic frontier production function models 27 The estimation problem is posed by assuming we have available a random sample of N observations and then forming the relevant loglikelihood function 12 In 9 yjI A 02 N lndT N In CI 2 In lFl 10 il which is almost exactly the form of the Amemiya 1973 p 1015 eq 102 Taking derivatives 1 ikelihood function considered by a In 2 jyr a In 8 i iI YiBxi9 an alns 1 N i iFI YiPiri i xi afi il i 11 12 where a k x 1 vector consisting of elements in the ith row of X and f and FT are respectively the standard normal density and distribution func tions evaluated at yi xi lo Given 12 we have that N fi ill 1 FT YiBXi 0 at the optimum Inserting this result into ll the ML estimator for CJ is determined through N tf 2a4 YiBxi O 2a2 rl 14 which yields A2 0 iil YiBxi2 15 28 D Aigner et al Stochastic frontier production function models the basis for the usual ML estimator of residual variance in a regression model But the determination of fi is not independent of 8 from other equations In any event this result can be used as a basis for an iterative solution scheme I premultiplied into 13 gives Adding to this j2 times eq 12 and simplifying we get 16 17 which in conjunction with 13 gives a system of k 1 equations that cor responds very closely to the system of firstorder equations encountered in the socalled Tobit model6 Since our density function is continuous in the range of E it is not anticipated that the difficulties encountered in the Tobit model will occur here and therefore we claim all the usual maximumlikelihood properties for the values of p 1 and a2 which simultaneously equate ll IZ and 13 to zero Formal proof of this claim and an examination of the regularity conditions that support it follow the analysis in Amemiya 1973 Various solution algorithms are available for finding the optimizing values of R and u Most of these the FletcherPowell algorithm for exampe require analytical first or secondorder derivatives in addition to the likelihood function itself for their best performance at reasonable cost in computer time Since such algorithms are now readily availabIe7 we will not devote any space to a discussion of the ML computational problem except to note that this likelihood function seems to be wellbehaved based on our experience Second order derivatives are presented in the appendix to this paper for that use and as a basis for calculating asymptotic standard errors of the ML estimates We note in passing that if estimation of p alone is desired all but the coefficient in 3 corresponding to a column of ones in X is estimated unbiasedly and con sistently by least squares Moreover the components of C can be extracted ie consistent estimators for them can be found based on the least squares results by utilizing eq 9 for VE in terms of 0 and C and a similar relation ship for a higherorder moment of e since VE and higher order meancorrected See Amemiya 1973 p 1011 eqs 72 73 7For a good discussion of the available algorithms as of a few years ago see Goldfeld Quandt 1971 D Aigner et al Stochastic frontier production function models 29 moments of E are themselves consistently estimable from the computed least squares residuals8 Similar comments andderivations would applyunder alternative distributional assumptions for ui For example we could choose the simple oneparameter exponential distribution for u lf9 where 4 2 0 is the mean of ui The variance is 4 A little algebra reveals that the distribution of ei viui is given by the density IFczexpb 19 where again F represents the cumulative distribution function of the standard normal distribution The likelihood function for the model follows immediately 5 Monte Carlo results In order to discover some specific information about the small sample be havior of the ML estimators discussed in the previous section we constructed two limited Monte Carlo experiments which rest entirely on artificial data Study 1 The model considered here is yi ei i 1 N where Ei is generated by eq 8 with various values of 0 and 0 The results and further details about the range of parameter values considered are reported in table 1 So no regression is involved in this instance indeed not even a mean is esti mated and we assume that the conventionally unobservable Ei is observable Some easily distinguished patterns of bias and precision in estimation emerge Considering table la 1 is apparently biased upward ol a and a2 seems to be estimated very well in all cases both with regard to small bias and MSE Moving to table lb increasing sample size to N 100 for the one set of results reported shows little if any basis for alteration of the qualitative con clusions reached above but also gives to indication of the reduction in bias and increase in precision promised by the asymptotic properties of ML sFor instance the thirdorder moment of E is 203 Er13 22jZj 1 In the Monte Carlo results that follow no attempt is made to evaluate the properties of these estimators C 30 D Aigner et al Stochastic frontier production function models Table la Monte Carlo results for the model yi si number of replications 100 sample size 50 value of A 166 124 083 166 124 083 141 b 136 008 122 i2 122 006 101 CP 102 004 1 177 028 li 131 020 x 093 013 1036 101 007 074 82 075 007 041 BZ 047 005 038 8 036 002 048 BUZ 047 002 060 8u2 055 003 188 a2 181 014 163 tP 164 011 135 8 136 008 n 180 033 i 135 017 t 094 013 138 aU2 134 016 099 82 103 011 055 aU2 062 009 050 b2 047 003 064 6 061 004 080 82 074 004 281 Bz 281 033 245 bZ 246 030 203 6 205 019 z 176 035 1 134 022 z 084 008 206 6u2 206 042 149 a2 153 030 083 6 085 018 075 82 075 007 096 BUZ 093 010 120 B2 120 009 Values in parentheses to the right of estimates are MSEs Values in parentheses to the left of2 b 2 u and 8 are the true values for these parameters used in the simulations Table lb Monte Car10 results for the model yt cl number of replications 100 sample size 100 value of Iz 166 124 0838 166 124 083 188 a2 188 014 163 82 170 012 135 6 132 008 z 181 037 5 134 023 x 090 012 138 8u2 140 015 099 62 106 014 055 a2 057 007 050 802 048 003 064 2 064 004 080 8 075 005 Values in parentheses to the right of estimates are MSEs Values in parentheses to the left of I and B are the true values for these parameters used in the simulations Study 2 Here the experiments of table 1 are repeated using the model yi jLLi i 1 IV Since si already possesses a nonzero mean the question is what effect does the extraction of an explicit intercept term have on the previous case From table 2a we see that qualitatively the results for 1 are altered somewhat with some cases of negative bias occurring Moreover the additional parameter to be estimated has a deleterious effect on 8 and hence a2 Whereas in the previous cases these estimators provided sharp results now there is a tendency for them both to be biased downward I is also biased downward but without any apparent pattern of relationship with the values of other parameters Again increasing sample size to N 100 has no perceptible effect on our results and conclusions from the N 50 case D Aigner et al Stochastic frontier production function models 31 Table 2a Monte Carlo results for the model y p cl number of replications 100 sample size 50 fl 10 value of 1 166 124 O838 166 124 083 141 P 134 022 1226 111 025 101 r 082 013 n 172 122 a 104 120 1 023 066 088 011 103 2 095 034 075 021 074 2 055 025 058 044 041 euz 012 023 038 Boz 039 004 048 a2 053 005 060 6 070 004 188 82 175 039 163 a2 161 051 135 a2 130 030 z 193 299 145 273 070 091 084 018 138 au2 125 065 082 025 099 zz oo 083 071 023 055 050 052 050 BZ 050 007 064 8 061 009 080 a2 080 010 81 8 256 089 245 8 240 093 203 s2 202 048 z 184 239 146 132 2 081 079 083 023 206 zU2 180 135 085 029 149 155 148 078 025 083 zU2 086 105 075 8 076 014 096 8 086 018 120 82 116 022 BValues in to the of estimates in parentheses to the of are the true values for these parameters used in the simulations Table 2b Monte Carlo results for the model y t8 number of replications 100 sample size 100 p 10 value of 1 166 124 083 166 124 083 188 B2 190 047 163 a2 179 055 135 a2 117 022 i 203 248 z 157 155 a 043 061 088 017 138 2 140 079 093 019 099 zUz 122 087 062 024 055 z 029 039 050 a 050 008 064 B 057 008 080 802 087 007 Values in parentheses to the right of estimates are MSEs Values in parentheses to the left ofbZ B 2andbZ t are the true values for these parameters used in the simulations Contrasting these results to similar experiments conducted by the Aigner Amemiya and Poirier 1976 the overall performance of common estimators 2 and S2 is roughly comparable We note that their measure of the relative variability in error sources is generally overstated but has small MSE compared to 1 9Cf their table 5 on page 20 The disturbance variance VE is set at 05 in their results whereas it varies in our table 2a and is generally much larger than 05 32 D Aigner et al Stochastic frontier production function models 6 Empirical examples In this section we present some examples of estimates of the stochastic frontier model and compare these estimates to other earlier results The first example uses 195758 data on the US primary metals industry SIC 33 consisting of observations across 28 states This data set was previously analyzed by Hildebrand and Liu 1965 and by Aigner and Chu 1968 The production function to be estimated is of the form In Y per lnZP In RlnKvu where Y is value added per establishment 1 is a measure of labor input per establishment K is the gross book value of plant and equipment per establish ment and R is the ratio of net to gross book value of plant and equipment Various sets of parameter estimates are given in table 3 The first three sets of entries correspond to our results running OLS and using the maximum Table 3 Estimates of eq 20 by various methods Method I 81 82 OLS Stochastic frontier exponential Stochastic frontier halfnormal HildebrandLiu OLS AignerChu OLS AignerChu LP AignerChu QPl AignerChu QP2 09146 09168 004164 204 731 219 09601 09144 004125 220 771 229 09600 09105 004208 206 768 234 0988 004208 0908 00333 0873 00031 1071 00269 0822 00219 likelihood technique of section 4 with the onesided disturbance assumed to be halfnormal and exponential respectively The numbers in parentheses under the OLS estimates are t ratios The numbers in parentheses under the stochastic frontier maximumlikelihood estimates are asymptotic t ratios That is they are the ratio of the coefficient estimate to the square root of the appropriate diagonal element of the inverse of the information matrix These are asymp totically distributed as N0 1 under the null hypothesis that the associated coefficient is zero D Aigner et al Stochastic frontier production function models 33 The remaining entries are taken from Aigner and Chu 1968 p 836 They consist of the OLS results of Hildebrand and Liu and of Aigner and Chu the linear programming results of Aigner and Chu and the results of two variants of the quadratic programming technique of Aigner and Chu We did not succeed in duplicating exactly the OLS results of either Hildebrand and Liu or Aigner and Chu although the discrepancies are not large What is most interesting is the extremely close agreement between our OLS estimates and both sets of stochastic frontier maximumlikelihood estimates In particular the stochastic frontier estimates are much closer to the OLS estimates than they are to Aigner and Chus programming results The reason for this is clear if we look at the estimates of the parameters of the distributions of the disturbances lo In the OLS case the estimated variance of the disturbances is 0077640 In the halfnormal case we have A2 0 0000686 asymptotic t ratio 005 A2 C 00692 asymptotic t ratio 364 and in the exponential case we have f 00180 asymptotic t ratio 016 A2 0 00691 asymptotic t ratio 368 In both cases the symmetric component of the disturbance effectively swamps the onesided component In the exponential case the mean of the onesided component is 0018 and its variance is 0000325 which is only 0468 percent of the total disturbance variance of 000032500691 00694 Similarly in the halfnormal case the mean of the onesided component is 0021 and its variance is 0000251 which is only 0361 percent of the total disturbance vari ance Therefore the picture that emerges is one of substantial variation in the frontier across states but relatively little variation of observed output beneath the frontier As a final note the maximized value of the logarithm of the likelihood func tion is 2372 in the exponential case and 2368 in the halfnormal case indicating a marginally better fit by the halfnormal distribution Our second example uses US agricultural data for six years 196065 and the 48 contiguous states The data set is the one used by Timmer 1971 exclud ing the years 1966 and 1967 The function to be estimated is Y ppl Labor Capitalp LandP Fertilizer Is LivestockP Seed v u 21 In the empirical examples used here the likelihood function was explicitly parameterized in terms of the parameters of the two error component distributions oU2 and uV2 rather than in the equivalent way discussed in section 4 34 D Aigner et al Stochastic frontier production function models where Y is gross agricultural output and the remaining variables are more or less selfexplanatory see Timmer 1971 for details All variables are in loga rithms and are on a perfarm basis Various sets of parameter estimates are given in table 4 The first two sets of estimates are our OLS and stochastic frontier results based on an exponential distribution for u The other four sets are taken from Timmer 1971 p 785 They represent respectively his OLS results his linear programming results and his linear programming results using 98 percent and 97 percent of the observations Table 4 Estimates of eq 21 by various methods Method OLS 18072 01149 02976 006061 01411 02581 01956 367 885 481 1240 1672 644 Stochastic frontier 18143 01148 02776 006061 01411 02581 01956 exponential 429 371 896 487 1256 1693 652 Timmer OLS 17350 01919 03726 00458 01484 02510 01579 538 67 117 42 160 195 54 Timmer LPI 0 0 16693 06015 04887 01334 02347 01043 Timmer LPg 8 18578 03287 03689 00298 01428 02045 02243 Timmer LPg7 18828 02679 04842 00099 01693 01885 01712 Our OLS results are probably as close as can be expected to Timmers OLS results given that we are missing 96 of his observations Our stochastic frontier estimates approximate the OLS estimates as in the previous example They are also reasonably close to Timmers OLS LP and LP estimates It would appear that the linear programming LPe estimates are the only ones very unlike the others Our estimates of the parameters of the distributions of the disturbances are f 000710 asymptotic t ratio 030 a 001005 asymptotic t ratio 112 Once again the symmetric component of the disturbance completely swamps the exponential component the variance of the exponential component 00000504 is only about one half of one percent of the total variance 7 Conclusions We have described a linear model with an error specification that is con sidered appropriate for the estimation of an industry production function using D Aigner et al Stochastic frontier production function models 35 crosssection data The specification does not prejudge the placement of the function as if an average function or a frontier function were to be fitted to the data Indeed the interesting feature of the framework discussed herein is precisely that placement of the function is estimated along with other model parameters This is also the contribution of the model presented in Aigner Amemiya and Poirier 1976 but that model is not capable of direct interpreta tion in terms of the sources of random error that may cause firms with identical input vectors to differ Whether from an applications viewpoint one model dominates the other is as yet not clear since the empirical evidence from both studies is limited What we find in the Monte Carlo results of the present paper is not particularly encouraging In the presence of an intercept the placement parameter A and the intercept itself are apparently difficult to estimate by the ML technique even when sample size is as large as 100 Further smallsample studies would help in forming a more definite opinion about the relative merits of the alter native specifications for tie placement and estimation of the industry production studied by Aigner Amemiya and Poirier and in the present paper Additional research is also required to evaluate the performance of the moment estimators based on leastsquares residuals mentioned at the end of section 4 Tests of our model on two realworld data sets indicated relatively small onesided components of the disturbance This in turn suggests high levels of efficiency relative to a stochastic frontier Whether this finding based on state wide perestablishment aggregates would continue to hold for the individual establishments themselves is yet another interesting question to be answered Appendix Below are reproduced the secondorder derivatives of In fZ with respect to fi 2 and o2 for use in computational algorithms and as a basis for estimating asymptotic standard errors aln2 1 iv 32 7 il 132 ifixi2 c fl I ff lfTyjBXi 1 A 1 aln2 apap f i xixl 642 36 D Aigner et al Stochastic frontier production function models 1 FryiPXi331FyiBXi b43 d21n9 1 N anap d il lOf c f i lF lnyififXi 1 FYijXi2 I Xi a2kz Tg ii1 yiXiXi L t L 3 il lj72 i lFT tfYifiXi 1 FTyiBXi2 645 a21n9 1 5 fi 12l 57 iCl 1 Fy I lFTYijXJ afyiflXi 1 FTyiBXi3 1 646 References Afriat SN 1972 Efficiency estimation of production functions International Economic Review 13 Oct 568598 Aigner DJ and SF Chu 1968 On estimating the industry production function American Economic Review 58826839 Aigner DJ T Amemiya and DJ Poirier 1976 On the estimation of production frontiers International Economic Review 17 377396 Amemiya Takeshi 1973 Regression analysis when the dependent variable is truncated normal Econometrica 41 9971016 Cohen AC Jr 1950 Estimating the mean and variance of normal populations from singly and doubly truncated samples Annals of Mathematical Statistics 21 557569 Cohen AC Jr 1957 On the solution of estimating equations for truncated and censored samples from normal populations Biometrika 44 225236 Dugger R 1974 An application of bounded nonparametric estimating functions to the analysis of bank cost and production functions unpublished PhD dissertation University of North Carolina Chapel Hill NC D Aigner et al Stochastic frontier production function models 37 Farrell MJ 1957 The measurement of productive efficiency Journal of the Royal Statistical Society A general 120 pt 3 253281 Goldfeld Stephen and Richard Quandt 1971 Nonlinear methods in econometrics North Holland Amsterdam Halperin M 1952 Maximum likelihood estimation in truncated samples Annals of Mathe matical Statistics 23 226238 Halperin M 1952 Estimation in the truncated normal distribution Journal of the American Statistical Association 47 457465 Hartley HO 1958 Maximum likelihood estimation from incomplete data Biometrics 14 174194 Hildebrand G and TC Liu 1965 Manufacturing production functions in the United States 1957 Cornell University Press Ithaca NY Marschak J and WJ Andrews 1944 Random simultaneous equations and the theory of production Econometrica 12 JulyOct 143205 Richmond J 1974 Estimating the efficiency of production International Economic Review 15 June 515521 Schmidt P 1976 On the statistical estimation of parametric frontier production functions Review of Economics and Statistics 58 238239 Timmer C Peter 1971 Using a probabilistic frontier production function to measure technical efficiency Journal of Political Economy 79 776794 Weinstein MA 1964 The sum of values from a normal and a truncated normal distribution Technometrics 6 104105 with some additional material 469470 Zellner A Kmenta J and J De 1966 Specification and estimation of CobbDouglas production functions Econometrica 34 Oct 784795 Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p2243 JanAbril 2023 22 REVISTA EVIDENCIAÇÃO CONTÁBIL FINANÇAS João Pessoa v 11 n 1 p 2243 JanAbril 2023 ISSN 23181001 DOI 1022478ufpb231810012023v11n158619 Disponível em httpperiodicosufpbbrojs2indexphprecfin INTERMEDIAÇÃO FINANCEIRA E EFICIÊNCIA UM ESTUDO DOS BANCOS E COOPERATIVAS DE CRÉDITO DO BRASIL1 FINANCIAL INTERMEDIATION AND EFFICIENCY A STUDY OF BRAZILS CREDIT BANKS AND COOPERATIVES Wesley Paulo Santos Mestre em Ciências Contábeis UFPE Faculdade Santa Helena wesleypaulolivecom Elenildo Santos Bezerra Doutorando em Ciências Contábeis UFPE Universidade Federal de Sergipe UFS elenildoconsultoriagmailcom Mércia de Lima Pereira Doutoranda em Ciências Contábeis UFPB Universidade Federal da Paraíba UFPB profamercialimagmailcom RESUMO Objetivo Analisar quais fatores influenciam a eficiência das instituições financeiras e verificar quais índices exigidos por Basileia possuem relação com a eficiência delas Fundamento A forma que os gestores de instituições financeiras utilizam para o reconhecimento de despesas impacta nos resultados do período e consequentemente na eficiência demonstrada por aquelas instituições financeiras aos demais agentes e órgãos reguladores Macedo Kelly 2016 Santos et al 2019 Método Foi realizada uma fronteira de eficiência estocástica bayesiana para determinar a eficiên cia dos custos e isolar o fator de ineficiência seguindo o que foi realizado por Tecles e Tabak 2010 e Barros e Wanke 2014 Em seguida foi comparada a forma com que os índices obrigatórios por Basileia afetaram o termo de ineficiência A amostra foi de seiscentas e setenta e uma instituições financeiras entre 2015 e 2018 Resultado Os resultados demonstraram que os fatores que afetaram a eficiência dos bancos foram as obrigações por empréstimos as despesas operacionais o preço pelo trabalho e nível do capital próprio Quanto às cooperativas de crédito os fatores que afetaram a eficiência foram os depósitos totais o market share e o nível de capital próprio 1 Artigo recebido em 23032021 Revisado por pares em 09052023 Reformulado em 01102021 Recomendado para publicação 18052023 por Anna Paola Fernandes Freire Editora Adjunta Publicado em 16082023 Organização res ponsável pelo periódico UFPB Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 23 Contribuições Este estudo contribui para a literatura por evidenciar que em linhas gerais a boa gestão do índice de Alavancagem proposto por Basiléia aumenta a eficiência dos custos para as cooperativas porém o mesmo não ocorre para os bancos sendo assim tal ineficiência dáse devi do a outros fatores Palavraschave Fronteira de eficiência estocástica bayesiana Eficiência Bancos Cooperativas de crédito ABSTRACT Objective To analyze which factors influence the efficiency of financial institutions and to verify which mandatory indexes by Basel interfere in efficiency Background The way that managers of financial institutions use to recognize expenses influences the result for the period and consequently the efficiency shown by that financial institution to other agents and regulatory bodies Method A Bayesian stochastic efficiency frontier was used to determine efficiency and isolate the inefficiency factor following what was done by Tecles and Tabak 2010 and Barros and Wanke 2014 Then the way in which the Basel mandatory ratios affected the inefficiency term was com pared The sample consisted of 671 financial institutions between 2015 and 2018 Result The results showed that the factors that affected banks efficiency were loan obligations operating expenses price for labor and level of equity For credit unions the factors that affected efficiency were total deposits market share and equity level Contributions This study contributes to the literature by showing that the higher the Basel index the lower the banking inefficiency and the greater the degree of financial leverage the greater the level of inefficiency Keywords Bayesian stochastic efficiency frontier Efficiency Banks Credit unions 1 INTRODUÇÃO O setor bancário assim como qualquer outro setor tem como objetivo gerir da melhor for ma possível seus recursos em um processo eficiente de utilização de insumos para a geração de riqueza Tecles Tabak 2010 Nesse sentido eficiência bancária deve ser entendida como as me lhores práticas na gestão dos custos empresariais e a ineficiência por outro lado é assumida como um fator que aumenta o custo das organizações e reduz seu lucro No entanto diferente de outras entidades a eficiência bancária não pode ser medida ape nas em relação aos custos e lucros gerados uma vez que os bancos estão expostos a variações ma croeconômicas o que pode fazer com que o lucro obtido durante um período não se confirme no período seguinte mas sim que haja prejuízo Santos Rodrigues Anjos 2019 Assim é importante que os bancos gerenciem a qualidade do capital responsável por gerar seus lucros Esse fator im plica uma dicotomia pois por um lado os bancos precisam captar recursos e emprestálos a outros agentes entretanto por outro lado eles precisam analisar de forma cuidadosa a quem concedem esses valores porque afinal podem não ser mais reavidos Nesse contexto a partir da segunda metade da década de 1970 os estudos sobre eficiência passaram a construir uma fronteira de eficiência capaz de determinar quais instituições estariam utilizando melhor seus recursos criando assim um ranking Os estudos de Aigner Lovell e Sch midt 1977 e Meeusen e Broeck 1977 foram os primeiros a associar a ineficiência a uma distribui ção de probabilidade estabelecendo uma Fronteira de Eficiência Estocástica FEE Se aplicado ao setor bancário a FEE poderá identificar quais bancos são os mais eficientes dentro do Sistema Fi Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 24 nanceiro Nacional SFN bem como quais fatores interferem de modo a aumentar o seu grau de ineficiência A técnica utilizada para criar a FEE foi de abordagem bayesiana Esta se diferencia da a bordagem clássica por selecionar a distribuição de probabilidade para um evento específico de forma prévia sem contudo saber o número total de casos desse evento Tecles Tabak 2010 Meeusen e Broeck 1977 expõem que a abordagem bayesiana é uma importante alternativa diante da falta de informações precisas sobre determinado evento ou população A distribuição de proba bilidade considerada neste estudo foi a proposta por Griffin e Stell 2007 e validada para as insti tuições financeiras brasileiras por Tecles e Tabak 2010 e Barros e Wanke 2014 Verificar a eficiência das instituições financeiras de um país ou região é importante pois elas são os agentes responsáveis por fomentar a economia na forma de concessão de empréstimos e créditos para que agentes deficitários possam operar A relevância que as entidades que fazem parte do SFN têm é reconhecida mundialmente tanto que elas são supervisionadas por bancos centrais e obrigadas a atender legislação própria emitida pelos acordos de Basileia Apesar da supervisão dos bancos centrais e da regulação proposta por Basileia o sistema financeiro pode passar por crises Quase 20 anos após o primeiro acordo de Basileia de 1988 que definiu o capital mínimo que as entidades financeiras deveriam ter para suportar a sua exposição ao risco e cerca de 4 anos depois do segundo acordo de Basileia de 2004 que teve como foco os riscos de crédito de mercado e operacional das instituições financeiras o mundo testemunhou a crise financeira ocorrida no ano de 2008 A crise financeira de 2008 entre outros fatores deuse pela falência de um banco de inves timentos o Lehman Brothers o que abalou a economia de vários países provocando um efeito devastador Após a crise financeira houve um novo acordo de Basileia chamado de Basileia III que entrou em vigor no Brasil no ano de 2013 Esse último acordo introduziu a obrigatoriedade do índice de alavancagem liquidez e reformulou a estrutura de capital das organizações Com as alte rações trazidas por Basileia III três indicadores ganharam especial atenção do sistema financeiro o índice de risco de Basileia o índice de Imobilização além do índice de Alavancagem Apesar de toda regulamentação e supervisão o setor financeiro carece de constante apri moramento uma vez que a falência de uma entidade financeira pode desestabilizar diversos agen tes econômicos Cardoso Campos Dantas Medeiros 2019 Para Ferreira Zanini e Alves 2019 a segunda metade do século XX foi de diversificação das receitas bancárias o que levou alguns paí ses a flexibilizarem suas leis para atender esse novo padrão Porém tal medida favoreceu uma maior adoção de um mix de receitas bancárias que privilegiasse o curto prazo Bonfim Kim 2012 Nesse sentido os acordos de Basileia têm entre seus objetivos reforçar o quadro global de capitais e preservar a liquidez do sistema por meio de requisitos mínimos obrigatórios de capital para suas entidades Cardoso et al 2019 No contexto brasileiro estudos como os de Pinheiro Savóia e Securato 2015 Goes Sheng e Schiozer 2016 e Cardoso et al 2019 analisaram o impacto de Basileia III sobre as entidades que compõe o SFN De forma anterior à Basileia III os estudos de Tecles e Tabak 2010 e Barros e Wanke 2014 criaram uma fronteira de eficiência para o setor bancário brasileiro No entanto ne nhum desses estudos estabeleceu a relação entre eficiência bancária e os índices obrigatórios por Basileia Assim o objetivo deste estudo é analisar a forma com que os índices obrigatórios por Basi leia afetam a eficiência das instituições financeiras brasileiras no período entre 2015 e 2018 De forma específica este estudo demonstra a quantidade de bancos e cooperativas que são eficientes e ineficientes além de verificar quais fatores influenciam a eficiência dessas instituições como por exemplo despesas operacionais market share entre outros Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 25 Para alcançar o objetivo foi realizada uma fronteira de eficiência estocástica para os bancos e outra para as cooperativas de crédito com a finalidade identificar e isolar o fator de ineficiência de cada instituição financeira da amostra Em seguida a ineficiência dessas instituições financeiras foram analisadas levando em consideração os índices obrigatórios pelos acordos de Basileia Estu dar os bancos e as cooperativas de crédito sob a ótica da eficiência financeira se justifica pois em bora ambas possuam finalidades distintas uma vez que os bancos possuem o propósito de gerar lucro aos seus acionistas enquanto as cooperativas de crédito possuem como propósito atender aos seus cooperados ambas as entidades possuem como principal atividade a intermediação fi nanceira através da facilitação do acesso ao crédito Os resultados deste estudo são úteis de três formas apontam quais fatores operacionais ou de mercado influenciam nos custos das entidades demonstram se a maior parte das entidades que compõe o SFN atua de forma eficiente e por fim evidenciam se os índices de estrutura de capital alavancagem e imobilização propostos por Basileia reduzem a ineficiência bancária o que tende a aumentar a estabilidade financeira e atende à finalidade com a qual eles foram estabelecidos Des sa forma depreendese que este estudo é importante para reguladores normatizadores clientes e demais agentes do mercado financeiro por evidenciar que a maioria dos bancos e cooperativas se preocupam em atender apenas ao mínimo exigido por Basileia Um dos possíveis motivos para isso está na exposição do capital dessas instituições a maiores níveis de risco 2 FUNDAMENTO 21 Sistema financeiro brasileiro No Brasil o Sistema Financeiro Nacional SFN é composto por um conjunto de entidades e instituições que promovem a intermediação financeira entre credores e tomadores de recursos O SFN possui três órgãos normativos o Conselho Nacional de Seguros Privados o Conselho Nacio nal de Previdência Complementar e o Conselho Monetário Nacional CMN sendo este último o responsável por formular políticas de moeda e crédito e todos esses órgãos são subordinados ao Governo Federal do Brasil BCB 2019 O CMN além de estabelecer políticas macroeconômicas também controla os seguintes ór gãos supervisores a Comissão de Valores Mobiliários CVM órgão responsável por supervisionar a bolsa de valores mercadorias e futuro e o Banco Central do Brasil BCB órgão responsável por supervisionar bancos caixas econômicas e cooperativas de crédito BCB 2019a De acordo com o BCB 2019b a quantidade de conglomerados financeiros e instituições independentes em dezem bro de 2018 era de 1356 instituições Desse total 134 instituições eram bancos 962 eram cooperati vas de crédito e 260 instituições eram constituídas na forma de outras entidades como instituições não bancárias de crédito não bancárias de mercado de capitais e instituições de pagamento Em relação às características do SFN em dezembro de 2018 existiam 962 cooperativas de crédito que possuíam um ativo total superior a R 271 trilhões enquanto apenas 6 instituições sendo todas bancárias em todo SFN eram classificadas como S1 ou seja que possuíam participa ção superior a 10 do mercado Os bancos que fizeram parte desse grupo foram Itaú Banco do Brasil Caixa Econômica Federal Bradesco Santander e BTG Pactual sendo a soma dos ativos des ses 6 bancos estimada no valor de R 62 trilhões enquanto o total de ativos de todas as 134 institu ições bancárias era de R 9 trilhões BCB2019b Outra característica do SFN é a existência de insti tuições financeiras públicas que operam no mercado comum Em 2018 existiam 28 instituições fi nanceiras nessa situação sendo o Banco do Brasil e a Caixa Econômica Federal bancos públicos que figuravam dentro do grupo das 6 maiores instituições financeiras nacionais BCB 2019b Em relação aos bancos públicos e privados no Brasil o estudo de Sanches Silva Júnior e Srisuma 2018 teve como objetivo estudar a privatização dos bancos públicos no país Os resulta dos apontam que os bancos privados possuem o objetivo de maximizar seus lucros enquanto os Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 26 bancos públicos tendem a fornecer serviços a segmentos de mercado que não são lucrativos ou carecem de financiamento diferenciado dada a sua importância Os autores citam como exemplo o Banco do Brasil que financia o setor agrícola brasileiro e a Caixa Econômica Federal que é a res ponsável pela distribuição dos recursos de fundos governamentais para habitação saneamento assistência social entre outros Assim Sanches et al 2018 discorrem que o desenho institucional do mercado brasileiro faz com que bancos públicos operem oferecendo crédito para infraestrutura hipotecas e investimentos enquanto os bancos privados ofereçam em sua maioria crédito pessoal e de curto prazo para empresas Sobre a fonte de receita dos bancos e cooperativas o estudo de Ferreira Zanini e Alves 2019 demonstrou que apesar do crescente movimento por diversificação de receita dessas institu ições a receita de intermediação financeira interest é a principal fonte de retorno dos bancos e cooperativas do Brasil A receita noniterest que são comissões e receitas em geral exceto de inter mediação financeira foi responsável em média por 274 do faturamento dos bancos brasileiros em 2014 o que demonstra que 726 das receitas obtidas foram de intermediação financeira con figurando ainda sua principal fonte de recurso Ao analisar especificamente as receitas de interme diação financeira Ferreira et al 2019 sugerem que os bancos atuantes no Brasil deem preferência às atividades de crédito em vez de atividades com títulos e valores mobiliários pois seriam as mais rentáveis Ao que se refere as cooperativas de crédito elas possuem também uma função social uma vez que o interesse da instituição é fornecer crédito com taxas mais atrativas para os seus coopera dos do que um banco comum Maia Bressan Lamounier e Braga 2013 comentam que os mem bros da cooperativa são os responsáveis por prover o capital necessário para a entidade quanto a demanda a ser consumida Assim no aspecto conceitual as cooperativas de crédito não devem se valer de altas taxas de juros com intermediação financeira pois sua finalidade principal é atender aos seus cooperados No entanto Ventura Fontes Filho e Soares 2009 demonstram que ainda assim as cooperativas de crédito podem ser submetidas a pressões por performance financeira uma vez que as sobras líquidas podem ser distribuídas entre seus cooperados Sobre a estrutura de capital das cooperativas de crédito o estudo de Carvalho Diaz Bia loskorski Neto e Kalatzis 2015 enfatiza que a gestão dessas instituições pode tornarse difícil uma vez que elas podem ser geridas por pessoas ligadas diretamente aos cooperados porém sem a de vida capacidade técnica A falta de capacidade técnica por parte da gestão pode por sua vez pre judicar a estrutura de capital das cooperativas Sob outra perspectiva os autores expõem que as cooperativas de crédito podem ter ganhos de escala por meio de fusões de pequenas cooperativas e o auxílio mútuo através de cooperativas centrais o que elevaria o seu retorno financeiro Quanto às diferenças apresentadas entre os bancos e as cooperativas dois estudos busca ram se aprofundar no tema no contexto brasileiro Matias Quaglio Lima e Magnani 2014 anali sou qual tipo de entidade foi mais eficiente em relação a gestão de suas receitas se comparado com as despesas administrativas e de pessoal para os anos entre 2002 e 2012 A conclusão chegada por Matias et al 2014 é a de que as cooperativas de crédito eram menos eficientes que os bancos pú blicos e privados na gestão de suas despesas Outro estudo que abordou de forma conjunta os ban cos e as cooperativas brasileiras foi o de Bittencourt V G Bressan Goulart A A Bressan Costa e Lamounier 2016 em que foi analisada a rentabilidade dos bancos e das cooperativas em relação ao retorno sobre o patrimônio líquido e o retorno sobre o ativo entre os anos de 2009 a 2013 Já Bittencourt et al 2016 concluem que não há diferença significativa entre o retorno sobre o patri mônio líquido dos bancos e das cooperativas no entanto em relação ao retorno sobre o ativo os bancos apresentam retornos maiores que as cooperativas sendo essa diferença estatisticamente significativa Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 27 Além do estudo de Cardoso et al 2019 outros estudos levaram em consideração o impac to causado pelas variáveis macroeconômicas para o setor bancário brasileiro O estudo de Araújo Lustosa e Paulo 2018 que teve como objetivo analisar se as provisões para crédito de liquidação duvidosa em bancos eram prócíclicas anticíclicas ou acíclicas concluiu que elas eram prócíclicas e que o efeito das variações macroeconômicas repercutia no mesmo período contábil Brito Lopes e Coelho 2012 e Dantas Paulo Medeiros 2013 analisaram o conservadorismo nos lucros dos bancos brasileiros Ambos os estudos demonstraram que os bancos tendem a ser conservadores condicionais o que implica reconhecer de forma antecipada as perdas se comparado com os ga nhos e que bancos estatais são mais propensos a tal procedimento Corroborando com o exposto Santos Rodrigues Anjos e Tavares 2019 analisaram o cená rio econômico interno e externo na prática de income smoothing entre os anos de 2011 a 2017 e i dentificaram que os bancos brasileiros são mais conservadores em seus lucros no final de cada se mestre ao aumentarem suas despesas com crédito de liquidação duvidosa Sobre o gerenciamento de resultado ao final de cada semestre Sousa 2018 demonstrou que essa prática é comum nos bancos brasileiros principalmente nos bancos de pequeno porte Santos et al 2019 também iden tificaram que o cenário econômico externo influencia de forma significativa a prática de suavização de resultado Os resultados de Santos et al 2019 corroboram o que foi exposto anteriormente por Macedo e Kelly 2016 ao sugerir que bancos de capital estrangeiro praticam mais gerenciamento de resultado do que os bancos de capital nacional 22 Fronteira de eficiência bayesiana Fronteiras de eficiência são ferramentas muito utilizadas em economia e finanças para de terminar quantidades ótimas como exemplo há a fronteira de eficiência de uma carteira de ações há a fronteira de eficiência do consumidor ou do produtor entre outros Apesar de amplamente divulgada as fronteiras de eficiência em sua forma clássica necessitam do prévio conhecimento do total de casos de um determinado evento sendo algo fechado em si mesmo Ntzoufras 2009 Apenas a partir dos estudos de Aigner et al 1977 e Meeusen e Broeck 1977 é que as fronteiras de eficiência passaram a ser consideradas de forma estocástica ou seja associada a uma distribui ção de probabilidade a partir de um evento aleatório gerado por simulação Aigner et al 1977 formularam e estimaram uma fronteira de produção associando o ter mo de erro do modelo a duas distribuições de probabilidade Em seguida uma série de simulações foram estimadas via Simulação de Monte Carlo sendo a principal contribuição do estudo a possi bilidade de estimar o modelo junto com sua respectiva função de eficiência Aigner Lovell Sch midt 1977 De forma similar Meeusen e Broeck 1977 também estimaram uma função de produ ção ao escolherem uma distribuição exponencial para o termo de erro Por fim concluíram que a fragilidade desse método estava na escolha prévia da distribuição de probabilidade sendo este o principal fator Meeusen Broeck 1977 Zhag 2000 ao comparar o modelo bayesiano com o modelo de máxima verossimilhança identificou que o primeiro foi superior ao estimar a fronteira de eficiência Assaf Matousek e Tsionas 2013 ao compararem o modelo Bayesiano com o método dos momentos generalizados GMM identificaram que o primeiro é mais estável possuindo me nos variações do que o segundo A escolha da distribuição de probabilidade pode ser realizada com base na análise da quan tidade obtida dos eventos observados ou pode se fazer o inverso primeiro se escolhe a distribui ção de probabilidade para em seguida se observar os eventos Essa segunda abordagem é deno minada de bayesiana e deve ser utilizada quando o número de casos para o evento é desconheci do porém já existem algumas evidências sobre eles Kinas Andrade 2010 De acordo com Te cles e Tabak 2010 em um modelo bayesiano são atribuídos aos parâmetros distribuições prévias Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 28 que contenham as informações já conhecidas sobre o evento sem contudo saber de forma precisa o número de casos para cada evento Nesse sentido o modelo bayesiano pode ser uma boa escolha se já existem evidências sobre o objeto de estudo ao mesmo tempo que não é o mais indicado caso o parâmetro não possua ne nhuma base teórica anterior De acordo com Kumbhakar e Tsionas 2005 a inferência bayesiana é realizada utilizando a técnica de Monte Carlo por Cadeia de Markov MCMC para os sorteios aleatórios A filosofia desse método é que a função de densidade posterior é dada por onde é o vetor do parâmetro sendo o objetivo simular sorteios aleatórios da distribuição posterior por Ntzoufras 2009 argumenta que o sistema bayesiano necessita de alguns parâmetros O primeiro deles é a seleção das variáveis dependentes e independentes Em seguida a escolha da distribuição de probabilidade utilizada que deve conter as características conhecidas dos números de casos já observados de um determinado evento ou ser validada de acordo com a literatura pre viamente estabelecida de estudos anteriores Ntzoufras 2009 Esse segundo passo visa contornar a fragilidade na determinação da distribuição de probabilidade apontada por Meeusen e Broeck 1977 Por último o passo apontado por Ntzoufras 2009 é a análise dos parâmetros de saída que são a fronteira de eficiência e a significância das variáveis que compõe o modelo Entre os estudos que utilizaram a abordagem bayesiana para analisar o setor bancário des tacamse para o contexto pretendido os trabalhos de Kumbhakar e Tsionas 2005 Behr 2010 e Assaf et al 2013 no âmbito internacional e o de Tecles e Tabak 2010 e de Barros e Wanke 2014 para o contexto brasileiro 23 Acordos de Basileia e eficiência bancária O Acordo de Capital de 1988 também denominado de International convergence of capital measurement and capital standards ou Basileia I teve como objetivo a convergência do setor financei ro mundial a partir da segunda metade do século XX Pinheiro Savóia Securato 2015 Ele estabe leceu de forma global o conceito de capital mínimo regulatório que na ocasião era de 4 para o nível 1 e de 8 no total Goes Sheng Schiozer 2016 O Brasil adotou as regras estabelecidas por Basileia apenas em 1994 período posterior ao da hiperinflação brasileira e quando adotou optou por níveis de capitais mais conservadores sendo de 8 para o capital mínimo e 11 para o ativo ponderado pelo risco Goes Sheng Schiozer 2016 Em 2005 foi estabelecido o segundo acordo de Basileia Ele promoveu mudanças qualitati vas uma vez que alterou o conceito de ativo ponderado pelo risco Pinheiro Savóia Securato 2015 Uma outra mudança considerável foi a dedução do ativo intangível na composição do capi tal mínimo regulatório Pinheiro Savóia Securato 2015 O terceiro acordo de Basileia realizado em 2011 foi uma revisão do Acordo de Capital De acordo com Pinheiro Savóia e Securato 2015 as alterações trazidas por Basileia III fizeram com que os bancos fossem obrigados a levar em consideração três indicadores financeiros sendo eles 1 O índice de imobilização que informa o quanto do capital total da instituição financeira está alocado em ativos permanentes 2 O índice de risco de Basileia que mede o risco de insolvência de uma instituição financeira assim quanto menor for esse índice mais propenso o banco estará de falir 3 O índice de alavancagem que é a razão entre o capital seguro nível 1 e a exposição total da instituição Para Goes Sheng e Schiozer 2016 o terceiro acordo de Basileia teve como objetivo aumentar a capacidade dos bancos de suportar possíveis perdas transferindo o possível ônus de decisões equivocadas por parte das instituições para investidores e proprietários enquanto livra o governo e a sociedade Diante da importância que o setor financeiro tem para a economia nacional além do risco operacional ao qual essas entidades estão expostas alguns estudos no Brasil trataram de verificar Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 29 como se deu a adoção de Basileia II nas instituições financeiras do Brasil O estudo de Santos Ma cedo e Rodrigues 2014 analisou os fatores que determinaram o nível de divulgação dos 100 maio res bancos do país no ano de 2010 Os autores chegaram à conclusão de que apenas um maior por te do banco e um melhor índice de Basileia foram capazes de explicar uma maior divulgação de informações financeiras por parte dos bancos De forma anterior Beneditto e Silva 2008 expuse ram que os únicos fatores que poderiam aumentar o nível de divulgação dos bancos nacionais além da regulação seriam através das auditorias independentes e por meio da pressão exercida por seus stakeholders Ainda sobre os estudos realizados que tomaram como base os acordos de Basileia o estudo de Pinheiro et al 2015 ao analisar o impacto trazido pelo terceiro acordo de Basileia no setor fi nanceiro brasileiro concluiu que os bancos devem buscar fontes de receita que ofereçam maior rentabilidade alinhados a uma maior eficiência no que diz respeito às suas despesas operacionais Para eles o acordo de Basileia III ao deixar mais rígido a ponderação pelo risco impactou direta mente no nível de resultado operacional dos bancos o que deve fazer com que esses agentes reve jam suas estratégias operacionais Uma vez que essa estrutura trazida por Basileia III diminuiu o valor líquido dos ativos das instituições financeiras o estudo de Cardoso et al 2019 analisou os fatores macroeconômicos re lacionados à liquidez estrutural dessas instituições Os autores identificaram que quanto maior a taxa de juros o risco Brasil e a saída de recursos na balança comercial menor será a liquidez do setor bancário brasileiro Ainda sobre a estrutura de capital dos bancos o estudo de F Malaquias A Malaquias e Hwang 2018 apontou que é crescente a parcela da população brasileira que dei xou de procurar as agências pois resolve suas demandas por meio de aplicativo fazendo com que instituições financeiras reduzissem o número de agências ou optassem por agências menores Outros estudos utilizaram a abordagem bayesiana para analisar o setor bancário de diferen tes países Kumbhakar e Tsionas 2005 analisaram 500 bancos comerciais dos Estados Unidos para o período de 1996 e 2000 e identificaram que os parâmetros técnicos e alocativos influenciavam na eficiência bancária Behr 2010 analisou a eficiência dos bancos caixas econômicas e cooperativas de crédito alemãs o resultado encontrado foi que as instituições mais eficientes foram aquelas cu jos custos diferem consideravelmente de sua média Assaf et al 2013 estimou a eficiência dos bancos turcos para os períodos entre 2002 e 2010 identificando que a eficiência média dos bancos turcos foi negativa e que isso pode ter sido agravado pela crise financeira de 2008 No contexto brasileiro o estudo de Tecles e Tabak 2010 analisou 156 bancos no período de 2000 a 2007 Os fatores considerados foram participação no mercado empréstimo e capital próprio e seus resultados apontam que não houve uma grande variação da eficiência dos custos no merca do brasileiro para o período O estudo de Barros e Wanke 2014 foi outro a analisar os bancos bra sileiros considerando os 40 maiores bancos brasileiros entre o período de 1998 e 2010 Obtiveram como resultado que quanto menor o porte dos bancos menos eficientes eles foram Também veri ficaram que bancos estrangeiros e públicos possuíram maior ineficiência Além dos fatores de entrada e saída de recursos que são sempre utilizados para determinar a eficiência das instituições financeiras outras características dessas instituições foram levadas em consideração por outros estudos para determinar a fronteira de eficiência estocástica Behr 2010 levou em consideração a rentabilidade do patrimônio líquido e o percentual de ativo fixo Tecles e Tabak 2010 consideraram como determinantes para a eficiência se os bancos eram públicos pri vados ou estrangeiros Barros e Wanke 2014 por sua vez levaram em consideração o porte dos bancos para determinar a eficiência se os bancos haviam recebido ajuda do governo durante o ano de 2008 e se os bancos haviam passado por processo de fusão recente Porém nenhum desses es tudos analisou o nível de eficiência obtida de forma bayesiana com o índice de eficiência trazido e homologado pelos acordos de Basileia Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 30 Para as cooperativas de crédito os estudos de Garden e Ralston 1999 e McKillop Glass e Ferguson 2002 utilizaram a análise envoltória de dados ou DEA para traçar a fronteira de eficiên cia Garden e Ralston 1999 ao traçar uma fronteira de eficiência para as cooperativas de crédito australianas entre os anos de 1992 e 1997 teve como objetivo verificar se na ocasião de fusões es sas novas cooperativas teriam maior eficiência em relação às entidades que a antecederam ou em relação ao mercado Os resultados encontrados por Garden e Ralston 1999 foram que as coopera tivas que foram objetivos de fusão não apresentaram maior grau de eficiência em nenhuma das duas situações De forma semelhante McKillop Glass e Ferguson 2002 analisaram as cooperati vas de crédito do Reino Unido eles se concentraram especificamente no ano de 1996 ano de esta bilidade econômica local onde as entidades financeiras conseguiram reverter uma recessão que o setor atravessava Os resultados obtidos por McKillop Glass e Ferguson 2002 demonstram que apesar do contexto a maior parte das cooperativas de crédito eram ineficientes atuando próximo ao limite aceitável pela legislação Em um contexto bayesiano o trabalho de Feng e Wang 2021 analisou os bancos comunitá rios dos Estados Unidos entre os anos de 2001 e 2017 Os bancos comunitários são semelhantes às cooperativas uma vez que ambos visam atender aos seus membros no entanto o que os difere é o fato de as cooperativas poderem oferecer outros serviços de intermediação financeira além de a tender aos seus segurados ou exclusivamente a uma comunidade local Feng e Wang 2021 chega ram à conclusão de que os bancos comunitários apresentaram um baixo desempenho técnico cus tos financeiros mais altos e menor rentabilidade se comparado com os bancos comerciais Em rela ção ao Brasil o trabalho de V G Bressan Braga e A A Bressan 2010 utilizou uma abordagem estocástica para avaliar a eficiência das cooperativas de crédito de Minas Gerais para o período de 2001 a 2003 os resultados obtidos foram que apenas 15 das cooperativas eram eficientes em rela ção aos seus custos 3 MÉTODO 31 População e Amostra Inicialmente a população foi constituída de 671 instituições financeiras que compreendiam o total das instituições financeiras listadas pelo Banco Central do Brasil BCB em dezembro de 2018 Os dados foram obtidos no endereço eletrônico do BCB do relatório de Dados Selecionados de Entidades Supervisionadas IFdata BCB 2019b Desse modo os dados considerados são de natureza secundária uma vez que a primazia deles era de fornecer informações ao BCB sobre a situação econômica e financeira das instituições financeiras que compõem o Sistema Financeiro Nacional Da população foram excluídas instituições que no período 1 não possuíam os 16 trimes tres entre março de 2015 e dezembro de 2018 2 não possuíam carteira de crédito ativa 3 não tinham operações realizadas 4 o seu ativo era igual a zero 5 eram considerados como bancos de desenvolvimento ou não bancárias de crédito Após as exclusões a amostra considerada foi de 52 bancos comerciais banco múltiplo com carteira comercial ou caixa econômica categoria b1 18 bancos múltiplos sem carteira comercial ou banco de investimento ou banco de câmbio categoria b2 543 cooperativas de crédito singulares categoria b3s 16 instituições centrais e confederações de cooperativas de crédito carteira b3c As sim a amostra final compreende 70 instituições bancárias e 559 cooperativas O período escolhido se justifica pela alteração em relação a forma de divulgação do BCB Antes de 2015 o BCB apresentava os dados financeiros apenas das 50 maiores instituições financei ras do país o que inviabilizava uma análise macro do setor O ano de 2018 foi escolhido como úl timo período pois em 2019 os bancos classificados como S1 foram obrigados a adotar de forma definitiva a Resolução 4616 de 30112017 do Conselho Monetário Nacional que reestruturou o Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 31 indicador de liquidez de longo prazo dessas instituições Assim o período escolhido contém a maior parte das instituições financeiras do país e a trata de forma homogênea em relação a sua obrigatoriedade legal 32 Modelagem A modelagem desenvolvida se baseia na fronteira estocástica bayesiana apresentada por Batesse e Coelli 1992 Griffin e Stell 2007 e adaptada para as entidades financeiras brasileiras por Tecles e Tabak 2010 Assim a eficiência de forma genérica é dada pela regressão em um painel da variável y que é o logaritmo natural do custo de intermediação financeira de uma firma i obser vada no tempo t com um conjunto de fatores de saída e de produção Na equação anterior é a variável dependente são fatores de produção e saída de recursos é um fator de ineficiência e pode assumir qualquer valor entre zero e um sendo que quanto mais próximo de um maior será a ineficiência da entidade i o que pode aumentar o valor da variável dependente e é o erro padrão da regressão Uma vez que serão consideradas distri buições de probabilidade em uma ótica bayesiana algumas modificações devem ser realizadas Os dois primeiros passos de acordo com Ntzoufras 2009 é a escolha das variáveis depen dentes e independentes além da seleção das distribuições de probabilidade Assim a junção de que é a variável dependente com fatores de saída entrada e outras características que queiram ser observadas formam uma função de eficiência que por sua vez é condicionada a uma distribuição de probabilidade conforme a expressão que segue Sendo µ corresponde a Ao realizar o segundo passo estabelecido por Ntzoufras 2009 buscouse validar se a dis tribuição de probabilidade normal utilizada por Tecles e Tabak 2010 e por Barros e Kanke 2014 era a mais indicada para a série de dados Por meio do software estatístico Oracle Crystal Ball foi verificada qual distribuição de probabilidade era capaz de melhor retratar a variável dependente para os bancos e cooperativas Para os bancos o teste de KolmogorovSmirnov foi aplicado à vari ável dependente que apresentou coeficiente de 0421 e pvalue de 000 No caso das cooperativas o teste de KolmogorovSmirnov apresentou coeficiente de 04944 e pvalue de 000 Desse modo a distribuição de probabilidade normal foi validada estatisticamente além de já ter sido adotada em estudos anteriores por Tecles e Tabak 2010 e Barros e Kanke 2014 As variáveis selecionadas de a tiveram como base os estudo de Koop Osiewalski e Steel 1997 Tecles e Tabak 2010 e Barros e Wanke 2014 A primeira variável compreende aos depósitos totais e é dada pela divisão entre o valor total dos depósitos pelo somatório das o brigações com empréstimos e repasses mais depósitos totais A proporção de obrigações por em préstimos é a fração entre as obrigações por empréstimos e repasses pelo total de ativos A taxa dos fundos corresponde à divisão das despesas com intermediação financeira pelo total de obrigações por empréstimos e repasses A proporção de despesas operacionais é a divisão Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 32 entre o total das despesas operacionais despesas de pessoal despesas administrativas despesas tributárias e outras despesas operacionais pelo total de ativos O preço do trabalho é a despe sa de pessoal dividida pelo ativo total O Market Share se dá pela divisão entre a quantidade de clientes ativos para cada banco pelo total de clientes ativos no sistema financeiro nacional O nível de capital próprio é a fração entre o patrimônio líquido de cada instituição pelo seu total de ativos Por fim a variável nível de atraso é a provisão sobre operações de crédito para clientes dividido pelo valor total das operações de crédito para clientes Os coeficientes de α e β também são modelados por uma distribuição de probabilidade eles seguem uma distribuição normal com média 0 e variância 1 conforme utilizado anteriormente por Griffin e Stell 2007 A ineficiência considerada na variável modela a diferença entre a melhor prática e a prá tica real e presume que tal medida tenha um comportamento exponencial e que a eficiência da firma varia ao longo do tempo Meeusen Broeck 1977 Griffin Stell 2007 Sendo assim a ine ficiência pode ser descrito da forma a seguir Onde nas instituições financeiras brasileiras é igual a 085 e é o logaritmo negativo que compõe uma distribuição exponencial que compreende esse mesmo argumento foi utilizado por Tecles e Tabak 2010 por sua vez segue também uma distribuição exponencial que modela que é a ineficiência parcial Já é encontrado conforme segue A variável corresponde a uma distribuição normal de média 0 e variância 04 conforme o utilizado por Battese e Coelli 1992 e Griffin e Stell 2007 e poderá indicar um incremento de efi ciência ao longo do tempo o que poderá resultar em um maior O representa a eficiência ao longo do tempo e se positivo poderá reduzir a ineficiência da instituição financeira dada por em que o resultado disso será a ineficiência total da instituição i no tempo t que é expressa por Uma vez estabelecida a ineficiência para determinar o seu oposto a eficiência deverá seguir a ex pressão A eficiência nesses termos segue o utilizado por Griffin e Stell 2007 e Tecles e Tabak 2010 e é a exponencial do inverso da ineficiência O numerador é a função de uma empresa livre de ineficiência por outro lado a base ou denominador é a função de uma empresa levando em consideração a ineficiência da mesma A métrica adotada para a eficiência resultará na criação de coeficientes que possibilitará a aferição e criação de uma fronteira de eficiência estocástica A fronteira de eficiência estocástica permitirá avaliar a quantidade de instituições que eficientes dadas as variáveis e premissas esco lhidas Esse processo de aferição e análise dos coeficientes e da fronteira de eficiência é o terceiro passo estabelecido por Ntzoufras 2009 Sobre o ranking é importante destacar que tal medida não tem como objetivo denegrir ou exaltar qualquer instituição financeira em especifico mas o de evidenciar o comportamento das instituições financeiras ao longo do tempo com base na metodologia aplicada motivo pelo qual os nomes das instituições financeiras serão preservados Por fim serão estimados dois painéis sepa radamente um para as instituições bancárias e outro para as cooperativas de crédito o que resul Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 33 tará em uma fronteira de eficiência para cada grupo em que medida permitirá a comparação entre esses dois macrossegmentos do SFN 33 Simulações Uma vez estabelecido o modelo foi cotado para cada painel 10000 interações os 1000 pri meiros números gerados foram descartados e a cada 25 valores gerados uma amostra era conside rada sendo tal medida necessária para diminuir possíveis correlações existentes entre as variáveis O software utilizado na modelagem foi o WinBUGS14 software grátis de abordagem bayesiana capaz de gerar números aleatórios por meio da simulação de Monte Carlo por Cadeia de Markov 34 Regressão Após determinar um índice de ineficiência de cada instituição financeira foi realizada uma regressão em um modelo logístico ordinal com o objetivo de elencar quais fatores influenciam na perda de produtividade das instituições financeiras O modelo logístico foi o escolhido por ser um dos mais indicados para tratar de variáveis dependentes que possuem como característica índices ou variáveis que exijam ranqueamento Wooldrige 2010 Greene 2018 A regressão foi focada em um único período de tempo dezembro de 2018 ou seja cortetransversal Assim temse a regres são Na regressão a variável dependente é o índice de ineficiência financeira de cada banco ou cooperativa compreende ao índice de Basileia compreende ao índice de imobilização compreende ao índice de alavancagem financeira Assim foram realizadas duas regressões uma contendo apenas os bancos e outra contendo apenas as cooperativas Para aumentar o rigor da comparação foi utilizada como metodologia também uma varia ção do Propensity Score Matching conforme discutido de modo técnico por Angrist e Pischke 2009 e já utilizado na área de contabilidade e finanças por Armstrong Jagolinzer Lacker 2010 e Gue dhami Pittman Saffar 2014 O score utilizado foi a ineficiência presente nos bancos e nas coopera tivas Desse modo ao considerar o coeficiente de ineficiência de um banco buscouse nas coopera tivas uma instituição que possuísse um coeficiente de ineficiência igual ou semelhante ao apresen tado pelo banco Tal método é denominado de vizinho mais próximo e foi por ele que foi realiza da a correspondência Matching entre os bancos e cooperativas Angrist Pischke 2009 Por esse motivo tanto a regressão aplicada aos bancos quanto a regressão aplicada às cooperativas possu em cada uma 70 instituições 4 RESULTADOS 41 Fatores que influenciam a eficiência dos custos com intermediação financeira dos bancos O painel realizado para as instituições bancárias não possuía presença de autocorrelação e sua série temporal foi estacionária A presença ou ausência de autocorrelação foi avaliada por meio da função de autocorrelação ou FAC seguindo o proposto por Griffin e Stell 2007 e Tecles e Ta bak 2010 e Santos 2022 Significa dizer que há indícios de autocorrelação quando apesar dos números de lags os valores não tendem a zero o que não foi o caso aqui todos coeficientes já ten diam a zero antes de 20 lags Sobre a estacionariedade ela foi garantida por meio da Simulação de Monte Carlo por Cadeia de Markov de acordo com o que foi apresentado por Lunn Thomas Best e Spiegelhalter 2000 e Santos 2022 uma vez que ela assegura por meio das simulações que a série convirja para uma única média o que também aconteceu na regressão abaixo Assim o painel apresentou os seguintes resultados Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 34 Tabela 1 Fronteira de eficiência bancária Variáveis Coeficientes Desviopadrão MC erro 5 95 Significância Constante 70900 02360 00070 67040 74710 Depósitos 02141 02698 00054 02274 06580 Obrigações por em préstimos 15140 03025 00045 10100 20140 Taxa dos fundos 00004 00004 00000 00002 00010 Despesas operacio nais 32950 09832 00121 16860 49100 Preço do Trabalho 78750 36470 00425 19070 138100 Market Share 00359 00448 00015 00357 01107 Nível de capital próprio 25580 04312 00056 32720 18500 Nível de atraso 00079 00122 00001 00122 00280 Notas significante a 1 significante a 5 significante a 10 Fonte Elaborado pelos autores 2020 Os resultados da tabela 1 demonstram que os custos com intermediação financeira dos ban cos brasileiros são influenciados positivamente pelas obrigações por empréstimos com coeficiente de 15140 proporção de despesas operacionais com 3295 preço do trabalho com 7875 e influen ciado negativamente pelo nível do capital próprio com coeficiente de 2558 sendo todos estatisti camente significantes a um nível de 10 Para os bancos quando eles captam dinheiro com tercei ros são obrigados a pagar uma taxa desse modo quanto maior o nível de obrigação com terceiros maior será o seu custo de intermediação financeira Esse resultado é semelhante ao encontrado por Tecles e Tabak 2010 ao considerar os empréstimos como um fator de saída de recursos para os bancos pois também encontraram uma associação positiva e significativa entre ele e os custos A despesa com mãodeobra também foi estatisticamente significativa Isso se deve ao fato de os bancos precisarem de pontos físicos agências para vender seus principais produtos Esse resultado também é semelhante ao encontrado por Tecles e Tabak 2010 que obtiveram um resul tado positivo e significativo para esta variável Ainda que a maioria dos bancos esteja investido em suas plataformas digitais como apontado por F Malaquias et al 2018 e que tenham surgido bancos exclusivamente digitais tal fator não é capturado pelo modelo uma vez que os bancos to talmente digitais por exemplo foram excluídos da amostra por não satisfazerem ao período anali sado de 16 trimestres A despesa com trabalho foi significativa e positiva cujo resultado é semelhante ao encon trado por Kumbhakar e Tsionas 2005 e Tecles e Tabak 2010 Barros e Wanke 2018 considera ram as despesas com trabalho de três formas diferentes a primeira sendo apenas o valor do traba lho ponderado pelo capital da empresa o segundo foi o valor do trabalho ponderado pelo capital da empresa ao quadrado e o terceiro foi o valor do trabalho ponderado pelo capital multiplicado pelos empréstimos concedidos a clientes Das três considerações realizadas por Barros e Wanke 2018 para o valor do trabalho a primeira apresentou sinal diferente do que foi encontrado en quanto o segundo e terceiro teve adoção semelhante ao que foi achado O resultado negativo e significativo encontrado para a proporção do capital próprio sobre o total de ativo foi também encontrada por Koop et al 1997 e por Tecles e Tabak 2010 Esse resul tado aponta que quanto maior é o nível de capital próprio menor será a ineficiência Isso se deve Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 35 ao fato de que quanto maior for o volume de capital próprio aplicado pelos bancos mais seguras serão suas estruturas de capitais pois eles não dependerão tanto de uma fonte originada de tercei ros Após determinados os fatores que influenciam os custos dos bancos foi realizado um ran king de quais instituições estão acima da fronteira de eficiência estocástica A seguir é apresentada a fronteira de eficiência de bancos Figura 1 Fronteira de eficiência dos Bancos Fonte Elaborado pelos autores 2020 O índice de eficiência se dá em um espaço amostral que está entre 0 e 1 concebido por meio de uma distribuição exponencial sendo esse espaço entre 0 e 1 a área da distribuição Assim quan to mais próximo de 1 mais eficiente será a instituição Já a fronteira de eficiência é determinada pela posição média que cada instituição financeira ocupa nesse espaço bem como sua variância sendo esse valor individual atribuído a cada instituição por meio da modelagem bayesiana reali zada Desse modo a fronteira de eficiência é traçada ao levar em consideração a posição que to das as instituições ocupam para a amostra e representa uma aproximação das melhores práticas para se ter um menor nível de custos de modo a maximizar o conjunto de saída Griffin Stell 2007 Para os bancos essa fronteira de eficiência foi de 00478 nessa escala que varia entre 0 e 1 o que demonstra uma fronteira de eficiência mais próxima de zero no entanto ao posicionar cada banco nesse espaço amostral a maioria ficou abaixo da fronteira de eficiência O modelo considerou os 70 bancos da amostra desses apenas 16 foram eficientes o que corresponde a 2286 enquanto 54 ou 7714 ficaram abaixo da fronteira de eficiência Ao anali sar os que ficaram abaixo da fronteira de eficiência foi constatado que os maiores bancos do Brasil até 2018 estavam nessa região ou seja apesar do ganho de escala tido pelo market share eles são os menos eficientes em relação a gestão dos seus custos O motivo para que bancos de pequeno e mé dio porte sejam mais eficientes do que os 5 maiores bancos do país para o momento analisado têm como causa o fato de eles necessitarem operar com mais recursos de terceiros do que com os próprios para atender a quantidade de clientes Isso faz com que eles tenham uma maior despesa remunerando o capital de terceiros além de terem um maior volume financeiro com inadimplên cia de seus tomadores Esses resultados são semelhantes aos encontrados por Barros e Wanke 2014 quando constataram que os maiores bancos brasileiros possuíam menor eficiência em rela ção aos seus custos Fronteira de eficiência Bancos 100E5 100E4 0001 001 01 Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 36 42 Fatores que influenciam a eficiência dos custos com intermediação financeira das cooperati vas de crédito O painel realizado para as cooperativas de crédito não apresentou autocorrelação e sua série temporal foi estacionária A ausência de autocorrelação foi avaliada por meio da função de autocorrelação ou FAC também seguindo o proposto por Griffin e Stell 2007 e Tecles e Tabak 2010 e Santos 2022 Foi verificado que todos os coeficientes já tendiam a zero antes de 20 lags indicando ausência de autocorrelação A estacionariedade foi garantida por meio da Simulação de Monte Carlo por Cadeia de Markov também de acordo com o que foi apresentado por Lunn Thomas Best e Spiegelhalter 2000 e Santos 2022 uma vez que ela assegura por meio das simu lações que a série convirja para uma única média o que também aconteceu na regressão que em breve será apresentada Sobre a regressão os depósitos totais e o market share foram estatisticamente significativos para as cooperativas de crédito e de forma positiva os depósitos totais com coeficiente de 06025 e o market share com coeficiente de 3949 Com relação negativa apresentouse o nível de capital pró prio com 00136 Todos foram significativos a 10 A variável taxa dos fundos e o nível de atraso não foram estatisticamente significativas nem para os bancos nem para as cooperativas Os resul tados da tabela 2 demonstram que com exceção da variável nível de capital próprio que é estatisti camente significativa e que reduz o custo de intermediação financeira para os bancos e para as cooperativas os fatores que influenciam no custo de intermediação financeira para os bancos e para as cooperativas são diferentes O sinal negativo para o nível de capital próprio para os bancos e para as cooperativas demonstra que quanto maior o nível de capital próprio menor é o custo de intermediação financeira dessas instituições Os resultados do painel são apresentados a seguir Tabela 2 Fronteira de eficiência das cooperativas de crédito Variáveis Coeficientes Desviopadrão MC erro 5 95 Significância Constante 29810 00756 00012 28550 31040 Depósitos 06025 00828 00015 04669 07400 Obrigações por em préstimos 00322 00253 00003 00104 00737 Taxa dos fundos 00000 00000 00000 00000 00000 Despesas operacionais 00272 01177 00013 01672 02181 Preço do Trabalho 01422 02474 00026 02594 05526 Market Share 394900 66010 01288 285800 504100 Nível de capital pró prio 00136 00051 00001 00220 00052 Nível de atraso 00098 00095 00001 00058 00253 Notas significante a 1 significante a 5 significante a 10 Fonte Elaborado pelos autores 2020 Para as cooperativas quanto maior o nível de depósitos totais maior é o custo de interme diação financeira isso se deve pois diferente dos bancos que a principal obrigação com terceiros vem de valores obtidos com outras instituições bancárias as cooperativas possuem como sua prin cipal fonte de obrigação as captações obtidas por seus cooperados que são classificados como de pósitos totais Maia et al 2013 Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 37 Em relação ao market share as cooperativas de crédito apresentaram resultados estatistica mente diferentes daqueles apresentados pelos bancos e mais parecidos com os encontrados por Barros e Wanke 2014 para os bancos durante o período de crise financeira de 2008 Enquanto eles identificaram que bancos que possuíam agências na maior parte do Brasil tinham uma maior efici ência em relação aos seus custos esse resultado também pôde ser encontrado em relação às coope rativas de crédito O motivo pelo qual o porte das cooperativas influencia na eficiência dos custos se deve ao fato dos clientes ativos nas cooperativas de crédito serem os seus cooperados Assim a variável market share é a mais significativa uma vez que quanto maior for a quantidade de coope rados também será maior a quantidade de clientes que a cooperativa terá que atender Os bancos e cooperativas de crédito em relação ao nível de capital próprio apresentaram resultados semelhantes isso se deve ao fato de que quanto maior forem os recursos próprios em pregados na operação bancária em linhas gerais menor será os custos de intermediação financei ra uma vez que não terá que remunerar nenhum terceiro Esse resultado está de acordo com o encontrado por Tecles e Tabak 2010 que identificaram que no Brasil quanto maior o nível de capital próprio mais eficiente será a instituição financeira Figura 2 Fronteira de eficiência das cooperativas Fonte Elaborado pelos autores 2020 Para as 559 cooperativas de crédito foi considerada uma fronteira de eficiência de aproxi madamente 00642 em uma escala que varia entre 0 e 1 Uma vez que essa escala se dá por meio de uma distribuição exponencial uma fronteira de eficiência mais próxima de 1 representa que as chances de se ter instituições ineficientes são maiores uma vez que a área que fica abaixo da efici ência necessária é maior Nesse sentido Griffin e Stell 2007 argumentam que a fronteira estocásti ca representa uma aproximação das melhores saídas com base em um menor custo Assim 102 instituições ou 1825 foram eficientes enquanto 457 instituições correspondendo a 8175 foram ineficientes Os resultados das fronteiras de eficiência demonstram que as cooperativas possuem uma fronteira de eficiência maior que a dos bancos sendo respectivamente 00642 e 00478 Nas duas amostras a maioria das instituições ficaram abaixo da fronteira de eficiência apesar do aumento de regulação implementado no setor com os acordos de Basileia Portanto 7714 dos bancos e 8175 das cooperativas ficaram abaixo da eficiência esperada para o setor Fronteira de eficiência Cooperativas 100E5 100E4 0001 001 01 Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 38 44 Fatores que influenciam a ineficiência das instituições financeiras Uma vez que a maioria dos bancos e cooperativas de crédito ficaram abaixo da fronteira de eficiência buscouse analisar quais fatores obrigados pelos acordos de Basileia reduzem a inefici ência das instituições financeiras Foram considerados em uma regressão o índice de Basileia o nível de Imobilização e o grau de Alavancagem todos propostos por Basileia como medida que poderia reduzir a ineficiência das instituições financeira Para a regressão foi utilizado o modelo logit ordinal sendo o coeficiente de ineficiência o fator de ordem entre as instituições A tabela abaixo foi dividida em dois grupos de modo a con templar os bancos e as cooperativas Em ambos os casos a segunda coluna contém a probabilidade de ocorrência que é a chance que determinada variável possui em afetar o termo de ineficiência a terceira coluna contém os coeficientes do modelo logit para cada variável a quarta traz o VIF que é o fator de inflação das variáveis a quinta e a sexta coluna respectivamente trazem o erropadrão e o pvalue de cada variável Tabela 3 Índices obrigatórios e ineficiência Bancos1 Probabilidade de Ocorrência2 Coeficiente VIF Erro padrão Significância3 64424 0594 1080 1177 29630 0865 1080 2630 59632 0390 1160 1957 Cooperativas1 Probabilidade de Ocorrência2 Coeficiente VIF Erro padrão Significância3 0089 7018 1150 1508 99998 11796 1050 4190 14273 1793 1100 2599 Notas 1 O número de observação para os bancos foi de 70 instituições financeiras igual ao número de cooperativas que também foram de 70 instituições Essas 70 cooperativas foram escolhidas por apresenta rem o termo de ineficiência igual ou similar ao apresentado pelos bancos na técnica de correspondência chamada de vizinho mais próximo 2 A probabilidade de ocorrência PO se deu por PO exp coeficiente 1 exp coeficiente 3 significante a 1 significante a 5 significante a 10 Fonte Elaborado pelos autores 2022 Os resultados das regressões acima demonstram que o índice de Basileia Imobilização e Alavancagem não possuem correlação com a ineficiência de custos das instituições bancárias por outro lado para as cooperativas os índices de Basileia e Alavancagem tiveram uma correlação significativa a 1 com o termo de ineficiência de custos O fator de inflação das variáveis ou VIF demonstrou que para ambas regressões as variáveis foram consistentes ou seja não inflacionadas o que permite a inferência caso ela seja significativa Em relação aos bancos embora as probabilidades de ocorrência tenham apresentado per centuais acima de 50 para os índices de Basileia e Imobilização tal percentual não possui respal do uma vez que essas variáveis não foram significativas Entre os possíveis motivos estão a dife rença que os bancos da amostra possuem entre si seja em relação ao porte onde apenas 5 bancos juntos possuíam market share superior a 65 do mercado em dezembro de 2018 enquanto os ou tros 65 bancos possuíam juntos 18 de market share os outros 17 de market share correspondem as cooperativas de crédito da amostra e a outros bancos e cooperativas que por algum dos motivos listados na metodologia não fizeram parte da amostra Além do porte os bancos poderiam ser di gitais físicos ou híbridos com representação digital e física o que influenciaria na sua estrutura de custos Um terceiro possível motivo está nos principais produtos e públicoalvo que os bancos pos suíam o que poderia influenciar a sua estrutura da carteira de crédito Assim existiram na amos Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 39 tra bancos cuja principal atividade era a exploração de sua carteira comercial outros possuíam como atividade principal sua carteira de investimento e outros bancos cuja finalidade era a de fi nanciar produtos para terceiros de outras subsidiarias não financeiras como financiamento de veí culos e máquinas agrícolas Desse modo uma vez que os bancos variavam em estrutura porte públicoalvo perfil da carteira de crédito além de outros fatores não observáveis a sua ineficiência de custos não se correlaciona diretamente com os índices de Basileia Alavancagem e Imobilização Em relação ao índice de Basileia e Alavancagem os bancos podem optar a quem vai conce der seus empréstimos bem como pedir garantias reais se esse crédito for para o financiamento de bens duráveis e uma vez que são poucos os bancos que concentram a maior parte do mercado logo devem ser eles os mais lembrados por clientes que possuem um perfil de crédito arriscado Em relação ao índice de Imobilização os bancos digitais tendem a possuir menos estruturas físicas enquanto os bancos físicos com maior capilaridade no mercado tendem a possuir maior número de agências unidades de atendimento e consequentemente de pessoal alocado nesses lugares po rém esses bancos são minoria No tocante as cooperativas de crédito o índice de Basileia e o índice de Alavancagem estão correlacionados com a ineficiência dos custos No entanto embora o índice de Basileia esteja corre lacionado sua probabilidade é inferior a 1 o que significa dizer que sua chance de ocorrência é quase nula o que é justificável visto que o índice de Basileia tem como objetivo promover a efici ência e estabilidade das instituições financeiras no mercado Por outro lado uma cooperativa de crédito mais alavancada possui uma probabilidade superior a 99 de afetar o índice de ineficiên cia dos custos Assim em linhas gerais quanto mais alavancada é uma cooperativa mais sua estru tura de custos será pressionada o que pode aumentar sua ineficiência de custos por ter gastos ex tras No entanto em relação as cooperativas uma vez que sua atividade principal é suprir as ne cessidades dos seus cooperados sua estrutura de custos por vezes pode ser pressionada em prol desse objetivo Outros fatores que podem justificar esses achados em relação as cooperativas são que elas de forma geral se mostraram mais homogêneas entre si haja vista que o número de cooperativas ativas no Sistema Financeiro Nacional em dezembro de 2018 excedia ao de bancos em mais de sete vezes enquanto o total de bancos era de 135 instituições a quantidade de cooperativas era de 962 instituições Alinhado a isso nenhuma cooperativa de crédito possuía market share suficiente de modo a dominar o mercado sendo antes instituições consideradas de pequeno porte para o Siste ma Financeiro Nacional E um terceiro fator que deve ser levado em consideração é o fato de as cooperativas existirem primordialmente para atenderem as necessidades dos seus cooperados o que em muitos casos pode limitar sua cesta de produtos e serviços financeiros além de pressionar sua estrutura de custos 5 CONCLUSÃO Este artigo teve como objetivo analisar a forma com que os índices obrigatórios por Basileia afetam a eficiência de instituições financeiras brasileiras no período entre 2015 e 2018 Para atingir tal propósito foram coletados dados de 629 instituições financeiras das quais 70 instituições são bancárias e 559 são cooperativas de crédito A modelagem utilizada para encontrar a ineficiência e a eficiência bancária seguiu a mesma utilizada pelos estudos de Griffin e Stell 2007 e Tecles e Ta bak 2010 que criaram uma fronteira de eficiência estocástica e determina um fator de ineficiência Os principais resultados encontrados sobre quais fatores influenciam os custos das entida des financeiras apontam que obrigações com empréstimo proporção das despesas operacionais e preço pago pelo trabalho aumentam o custo dos bancos enquanto um maior nível de capital pró prio o reduz Para as cooperativas de crédito os fatores que elevaram seus custos foram os depósi tos totais e o market share e de modo negativo o nível de capital próprio Esses resultados demons Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 40 tram que bancos e cooperativas apesar de fazerem parte do SFN possuem diferentes fatores res ponsáveis por elevar seus custos mas que em ambos os casos quanto maior for o nível do capital próprio menor será o custo dessas entidades No que diz respeito à fronteira de eficiência foi demonstrado que as cooperativas de crédi to possuem uma zona de ineficiência maior que os bancos Também foi encontrado que os bancos foram proporcionalmente mais eficientes que as cooperativas sendo 7714 dos bancos e 8175 das cooperativas Esses resultados concordam de certo modo com o que foi encontrado nas regres sões logísticas para os bancos e cooperativas em que se evidenciou que uma maior alavancagem das cooperativas de crédito possui elevada probabilidade de influenciar no termo de ineficiência dos custos Ou seja as cooperativas de crédito que possuíam ineficiência de custos similares àque las apresentadas pelos bancos apesar de similares eram mais sensíveis a fatores que influencia vam seu nível de alavancagem como volume de depósito e nível de capital próprio o que afetava diretamente a eficiência em relação aos custos enquanto para os bancos não houve uma relação clara entre a gestão dos índices propostos por Basileia e a redução ou aumento do termo de inefici ência dos custos Esses resultados corroboram o que foi exposto por Pinheiro Savóia e Securato 2015 Goes Sheng e Schiozer 2016 e Cardoso et al 2019 sobre a gestão dos recursos de posse das entidades bancárias Desse modo concluise que o índice de imobilização não influencia a ineficiência dos bancos e das cooperativas enquanto o grau de alavancagem aumenta a ineficiência das cooperati vas de crédito Em relação ao índice de Basileia em si seus resultados não foram conclusivos uma vez que para os bancos ele apresentou elevada probabilidade mas não foi significativo enquanto para as cooperativas embora significativo seu efeito era mínimo em relação à ineficiência Vale ressaltar o que foi exposto por Maia et al 2013 que de forma geral as cooperativas de crédito são menos alavancadas uma vez que parte do seu capital é subsidiada pelos seus cooperados Nesse sentido Ventura Fontes Filho e Soares 2009 argumentam que o contexto das cooperativas é complexo uma vez que seus cooperados possuem uma relação de cliente e de fornecedor com as instituições fazendo com que para eles o lucro não seja o objetivo principal mas que apesar desse contexto desafiador as cooperativas não devem se acomodar em relação as suas ineficiências No mais por meio desses resultados é possível concluir que de modo geral para as coope rativas de crédito a observação e a boa gestão dos Índices de Basiléia Alavancagem Imobilização e Basiléia produzem um efeito positivo nos custos das cooperativas de modo a aumentar sua efi ciência Por outro lado para os bancos de forma geral a gestão dos Índices de Basiléia não reper cute diretamente no grau de ineficiência dos custos Possivelmente fatores não observados por este estudo como o seguimento do banco o tipo do consolidado bancário o tipo da carteira de crédito e a presença física eou digital da instituição são fatores que influenciam na sua estrutura de custos e consequentemente na ineficiência bancária Por fim este estudo possuiu como limitação o fato de a fronteira de eficiência bayesiana pressupor uma série de distribuições de probabilidade como o argumentado por Meeusen e Bro eck 1977 para este tipo de estudo Como sugestão para pesquisa futura podese verificar a forma que o cenário econômico local interfere na eficiência bancária REFERÊNCIAS Aigner D Lovell CAK Schmidt P 1977 Formulation and estimation of stochastic frontier production fuction models Journal of Econometrics 6 p 2137 httpsdoiorg1010160304 407677900525 Araújo A M H B Lustosa P R B Paulo E 2018 A ciclicidade da provisão para créditos de liquidação duvidosa sob três diferentes modelos contábeis Reino Unido Espanha e Brasil Re vista de Contabilidade e Finanças USP 29 76 p 97113 https1015901808057x201804490 Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 41 Armstrong C S Jagolinzer A D Larcker D F 2010 Chief Executive Officer Equity Incentives and Accounting Irregularities Journal of Accounting Research 48 2 p 225271 https 101111j1475679X200900361x Angrist J D Pischke J S 2009 Mostly Harmless Econometrics An Empiricistís Companion Princeton University Press Assaf A G Matousek R Tsionas E G 2013 Turkish bank efficiency Bayesian estimation with undesirable outputs Journal of Banking Finance 37 p 506517 httpsdxdoiorg10116jbankfin201209009 Barros C P Wanke P 2014 Banking efficiency in Brazil Journal of International Financial Mar kets Institutions Money 28 p 5465 httpsdxdoiorg101016jintfin201310004 Batesse G E Coelli T J 1992 Frontier production functions technical efficiency and panel data With application to paddy farmers in India Journal of Productivity analysis 3 12 p 153 169 httpsdoiorg101007BF00158774 BCB 2019a Organização do Sistema Financeiro Disponível em httpswwwbcbgovbrestabilidadefinanceiraorganizacao Acesso em 24 de set de 2019 BCB 2019b Dados selecionados de instituições financeiras supervisionadas Disponível em httpswww3bcbgovbrifdata Acesso em 24 de set de 2019 Behr A 2010 Quantile regression for robust bank efficiency score estimation European Journal of Operational Reseach 200 p 568581 httpsdoiorg101016jejor200812033 Beneditto E A N Silva R N S 2008 Análise do nível de transparência das instituições fi nanceiras brasileiras em relações ao acordo com o novo acordo de capitais Basiléia II Um es tudo de caso múltiplo Brazilian Business Review 5 3 p 192208 httpsdxdoiorg1015728bb2008532 Broeck J D Koop G Osiewalski J Steel MFJ 1994 Stochastic frontier models A Bayesian perspective Journal of Econometrics 61 p 273303 httpsdoiorg1010160304407694900876 Brito G A Lopes A B Coelho A C D 2012 Conservadorismo nos lucros contábeis dos bancos no Brasil Influência do controle estatal Revista Universo Contábil 8 3 p 1939 httpsdoiorg104270ruc2012429 Cardoso V R S Campos J A Dantas J A Medeiros O R 2019 Fatores relacionados à li quidez estrutural dos bancos no Brasil Revista de Contabilidade e Finanças USP 30 80 p 252 267 https1015901808057x201806350 Carvalho F L Diaz M D M Bialoskorski neto S Kalatzis A E G 2015 Saída e insucesso das cooperativas de crédito no Brasil Uma análise do risco Revista de Contabilidade e Finanças USP 26 67 p 7084 https1015901808057x201411390 Ferreira J H Zanini F A M Alves T W 2019 A diversificação de receitas bancárias seu impacto sobre risco e retorno dos bancos brasileiros Revista de Contabilidade e Finanças USP 30 79 p 91106 https1015901808057x201805810 Goes K C Sheng H H Schiozer R F 2016 Contingentes conversíveis e seus impactos na otimização da estrutura de capital de bancos brasileiros sob Basileia III Revista de Contabilidade e Finanças USP 27 70 p 8097 https1015901808057x201501350 Greene W H 2018 Econometric Analysis Pretince Hall Griffin J Stell M 2004 Semiparametric Bayesian inference for stochastic frotier models Jour nal of Econometrics 123 p 121152 httpsdoiorg101016jeconom200311001 Griffin J Stell M 2007 Bayesian stochastic frontier analysis using winbugs Journal of Produc tivity Analysis 27 p 163176 httpsdoiorg101007s111230070033y Guedhami O Pittman J A Saffar W 2014 Auditor Choice in Politically Connected Firms Jour nal of Accounting Research 52 nº1 p 107167 httpsdoiorg1011111475679X12032 Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 42 Kinas P G Andrade H A 2010 Introdução à análise bayesiana com R Porto Alegre maisQnada Editora Koop G Osiewalski J Steel M 1997 Bayesian efficiency analysis through individual effects Hospital cost frontiers Journal of Econometrics 76 77105 httpsdoiorg1010160304 407695017836 Kumbhakar S C Tsionas G T 2005 Measuring technical and allocative inefficiency in the translog cost system a Bayesian approach Journal of Econometrics 126 p 355384 httpsdoiorg101016jjeconom200405006 Lunn DJ Thomas A Best N Spiegelhalter 2020WinBUGS A Bayesian modelling frame work Concepts structure and extensibilityStatistics and Computing10pp 325337 httpsdoiorg101023A1008929526011 Macedo M A S Kelly V L A 2016 Gerenciamento de resultado em instituições financeiras no Brasil Uma análise com base em provisões para crédito de liquidação duvidosa Revista E videnciação Contábil finanças 4 2 p 8296 httpsdoiorg1018405recfin20160206 Maia S C Bressan V G F Lamounier W M Braga M J 2013 Gerenciamento de resultado em cooperativas de crédito no Brasil Brazilian Business Review 10 4 p 96116 httpsdxdoiorg1015728bb20131044 Malaquias F Malaquias R Hwang Y 2018 Understanding the determinants of mobile bank ing adoption a longitudinal study in Brazil Electronic Commerce Research and Applications 30 p 17 httpsdoiorg101016jelerap201805002 Meeusen W Broeck J D 1977 Efficiency estimation from CobbDouglas production func tions with composed error International Economic Review 18 2 p 435444 httpsdoiorg1023072525757 Ntzoufras I 2009 Bayesian Modeling Using WinBUGS New Jersey 1ª ed John Wiley Sons https 1010029780470434567 Pinheiro F A PSavóia J R F Securato J R 2015 Basileia III Impacto para os Bancos no Brasil Revista de Contabilidade e Finanças USP 26 69 p 345361 https1015901808 057x201500720 Sanches F Silva Júnior D Srisuma S 2018 Banking privatization and market strutucture in Brazil a dynamic structural analysis Journal of Economics 49 4 p 936963 httpsdoiorg1011111756217112257 Santos L J Macedo M A S Rodrigues A 2014 Determinantes do nível de divulgação das recomendações do Pilar 3 do acordo de Basileia 2 nas demonstrações financeiras de institui ções bancárias que atuam no Brasil Brazilian Business Review 11 1 p 2650 httpsdxdoiorg1015728bb20141112 Santos WP Rodrigues R N Anjos L C M Tavares M F N 2019 Análise da influência do cenário econômico interno e externo na prática de income smoothing em bancos brasileiros Re vista Evidenciação Contábil Finanças 7 1 p 7893 httpsdoiorg1022478ufpb2318 10010v0n037439 Santos W P 2022 Adeus Floresta Amazônica Um Estudo Sobre o Desmatamento na Amazônia Legal à Luz do Gasto Público Estadual e dos Fatores Socioeconômicos Locais Revista Evidenci ação Contábil Finanças 102 4263 httpsdoiorg1022478ufpb231810012022v10n257591 Sousa LO 2018 Gerenciamento de resultado em bancos Indícios relacionados à aversão à di vulgação de prejuízos Revista Evidenciação Contábil Finanças 6 1 p 83100 httpsdoiorg1018405recfin20180105 Tecles P L Tabak B M 2010 Determinants of bank efficiency the case Brazil European Jour nal of Operational Reseach 207 p 15871598 httpsdoiorg101016jejor201006007 Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 43 Ventura E C F Fontes Filho J R Soares M M 2009 Governança Corporativa Diretrizes e me canismos para fortalecimento da governança em cooperativas de crédito Brasília BCB 257 Wooldridge J M 2010 Econometric analysis of cross section and panel data Massachusetts Institute of Technology United States 2ª Ed Zhag X 2000 A Monte Carlo study on the finite sample properties of the Gibbs sampling meth od for a stochastic frontier model Journal of Productivity Analysis 14 p 7183 httpsdoiorg101023A1007895912705 Citation Shen Bingbing Aleksandr Aleksandrovich Perfilev Lidiya Pavlovna Bufetova and Xueyan Li 2023 Bank Profitability Analysis in China Stochastic Frontier Approach Journal of Risk and Financial Management 16 243 https doiorg103390jrfm16040243 Academic Editor Yong Tan Received 13 March 2023 Revised 7 April 2023 Accepted 13 April 2023 Published 16 April 2023 Copyright 2023 by the authors Licensee MDPI Basel Switzerland This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution CC BY license https creativecommonsorglicensesby 40 Journal of Risk and Financial Management Article Bank Profitability Analysis in China Stochastic Frontier Approach Bingbing Shen 1 Aleksandr Aleksandrovich Perfilev 1 Lidiya Pavlovna Bufetova 1 and Xueyan Li 2 1 Department of Economics Novosibirsk State University 2 Pirogov Str 630090 Novosibirsk Russia aperfilevgnsuru AAP lidabufgmailcom LPB 2 School of Economics and Business Administration Heilongjiang University Harbin 150080 China lixueyan0451aliyuncom Correspondence shenbingbing724gmailcom Abstract Chinas banking system has a relatively high level of state control while an important task in regulating the banking system is to manage the profitability of banks Using the stochastic frontier approach to assess the profitability of commercial banks not only allows for the banks ability to generate profits relative to the leading banks in the industry to be assessed but also takes into account the specifics of the management technologies used and the influence of the market environment This article analyzes the profitability of the Chinese banking system for the period 20122020 using the stochastic frontier approach from the position of the central bank The specifics of the analysis from the banks perspective imply a focus on the position of most banks regarding the level of best practices and trends in changing the overall level of profitability Analysis may be of interest to banking regulators and researchers In general the Chinese banking system demonstrates a high level of profit efficiency and cost efficiency although the dynamics of these indicators are negative The reason for the negative dynamics is a decrease in the economic growth rate of the economy the instability of the financial market and ongoing reforms Stateowned commercial banks are becoming highly profitable while national jointstock commercial banks are facing increasing competition and reducing efficiency of profitability City and rural commercial banks maintain a high level of profitability due to state support Keywords bank profitability stochastic frontier approach commercial bank ROA ROE profit efficiency cost efficiency JEL Classification G21 G31 1 Introduction The banking systems of developing countries tend to have a relatively high level of state control Many scholars believe that this inhibits competition and the efficient redistribution of financial resources according to Lee et al 2020 and Le and Ngo 2020 According to Huang et al 2021 and Mai et al 2022 developing countries should pursue a policy of deregulation and develop market mechanisms However the 2008 crisis and subsequent prolonged stagnation in the banking sector of developed countries against the background of Chinas significant economic success force a new look at the regulation of financial institutions competently combining administrative and market methods One important condition that central banks observe when regulating the banking system is the need to keep commercial banks interested in making profits Control over the profitability of credit institutions can help in understanding how effectively banks of different types and sizes function whether they need help and whether changes in the existing mechanisms of market regulation are necessary Central banks evaluate the profitability of individual banks using indicators such as return on assets and return on J Risk Financial Manag 2023 16 243 httpsdoiorg103390jrfm16040243 httpswwwmdpicomjournaljrfm J Risk Financial Manag 2023 16 243 2 of 20 capital and also pay attention to the share of profitable banks and the profitability of banks of different sizes and types as indicated by Capraru and Ihnatov 2014 and Lin and Zhang 2009 Statistical and regression methods are used to assess the degree of influence of banking determinants on profitability Djalilova and Piesse 2016 Saona 2016 Tan 2017 This allows for us to understand the extent to which the central banks regulatory methods and changes in the economy will affect the activities of credit institutions It is important for the central bank to make sure that commercial banks are able to generate the expected profits for their owners and are interested in developing the banking business To organize profitability control the central bank needs to develop appropriate methods of analysis The profitability of the banking sector has a direct impact on the effective use of financial resources and the rational allocation of public resources Athanasoglou et al 2008 The profitability of banks is a key element in ensuring the sustainability of the banking sector de Guindos 2019 Profitability measures the overall level of development and sustainable efficiency of commercial banks VeraGilces et al 2020 The need to be profitable is spelled out in the Commercial Bank Law of China1 the need to control the profitability of banks follows from the law on the Peoples Bank of China2 In recent years the profitability of commercial banks in China has significantly decreased as a result of slowing economic growth increased competition from nonfinancial organizations falling rates on financial markets and tightening capital adequacy requirements The banking sector regulator the Peoples Bank of China began to tighten capital adequacy requirements in 2011 and another liberalization of interest rates occurred in 2015 This complicates the activities of banks and requires efforts to ensure that commercial banks comply with regulatory requirements on the one hand and on the other hand retain the ability to operate on market principles and above all create profits The need to assess profitability from the position of regulatory and supervisory au thorities and identify trends and reasons for changing the behavior of banks requires the development of scientific approaches that allow for these problems to be solved As the results of a study by Chinese and foreign authors show the stochastic frontier approach helps to significantly improve the assessment of the profitability of commercial banks from the perspective of their technological capabilities and the influence of managerial and environmental factors A feature of the stochastic frontier approach is the possibility to construct the produc tion possibilities frontier of companies and determine each companys distance from this frontier Thus it becomes possible to evaluate the effectiveness of all or most firms in the industry The application of the stochastic frontier approach to assess the profitability of the Chinese banking sector requires that local features are taken into account This is due to the allocation of groups of banks that focus on certain tasks and types of clients and have different levels of access to resources and different opportunities for the placement of assets These groups are the conductors of the monetary policy of the state in certain areas of the economy We chose the period from 2012 to 2020 for this study This period is characterized by important changes in the economy and the banking sector In 2013 the Chinese government announced the launch of projects related to the export of capital to the countries of Asia Africa and Latin America3 During this period there was a slowdown in economic growth a significant decrease in interest rates in financial markets the development of banking technologies on a digital basis and the emergence of shadow banking The noted changes affected the activities of banks and forced them to adapt to new conditions which inevitably affected their profitability The movement of capital especially within the framework of international and na tional infrastructure projects contributed to the development of systemic banks and in fluenced their growth Jointstock banks also received the opportunity to increase the volume of their operations by servicing the growth of large companies the growth of household incomes and their operations Urban and rural banks were stimulated to grow J Risk Financial Manag 2023 16 243 3 of 20 by participating in programs to promote the development of urban and rural areas Foreign banks mainly paid attention to the growing integration of the Chinese economy into the world economy Chinese and foreign scientists have paid attention to assessing the effectiveness of the banking system in different periods of its development Fu and Heffernan 2007 One can single out works Horrocks et al 2008 that assess the success of Chinese reforms in the period up to 2012 They draw conclusions regarding the beginning of the process of transformation of the national banking system to market conditions We did not find works on a later period although serious changes occurred at this time both in the PRC and in the world economy One feature of our analysis is an attempt to analyze the profitability of Chinese banks in the period from 2012 to 2020 taking into account internal specifics from the position of the central bank of China This specificity is associated with the following features the presence of different groups of banks with different levels of involvement in state regulation the implementation of financial market reforms which contributed to a radical reduction in market rates the secrecy of data on the financial markets of China Therefore in order to optimize the commercial profitability and profit efficiency of the Chinese banking sector it is necessary to manage and improve the quality of banking assets Therefore this work is of great importance when studying the performance of listed banks in China and the factors influencing this The stochastic frontier approach can significantly improve the analysis of the prof itability of the banking system For China this is of particular importance because of the need to evaluate the results of both reforms and the results of economic regulation which may affect the interests of commercial banks The analysis of the profitability of the banking system should be based on its current level of profitability which is determined by the current frontal boundary The deviation of banks from this boundary allows for profitability trends to be determined When analyzing profitability it is important to consider the state of costs which will allow us to determine the reason for the change in the state of profitability 2 Characteristics of the Chinese Banking System from 2012 to 2020 21 The Structure of the Chinese Banking System Chinas banking system is characterized by a high degree of centralization and the presence of banks of various types Mai et al 2022 Half of the assets of the banking system belong to large banks The share of foreign banks is insignificant Big banks also hold more than 60 of deposits and up to 50 of loans The big banks are the core of the banking system and they should be the focus of economic analysis Basically these are large national jointstock commercial banks which are owned by the state and are called state commercial banks4 SO Their shares are traded on various world stock exchanges The peculiarity of stateowned commercial banks is their focus on the financing of large infrastructure and projects that are important to the country However they successfully compete with other banks carrying out banking operations with households and firms Political banks PB5 are directly managed by the state and mainly provide banking services to stateowned companies Their purpose is to stimulate the development of specific industries Political banks do not operate with the aim of maximizing profits National jointstock commercial banks NJ6 meet the financing and savings needs of enterprises and households they contribute to the formation of a competitive banking system These banks are majorityowned by the state either through government agencies or stateowned enterprises These are relatively large credit institutions that have been corporatized by government order with the aim of creating marketoriented organizations without government administration for the benefit of shareholders Foreign commercial banks FB established as branches subsidiaries or independent banks with foreign participation focus on servicing projects involving the combination of domestic and foreign enterprises they can help the customer develop a joint lending J Risk Financial Manag 2023 16 243 4 of 20 structure for domestic and foreign enterprises and provide services to customers related to foreign trade operations Rural commercial banks RB are jointstock local financial institutions established by farmers rural industrial and commercial enterprises and corporate entities Their main objective is to provide financial services for local farmers agriculture and rural economic development and to promote the coordinated development of the urban and rural economy All the rural cooperative banks have been converted into rural commercial banks City commercial banks CB used to exist as urban credit cooperatives Since 1995 the state council of China decided to establish city commercial banks in large and medium sized cities Most city commercial banks show flexibility and responsiveness to market changes compared to large banks creating a strong attraction for enterprises with urgent capital needs City commercial banks have the advantage that they are closely linked to the local economy and have a more detailed and accurate understanding of the lending capacity and performance of local customers The government can stimulate certain groups of banks to develop Infrastructure projects state commercial banks play an important role here to increase the wealth of rural regions rural commercial banks play an important role here and to develop business in cities urban commercial banks Therefore the profitability of some banks can be high not only because of their efficient operation but also as a result of incentives from the state to promote their development Between 2012 and 2020 Chinas banking system underwent significant changes The period began with a slowdown in economic growth and a search for new ways to develop the economy These paths were related to capital exports and implementation of large scale infrastructure projects Li et al 2023 The crash of 20152016 revealed problems that had accumulated in the financial sector which required decisive intervention by the government7 In 2018 trade relations with the US were deteriorating leading to a decline in exports Yang et al 2023 At the same time foreign commercial banks were sharply reducing service revenues During this period the mobile internet became increasingly popular and mobile payments and mobile ecommerce were becoming a way of life The financial innovations represented by the mobile internet are having an impact on the banking industry The work of Scotta et al 2017 showed that digital innovation has had a significant impact on the longterm profitability of banks The authors of Zhao et al 2022 investigated the impact of fintech innovation on the performance of Chinese banks They found that fintech innovation reduces bank profitability and overall asset quality which is more pronounced among large stateowned commercial banks but improves capital adequacy and governance efficiency 22 Profitability of Chinese Banks in the Period 20122020 Chinas economic growth has slowed since 2012 and longstanding structural con tradictions continue to emerge declining export prices for manufactured goods lower real enterprise profits lower fiscal revenue growth and increased economic risk potential Li 2020 In July 2013 the lending rates for banks and other financial institutions were fully liberalized Banks were allowed to set their own lending rates and on 22 November 2014 the Peoples Bank of China reduced the RMB lending and deposit rates of finan cial institutions using the base rate In 2015 deposit insurance rules were announced Lee and Hsieh 2013 and China formally established a deposit insurance system Despite a slowdown in economic growth Chinas interest rate liberalization reform was completed with further expansion of floating range of deposit rates and the introduction of a deposit insurance system We show in Figure 1 average ROA and ROE for Chinas banking sector decline over the period 20132020 As a result of interest rate liberalization bank profitability has declined Sun 2020 but the level of decline is in line with the standards of developed countries J Risk Financial Manag 2023 16 243 5 of 20 J Risk Financial Manag 2023 16 x FOR PEER REVIEW 5 of 21 Figure 1 Evolution of average ROA and ROE in Chinas banking sector over the period 20132020 Source authors preparation httpswwwtheglobaleconomycomChina accessed on 28 Decem ber 2022 Notes abscissa axisyears left ordinate axisaverage return on assets of Chinas banking system right ordinate axisaverage return on equity of Chinas banking system Interest rate liberalization Li et al 2021 has allowed banks the autonomy to set their own deposit and lending rates In a highly competitive environment this has narrowed the spreads on the yield differential between attracted and lent resources and negatively affected banks profitability Since 2012 the Chinese government has understood that the further development of the economy and society cannot be confined to the production of goods and must move to a new level the export of capital The Belt and Road Initiative is a largescale project to improve regional cooperation by improving connectivity between countries along the an cient Silk Road and beyond Li et al 2023 As of February 2019 the Belt and Road Initia tive has triggered an increase in transportation infrastructure investment in 96 countries along the Belt and Road since its inception in 2013 To service the construction of the One Belt One Road initiative Chinese banks are strengthening their financial support by de veloping relevant strategic plans improving the structure of their institutions and service network facilitating credit support for major projects along the One Belt One Road re gion and implementing business innovation institutional innovation and managerial in novation During the period under study banks in China had competitors in the form of com panies that along with their core business offer a range of banking services to customers We are talking primarily about the Alipay platform and Yue Bao which have had a pro found impact on the financial industry but there are also other competitors Zhao et al 2022 The growth of internet financing has attracted a significant portion of depositors funds to the companies in question which will undoubtedly put significant pressure on banks whose main form of income is interest For central banks ROA and ROE are the most important indicators of profitability as they show how banks generate profits under regulatory conditions In the chart below we analyze the average return on assets ROA how much net profit a bank generates per unit of assets and average the return on equity ROE the banks ability to generate net profit from its net assets for commercial banks in China over the period 20122020 As noted there has been a steady decline in bank asset returns over the past decade but this has varied across bank types There is a spread between rural banks urban banks and jointstock banks possibly related to the stimulus processes undertaken by the gov ernment National jointstock commercial banks show a slower rate of decline than others which suggests a relative increase in return on assets relative to other types of banks City commercial banks show a steeper decline in returns than others they are perhaps nega tively influenced by the processes related to shadow banking and intrabank competition as discussed above The Central Bank should perhaps pay attention to the development 166 155 134 117 115 103 104 096 2667 2431 1884 1649 1599 1369 1311 1173 0 05 1 15 2 0 5 10 15 20 25 30 2013 2014 2015 2016 2017 2018 2019 2020 ROAin percent ROEin percent Year Bank return on assetsROA Bank return on equityROE Figure 1 Evolution of average ROA and ROE in Chinas banking sector over the period 2013 2020 Source authors preparation httpswwwtheglobaleconomycomChina accessed on 28 December 2022 Notes abscissa axisyears left ordinate axisaverage return on assets of Chinas banking system right ordinate axisaverage return on equity of Chinas banking system Interest rate liberalization Li et al 2021 has allowed banks the autonomy to set their own deposit and lending rates In a highly competitive environment this has narrowed the spreads on the yield differential between attracted and lent resources and negatively affected banks profitability Since 2012 the Chinese government has understood that the further development of the economy and society cannot be confined to the production of goods and must move to a new level the export of capital The Belt and Road Initiative is a largescale project to improve regional cooperation by improving connectivity between countries along the ancient Silk Road and beyond Li et al 2023 As of February 2019 the Belt and Road Initiative has triggered an increase in transportation infrastructure investment in 96 countries along the Belt and Road since its inception in 2013 To service the construction of the One Belt One Road initiative Chinese banks are strengthening their financial support by developing relevant strategic plans improving the structure of their institutions and service network facilitating credit support for major projects along the One Belt One Road region and implementing business innovation institutional innovation and managerial innovation During the period under study banks in China had competitors in the form of compa nies that along with their core business offer a range of banking services to customers We are talking primarily about the Alipay platform and Yue Bao which have had a profound impact on the financial industry but there are also other competitors Zhao et al 2022 The growth of internet financing has attracted a significant portion of depositors funds to the companies in question which will undoubtedly put significant pressure on banks whose main form of income is interest For central banks ROA and ROE are the most important indicators of profitability as they show how banks generate profits under regulatory conditions In the chart below we analyze the average return on assets ROA how much net profit a bank generates per unit of assets and average the return on equity ROE the banks ability to generate net profit from its net assets for commercial banks in China over the period 20122020 As noted there has been a steady decline in bank asset returns over the past decade but this has varied across bank types There is a spread between rural banks urban banks and jointstock banks possibly related to the stimulus processes undertaken by the government National jointstock commercial banks show a slower rate of decline than others which suggests a relative increase in return on assets relative to other types of banks City commercial banks show a steeper decline in returns than others they are perhaps negatively influenced by the processes related to shadow banking and intrabank competition as discussed above The Central Bank should perhaps pay attention to the J Risk Financial Manag 2023 16 243 6 of 20 development of banks in this sector Foreign commercial banks show the movement of asset yields around the average value of 05 This can be regarded as the lower boundary within which the profitability of local banks can be reduced without affecting the financial stability of the credit institution Return on equity ROE considers the level of financial leverage and characterizes the profitability of bank owners reflecting their level of interest in business development The Central Bank is interested in creating the necessary conditions for the banking sector to be attractive to investors The right part of Figure 2 shows the dynamics of ROE for the observed period It can be noted that foreign commercial banks show a level in the range of 3 Chinese banks show a decline in ROE which as noted was affected by the general decline in interest rates in the economy The highest level of return on capital is shown by national jointstock commercial banks which seems to be due to their more effective management of the capital structure This shows the success of measures to introduce market mechanisms into the banking sector Rural commercial banks demonstrate good adaptation to market conditions They manage to compensate for the overall decline in return on assets and show a weaker decline in ROE than other banks Urban commercial banks show similar ROA and ROE dynamics J Risk Financial Manag 2023 16 x FOR PEER REVIEW 6 of 21 of banks in this sector Foreign commercial banks show the movement of asset yields around the average value of 05 This can be regarded as the lower boundary within which the profitability of local banks can be reduced without affecting the financial stabil ity of the credit institution Return on equity ROE considers the level of financial leverage and characterizes the profitability of bank owners reflecting their level of interest in business development The Central Bank is interested in creating the necessary conditions for the banking sector to be attractive to investors The right part of Figure 2 shows the dynamics of ROE for the ob served period It can be noted that foreign commercial banks show a level in the range of 3 Chinese banks show a decline in ROE which as noted was affected by the general decline in interest rates in the economy The highest level of return on capital is shown by national jointstock commercial banks which seems to be due to their more effective man agement of the capital structure This shows the success of measures to introduce market mechanisms into the banking sector Rural commercial banks demonstrate good adapta tion to market conditions They manage to compensate for the overall decline in return on assets and show a weaker decline in ROE than other banks Urban commercial banks show similar ROA and ROE dynamics a b Figure 2 a Average return on assets by bank type b average return on equity by bank type Source authors preparation Notes left axis of ordinatesaverage return on assets for a group of banks abscissa axisyears right axis of ordinatesaverage return on equity for a group of banks abscissa axisyears categoriesgroups of banks state commercial banks SO national joint stock commercial banks NJ foreign commercial banks FB rural commercial banks RB urban com mercial banks CB Thus we can observe a decline in profitability indicators in the banking system of China over the observed period Most experts Huang 2020 see this as positive associated with economic development and risk reduction In this case the use of SFA which allows us to understand how banks are positioned relative to the production capacity boundary allows for us to assess the effectiveness of banks in generating profits more accurately than traditional methods 3 Literature Review on Banking Profit Efficiency and Cost Efficiency In practice two boundary methods are the most popular the stochastic frontier ap proach SFA Aigner et al 1997 and data envelopment analysis DEA Charnes et al 1978 The SFA method is parametric it describes the inputoutput relationship using pro duction functions such as the CobbDouglas function The DEA method determines the production possibility boundary by constructing linear combinations of input and output parameters Both methods show similar results and the choice of one of them can be re lated both to the preference of the researcher and the specifics of the data Avkiran 2011 Figure 2 a Average return on assets by bank type b average return on equity by bank type Source authors preparation Notes left axis of ordinatesaverage return on assets for a group of banks abscissa axisyears right axis of ordinatesaverage return on equity for a group of banks abscissa axisyears categoriesgroups of banks state commercial banks SO national joint stock commercial banks NJ foreign commercial banks FB rural commercial banks RB urban commercial banks CB Thus we can observe a decline in profitability indicators in the banking system of China over the observed period Most experts Huang 2020 see this as positive associated with economic development and risk reduction In this case the use of SFA which allows us to understand how banks are positioned relative to the production capacity boundary allows for us to assess the effectiveness of banks in generating profits more accurately than traditional methods 3 Literature Review on Banking Profit Efficiency and Cost Efficiency In practice two boundary methods are the most popular the stochastic frontier approach SFA Aigner et al 1997 and data envelopment analysis DEA Charnes et al 1978 The SFA method is parametric it describes the inputoutput relationship using production functions such as the CobbDouglas function The DEA method determines the production possibility boundary by constructing linear combinations of input and output parameters Both methods show similar results and the choice of one of them can be related both to the preference of the researcher and the specifics of the data Avkiran 2011 Wang et al 2014 Henriques et al 2020 Since the late 1980s several studies have J Risk Financial Manag 2023 16 243 7 of 20 been conducted to measure the profit efficiency and cost efficiency of banks Estimates of banks profit efficiency and cost efficiency are important and the results of the studies have important implications for bank practitioners and policymakers Silva et al 2017 DEA is a nonparametric approach and there is no concept of a production function in DEA analysis but there is the concept of a production frontier The DEA method determines the production possibility frontier by constructing linear combinations of input and output parameters Avkiran 2011 Using the DEA method to study the efficiency of 16 major commercial banks during banking reform in China 20032011 Wang et al 2014 concluded that the overall efficiency of stateowned commercial banks was higher than that of jointstock commercial banks before the reform while the efficiency gap between stateowned and jointstock commercial banks narrowed after the reform The SFA method is parametric and uses production functions such as the Cobb Douglas function to describe the inputoutput relationship Using the stochastic frontier approach authors Pasiouras et al 2009 studied 615 commercial banks operating in 74 countries over the period 20002004 The research showed that stronger banking su pervision increases the cost and profit efficiency of banks The authors of Ngan 2014 measured the cost and profit efficiency of 45 Vietnamese commercial banks in the period 20072012 using the stochastic frontier analysis SFA method and the results showed that stateowned commercial banks SCOBs are more efficient in terms of profit efficiency than other domestic commercial banks JSCBs and foreign banks while foreign banks are ahead of stateowned banks in terms of cost efficiency The authors Aielloa and Bonanno 2016 using the stochastic frontier analysis SFA method showed that cost efficiency was not related to market risk in Italian banks in 20062011 while profit efficiency was related to market risk The cost efficiency of commercial banks is inversely related to loan diversification The profits of Italian commercial banks are negatively affected by size The authors Dong et al 2016 examined the cost and profit efficiency of four types of commercial banks in China over the period from 2002 to 2013 The study shows that cost and profit efficiency increased for stateowned banks jointstock banks foreign banks and city banks and that banks were more profitefficient than costefficient Foreign banks are the most costefficient but the least profitefficient Using the stochastic frontier approach Bitar et al 2018 analyzed 1992 banks in 39 OECD countries over the period 19992013 and found that higher capital ratios may harm the banking efficiency and profitability of highly liquid banks In summary our review of the literature on profit efficiency and cost efficiency in the banking sector reveals some mixed and contradictory results and research gaps Since 2012 rural commercial banks have played an increasingly important role in the Chinese banking system but existing studies have underresearched rural commercial banks in China Commercial banks in China vary in their ability to effectively apply banking technol ogy The choice of methodology may be related to the specifics of the data In our study we added the classification of Chinese commercial banks in China to the existing research analysis of rural commercial banks and used the SFA methodology because its results are illustrative and consider the inefficiency term 4 Methodology 41 Empirical Method In our study we used the SFA method because its results are quite clear it is robust to outliers and it allows for us to build a boundary based on the most widespread use of the technology Commercial banks in China differ in their effective application of banking technology and in their access to banking product markets as well as having different sizes In this case focusing on the most widespread use of technology would provide a fairer estimate Note that this assessment would be fair to banks when considered as separate entities without taking into account their importance to the banking sector This is important when creating a fair environment for all participants in the banking sector The stochastic frontier approach originally relied on a neoclassical theory of the firm which attempts to explain the relationship between production profits and costs Aigner et al 1997 Coelli et al 2005 Firms were assumed to operate in perfect outputcost markets and input prices were defined as exogenous Competition leads to the price equalization of inputs and outputs Subsequently it was argued that profitability cannot be explained by production factors alone but also by factors related to management and the behavior of key stakeholders Silva et al 2017 Lee and Hsieh 2013 The application of frontier methods requires compliance with a number of requirements that are difficult to capture in real practice all firms in the analysis are competing the products are homogenous all firms produce identical products and use identical resources and firms have an identical function in converting resources into final products The stochastic frontier approach considers the set of firms as a set of technologies which are characterized by the terms input and output where inputs are resources and outputs are the products obtained by using these resources They can be used to determine the boundary of production capacity at which firms use the minimum amount of resources to obtain the maximum volume of production Battese and Coelli 1995 Berger and Humphrey 1997 emphasized the advantages of effective frontier approaches compared to traditional performance measures such as return on assets because they take into account both the banks respective inputs and outputs and differences in the prices of inputs The SFA method assumes that not all components of the residuals of the production function built on the panel data are of a random nature and distinguishes nonrandom u and random v components in the residuals yit fxit β vit uit 1 vit N0 σ2v uit N0 σ2u 2 where yit denotes the production at observation t 1 2 T xit is a 1k vector of the values of known functions of production inputs and other explanatory variables associated with the ith firm at the tth observation β is a k1 vector of unknown parameters to be estimated8 vit are random errors which are distributed independently of the uit uit are nonnegative random variables associated with the technical inefficiency of production which are assumed to be independently distributed In most works on similar topics it is assumed that the random component v has a normal distribution with a mean of zero and component u has a seminormal distribution both quantities are independent The parameter u is used to measure the production efficiency of individual organizations as part of their cumulative total namely to estimate the difference between the output of an individual company and the maximum possible output The efficiency score is estimated using a stage stochastic frontier model Battese and Coelli 1995 Coelli et al 2005 in which the effect of environmental variables on inefficiency is modeled Historically the scientific basis for the construction of production functions was the neoclassical theory of the firm according to which in a perfect market firms had equal access to both resources and sales and competed with each other by reducing individual costs As was shown later this was not consistent with practice in particular the role of agents was not taken into account The development of management theory led to an understanding of the role that managerial factors play in firm performance The behavioral theory results led to an understanding of the importance of the harmonious development of an organization consideration of the interests of all interested groups and influence of external factors This is especially important for banks because the activity of these organizations is socially important it is based on the trust of clients is regulated and requires careful consideration of stakeholders opinions9 The impact of management and environmental factors can be accounted for in two ways by adding determinants to the function Aiello and Bonanno 2013 by estimating J Risk Financial Manag 2023 16 243 9 of 20 the efficiency derived from the function using the production parameters and then esti mating the impact of management determinants and environmental factors on efficiency Dong et al 2016 Based on our data we adopted the first approach adding management parameters Z and environmental factors E to the function The effect of environmental variables eg GDP inflation rate on inefficiency were modeled together with a profit or cost frontier model The profit and cost frontier models are represented below Profitit Fpy w β Z Eevpeup 3 Costit Fcy w β Z Eevceuc 4 where y is the product service produced w is the price β is the function parameters Profitit is the profit of object i at time t and Costit is the total cost of object i at time t Z denotes the control parameters vp and vc represent random components in the statistical error of Equations 3 and 4 up and uc represent nonrandom components in the statistical error of Equations 3 and 4 associated with inefficiency E denotes environmental parameters The SFA method distinguishes between the nonrandom u and random v components of the residuals The random component v has a normal distribution and the component u has a seminormal distribution with both quantities being independent There are two different versions of the concept of profit efficiency namely standard profit efficiency SPE and alternative profit efficiency APE Berger and Mester 1997 Standard profit efficiency indicates how close a bank is to maximizing profits at a specified input and output price It assumes that output markets are perfectly competitive Alterna tive profit efficiency assumes that there is imperfect competition and therefore banks have some market power APE measures how close a bank is to earning maximum profit by adjusting output prices and input quantities given a specific level of output and a specific input price Due to the absence of reliable data on banks output prices and the existence of imperfect competition in the Chinese banking sector we used alternative profit efficiency APE analysis in our study From Equations 3 and 4 profit efficiency is the ratio of actual observed bank profits to the maximum level of potential efficient bank profits Cost efficiency is the ratio of the minimum level of potentially efficient banks costs to the actual observed level of banks costs Profit efficiency and cost efficiency have a value between 0 and 1 Profit efficiency APE Fpy w β Z Eevp Fpy w β Z Eevpeup eup 5 Cost efficiency Fcy w β Z Eevc Fpy w β Z Eevcec euc 6 42 Variable Definitions There is no single approach to deciding which parameters should be used in the profitability function and the cost function Ariff and Luc 2008 Foroutan et al 2015 Gaganis et al 2020 Aiello and Bonanno 2013 Ding and Sickles 2018 Almanidis et al 2019 Le et al 2020 The researcher agrees that the set of these parameters can be the same As a rule there are basic parameters that characterize the technology control parameters and parameters of the environment external environment In the literature on bank efficiency there are many ways to measure the inputs and outputs of banks the production approach the intermediation approach and the operating approach Coelli et al 2005 Our study uses the intermediation approach The intermediation approach proposed by Sealey and Lindley 1977 views the bank as an intermediary that collects customer deposits and grants loans The bank acts as an intermediary collecting funds from savers and converting these funds into profitgenerating items loans etc The approach is J Risk Financial Manag 2023 16 243 10 of 20 consistent with the concept of the bank as a financial intermediary and the data required to implement the approach are readily available Berger and Humphrey 1997 highlighted the advantages of stochastic frontier methods compared to traditional measures of efficiency eg return on assets because they take into account both relevant bank inputs and outputs and differences in input prices they believe that the intermediary approach is more appropriate for measuring the efficiency of the bank as a whole Profit before tax P is used as a measure of profit Total costs C ie the sum of overheads interest costs and other operating expenses are used as a measure of costs There is no consensus as to which parameters should be used in the profitability function and the cost function The opinion of the researchers is that some basic parameters characterize the production parameters the management parameters and the environmental parameters Basic parameters production parameters The basic parameters outline the process of turning resources into final products In general the profitability and cost functions use loans investments and noninterest income as output parameters The input parameters are labor prices and deposit prices Three groups of earning assets were used loans other earning assets and noninterest income comprising mainly fee and commission income and investment income which is consistent with current practice These variables serve as the output parameters of the model The input parameters input prices the price of loanable funds the price of capital described in the model are as follows Table 1 shows the selected variables outputs and inputs total costs C and total profits P Table 1 Definition of the variables included in the cost and profit functions Variables Name Description Outputs y1 Loans Loans y2 Other earning assets Total earning assets minus loans y3 Noninterest income Other operating income Inputs w1 Price of loanable funds Interest expensetotal assets equity w2 Price of capital Overheadsfixed assets Cost C Total costs Overheads interest expense other operating expenses Profit P Total profits Profit before tax Source authors preparation We believe that the stochastic frontier approach to assessing the profitability of com mercial banks not only assesses the banks profitability relative to the leading banks in the industry but also takes into account the specific circumstances of the management techniques used and the market environment Control parameters Risk management is a fundamental element of commercial banks management Central banks regulate risks in the activities of commercial banks using capital adequacy risk as recommended by BASEL 3 Wanke et al 2022 provided a new perspective on the performance of banks in ASEAN member countries arguing that capital adequacy and asset quality especially credit quality are key determinants of banks financial position and profitability Bank management may expose the organization to high risk to provide higher profitability to investors Bank liquidity and bank profitability interact Nicolae et al 2015 AbuAlkheil et al 2018 found that liquidity risk affects the placing of some assets in a nonrecoverable form and reducing liquidity risk leads to lower profitability Klein and Weill 2022 argued that banks profitability is affected by their asset size which promotes economies of scale and larger banks have access to cheaper resources and more reliable funding VeraGilces et al 2020 Table 2 shows the control parameters Table 2 Definition of control variables Variables Name Description z1 CAR Capital adequacy ratio capitalrisk weighted assets z2 RQA Assets quality ratios loan lossesgross loans z3 LA Liquid assets ratio liquid assetsdeposits and short funding z4 TA Bank size total assets in logs Source authors preparation Environmental parameters Environmental parameters are not controlled by bank management but they affect a banks profitability Le and Ngo 2020 and Bitar et al 2018 argued that GDP growth leads to increased demand for production financing and affects the profitability of banks VeraGilces et al 2020 and Nicolae et al 2015 argued that inflation and deregulation affect bank profitability Davis et al 2022 argued that market competition affects bank profitability Dong et al 2016 used time trends to reflect the technological change Table 3 shows the environmental parameters used in the analysis of the banking system Table 3 Definition of environmental variables Variables Name Description e1 GDP GDP in logs e2 INF Inflation rate e3 SMCapitalizationGDP Stock market capitalizationGDP e4 T Year2011 Source authors preparation 43 Model Specification We used the transcendental logarithm translog from Christensen et al 1973 which is the most commonly used functional form in terms of bank efficiency Kumbhakar and Lovell 2000 Our profit efficiency frontier model is as follows lnP β0 i13 βi ln yi γ1 ln W 12 i j13 φij ln yi ln yj 13 μ11 ln W ln W i13 αi ln yi n14 θn zn ε14 εn en u v 7 where y1 y2 y3 are outputs w1 w2 are the price of inputs βi γ1 φij μ11 αi θn εn are the parameters to be estimated Using a translog linear homogeneity also requires standard symmetry φij φji u is the inefficiency of resource use v is random error lnP is the logarithm of total profits The total profits P and input price terms are normalized by one of the input prices w2 in order to impose a linear homogeneity of degree one on the input prices W w1w2 When using the SFA model to analyze profitability the observations in the panel data should not contain negative values To ensure this the profit values in the sample were adjusted based on the results of the study Aielloa and Bonanno 2016 Profit10 figures were transformed by adding the absolute value of the minimum profit Profitmin for each bank plus one After the transformation the right side of Equation 7 looks as follows lnP ln Profit Profitmin 1 w2 8 By substituting profit for cost11 in the formula we can obtain a cost function formula 44 Data Our sample covers 239 Chinese banks from 2012 to 2020 with annual data of 1719 observations in total All banks with complete data for at least eight years were included J Risk Financial Manag 2023 16 243 12 of 20 in the analysis We checked our data against the theoretical requirements and ensured that they were correct incorrect data were removed The sample included 6 stateowned commercial banks 12 national jointstock commercial banks 61 rural commercial banks 130 urban commercial banks and 30 foreign commercial banks see Table 4 The data for China for 20122020 were representative covering more than 95 of the banking sector by asset size The data were mainly taken from the ORBIS international database All financial variables were measured in thousands of US dollars Data on GDP in flation and stock market capitalization were taken from the Global Economy website httpswwwtheglobaleconomycomChinaeconomicgrowth accessed on 28 Decem ber 2022 Table 4 Data characteristics by type of bank Types Number of Banks Abbreviation Average Assets as of 2020 USD Billion State commercial banks 6 SO 3 447 National jointstock commercial banks 12 NJ 759 Foreign commercial banks 30 FB 17 Rural commercial banks 61 RB 34 City commercial banks 130 CB 60 Total 239 Source authors calculations The sample included all stateowned commercial banks national jointstock commer cial banks and most foreign and urban banks Rural commercial banks were represented by the most characteristic banks The sample included banks whose total asset value was more than 95 of the total asset value of the banking system PRC banks vary greatly in size The average size of assets of systemic banks is 3447 billion US dollars many of them top the list of the largest banks in the world Jointstock banks are much smaller but are nevertheless large and play an important role in the financial market Foreign commercial banks are small in size they are not involved in the transformation of assets but mainly focus on service functions The main focus of our study was to analyze the efficiency of the profitability and costs of Chinas commercial banks from the perspective of a regulator The regulator is interested in the number of banks that are close to the production possibilities frontier and banks that are far from this frontier It is important that as many banks as possible can generate profits at the level of the best technologies and management practices Outsiders are the object of analysis of the determinants that affect their weak position The regulator is interested in the dynamics of profitability and cost efficiency which can help to reveal negative or positive trends as well as the reaction of banks to reforms and external influences We did not consider issues related to the fulfillment of the theoretical requirements for the cost function and profitability function in detail We checked our data for compliance with theoretical requirements and made sure they were correct Outliers and incorrect data were removed however this did not significantly affect the sample size 5 Results Constructed regression equations showed high significance for the majority of de terminants see Table 5 which allows us to conclude that we correctly determined the input and output production parameters as well as the control parameters and external parameters12 J Risk Financial Manag 2023 16 243 13 of 20 Table 5 Parameter estimates for stochastic frontier profit function and cost function Variables Parameters Profit Function Cost Function Coefficient Signif Codes Coefficient Signif Codes Constant β0 88906 15309 lny1 β1 2482 0149 lny2 β2 1220 0098 lny3 β3 0143 0017 lnw1w2 γ1 1034 1097 05 lny1ˆ2 ϕ11 0285 0008 lny1 lny2 ϕ12 0409 0006 lny1 lny3 ϕ13 0008 0006 05 lny2ˆ2 ϕ22 0382 0010 lny2 lny3 ϕ23 0000 0005 05 lny3ˆ2 ϕ33 0001 0003 05 lnw1w2ˆ2 µ11 0023 0016 lnw1w2 lny1 α11 0026 0003 lnw1w2 lny2 α21 0043 0002 lnw1w2 ny3 α31 0006 0006 z1 CAR θ1 0001 0002 z2 RQA θ2 0018 0008 z3 LA θ3 0002 0001 z4 logTotalAssets θ4 2955 1268 e1 logGDP ε5 1774 0426 e2 Inflation ε6 0006 0000 e3 Stock marketGDP ε7 0005 0101 e4 T ε8 0120 0002 σ2 σ2v σ2 u σ2 0148 0051 Gamma γ 0651 0740 Mean efficiency 0797 0868 Total number of observations 1719 Notes the Variables column presents the determinants and their combinations that make up Equation 7 the Parameters column presents the coefficient notation of Equation 7 and the corresponding variables the profitability functions and cost functions blocks present the coefficient values of the parameters column Coefficient and the significance13 level estimates Significance respectively Signif codes 0 0001 001 005 01 1 Source authors calculations To assess the significance of efficiency factors we used the parameter γ Battese and Coelli 1992 1995 which is the ratio of the change due to inefficiency to the total change in the residuals of the regression model This indicates that the banks deviation from the efficiency frontier is due to inefficiency in the use of available resources γ σ2 u σ2v σ2 u σ2 u σ2 9 Σ2 v and σ2 u represent the random error term v and the nonnegative error term u with a value γ between 0 and 1 Le et al 2020 Aiello and Bonanno 2013 The closer the rate of change γ is to 1 the smaller the effect of the random error term v which means that γ is determined by the nonnegative error term u Conversely the closer the rate of change γ is to 0 the more that the profit function and the cost function fail the test The high value of the estimated γ parameter 065 for the profit function and 074 for the cost function indicates that the vast majority of the residual variation is due to profit inefficiency and cost inefficiency effects and strongly supports the use of the stochastic frontier model rather than the standard ordinary least squares OLS model The average profit efficiency of the banks in the entire sample was 08 indicating that most Chinese banks were close to the maximum possible profitability However the average cost efficiency of banks was 087 This suggests that banks are managing their costs efficiently The difference in average profitability and cost efficiency can be explained J Risk Financial Manag 2023 16 243 14 of 20 by the particularities of the period under study during which interest rates and capital adequacy requirements changed dramatically Deposits which account for the majority of commercial banks resources approximately 70 are classified as riskfree Interest rates on loans which also accounted for a large proportion of operating assets included a risk premium During this period reducing financial risk would have had a greater impact on bank profits than on costs To analyze the dynamics of profit efficiency and cost efficiency graphs were drawn up to show how these indicators change over time for different types of banks This is important because different types of banks have different abilities regarding the use of banking technology different levels of access to resources and different asset allocation capabilities For each group of banks the average values14 of the profit and cost efficiency indicator in a given year were estimated Figure 3 on the left shows the dynamics of average profit efficiency by bank type and on the right the average cost efficiency by bank type As can be seen from the left panel the dynamics of profit efficiency can be divided into two phases The first phase from 2012 to 2016 is characterized by growth and then the profit efficiency of Chinese banks starts to decline In our view this can be explained by the 20152016 stock market crisis in the Chinese financial market and the interest rate liberalization reforms Foreign banks were less affected by this process Before 2014 Chinese banks showed a multidirectional change in terms of cost efficiency Between 2014 and 2016 cost efficiency declined sharply Then as the financial markets stabilized an increase in cost efficiency could be seen in Chinese banks J Risk Financial Manag 2023 16 x FOR PEER REVIEW 14 of 21 closer the rate of change γ is to 0 the more that the profit function and the cost function fail the test The high value of the estimated γ parameter 065 for the profit function and 074 for the cost function indicates that the vast majority of the residual variation is due to profit inefficiency and cost inefficiency effects and strongly supports the use of the stochastic frontier model rather than the standard ordinary least squares OLS model The average profit efficiency of the banks in the entire sample was 08 indicating that most Chinese banks were close to the maximum possible profitability However the average cost efficiency of banks was 087 This suggests that banks are managing their costs efficiently The difference in average profitability and cost efficiency can be explained by the particularities of the period under study during which interest rates and capital adequacy requirements changed dramatically Deposits which account for the majority of commercial banks resources approximately 70 are classified as riskfree Interest rates on loans which also accounted for a large proportion of operating assets included a risk premium During this period reducing financial risk would have had a greater impact on bank profits than on costs To analyze the dynamics of profit efficiency and cost efficiency graphs were drawn up to show how these indicators change over time for different types of banks This is important because different types of banks have different abilities regarding the use of banking technology different levels of access to resources and different asset allocation capabilities For each group of banks the average values14 of the profit and cost efficiency indicator in a given year were estimated Figure 3 on the left shows the dynamics of average profit efficiency by bank type and on the right the average cost efficiency by bank type As can be seen from the left panel the dynamics of profit efficiency can be divided into two phases The first phase from 2012 to 2016 is characterized by growth and then the profit efficiency of Chinese banks starts to decline In our view this can be explained by the 20152016 stock market crisis in the Chinese financial market and the interest rate liberalization reforms Foreign banks were less affected by this process Before 2014 Chinese banks showed a multidirectional change in terms of cost efficiency Between 2014 and 2016 cost efficiency declined sharply Then as the financial markets stabilized an increase in cost efficiency could be seen in Chinese banks a b Figure 3 a Average profit efficiency by bank type b average cost efficiency by bank type Source authors preparation Notes axis of ordinatesaverage efficiency per group of banks axis of abscissayears categoriesgroups of banks as in Table 4 Stateowned commercial banks which had the highest profit efficiency relative to other types of banks during the study period are steadily declining Furthermore as can be seen in Figure 4 there is a clear distinction in size between stateowned commercial banks and other banks which are significantly larger and do not deviate much from the average efficiency values except for the Postal Savings Bank of China the youngest state Figure 3 a Average profit efficiency by bank type b average cost efficiency by bank type Source authors preparation Notes axis of ordinatesaverage efficiency per group of banks axis of abscissayears categoriesgroups of banks as in Table 4 Stateowned commercial banks which had the highest profit efficiency relative to other types of banks during the study period are steadily declining Furthermore as can be seen in Figure 4 there is a clear distinction in size between stateowned commercial banks and other banks which are significantly larger and do not deviate much from the average efficiency values except for the Postal Savings Bank of China the youngest stateowned bank The decline in profit efficiency intensified after 2018 which can be attributed to rising problems in Chinas international investments Belt and Road Initiative both for political reasons such as trade issues between the US and China and the general volatility of global markets The cost efficiency of stateowned commercial banks has been relatively stable but not the best Stateowned commercial banks do not need to improve their cost management and competitive resources They have a high reputation for household and corporate activities and can also rely on national policy support J Risk Financial Manag 2023 16 243 15 of 20 J Risk Financial Manag 2023 16 x FOR PEER REVIEW 15 of 21 owned bank The decline in profit efficiency intensified after 2018 which can be attributed to rising problems in Chinas international investments Belt and Road Initiative both for political reasons such as trade issues between the US and China and the general volatility of global markets The cost efficiency of stateowned commercial banks has been relatively stable but not the best Stateowned commercial banks do not need to improve their cost management and competitive resources They have a high reputation for household and corporate activities and can also rely on national policy support Figure 4 Profit efficiency of different types of banks Source authors preparation Notes ordinate axisthe average value of efficiency for the group of banks abscissa axisyears categoriesgroups of banks according to Table 4 where the size of the marker characterizes the size of bank assets and its representation corresponds to the scale Total Assets at the bottom of the figure National jointstock commercial banks showed the lowest level of profit efficiency and negative dynamics for 20122020 overall the decline has intensified since 2016 It is difficult for jointstock banks which cannot count on serious state support to adapt to changes in a short time moreover they are forced to compete for markets with both systemic and city banks as well as with shadow banking However these banks have the highest cost efficiency It can be concluded that national jointstock commercial banks are able to effectively manage their business but due to drastic changes in financial markets and a slowdown in economic growth they cannot fully control profitability Figure 5 shows that this trend is not typical for all national jointstock commercial banks the general trend was affected by the decrease in the efficiency of some banks It can be seen that over time the variability in both the profit and cost efficiency of national jointstock commercial banks increases relative to the average Figure 5 Cost efficiency of different types of banks Source authors preparation Notes ordinate axisthe average value of efficiency for the group of banks abscissa axisyears categoriesgroups of banks according to Table 4 where the size of the marker characterizes the size of bank assets and its representation corresponds to the scale Total Assets at the bottom of the figure Figure 4 Profit efficiency of different types of banks Source authors preparation Notes ordinate axisthe average value of efficiency for the group of banks abscissa axisyears categoriesgroups of banks according to Table 4 where the size of the marker characterizes the size of bank assets and its representation corresponds to the scale Total Assets at the bottom of the figure National jointstock commercial banks showed the lowest level of profit efficiency and negative dynamics for 20122020 overall the decline has intensified since 2016 It is difficult for jointstock banks which cannot count on serious state support to adapt to changes in a short time moreover they are forced to compete for markets with both systemic and city banks as well as with shadow banking However these banks have the highest cost efficiency It can be concluded that national jointstock commercial banks are able to effectively manage their business but due to drastic changes in financial markets and a slowdown in economic growth they cannot fully control profitability Figure 5 shows that this trend is not typical for all national jointstock commercial banks the general trend was affected by the decrease in the efficiency of some banks It can be seen that over time the variability in both the profit and cost efficiency of national jointstock commercial banks increases relative to the average J Risk Financial Manag 2023 16 x FOR PEER REVIEW 15 of 21 owned bank The decline in profit efficiency intensified after 2018 which can be attributed to rising problems in Chinas international investments Belt and Road Initiative both for political reasons such as trade issues between the US and China and the general volatility of global markets The cost efficiency of stateowned commercial banks has been relatively stable but not the best Stateowned commercial banks do not need to improve their cost management and competitive resources They have a high reputation for household and corporate activities and can also rely on national policy support Figure 4 Profit efficiency of different types of banks Source authors preparation Notes ordinate axisthe average value of efficiency for the group of banks abscissa axisyears categoriesgroups of banks according to Table 4 where the size of the marker characterizes the size of bank assets and its representation corresponds to the scale Total Assets at the bottom of the figure National jointstock commercial banks showed the lowest level of profit efficiency and negative dynamics for 20122020 overall the decline has intensified since 2016 It is difficult for jointstock banks which cannot count on serious state support to adapt to changes in a short time moreover they are forced to compete for markets with both systemic and city banks as well as with shadow banking However these banks have the highest cost efficiency It can be concluded that national jointstock commercial banks are able to effectively manage their business but due to drastic changes in financial markets and a slowdown in economic growth they cannot fully control profitability Figure 5 shows that this trend is not typical for all national jointstock commercial banks the general trend was affected by the decrease in the efficiency of some banks It can be seen that over time the variability in both the profit and cost efficiency of national jointstock commercial banks increases relative to the average Figure 5 Cost efficiency of different types of banks Source authors preparation Notes ordinate axisthe average value of efficiency for the group of banks abscissa axisyears categoriesgroups of banks according to Table 4 where the size of the marker characterizes the size of bank assets and its representation corresponds to the scale Total Assets at the bottom of the figure Figure 5 Cost efficiency of different types of banks Source authors preparation Notes ordinate axisthe average value of efficiency for the group of banks abscissa axisyears categoriesgroups of banks according to Table 4 where the size of the marker characterizes the size of bank assets and its representation corresponds to the scale Total Assets at the bottom of the figure The profit and cost efficiency of city commercial banks are similar to those of national jointstock commercial banks There is a plausible explanation for this These banks have similar customers and provide similar services However city commercial banks and national jointstock commercial banks have different custom sizes City banks work with small companies and individuals in smaller cities where there is less competition from jointstock banks and systemic banks The volatility of profit efficiency increases for city commercial banks Figure 4 while the volatility of cost efficiency decreases Figure 5 The profit and cost efficiency of agricultural commercial banks remained high and stable In our opinion the reason for this stability is related to the fact that the customers of J Risk Financial Manag 2023 16 243 16 of 20 these banks are linked to the domestic market and do not depend on external factors Rural commercial banks use rather conservative products such as loans and deposits and less often securities depending on market factors Rural development schemes are subsidized by the state and this provides a stable income for rural commercial banks Figures 4 and 5 show that the assets of rural banks significantly increased over the observation period with little change in the volatility of profitability or cost performance The efficiency of foreign commercial banks was stable and has been increasing since 2015 This can be explained by the fact that their core business is not tied to Chinas lending rates they do not need to adapt and the demand for their services from foreign companies is growing As can be seen from the data in Figures 4 and 5 for foreign banks the variability in both profit efficiency and cost efficiency significantly increased This indicates that foreign banks show different behavioral strategies in the Chinese market and have different competencies related to the characteristics of their customers Thus different types of Chinese banks depending on their characteristics show different behaviors as well as common variations Profitability declined for both external economic and political reasons as well as internal factors In terms of the operation of the market mechanism and the development of competition increased efficiency should be linked to both higher profitability and lower costs Sound regulatory policies should work in tandem with market mechanisms Therefore if we want to assess the quality of banks and the quality of regulatory action in terms of profitability it is necessary to examine the link between efficiency and related indicators eg ROA Figure 6 shows the relationship between return on assets ROA and profit efficiency performance There is a strong positive correlation between return on assets ROA and profit efficiency for national jointstock commercial banks National jointstock commercial banks were established as an important part of Chinas marketbased banking system to measure the role of marketbased approaches in financial reform and banking sector reform Stateowned commercial banks also show high correlation between bank return on assets ROA and profit efficiency However rural commercial banks foreign commercial banks and city commercial banks show no such relationship It can be argued that foreign banks concentrate on service functions such as foreign and domestic trade settlement and do not pursue the goal of increasing profitability Rural and city banks are quite heterogeneous and numerous and operate in distinctive niches cities small towns and rural settlements The profitability of these banks depends on a variety of unique reasons including innovation in banking financial products and government support These banks have a short history and are undergoing a development process the results will manifest in the future J Risk Financial Manag 2023 16 x FOR PEER REVIEW 17 of 21 Figure 6 The relationship between profit efficiency and return on assets ROA for different types of banks Source authors preparation Notes ordinate axisaverage value of return on assets for the group of banks abscissa axisvalue of efficiency categoriesgroups of banks according to Table 4 each point characterizes the position of a particular bank in the coordinate system used while the blue line determines the statistical relationship of data 6 Conclusions The Chinese banking system includes different types of banks with different resource acquisitions asset allocations and technological capabilities This specificity must be taken into account when analyzing profitability The studied period is characterized by instability due to the slowing economic growth financial market instability ongoing banking reforms and reorientation of the economy toward capital exports It is important to understand how the banking system functioned during this period and whether regulatory intervention was necessary The findings suggest that Chinese commercial banks are adapting to the changes brought about by external changes and banking reforms Liu 2017 Li and Liu 2019 and Mott 2022 have shown that interest rate policy affects the profitability of banks When characterizing the dynamics of profitability and cost efficiency by type of bank the following points can be emphasized Stateowned commercial banks lead in terms of profit efficiency cost efficiency is also high After 2015 national jointstock commercial banks had to compete for the market with systemic municipal and shadow banks This harmed their profitability It can be concluded that national jointstock commercial banks were able to manage their business efficiently but could not fully control their profitability due to dramatic changes in financial markets and the slowdown in economic growth The profitability and economic efficiency of city banks are similar to those of joint stock banks City banks work with small companies and individuals in smaller cities where there is less competition from jointstock and stateowed commercial banks Chinas cities have a large concentration of manufacturing logistics and financial institutions all of which are increasing City residents are financially literate and responsive to new banking products This has forced city banks to improve their customer service techniques which has affected their efficiency The profitability and economic efficiency of city commercial banks are similar to that of jointstock banks Urban commercial banks work with small companies and individuals in smaller cities where there is less competition from jointstock and system banks Foreign banks performance has remained stable and has been increasing since 2015 This can be explained by the fact that their core business is not tied to Chinese lending rates they do not need to adapt to Chinese reforms and the demand for services from foreign companies is increasing The profit efficiency and cost efficiency of foreign commercial banks increased between 2012 and 2020 Figure 6 The relationship between profit efficiency and return on assets ROA for different types of banks Source authors preparation Notes ordinate axisaverage value of return on assets for the group of banks abscissa axisvalue of efficiency categoriesgroups of banks according to Table 4 each point characterizes the position of a particular bank in the coordinate system used while the blue line determines the statistical relationship of data J Risk Financial Manag 2023 16 243 17 of 20 6 Conclusions The Chinese banking system includes different types of banks with different resource acquisitions asset allocations and technological capabilities This specificity must be taken into account when analyzing profitability The studied period is characterized by instability due to the slowing economic growth financial market instability ongoing banking reforms and reorientation of the economy toward capital exports It is important to understand how the banking system functioned during this period and whether regulatory intervention was necessary The findings suggest that Chinese commercial banks are adapting to the changes brought about by external changes and banking reforms Liu 2017 Li and Liu 2019 and Mott 2022 have shown that interest rate policy affects the profitability of banks When characterizing the dynamics of profitability and cost efficiency by type of bank the following points can be emphasized Stateowned commercial banks lead in terms of profit efficiency cost efficiency is also high After 2015 national jointstock commercial banks had to compete for the market with systemic municipal and shadow banks This harmed their profitability It can be concluded that national jointstock commercial banks were able to manage their business efficiently but could not fully control their profitability due to dramatic changes in financial markets and the slowdown in economic growth The profitability and economic efficiency of city banks are similar to those of jointstock banks City banks work with small companies and individuals in smaller cities where there is less competition from jointstock and stateowed commercial banks Chinas cities have a large concentration of manufacturing logistics and financial institutions all of which are increasing City residents are financially literate and responsive to new banking products This has forced city banks to improve their customer service techniques which has affected their efficiency The profitability and economic efficiency of city commercial banks are similar to that of jointstock banks Urban commercial banks work with small companies and individuals in smaller cities where there is less competition from jointstock and system banks Foreign banks performance has remained stable and has been increasing since 2015 This can be explained by the fact that their core business is not tied to Chinese lending rates they do not need to adapt to Chinese reforms and the demand for services from foreign companies is increasing The profit efficiency and cost efficiency of foreign commercial banks increased between 2012 and 2020 Thus different types of Chinese banks depending on their characteristics show different behaviors but follow some common trends The decline in profitability is due to both external economic and political reasons as well as internal factors related to changes in the structure of the stock market and financial sector reforms From the regulators perspective the focus should be on the performance of joint stock banks as well as on the factors of deterioration in the efficiency of different types of banks Given the increasing volatility of profitability and cost efficiency it is necessary to understand the position of outsiders and the reasons for their declining efficiency Author Contributions Conceptualization AAP and LPB methodology AAP software BS and AAP validation BS and AAP formal analysis BS and AAP investigation BS resources BS data curation BS writingoriginal draft preparation BS writingreview and editing BS and AAP visualization BS and LPB supervision XL funding acquisition BS All authors have read and agreed to the published version of the manuscript Funding This research was funded by China Scholarship Council grant number 202008230139 The APC was paid by China Scholarship Council grant number 202008230139 Data Availability Statement Data sources are contained within the article Data are freely available and can be accessed by everyone J Risk Financial Manag 2023 16 243 18 of 20 Conflicts of Interest The authors declare no conflict of interest Notes 1 Commercial Banking Law of the Peoples Republic of China httpwwwgovcntest20050628content10576htm accessed on 4 January 2023 2 Law of the Peoples Republic of China on the Peoples Bank of China httpwwwgovcntest20050628content10577htm accessed on 4 January 2023 3 Belt and Road Initiative httpswwwyidaiyilugovcn accessed on 4 January 2023 4 Bank of China China Construction Bank Industrial and Commercial Bank China Agricultural Bank and China Communications Bank PostalSavings Bank of China 5 Agriculture Development Bank of China China Development Bank The ExportImport Bank of China 6 There are 12 national jointstock commercial banks in China China Merchants Bank Pudong Development Bank CITIC Bank China Everbright Bank Huaxia Bank China Minsheng Bank Guangfa Bank Industrial Bank Ping An Bank Zhejiang Commercial Bank Hengfeng Bank и Bohai Bank 7 In 2015 the collapse of the Shanghai and Shenzhen stock exchanges evaporated about 5 trillion USD in market value Investors suffered heavy losses each losing more than 15000 USD on average Shi et al 2022 In 2015 the stock market problems pushed the Peoples Bank of China to raise its deposit and lending rates Rathnayake et al 2022 8 β Vector dimension number of resources used 1 9 Berger and Mester 1997 showed that it is necessary to take into account the methods of attracting and allocating resources the levels of interest rates and the risks taken In addition for banks of different sizes and specializations it is necessary to take into account the available opportunities for creating banking products A number of researchers believe that it is important for banks to take into account not only internal but also external factors such as market influences and banking regulation Bhattacharyya et al 1997 This influence extends to all banks and will lead to a parallel change in the border production function 10 According to Table 1 Profit Total profits Profit before tax 11 According to Table 1 Cost Total costs overheads interest expense other operating expense 12 The significance of individual factors characterizes the quality of the created regression model which can explain the estimated parameter This does not allow for us to draw unambiguous conclusions about the real influence of individual factors on the estimated parameter but characterizes only the statistical relationship Conclusions about the real influence of individual factors require additional reasoning and analysis 13 The criteria for determining significance are given in the last row of Table 5 14 The arithmetic mean References AbuAlkheil Ahmad Ghadeer Khartabiel and Nuradli Ridzwan Shah Mohd Dali 2018 A TwoStage Parametric Stochastic Frontier Analysis SFA of the Efficiency Performance of Shariah Compliant Banks A Global Evidence American Journal of Finance and Accounting 5 85110 CrossRef Aiello Francesco and Graziella Bonanno 2013 Profit and Cost Efficiency in the Italian Banking Industry 20062011 Economics and Business Letters 2 190205 CrossRef Aielloa Francesco and Graziella Bonanno 2016 Bank Efficiency and Local Market Conditions Evidence from Italy Journal of Economics and Business 83 7090 CrossRef Aigner Dennis C A Knox Lovell and Peter Schmidt 1997 Formulation and Estimation of Stochastic Frontier Production Function Models Journal of Econometrics 6 2137 CrossRef Almanidis Pavlos Mustafa Karakaplan and Levent Kutlu 2019 A Dynamic Stochastic Frontier Model with Threshold Effects US Bank Size and Efficiency Journal of Productivity Analysis 52 116 CrossRef Ariff Mohamed and Can Luc 2008 Cost and Profit Efficiency of Chinese Banks A NonParametric Analysis China Economic Review 19 26073 CrossRef Athanasoglou Panayiotis P Sophocles N Brissimis and Matthaios D Delis 2008 BankSpecific IndustrySpecific and Macroeconomic Determinants of Bank Profitability Journal of International Financial Markets Institutions and Money 18 12136 CrossRef Avkiran Necmi K 2011 Association of DEA SuperEfficiency Estimates with Financial Ratios Investigating the Case for Chinese Banks Omega 39 32334 CrossRef Battese George Edward and Tim J Coelli 1992 Frontier Production Functions Technical Efficiency and Panel Data With Application to Paddy Farmers in India Journal of Productivity Analysis 3 15369 Available online httpswwwjstororgstable41770578 accessed on 12 April 2023 CrossRef Battese George Edward and Tim J Coelli 1995 A Model for Technical Inefficiency Effects in a Stochastic Frontier Production Function for Panel Data Empirical Economics 20 32532 CrossRef J Risk Financial Manag 2023 16 243 19 of 20 Berger Allen N and David B Humphrey 1997 Efficiency of Financial Institutions International Survey and Directions for Future Research European Journal of Operational Research 98 175212 CrossRef Berger Allen N and Loretta J Mester 1997 Inside the Black Box What Explains Differences in the Efficiencies of Financial Institutions Board of Governors of the Federal Reserve System Working Paper 21 1059 CrossRef Bhattacharyya Arunava C A Knox Lovell and Pankaj Sahay 1997 The Impact of Liberalization on the Productive Efficiency of Indian Commercial Banks European Journal of Operational Research 98 33245 CrossRef Bitar Mohammad Kuntara Pukthuanthong and Thomas Walker 2018 The Effect of Capital Ratios on the Risk Efficiency and Profitability of Banks Evidence from OECD Countries Journal of International Financial Markets Institutions and Money 53 22762 CrossRef Capraru Bogdan and Iulian Ihnatov 2014 Banks Profitability in Selected Central and Eastern European Countries Procedia Economics and Finance 16 58791 CrossRef Charnes Abraham William W Cooper and Edwardo Rhodes 1978 Measuring the Efficiency of Decision Making Units European Journal of Operational Research 2 42944 CrossRef Christensen Laurits Dale Jorgenson and Lawrence Lau 1973 Transcendental Logarithmic Production Frontiers The Review of Economics and Statistics 55 2845 CrossRef Coelli Timothy J D S Prasada Rao Christopher J ODonnell and George E Battese 2005 An Introduction to Efficiency and Productivity Analysis 2nd ed New York Springer CrossRef Davis Philip Dilruba Karim and Dennison Noel 2022 The Effects of Macroprudential Policy on Banks Profitability International Review of Financial Analysis 80 127 CrossRef de Guindos Luis 2019 Challenges for Bank Profitability OMFIF City Lecture Available online httpswwwecbeuropaeupress keydate2019htmlecbsp1905017733ecc1a9enhtml accessed on 28 December 2022 Ding Dong and Robin Sickles 2018 Frontier Efficiency Capital Structure and Portfolio Risk An Empirical Analysis of US Banks Business Research Quarterly 21 715 CrossRef Djalilova Khurshid and Jenifer Piesse 2016 Determinants of Bank Profitability in Transition Countries What Matters Most Research in International Business and Finance 38 6982 CrossRef Dong Yizhe Michael Firth Wenxuan Hou and Weiwei Yang 2016 Evaluating the Performance of Chinese Commercial Banks A Comparative Analysis of Different Types of Banks European Journal of Operational Research 252 28095 CrossRef Foroutan Farnaz Homayoun Ranjbar and Mehdi HampaeyanMiandoab 2015 Estimation of Technical Efficiency in Banking Industry Using Stochastic Frontier Analysis International Journal of Productivity and Quality Management 16 5469 CrossRef Fu Xiaoqing and Shelagh Heffernan 2007 Cost XEfficiency in Chinas Banking Sector China Economic Review 18 3553 CrossRef Gaganis Chrysovalantis Emilios Galariotis Fotios Pasiouras and Christos Staikouras 2020 Bank Profit Efficiency and Financial Consumer Protection Policies Journal of Business Research 118 98116 CrossRef Henriques Iago Cotrim Vinicius Amorim Sobreiro Herbert Kimura and Enzo Barberio Mariano 2020 TwoStage DEA in Banks Terminological Controversies and Future Directions Expert Systems with Applications 161 131 CrossRef Horrocks Michael Venkataramanan Ramaswamy and Keri Rupp 2008 Financial Reforms in Chinese Banking The Impact on Personal Lending and Operational Efficiency Business Horizons 51 51117 CrossRef Huang Qiubin 2020 Ownership Concentration and Bank Profitability in China Economics Letters 196 13 CrossRef Huang Wen Ying Wu and Li Deng 2021 Does Banking Competition Stimulate Regional Innovation Evidence from China Pacific Basin Finance Journal 70 115 CrossRef Klein PaulOlivier and Laurent Weill 2022 Bank Profitability and Economic Growth Quarterly Review of Economics and Finance 84 18399 CrossRef Kumbhakar Subal C and C A Knox Lovell 2000 Stochastic Frontier Analysis Cambridge Cambridge University Press pp 279307 CrossRef Le Minh VietNgu Hoang Clevo Wilson and Shunsuke Managi 2020 Net Stable Funding Ratio and Profit Efficiency of Commercial Banks in the US Economic Analysis and Policy 67 5566 CrossRef Le Tu and Thanh Ngo 2020 The Determinants of Bank Profitability A CrossCountry Analysis Central Bank Review 20 6573 CrossRef Lee Chien Chiang and Meng Fen Hsieh 2013 The Impact of Bank Capital on Profitability and Risk in Asian Banking Journal of International Money and Finance 32 25181 CrossRef Lee Chien Chiang Chih Wei Wang and Shan Ju Ho 2020 Financial Innovation and Bank Growth The Role of Institutional Environments North American Journal of Economics and Finance 53 123 CrossRef Li Fangmin Tianle Yang Min Du and Miao Huang 2023 The Development Fit Index of Digital Currency Electronic Payment between China and the One Belt One Road Countries Research in International Business and Finance 64 114 CrossRef Li Jingya and MingHua Liu 2019 Interest Rate Liberalization and Passthrough of Monetary Policy Rate to Bank Lending Rates in China Frontiers of Business Research in China 13 219 CrossRef Li XiaoLin DengKui Si and Xinyu Ge 2021 Chinas Interest Rate Passthrough after the Interest Rate Liberalization Evidence from a Nonlinear Autoregressive Distributed Lag Model International Review of Economics and Finance 73 25774 CrossRef Li Yang 2020 Analyzing Efficiencies of City Commercial Banks in China An Application of the Bootstrapped DEA Approach Pacific Basin Finance Journal 62 112 CrossRef J Risk Financial Manag 2023 16 243 20 of 20 Lin Xiaochi and Yi Zhang 2009 Bank Ownership Reform and Bank Performance in China Journal of Banking and Finance 33 2029 CrossRef Liu Kerry 2017 Chinas Interest Rate Passthrough to Commercial Banks Before and After Interest Rate Liberalisation Economic Affairs 37 27987 CrossRef Mai Vo Phuong Kent Matthews David Meenagh Patrick Minford and Zhiguo Xiao 2022 Regulatory Arbitrage Shadow Banking and Monetary Policy in China Journal of International Financial Markets Institutions and Money 80 113 CrossRef Mott Carey 2022 China Reserve Requirements 20152016 Journal of Financial Crisis 4 393428 Available online httpselischolar libraryyaleedujournaloffinancialcrisesvol4iss419 accessed on 12 April 2023 Ngan Le Thi Thanh 2014 Profit and Cost Efficiency Analysis in Banking Sector A Case of Stochastic Frontier Approach for Vietnam Journal of Knowledge Management Economics and Information Technology IV 117 Available online wwwscientificpapersorg accessed on 28 December 2022 Nicolae Petria Bogdan Capraru and Iulian Ihnatov 2015 Determinants of Banks Profitability Evidence from EU 27 Banking Systems Procedia Economics and Finance 20 51824 CrossRef Pasiouras Fotios Sailesh Tanna and Constantin Zopounidis 2009 The Impact of Banking Regulations on Banks Cost and Profit Efficiency CrossCountry Evidence International Review of Financial Analysis 18 294302 CrossRef Rathnayake Dilesha Yang Bai Pierre Louembé and Li Qi 2022 Interest Rate Liberalization and Commercial Bank Performance New Evidence From Chinese AShare Banks SAGE Open 12 112 CrossRef Saona Paolo 2016 Intra and ExtraBank Determinants of Latin American Banks Profitability International Review of Economics and Finance 45 197214 CrossRef Scotta Susan V John Van Reenenb and Markos Zachariadis 2017 The LongTerm Effect of Digital Innovation on Bank Performance An Empirical Study of SWIFT Adoption in Financial Services Research Policy 46 9841004 CrossRef Sealey Calvin W and James T Lindley 1977 Inputs Outputs and a Theory of Production and Cost at Depository Financial Institutions Journal of Finance 32 125166 CrossRef Shi Yang Shu Chen and Ruiming Liu 2022 Fund Renaming and Fund Flows Evidence from Chinas Stock Market Crash in 2015 Economic Modelling 108 113 CrossRef Silva Thiago Christiano Benjamin Miranda Tabak Daniel Oliveira Cajueiro and Marina Villas Boas Diasd 2017 A Comparison of DEA and SFA Using Micro and MacroLevel Perspectives Efficiency of Chinese Local Banks Physica A Statistical Mechanics and Its Applications 469 21623 CrossRef Sun Rongrong 2020 Monetary Policy Announcements and Market Interest Rates Response Evidence from China Journal of Banking and Finance 113 105766 CrossRef Tan Yong 2017 The Impacts of Competition and Shadow Banking on Profitability Evidence from the Chinese Banking Industry North American Journal of Economics and Finance 42 89106 CrossRef VeraGilces Paul Segundo CaminoMogro Xavier OrdẽnanaRodríguez and Gino CornejoMarcos 2020 A Look inside Banking Profitability Evidence from a Dollarized Emerging Country Quarterly Review of Economics and Finance 75 14766 CrossRef Wang Ke Wei Huang Jie Wu and Yingnan Liu 2014 Efficiency Measures of the Chinese Commercial Banking System Using an Additive TwoStage DEA Omega 44 520 CrossRef Wanke Peter Amir Yazdi and Ramona Birau 2022 Revisiting CAMELS Rating System and the Performance of ASEAN Banks A Comprehensive MCDMZNumbers Approach IEEE Access 10 54098109 CrossRef Yang Shanran Benye Shi and Fujia Yang 2023 Research in International Business and Finance Macroeconomic Impact of the SinoU S Trade Frictions Based on a TwoCountry TwoSector DSGE Model Research in International Business and Finance 65 117 CrossRef Zhao Jinsong Xinghao Li ChinHsien Shi Chen and ChiChuan Lee 2022 Riding the FinTech Innovation Wave FinTech Patents and Bank Performance Journal of International Money and Finance 122 102552 CrossRef DisclaimerPublishers Note The statements opinions and data contained in all publications are solely those of the individual authors and contributors and not of MDPI andor the editors MDPI andor the editors disclaim responsibility for any injury to people or property resulting from any ideas methods instructions or products referred to in the content Este artigo está licenciado com uma Licença Creative Commons Atribuição Não Comercial 40 Internacional EFICIÊNCIA BANCÁRIA NO BRASIL 20102019 UMA ANÁLISE DE DOIS ESTÁGIOS Fábio Lucas Takahashi Graduado em Ciências Econômicas pela Universidade Estadual de Maringá UEM e mestrando em Economia da mesma instituição Email flucastakahashioutlookcom httpsorcidorg0000000171323876 Marcos Roberto Vasconcelos Graduado em Ciências Econômicas pela Universidade Federal do Paraná UFPR e doutor em Ciência Econômica pela Universidade Estadual de Campinas Unicamp Professor do Departamento de Economia DCO da Universidade Estadual de Maringá UEM Email mrvasconcelosuembr httpsorcidorg0000000314764899 Como citar este artigo Takahashi F L Vasconcelos M R 2022 Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Revista de Economia Mackenzie 192 165187 doi10593518082785remv19n2p165187 Recebido em 14042022 Aprovado em 16082022 166 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 Resumo O objetivo deste trabalho é analisar o desempenho do setor bancário brasileiro e seus determinantes A presente pesquisa foi dividida em duas partes Na primei ra buscase determinar a eficiência técnica pelo modelo DEABCC nos principais bancos brasileiros no período de 2010 a 2019 Na segunda parte são avaliados os determinantes que podem influenciar o desempenho dos bancos As estimati vas indicam que os grandes bancos são em média mais eficientes do que os bancos de pequeno e médio portes Além disso quando se observa o tipo de propriedade os bancos estrangeiros tendem a apresentar melhor desempenho em comparação com os bancos nacionais Ademais os determinantes apontam que a margem de lucro líquido e a renda sem juros são significantes e contribuem posi tivamente para o nível de eficiência Por sua vez o índice de adequação de capi tal os empréstimos inadimplentes e a inflação têm impacto negativo sobre a efi ciência técnica Palavraschave Análise Envoltória de Dados determinantes eficiência técnica instituições financeiras setor bancário Classificação JEL B23 B26 C24 G20 INTRODUÇÃO O aumento da incidência de crises bancárias e financeiras nas últimas dé cadas desencadeou uma agenda de pesquisas sobre o comportamento do setor bancário não apenas sobre as causas subjacentes das crises mas também so bre o impacto na economia real Kose et al 2010 Ahmad et al 2019 Nesse sentido a literatura de eficiência bancária tem recebido grande atenção prin cipalmente pelo impacto das reformas regulatórias da competição bancária e das novas tecnologias de informação que foram inseridas no sistema bancário Diallo 2018 A importância do setor bancário baseiase no fato de que os bancos são os principais canais de poupança e alocação de crédito em muitas economias entre elas a brasileira Tecles Tabak 2010 Isto é o setor bancário desem penha importante função ao converter depósitos em empréstimos e investi mentos Claessens Laeven 2005 Em razão do crescente volume e da com plexidade das transações e dos ativos negociados tornouse cada vez mais necessário avaliar os riscos da atividade bancária Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 167 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 É razoável esperar que um setor bancário eficiente proporcione maior esta bilidade ao sistema financeiro e conduza para o melhor desenvolvimento eco nômico do país Levine 1997 Dessa forma o conhecimento dos fatores que influenciam o desempenho do setor bancário é essencial para que os gestores das instituições financeiras reguladores e supervisores possam formular polí ticas mais apropriadas aos problemas de ineficiência e fragilidade bancária Abreu et al 2019 Parte considerável da eficiência das instituições financeiras depende dire tamente do gerenciamento eficiente da combinação de insumos inputs e pro dutos outputs nas operações bancárias Entretanto os fatores que determi nam a eficiência das instituições financeiras podem estar associados tanto a fatores internos por exemplo gestão ineficiente e fraudes como a fatores externos por exemplo mudanças regulatórias e variáveis macroeconômicas Berger DeYoung 1997 Diante do exposto o presente trabalho examina a eficiência técnica das instituições financeiras do setor bancário brasileiro no período de 2010 a 2019 e analisa seus determinantes Em particular concentrase em explorar as diferenças nas eficiências médias entre os tipos de propriedade público na cional privado nacional e estrangeiro e os diferentes tamanhos de banco pe queno médio e grande portes Para tanto utilizouse a metodologia de Análise Envoltória de Dados Data Envelopment Analysis DEA para estimar o nível de eficiência técnica das instituições financeiras Adicionalmente para avaliar os determinantes da efi ciência técnica foi empregada a regressão Tobit que permite mensurar variá veis com dados censurados ou limitados Nesse aspecto esperase que os re sultados permitam agregar novas evidências sobre o desempenho do setor bancário brasileiro Este trabalho tem implicações importantes Em primeiro lugar adicionamse evidências sobre a eficiência bancária e seus determinantes Em segundo lu gar ampliase a compreensão de como os indicadores de performance bancária e as variáveis macroeconômicas moldam o nível de eficiência das instituições financeiras Vários estudos investigaram a evolução do setor bancário brasilei ro e seus possíveis determinantes Staub et al 2010 Tecles Tabak 2010 Wanke Barros 2014 Embora a literatura de eficiência bancária no Brasil aborde as diferenças de eficiência entre o tipo de propriedade e o tamanho dos bancos notase que há poucas evidências do impacto dos empréstimos inadimplentes no desempe nho bancário Os resultados deste trabalho sugerem que os bancos brasileiros 168 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 melhoram seus desempenhos ao aumentarem as margens de lucros das recei tas provindas de juros e sem juros Entretanto podese observar que a menor eficiência do setor bancário brasileiro está associada à presença de maior capi talização inflação e empréstimos inadimplentes O artigo prossegue da seguinte forma a seção 1 apresenta o referencial teó rico a seção 2 contém a revisão de literatura de eficiência bancária a seção 3 exibe os dados utilizados neste trabalho a seção 4 expõe os métodos de pes quisa a seção 5 relata os resultados e por fim há as considerações finais 1 REFERENCIAL TEÓRICO Nos últimos anos as preocupações com a mensuração de desempenho das instituições financeiras têm adquirido maior atenção dos pesquisadores Ahmad et al 2019 Os motivos são diversos como preocupações com os riscos de falência bancária problemas de estabilidade financeira fatores que influen ciam o desenvolvimento financeiro etc Abreu et al 2019 Ahmad et al 2019 Nesses estudos temse utilizado com frequência a DEA para avaliar a eficiên cia das instituições financeiras A metodologia não paramétrica da DEA utiliza a combinação de insumos inputs e produtos outputs sem precisar necessariamente de suposições so bre sua função de produção Os modelos DEA podem avaliar a eficiência rela tiva de um conjunto de entidades denominadas Unidades Tomadoras de De cisão DecisionMaking Units DMU sendo uma DMU qualquer entidade que deve ser avaliada em termos de sua capacidade de converter múltiplos insumos em múltiplos produtos Cooper et al 2011 Inicialmente o modelo DEA introduzido por Charnes Cooper e Rhodes 1978 baseado principalmente no trabalho de Farrell 1957 proporcio nou a formação do modelo CCR ou Constant Return to Scale CRS Nesse as pecto o modelo CCR CRS assume que as DMU devem operar com retornos constantes de escala isto é se um aumento nos insumos inputs resultar em um aumento proporcional nos níveis do produto outputs Esse modelo men sura a eficiência relativa das DMU e produz resultados consistentes em termos de avaliação global das unidades produtivas Charner et al 1978 Entretanto os modelos DEA do tipo CCR CRS requerem a suposição de total proporcionalidade entre insumos e produtos o que pode não ser apro Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 169 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 priado em alguns contextos Para superar essa dificuldade Banker Charnes e Cooper 1984 desenvolveram o modelo BCC ou Variable Return to Scale VRS que mede as DMU com retornos variáveis de escala que ajuda a estimar se um aumento ou uma diminuição nos insumos e produtos não resulta em mudança proporcional dos insumos e produtos respectivamente Cooper et al 2011 Nesse sentido o modelo BCC pode ser útil para avaliar DMU com diferen tes tamanhos de economias de escala cujas operações são proporcionalmente divergentes entre as DMU Ou seja a escala das operações pode ter impacto nos níveis de eficiência criando economias ou deseconomias de escala Por isso o modelo proposto por Banker et al 1984 com retornos de variáveis de escala poderia apresentar resultados mais seguros A partir do desenvolvimento dos modelos CCR CRS e BCC VRS hou ve um crescimento extraordinário na metodologia DEA Como resultado surgiram trabalhos em quantidade elevada sobre as aplicações da DEA para avaliar a eficiência de unidades produtivas conforme Emrouznejad et al 2008 e Emrouznejad e Yang 2018 Um dos fatores para o elevado volume de trabalhos na literatura sobre DEA é a capacidade de os modelos inovarem tanto na teoria como nos resultados empíricos o que reforça sua aplicabili dade Emrouznejad Yang 2018 Na literatura de eficiência bancária por exemplo observamse diversos trabalhos sobre a eficiência relativa que podem envolver também temas sobre rentabilidade produtividade e custos entre as instituições financeiras Ahmad et al 2019 Abreu et al 2019 Esses estudos permitiram investigar o com portamento das instituições no setor bancário aprimorar a metodologia e pro porcionar evidências sobre as ineficiências dessas instituições Nesse sentido parte da literatura de eficiência bancária utiliza estudos ba seados em duas etapas pois os resultados fornecidos pela DEA podem não ser suficientes para fornecer diagnósticos mais precisos sobre o comportamento do setor bancário Entre esses modelos podese utilizar a regressão Tobit para estimar os efeitos de variáveis explicativas a fim de fornecer evidências sobre os possíveis fatores que influenciam a eficiência das instituições financeiras 2 LITERATURA DE EFICIÊNCIA BANCÁRIA Na literatura de eficiência bancária há estudos que aplicam para os Estados Unidos a Europa e outros países técnicas econométricas e não paramétricas 170 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 com o propósito de examinar a eficiência e a evolução de produtividade dos bancos Claessens et al 2001 Miccoel et al 2007 Nguyen 2018 Embora existam vastos estudos sobre eficiência bancária poucos se referem aos mode los de dois estágios DEA e análise dos determinantes no Brasil Staub et al 2010 Tecles Tabak 2010 Wanke Barros 2014 Como o foco do pre sente trabalho é o sistema bancário brasileiro são discutidos a seguir os estu dos referentes a economias emergentes Por exemplo GarzaGarcía 2012 investiga os determinantes da eficiência bancária no México durante o período de 2001 a 2009 O resultado da DEA indica que o setor bancário mexicano experimentou baixos níveis de eficiên cia técnica 15 eficiência técnica pura 29 e eficiência de escala 14 Além disso os principais determinantes do aumento da eficiência bancária estão associados aos aumentos dos empréstimos ao crescimento econômico variação do produto interno bruto PIB e à participação das instituições estrangeiras Os resultados de Defung et al 2016 fornecem uma análise da eficiência técnica do setor bancário da Indonésia no período de 1993 a 2011 A partir do modelo DEABCC foi possível verificar que as instituições financeiras dos tipos público nacional e estrangeiro por exemplo são consideradas mais efi cientes do que qualquer outro tipo de instituição Na análise dos determinan tes realizada pela regressão Tobit pôdese identificar que o impacto regulatório é geralmente positivo e estatisticamente significativo Para Kutlar et al 2017 que avaliaram os bancos comerciais da Turquia para o período de 2003 a 2012 por meio da metodologia DEA CCR e BCC os resultados observados foram mistos Na análise de eficiência tanto do modelo CCR quanto do BCC houve ao longo do período examinado uma redução do número de bancos tecnicamente eficientes no entanto a eficiência alocati va dos bancos turcos aumentou nesse período Complementarmente no mo delo de regressão Tobit os autores determinaram qual tipo de eficiência téc nica ou alocativa representava melhor o comportamento do produto output De acordo com os resultados as pontuações de eficiência alocativa represen taram melhor a reação do produto output Jayaraman e Srinivasan 2019 examinaram a eficiência de custo receita e lucro dos bancos da Índia de 2004 a 2013 também por meio da abordagem de dois estágios DEA e regressão Tobit Os resultados sugerem que a eficiên cia de custos e lucros dos bancos indianos está correlacionada positivamente e revelam que se os bancos são eficientes em termos de custos também o são em termos de lucro Da mesma forma os autores evidenciaram que a eficiência Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 171 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 de lucro é o tipo de eficiência que diferencia melhor o desempenho entre as instituições analisadas Por fim concluíram que os principais determinantes da eficiência bancária foram o tamanho total de ativos e a variável da razão de despesas operacionais em relação ao total de ativos Por sua vez Staub et al 2010 analisaram as eficiências técnica alocativa e de custos para as instituições financeiras do Brasil de 2000 a 2007 com a metodologia DEA Inicialmente os resultados indicaram que em média os bancos no Brasil apresentam níveis baixos de eficiência de custo em compara ção com bancos da Europa e dos Estados Unidos Para o período de alta vola tilidade macroeconômica de 2000 a 2002 a ineficiência dos bancos brasilei ros pode ser atribuída à eficiência técnica e não à eficiência alocativa Além disso os bancos públicos brasileiros mostraramse mais eficientes em termos de custo do que os privados nacionais e estrangeiros Tecles e Tabak 2010 por exemplo ao avaliarem a eficiência do setor ban cário brasileiro no período de 2000 a 2007 optaram por utilizar a abordagem de fronteira estocástica bayesiana para estimar a eficiência de custo e lucro Os resultados sugerem que os grandes bancos são mais eficientes em custo e lucro o que pode estar associado com a crescente concentração bancária observada no período de análise Por fim quando se analisaram os determinantes veri ficouse que a capitalização tem efeito positivo na eficiência de custo e lucro O estudo de Wanke e Barros 2014 para o setor bancário brasileiro con siderando dados relativos ao ano de 2012 estimou medidas de eficiência em um processo de dois estágios 1 a eficiência de custos e a produtiva e 2 os determinantes de cada uma dessas medidas de eficiência Os resultados su gerem que os bancos brasileiros são heterogêneos ou seja alguns são consi derados mais eficientes em termos de custo e outros em eficiência produtiva O tamanho do ativo bancário mostrouse o principal determinante da eficiên cia de custo enquanto a eficiência produtiva parece ser impactada positiva mente por eventos de fusão e aquisição bem como pela presença de um controlador privado Nesse sentido os modelos propostos em DEA podem apresentar diferentes resultados conforme o tamanho da amostra das variáveis escolhidas nos insu mos inputs e produtos outputs da orientação etc Da mesma forma os de terminantes da eficiência podem apresentar resultados mistos ou seja não há consensos definitivos Abreu et al 2019 Portanto o presente estudo tem como objetivo agregar novas evidências à literatura de eficiência bancária no Brasil e a seus possíveis determinantes de ineficiência 172 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 3 ANÁLISE DE DADOS A amostra foi composta pelas 39 maiores instituições financeiras que ope raram no Brasil no período de 2010 a 2019 Especificamente as instituições financeiras foram selecionadas de acordo com o tamanho total de ativos e o tipo de consolidado bancário tipo B1 que possuem carteira comercial As sim coletaramse os dados contábilfinanceiros das instituições financeiras no banco de dados IFdata disponibilizado pelo Banco Central do Brasil BCB Dessa forma a Tabela 1 mostra as variáveis utilizadas pelo presente trabalho Tabela 1 Descrição das variáveis do modelo de eficiência Variável Descrição Insumos x x1 Capital x1 patrimônio líquido x2 Depósito x2 total de depósitos Produtos y y1 Empréstimos e recebíveis y1 Operações de crédito líquidas de provisão y2 Total de títulos y2 aplicações interfinanceiras de liquidez Títulos e Valores Mobiliários TVM e instrumentos financeiros derivativos y3 Receita não financeira y3 rendas de prestação de serviços rendas tarifárias bancárias resultado de participações outras receitas operacionais Fonte Elaborada pelos autores com base nos dados obtidos no site do Banco Central do Brasil BCB 4 MÉTODOS DE PESQUISA A aplicação da metodologia neste trabalho está dividida em dois estágios No primeiro estágio a DEA é utilizada para medir o nível de eficiência opera cional das DMU representada pelas instituições financeiras No segundo estágio o modelo Tobit é usado para calcular os determinantes da eficiência técnica Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 173 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 41 Análise envoltória de dados Como já exposto melhorar a gestão das instituições financeiras tornouse um objetivo cada vez mais importante para diminuir os riscos no sistema fi nanceiro Uma das ferramentas mais utilizadas para mensuração de eficiência no setor bancário é a DEA A principal vantagem desse método é sua capaci dade de acomodar múltiplas entradas inputs e saídas outputs Também é útil porque permite avaliar a eficiência relativa comparando cada DMU com outras DMU com melhores pontuações de eficiência Nesse sentido a solução pode produzir unidades tecnicamente eficientes ou unidades com melhores práticas benchmark que são identificadas pelas DMU que se encontram na fronteira de melhores práticas As unidades tecni camente eficientes possuem pontuação igual a 1 ou 100 Por sua vez uni dades tecnicamente ineficientes ou inferiores à fronteira de melhores práticas são inferiores a 1 menor que 100 Cabe ressaltar que um dos objetivos do trabalho é entender a capacidade das instituições financeiras em transformar os insumos em produtos isto é como converter capital e depósito em títulos receitas não financeiras e em préstimos As pontuações de eficiência nos modelos DEA são definidos es sencialmente pela orientação à entrada insumos ou ao produto output A orientação para entrada representa a redução equiproporcional máxima permitida de suas entradas que ainda é capaz de produzir o mesmo nível de produtos outputs Por sua vez a orientação para produto output reflete a expansão equiproporcional máxima de seus produtos outputs que pode ser feita pelo mesmo nível de insumos inputs O sentido da orientação é especi ficar a forma de otimização do modelo de retorno de escala CRS ou VRS Cooper et al 2011 Para avaliar a eficiência bancária o presente trabalho utiliza o modelo DEA BCC ou VRS orientado para o produto output Nesse sentido a motivação para a escolha do modelo DEABCC decorre do fato de que há a necessidade de considerar os retornos variáveis de escala entre as instituições financeiras Além disso o modelo de eficiência do tipo CCR ou CRS que permite avaliar retornos constantes de escala só deve ser usado em contextos em que todas as instituições financeiras operem em condições semelhantes Desse modo dado um conjunto de DMU i n DMU i 12 i no qual cada unidade consome x insumos inputs 1i 2i ei x x x para produzir y produtos outputs 1i 2i fi y y y Assim de acordo com Cooper et al 2011 a eficiência do modelo BCC orientado para produto output da unidade θ k é dada pela seguinte programação linear 174 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 θ k max λ n j ij io j 1 x x i λ θ n j rj ro j 1 y y r λ j 0 j λ n j 1 j 1 1 O modelo BCC VRS introduzido por Banker et al 1984 mede a eficiên cia técnica considerando a restrição de convexidade que garante que a DMU tenha o tamanho de escala proporcional à unidade que está sendo medida Ao contrário do modelo CCR CRS o modelo BCC VRS permite retornos variá veis de escala Como as instituições no Brasil variam drasticamente de tama nho e mix de produtos e serviços financeiros uma estimativa com retornos constantes de escala seria inadequada para nossa análise e por isso o presen te trabalho optou pelo modelo BCC VRS 42 Análise de regressão Tobit Tendo em vista os resultados de eficiência técnica é possível avaliar a par tir do modelo de regressão Tobit as variáveis que influenciaram a variável dependente Isto é as pontuações de eficiência obtidas no modelo DEABCC são usadas como variável dependente E como a pontuação de eficiência é li mitada ao intervalo entre 0 e 1 tornase o uso da técnica de regressão dos mínimos quadrados um modelo inadequado pois pode produzir resultados tendenciosos e inconsistentes Por conta disso a regressão Tobit permite o uso de variável dependente de alcance limitado ou seja variável dependente censurada O modelo Tobit tem sido amplamente utilizado nos estudos para verificar os fatores que afetam o nível de eficiência das unidades de produção Defung et al 2016 A diferen ça de eficiência entre as instituições financeiras não é afetada apenas pelos insumos e produtos mas também por outros fatores condicionais para a ges tão delas Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 175 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 Assim o presente trabalho apresenta um modelo de regressão Tobit para entender melhor a influência dos indicadores de qualidade bancária na eficiên cia técnica dessas instituições financeiras GarzaGarcía 2012 O modelo Tobit do presente trabalho é descrito da seguinte forma α β β β β β β ε 1 2 3 4 5 6 i it it it it it it it ET CAR NIM NII ROA INF PIB 1 39 1 i t n 2 A variável dependente i ET na Equação 2 é a pontuação de eficiência téc nica obtida no modelo DEABCC orientado ao produto output Assim i re presenta a observação instituição financeira t referese ao período α é o termo constante β reflete o coeficiente de cada variável e ε é o termo de erro da regressão As variáveis explicativas estão descritas na Tabela 2 Tabela 2 Descrição das variáveis explicativas da regressão Tobit Variável Descrição CAR Índice de adequação de capital CAR patrimônio líquido total de ativos NIM Margem de lucro líquida NIM receita de intermediação financeira total de depósitos NII Renda sem juros NII rendas de prestação de serviços rendas de tarifas bancárias resultados de participação outras receitas operacionais total de ativos ROA Retorno sobre ativos ROA lucro líquido total de ativos NPLR Razão de índice de inadimplência NPLR provisão sobre operações de crédito operações de crédito líquidas de provisão INF Taxa de inflação anual INF Índice de Preço ao Consumidor Amplo IPCA PIB Taxa de crescimento econômico anual PIB taxa de crescimento econômico real Fonte Elaborada pelos autores com base nos dados obtidos nos sites do BCB Instituto Brasileiro de Geografia e Estatística Banco Mundial e Instituto de Pes quisa Econômica Aplicada 176 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 5 RESULTADOS EMPÍRICOS 51 Eficiência técnica do setor bancário brasileiro Os resultados de eficiência técnica do modelo DEA das 39 instituições fi nanceiras analisadas entre 2010 e 2019 foi em média de 07964 7964 O ano de melhor performance foi em 2018 com 08846 8846 e a pior performance ocorreu em 2017 com 06153 6153 As pontuações de eficiên cia técnica média de cada instituição financeira estão apresentadas na Tabela 3 Tabela 3 Eficiência técnica média das instituições financeiras de 2010 a 2019 Instituição financeira Tipo de propriedade Tamanho Eficiência técnica média Desvio padrão BB Público nacional Grande 10000 00000 CEF Público nacional Grande 10000 00000 Alfa Privado nacional Pequeno 10000 00000 Sicredi Privado nacional Pequeno 10000 00000 Itaú Unibanco Privado nacional Grande 10000 00000 Bradesco Privado nacional Grande 09931 00169 Votorantim Privado nacional Médio 09756 00770 Bancoob Privado nacional Pequeno 09650 00564 Safra Privado nacional Médio 09537 00850 CSF Estrangeiro Pequeno 09302 02207 Rabobank Estrangeiro Pequeno 09277 02125 John Deere Estrangeiro Pequeno 09241 01946 Volkswagen Estrangeiro Pequeno 08943 02229 Santander Estrangeiro Grande 08812 01564 BNP Paribas Estrangeiro Pequeno 08714 01078 Nordeste Público nacional Médio 08529 01119 J P Morgan Estrangeiro Pequeno 08481 01617 HSBC Estrangeiro Grande 08342 01207 Pan Privado nacional Pequeno 08203 02019 continua Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 177 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 Instituição financeira Tipo de propriedade Tamanho Eficiência técnica média Desvio padrão MercedesBenz Estrangeiro Pequeno 08185 02630 BTG Pactual Privado nacional Grande 08129 02009 Banestes Público nacional Pequeno 08103 01540 CCB Estrangeiro Pequeno 07774 02116 BMG Privado nacional Pequeno 07580 02452 Société Générale Estrangeiro Pequeno 07326 01709 BofA Securities Estrangeiro Pequeno 07320 01813 Citibank Estrangeiro Médio 07181 01366 ING Estrangeiro Pequeno 06988 03069 BRB Público nacional Pequeno 06790 01369 Daycoval Privado nacional Pequeno 06684 01755 Banpará Público nacional Pequeno 06666 01749 GM Estrangeiro Pequeno 06474 01844 Banrisul Público nacional Médio 06133 01090 ABC Estrangeiro Pequeno 05913 00837 Pine Privado nacional Pequeno 05581 02035 Amazônia Público nacional Pequeno 05465 01480 Credit Suisse Estrangeiro Pequeno 05265 02191 Crédit Agricole Estrangeiro Pequeno 04966 01656 Original Privado nacional Pequeno 03328 01524 Fonte Elaborada pelos autores com base nos dados obtidos no site do BCB Os resultados apontam que as únicas instituições financeiras tecnicamente eficientes ET 1 em todo o período analisado de 2010 a 2019 foram Banco do Brasil BB Caixa Econômica Federal CEF Alfa Sicredi e Itaú Unibanco Cabe ressaltar que a eficiência técnica ET é relacionada à qualidade da ges tão ou seja as instituições financeiras tecnicamente eficientes ET 1 são consideradas as unidades com as melhores práticas ou o benchmark do setor analisado Entretanto avaliar cada instituição apenas revela quais unidades de produ ção ou DMU utilizam de forma eficiente seus recursos inputs na elaboração Tabela 3 Eficiência técnica média das instituições financeiras de 2010 a 2019 conclusão 178 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 de seus produtos outputs Ou seja apenas se avalia a qualidade de gestão de uma determinada instituição Assim para obtermos avaliações mais agrega das podemos categorizar as instituições financeiras o que permite examinar eficiência por agrupamentos bancários Nesse sentido na literatura de eficiência bancária há o interesse de verificar se o tamanho total de ativos influencia o desempenho das instituições finan ceiras A Tabela 4 mostra quais grupos bancários por tamanho pequeno mé dio e grande são em média mais eficientes no mercado bancário Tabela 4 Eficiência técnica média das instituições financeiras por tipo de tamanho Ano Pequeno Médio Grande Média DP 2010 08090 08923 09931 08981 00922 2011 07268 08328 09407 08335 01070 2012 07956 08453 09692 08700 00894 2013 07625 07915 09610 08384 01072 2014 07601 07709 09248 08186 00921 2015 06794 07952 08493 07747 00868 2016 07692 07973 09578 08414 01017 2017 05509 06978 08259 06915 01376 2018 08599 09002 09786 09129 00604 2019 08347 09038 09729 09038 00691 Média 07548 08227 09373 DP 00884 00656 00562 Fonte Elaborada pelos autores com base nos dados obtidos no site do BCB DP desvio padrão Em relação ao tamanho das instituições financeiras apontase que as de grande porte são em média mais eficientes 9373 do que aquelas de mé dio 8227 e pequeno 7548 portes Isso corrobora a ideia de que as instituições de grande porte por possuírem quantidade maior de agências bancárias produtos mais diversificados e melhor aplicação de tecnologias po dem alcançar níveis de eficiência mais elevados Defung et al 2016 Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 179 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 No período de desaceleração econômica em 2014 e com recessão econô mica de 20152016 notase redução de 537 da eficiência técnica média entre 2014 e 2015 puxada principalmente pelas instituições de pequeno e grande portes Essa queda da eficiência pode ser atribuída à queda do total de depósitos 744 no setor bancário que ao reduzir os insumos inputs ao mesmo tempo teve redução no volume total dos empréstimos 024 A crise econômica no Brasil entre 2014 e 2016 originouse principalmen te da redução da capacidade de crescimento econômico e da crescente eleva ção da dívida pública federal Vartanian Garbe 2019 Essas mudanças no ambiente macroeconômico podem ter impactado o setor bancário em diferen tes graus dependendo do tamanho das perdas esperadas dos empréstimos concedidos e da desaceleração das operações de crédito O pior desempenho do setor bancário em 2017 se deve sobretudo à queda de depósitos 2258 das instituições de grande porte Além disso os pro dutos outputs das instituições de pequeno médio e grande portes sofreram reduções expressivas Essa redução dos produtos outputs pode ser auferida pelas perdas nos empréstimos 232 nos investimentos de títulos e instru mentos derivativos 790 e nas receitas não financeiras 2327 Outra classificação de bancos que possui relevância nos estudos de eficiên cia bancária está associada ao tipo de propriedade ou tipo de controle Nessa classificação é comum debater se os gestores das instituições públicas são mais ineficientes em relação às instituições privadas e estrangeiras o que é normalmente justificado pela interferência do governo ou pelos incentivos inadequados para a gestão desse tipo de instituição Jayaraman Srinivasan 2019 Claessens et al 2001 Sturm Williams 2004 Na Tabela 5 apresentase a eficiência técnica média das instituições finan ceiras por tipo de propriedade público estadual e federal privado nacional e estrangeiro O grupo de bancos públicos federais BB CEF Banco do Nordeste Banco da Amazônia é considerado o tipo de propriedade com melhor desempenho ao longo do período amostral com pontuação média de 08498 8498 Apesar do baixo desempenho do Banco da Amazônia com média de 05465 5465 o BB a CEF e o Banco do Nordeste apresentaram pontuações ele vadas durante todo o período analisado o que proporcionou a alta eficiência técnica média para o grupo 180 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 Tabela 5 Eficiência técnica média das instituições financeiras por tipo de propriedade Ano Público estadual Público federal Privado nacional Estrangeiro Média DP 2010 07174 08463 09534 08216 08347 00969 2011 05834 08064 08626 07592 07529 01207 2012 07730 08313 08740 08201 08246 00415 2013 07630 08021 08777 07553 07995 00560 2014 06174 07868 08078 08258 07594 00961 2015 05309 07947 07444 07381 07020 01169 2016 07489 08272 08181 07995 07984 00350 2017 05822 09052 06556 05185 06654 01694 2018 08644 09405 09093 08556 08924 00397 2019 07423 09578 08819 08604 08606 00893 Média 06923 08498 08385 07754 DP 01070 00624 00864 00994 Fonte Elaborada pelos autores com base nos dados obtidos no site do BCB DP desvio padrão No entanto se os bancos públicos federais e estaduais fossem analisados conjuntamente as instituições mais eficientes seriam os bancos estrangeiros Isto é os bancos públicos estaduais e federais teriam pontuação de 07711 7711 que seria inferior à pontuação média de 08385 8385 dos ban cos estrangeiros Dessa forma notamse diferenças significativas entre insti tuições públicas estaduais e federais A possível justificativa para o desempenho inferior das instituições públi cas estaduais pode estar associada aos incentivos errados como remuneração ou bônus inadequados aos gestores falta de rigor na avaliação de desempe nho não transparência etc Esses problemas refletemse nas operações ban cárias como falhas na supervisão da avaliação de risco dos empréstimos e volume mais expressivo de inadimplências Quando se analisam os períodos de melhor ou pior desempenho por tipo de propriedade notase que as instituições públicas estaduais privadas nacio nais e estrangeiras são mais sensíveis aos ciclos econômicos Por exemplo no período de recessão de 2015 a 2016 as instituições financeiras exceto os Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 181 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 bancos públicos federais tiveram quedas significativas no desempenho de modo que a atividade econômica pode influenciar a composição de insumos inputs e produtos outputs do setor bancário Para tanto as variáveis macroeconômicas por exemplo taxa de cresci mento e inflação podem ter efeitos significativos em determinadas institui ções financeiras Delis Papanikolaou 2009 Assim não só os fatores inter nos relacionados à gestão dos bancos influenciam os níveis de eficiência mas também os fatores externos por exemplo reformas na regulação e variáveis macroeconômicas a depender do grau impactam o desempenho dos bancos Berger DeYoung 1997 O fato de os bancos públicos federais exceto o Banco da Amazônia terem apresentado melhor desempenho nos períodos de desaceleração econômica pode estar associado às políticas de fomento ao crédito pelo governo federal Isto é as instituições públicas federais impulsionam suas carteiras de crédito nos períodos de desaceleração econômica o que permite elevar seus produtos outputs O ano com o pior desempenho foi 2017 e isso ocorreu em grande medi da por causa da redução dos produtos outputs dos bancos públicos esta duais 903 bancos públicos federais 503 bancos privados nacionais 903 e bancos estrangeiros 769 Parte considerável das perdas no setor bancário nesse período está associada com as receitas não financeiras que em média tiveram uma redução de 2326 Os resultados do baixo desempenho em 2017 podem estar relacionados aos efeitos tardios da recessão econômica de 20152016 e para que pudes sem mitigar maiores perdas dos produtos outputs as instituições reformula ram suas operações o que reduziu seus níveis de eficiência Consequente mente em 2018 é possível observar que essa reestruturação das operações elevou expressivamente a eficiência técnica para todos os tipos de instituição financeira 52 Determinantes da eficiência técnica do setor bancário brasileiro Para ampliar a avaliação de eficiência bancária é comum complementar as análises com os determinantes da eficiência Esses determinantes podem in cluir indicadores relacionados com as informações contábeis das instituições financeiras ou estabelecer relações com variáveis micro e macroeconômicas 182 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 Neste trabalho avaliaramse os indicadores bancários para gestão interna como o índice de adequação de capital capital adequacy ratio CAR margem financeira líquida net interest margin NIM renda sem juros noninterest income NII retorno sobre ativos return on assets ROA proxy para perdas de crédito nonperformance loans ratio NPLR e variáveis macroeconômicas como a taxa de inflação INF e taxa de crescimento econômico PIB A Tabe la 6 apresenta os determinantes da eficiência técnica a partir da regressão Tobit que permite avaliar variáveis censuradas Tabela 6 Determinantes da eficiência técnica Coeficiente Estimativa Erro padrão CAR Índice de adequação de capital 18469 03166 NIM Margem financeira líquida 02333 00436 NII Renda sem juros 113612 23265 ROA Retorno sobre ativos 54678 31021 NPLR Proxy para perdas de crédito 17106 04420 INF Taxa de inflação anual 35595 14149 PIB Taxa de crescimento econômico anual 06208 09230 Fonte Elaborada pelos autores com base em dados obtidos no site do BCB e denotam significância aos níveis de 1 5 e 10 respectivamente O CAR foi negativamente significante para o modelo Essa variável é padrão nos trabalhos sobre determinantes de eficiência bancária e pode ter resultados mistos Defung et al 2016 Nesse caso o resultado negativo do CAR pode indicar que os bancos tendem a ser excessivamente conservadores com a con cessão das operações de crédito o que pode levar a uma menor eficiência Nesse contexto o trabalho de Altunbas et al 2007 indica que há uma rela ção inversa entre eficiência e capitalização isto é um alto índice de capitalização pode reduzir o risco na concessão dos empréstimos No entanto pode restrin gir a alavancagem e os retornos financeiros que consequentemente afetam negativamente o nível de eficiência A variável NIM foi considerada positivamente significativa assim como os resultados apresentados por GarzaGarcía 2012 A relação positiva da NIM Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 183 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 aponta que as receitas líquidas de juros são importantes para o nível de eficiên cia isto é os lucros que provêm de empréstimos e investimentos contribuem para o desempenho positivo da instituição financeira Podese sugerir que as instituições financeiras que procuram aumentar as margens de receitas com os empréstimos e investimentos tendem a ter maiores pontuações de eficiência Da mesma forma a NII também foi significativa para a eficiência técnica A variável NII indica como as instituições financeiras geram receitas não decor rentes de juros com o objetivo de ampliar suas receitas e melhorar os níveis de liquidez em caso de aumento de perdas esperadas da inadimplência e outros investimentos Nesse aspecto o resultado pode indicar que as instituições fi nanceiras com maior capacidade de geração de receitas sem juros podem as segurar melhor desempenho O ROA não foi significativo para o modelo apesar de ser um importante indicador para a rentabilidade das instituições financeiras Em relação aos empréstimos inadimplentes observase que a NPLR é ne gativamente significativa para o nível de eficiência do setor bancário brasileiro Esse resultado sinaliza que as instituições financeiras são sensíveis ao aumen to da inadimplência e prejudicam a rentabilidade dos negócios com emprésti mos Assim o volume elevado de empréstimos inadimplentes pode reduzir a capacidade das instituições financeiras de produzir novos empréstimos e pa gar os custos operacionais Berger DeYoung 1997 Por fim as variáveis macroeconômicas selecionadas neste trabalho como a taxa de inflação INF e taxa de crescimento econômico anual PIB alcança ram resultados distintos A inflação INF obteve resultado significativo e cor respondeu ao esperado isto é a inflação teve impacto negativo sobre as ope rações bancárias alterando os preços dos ativos e serviços o que aumentou os custos para manter a rentabilidade dos negócios bancários Tan 2016 Já a taxa de crescimento econômico PIB que na literatura comumente parece ter impacto significativo e positivo sobre a eficiência das instituições financeiras ver por exemplo GarzaGarcía 2012 no presente trabalho não apresentou significância estatística CONSIDERAÇÕES FINAIS O objetivo deste artigo foi comparar a eficiência entre instituições financeiras no Brasil durante o período de 2010 a 2019 As estimativas de eficiência téc nica para cada banco foram calculadas usandose o modelo BCCVRS orientado 184 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 ao produto output da DEA Complementarmente no presente trabalho utili zouse a regressão Tobit para encontrar estimativas sobre os determinantes da eficiência técnica obtidos na DEA Os resultados empíricos sobre o tamanho dos bancos sugerem que as ins tituições de grande porte são mais eficientes tecnicamente em comparação com as instituições de pequeno e médio portes Uma explicação possível pode estar associada a melhor uso da tecnologia maior escala de suas operações e produtos mais diversificados como destacado por Defung et al 2016 Em relação ao tipo de propriedade ou controle os bancos públicos fede rais em média apresentaramse como mais eficientes que os outros tipos de propriedade No entanto quando se comparam as instituições públicas ou seja estaduais e federais no mesmo conjunto os bancos estrangeiros passam a ser o grupo com maior nível de eficiência Parte considerável da queda de eficiência técnica dos bancos públicos nacionais está associada ao baixo de sempenho dos bancos estaduais Banestes BRB Banpará e Banrisul Os determinantes da eficiência técnica tiveram resultados mistos As variá veis internas como NIM e NII foram positivamente significativas e correspon deram à capacidade de os bancos obterem receitas de juros e sem juros res pectivamente Isso indica que a geração de receita é um importante atributo para melhorar o nível de eficiência das instituições financeiras O impacto negativo do CAR pode sugerir que o maior nível de capitaliza ção assegura maior solidez financeira contra desacelerações econômicas mas restringe a eficiência técnica das instituições financeiras A NPLR por sua vez foi negativamente significativa para o nível de eficiência que indica que o vo lume elevado de empréstimos inadimplentes reduz a capacidade de os bancos operarem de forma eficiente Quando se analisaram as variáveis macroeconômicas observouse que a inflação INF pode produzir efeitos negativos sobre a eficiência do setor ban cário brasileiro No entanto a taxa de crescimento econômico PIB não apre sentou significância em relação ao nível de eficiência o que não descarta a possibilidade de que as desacelerações na economia possam reduzir as opera ções bancárias e consequentemente a eficiência das instituições financeiras As evidências do presente trabalho são importantes para avaliação dos ges tores dos diferentes tipos de instituição financeira da mesma forma que apon tam os agrupamentos bancários com melhor e pior desempenho As pesquisas futuras por exemplo poderiam concentrarse nas reformas regulatórias sobre o setor bancário Esses resultados podem evidenciar se o nível de capital re querido tem influência no desempenho das instituições financeiras Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 185 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 BANKING EFFICIENCY IN BRAZIL 20102019 A TWOSTAGE ANALYSIS Abstract The objective of this work is to analyze the performance of the Brazilian banking sector and its determinants The present research was divided into two parts the first part seeks to determine the technical efficiency by the DEABCC model in the main Brazilian banks in the period from 2010 to 2019 In the second part the determinants that can influence the performance of banks are evaluated Esti mates indicate that large banks are on average more efficient than small and mediumsized banks In addition when looking at the type of ownership foreign banks tend to perform better compared to national banks In addition the deter minants point out that the net interest margin and noninterest income are sig nificant and contribute positively to the level of efficiency In turn the capital adequacy ratio nonperforming loans and inflation have a negative impact on technical efficiency Keywords Data Envelopment Analysis determinants technical efficiency financial institutions banking sector Referências Abreu E S de Kimura H Sobreiro V A 2019 What is going on with studies on banking efficiency Research in International Business and Finance 47 195219 httpsdoiorg101016j ribaf201807010 Ahmad N Naveed A Ahmad S Butt I 2019 Banking sector performance profitability and efficiency A citationbased systematic literature review Journal of Economic Surveys 341 185 218 httpsdoiorg101111joes12346 Altunbas Y Carbo S Garderner E P M Molyneux P 2007 Examining the relationship between capital risk and efficiency in European banking European Financial Management 131 4970 httpsdoiorg101111j1468036X200600285x Banker R D Charnes A Cooper W W 1984 Some models for estimating technical and scale inefficiencies in Data Envelopment Analysis Management Science 309 10311142 https doiorg101287mnsc3091078 Berger A N DeYoung R 1997 Problem loans and cost efficiency in comercial banks Journal of Banking and Finance 216 849870 httpsdoiorg101016S0378426697000034 186 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 Charnes A Cooper W W Rhodes E 1978 Measuring the efficiency of decision making units European Journal of Operational Research 26 429444 httpsdoiorg10101603772217 78901388 Claessens S DemirgüçKunt A Huizinga H 2001 How does foreign entry affect domestic banking markets Journal of Banking and Finance 255 891911 httpsdoiorg101016S0378 426600001023 Claessens S Laeven L 2005 Financial dependence banking sector competition and eco nomic growth Journal of the European Economic Association 31 179207 httpsdoiorg101162 1542476053295322 Cooper W W Seiford L M Zhu J Orgs 2011 Handbook on Data Envelopment Analysis Springer ScienceBusiness Media Defung F Salim R Bloch H 2016 Has regulatory reform had any impact on bank efficiency in Indonesia A twostage analysis Applied Economics 4852 50605074 httpsdoiorg101080 0003684620161170934 Delis M D Papanikolaou N I 2009 Determinants of bank efficiency Evidence from a semiparametric methodology Managerial Finance 353 260275 httpsdoiorg101108030 74350910931771 Diallo B 2018 Bank efficiency and industry growth during financial crises Economic Modelling 68 1122 httpsdoiorg101016jeconmod201703011 Emrouznejad A Parker B Tavares G 2008 Evaluation of research in efficiency and produc tivity A survey and analysis of the first 30 years of scholarly literature in DEA SocioEconomic Planning Sciences 423 151157 httpsdoiorg101016jseps200707002 Emrouznejad A Yang G 2018 A survey and analysis of the first 40 years of schorlarly literature in DEA 19782016 SocioEconomic Planning Sciences 61 48 httpsdoiorg101016jseps2017 01008 Farrell M J 1957 The measurement of productivity efficiency Journal of the Royal Statistical Society Serie A 1203 253281 httpsdoiorg1023072343100 GarzaGarcía J G 2012 Determinants of bank efficiency in Mexico A twostage analysis Applied Economic Letters 1917 16791682 httpsdoiorg101080135048512012665589 Jayaraman A R Srinivasan M R 2019 Determinants of Indian banks efficiency A twostage approach International Journal of Operational Research 362 270291 httpsdoiorg101504 IJOR2019102414 Kose M A Nagle P Ohnsorge F Sugawara N 2010 Global waves of debt Causes and conse quences World Bank Kutlar A Kabasakal A Ekici M S 2017 Efficiency of commercial banks in Turkey and their comparison Application of DEA with Tobit analysis International Journal of Mathematics in Operational Research 101 84103 httpsdoiorg101504IJMOR2017080746 Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 187 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 Levine R 1997 Financial development and economic growth Views and agenda Journal of Economic Literature 352 688726 httpwwwjstororgstable2729790 Micco A Panizza U Yañez M 2007 Bank ownership and performance Does politics matter Journal of Banking and Finance 311 219241 httpsdoiorg101016jjbankfin200602007 Nguyen T L A 2018 Diversification and bank efficiency in six ASEAN countries Global Finance Journal 37 5778 httpsdoiorg101016jgfj201804004 Staub R B Souza G S Tabak B 2010 Evolution of bank efficiency in Brazil A DEA approach European Journal of Operational Research 2021 204213 httpsdoiorg101016jejor2009 04025 Sturm J Williams B 2004 Foreign bank entry deregulation and bank efficiency Lessons from the Australian experience Journal of Banking and Finance 281 17751799 httpsdoiorg 101016jjbankfin200306005 Tan Y 2016 The impacts of risk and competition on bank profitability in China Journal of International Financial Markets Institutions and Money 40 85110 httpsdoiorg101016jintfin 2015 09003 Tecles P L Tabak B M 2010 Determinants of bank efficiency The case of Brazil European Journal of Operational Research 2073 15871598 httpsdoiorg101016jejor201006007 Vartanian P R Garbe H de S 2019 The Brazilian economic crisis during the period 20142016 Is there precedence of internal or external factors Journal of International and Global Economic Studies 121 6686 Wanke P Barros C 2014 Twostage DEA An application to major Brazilian banks Expert System with Applications 415 23372344 httpsdoiorg101016jeswa201309031 Copyright of Revista de Economia Mackenzie is the property of Universidade Presbiteriana Mackenzie Revista de Economia Mackenzie and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holders express written permission However users may print download or email articles for individual use Accepted Manuscript Efficiency Factors in OECD Banks A TenYear Analysis Peter Wanke Abul Kalam Azad CP Barros PII S0957417416303657 DOI 101016jeswa201607020 Reference ESWA 10766 To appear in Expert Systems With Applications Received date 25 January 2016 Revised date 13 June 2016 Accepted date 14 July 2016 Please cite this article as Peter Wanke Abul Kalam Azad CP Barros Efficiency Fac tors in OECD Banks A TenYear Analysis Expert Systems With Applications 2016 doi 101016jeswa201607020 This is a PDF file of an unedited manuscript that has been accepted for publication As a service to our customers we are providing this early version of the manuscript The manuscript will undergo copyediting typesetting and review of the resulting proof before it is published in its final form Please note that during the production process errors may be discovered which could affect the content and all legal disclaimers that apply to the journal pertain ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 1 Highlights This paper assesses performance of 128 OECD banks from 20042013 An analytical approach that emulates the CAMELS rating system is used Results reveal that contextual variables have a prominent impact on efficiency levels ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 2 Efficiency Factors in OECD Banks A TenYear Analysis Peter Wanke COPPEAD Graduate Business School Federal University of Rio de Janeiro Rua Paschoal Lemme 355 21949900 Rio de Janeiro Brazil Email petercoppeadufrjbr Phone 55 21 39389494 Corresponding author Md Abul Kalam Azad University of Malaya Department of Applied Statistics Faculty of Economics and Administration 50603 Kuala Lumpur Malaysia Email azadiiucgmailcom Phone60176181759 CP Barros ISEG Lisbon School of Economics and Management University of Lisboa and CEsA Research Centre on African Asian and Latin American Studies Rua Miguel Lupi 20 1249078 Lisboa Portugal Email address cbarrosisegutlpt Phone 351 213 016115 Fax 351 213 925 912 Research made with support of Calouste Gulbenkian Foundation Abstract This paper presents a performance assessment of 128 banks from 23 OECD countries from 20042013 using different financial criteria that emulate the CAMELS rating system A robust TOPSIS approach for assessing bank efficiency is also developed and presented First alternative variable reduction techniques are employed to extract the major factors within each CAMELS criterion This is done to mitigate collinearity issues Then TOPSIS is used to measure bank performance based upon these factors equally weighted A comprehensive analysis based on a weighted linear optimization model for multicriteria classification is also performed which detects any discrepancies from the original scores Lastly censored quantile regressions are combined with bootstrapped TOPSIS scores to produce a model for predicting the impact of different contextual variables on different efficiency quantiles Results reveal that the effects of ownership trend and origin of the bank may vary with respect to efficiency levels whether high or low Keywords Banks OECD factor extraction TOPSIS censored quantile regression ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 3 1 Introduction Since the inception of efficiency measurement tools in banking most studies have focused on developed economies at the country level see for instance Matousek et al 2014 and the reference therein In this sense Paradi et al 2011 listed the ten countries most targeted by researchers worldwide all were developed economies except for India Some studies following a different perspective focused on economic blocks and analyzed specific issues such as publicprivate ownership bank origin and merger and acquisitions MA processes considered relevant to understanding banking performance Sturm Williams 2010 Ebrahim Girma Shah Williams 2014 In this regard OECD banks are of particular interest A handful of papers have analyzed different aspects of banking activity between and within OECD countries Jaffee Levonian 2001 Levintal 2013 Müller Uhde 2013 However to the best of our knowledge comparative performance studies within the realm of OECD banks are still scarce Thus this paper aims to contribute to the current body of knowledge by exploring the use of a robust TOPSIS Technique for Order of Preference by Similarity to Ideal Solution approach applied to a set of financial criteria that emulates the CAMELS rating system for purposes of assessing the performance of 128 banks from 23 OECD countries from 20042013 Although numerous studies on bank efficiency performed at the country level have been produced they are still concentrated in two main approaches best known as parametric and nonparametric Berger Humphrey 1997 De Borger Ferrier Kerstens 1998 Athanassopoulos 1998 Brandouy Briec Kerstens Van de Woestyne 2010 Brissimis Delis Tsionas 2010 Kerstens Mounir Van de Woestyne 2011 Briec Liang 2011 Lampe Hilgers 2014 Cummins RubioMisas 2006 The most popular parametric method is the Stochastic Frontier Approach or SFA whereas the most popular nonparametric method is the Data Envelopment Analysis or DEA Gunay 2012 Yang 2009 Chen 2002 Tözüm 2002 Li Liu Liu Whitmore 2001 Deliktas Balcilar 2005 Nonparametric methods however are more widely used in different banking industries in various countries and regions around the globe Wanke et al 2015 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 4 An emerging trend in banking performance measurement entails the use of multi criteria decisionmaking MCDM methods such as TOPSIS The TOPSIS method suggested by Hwang and Yoon 1981 belongs to the group of linear ranking methods of multidimensional objects Broadly speaking the ranking of objects from best to worst considering an assumed synthetic measure which is not subjected to a direct measurement is a task of linear ordering Dudek and Jefmanski 2015 According to Behzadian Khanmohammadi Otaghsara Yazdani and Ignatius 2012 TOPSIS studies in banking are still scarce For example Shaverdi et al 2011 applied TOPSIS to rank the performance of banks based on criteria weights obtained using a fuzzy analytical hierarchy process FAHP Similarly Mandic Delibasic Knezevic and Benkovic 2014 examined Serbian bank performance by first applying FAHP and then TOPSIS in a two stage approach Among earlier studies İÇ and Yurdakul 2010 proposed a ranking model using financial ratios and fuzzy TOPSIS for computing credit risk Similarly Tansel İç 2012 proposed a model combining fuzzy TOPSIS with linear programming for determining credit risk Most recently Wanke Barros and Macanda 2015 studied Angolan bank efficiency using TOPSIS and neural networks in a twostate approach This paper differs from previous studies and contributes to the state of art of the methodology in efficiency measurement by developing and presenting a robust TOPSIS approach using 10 years of panel data of OECD banks The robust TOPSIS approach adopted here is anchored in three key aspects which are further detailed in the methodology section First of all two different variable reduction techniques Principal Component Analysis PCA and Independent Component Analysis ICA are alternatively used to extract the main factors embedded in the several financial ratios that compose the different CAMELS criteria This is necessary not only to reduce the number of criteria to a manageable number of factors but also to avoid collinearity issues between different financial criteria Second the weighted linear optimization model for multi criteria classification proposed by Ng 2007 used an auxiliary tool of which different financial criteria orders and therefore their respective set of weights are comprehensively tested In this case any discrepancies can be measured and compared against the original TOPSIS scores computed with equal weights for the PCA and ICA extracted factors In contrast with other MCDM methods decision makers are free to determine their own set of ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 5 criteria weights in TOPSIS Put differently because criteria weights are defined exogenously in TOPSIS sensitivity analyses can be performed on different weighting criteria to assess the robustness of the results Thus the role of the Ng model within the ambit of TOPSIS to provide TOPSIS the weights for each criteria computed during the optimization stage Third bootstrapped truncated regressions similarly to what is done with classical DEA models are used to resample original TOPSIS scores and then in a censored quantile regression model to explore the impact of different contextual variables on different efficiency quantiles Results presented in this research contribute to the growing literature on banking efficiency especially by shedding light on how country origin ownership and MA may affect financial performance These results are consistent with and extend the findings of recently conducted studies Dima Dincă Spulbăr 2014 Du Sim 2015 Fungáčová Poghosyan 2011 Gardener Molyneux Hoai 2011 GarzaGarcia 2012 CC Lee Hsieh 2014 Sufian 2007 thus adding to the body of knowledge The remainder of the paper is organized as follows Section 2 presents the contextual setting Section 3 covers the literature review and Section 4 presents the methodology Data is presented in Section 5 while empirical results are presented and discussed in terms of policy implications in Section 6 Conclusions follow in Section 7 2 Contextual Setting Efficiency in OECD banks appears to be substantially impacted at the country level thus suggesting that regulatory barriers cultural aspects and the economic dynamics within each country are possible drivers for understanding higher or lower levels of banking performance Furthermore temporal analyses of basic financial ratios suggest that differences in banking indicators between OECD countries seem to spread extending over several years especially after the world financial crisis This observation also indicates that each country responded to the financial crisis in an individual way Thus four key development indicators and bank information are presented in Figure 1 and Table 1 First an inspection of the nonperforming loans to total gross loans in Figure 1 makes it clear that not all the selected countries pattern similarly along the time ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 6 series According to world development indicators nonperforming loans usually express the ratio between the value of nonperforming loans and the total value of the loan portfolio including nonperforming loans before the deduction of specific loanloss provisions Figure 1 depicts a common pattern of increasing frequency of nonperforming loans among the main six OECD countries Greece has the highest percentage followed by Ireland Italy Hungary Slovakia and Portugal which range from 343 to 1052 On the other hand Poland has successfully managed to decrease its level of nonperforming loans from 215 in 2001 to 57 in 2014 Figure 1 also reveals that all the major shifts of non performing loans among the above countries started after the global financial crisis of 2008 For the other selected countries bank risk for nonperforming loans remains relatively unchanged compared with the period prior to the financial crisis Figure 1 Nonperforming loans to total gross loans Source World development indicators retrieved on 251015 On the other hand Figure 2 presents liquid reserves to assets ratio for the selected OECD countries The significant finding related to Figure 2 is that a reasonable number of anomalies can be observed after the global financial crisis of 2008 All of the countries except Greece 95 and Denmark 54 had been experiencing an increase of liquid reserves to bank assets ratio within a limit of 08 to 301 from 2001 to 2007 But after 2008 these countries experienced significant variation throughout the following ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 7 years and to date the range between 06 and 62 The exceptional countries with banks having a high percentage of liquid assets are United States with 167 and Poland with 132 This suggests that these countries have economic responses in place to enhance their ability to overcome the worse economic conditions in the postcrisis period Figure 2 Banks liquid reserves to bank assets ratio Source World development indicators accessed 10252015 The remaining two variables bank capital to assets ratio and GDP per capita growth annual are presented in detail in Table 1 The data pertaining to bank capital to assets ratio show that all the selected countries exhibit similar patterns over the years Interestingly the United States has achieved a comparatively higher rate in all years However for the remaining member countries results range from 117 Ireland to 43 Finland This particular issue has revealed that OECD banks have been achieving a standard ratio in all these years Last but not the least GDP growth has a positive correlation pattern among these countries as presented in Table 1 In the year 2015 the GDP growth stayed within a range of 45 for Ireland to 22 for Italy The world economic crisis of 2008 is still having an impact here The only exception is Greece which is still struggling to recover economically In summary these four indicators exemplify the importance of the country effect and its underlying variables regulation economic dynamics culture for comprehending certain aspects of banking efficiency in OECD countries Section 3 further explores i how these financial indicators and others are ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 8 combined to emulate the CAMELS rating system and ii the expected impact of other contextual variables such as ownership and merger and acquisitions on efficiency levels in light of the current literature Table 1 Country level data of OECD Countries Country Name 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Bank capital to assets ratio 1 Australia 5200 NA 6900 5200 5000 5400 6289 6286 6070 6140 6004 5880 Austria 4900 NA NA 5240 6484 6284 6977 7493 7186 7770 7986 6922 France 5400 NA NA NA NA 3728 4098 4878 4754 5201 5818 NA Greece 6900 NA NA 7000 6800 6400 6959 6665 NA 5758 7548 9114 Luxembourg 5800 NA NA 5000 5000 4796 5537 5234 4998 6284 6392 6099 New Zealand NA NA NA NA 5300 6400 5500 6000 6500 7100 7300 7400 Poland 8300 NA NA 7800 8000 7500 8084 8198 7818 8693 9098 9253 Norway 5900 NA NA 7000 6400 5900 6000 6700 6800 NA NA NA Korea Rep 7000 NA NA NA NA 6295 7294 7595 8114 8231 8273 8036 Italy 6400 NA 4600 4600 4600 4100 4816 4979 5448 5386 5404 5922 Ireland 5300 NA 4329 4508 4395 3746 5435 5323 6440 7279 7749 11486 Finland 10900 NA 9400 9400 8000 6200 6407 5525 4354 4417 5011 4279 Denmark 5900 NA 4900 5700 6200 5500 5493 5430 5054 5266 7061 7466 Belgium 3100 NA NA 3300 4300 3300 4628 4964 4600 5761 6390 6335 Germany 4200 NA 4070 4329 4267 4500 4807 4298 4362 4727 5453 5612 Spain 7300 NA 6300 6400 6700 5900 6418 6095 5924 5757 6792 7242 Slovenia 8300 NA 8500 8400 8400 8400 8300 8200 NA NA NA NA Sweden 5000 NA NA 4900 4800 4700 5000 NA NA NA NA NA Portugal 5800 NA 6152 6641 6526 5800 6494 6697 5327 6701 6929 4651 United Kingdom 6600 NA NA 6100 5500 4400 5393 5366 5100 5509 6345 5865 United States 9200 NA 10300 10500 10300 9300 12375 12739 12229 11963 11776 11719 Switzerland 5700 NA NA 4900 4600 4837 5551 5401 5548 5514 5707 NA GDP per capita growth2 Australia 1815 2956 1863 1482 3114 1645 0343 0389 0910 1956 0749 0883 Austria 0267 2071 1447 2841 3286 1230 4051 1635 2725 0425 0356 0347 France 0107 2032 0845 1664 1730 0363 3439 1463 1587 0271 0221 0246 Greece 6303 4597 0553 5482 3212 0653 4400 5164 8616 6315 3327 1416 Luxembourg 0032 3445 2543 3222 4832 1293 7071 3243 0354 2532 0340 NA New Zealand 2587 2285 2250 1519 2004 2447 1244 0320 1442 1619 1691 NA Poland 3632 5197 3593 6267 7215 3852 2553 4006 4713 1821 1775 3491 Norway 0330 3347 1928 1574 1870 0859 2855 0643 0333 1408 0469 1103 Korea Rep 2423 4507 3711 4668 4974 2092 0230 6005 2913 1832 2455 2892 Italy 0292 0928 0455 1700 0963 1703 5911 1398 0414 3032 2831 2216 Ireland 1294 2684 3397 2665 1942 4575 7317 0817 2399 0533 0075 4465 Finland 1751 3625 2429 3657 4738 0253 8707 2522 2096 1894 1775 0564 Denmark 0117 2374 2155 3456 0378 1300 5594 1175 0737 1028 0900 0693 Belgium 0468 2988 1335 1955 2247 0159 3399 1339 0446 0631 0198 0685 Germany 0775 1203 0764 3827 3408 1244 5399 4250 3564 2089 0167 1298 1 Bank capital to assets is the ratio of bank capital and reserves to total assets Capital and reserves include funds contributed by owners retained earnings general and special reserves provisions and valuation adjustments Capital includes tier 1 capital paidup shares and common stock which is a common feature in all countries banking systems and total regulatory capital which includes several specified types of subordinated debt instruments that need not be repaid if the funds are required to maintain minimum capital levels these comprise tier 2 and tier 3 capital Total assets include all nonfinancial and financial assets Definition retrieved from httpdatabankworldbankorg on 10252015 2 Annual percentage growth rate of GDP per capita based on constant local currency Aggregates are based on constant 2005 US dollars GDP per capita is gross domestic product divided by midyear population GDP at purchasers prices is the sum of gross value added by all resident producers in the economy plus any product taxes and minus any subsidies not included in the value of the products It is calculated without making deductions for depreciation of fabricated assets or for depletion and degradation of natural resources Definition retrieved from httpdatabankworldbankorg on 10252015 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 9 Spain 1338 1402 1986 2428 1866 0484 4424 0446 0970 2152 0906 1860 Slovenia 2780 4285 3823 5320 6345 3137 8627 0781 0404 2844 1133 2524 Sweden 2005 3911 2408 4101 2641 1329 5989 5089 1892 1021 0428 1337 Portugal 1305 1568 0580 1370 2291 0055 3071 1852 1682 3638 1063 1472 United Kingdom 3816 1873 2103 2287 1760 1113 5032 1116 0854 0038 1020 1911 United States 1927 2830 2397 1681 0815 1230 3624 1678 0827 1542 1449 1631 Switzerland 0690 2138 2380 3362 3214 0986 3345 1887 0679 0041 0750 NA Source World development indicators World bank data retrieved 10252015 3 Literature Review This section is subdivided into two subsections The first reports on previous studies that used financial criteria to emulate rating systems and compute efficiency scores The second presents the findings of the previous studies that focused on the impact of exogenous contextual variables on efficiency scores The major research proposition is presented at the end of this section 31 Financial Indicators and Banking Performance Considering that the different efficiency measurement methods as discussed in the introduction can provide efficiency scores it is crucial to establish the linkages between banking efficiency and financial indicators Wanke Barros Faria 2015 Indeed these methods should be capable of indicating how effective a bank is in minimizing financial indicators related to decreasing performance and in maximizing others related increasing performance Tsai Chang 2010 This finetuning between efficiency scores and financial indicators is often accomplished by selecting a suitable set of criteria and the attendant impacts positive or negative on banking performance Several financial criteria or indicators are thought to be associated with banking performance For instance the prediction of failures poor performance using firmspecific characteristics and financial indicators dates back to the seminal works of Altman 1968 and Altman Haldeman and Narayanan 1977 which applied discriminant analysis to financial ratios in order to derive the Zscore approach More recently Männasoo and Mayes 2009 presented a comprehensive literature review on this subject According to these authors although there is no universal set of indicators across previous studies the ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 10 CAMELS criteria appear to have a significant capacity for detecting distress CAMELS stands for capital adequacy C asset quality A management efficiency M earnings E liquidity L and sensitivity to market risk S In recent decades several studies reported on the use of the CAMELS variables in risk measurement and monitoring Examples can be found in Cole and Gunther 1995a DeYoung 1998 Oshinsky and Olin 2006 Kumar and Ravi 2007 Poghosyan and Cihak 2011 and Ravisankar Ravi and Bose 2010 More recently Wanke Barros and Faria 2015 presented a practical application that emulated the CAMELS rating systems in the Brazilian banking industry using DEA dynamic slacks The fundamental ideas behind this practical application are embedded in the close relationship between efficiency levels and distance to the frontier of best practices the latter may be considered as proxies of looming financial distress In other words a consistent movement towards lower efficiency levels over the course of time may constitute a leading distress indicator It should be noted however that the original criteria used to determine the CAMELS ratings are not disclosed to the general public Jin Kanagaretnam Lobo 2011 Therefore proxies are often selected accordingly based both on prior studies and availability of data Table 2 contains a list of the major financial criteria or indicators commonly used to emulate the CAMELS rating system as found in recent papers Table 2 CAMELS sub criteria commonly used Betz Oprică Peltonen and Sarlin 2014 Maghyereh and Awartani 2014 Wang Lu and Wang 2013 Wang Lu and Lin 2012 Doumpos and Zopounidis 2010 Secme Bayrakdaroglu and Kahraman 2009 Zhao Sinha and Ge 2009 Cole and Gunther 1995 Cole and Gunther 1998 Capital adequacy ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 11 Total Regulatory Capital Ratio Equitiestotal assets Equities Fixed assetsTotal assets BASEL ratio Assets Quality Loan Loss Res Gross Loans Loan Loss Provision Net Interest Rev Loan Loss Res Impaired Loans Impaired Loans Gross Loans NCO Average Gross Loans Impaired Loans Equity Tier 1 Ratio Management Net Interest Margin Net Interest Revenue Average Assets Other Operating Income Average Assets NonInterest Expense Average Assets Earning quality Return on Average Assets ROAA Return on Average Equity ROAE NonOperating Items Net Income Cost to Income Ratio Liquidity Interbank ratio Net Loans Total Assets Net Loans Deposits ST Funding Liquid Assets Total Deposit Borrowings Sensitivity of Market risk Rate sensitive assetsrate sensitive liabilities Average earning asset Rate sensitive assetsrate sensitive liabilities Risk weighted asset II Risk weighted asset I II Share of trading income ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 12 32The impact of exogenous contextual variables on efficiency scores Exogenous contextual variables are thought to affect banking efficiency in a unidirectional way and therefore are directly linked to policy formulation Put differently exogenous contextual variables affect efficiency levels without being affected by them As a matter of fact exogenous contextual variables represent decision variables based on the banking discretion rather than endogenous variables generated within the ambit of an efficiency model or a production process such as the country effect discussed in Section 2 Past studies in banking have explored these exogenous contextual variables and their impacts on efficiency levels Assaf Barros Matousek 2011 Assaf Barros Ibiowie 2012 Assaf Matousek Tsionas 2013 We next discuss the impacts of these exogenous contextual variables on banking efficiency Here the comprehensive theoretical framework presented in Barros and Wanke 2014 and Wanke et al 2015 is adopted Broadly speaking merger and acquisition processes in the financial sector are well known for leveraging banks financial and operational indicators The same basic economic principles apply to smaller banks in emerging countries which are the first to suffer the consequences of systemic financial crises On the other hand private instead of public ownership may be accountable for differences in production technology and managerial style thus affecting efficiency levels More specifically public banks are common in some countries and whether they are more efficient than private banks is a matter of debate Dinc 2005 Galindo and Micco 2004 Although they are not usually tested in efficiency analyses in the OECD context public banks are representative justifying therefore the testing of whether or not they are less efficient than private banks This hypothesis was previously examined by Kumar and Gulati 2010 and Mohan and Ray 2004 On the other hand mergers and acquisitions between banks are known as horizontal mergers Andrade et al 2001 and aim to improve cost performance and synergy through a larger market share In the former case the merged ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 13 companies reduce operating costs but retain the facilities of the merged or acquired company Garette and Dussauge 2000 This discussion leads us to the major proposition of this study efficiency levels in the OECD banks are significantly affected by endogenous contextual variables related to the bank origin country effect which may reflect regulatory issues cultural barriers and economic dynamics as discussed in Section 2 and by exogenous contextual variables that is subject to managerial discretion such as ownership publicprivate and MAs In this regard in examining the US banking industry AlSharkas Hassan and Lawrence 2008 found that merging has a positive influence on banks cost and profit efficiency They posited the technological advantage of allocative efficiency as the leading driver of bank mergers in the US Moreover Pancurova and Lyocsa 2013 examined the banking sector of 11 Central and Eastern European Countries in order to find the determinants bank efficiency Their study revealed that both bank ownership and country of origin have significant influence on bank efficiency performance A similar study by Sufian and Kamarudin 2015 examined determinants of revenue efficiency of Islamic banks among the Southeast Asian countries and showed that bank ownership has significant influence on banks revenue efficiency An analogous result is also available in some influential studies including Lee Kim 2013 and Minh Long Hung 2013 4 Methodology This section describes the building blocks upon which the robust TOPSIS analysis is developed in this research The fundamentals of the TOPSIS method are presented in Section 41 then a robust approach for criteria weighting in TOPSIS is presented in Section 42 using the Ng model 2007 as cornerstone Section 43 discusses why data reduction techniques such as PCA and ICA are deemed necessary to extract the major factors from a set of several financial indicators so that collinearity can be properly handled in TOPSIS Lastly the fundamentals of censored quantile regressions and the underlying role of bootstrapped truncated regressions similar to those applied to traditional DEA models are discussed in Section 44 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 14 41 TOPSIS The design and application of Multi Criteria Decision Making MCDM methods has been a prolific research field in business analytics and engineering As a matter of fact over the course of the last three decades a variety of different MCDM have been developed and empirically employed in field research This list includes the AHP Saaty 1980 Ramanathan 2013 Promethee Brans Vincke 1985 Corrente Greco Slowinski 2013 Electre HatamiMarbini Tavana 2011 Corrente et al 2013 Dematel Tsay Chou Hsu 2009 Vikor Opricovic 1998 Opricovic Tzeng 2007 and Uta Siskos Askounis Psarras 2014 TOPSIS is one of the MCDM approaches used when the weighting criteria for choosing the most adequate method is exogenously defined Wang Nistor Murty Wei 2014 BilbaoTerol et al 2014 Feng Wang 2001 Pirdavani Brijs Geert 2010 Chen Leng Mao Liu 2014 Hassan Hawas Ahmed 2013 Barros Wanke 2015 Its basic principle assumes that the chosen alternative should simultaneously be closest to the positiveideal solution and farthest from the negativeideal solution Hwang Yoon 1981 Ertugrul Karakasoglu 2009 More precisely the positiveideal solution is the one that maximizes the benefit and simultaneously minimizes the total costs while the negativeideal solution is the one that minimizes the benefit and simultaneously maximizes the cost Lai Liu Hwang 1994 Wu Lin Lee 2010 In spite of TOPSIS general resemblance to DEA objectives where outputs may be maximized and or inputs minimized in nonradial radial models the determination of the weights relative to the importance of each criteria is a key step in TOPSIS In DEA in contrast these weights are defined within the ambit of the model As a matter of fact while DEA models rely on the optimization of the distance from each firm to the convexefficient production frontier by finding a proper set of weights which are endogenously defined TOPSIS employs purely analytical methods of Euclidean distance functions on normalized vectors of positive outputs and negative inputs criteria given that the weights have been previously defined by the decisionmaker Barros Wanke 2015 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 15 The TOPSIS technique is built on an evaluation matrix consisting of m alternatives and n criteria with the intersection of each alternative and criteria given as Therefore one obtains a matrix This matrix should be firstly normalized from a regulated matrix as demonstrated in Eq 1 1 After normalization the weighted normalized decision matrix for efficiency assessment should be calculated observing Eq 2 2 where is the weight given to the criteria and Once the weighting criteria are defined the worst alternative the negative ideal assessment unit and the best alternative the positive ideal assessment unit should be defined using Eqs 3 and 4 3 4 where and which are a set of positive benefit and negative cost attributes respectively Given the best and the worst alternatives the distance between the target alternative i and the worst condition should calculated using Eq 5 5 and the distance between the alternative i and the best condition Ab by observing Eq 6 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 16 6 where dia and dib are the Euclidean distance from the target alternative i to the worst and best conditions respectively Then the similarity of alternative i to the worst condition the inefficient best conditions should be computed respectively as 7 where 0 Si 1 i 12 m if and only if the alternative solution has the worst condition if and only if the alternative solution has the best condition According to Jahanshahloo Lotfi and Davoodi 2009 represents the efficiency scores for each alternative that is for each bank of the sample over the course of the years as determined by the decision making criteria Lastly the alternatives should be ranked according to a higher value of indicates a better solution with respect to higher efficiency levels within the environment of 128 OECD banks thereby enabling the subsequent assessment of the impact of the contextual variables 42 Robust TOPSIS weighting Applying the Ng model With respect to the definition of the weighing criteria several different methods can be found in literature thus there is not a single methodological procedure to be followed Madeira Junior Cardoso Junior Belderrain Correia Schwanz 2012 Hemmati et al 2013 for instance used the Entropy technique Seçme et al 2009 and Shaverdi et al 2011 in contrast used fuzzy AHP for determining the weights of main and subcriteria In our study the 21 financial indicators or criteria presented in Table 3 were initially given the same weight since the literature review provides no clear indication as to which criteria should be prioritized for ranking banks in the OECD context ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 17 Importantly this study conducts a robustness analysis based on the guidelines provided in Madeira Junior et al 2012 The underlying idea is to assess how different weighting criteria would impact the computation of TOPSIS scores compared to those derived from the initial ones where the equal weighting assumption was considered The steps taken in generating different weightings are further detailed in the next paragraphs and presented in the discussion section Readers should note however that despite the weighting criteria adopted the contextual variables are further used as performance predictors in the Censored Quantile Regression CQR analysis for the base case where all weights are assumed to be equal The Ng weighting model applied to this research assumes that there are I bank observations and that they should be ranked in terms of J criteria or financial indicators Further let the performance of i th bank observation in terms of each of the j th criteria or financial indicator be denoted as ij y The purpose is to aggregate multiple performance scores of a bank observation with respect to different criteria or financial indicators into a single score for the subsequent ranking In the Ngmodel the author first transforms all measures to a baseline for purposes of comparison Using the transformation min max min ij ij ij ij y y y y 8 The Ngmodel converts all measurement in a 01 scale for all items To facilitate the bank ranking under multiple criteria Ng defines a nonnegative weight w ij which is the weight of contribution of performance of the i th item under the j th criteria to the score of the item The criteria are assumed to be ranked in a descending order such that 1 2 i i iJ w w w for all item i The Ng 2007 for aggregation purposes is given as follows max 1 J i ij ij j S y w st 1 1 J ij j w 9 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 18 1 2 i i iJ w w w ij 0 w 1 1 i I j J 43 Techniques for Factor Extraction Efficiency measurement techniques often employ some kind of weighting on the inputs and outputs of each observation This being the case whenever inputs and outputs are strongly correlated the accuracy of the scores computed for each observation may be biased Madeira Junior et al 2012 Another frequent problem found in some efficiency measurement techniques resides on their reduced discriminatory capability that is sometimes efficiency scores are concentrated towards one Bian 2012 Jenkins Anderson 2003 To mitigate these problems a comparative PCAICA factor extraction is proposed in this research within the ambit of the robust TOPSIS Kao Lu and Chiu 2011 verified that ICA stands next to PCA in terms of effectiveness as a solution to the discrimination problem of the DEA model Pioneering works that applied factor extraction and efficiency measurement techniques include Adler and Golany 2001 2002 and Adler and Yazhemsky 2010 who used PCA to increase the discriminatory capability of DEA scores and therefore reduce the so called curse of dimensionality Jenkins Anderson 2003 Recent applications of PCA in conjunction with TOPSIS can be found for instance in Wanke Barros and Macanda 2015 It is worth mentioning within the realm of the innovative applications of this research that ICA is a statistical signal process technique for finding independent sources given only observed data that are mixtures of the unknown sources without any prior knowledge of the mixing mechanisms Lee 1998 Hyvarinen Karhunen Oja 2001 The aim is to extract the hidden information from the observed multivariate statistical data where no relevant signal mixture mechanisms are available This hidden information is called the independent components ICs of the observable data The ICs are properly selected based on their statistical independency and are regarded as the key factors affecting the efficiency measurement results Shi Liu Sun 2006 Zheng Huang ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 19 Shang 2006 There are few papers that integrate ICA with efficiency measurement techniques such as DEA Kao et al 2011 So far to the best of our knowledge ICA has not been used in conjunction with TOPSIS in previous research The estimated ICs are regarded as more meaningful variables representing the independent sources or hidden information of the inputoutput variables These issues are further explored below 431 Principal Component Analysis PCA PCA explains the variance structure of a data matrix through linear combinations of variables consequently reducing the data to a few principal components which generally account for 8090 of data variance If most of the population variance can be attributed to the first few components then they can replace the original variables with minimum loss of information Per Hair Anderson Tatham and Black 1995 let the random vector ie the original inputsoutputs chosen to be aggregated have the correlation matrix with eigenvalues and normalized eigenvectors Consider the linear combinations where the superscript t represents the transpose operator 10 11 12 The principal components PCs are the uncorrelated linear combinations ranked by their variances in descending order The PCA used here is based on correlation rather than on covariance because the variables used in TOPSIS are often quantified in different units of measure In general the inputs and outputs of a TOPSIS need to be strictly positive however PCs can contain negative values Readers should note that PCA identify underlying variables that are uncorrelated with each other Intuitively this is desirable because the underlying variables that account for a set of measured variables should correspond to physically different processes which ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 20 in turn should have outputs that are uncorrelated with each other However specifying that the underlying variables should be uncorrelated imposes quite weak constraints on the form these variables take It is the weakness of these constraints that ensures the factors extracted can be rotated in order to find a more interpretable set of factors without affecting the zero correlations between factors Most factor rotation methods yield a statistically equivalent set of uncorrelated factors The variety of factor rotation methods available is regarded with relative skepticism by some researchers This is because it is sometimes possible to use factor rotation to obtain new factors that while easily interpreted are statistically no more significant than results obtained with other factor rotations By contrast any attempt to rotate the factors independent components extracted by ICA would yield nonindependent factors Thus the independent components of ICA do not permit postICA rotations because such factors are statistically independent as further detailed in the next section 432 Independent Component Analysis ICA is based on the assumption that source signals are not only uncorrelated but also statistically independent Essentially if two variables are independent then the value of one variable provides absolutely no information about the value of the other By contrast even though two variables are uncorrelated the value of one variable can still provide information about the value of the other ICA seeks a set of statistically independent signals amongst a set of signal mixtures on the assumption that such statistically independent signals are derived from different physical processes The objective of finding such a set of statistically independent signals is achieved by maximizing a measure of the joint entropy of the extracted signals In the basic ICA model the observed random variables are generally assumed to be a linear mixture of the unknown sources from a mixing matrix It can be expressed as Hyvarinen et al 2001 13 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 21 where is the vector of observed random variables is the vector of the unknown latent variables meaning that they cannot be directly observed from the mixture variables is an unknown mixing matrix of size n n The problem is then to estimate both the mixing matrix A and latent variables b using observations of z alone In order to estimate the unknown latent variables b the de mixing matrix W is used to transform the observed random variables z to yield the independent components ICs The independent components g are used as the estimates of the latent variables b The ICA modeling is formulated as an optimization problem by setting up the measure of independence of ICs as an objective function and using some optimization techniques to solve for the demixing matrix W There exist several algorithms performing ICA Bell Sejnowski 1995 Hyvarinen Oja 2000 In general the ICs are obtained by using the demixing matrix W to multiply the mixture variables The de mixing matrix W can be determined using an unsupervised learning algorithm with the objective of maximizing the statistical independence of ICs The ICs with nonGaussian distributions imply the statistical independence Bell Sejnowsky 1995 Hyvarinen Oja 2000 The nonGaussianity of the ICs can be measured by the negentropy Hyvarinen et al 2001 14 where is a Gaussian random vector of the same covariance matrix as g H is the entropy of a random vector g with density pg η defined as 15 The negentropy is always nonnegative and is zero if and only if g has a Gaussian distribution Since the problem in using negentropy is computationally difficult an approximation of negentropy is proposed as follows Hyvarinen Oja 2000 16 where o is a Gaussian variable of zero mean and unit variance and g is a random variable of zero mean and unit variance F is a nonquadratic function and is given by ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 22 in this study The FastICA algorithm proposed by Hyvärinen and Oja 2001 is adopted in this paper to solve for the demixing matrix W and independent components 44 Censored Quantile Regression Traditional regression models cannot answer an important question do contextual variables influence efficiency levels differently for low efficiency OECD banks than for those with high or average efficiency levels A more comprehensive picture of the effect of the contextual variables on the efficiency levels can be obtained via CQR Quantile regression models the relation between a set of contextual variables and specific percentiles or quantiles of the response variable it specifies changes in the quantiles of the efficiency For example a median regression 50th percentile of efficiency on a banks contextual variables specifies the changes in the median efficiency as a function of these predictors In this sense the effect of private ownership on median efficiency levels can be compared to its effect on other quantiles of efficiency in OECD banks As a matter of fact the quantile regression parameter estimates the change in a specified quantile of the response variable produced by a one unit change in the predictor variable This allows one to compare how some percentiles of the efficiency levels may be more affected by certain bank characteristics or contextual variables than other percentiles This is reflected in the change in the size of the regression coefficient According to Leng and Tong 2013 the quantile regression was introduced by Koenker and Bassett 1978 and has become an increasingly important tool in statistical analysis Indeed their general quantile regression QR estimation became the most popular approach Chernozhukov Hong 2002 Contrary to the usual model for the conditional mean it provides distributional information on the dependence of T on Z The Ƭth conditional quantile function of the dependent variable T given covariates Z QT ƬZ is defined as QT ƬZ infv F0vZ Ƭ where F0 is the cumulative conditional distribution function of T given Z Correspondingly a quantile regression model for QT ƬZ with Ƭ 0 1 can be denoted as QT ƬZ Z 15 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 23 When data are subject to censoring statistical estimation and inference for quantile regression is more involved Indeed a naive procedure which completely ignores censoring may give highly biased estimates Koenker dOrey 1987 Equivariance to monotone transformations is an important property of quantile regression models Powell 1986 Powell 1986 1984 first studied Censored Quantile Regression CQR with fixed censoring For random censoring Ying Jung and Wei 1995 proposed a semiparametric median regression model Despite the simplicity of their method this procedure requires the unconditional independence of the survival time and censoring time This assumption is often restrictive as conditional independence given the covariates is more natural Kalbfleisch Prentice 2002 In addition the estimating equation approach proposed by Ying et al 1995 involves solving nonmonotone discrete equations creating optimization difficulty As a consequence inferential procedures such as the resampling approach by Jin Ying and Wei 2001 or the bootstrap method can be computationally unattainable Chernozhukov and Hong 2002 argue that the CQR models allow covariates to shift location scale and the entire shape of the distribution and permit distributionfree specifications As such CQR models compare favorably to the normal AmemiyaTobin Cox BuckleyJames and other approaches According to the authors in this model the latent variable is left censored by the observable possibly random censoring points Ci and we observe Yi Ci Xi Ci σi 1 Yi Ci 16 is assumed to be conditionally independent of the censoring point Ci that is for all y IR P Y yXi Ci PY yXi so that TZ Z 17 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 24 Conditioning on Ci equation 5 and the equivariance transformation yield the following CQR model Powell 1986 QYi Xi Ci TZ Z Ci 18 More recently Bose and Chatterjee 2003 and Portnoy 2013 offered alternatives for resampling xy pairs based on either conventional multinomial resampling or using exponential weighting and on a deleted jackknife respectively Similarly to what is done with bootstrapped truncated regressions for DEA scores please refer to Wanke et al 2015 for a review on the subject the bootstrapping approach is adopted here uses the original TOPSIS scores within the ambit of censored quantile regressions Resamples with a size of 200 replications are adopted here Readers should refer to Koenker 2008 for more details on how implementing this procedure in R 5 The Data The data on OCDE banks was obtained from BankScope database3 As previously mentioned the TOPSIS alternatives consisted of each one of the 1280 samples formed by the combination of 128 banks over 10 years The underlying idea is to develop a model for assessing banking performance in OECD countries based on the criteria or financial indicators that could emulate the CAMELS rating system The model is based on the premise that banking performance should be viewed as a result of capital adequacy C assets quality A management M earning quality E liquidity L and sensitivity to market risk R The criteria observed the financial indicators commonly found in the literature review presented in Section 31 and data availability Financial indicators have been grouped observing the major CAMELS criteria as presented in Table 3 which also gives their major descriptive statistics and their expected impact on efficiency levels Negative impacts represent criteria that should be minimized while positive impacts represent 3 Available at httpsbankscopebvdinfocomversion0151021homeservproductscope2006 retrieved 05282015 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 25 criteria that should be maximized when deriving the ideal and antiideal solutions Figure 3 depicts the correlations between the financial ratios used to emulate the CAMELS rating system as presented in Table 3 The data source also provided information on contextual variables for each bank such as ownership public or private country of origin and whether the bank underwent an MA process All variables were coded as dummies Table 3 Descriptive statistics for the TOPSIS criteria and the contextual variables TOPSIS Criteria Subcriteria Impact Min Max Mean SD C Total Capital Ratio 9860 223800 16455 13687 Equity Tot Assets 9588 99461 14718 19255 Equity Net Loans 22116 847475 29831 70886 Equity Cust Short Term Funding 10298 965997 41039 117182 Equity Liabilities 9284 984648 31385 100536 A Loan Loss Res Gross Loans 0005 44915 3171 4189 Loan Loss Prov Net Int Rev 959933 580576 23865 60821 NCO Net Inc Bef Ln Lss Prov 682609 612582 21479 105358 Tier 1 Ratio 9860 223000 12767 9938 M Net Interest Margin 18003 71334 3447 7025 Net Int Rev Avg Assets 2744 67069 3130 6274 Oth Op Inc Avg Assets 30116 92011 3601 10480 Non Int Exp Avg Assets 0151 73629 5087 8324 E Return on Avg Assets ROAA 32517 31553 1163 4396 Return on Avg Equity ROAE 992293 185708 3773 47291 Cost to Income Ratio 2585 934541 72592 57085 Recurring Earning Power 25896 39906 2284 5285 L Net Loans Tot Assets 0466 94423 59299 21614 Net Loans Dep ST Funding 3354 974381 99286 72694 Liquid Assets Dep ST Funding 0852 997421 48584 118879 S Net Income Risk Weighted Assets 30470 17420 0507 2504 Contextual Variables Trend 1000 10000 5500 2873 Private bank 1yes0no 0000 1000 0327 0469 Merger and acquisition 1yes0no 0000 1000 0992 0088 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 26 Legend C1 Total Capital Ratio C2 Equity Tot Assets C3 Equity Net Loans C4 Equity Cust Short Term Funding C5 Equity Liabilities A1 Loan Loss Res Gross Loans A2 Loan Loss Prov Net Int Rev A3 NCO Net Inc Bef Ln Lss Prov A4 Tier 1 Ratio M1 Net Interest Margin M2 Net Int Rev Avg Assets M3 Oth Op Inc Avg Assets M4 Non Int Exp Avg Assets E1 Return On Avg Assets ROAA E2 Return On Avg Equity ROAE E3 Cost To Income Ratio E4 Recurring Earning Power L1 Net Loans Tot Assets L2 Net Loans Dep ST Funding L3 Liquid Assets Dep ST Funding S1 Net Income Risk Weighted Assets Figure 3 Graphic correlogram for the financial ratios 6 Results and Discussion The efficiency levels calculated for 128 selected OECD banks from 2004 to 2013 using the TOPSIS approach and considering different grouping criteria are given in Figs 2 3 and 4 The plot of the kernel density of the TOPSIS efficiency scores not only reflects the fact that these scores are censored at 0 and 1 but also suggests a Gaussian shape around C1 C2 C3 C4 C5 A1 A2 A3 A4 M1 M2 M3 M4 E1 E2 E3 E4 L1 L2 L3 S1 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 27 the mean see Figure 4 left especially when looking at the scores computed from PCA and ICA factor extraction As a matter of fact the KullbackLeibler KL divergence Cobb 2013 Langseth Nielsen PerezBernabe Salmeron 2014 between the inverse cumulative distribution function of the standardized TOPSIS scores obtained from the original set of criteria and the standard Gaussian is 0295 thus suggesting a considerable overlap between both distributions see Figure 4 right A KL divergence value of 0 indicates both distributions are equal Although the median efficiency levels are quite stable over the analyzed period the KruskalWallis test for differences between medians showed no significant results there are substantial differences when efficiency levels are grouped by country see Figures 5 and 6 For instance efficiency tends to be highly dispersed in emerging countries such as Greece Mexico Slovenia and South Korea and higher and less dispersed in developed countries such as the US Germany and the UK A possible explanation for this phenomena may be found in Ebrahim Girma Shah and Williams 2014 who examined the perspective of the Brazil Turkey India and China group Since banking in emerging markets is striving to create premium value for their customers banks tend to have substantial cash to acquire other institutions and invest in other possible growth opportunities Similar results are obtained by Faria and McAdam 2015 who examined growth under regime shift in Arab economies and their macroeconomic stabilization Although this result cannot be taken as a general rule French and Italian banks present high efficiency dispersion distinct from German and British ones the eventual impact of contextual variables may be embedded within these grouping schemes As a matter of fact differences between countries may involve both differences in regulatory standards in finance as well as diverse cultural aspects with respect to business practices ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 28 Figure 4 Kernel densities of the TOPSIS efficiency scores left and the inverse cumulative distributions for the standardized TOPSIS scores and for the standard Gaussian Figure 5 Efficiency levels grouped by year ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 29 Figure 6 Efficiency levels grouped by country ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 30 In regards to the factor extraction Figure 3 reveals that the criteria used to emulate the CAMELS system are strongly correlated Considering that correlations are not taken into account in several efficiency measurement techniques including TOPSIS factor extraction techniques such as PCA and ICA should be alternatively performed on the data as suggested for instance in Madeira Junior et al 2012 Adler and Yazhemsky 2010 and Kao et al 2011 This would allow the extraction of factors that are not only correlated with each other in the case of PCA results but also statistically independent from each other in the case of ICA results Therefore factor extraction was conducted on each of the CAMELS criteria using the PCA with a Varimax standardized rotation for data collected from the 128 OECD banks According to Tabachnick and Fidell 2001 only factor loads greater than 050 deserve to be interpreted and in these cases the variable represents a good factor measure Results presented in Table 4 cover only autovalues greater than 1 With the exception of the S criteria formed by a single variable for the sake of simplicity two main factors were extracted for the other criteria In a similar way factor extraction was conducted with ICA and their results are presented in Table 5 In order to select the most meaningful ICs their statistical independency is evaluated by computing kurtosis values Note that kurtosis is one of the methods for important ICs selection Hyvarinen et al 2001 Kao et al 2011 suggested that ICs with positive kurtosis values higher than three should be selected as the candidate variables ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 31 Table 4 Factor extraction from PCA C Capital Adequacy Variables Factors A Asset Quality Variables Factors L Liquidity Variables Factors PC1 PC2 PC1 PC2 PC1 PC2 Total Capital Ratio 024 068 Loan Loss Res Gross Loans 084 007 Net Loans Tot Assets 086 023 Equity Tot Assets 046 002 Loan Loss Prov Net Int Rev 083 012 Net Loans Dep ST Funding 001 097 Equity Net Loans 001 048 NCO Net Inc Bef Ln Lss Prov 013 071 Liquid Assets Dep ST Funding 084 029 Equity Cust Short Term Funding 019 026 Tier 1 Ratio 009 071 KMO Measure of Sampling Adequacy 0422 Equity Liabilities 054 025 KMO Measure of Sampling Adequacy 049 KMO Measure of Sampling Adequacy 0633 Bartletts Test of Sphericity Approx ChiSquare 377145 Bartletts Test of Sphericity Approx Chi Square 2212387 Bartletts Test of Sphericity Approx Chi Square 16605 df 3 df 6 Sig 0 df 10 Sig 0 Percent of variance explained by the factors 84 Sig 0 Percent of variance explained by the factors 60 Percent of variance explained by the factors 67 M Management Capacity Variables Factors E Earning Quality Variables Factors PC1 PC2 PC1 PC2 Net Interest Margin 099 006 Return on Avg Assets ROAA 079 051 Net Int Rev Avg Assets 099 006 Return on Avg Equity ROAE 008 095 Oth Op Inc Avg Assets 015 097 Cost to Income Ratio 072 009 Non Int Exp Avg Assets 036 091 Recurring Earning Power 086 024 KMO Measure of Sampling Adequacy 0503 KMO Measure of Sampling Adequacy 0505 Bartletts Test of Sphericity Approx Chi Square 76981 Bartletts Test of Sphericity Approx Chi Square 2285119 df 6 df 6 Sig 0 Sig 0 Percent of variance explained by the factors 98 Percent of variance explained by the factors 78 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 32 Table 5 Factor extraction from ICA C Capital Adequacy Variables Factors A Asset Quality Variables Factors L Liquidity Variables Factors IC1 IC2 IC1 IC2 IC1 IC2 Total Capital Ratio 000016 000005 Loan Loss Res Gross Loans 000001 000044 Net Loans Tot Assets 0000652 0001039 Equity Tot Assets 000078 000087 Loan Loss Prov Net Int Rev 000096 001653 Net Loans Dep ST Funding 0001333 0013856 Equity Net Loans 000227 000069 NCO Net Inc Bef Ln Lss Prov 000941 000169 Liquid Assets Dep ST Funding 0008183 0002174 Equity Cust Short Term Funding 000628 000545 Tier 1 Ratio 000001 000000 Kurtosis value 65303 28689 Equity Liabilities 000369 000952 Kurtosis value 16853 88823 Kurtosis value 53227 27609 M Management Capacity Variables Factors E Earning Quality Variables Factors IC1 IC2 IC1 IC2 Net Interest Margin 001021 006958 Return on Avg Assets ROAA 000061 000062 Net Int Rev Avg Assets 000925 006216 Return on Avg Equity ROAE 000609 002054 Oth Op Inc Avg Assets 005963 003564 Cost to Income Ratio 001591 000789 Non Int Exp Avg Assets 004834 001737 Recurring Earning Power 000055 000014 Kurtosis value 64363 34831 Kurtosis value 85933 80476 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 33 The computation of the Spearmans rank correlation between the TOPSIS scores calculated from the original inputs and outputs and those calculated from the factors extracted using PCA and ICA methods revealed significant adherence between each pair of results with a correlation score of 0321 p 00001 for the PCA extraction and 0370 p 00001 for the ICA extraction This indicates that both sets of TOPSIS scores computed from extracted factors are isotonic with respect to the original score set that is their trends are strictly increasing and the orders of distances between points in the embedding matches are low Kruskal 1964 Figure 7 depicts the results for the isotonic regression between both sets of TOPSIS scores in accordance with the fundamentals provided in De Leeuw Hornik and Mair 2009 Figure 7 Results for the isotonic regression Despite the correlation analysis and the isotonic regression between the original sets of TOPSIS scores and those computed from PCA and ICA factor extraction it is still necessary according to Madeira Junior et al 2012 to assess how different weighting criteria impact isotonicity considering equal weights as the base case In this research we use the weighted linear optimization model proposed by Ng 2007 and inspired in Pearman 1977 to generate various sets of weights for the inputs and the output extracted ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 34 factors before conducting the sensitivity analysis on TOPSIS scores with respect to the weighting criteria Under Ngs approach the weights are endogenous that is they are automatically generated when the model is optimized and are therefore not subjective Moreover Ngs approach can be used in a supplementary manner along with several MCDM techniques such as TOPSIS so that weights can be specified exogenously by decisionmakers regardless of some degree of subjectivity However as presented in Ng 2007 the decision maker has to rank the importance of criteria before solving the model We used combinatorial analysis to generate within the ambit of the R statistical package the full universe of combinations where the extracted factors are ranked in different orders of importance The model presented in Ng 2007 was run 9 362880 2 2 times for the full set of the extracted factors for their combinations in different orders of importance both for PCA and ICA extractions In every round the sum of the weights for the extracted factors was equal to one observing the constraints of the linear program Then as a final step this full set of extracted factors and their weight combination 92 725760 was used to recompute the TOPSIS scores Table 6 depicts the results for the mean TOPSIS criteria weights with respect to different groups of results in terms of the Spearmans rank correlations and their respective significances between the original TOPSIS scores and each one of the 725760 weight combination under the PCA and the ICA extractions Four groups were defined for each extraction Considering the PCA extraction in the first group the correlations are found to be positive and significant at 005 This group is not only the most numerous with 4139 of the total 725760 cases but also within this group it is possible to affirm that this set of weights is robustly represented by the original TOPSIS scores with equal weights On the other hand in the second group although correlations are still positive they are not significant With respect to the second largest group 4077 of cases it is possible to affirm that this set of weights is weakly represented by the original TOPSIS scores with equal weights Isotonicity still holds albeit weakly The third and the fourth groups however presented no isotonicity and corresponded to almost 18 of cases On the other hand considering the ICA extraction 100 of results fell within the first group where correlations are significant and isotonicity is high These results suggest that original TOPSIS scores are more robustly represented by those scores computed from the factors ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 35 derived by ICA than those obtained from the factors computed using PCA whenever a different set of weighting criteria is taken into consideration Readers should also observe that the aggregate weighing of factors obtained from each factor extraction differs substantially in response to a given factor extraction method Table 6 Mean TOPSIS criteria weights per group of result PCA TOPSIS Criteria Positive Correlation Negative Correlation Significant Not Significant Not Significant Significant Inputs C1 0090892 0042534 0069463 0119357 C2 0078247 0026512 0059668 0124385 A1 0103201 0257382 0078912 0127316 A2 0077841 0044819 0049765 0076716 M1 0074107 0055108 0036662 0042762 M2 0079608 0043263 005836 0086392 L1 0119384 0335251 0212249 0174819 L2 0076334 0012693 0250664 0065385 S1 0118568 0000621 0002439 0001049 Outputs E1 0051433 0057338 0047173 0035785 E2 0130385 012448 0134645 0146034 Number of Cases 300423 295902 41732 87703 of Cases 4139 4077 575 1208 ICA TOPSIS Criteria Positive Correlation Negative Correlation Significant Not Significant Not Significant Significant Inputs C1 0073393 NA NA NA C2 0061668 NA NA NA A1 0167604 NA NA NA A2 0062635 NA NA NA M1 0060395 NA NA NA M2 0064396 NA NA NA L1 0051711 NA NA NA L2 0130107 NA NA NA S1 02194 NA NA NA Outputs E1 0059089 NA NA NA E2 00496 NA NA NA Number of Cases 725760 0 0 0 of Cases 10000 000 000 000 NA not available ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 36 Results for the CQR of the efficiency scores on the contextual variables for selected efficiency percentiles are presented in Table 7 tau 020 040 060 and 080 Their significances are given in the Appendix The full set of results for the percentile range from 005 to 095 in intervals of 005 is depicted in Figure 8 Significances and bootstrapped lower and upper confidence intervals can also be found in the Appendix The standard errors and confidence limits for the CQR coefficient estimates were obtained with asymptotic and bootstrapping methods Both methods provide robust results Koenker Hallock 2001 though the bootstrap method is preferred for being more practical Hao Naiman 2007 As easily observed in Table 7 the signs and the magnitude of the relationships between the PCA extraction method and the efficiency levels remained quite unchanged for different quantiles These results indicate that efficiency levels computed from PCA factors are systematically higher than those computed from ICA extraction for different efficiency quantiles In conjunction with results presented in Table 6 these results also corroborate the robustness of ICA extraction over PCA extraction within the ambit of efficiency computation since their scores are lower and more discriminatory On the other hand for several countryrelated variables besides trend ownership and MA changes in magnitude and sign reversion were verified In fact this effect occurs because rather than predicting the mean of the dependent variable CQR examines the quantiles of the dependent variable By choosing tau 04 or 06 the 40th and 60th percentiles of the data are being used to compute the regression Therefore CQR can answer the question For which type of OECD bank high or low efficiency does the impact of a given contextual variable prevail Sign reversion may suggest that local cultural and regulatory aspects ie the specifics of each country may be impacting banking operations differently at distinct efficiency levels On the other hand sign reversions in trend ownership and MA suggest that the beneficial effects of privatization mergers and learning curves may or may not be present depending upon the banks efficiency level In other words in light of such exogenous contextual variables higher efficiency banks may be affected differently than lower efficiency ones ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 37 Table 7 Coefficients for the CQR for selected percentiles tau 02 tau 04 tau 06 tau 08 Intercept 4122926e01 04221409424 4175025e01 04244353720 Trend 2102790e03 00023308742 1588746e03 00011177511 Trend2 1336560e04 00001488903 9157415e05 00001904028 Private Bank 2119663e03 00008224629 9959784e04 00090363456 MA 9548725e03 00159760743 6810100e03 00041106571 Austria 7133545e03 00046903075 3914858e03 00021778313 Belgium 7263434e03 00069033255 7021548e03 00113366396 Denmark 1301956e03 00044493096 7326106e03 00083083090 Finland 1912597e06 00068622923 1405544e02 00151699111 France 1508672e03 00014877053 2344498e03 00010760138 Germany 2691382e04 00024007925 2696792e03 00061903361 Greece 1224619e03 00080806710 1306467e02 00402785660 Ireland 4015955e03 00053940041 2822797e02 00417585709 Italy 9326014e03 00045616133 2763637e03 00037587382 Luxembourg 2098646e02 00243525250 2958012e02 00470656408 Mexico 3675285e02 00329281231 2922368e02 00380989958 Netherlands 5443823e03 00010127235 1634528e03 00010748266 Norway 1194644e03 00005866565 2560771e03 00039022218 Poland 1865862e02 00168202125 1074817e02 00087141654 Portugal 4481474e03 00067766857 7775031e03 00289144122 S Korea 6186100e03 00068988726 1974462e02 00285323970 Slovenia 3640998e03 00042373254 5149724e03 00076564991 Spain 1218051e03 00027374663 4024112e03 00021278413 Sweden 1171713e03 00070868866 1003613e02 00081081652 Turkey 1693147e02 00137878866 7654560e03 00101859648 UK 1430675e02 00055746423 3122834e03 00018982412 USA 8695248e03 00118017731 1440764e02 00276463442 Method PCA 2214945e02 00221114694 2200379e02 00215869525 The CQR results presented in Table 7 indicate that the effect of private ownership is negative for lower efficiency banks and positive for higher efficiency banks thus suggesting that public OECD banks should be minimally competitive in terms of efficiency if willing to benefit from a privatization process In other words these banks should first ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 38 become more efficient then privatize and not vice versa Conversely the aggregate effects of trend are negative in higher quantiles of banking efficiency and positive in lower ones thus suggesting that less efficient banks take more advantage of learning curves than do more efficient banks Regarding MA the results presented in the Appendix indicate that this variable was not significant for different efficiency quantiles Figure 8 Censored Quantile Regression Coefficients Plots The solid blue line indicates the quantile regression point estimates the lighter blue region is a pointwise 95 confidence band Xaxis represents the quantiles and Yaxis represents the coefficients for each contextual variable It is important to notice in Figure 8 that CQR coefficients for the country endogenous variables cross the horizontal effect equals zero axis in some quantiles more than once Such crossovers are however sometimes quite plausible and an advantage of the CQR approach is that they are more easily revealed Koenker Geling 2001 In this research crossover occurs for different efficiency quantiles in different countries suggesting that cultural and regulatory aspects related to the country origin may be negative positive for high low efficiency banks and that their efficiency levels may even decrease increase possibly fostering a situation where decreasing increasing returns to scale prevails This appears to be the case of banks located in Germany Spain and Slovenia Greece USA and Ireland For several countries however no significant results were found for different quantile levels eg Portugal Norway Netherlands and France ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 39 7 Conclusions This article presented and developed a robust TOPSIS approach to assess the efficiency of OECD banks in light of different contextual variables related to merger and acquisition ownership publicprivate trend learning curve and country location Three distinctive aspects characterize the robust TOPSIS approach used here i the use of data reduction techniques to handle collinearity issues present in the financial indicators ii a comprehensive criteria weight testing based on linear programming and iii the use of bootstrapped censored quantile regressions to assess how efficiency drivers vary between highlow efficiency banks The methodology used in this research contributed to the efficiency literature in general not only by the enhancing the potential use of multicriteria decision making techniques such as TOPSIS but also by offering an alternative to the two major shortcomings of the state of the art inherent in twostage bootstrapped DEA models The first one is related to the fact that in DEA weights are endogenously defined within the ambit of the linear optimization model thus making it difficult for the analyst to perform comprehensive sensitivity analysis based on the inputoutput here called criteria of financial indicators weights The other one is the shorter computation time involved in computing TOPSIS scores compared to those verified when computing DEA scores especially under intensive sensitivity analysis of weights Results derived in this research also represent a contribution to the literature in banking efficiency They indicate that higher and lower efficiency banks require different courses of action highlighting the importance of market segmentation when analyzing performance Differently from traditional efficiency studies where size is considered the common ground for banking segmentation this research offers a perspective of segmentation based on efficiency quantiles Although the quantile results obtained for the less efficient banks are quite intuitive and corroborate previous literature positive impact of trend and negative impact of private ownership the paths for maintaining high efficiency levels in more efficient banks is not trivial ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 40 Additionally the results presented here suggest a number of policy implications for OECD banks implying different courses of action for banks with higher and lower efficiency levels Lower efficiency banks clearly benefit more from learning curves and are less impacted by privatization than higher efficiency banks although nothing conclusive was found with respect to mergers and acquisitions While less efficient banks should focus on ownership and continuous operational improvements more efficient banks should pay attention to the cultural and regulatory specifics of the country where they operate perhaps choosing efficiencyfriendly OECD countries for expanding their operations In this sense worth mentioning is the substantial endogenous impact of the country effect suggesting that cultural and regulatory issues can be as relevant as other exogenous factors in understanding efficiency Lower dispersion on efficiency results may suggest that banks in developed countries are regulated more tightly than those operating in developing ones Future research should target higher efficiency banks in OECD countries in order to carefully analyze the role of their cultural and regulatory specifics in efficiency levels Possible venues of future studies could also try to overcome some limitations of the current method such as the testing of an alternative set of financial criteria used to emulate the CAMELS rating system and that tends to be limited by the characteristics of the secondary database used Comparative research venues on DEA and TOPSIS scores could also be considered by authors in future studies References Adler N Golany B 2001 Evaluation of deregulated airline networks using data envelopment analysis combined with principle component analysis with an application to Western Europe European Journal of Operational Research 1322 260273 Adler N Golany B 2002 Including principle component weights to improve discrimination in data envelopment analysis Journal of the Operational Research Society 539 985991 Adler N Yazhemsky E 2010 Improving discrimination in data envelopment analysis PCADEA or variable reduction European Journal of Operational Research 202 1 273284 Albayrak E Erensal Y C 2005 A study bank selection decision in Turkey using the extended fuzzy AHP method Proceedings of the 35th International Conference on Computers and Industrial Engineering Istanbul Turkey AlSharkas A A Hassan M K Lawrence S 2008 The impact of mergers and acquisitions on the efficiency of the US banking industry Further evidence Journal of Business Finance Accounting 3512 5070 httpdxdoiorg101111j14685957200702059x Altman E I 1968 Financial ratios discriminant analysis and the prediction of corporate bankruptcy Journal of Finance 234 589609 doi101111j154062611968tb00843x Altman E I Haldeman R G Narayanan P 1977 ZETATM analysis a new model to identify bankruptcy risk of corporations Journal of Bank Finance 11 2954 doi1010160378426677900176 Andrade G Mitchell M Stafford E 2001 New evidence and perspectives on mergers Journal of Economic Perspectives ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 41 152 103120 Assaf AG Barros CP Ibiwoye A 2012 Performance assessment of Nigerian Banks prior and post consolidation Evidence from a Bayesian approach The services Industries Journal 322 215229 doi101080026420692010529135 Assaf AG Barros C P Josiassen A 2010 Hotel efficiency A bootstrapped metafrontier approach International Journal of Hospitality Management 293 468475 doi101016jijhm200910020 Assaf AG Barros C P Matousek R 2011 Technical efficiency in Saudi banks Expert Systems with Applications 385 57815786 doi101016jeswa201010054 Assaf A Barros CP and Matousek R 2011 Productivity and efficiency analysis of shinkin banks Evidence from bootstrap and Bayesian approaches Journal of Banking and Finance 35 331342 Assaf A G Matousek R Tsionas E G 2013 Turkish bank efficiency Bayesian estimation with undesirable outputs Journal of Banking Finance 372 506517 doi101016jjbankfin201209009 Athanasoglou P P Brissimis S N Delis M D 2008 Bankspecific industryspecific and macroeconomic determinants of bank profitability Journal of International Financial Markets Institutions and Money 182 121136 doihttpdxdoiorg101016jintfin200607001 Athanassopoulos A D 1998 Nonparametric frontier models for assessing the market and cost efficiency of largescale bank branch networks Journal of Money Credit and Banking 302 172192 Retrieved from Go to ISI000073439900003 httpwwwjstororgstablepdfplus2601208pdf Azam J P Biais B Dia M 2004 Privatisation versus regulation in developing economies The case of West African Banks Journal of African Economies 133 361394 doi101093jaeejh022 Barros CP Ferreira C Williams J 2007 Analysing the determinants of performance of best and worse European banks A mixed logit approach Journal of Banking and Finance 31 21892203 Barros C P Managi S Matousek R 2012 The technical efficiency of the Japanese banks Nonradial directional performance measurement with undesirable output Omega 401 18 doi101016jomega201102005 Barros C P Peypoch N 2009 An evaluation of European airlines operational performance International Journal of Production Economics 1222 525533 doi101016jijpe200904016 Barros C P Peypoch N Williams J 2010 A note on productivity change in European cooperative banks The Luenberger indicator approach International Review of Applied Economics 242 137147 doi10108002692171003590047 Barros CP and P Wanke 2014 Banking efficiency in Brazil Journal of International Financial Markets Institutions Money 28 5465 Barros CP Wanke P 2015 An analysis of African airlines efficiency with twostage TOPSIS and neural networks Journal of Air Transport Management httpdxdoiorg101016jjairtraman201503002 Bauer P W Berger A N Humphrey D B 1993 Efficiency and productivity growth in US banking In HO Fried CAK Lovell SS Schmidt Eds The measurement of productive efficiency Techniques and applications pp 386413 Oxford Oxford University Press Behzadian M Khanmohammadi Otaghsara S Yazdani M Ignatius J 2012 A stateof theart survey of TOPSIS applications Expert Systems with Applications 3917 1305113069 Bell JA Sejnowski TJ 1995 An InformationMaximization Approach to Blind Separation and Blind Deconvolution Neural Computation 7 11291159 Berger A N Hanweck GA Humphrey D B 1987 Competitive viability in banking Scale scope and product mix economies Journal of Monetary Economics 203 501520 doi1010160304393287900390 Berger A N Humphrey D B 1997 Efficiency of financial institutions international survey and directions for future research European Journal of Operational Research 982 175212 doi101016S0377221796003426 Berger A N Hasan I Zhou M 2009 Bank ownership and efficiency in China What will happen in the worlds largest nation Journal of Banking and Finance 331 113130 doi101016jjbankfin200705016 Betz F Oprică S Peltonen T A Sarlin P 2014 Predicting distress in European banks Journal of Banking Finance 45 225241 doi101016jjbankfin201311041 Bian Y 2012 A GramSchmidt process based approach for improving DEA discrimination in the presence of large dimensionality of data set Expert Systems with Applications 39 3 37933799 BilbaoTerol A Arenas Parra M Cañal Fernandes V 2014 Using TOPSIS for assessing the sustainability of government bond funds Omega 49 117 doi101016jomega201404005 Bose A and S Chatterjee 2003 Generalized bootstrap for estimators of minimizers of convex functions J Stat Planning and Inf 117 225239 Brandouy O Briec W Kerstens K Van de Woestyne I 2010 Portfolio performance gauging in discrete time using a Luenberger productivity indicator Journal of Banking Finance 348 18991910 doi101016jjbankfin200912015 Brans J P Vincke P H 1985 Note A preference ranking organization method Management Science 316 647656 doi101287mnsc316647 Briec W Liang Q L 2011 On some semilattice structures for production technologies European Journal of Operational Research 2153 740749 doi101016jejor201106012 Brissimis S N Delis M D Tsionas E G 2010 Technical and allocative efficiency in European banking European Journal of Operational Research 2041 153163 doi101016jejor200909034 Chen S Leng Y Mao B Liu S 2014 Integrated weightbased multicriteria evaluation on transfer in large transport terminals A case study of the Beijing South Railway Station Transportation Research Part A Policy and Practice 66 1326 Chen TY 2002 A comparison of chanceconstrained DEA and stochastic frontier analysis Bank efficiency in Taiwan Journal of the Operational Research Society 535 492500 doi101057palgravejors2601318 Chen X Skully M Brown K 2005 Banking efficiency in China Application of DEA to pre and postderegulations eras 19932000 China Economic Review 163 229245 doi101016jchieco200502001 Chernozhukov V Hong H 2002 Threestep censored quantile regression and extramarital affairs Journal of the American Statistical Association 97459 872882 Cobb B 2013 Mixture distributions for modelling demand during lead time Journal of the Operational Research Society 64 217 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 42 228 Cole R A Gunther J W 1995a A CAMEL ratings shelf life Federal Reserve Bank of Dallas Review 1320 doihttpdxdoiorg102139ssrn1293504 Cole R A Gunther J W 1995b Separating the likelihood and timing of bank failure Journal of Bank Finance 196 10731089 doi101016037842669598952M Cole R A Gunther J W 1998 Predicting bank failures A comparison of onand offsite monitoring systems Journal of Financial Services Research 132 103117 doihttpdxdoiorg101023A1007954718966 Corrente S Greco S Slowinski R 2013 Multiple criteria hierarchy process with ELECTRE and PROMETHEE Omega 415 820846 doi101016jomega201210009 Cummins J D RubioMisas M 2006 Deregulation consolidation and efficiency Evidence from the Spanish insurance industry Journal of Money Credit and Banking 382 323355 Retrieved from Go to ISI000236422700002 De Borger B Ferrier G D Kerstens K 1998 The choice of a technical efficiency measure on the free disposal hull reference technology A comparison using US banking data European Journal of Operational Research 1053 427446 doi101016S026322419800013X De Leeuw J K Hornik P Mair 2009 Isotone Optimization in R PoolAdjacentViolators Algorithm PAVA and Active Set Methods Journal of Statistical Software 325 124 Deliktaş E Balcilar M 2005 A Comparative Analysis of Productivity Growth CatchUp and Convergence in Transition Economies Emerging Markets Finance Trade 411 628 De Young R 1998 Management quality and Xinefficiency in national banks Journal of Financial Services Research 131 522 doi101023A1007965210067 Dima B Dincă M S Spulbăr C 2014 Financial nexus Efficiency and soundness in banking and capital markets Journal of International Money and Finance 47 100124 doihttpdxdoiorg101016jjimonfin201405002 Dinc S 2005 Politicians and Banks Political Influences on GovernmentOwned Banks in Emerging Countries Journal of Financial Economics Vol 77 No 2 pp 453479 Doumpos M Zopounidis C 2010 A multicriteria decision support system for bank rating Decision Support Systems 501 5563 doihttpdxdoiorg101016jdss201007002 Du K Sim N 2016 Mergers acquisitions and bank efficiency CrossCountry evidence from emerging markets Research in International Business and Finance 36 499510 doi101016jribaf201510005 Dudek A Jefmański B 2015Tthe Fuzzy TOPSIS method and its implementation in the r programme Business Informatics 135 Ebrahim M S Girma S Shah M E Williams J 2014 Rationalizing the value premium in emerging markets Journal of International Financial Markets Institutions and Money 29 5170 doi101016jintfin201311005 Ertugrul I Karakasoglu N 2009 Performance evaluation of Turkish cement firms with fuzzy analytic hierarchy process and TOPSIS methods Expert Systems with Applications 361 702715 doi101016jeswa200710014 Faria J R McAdam P 2015 Macroeconomic adjustment under regime change From social contract to Arab Spring Journal of International Money and Finance 56 122 doihttpdxdoiorg101016jjimonfin201504001 Feng CM Wang RT 2001 Considering the financial ratios on the performance evaluation of highway bus industry Transport Reviews 214 449467 Figueira C Nellis J Parker D 2006 Does Ownership affect the efficiency of African banks Journal of Developing Areas 401 3863 doi101353jda20070004 Frei F X Harker P T 1999 Measuring aggregate process performance using AHP European Journal of Operational Research 1162 436442 Fungáčová Z Poghosyan T 2011 Determinants of bank interest margins in Russia Does bank ownership matter Economic Systems 354 481495 doihttpdxdoiorg101016jecosys201011007 Galindo A and A Micco 2004 Do StateOwned Banks Promote Growth CrossCountry Evidence for Manufacturing Industries Economic Letters Vol 84 pp 371376 Gardener E Molyneux P Hoai N L 2011 Determinants of efficiency in South East Asian banking Service Industries Journal 3116 26932719 doihttpdxdoiorg101080026420692010512659 Garette B Dussauge P 2000 Alliances versus acquisitions Choosing the right option European Management Journal 181 6369 GarzaGarcia J G 2012 Determinants of bank efficiency in Mexico a twostage analysis Applied Economics Letters 1917 1679 1682 doihttpdxdoiorg101080135048512012665589 Gunay E 2012 Risk Incorporation and Efficiency in Emerging Market Banks During the Global Crisis Evidence from Turkey 20022009 Emerging Markets Finance Trade 485 91102 Hair JF Anderson RE Tatham RL Black WC 1995 Multivariate Data Analysis Englewood Cliffs NJ PrenticeHall Hao L Naiman D Q 2007 Quantile Regression Thousand Oaks Sage Publications Hassan M Hawas Y Ahmed K 2013 A multidimensional framework for evaluating the transit service performance Transportation Research Part A Policy and Practice 50 4761 HatamiMarbini A Tavana M 2011 An extension of the Electre 1 method for group decision making under a fuzzy environment Omega 394 37386 doi101016jomega201009001 Hemmati M Dalghandi S A Nazari H 2013 Measuring relative performance of banking industry using a DEA and TOPSIS Measurement Science Letters 32 499503 Hwang C L Yoon K 1981 Multiple attribute decision making methods and applications a state of the art survey Berlin New York SpringerVerlag Hyvarinen A Oja E 2000 Independent Component Analysis Algorithms and Applications Neural Computation 13 411430 Hyvarinen A Karhunen J Oja E 2001 Independent Component Analysis New York NY John Wiley and Sons İÇ Y T Yurdakul M 2010 Development of a quick credibility scoring decision support system using fuzzy TOPSIS Expert Systems with Applications 371 567574 doihttpdxdoiorg101016jeswa200905038 Ikhide S I 2008 Measuring the operational efficiency of commercial banks in Namibia South African Journal of Economics 764 586595 doi101111j18136982200800214x Jaffee D Levonian M 2001 The Structure of Banking Systems in Developed and Transition Economies European Financial ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 43 Management 72 161181 doi1011111468036X00151 Jahanshahloo GR Lotfi FH Davoodi AR 2009 Extension of TOPSIS for decisionmaking problems with interval data Interval efficiency Mathematical and Computer Modelling 4956 11371142 Jenkins L Anderson M 2003 A multivariate statistic approach to reducing the number of variables in data envelopment analysis European Journal of Operational Research 1471 5161 Jin JY Kanagaretnam K Lobo G J 2011 Ability of accounting and audit quality variables to predict bank failure during the financial crisis Journal of Banking Finance 3511 28112819 doi101016jjbankfin201103005 Jin Z Ying Z Wei LJ 2001 A simple resampling method by perturbing the minimand Biometrika 88 381390 Kalbfleisch J D and Prentice R L 2002 Analysis of Correlated Failure Time Data in The Statistical Analysis of Failure Time Data Second Edition John Wiley Sons Inc Hoboken NJ USA doi1010029781118032985ch10 Kao LJ Lu CJ Chiu CC 2011 Efficiency measurement using independent component analysis and data envelopment analysis European Journal of Operational Research 210 2 310317 Kebede H A Wassie B 2013 How efficient are the Ethiopian microfinance institutions in extending financial services to the poor A comparison with the commercial banks Journal of African Economies 221 112135 doi101093jaeejs012 Kerstens K Mounir A Van de Woestyne I 2011 Nonparametric frontier estimates of mutual fund performance using C and Lmoments some specification tests Journal of Banking Finance 355 11901201 doi101016jjbankfin201009030 Kirkpatrick C H Murinde V Tefula M 2008 The measurement and determinants of xinefficiency in commercial banks in SubSaharan Africa European Journal of Finance 147 625639 doi10108013518470701705769 Kiyota H 2009 Efficiency of Commercial Banks in SubSaharan Africa A Comparative Analysis of Domestic and Foreign Banks Proceedings of the CSAE Conference Oxford UK Koenker R Bassett G Jr 1978 Regression quantiles Econometrica 461 3350 Koenker R dOrey V 1987 Computing Regression Quantiles Applied Statistics 36 383393 Koenker R Geling O 2001 Reappraising medfly longevity A quantile regression survival analysis Journal of the American Statistical Association 96 458468 Koenker R Hallock K 2001 Quantile regression An introduction Journal of Economic Perspectives 15 143156 Koenker R 2008 Censored Quantile Regression Redux Journal of Statistical Software 27 125 Kontolaimou A Tsekouras K 2010 Are cooperatives the weakest link in European banking A nonparametric metafrontier approach Journal of Banking Finance 348 19461957 Kruskal J B 1964 Nonmetric Multidimensional Scaling A numerical method Psychometrika 292 115129 doi101007BF02289694 Kumar P R Ravi V 2007 Bankruptcy prediction in banks and firms via statistical and intelligent techniques a review European Journal of Operational Research 1801 128 doi101016jejor200608043 Kumar S Gulati R 2010 Measuring efficiency effectiveness and performance of Indian public sector banks International Journal of Productivity and Performance Management 591 5174 Kumbhakar SC Wang D 2007 Economic reforms efficiency and productivity of Chinese banking Journal of Regulatory Economics 322 105129 Lai Y J Liu T Y Hwang C L 1994 TOPSIS for MCDM European Journal of Operational Research 76 486500 doi1010160377221794902828 Lampe H W Hilgers D 2014 Trajectories of efficiency measurement A bibliometric analysis of DEA and SFA European Journal of Operational Research 2401 121 doi101016jejor201404041 Langseth H Nielsen TD PerezBernabe I Salmeron A 2014 Learning mixtures of truncated basis functions from data International Journal of Approximate Reasoning 55 940956 Lee J Y Kim D 2013 Bank performance and its determinants in Korea Japan and the World Economy 270 8394 doihttpdxdoiorg101016jjapwor201305001 Lee T W 1998 Independent Component Analysis Theory and Application Boston MA Kluwer Academic Publishers Lee CC Hsieh MF 2014 Bank reforms foreign ownership and financial stability Journal of International Money and Finance 400 204224 doihttpdxdoiorg101016jjimonfin201309001 Leng C Tong X 2013 A quantile regression estimator for censored data Bernoulli 191 344361 Levintal O 2013 The real effects of banking shocks Evidence from OECD countries Journal of International Money and Finance 32 556578 doihttpdxdoiorg101016jjimonfin201205028 Li S Liu F Liu S Whitmore G A 2001 Comparative performance of Chinese commercial banks findings and policy implications Review of Quantities Finance and Accounting 162 149170 Madeira Junior AG Cardoso Junior M Belderrain M N Correia AR Schwanz SH 2012 Multicriteria and multivariate analysis for port performance evaluation International Journal of Production Economics 140 1 450456 Matousek R Rughoo A Sarantis N Assaf GA 2014 Bank performance and convergence during the financial crisis Evidence from the old European Union and Eurozone J of Bank Finance doi101016jjbankfin201408012 Maghyereh A I Awartani B 2014 Bank distress prediction Empirical evidence from the Gulf Cooperation Council countries Research in International Business and Finance 30 126147 doihttpdxdoiorg101016jribaf201307001 Mandic K Delibasic B Knezevic S Benkovic S 2014 Analysis of the financial parameters of Serbian banks through the application of the fuzzy AHP and TOPSIS methods Economic Modelling 430 3037 doihttpdxdoiorg101016jeconmod201407036 Männasoo K Mayes D G 2009 Explaining bank distress in Eastern European transition economies Journal of Banking Finance 332 244253 doi101016jjbankfin200807016 Minh N K Long G T Hung N V 2013 Efficiency and SuperEfficiency of Commercial Banks in Vietnam Performances and Determinants AsiaPacific Journal of Operational Research 301 doihttpdxdoiorg101142S0217595912500479 Mohan TR Ray SC 2004 Technical efficiency in public and private sectors in India evidence from the postreform years Working Paper Nº 200322 Economics Working Paper University of Connecticut Department of Economics Müller O Uhde A 2013 Crossborder bank lending Empirical evidence on new determinants from OECD banking markets ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 44 Journal of International Financial Markets Institutions and Money 23 136162 doihttpdxdoiorg101016jintfin201209004 NgWL2007A simple classifier for multiple criteria ABC analysis Eur J Oper Res 177 1 344353 ODonnell G J Westhnizen G V D 2002 Regional comparison of banking performance in South Africa South African Journal of Economics 802 246263 doi101111j181369822002tb01301x Okeahalam C C 2008 Internationalisation and firm performance Evidence from estimates of efficiency in banking in Namibia and Tanzania Journal of International Development 207 942964 doi101002jid1455 Opricovic S 1998 Multicriteria optimization of civil engineering systems Serbia Belgrade Faculty of Civil Engineering Opricovic S Tzeng G H 2007 Extended Vikor method in comparison with outranking methods European Journal of Operational Research 1782 514529 doi101016jejor200601020 Oshinsky R Olin V 2006 Troubled Banks Why dont they all fail FDIC Banking Review 181 2344 Pancurova D Lyocsa S 2013 Determinants of Commercial Banks Efficiency Evidence from 11 CEE Countries Finance a UverCzech Journal of Economics and Finance 632 152179 Retrieved from Go to ISIWOS000319494900003 Paradi JC Yang Z Zhu H 2011 Assessing bank and bank branch performance In Cooper WW Seiford LM Zhu J Eds Handbook on Data Envelopment Analysis pp 315361 Springer New York Pearman AD 1977 A weighted maximin and maximax approach to multiple criteria decision making Operational Research Quarterly 28 3 584587 Poghosyan T Cihák M 2011 Distress in European banks An analysis based on a new dataset Journal of Financial Services Research 403 163184 doi101007s1069301101031 Portnoy S 2013 The Jackknifes Edge Inference for Censored Quantile Regression CSDA forthcoming Powell J L 1984 Least absolute deviations estimation for the censored regression model Journal of Econometrics 25 3 303 325 Powell J L 1986 Censored regression quantiles Journal of Econometrics 32 1 143155 Ramanathan U 2013 Aligning supply chain collaboration using analytical hierarchy Process Omega 412 431440 doi101016jomega201203001 Ravisankar P Ravi V Bose I 2010 Failure prediction of dotcom companies using neural networkgenetic programming hybrids Journal of Information Science 1808 12571267 doi101016jins200912022 Saaty T L 1980 The analytical hierarchy process New York MacGraw Hill Secme N Y Bayrakdaroglu A Kahraman C 2009 Fuzzy performance evaluation in Turkish Banking Sector using Analytic Hierarchy Process and TOPSIS Expert Systems with Applications 369 1169911709 doihttpdxdoiorg101016jeswa200903013 Shaverdi M Akbari M Tafti S F 2011 Combining fuzzy MCDM with BSC approach in performance evaluation of Iranian private banking sector Advances in Fuzzy Systems112 doi1011552011148712 Shi J Liu X Sun Y2006 Melt index prediction by neural networks based on independent component analysis and multi scale analysis Neurocomputing 70 13 280287 Siskos E Askounis D Psarras J 2014 Multicriteria decision making support for global egoverning evaluation Omega 40 5163 doi101016jomega201402001 Staub R B Souza G S Tabak B M 2010 Evolution of bank efficiency in Brazil A DEA approach European Journal of Banking and Finance 2021 204213 doi101016jejor200904025 Sturm JE Williams B 2010 What determines differences in foreign bank efficiency Australian evidence Journal of International Financial Markets Institutions and Money 203 284309 doi101016jintfin201003008 Sufian F 2007 Mergers and acquisitions in the Malaysian banking industry technical and scale efficiency effects International Journal of Financial Services Management 24 304326 Sufian F Kamarudin F 2015 Determinants of revenue efficiency of Islamic banks Empirical evidence from the Southeast Asian countries International Journal of Islamic and Middle Eastern Finance and Management 81 130 doi101108IMEFM1220120114 Tabachnick BG Fidell LS 2001 Using multivariate statistics Boston Allyn and Bacon Tansel İç Y 2012 Development of a credit limit allocation model for banks using an integrated Fuzzy TOPSIS and linear programming Expert Systems with Applications 395 53095316 doihttpdxdoiorg101016jeswa201111005 Tözüm H 2002 Performance evaluation of banks Journal of Banking and Finance 27 19 Tsai BH Chang CH 2010 Predicting financial distress based on the credit cycle index A twostage empirical analysis Emerging Markets Finance and Trade 463 6779 Tsay W H Chou WC Hsu W 2009 The sustainability balanced scorecard as a framework for selecting social responsible investments an effective MCDM model Journal of Operational Research Society 6010 1396410 doi101057jors200891 Wang B Nistor I Murty T Wei YM 2014 Efficiency assessment of hydroelectric power plants in Canada A multi criteria decision approach Energy Economics 46 112121 Wang W K Lu W M Lin Y L 2012 Does corporate governance play an important role in BHC performance Evidence from the US Economic Modelling 293 751760 doihttpdxdoiorg101016jeconmod201201021 Wang W K Lu W M Wang Y H 2013 The relationship between bank performance and intellectual capital in East Asia Quality Quantity 472 10411062 doihttpdxdoiorg101007s1113501195822 Wanke P Barros C 2014 Twostage DEA An application to major Brazilian banks Expert Systems with Applications 415 23372344 doi101016jeswa201309031 Wanke P Barros C Faria J R 2015 Financial distress drivers in Brazilian banks a dynamic slacks approach European Journal of Operational Research 2401 258268 doi101016jejor201406044 Wanke P Barros C Macanda N P J 2015 Predicting Efficiency in Angolan Banks A TwoStage TOPSIS and Neural Networks Approach South African Journal of Economics doi101111saje12103 Worthington A C 1999 Malmquist indices of productivity change in Australian financial services Journal of International Financial Markets Institutions and Money 93 303320 doihttpdxdoiorg101016S104244319900013X Wu CS Lin CT Lee C 2010 Optimal marketing strategy A decisionmaking with ANP and TOPSIS International Journal of ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 45 Production Economics 1271 90196 Yang Z 2009 Assessing the performance of Canadian bank branches using data envelopment analysis Journal of the Operational Research Society 606 771780 doi101057palgravejors2602619 Ying Z Jung SH Wei LJ 1995 Survival analysis with median regression models Journal of the American Statistical Association 90 178184 Yurdakul M Iç Y T 2004 AHP approach in the credit evaluation of the manufacturing firms in Turkey International Journal of Production Economics 883 269289 doi101016S0925527303001890 Zhao H Sinha A P Ge W 2009 Effects of feature construction on classification performance An empirical study in bank failure prediction Expert Systems with Applications 362 26332644 doihttpdxdoiorg101016jeswa200801053 Zheng CH Huang DS Shang L 2006 Feature selection in independent component subspace for microarray data classification Neurocomputing 69 1618 24072410 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 46 Appendix Results for the Censored Quantile Regressions tau 1 02 Coefficients Value Lower Bd Upper Bd Std Error T Value Prt Intercept 041229 038766 043693 001257 3280455 000000 Trend 000210 000012 000409 000101 207473 003801 Trend2 000013 000031 000004 000009 148947 013636 Private Bank 000212 000520 000096 000157 134991 017705 MA 000955 003153 001243 001121 085143 039453 Austria 000713 000321 001747 000527 135236 017626 Belgium 000726 000465 001918 000608 119499 023209 Denmark 000130 001009 001269 000581 022408 082270 Finland 000000 002845 002845 001452 000013 099989 France 000151 001473 001171 000674 022368 082301 Germany 000027 001211 001265 000632 004261 096601 Greece 000122 001159 001403 000654 018737 085137 Ireland 000402 000659 001462 000541 074196 045811 Italy 000933 002108 000243 000600 155470 012002 Luxembourg 002099 000498 003700 000817 256897 001020 Mexico 003675 006825 000526 001607 228709 002219 Netherlands 000544 001803 000714 000642 084769 039661 Norway 000119 001153 000914 000527 022652 082080 Poland 001866 003018 000714 000588 317482 000150 Portugal 000448 000575 001471 000522 085837 039069 S Korea 000619 000772 002009 000710 087176 038334 Slovenia 000364 000964 001692 000677 053743 059097 Spain 000122 001053 001297 000600 020315 083901 Sweden 000117 007572 007807 003923 002987 097617 Turkey 001693 003048 000338 000691 244863 001434 UK 001431 002553 000309 000573 249881 001246 USA 000870 000255 001994 000574 151580 012957 Method PCA 002215 002061 002369 000079 2811852 000000 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 47 tau 1 04 Coefficients Value Lower Bd Upper Bd Std Error T Value Prt Intercept 042214 041528 042901 000350 12053231 000000 Trend 000233 000108 000358 000064 366400 000025 Trend2 000015 000025 000005 000005 299107 000278 Private Bank 000082 000295 000131 000109 075741 044880 MA 001598 002024 001171 000218 733853 000000 Austria 000469 000335 000603 000068 687293 000000 Belgium 000690 002012 003393 001379 050066 061661 Denmark 000445 000057 000833 000198 224556 002473 Finland 000686 000125 001247 000286 239660 001655 France 000149 000171 000468 000163 091301 036124 Germany 000240 001099 000618 000438 054807 058365 Greece 000808 000239 001377 000290 278335 000538 Ireland 000539 000306 000772 000119 453906 000001 Italy 000456 000904 000009 000228 199759 004576 Luxembourg 002435 000166 005036 001327 183506 006650 Mexico 003293 004770 001816 000754 436926 000001 Netherlands 000101 000518 000315 000213 047628 063388 Norway 000059 000161 000279 000112 052290 060105 Poland 001682 002380 000984 000356 472079 000000 Portugal 000678 000104 001460 000399 169863 008939 S Korea 000690 000205 001175 000248 278605 000534 Slovenia 000424 000117 000731 000157 270645 000680 Spain 000274 000001 000547 000139 196399 004953 Sweden 000709 000492 000925 000111 640780 000000 Turkey 001379 001661 001097 000144 957334 000000 UK 000557 000785 000330 000116 480384 000000 USA 001180 000905 001455 000140 841840 000000 Method PCA 002211 002094 002328 000060 3711471 000000 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 48 tau 1 06 Coefficients Value Lower Bd Upper Bd Std Error T Value Prt Intercept 041750 041090 042411 000337 12391415 000000 Trend 000159 000054 000264 000053 297613 000292 Trend2 000009 000017 000001 000004 220184 002768 Private Bank 000100 000095 000294 000099 100360 031557 MA 000681 001171 000191 000250 272311 000647 Austria 000391 000204 000579 000096 408682 000004 Belgium 000702 000461 000943 000123 570609 000000 Denmark 000733 000493 000972 000122 600373 000000 Finland 001406 000719 002092 000350 401079 000006 France 000234 000010 000459 000114 204856 004051 Germany 000270 001166 001706 000733 036805 071284 Greece 001306 000898 001714 000208 627651 000000 Ireland 002823 002593 003052 000117 2411375 000000 Italy 000276 000042 000595 000162 170122 008890 Luxembourg 002958 001654 007570 002353 125712 020871 Mexico 002922 003391 002454 000239 1222347 000000 Netherlands 000163 000077 000404 000123 133060 018332 Norway 000256 000031 000543 000146 174879 008033 Poland 001075 002066 000084 000506 212623 003348 Portugal 000778 001000 002555 000907 085742 039121 S Korea 001974 001119 002830 000437 452185 000001 Slovenia 000515 000267 000763 000127 406747 000005 Spain 000402 000206 000599 000100 402063 000006 Sweden 001004 000824 001183 000091 1097864 000000 Turkey 000765 001063 000468 000152 504298 000000 UK 000312 000558 000066 000126 248600 001292 USA 001441 001202 001679 000122 1184853 000000 Method PCA 002200 002084 002317 000059 3709730 000000 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 49 tau 1 08 Coefficients Value Lower Bd Upper Bd Std Error T Value Prt Intercept 042444 026556 058331 008106 523610 000000 Trend 000112 000382 000159 000138 081011 041787 Trend2 000019 000005 000043 000012 155552 011982 Private Bank 000904 000255 001552 000331 273049 000632 MA 000411 016150 015328 008030 005119 095918 Austria 000218 000307 000743 000268 081288 041629 Belgium 001134 011282 013549 006335 017897 085796 Denmark 000831 000355 001307 000243 342048 000063 Finland 001517 001052 001982 000237 638889 000000 France 000108 000476 000692 000298 036117 071797 Germany 000619 001918 000680 000663 093395 035033 Greece 004028 003005 005051 000522 771950 000000 Ireland 004176 003298 005054 000448 932039 000000 Italy 000376 000257 001009 000323 116333 024469 Luxembourg 004707 004053 013466 004469 105315 029227 Mexico 003810 004652 002968 000430 887045 000000 Netherlands 000107 000870 000655 000389 027643 078222 Norway 000390 000068 000712 000164 237621 001749 Poland 000871 001610 000133 000377 231174 002079 Portugal 002891 002063 003720 000423 684039 000000 S Korea 002853 000986 006692 001959 145661 014522 Slovenia 000766 000167 001364 000305 250860 001212 Spain 000213 000187 000612 000204 104431 029634 Sweden 000811 000490 002112 000664 122125 022199 Turkey 001019 002422 000385 000716 142276 015481 UK 000190 001019 000640 000423 044857 065374 USA 002765 001634 003896 000577 479056 000000 Method PCA 002159 001905 002412 000129 1667546 000000 OMEGA Int J of Mgmt Sci Vol 17 No 3 pp 237250 1989 03050483j89 300 000 Printed in Great Britain All rights reserved Copyright 1989 Pergamon Press plc An Application Procedure for DEA B GOLANY Y ROLL Technionlsrael Institute of Technology Israel Received October 1987 in revised form November 1988 Data Envelopment Analysis DEA has become an accepted approach for assessing efficiency in a wide range of eases The present paper suggests a systematic application procedure of the DEA methodology in its various stages Attention is focused on the selection of decision making units DMUs to enter the analysis as well as the choice and screening of factors The application of several DEA models in different versions and formulations is demonstrated in the process of determining relative eMciencies within the compared DMUs I INTRODUCTION I EFFICIENCY OR PRODUCTIVITY analyses are vital managerial control tools for assessing the degree to which inputs are utilized in the process of obtaining desired outputs The abundant available literature on productivity measure ment tackles this problem from various points of view applying different approaches The economic approach assumes the existence of specific mostly constant over time input output relations which can be identified by the analysis of a large body of data eg national accounts Productivity due to this approach is evaluated against production functions which formulate the assumed relations 27 An engin eering approach is one where productivity is measured by comparing performance to suit ably set engineering standards 29 Other ap proaches assess productivity by means of ratio analysis or by variations on accepted accounting techniques None of these approaches is satis factory for measuring productivity in the service and public sectors or in nonprofit organiz ations This statement holds also for private sector companies if some of the pertinent factors are not readily expressed in economic terms The main reasons for the failure of traditional productivity measurement approaches in such cases are the following Most traditional approaches for pro ductivity assessment are based on process measures with little or no attention to important outcome measures eg 31 2 Such outcome measures as well as some input factors are typically quali tative in nature In addition to the problems involved in quantifying such factors it is usually extremely difficult to assign them proper relative weights 3 It is very difficult to formulate an ex plicit functional relationship between inputs and outputs with fixed weights on the various factors 4 Averaging performance across many DMUs as in statistical regressions fails to explain the behavior of individ ual DMUs The Data Envelopment Analysis DEA approach 11 15 tries to overcome these diffi culties It provides a means for assessing relative efficiencies of Decision Making Units DMUs with minimal prior assumptions on input output relations in these units Such relative efficiencies can be evaluated among a group of 237 238 Golany RollmAn Application Procedure for DEA a single period or in a sequence of periods Alternatively efficiencies can be evaluated with respect to known performance by competitors or measured against predetermined goals or standards Objectives for carrying out DEA vary widely including typically the following Identification of the sources and amounts of relative inefficiency in each of the compared units in any of its inputoutput dimensions eg 8 Ranking of the units by their efficiency outcomes Evaluation of management heading the compared units Evaluating the effectiveness of pro grams or policies which are outside the control of the units and differentiating between program inefficiency and managerial inefficiency eg 16 Creating a quantitative basis for re allocating resources among the units under evaluation Reallocation poli cies may take into account the slacks found in some units as well as the virtual multipliers obtained for the various inputs at different units see Appendix The general purpose of such reallocation policies is to shift limited resources to units where they would be utilized more effectively in generating desired outputs eg 7 Identification of efficient units or efficient inputoutput relations for pur poses not directly related to compari son among the units An example for such an objective is the use of DEA outcomes to determine test markets for the trial of new products Other objectives are cited in 36 Analysis and scrutiny of prevailing standards on specific inputoutput relations against actual performance eg 32 w Comparison and contrasting of results from previous studies eg 3 9 Since first publications on DEA in 1978 the literature has expanded rapidly reporting further conceptual and theoretical develop ments as well as numerous applications 14 However in most cases DEA applications are reported as demonstrations of or additions to the theory Thus they tend to emphasize a particular point rather than give an overall view of the application procedure Furthermore some steps in the procedure are not dealt with in a methodical manner Typical examples of such steps are the choice of factors to enter the analysis and the quantification of qualitative factors 34 Also little is published on the comparison between various versions of the DEA approach and on the significance of differences in outcomes resulting from employ ing different models The purpose of this paper is to bring together some of the developments in DEA and organize the experience gained into a systematic applica tion procedure Because of the extremely wide and heterogeneous areas for which DEA may be appropriate the proposed procedure can only serve as a general guideline Where applicable some of the different options are presented and discussed An attempt is made towards minimal repetition of material published elsewhere for which frequent reference is made However the paper is designed to be to a degree selfcon tained so that the proposed procedure can be readily followed by users with little previous experience of DEA To this end the most common DEA models are presented in the Appendix The three main phases in carrying out an efficiency study by means of DEA are the following i Definition and selection of DMUs to enter the analysis ii Determination of input and output factors which are relevant and suitable for assessing the relative efficiency of the selected DMUs iii Application of the DEA models and analysis of outcomes Each of these phases comprises several steps The entire proposed procedure is presented in a flowchart form in Fig 1 The different phases are discussed in some detail in the next three sections Then a short illustrative example is worked out followed by some concluding remarks Omega Vol 17 No 3 239 2 SELECTION OF DMUs DEA is a technique for assessing the relative efficiency of comparable units with a view to improving their performance This implies a basic assumption that differences in perform ance among like units exist and are measurable 23 25 Even under quite similar conditions one always finds differences in the way units are managed if only because they are led by dif ferent decision makers Thus on the one hand we look for a homogeneous set of units where comparison makes sense and on the other we try to identify the differences between them These contradicting considerations accompany every step of a DEA application They are however most prominent during the stages of choosing the DMUs to be compared and identi fying the factors affecting them A homogeneous group of units for our purposes is one where the units under consideration perform the same tasks with similar objectives all the units perform under the same set of market conditions this is of special importance in the analysis of nonprofit organizations such as schools army units state hospitals courts etc the factors both inputs and outputs characterizing the performance of all units in the group are identical except for differences in intensity or magni tude The next step is to determine the size of the comparison group Again two conflicting con siderations present themselves On the one hand there is a tendency to increase the number of units because the larger the population the larger the probability of capturing high per formance units which would determine the efficiency frontier Also a large set of units enables a sharper identification of typical relations between inputs and outputs in the set Furthermore as the number of units increases it is possible to incorporate more factors into the analysis A rule of thumb established here see 1 8 is that the number of units should be at least twice the number of inputs and outputs considered On the other hand the larger the number of units in the analyzed set the lower the homogeneity within the set generally increasing the possibility that results may be affected by some exogeneous factors which are not of interest The determination of DMUs to enter the DEA evalualtion process is affected by two kinds of boundaries One comprises the organiz ational physical or regional boundaries which define the individual units The other relates to the time periods used in measuring the DMUs activities Preferably the time periods to be considered should be natural ones corre sponding to seasonal cycles and budgeting or auditing periods Regarding the length of such periods it should be borne in mind that long periods may obscure important changes occur ing within them while short periods may give an incomplete picture of the DMUs activities Another consideration is how far back to go without distorting the comparison 26 and whether to create an overlap of data by means of window analysis 10 A final step in deter mining the set of units is sifting out DMUs which may be considered as outliers ie units or time periods deviating from the general characterization of the group to be analyzed This should be carried out with utmost care and if needed with the advice and intervention of managers 13 One should bear in mind that efficiency is measured with respect to the DMUs and factors selected There is no guarantee that the initial selection is correct in the sense that it serves best the purposes of the analysis Thus the considerations mentioned in this section may require the application of parts of the proposed procedure in an iterative fashion as indicated in Fig 1 No practical stopping rule is suggested for these iterations and they depend entirely on the specific application Furthermore it may be useful in many applications to present more than one set of results derived from different selections of DMUs factors or models 3 SELECTION OF FACTORS The initial list of factors to be considered for assessing DMU performance should be as wide as possible Every dimension the changes in which may affect the DMUs to be evaluated should be included in the initial list Such factors could be either fully or partially controllable by 240 Golany RollmAn Application Procedure for DEA I J VL I Set Factors Measurement Scales I Ralationshil Formalize Inial Model Computer Environment I Analyze by Factor I Define Population of DMUs I Set Goals for Analysis I Select DMUs to be Compared m LIS Relevant Factors I I Examine Factors I Judgement I j Examine Factors II Correlations Examine Factors III Trial Runs Formalize Final Model I Presenl Initial Results 1 i Special Analyses I I I Analyze by Individual DMU Fig DEA application flow chart the DMUs or they may be environmental factors outside the control of the DMUs Some of the factors would be quantitative ie readily available while other factors may be qualita tive in nature with different degrees of difficulty to be accorded numerical values The factors could be inputs or outputs or factors placed on either side of a production relationship All the factors which may possibly have any bearing on the performance of the DMUs to be analyzed should be listed at this stage without any numerical treatment Clearly this may result in quite a large number of factors in the initial list for example Thomas 35 started his study with an initial list of 92 factors which was eventually brought down to only 14 see also 16 22 The introduction of a large number of factors into the analysis will result in explaining away a larger portion of the differences among DMUs This will tend to shift the compared units towards the efficiency frontier resulting in a relatively large number of units with high efficiency scores In the following procedure it is proposed to introduce initially only a limited number of carefully selected factors thus accen tuating the basic differences among units At a later stage while analyzing the outcomes additional factors can be brought into the model to examine whether they explain some of the differences Thus the next steps are directed towards the reduction of the typically lengthy initial list to one that includes only the most relevant factors The factors chosen should distinguish clearly among the compared units and serve effectively the objectives of the analysis This refinement of the list can be carried out in three stages i judgemental screening ii nonDEA quantitative analysis iiiDEA based analysis 1 Judgemental process The first stage in reducing the list of factors is its critical examination by expert decision makers in the field where the DMUs operate Omega Vol 17 No 3 241 Since a large list of relevant factors is usually compiled some may be repeating virtually the same information some may not be regarded as crucial while others may appear to be con flicting or confusing A problem often encountered at this stage is the proper distinction between factors determin ing efficiency and factors explaining efficiency gaps For example labor input may serve to determine efficiency while the scale at which the unit operates may be an explaining factor Entering such explaining factors into the initial analysis may blur the overall picture and reduce the distinction between compared units It may also obscure our understanding of the way in which some factors affect performance and generally diminish the usefulness of efficiency analyses as tools for improving performance In public sector units or in nonprofit organiz ations where factors governing performance are not always welldefined special care should be exercised to distinguish between inputs and explaining factors Judgement may be exercised inter alia along the following lines Is the factor related to or contributing to one or more of the objectives set for the application Is the factor conveying pertinent infor mation not included in other factors Does the factor contain elements eg price which interfere with the notion of technical efficiency 23 Are data on the factor readily available and generally reliable Possible systematic procedures to be applied at this stage for structuring the judgemental process are Delphilike techniques 35 or vari ations of the Analytic Hierarchy Process 30 Ad hoc methods using expert knowledge are mentioned in 9 where a panel of experts was used to formalize the model and later check its results and in 33 where an official committee was formed to scrutinize the factors 2 NonDEA Quantitative Methods The first step here is to assign numerical values to the various factors For many of these factors the natural choice would be the physical units by which they are measured Such measures can be in dollar terms for economic factors in number of persons KWH of gener ated electricity gallons of fuel etc A non trivial consideration in this regard is whether to aggregate all or some factors which can be measured in economic terms into one dollar factor in other words assign fixed relations among the weights of these factors see discus sion in section 4 Whether or not to do such aggregations depends strongly on the objectives of the analysis Another issue concerning quan titative factors is the handling of cases where zero values are encountered for some factors in some DMUs or some time periods This may happen if the periods chosen do not correspond to natural cycles of operation or for reasons stemming from the datagathering procedure In such cases periods could be redefined or data accumulated across several periods In principle the DEA models can handle cases with zero values for some of the factors as long as there exists at least one input and one output for each DMU which is nonzero see Appendix However such cases should be handled with care as the computational algorithms may be sensitive to zero values 19 A final remark concerns the isotonicity relations which are assumed for DEA ie an increase in any input should not result in a decrease in any output Consequently the values of some factors may have to be inverted before they are entered into the analysis see for example 10 Another group of factors is the qualitative ones Indeed the inclusion of such factors is one of the novelties of DEA Nevertheless they have to be asigned numerical values in order to participate in the mathematical evaluation of efficiency The usual practice here is to locate some measurable surrogate variable which is assumed to bear a known relation to varying levels of the qualitative factor Typically several possible surrogates may be tried out for each qualitative factor until a suitable one is located Criteria for the choice of surrogate factors are the degree of correspondence between vari ations in the surrogate data and the examined factor the ability to express this correspondence in a functional form and the general compliance of the results to the analysis objectives An example of the selection process is described in 32 242 Golany RollAn Application Procedure for DEA The next step within this stage is to describe the production relations governing the DMUs to be analysed and classify the factors into inputs and outputs Resources utilized by the units or conditions affecting their operation are typical inputs while measurable benefits gener ated constitute the outputs In most cases this distinction is straightforward However some factors may be interpreted in both ways their classification depending on the analysts point of view eg 10 A useful procedure here may be to carry out a series of regression analyses of such factors one at a time on the factors known to be inputs and outputs A weak relation to inputs and strong relation to outputs indicates a preference towards classifying the factor as an input while a reversed outcome will point towards viewing the factor as an output A weak relation to all the factors may indicate a need to reexamine the factor and possibly delete it Alternatively strong relations may indicate that the information contained in that factor is already represented by other factors and again its deletion should be considered Similar regression analyses should be carried out on factors clearly defined as inputs and outputs with a view to eliminating redun dancies and reducing the list of factors eg 26 However as observed in 28 one should not regard these oneatatime regression tests as reliable rules but merely as indicators for a need to examine some of the factors more closely see 9 for an overview of contrasts between DEA and regression analyses It may be useful at this stage to use each factor separately to rank order all the DMUs These rankings can then be aggregated into a single ranking which best describes them all eg a ranking as in 20 Factors with rankings which differ sharply from the others are again candi dates for possible deletion Throughout this stage attempts can be made to refine or replace the surrogates selected earlier for the qualitative factors 3 DEAbased analyses The last step in the process of examining and refining the list of factors consists of trial runs of DEA models Within the family of the DEA models the one initially proposed by Charnes Cooper and Rhodes CCR 15 is the model which points out differences among DMUs in the most critical way Some of the subsequent models notably the one introduced by Banker Charnes and Cooper 2 incorporate some of the explanations to efficiency differences into the models themselves In the procedure proposed here analysis is started with the most strict model namely CCR Other DEA models are introduced in the subsequent stage when out comes are analyzed in an attempt to explain the resulting efficiency differences Factors which have remained in the list so far are now entered into the model and outcomes are examined closely Factors which are consist ently associated with very small multipliers ie have little impact on the efficiency scores may be dropped Basically we seek the ability to discriminate between the DMUs using the selected factors Hence factors which do not contribute to this end are candidates for elimination To test the discriminating power of the different factors the model is run with a series of combinations of these factors Then various grouping techniques can be applied to the DMUs using the resulting efficiency scores Factors which do not alter such group ings significantly should be examined closely Special attention should be given to factors which could not be easily classified as inputs or outputs They can be tried at both sides of the efficiency ratio and classified finally according to the results Some factors typically the nondiscretionary ones can be used to rescale all other factors in the analysis Thus non discretionary factors are accounted for in an indirect manner and the total number of factors reduced for a different approach see 4 Following a series of such iterations the list of factors to enter the analysis is decided upon Outcomes of the final iteration provide at the same time the first and basic set of efficiency scores for the compared DMUs The following section describes the process of further analysis and examination of these results 4 PRESENTATION AND ANALYSIS OF RESULTS Further analysis of the relative efficiency values obtained from the basic CCR model can be performed inter alia by applying different models of DEA and introducing addi tional information into the respective models There are two possibilities of formulating the CCR model One puts the emphasis on input Omega Vol 17 No 3 243 reduction and the other on output enhance ment Both formulations yield identical results which is not the case with other DEA models such as the BCC The choice between the two formulations is made according to prevailing circumstances In some applications the inputs are rather inflexible eg determined to a certain extent by higher managerial levels in which case the output formulation would be more suitable In other applications outputs are matched closely with goals set by management or restricted by environmental conditions Then the input formulation is more appro priate The models can be presented in either of two versions the primal version highlighting the virtual multipliers of inputs and outputs or the dual version emphasizing the relations between the examined DMU and the other members in the analyzed set A brief summary of some of the DEA models is presented in the Appendix For the demonstration of possible efficiency analyses a small illustrative example is worked out Suppose that the screening process of the relevant factors for a group of ten DMUs resulted in locating two outputs and three inputs as the most meaningful for assessing the relative efficiencies within the group The first ten lines in Table 1 list the numerical values of the five factors for each of the ten DMUs Table 2 presents the outcomes from applying the CCR model to this set of data The outcomes contain four components a An efficiency score for the specific DMU relative to a subgroup of DMUs those in the reference facet to which the examined DMU is best compared b The slacks output shortfalls and input surpluses associated with the examined Table I Illustrative example data DMU No Outputs Y2 Inputs XI X 2 X 3 I 090 70 I0 08 540 2 100 95 15 10 480 3 080 75 12 21 510 4 090 90 I0 06 420 5 070 80 18 05 600 6 100 50 7 09 520 7 080 70 I0 03 500 8 075 75 12 15 550 9 065 55 14 18 570 10 085 90 8 09 450 I I 095 I00 6 03 450 12 100 100 7 03 420 13 100 95 8 03 400 DMU in addition to the increase of all outputs or the decrease in all inputs by a factor equal to the efficiency score For example observing the outcomes for DMU all inputs could be de creased by a factor of 0847 and in addition input 3 decreased by 92 units in order to bring the efficiency of DMU to that of DMUs 4 6 and 7 c The DMUs in the reference facet for the examined DMU d The virtual multipliers for the examined DMU These are the factors weights determined by the model which when applied to the said DMU will render its best relative efficiency score by the fol lowing expression Oo uyo 3 see the Appendix for the notation A similar set of results can be obtained by applying other DEA models Three examples are given i The BCC model 5 and 6 in the Appendix This model takes into account the effect of returns to scale Table 2 Summary of outcomes across DMUs the CCR model Slacks DMU Outputs Inputs DMUs No Elciency I 2 I 2 3 in facet Virtual multipliers Outputs Inputs 1 2 I 2 3 1 0847 92 4 6 7 2 0972 50 35 031 4 3 0734 40 10 4 6 4 10 4 5 0829 017 44 4 7 6 10 6 7 0 7 8 0660 041 4 6 I0 9 0532 I0 029 052 4 I0 10 I0 0695 00032 0064 044 00000 0971 0 0 0 00021 0917 0 0020 0 00015 II I 0 0024 0 00018 0 00104 0 093 00109 0897 00021 0043 0 00013 1249 0 0045 061 00007 0715 00016 0034 0 00011 0818 0 0 0 00018 0 00111 0026 0 00018 244 Golany RollAn Application Procedure for DEA Table 3 Summary of outcomes across models DMU Model Efficiency Slacks Outputs Inputs DMUs I 2 1 2 3 in facet Vimtal multipliers Outputs Inputs I 2 I 2 3 CCR inputs CCR outputs BCC inputs BCC outputs Multipl 8 Additive 9 0847 92 4 6 7 0847 92 4 6 7 0937 025 676 4 6 2 0870 48 4 6 7 I0 0064 073 651 2 4 6 20 02 120 4 0695 00032 0064 0445 0 0695 00032 0064 0445 0 1063 00010 0 0275 0 0050 0 0044 0211 0001 0 0 0 0 0 OOI I 0 0 0 0 within the analyzed group of DMUs attempting to point out the most pro ductive scale size for each DMU while simultaneously identifying its technical inefficiency 2 Note however that the BCC model yields different results with the input and output formulations ii The unit invariant multiplicative model 7 in the Appendix This model links the DEA approach to some accepted CobbDouglaslike production func tions where production relationships are characterized by the product of factors raised to appropriate powers rather than the weighted sum of factors as in the common engineering approaches 18 iiiThe additive model 8 in the Appendix This model constitutes a Paretooptimal empirical production Table4 Effciencyrankingsacross models DMU CCR BCC CCSS CCGSS I 6 7 6 7 2 5 I5 I5 I5 3 8 8 9 6 4 14 I5 I5 I5 5 7 6 7 9 6 14 I5 I5 I5 7 I I5 I5 I5 8 9 9 8 8 9 10 10 10 10 10 14 I5 I5 15 function for the given group of DMUs 12 As mentioned above not all DEA models should be attempted for every application Rather some models are suitable only for specific cases or purposes of the analysis see the Appendix Here however only one example is given and the different models are applied to the same set of data The outcomes of these models for DMU are presented in Table 3 No efficiency score is given for the additive and multiplicative models since in these models the objective values represent sums of slacks rather than efficiency ratios see the Appendix In order to compare the efficiency evaluation of all the DMUs across the different models in spite of the difficulty men tioned earlier a summary of results in terms of optimal rankings is given in Table 4 In general most DMUs were ranked similarly by the different models which is an indication of the robustness of the technique However different rankings due to differences among the models should be expected as evident for DMUs whose ranking according to the CCGSS model is worse than its rankings by the other models A sum mary of DMUs appearing in the facets with the different models is given in Table 5 The last column in this table sums up the number of times each DMU was selected by the models run in this example to form a reference set for Table 5 DMUs appearing in facets with the different models DMU No CCR DMUs in reference facet BCCin BCCout Multipl Additive No of times this DMU is in a facet I 467 2 46 46 7 10 246 4 0 2 4 2 2 2 2 12 3 46 246 4 10 26 4 0 4 4 4 4 4 4 26 5 47 47 47 467 47 0 6 6 6 6 6 6 17 7 7 7 7 7 7 12 8 4 6 I0 2 I0 4 I0 2 6 4 0 9 4 2 6 4 2 6 4 0 I0 I0 lO 10 I0 I0 I0 Omega Vol 17 No 3 Table 6 Summary of outcomes 245 Efficiency scores CCR CCR BCCout CCR within constrained CCR within with CCR BCCout DMU original output aggregated original standard within within No set weights inputs set set category category j I 0847 0712 0844 0870 0681 10 10 2 0972 0924 0972 10 0833 0972 10 3 0734 0689 0732 0826 0627 0741 0833 4 10 10 10 10 0900 10 10 5 0829 0829 0622 0900 0560 10 10 6 10 0642 10 10 0906 10 10 7 10 10 10 10 0800 10 10 8 0660 0651 0636 0781 0573 0721 0818 9 0532 0451 0532 0737 0456 0600 0789 10 10 10 0941 10 0840 10 10 DMUs I5 were assumed in category A and DMUs 610 in category B other DMUs The above are only examples of the different DEA models There are several more such models each with its specific charac terisations for example the multiplicative model 17 and the weighted additive model Another direction in the analysis of efficiency outcomes is partitioning the group of DMUs into categories according to some characteristic which was not entered into the model as a factor determining inputoutput relationships The purpose of such categorization is twofold one is to gain a better relative assessment of efficiency by comparing performance within subgroups of units operating under similar conditions eg the same geographical region see 5 6 for related developments The other is a compari son between categories such as in the case where a category signifies a program under which a subgroup of DMUs operates 16 Table 6 shows the results of such a partitioning into two categories Further insight into the relative efficiency scores within the examined group of DMUs can be gained by imposing additional constraints on the standard models Such constraints can be added if additional information is available or specific hypotheses are to be tested about relationships among the factors Following are several examples ii iii i A known relationship between the weights of inputs This may be the ease when some of the inputs have market values and it is assumed that the virtual multipliers should relate to each other as the respective prices The additional constraint here will take the form of 9 in the Appendix In the illustrative example a situation is tested where the price of one unit of input 2 is ten times as high as that of one unit of input 1 The outcome given in Table 6 is identical to the one obtained without the additional constraint but with in puts l and 2 replaced by a single input reflecting their combined cost An assumed directional relationship between outputs 24 Such a condition can be imposed for example when there is a clear preference of one quali tative output over another The corre sponding constraint will take the form 10 in the Appendix Table 6 illustrates the case where there is a preference of output 2 over output 1 in the example This was effected by setting the virtual multiplier of output 2 at least as high as that of output 1 Bounding the range within which factor weights may vary 21 The argument here is that when we let the weights vary freely some DMUs may be as sessed by only a subset of the inputs and outputs which are considered rele vant Moreover a completely free choice of weights can be viewed in some cases as covering up the most serious deficiencies low outputs andor high inputs of the DMU being analyzed The difficulty here is in locating such suitable bounds One suggestion is to set the bounds on factor weights by observing the weight variation when the unbounded model is run Another option is to impose bounds on ratios between weights reflecting the 246 Golany RollAn Application Procedure for DEA importance attached by the analyst to the various input and output factors It should be noted that adding constraints will result only in downward changes in the efficiency scores On the other hand reducing the number of compared DMUs will result in upward changes in efficiency scores An additional perspective of the efficiency picture can be gained by adding to the observed data some external reference data Such refer ence data can be obtained from comparable DMUs outside the analyzed group or through setting standards of performance Adding ex ternal reference data serves to overcome the limitation of having only relative efficiencies within the analyzed group In the illustrative example three theoretical sets of data were added lines 1113 in Table 1 and the resulting efficiencies evaluated Table 6 summarizes the efficiency scores for the ten DMUs in the example when applying several models and model modifications It may be seen that substantially different efficiencies are obtained under each set of different con ditions The appropriateness of the different approaches and the significance of the respective changes in the efficiency scores should be judged in light of the specific circumstances of each application Efficiency outcomes can be pre sented graphically in the form of a bar chart as shown in Fig 2 for DMU8 in the example The total efficiency gap between the performance of DMU8 and the set of standards is partitioned into portions explained in different ways and an unexplained portion Again partitioning of the efficiency gap and the explanations offered depend strongly on the particular circumstances and the purpose of the analysis Also it should be noted that in most cases not all the evalu ations presented in Table 6 are simultaneously relevant Hence Fig 2 demonstrates a situation where only the first and the last four columns in Table 6 are of interest Similar barcharts can be constructed for the other evaluations when necessary An important aspect of any DEA implemen tation is the use of appropriate computer codes for handling the heavy computation needed in the analysis Since this technical issue is outside the main scope of this paper we find it sufficient to propose a general scheme for a DEA decision support system as in Fig 3 5 SUMMARY AND CONCLUSIONS DEA is an effective approach to assess the efficiency of organizations in cases where tra ditional approaches to efficiency analysis or pro ductivity measurement either fail or are difficult or impossible to apply Such are the cases typical within the public sector or the non Total Effic lency Gap GAp Total Within Compa Cetegor rlson Gap 10 818 ½ IIIIllNHHIIIIII Fig 2 Pardfioning the eiciency llun accounted for portion portlon explained by assumed diseconomies of scale within category Ioportlon explained by difference betwlsn best overall performance between standard and actual best performance gap DMU No 8 Omega Vol 17 No 3 I Input File Repeated LP IIi J Data Base Handler I Input File t Stasoa J Solver regr essions etc Tasks input I nput Fie l l Generator I Data Base handling package carble of storing retrieving errorchecking etc 2 Repeated Linear Programming mehanizm 3 Statistical Package 4 Graphical package 5 Report generator Fig 3 DEA decision support system 247 profit organizations where no strict functional relationship can be formed a priori between factors of production and where the relative weights of these factors inputs and outputs in the production processes are not welldefined However tackling such vaguely defined situ ations has its problems Prominent among these is the fact that DEA yields relative efficiencies only within the examined group of DMUs Outcomes are also dependent on the factors entered into the analysis and the numerical values accorded to qualitative factors Con sequently attention should be focused on the differences between efficiency scores rather than their absolute values It is also important to remember that the typical system under exam ination is dynamic in nature undergoing changes continuously Therefore special atten tion should be given to the identification of changes or trends in efficiencies over time Finally sensitivity of outcomes to the various assumptions made eg the mix of pertinent factors and the version of the DEA model employed should be carefully tested The example given in the previous section does not exhaust the possibilities for special analyses based on the DEA outcomes Other aspects may include statistical evaluation of the efficiency scores multiphase DEA sensitivity analysis techniques etc A possible approach for obtaining a general efficiency picture within the analyzed group is to divide it into several broad efficiency categories Members of the highest category would be the DMUs most often present in the reference facets and relatively insensitive to the different assumptions The lowest category would con tain the DMUs with consistently low efficiency scores with as many intermediate categories as befit the circumstances A valuable measure of change could be the shift of particular DMUs from one such category to another In recent years a growing number of researchers have criticized the field of OR for solving only limited and unrealistic problems while ignoring wicked problems which are common in real life situations The DEA approach is perhaps a step in the right direction OME 7 D 248 Golany RollAn Application Procedure for DEA It attempts to address tough problems and offers a way of tackling them The procedure proposed in this paper is meant to assist in extending the use of this method to areas which have not been explored so far as well as to strengthen existing applications APPENDIX DEA Models z DEA measures efficiency by estimating an em pirical production function which represents the highest values of outputs that could be gener ated by relevant inputs as obtained from ob served inputoutput vectors for the analyzed DMUs The inefficiency of a DMU is then measured by the distance from the point repre senting its input and output values to the corre sponding reference point on the production function There are a number of mathematical formulations of DEA all sharing the principle of envelopment An output vector Yk for DMUk a specific unit under evaluation is enveloped from above when the model identifies a combi nation of other output vectors for the same input vector X whose values are equal to or greater than all the elements in Yk Similarly the input vector Xk is enveloped from below when the model finds a combination of other input vectors whose values are smaller than or equal to all the elements in Xk If the pair Xk Yk cannot be enveloped simultaneously by a com bination of other DMUs other than DMUk itself then DMUk is efficient In general the set of efficient DMUs selected for evaluating an analyzed DMU defines one facet of the piece wise linear empirical production function A linear combination of these DMUs serves as a reference point for the measurement of the inefficiency of DMUk The data which DEA uses are only the observed Yj vectors for all the DMUs in the analysis By solving a series of linear programming LP optimizations one for each DMU DEA is able to identify those DMUs that are efficient on the facets and the This Appendix is not a comprehensive survey of the DEA methodology and is intended for readers who have already had some experience with DEA It highlights important differences in the meaning of the models and their suitability for different scenarios Readers are referred to 11 for detailed presentations of the methodology remaining inefficient DMUs along with their efficient reference points DEA was first developed as an extension to the classical engineering ratio notion of efficiency It determines for each DMU the maximal ratio of the sum of its weighted outputs divided by the sum of its weighted inputs where the weights are determined by the model This fractional programming problem is transformed into an LP model 1 below 11 Applying the duality theory of LP a dual program 2 can be derived for the transformed program The con straints of the dual lead to the envelopment interpretation given above Common notation used in the sequel is sum marized below Indices Data Variables jDMUs j n routputs r I t inputs i ffi 1 m yjthe value of the rth output of the jth DMU xgthe value of the ith input for the jth DMU a small positive number s aslacks corresponding to input i out put r respectively t0 2jweight of DMUj in the facet for the evaluated DMU 0 vvirtual multipliers for output r input i respectively hrelative efficiency of DMUk a The CCR model 15 For each DMU solve Max hk Yk st vxk 1 i r i The objective here is to find the largest sum of weighted outputs of DMU while keeping the sum of its weighted inputs at unit value and forcing the ratio of the sum of weighted outputs to the sum of weighted inputs for any DMU to be less than one This ratio corresponds to the classical Engineering ratio definition of efficiency 15 The dual program solves for each DMU St E y2j o yk J Y xj Okxk s 0 2 1 Omega Vol The objective function of this model attempts to find a minimal value for an intensity factor 0k which indicates the potential of a proportional reduction in all the inputs of DMUk In addition the objective function seeks the largest slack values in all inputoutput dimensions In other words it finds the reference point on the empirical production function which portrays DMUk in the worst efficiency characterization The constraints of the model represent the envelopment principle explained earlier We started in 1 with an emphasis on output enhancement holding the inputs at a fixed unit value Similar formulations are written for the case where input reduction is emphasized Min h k vxk i st ta y 1 r tatj Y zo t o 3 i and Max St jii fl 0 J i 4 b The BCC model 121 This model adds another restriction to the envelopment requirements It requires that the reference point on the production function for DMUk will be a convex combination of the observed efficient DMUs The dual formulation for DMUk is written in this case as St Z ij ff t yr i Y xoOkxsO E 2j 1 5 The corresponding primal has a slightly different objective from I Max h taYk uk st viz l i ta yj v Zo uk 0 6 t i the term uk was interpreted by BCC as an indicator of returns to scale 17 No 3 249 C The CCSS model 181 In formulations 2 4 and 6 the measure of relative efficiency is derived from sums of weighted inputs and outputs Alternatively this measure can be generated by taking the ratio of a product of outputs raised to virtual weights divided by the product of inputs raised to the virtual weights The envelopment formulation which results from this definition was shown in 18 to be associated with the CobbDouglas type production functions The model solves for each DMU Min hkas r i st logyoA j a logyk J IogxA j s logxk 1 E 7 J d The CCGSS model 121 Here DEA can be used to test for Pareto Optimality of the analyzed DMUs In 12 it was shown that DMUk is Paretooptimal if and only if the objective value in the next model is zero Min hk o si i st Y yjjy J Xo j S Xk J Z J t 8 e Modified DEA models Example I Model 1 with additional con straints such as c2vt c t v 2 0 9 These constraints specify known relations between virtual multipliers Adding such a con straint is equivalent to combining inputs 1 and 2 into one aggregate input eg as when con sidering economic inputs and testing the effect of replacing the physical quantities x and x2 with their combined cost clx c2x Example 2 Model 1 with additional con straints such as ta t ta 0 10 A constraint of this kind imposes an order of importance on qualitative variables 250 Golany RollAn Application Procedure for DEA REFERENCES 1 Aii I Charnes A Cooper WW Divine D Klopp GA and Stutz J 1988 An application of data envelopment analysis to US Army recruitment districts Applics Mgmt Sci Res A Edited by Schultz RL JAI Press 2 Banker RD Charnes A and Cooper WW 1984 Models for estimating technical and scale efficiencies in data envelopment analysis Mgmt Sci 309 10781092 3Banker RD Conrad RF and Strauss RP 1986 A comparative application of DEA and translog methods An illustrative study of hospital production Mgmt Sci 32i 3044 4 Banker RD and Maindiratta A 1986 Piecewise log linear estimation of efficient production surfaces Mgmt Sci 321 126135 5 Banker RD and Marcy RC 1986 Data envelopment analysis with categorical inputs and outputs Mgmt Sci 3212 16131627 6 Banker RD and More RC 0986 Efficiency analysis for exogenously fixed inputs and outputs Ops Res 344 5035 2 7 lkssent A Bessent W Charnes A Cooper WW and Thorogood R 1983 Program review for planning decisions through data envelopment analysis Educ Adm Q 192 5775 8 Bowlin WF 1987 Evaluating the efficiency of US AirForce real property maintenance activities J Opl Res Sac 382 127135 9 Bowlin WF Charnes A Cooper WW and Sherman HD 1985 Data envelopment analysis and regression approaches to efficiency estimation and evaluation Ann Ops Res 2 113138 10 Charnes A Clark T Cooper WW and Golany B 1985 A developmental study of data envelopment analysis in measuring the efficiency of maintenance units in the US Air Force Ann Ops Res 2 95112 1 I Charnes A and Cooper WW 1985 Preface to topics in data envelopment analysis Ann Ops Res 2 5994 12 Charnes A Cooper WW Golany B Sciford L and Stutz J 1985 Foundations of data envelopment analy sis for ParetoKoopmans efficient empirical production functions J Econometrics 30 91107 13 Charnes A Cooper WW Learner DB and Phillips FY 1985 Management science and marketing manage ment J Marketing 493 14 Charnes A Cooper WW Lewin AY Marcy RC and Rousseau JJ 1985 Sensitivity and stability analysis in DEA Ann Ops Res 2 139156 15 Charnes A Cooper WW and Rhodes E 1978 Measur ing efficiency of decision making units Fur J Ops Res 26 429444 16 Charnes A Cooper WW and Rhodes E 0981 Data envelopment analysis as an approach for evalualting program and managerial efficiency with an illustrative application to the program follow through experiment in US public school education Mgmt Sci 276 668697 17 Charnes A Cooper WW Seiford L and Stutz J 1982 A multiplicative model for efficiency analysis Socio Econ Plan Sci 165 222224 18 Charnes A Cooper WW Seiford L and Stutz J 1983 Invarlant multiplicative efficiency and piecewise CobbDouglas envelopments Ops Res Let 23 101103 19 Charnes A Cooper WW and Thrall RM 1986 Classi fying and characterizing efficiencies and inefficiencies in data envelopment analysis Ops Res Let 3 105110 20 Cook W and Seiford L 1978 Priority ranking and consensus formation Mgmt Sci 2416 17211732 21 Dyson RG and Thanassoulis E 1988 Reducing weight flexing in data envelopment analysis Y Opl Res Sac 396 563576 22 Fire R and Loveli CAK 1978 Measuring the technical efficiency of production J Econ Theory 191 150162 23 Farrell MJ 1957 The measurement of productive efficiency J R Statist SocA 1203 25328 24 Golany B 1988 A note on including ordinal relations among multipliers in data envelopment analysis Mgmt Sci 348 10291033 25 Leibenstein H 1966 Allocative efficiency versus X efficiency Am Econ Rev 563 392415 26 Lewin AY Marcy RC and Cook TJ 1982 Evaluating the administrative efficiency of courts Omega I04 401411 27 Nelson RR 1981 Research on productivity growth and productivity differences Dead ends and new departures J Econ Lit 193 102964 28 Rhodes E 1978 Data envelopment analysis and related approaches for measuring the efficiency of decision making units with an application to program following through in US education PhD dissertation Carnegy Melon University 29 Roll Y and Sachish A 1981 Productivity measurement at the plant level Omega 91 3742 30 Saaty TL 1980 The Analytic Hierarchy Process Planning Priority Setting Resource Allocation McGrawHill New York 31 Scott WR 1979 Measuring outputs in hospitals Measuring and Interpretation of Productivity National Academy of Science Washington DC 32 Scrrussi D 1987 Applying DEA to the IDF Air Force maintenance units MSc thesis Technionlsrael Institute of Technology 33 Sherman HD 1981 Measurement of hospital technical efficiencyA comparative evaluation of data envelop ment analysis and other approaches for locating inefficiencies in health care organizations DBA thesis Harvard University Graduate School of Business Administration 34 Smith P and Mayston D 1987 Measuring efficiency in the public sector Omega 153 181189 35 Thomas DL 1985 Auditing the efficiency of regulated companies through the use of data envelopment analy sis An application to electric cooperatives PhD thesis University of Texas Graduate School of Business Austin 36 Thompson RG Singleton FD Thrall RM and Smith BA 1986 Comparative site evaluations for locating a highenergy physics lab in Texas Interfaces 166 3549 ADDRESS FOR CORRESPONDENCE Dr B Goany Faculty of Industrial Engineering and Management Technion Israel Institute of Technology Hails 32 070 Israel Contents lists available at ScienceDirect European Journal of Operational Research journal homepage wwwelseviercomlocateejor Invited Review Assessing bank efficiency and performance with operational research and artificial intelligence techniques A survey Meryem Duygun Fethi a Fotios Pasiouras b a School of Management University of Leicester UK b School of Management University of Bath UK ARTICLE INFO ABSTRACT Article history Received 25 February 2009 Accepted 5 August 2009 Available online 11 August 2009 This paper presents a comprehensive review of 196 studies which employ operational research OR and artificial intelligence AI techniques in the assessment of bank performance Several key issues in the literature are highlighted The paper also points to a number of directions for future research We first discuss numerous applications of data envelopment analysis which is the most widely applied OR technique in the field Then we discuss applications of other techniques such as neural networks support vector machines and multicriteria decision aid that have also been used in recent years in bank failure prediction studies and the assessment of bank creditworthiness and underperformance 2009 Elsevier BV All rights reserved Keywords Artificial intelligence Banks Data envelopment analysis Efficiency Operational research Literature review 1 Introduction Banks play a central role in the economy They keep the savings of the public and finance the development of business and trade Furthermore numerous studies argue that the efficiency of financial intermediation affects economic growth while others indicate that bank insolvencies can result in systemic crises which have adverse consequences for the economy as a whole Thus the performance of banks has been an issue of major interest for various stakeholders such as regulators customers investors and the general public While bank performance has been traditionally evaluated on the basis of financial ratios advances in operational research OR and artificial intelligence AI have resulted in a shift towards such quantitative techniques Of course this is not surprising since OR has been extensively used in other applications in finance during the last half century Board et al 2003 This paper presents a comprehensive review of the use of OR and AI techniques in the assessment of bank efficiency and performance The rest of the paper is structured as follows Section 2 positions the survey within the existing literature and discusses our framework Section 3 discusses applications of data envelopment analysis DEA in the estimation of bank efficiency and productivity growth Section 4 presents applications of other OR and AI techniques in the prediction of bank failure and the assessment of bank creditworthiness and underperformance Section 5 summarizes our conclusions Corresponding author Tel 44 0 116 252 5328 fax 44 0 116 252 5515 Email addresses mfethileacuk MD Fethi fpasiourasbathacuk F Pasiouras 03772217 see front matter 2009 Elsevier BV All rights reserved doi101016jejor200908003 2007 discusses more recent applications of frontier techniques but his survey focuses only on studies that provide international comparisons of bank efficiency We differentiate our review from the above surveys by discuss ing applications of OR and AI techniques over the period 1998 early 2009 while focusing on bank performance We searched for papers in Scopus which is considered to be one of the largest ab stract and citation databases We consider only journal articles and we do not include working papers monographs dissertations or other publication outcomes Furthermore our search is limited to articles written in English We use a combination of various key words such as bank efficiency bank and data envelopment anal ysis bank performance bank and neural networks bank and artificial intelligence bank and operational or operations re search Additional studies were identified from crossreferencing and were manually collected We reviewed a total of 196 studies DEA is by far the most commonly used ORAI technique in assessing bank performance and we identified 151 studies that use DEAlike techniques to estimate various measures of bank efficiency and productivity growth and 30 studies that provide similar estimates at the branch level2 We also identified 15 studies that use classification techniques such as neural networks support vector machines mul ticriteria decision aid decision trees nearest neighbours to predict bank failure or assess bank creditworthiness and bank underper formance These studies were published in a total of 73 journals however around 58 of them appeared in just 12 journals The most frequent sources of publication are the European Journal of Operational Research 19 and the Journal of Banking and Finance 15 followed by Applied Financial Economics 13 Managerial Fi nance 11 Applied Economics 9 Expert Systems with Applications 9 the Journal of Productivity Analysis 9 and the Journal of Eco nomics and Business 83 3 DEA and bank efficiency DEA is a mathematical programming technique for the develop ment of production frontiers and the measurement of efficiency relative to these frontiers Each bank is assigned an efficiency score between 0 and 1 with higher scores indicating a more efficient bank relatively to other banks in the sample One of the wellknown advantages of DEA is that it works rela tively well with small samples Other advantages of DEA are that it does not require any assumptions to be made about the distribu tion of inefficiency and it does not require a particular functional form on the data in determining the most efficient banks However DEA is also subject to few limitations Two of the bestknown shortcomings are that DEA assumes data to be free of measure ment error and that it is sensitive to outliers Coelli et al 2005 also point out that i having few observations and many inputs andor outputs will result in many firms appearing on the DEA frontier ii treating inputsoutputs as homogenous commodities when they are heterogeneous may bias the results iii not accounting for differences in the environment may give misleading results iv standard DEA does not control for multiperiod optimi sation or risk managerial decision making Our survey shows that recent DEA studies have examined al most all of the banking sectors around the world A few recent studies provide crosscountry evidence Most of them examine banks from the large EU banking sectors Casu and Molyneux 2003 Beccalli et al 2006 LozanoVivas et al 2002 examine 10 EU countries Bergendahl 1998 focuses on Nordic countries while Pasiouras 2008a and Tanna 2009 examine international datasets 31 Methodological issues 311 Efficiency measures Most of the studies focus on the technical efficiency of banks eg LozanoVivas et al 2002 Drake et al 2006 Pasiouras 2008ab This efficiency measure indicates whether a bank uses the minimum quantity of inputs to produce a given quantity of outputs or maximizes the output quantity given a certain quantity of inputs However when price data for the inputs andor outputs are available one can also estimate cost andor profit efficiency mea sures4 Cost efficiency is the product of technical efficiency and allocative efficiency The latter refers to the ability of a bank to use the optimum mix of inputs given their respective prices Con sequently cost efficiency shows the ability of a bank to provide services without wasting resources as a result of technical or allo cative inefficiency As can be seen in Table 2 of the Supplementary material Appendix A around 35 studies present measures of DEA cost efficiency eg TortosaAusina 2002a Isik and Hassan 2002 2003a Pastor and Serrano 2006 propose the decomposition of cost inefficiency into composition inefficiency and intraspecialisation inefficiency The first component indicates the part of inefficiency due to the composition of specialisations of the banks in each banking sector The second component reveals the inefficient use of resources within each of the specialisation selected Prior 2003 also deviates from the above studies by calculating mea sures of short and longrun cost inefficiency as well as capacity inefficiency for Spanish banks The first refers to the case that a subset of inputs are fixed and impossible to modify in the short run Longrun inefficiency estimates are obtained under the assumption that inputs are variable and under the control of the company Finally capacity inefficiency obtained by the ratio of longrun to shortrun inefficiency refers to excess in costs as a re sult of inappropriate level in fixed inputs Similar concepts along with an application in the Indian banking sector are discussed in Sahoo and Tone 2009 Estimations of profit efficiency with DEA are rather limited in the literature One potential reason is the difficulty in collecting reliable and transparent information for output prices Further more the decomposition of profit efficiency into technical and allocative efficiency is not straightforward Coelli et al 2005 Fare et al 2004 propose the solution of two sets of linear programmes In the first a profit maximizing DEA is solved to measure profit efficiency In the second DEA problem technical efficiency is mea sured on the basis of a directional distance function that allows the simultaneous adjustment of inputs and outputs Kirkwood and Nahm 2006 also estimate profit efficiency although they use in put prices only Therefore in a sense they calculate a measure of efficiency that is similar to Berger and Mester 1997 alternative profit efficiency which is commonly used in the stochastic frontier analysis literature The studies of Maudos and Pastor 2003 and 2 Our survey focuses on studies that examine banking institutions as a whole however we also discuss studies on branch efficiency in Section 327 as one could argue that the efficiency of individual branches can influence the performance of banks as a whole 3 In the sections that follow we discuss various issues surrounding these studies while additional information is available in the Supplementary material Appendix A 4 One can also estimate revenue efficiency which is similar to profit efficiency In both cases both inputs and output prices are required The difference is that in the former measure the aim is to maximize revenues rather than profits ie revenues minus costs We are no aware of DEA studies focusing on revenue efficiency so we do not discuss this issue further Readers interesting in revenue efficiency could see Coelli et al 2005 for further details 190 MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 Ariff and Can 2008 provide estimates of both standard and alter native profit efficiency Finally around 40 studies obtain estimates of total factor pro ductivity TFP growth eg Casu et al 2004 Tanna 20095 This measure is usually decomposed further into technological change ie shift in the best practice frontier and technical efficiency change Furthermore in most cases technical efficiency change is disaggregated into pure technical efficiency change and scale effi ciency change under the variable returns to scale assumption dis cussed below 312 Constant vs variable returns to scale DEA can be implemented by assuming either constant returns to scale CRS or variable returns to scale VRS In their seminal study Charnes et al 1978 proposed a model that had an input orientation and assumed CRS This model returns a score that indi cates the overall technical efficiency OTE of each bank Banker et al 1984 suggested the use of variable returns to scale VRS that decomposes OTE into product of two components pure tech nical efficiency PTE and scale efficiency SE The former relates to the ability of managers to utilize firms given resources while the latter refers to exploiting scale economies by operating at a point where the production frontier exhibits CRS In most of the recent papers DEA models are estimated using the assumption of VRS while arguing that CRS is only appropriate when all firms are operating at an optimal scale6 Nevertheless other studies argue in favour of CRS rather than VRS or mention that caution is necessary when using the VRS formulation eg Noulas 1997 Avkiran 1999 Soteriou and Zenios 1999a Consequently many studies report the results obtained under both CRS and VRS assumptions eg Canhoto and Dermine 2003 Casu and Molyneux 2003 313 Outputinput orientation Technical efficiency can be estimated under either an inputori ented or outputoriented approach7 As Coelli et al 2005 point out the inputoriented technical efficiency measures address the question By how much can input quantities be proportionally reduced without changing the output quantities produced p 137 In con trast the outputoriented measures of technical efficiency address the question By how much can output quantities be proportionally expanded without altering the input quantities used p 137 By far studies in banking obtain efficiency estimates under the inputoriented approach8 This is most likely due to the assumption that bank managers have higher control over inputs eg personnel expenses rather than outputs eg loans income etc However there are also some studies that adopt the outputoriented approach eg Ataullah et al 2004 Ataullah and Le 2006 or report the results from both eg Casu and Molyneux 2003 Beccalli et al 20069 Fi nally others adopt a nonoriented slackbased measure SBM that assumes the simultaneous contraction of inputs and expansion of outputs eg Avkiran 2009ab 314 Selection of inputs and outputs There is an ongoing discussion in the banking literature regard ing the proper definition of inputs and outputs Using the words of Bergendahl 1998 There have been almost as many assumptions of inputs and outputs as there have been applications of DEA p 235 Berger and Humphrey 1997 identify two main approaches for the selection of inputs and outputs These are the production ap proach and the intermediation approach The first assumes that banks produce loans and deposits account services using labour and capital as inputs and that the number and type of transactions or documents processed measure outputs The second approach perceives banks as financial intermediaries between savers and investors Berger and Humphrey 1997 argue that neither of these two approaches is perfect because they cannot fully capture the dual role of financial institutions as providers of transactionsdoc ument processing services and also being financial intermediaries They point out that the production approach may be somewhat better for evaluating the efficiencies of bank branches and the intermediation approach may be more appropriate for evaluating financial institutions as a whole Furthermore there are difficulties in collecting the detailed transaction flow information required in the production approach As a result the intermediation approach is the one favoured in the literature However there is a controversy even within this approach con cerning the role of deposits Berger and Humphrey 1997 Conse quently some studies use only earning assets as outputs a selection that is in line with the asset approach of Sealey and Lind ley 1977 while others consider deposits as an additional output a selection that is more closely related to the socalled valueadded approach We find around 95 applications in bank efficiency where the monetary value of deposits is part of the input vector and 20 applications where deposits are part of the output vector10 Around 30 studies use interest expenses as an input without using the stock of deposits eg Weill 2004 Avkiran 2009ab In another 7 applica tions the stock of deposits is used as an output and the interest ex pense paid on deposits constitutes an input eg Maudos et al 2002 Chen et al 2005 Furthermore 7 studies use time deposits and sav ing deposits as input and demand deposits as output eg Bauer et al 1998 Gilbert and Wilson 1998 Finally in a few applications the deposits are included as both an input and an output eg TortosaAusina 2002a More recently some studies adopted another variation of the intermediation approach This is the socalled profitoriented or operating approach which defines revenue components eg interest income noninterest income etc as outputs and cost components eg personnel expenses interest expenses etc as in puts11 Drake et al 2006 mention that from the perspective of an 5 Almost all studies use the DEAlike Malmquist index Zen and Baldan 2008 use the Luenberger Indicator that is a generalization of the Malmquist Index 6 Reasons that may not allow a firm to operate at optimal scale include among others imperfect competition government regulations constrains on finance etc Coelli et al 2005 7 If price data are available then under the under cost minimization objective ie cost efficiency one obtains inputoriented measures of technical efficiency and input mix allocate efficiency Revenue maximization ie revenue efficiency results is outputoriented technical efficiency and outputmix allocative efficiency measures In the case of profit maximization profit efficiency technical efficiency can be obtained under either the input or outputoriented assumption Coelli et al 2005 8 Additional information is available in the Supplementary material Appendix A In contrast to efficiency estimates productivity measures are in several cases obtained using an outputoriented Malmquist index However as discussed in Coelli et al 2005 p 80 although the values of the components and consequently their contribution to overall productivity change may differ The overall TFP change measure will be the same regardless of the orientation ie input or output that is imposed 9 It should be mentioned that the inputoriented and outputoriented measures always provide the same value under CRS but they are unequal when VRS is assumed However Coelli et al 2005 mention that since linear programming does not suffer from statistical problems the choice of an appropriate orientation is not as important as in the case of econometric approaches Furthermore in many instances the choice of orientation has only a minor influence upon the scores obtained Coelli and Perelman 1996 10 This discussion refers to banks as a whole and not branches The number of applications does not match the number of studies as in several cases there are numerous models developed in each study Furthermore in some cases deposits or noninterest expenses are not considered in the analysis In one case Prior 2003 the output was the number of current and saving accounts rather than the monetary value of deposits ie stock 11 See Chu and Lim 1998 Avkiran 1999 Sturm and Williams 2004 Das and Ghosh 2006 Drake et al 2006 Ataullah and Le 2006 and Pasiouras 2008ab MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 191 inputoriented DEA relative efficiency analysis the more efficient units will be better at minimizing the various costs incurred in generating the various revenue streams and consequently better at maximizing profits p 1451 They also argue that this approach can be more appropriate in capturing the diversity of strategic responses by financial firms in the face of dynamic changes in competitive and environmental conditions Furthermore Luo 2003 calculates a measure of marketability efficiency in an attempt to capture the va lue of the bank in the stock market In this case revenue and profits are considered inputs whereas market value earnings per share and stock price are outputs As discussed before with the exception of deposits there is a general agreement about the main categories of inputs and out puts however this does not necessarily imply that there is consis tency with respect to the specific inputsoutputs used in various studies For instance the traditional inputs are fixed assets person nel12 and in many cases they are deposits eg Isik and Hassan 2002 Maudos and Pastor 2003 Casu and Girardone 2004 Havryl chyk 2006 However some studies use branches eg Chen 2001 loan loss provisions eg Drake et al 2006 Pasiouras 2008b and equity eg Chu and Lim 1998 Mukherjee et al 2001 Sturm and Williams 2004 Pasiouras 2008a as additional or alternative inputs Chen 2001 disaggregates deposits into current deposits and time deposits while Das and Ghosh 2006 use demand savings and fixed deposits Casu and Girardone 2006 and Beccalli et al 2006 use to tal costs as a single input while Casu and Molyneux 2003 use two inputs namely total costs and total deposits ie customers and shortterm funding Several studies use two outputs usually loans and other earn ing assets eg Casu and Molyneux 2003 Casu and Girardone 2004 200613 However some studies disaggregate loans into var ious categories such as housing loans and other loans eg Sturm and Williams 2004 real estate loans commercial loans and per sonal loans eg Mukherjee et al 2001 Fare et al 2004 or shortterm and longterm loans Isik and Hassan 2002 Others disaggregate other earning assets into investments and liquid as sets Tsionas et al 2003 or investment in government securities and investments in public and private enterprises eg Chen 2001 Canhoto and Dermine 2003 use the number of branches as an additional output under the assumption that it represents an additional value for retail customers Finally recent studies in clude noninterest income or offbalancesheet items as additional outputs eg Isik and Hassan 2002 2003a Sturm and Williams 2004 TortosaAusina 2003 Havrylchyk 2006 Pasiouras 2008b14 Finally Halkos and Salamouris 2004 propose an approach that deviates from the above literature They use an output vector that consists of five financial ratios and no inputs Their underlying hypothesis is that inputs are considered similar and equal for all banks as they operate in the same markets for money and service The comparison with standard DEA models developed under either CRS or VRS assumptions shows that the DEA ratio model has the highest discrimination power 315 Adjusting for the environment Coelli et al 2005 discuss four approaches that can be used to incorporate environmental variables in DEA applications15 The first method by Banker and Morey 1986 requires the environmen tal variables to be ordered from the least to the most harmful ones for efficiency Then the efficiency of a given firm is compared with those firms in the sample that have a value of the environmental var iable which is less than or equal to the given firm This ensures that banks are not compared with peers operating in a more favourable environment The second method by Charnes et al 1981 requires the decisionmaker to i divide the sample into subsamples and solve DEA problems for each subsample ii project all observed data points into their prospective frontiers and iii solve a single DEA using the projected points and assess any difference in the mean efficiency of the two subsamples According to Coelli et al 2005 the following two problems are common in both methods i sam ple splitting reduces the comparison set and ii only one environ mental variable can be considered in each case thereby limiting the scope of the analysis Under the third method the environmental variables are in cluded directly into the DEA problem as nondiscretionary inputs if it is believed to have a positive effect on efficiency or out puts if they have a negative effect on efficiency The disadvan tage of this approach is that one must know a priori the direction of the influence a shortcoming that is also applicable in the case of the first method Alternatively the environmental variables can be included as nondiscretionary neutral variables using an equality form The shortcoming of this approach is that it can reduce the reference set for each firm Recent applications of the above approaches in banking can be found in Pastor 1999 and LozanoVivas et al 2001 2002 The fourth method that is discussed in Coelli et al 2005 is the twostage approach This involves a DEA problem with traditional inputs and outputs in the first stage In the second stage the effi ciency scores obtained are regressed on the environmental vari ables While this approach has been frequently used in the banking literature with numerous applications it fails to adjust the efficiency measures Therefore it is more suitable when the objective is to examine the correlations of efficiency with various factors see Section 321 rather than provide the basis for absolute comparisons across different environments Finally Pastor 2002 Drake et al 2006 Avkiran 2009b and Thoraneenitiyan and Avkiran 2009 adjust the bank efficiency scores for risk andor external environmental factors using a mul tistage DEA In this case the application starts with the estimation of a DEA model with traditional inputs and outputs Then using the slacks from the DEA model they quantify the effect of the operat ing environment and adjust the initial dataset inputs andor out puts Finally they rerun the initial DEA model using the adjusted data 32 Topics of interest 321 Determinants of efficiency Several studies attempt to investigate the factors that influence the efficiency of banks Some studies examine only bankspecific factors and others examine both bankspecific attributes and envi ronmental determinants Commonly found bankspecific factors are size profitability capitalisation loans to assets Casu and Molyneux 2003 Casu and Girardone 2004 Ataullah and Le 2006 Ariff and Can 2008 12 Some studies use the number of personnel eg Maudos and Pastor 2003 Pasiouras 2008b However others rely on personnel expenses due to data unavailability eg Bergendahl 1998 LozanoVivas et al 2002 Drake et al 2006 13 Other earning assets normally include various items such as government securities investment securities trading securities other securities equity invest ments and other investments 14 Isik and Hassan 2003a mention among others that ignoring such nontraditional outputs may penalize banks that are heavily involved in these activities The reason is that while the resources used to produce these nontraditional outputs are part of the input vector the outputs generated using these inputs are not included in the output vector 15 According to Coelli et al 2005 environmental variables can include ownership differences location characteristics labour union power and government regulations and in a sense any factors that can influence the efficiency of the firm without being traditional inputs or under the control of managers 192 MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 Isik and Hassan 2003a examine additional characteristics such as the educational profile of bank personnel and CEOChairman affiliation Some studies examine whether age is related to efficiency eg Isik and Hassan 2003a Canhoto and Dermine 2003 Isik 2008 Furthermore Isik and Hassan 2002 Casu and Girardone 2004 and Pasiouras 2008b examine the international presence of Turkish Italian and Greek banks respectively Countryspecific factors include market concentration presence of foreign banks ratio of private investments to GDP fiscal deficits to GDP GDP growth Hauner 2005 Ataullah and Le 2006 Pasiouras 2008a examined the relationship between technical efficiency and regulations related to capital adequacy private monitoring banks activities deposit insurance schemes supervisory power and bank entry into the industry More recently Tanna 2009 investigated the impact of foreign direct investments on TFP growth These studies use the twostage approach discussed above First they use DEA to obtain efficiency estimates Then in a second stage the DEA scores are regressed on a number of explanatory variables using Tobit eg Hauner 2005 OLS Ataullah and Le 2006 GMM Ataullah and Le 2006 or GLS Isik and Hassan 2003a regression The rationale to use Tobit lies on the fact that the efficiency scores are bounded between 0 and 1 and hence noncensored estimates will be biased Ataullah and Le 2006 among others mention that it is not necessary to use Tobit as long as the efficiency scores can be transformed by taking the natural logarithm of efficiency score1 efficiency score Whether or not censored regression is used the most important argument comes from Simar and Wilson 2007 who point out that the covariates in the secondstep regression are obviously correlated with the one side error terms from the firststep as otherwise there would be no need for the secondstep regression Furthermore the covariates in the secondstep are likely to be highly correlated with the covariates in the firststep This means that the errors and the covariates in the firststep cannot be independent Thus Simar and Wilson conclude that the likelihood that is maximized is not the correct one unless one takes account of the correlation structure Casu and Molyneux 2003 present an early attempt to account for the problems that can emerge in the twostage DEA method using a bootstrap approach However Simar and Wilson 2007 mention that the simple bootstrap is not enough to overcome the drawbacks pointed out above They propose an algorithm that uses a double bootstrap procedure and they present an application to US banks Brissimis et al 2008 and Delis and Papanikolaou 2009 adopt this approach to examine the determinants of efficiency in the new EU banking sectors 322 Stock returns and efficiency The relationship between stock returns and publicly available information has attracted considerable attention in the accounting and finance literature Whilst most of the studies examine whether earnings reflect some of the information in stock prices recent research has however shifted towards the use of additional data such as accruals revenues economic value added and efficiency to understand how they affect stock prices and returns DEA applications that were published over the period of our survey and investigate the relationship between bank efficiency and stock returns examine Australia Kirkwood and Nahm 2006 Greece Pasiouras et al 2008 Malaysia Sufian and Majid 2006 Singapore Chu and Lim 1998 Turkey Erdem and Erdem 2008 and Spain Guzman and Reverte 2008 Furthermore in a crosscountry setting Beccalli et al 2006 provide evidence from France Germany Italy Spain UK that is the five principal EU banking sectors In most cases the results of these studies indicate a positive relationship between stock returns and efficiency changes Furthermore Beccalli et al 2006 reveal that the explanatory power of the model with DEA scores is higher than that of a model that uses the return on equity ROE as a measure of performance 323 Bank ownership A number of studies compare the efficiency of banks across different ownership types One approach used in the literature is to split the sample and compare the means of the different ownership groups Another approach is to incorporate dummy variables in a second stage analysis as the one described in Section 321 Some studies compare domestic and foreign banks Havrylchyk 2006 finds that greenfield banks are more efficient than domestic banks in Poland whereas foreign banks that acquired domestic ones have not successfully increased their efficiency Sturm and Williams 2004 report that foreign banks in Australia are more efficient than domestic ones Isik and Hassan 2003a find that foreign banks are more efficient than private domestic Turkish banks while Isik 2008 reports similar results for TFP growth estimates However Ataullah and Le 2004 find that prior to the financial liberalization of 19911992 foreign banks were less efficient than domestic ones in India and Pakistan nevertheless the opposite picture emerged after this period Other studies examine the efficiency of stateowned banks Many of them find that stateowned banks are less efficient than other banks For instance GarciaCestona and Surroca 2008 find that Spanish banks controlled by insiders ie managers and workers are more efficient than the ones controlled by public administrations Ariff and Can 2008 find that jointstock banks in China are more cost and profitefficient than stateowned banks Chen 1998 in Taiwan and Mercan et al 2003 in Turkey also report that the efficiency of privatelyowned banks is higher than that of stateowned banks Similarly Isik 2008 focuses on TFP growth and finds that the growth of private banks is more than double that of state banks In contrast other studies report that the efficiency of private banks is lower than that of stateowned banks in Turkey Isik and Hassan 2003a India Sathye 2003 and Austria and Germany Hauner 2005 324 Corporate events and efficiency Another part of the literature relates DEA efficiency estimates to corporate events such as mergers and acquisitions andor bankruptcy Some of the studies compare the efficiency of acquirers and their targets and examine whether mergers improve efficiency Avkiran 1999 reports that acquiring banks are more efficient than acquired Australian banks However the results of this study also indicate that acquiring banks do not always maintain their premerger efficiency and there is mixed evidence on the extent to which the benefits of efficiency are passed to the public The results of AlSharkas et al 2008 indicate that merged US banks are on average more technically efficient and that they experience higher productivity growth than nonmerged banks Hahn 2007a finds evidence that Austrian banks which engaged in domestic merger deals achieved a higher productive efficiency level than banks which did not participate in such deals The results also show that merger gains remain significant over a longer period of time more than 5 years but there is slight tendency to level off Sherman and Rupert 2006 also focus on the efficiency gains from bank mergers but they concentrate on the branch rather than the banklevel They find that there are opportunities to reduce operating costs Such benefits however are not realised until 4 years after the merger Kohers et al 2000 follow a different approach to test whether Xefficiencies influence the markets assessment of bank mergers They find that the abnormal returns of bidders around the announcement period were associated with both the target banks profit efficiency and cost efficiency as well as the difference between the average peer cost efficiency score and the corresponding targets cost16 Wheelock and Wilson 2000 also relate efficiency with bank acquisitions However in their case DEA efficiency scores are used as an input in hazard models that capture the acquisition and failure likelihood They find that inefficiency increases the risk of failure while reducing the probability of a US banks being acquired Using a sample of 245 US banks Luo 2003 also reports that overall technical efficiency can be useful in predicting bank failures 325 Regulatory reformliberalization and efficiency A number of studies examine the impact of regulatory reform and liberalization initiatives on bank efficiency and productivity The studies that we review seem to indicate a positive relationship in most countries including India and Pakistan Ataullah et al 2004 Australia Avkiran 1999 Sturm and Williams 2004 US Mukherjee et al 2001 China Chen et al 2005 Greece Tsionas et al 2003 Rezitis 2006 Taiwan Wang and Huang 2007 Korea Gilbert and Wilson 1998 Turkey Isik and Hassan 2003b and the new EU countries Brissimis et al 2008 However Hauner 2005 finds no evidence that productivity is related to deregula tion in Germany and Austria while Sathye 2002 argues that there can be a limit of deregulation after which no further productivity gains could be possible due to deregulation Some other interesting observations can be summarized as fol lows The banking sector reform in the newly acceded EU countries had a positive impact on bank efficiency while the effect of reform on TFP growth was significantly only toward the end of the reform process Brissimis et al 2008 The implementation of financial liberalization programmes may enable foreign banks to overcome the liability of foreignness and enhance their resource utilization Ataullah and Le 2004 The impact of deregulation will not neces sarily be the same across different bank ownerships such as state domestic foreign Isik and Hassan 2003b 326 Comparison of frontier techniques A few studies compare alternative frontier techniques over the period of our survey Bauer et al 1998 use a sample of US banks to compare DEA SFA TFA thick frontier approach and DFA dis tributionfree approach over six consistency criteria They find that DEA yields much lower average efficiencies ranks the banks differently and identifies the best and worst banks differently from parametric methods Furthermore compared to DEA the parametric measures were more highly correlated with the tradi tional nonfrontier performance measures Huang and Wang 2002 provide another comparison of SFA DFA and DEA using data from Taiwanese banks They also argue that the choice of the frontier approach can result in different conclusions Weill 2004 provides more recent evidence using a sample from five European countries However this crosscountry setting does not alter the main conclusion of lack of robustness across different approaches confirming the results of the previous two studies Delis et al 2009 provide further support to these findings using a dataset of Greek banks The study of Beccalli et al 2006 also provides useful insights for the differences of SFA and DEA While this study does not focus on the comparison between these two methods it finds that while changes in the stock prices of banks reflect percentage changes in DEA cost efficiency scores this trend is less clear with SFA efficiency estimates In contrast Fiordelisi 2008 reports that SFA cost efficiency estimates explain better the variations in shareholder value creation than those derived from DEA Casu et al 2004 compare productivity growth estimates ob tained through parametric and nonparametric approaches Their study seems to indicate that in this case the differences are not as large as in the efficiency studies They conclude that although the alternative methodologies produce in some cases opposing findings as for the sources of productivity for individual years in general they do not yield noticeably different results in terms of identifying the components of the productivity growth of EU banks during the period of their study 327 Efficiency of bank branches While most of the above studies focus on the efficiency of bank ing institutions as a whole a related strand of the literature exam ines the efficiency of bank branches17 In general branches are predominantly regarded as production units see eg Camanho and Dyson 2005ab 2006 Soteriou and Zenios 1999b Golany and Storbeck 1999 Athanassopoulos and Giokas 2000 where output is measured by the number of various accounts andor transactions as contrast to the intermediation approach employed by Portela and Thanassoulis 2005 2007 and Portela et al 2003 where branches are seen as intermediaries and thus outputs are measured in mone tary terms There are studies which consider the changing role of bank branches from a predominantly transactionbased one to a sales oriented role Cook et al 2000 and Cook and Hababou 2001 dis tinguish sales and services transactions functions of bank branches in a Canadian bank where inputs are usually shared be tween these two functions To model the shared resources Cook et al 2000 extend the usual methodology to develop a model which incorporates the best resource split and optimises the aggre gate efficiency score Further Cook and Hababou 2001 model the shared resources concept by adopting the Additive DEA model In a recent novel application Portela and Thanassoulis 2007 assess the branches of a Portuguese bank in terms of their performance in their new roles by specifying efficiency measures which con sider sales transactional activities as well as bank branch profit efficiency Service quality is also a significant dimension of bank branches performance Two different ways are developed in empirical applications to consider service quality For instance Soteriou and Zenios 1999b Golany and Storbeck 1999 and Athanassopo ulos and Giokas 2000 include quality variables in efficiency anal ysis whereas Portela and Thanassoulis 2007 use a post hoc analysis to compare service quality index with efficiency measures They find that service quality is positively related with operational and profit efficiency The diversity of environments in which the branches are oper ating is also considered in the literature Paradi and Schaffnit 2004 considered the role of the environmental parameters that are outside the management control in the commercial branches of a large Canadian bank and incorporated risk and economic growth rate of the region as two nondiscretionary factors On the other hand Camanho and Dyson 2006 assess the impact of environmental factors and regional managerial policies on branches productivity through the construction of an index that reflects the relative performance of bank branches operating in four different regions in Portugal Das et al 2009 introduce the concept of spatial efficiency for each region relative to the nation and thus measure the effects of differences in the regional characteristics on the efficiency of bank branches across four 16 Only cost efficiency was estimated with DEA in this study Profit efficiency was estimated using stochastic frontier analysis As in previous studies Kohers et al 2000 also provide comparisons of efficiency across different groups They find that acquiring US bank holding companies BHCs are less costefficient compared to their targets Also industry peers operate with greater cost efficiencies that then bidder and target BHCs 17 See Table 3 in the Supplementary material Appendix A for information on branch efficiency studies 194 MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 195 MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 associated with nonfailed banks and the unsafe traits failed banks known as features Then these features are used to vote on each bank and classify it as failed or nonfailed Lanine and Vander Vennet 2006 argue that the general rule of classifying safe and unsafe cells and assigning default probabilities works poorly when the number of banks falling into mixed cells is high To avoid this problem they suggest a formula to compute the probability of default directly without preliminary classification of cells as safe or unsafe Some researchers use soft computing approaches with the aim of synthesizing the human ability to tolerate and process uncer tain imprecise and incomplete information during the decision making Tung et al 2004 Soft computing usually integrates var ious techniques that may originate from different disciplines such as fuzzy logic19 neural networks machine learning etc in various combinations that will allow the researcher to exploit the strengths of individual techniques Ravi Kumar and Ravi 2007 A popular ap proach is the integration of neural network and fuzzy sets to develop a neural fuzzy network For example Tung et al 2004 propose a Generic Selforganizing Fuzzy Neural Network GenSoFNN based on the compositional rule of inference CRI to predict bank failure Multiple discriminant analysis MDA and logistic regression analysis LRA are two traditional techniques that are commonly used as benchmarks in classification studies eg Swicegood and Clark 2001 Pasiouras et al 2007 In contrast Tung et al 2004 benchmark their models against Cox regression which estimates the dependence of the risk of failure on banks characteristics and of the evolution of that risk over time 42 Variables In most of the cases the variables are selected on the basis of the CAMEL model and include measures for Capital strength Asset quality Earnings and Liquidity20 The number of variables differs significantly among studies Some researchers start from a large list of variables and then use statistical screening eg KruskalWallis or dimension reduction eg factor analysis to end up with a re duced set of variables In addition to the financial variables some studies include additional characteristics related to ownership eg of shares held by the government market information eg stock price in previous year auditing and country eg Heritage index 43 Classification results Many of the studies review in this section report quite satisfac tory classification accuracies and in several cases the proposed OR and AI methods outperform the traditional techniques ie MDA LRA Cox regression Furthermore the studies that focus on bankruptcy prediction in general perform more accurately than those that deal with credit ratings However this is not sur prising for two reasons First while classifying the banks in more groups is more informative it becomes more difficult to discrimi nate among groups in the intermediate categories Second credit ratings may incorporate additional qualitative information and hu man judgment that cannot be captured by the conventional vari ables used in most studies In any case comparisons across different studies should be trea ted with extreme caution One reason is that different studies make different assumptions about the prior probabilities of group membership as well as misclassification costs A second reason is that different studies use different approaches to validate the developed models For instance some researchers simply split the total sample in training and holdout datasets without account ing for the stability of models over time others use resampling techniques eg kfold crossvalidation and others use a holdout sample from a future period However it should kept in mind that classification ability is likely to overstate predictive ability Thus a superior approach would require that the model be validated against a future period as this approach more closely reflects a real world setting 5 Concluding remarks We have presented a comprehensive review of applications of operational research and artificial intelligence techniques in the assessment of bank performance by discussing a total of 196 stud ies published between 1998 and early 2009 We have classified the studies in two main categories The first uses DEA and the DEAlike Malmquist index to esti mate the efficiency and productivity growth of banks and bank branches We have discussed various methodological issues such as the estimated measures of efficiency the underlying assump tions of the estimated models and the selection of inputs and out puts We also discussed the main topics of interest including the relationship between ownership and efficiency stock returns and efficiency the determinants of efficiency the efficiency of bank branches amongst others Three of the main conclusions are that i profit efficiency and capacity efficiency have received quite lim ited attention in DEA studies in banking ii most studies that use a twostage DEA do not employ appropriate bootstrapping tech niques and their results may be biased iii there is much diversity among studies with respect to the selection of input and outputs As in the case of banks as a whole cost and profit efficiency has re ceived considerably less attention in branch efficiency studies Fur thermore an area of research deserving attention would be the estimation of bank branches efficiency over successive time periods Studies falling into the second category attempt to develop classification models to predict failure the credit ratings of banks and identify underperformers We have found both countryspe cific and crosscountry studies and applications of techniques that originate from various disciplines Most of the studies rely heavily on financial information although in some cases nonfinancial vari ables are also used Given the differences in the approaches em ployed to validate the models comparisons of classification accuracies across studies should be treated with caution We find only a few studies that propose the combination of the predictions of individual models into integrated metaclassifiers and we be lieve that this is an area of research that is worthy of further attention Appendix A Supplementary data Supplementary data associated with this article can be found in the online version at doi101016jejor200908003 References AlSharkas AA Hassan MK Lawrence S 2008 The impact of mergers and acquisitions on the efficiency of the US banking industry Further evidence Journal of Business Finance and Accounting 35 5070 Ariff M Can L 2008 Cost and profit efficiency of Chinese banks A nonparametric analysis China Economic Review 19 260273 Ataullah A Cockerill T Le H 2004 Financial liberalization and bank efficiency A comparative analysis of India and Pakistan Applied Economics 36 19151924 Ataullah A Le H 2004 Financial repression and liability of foreignness in developing countries Applied Economics Letters 11 545549 19 The fuzzy set theory aims to classify subjective reasoning and assign degrees of possibilities in reaching conclusions Fuzzy logic can also be used to obtain fuzzy if then rules 20 M in CAMEL refers to management quality and it is rarely used in empirical studies due to difficulties in assessing management quality 196 MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 Ataullah A Le H 2006 Economic reforms and bank efficiency in developing countries The case of the Indian banking industry Applied Financial Economics 16 653663 Athanassopoulos AD Giokas D 2000 The use of data envelopment analysis in banking institutions Evidence from the commercial bank of Greece Interfaces 30 2 8195 Avkiran NK 1999 The evidence on efficiency gains The role of mergers and the benefits to the public Journal of Banking and Finance 23 9911013 Avkiran NK 2009a Opening the black box on efficiency analysis An illustration with UAE banks Omega 37 930941 Avkiran NK 2009b Removing the impact of environment with unitsinvariant efficient frontier analysis An illustrative case study with intertemporal panel data Omega 37 535544 Banker RD Charnes A Cooper WW 1984 Some models for estimating technical and scale inefficiencies in data envelopment analysis Management Science 30 10781092 Banker RD Morey RC 1986 Efficiency analysis for exogenously fixed inputs and outputs Operations Research 34 513521 Bauer PW Berger AN Ferrier GD Humphrey DB 1998 Consistency conditions for regulatory analysis of financial institutions A comparison of frontier efficiency methods Journal of Economics and Business 50 85114 Beccalli E Casu B Girardone C 2006 Efficiency and stock performance in european banking Journal of Business Finance and Accounting 33 245262 Bergendahl G 1998 DEA and benchmarks An application to Nordic banks Annals of Operations Research 82 233249 Berger AN 2007 International comparisons of banking efficiency Financial Markets Institutions and Instruments 16 119144 Berger AN Humphrey DB 1997 Efficiency of financial institutions International survey and directions for future research European Journal of Operational Research 98 175212 Berger AN Mester LJ 1997 Inside the black box What explains differences in the efficiencies of financial institutions Journal of Banking and Finance 21 895947 Board J Sutcliffe C Ziemba WT 2003 Applying operations research techniques to financial markets Interfaces 33 1224 Boyacioglu MA Kara Y Baykan OK 2009 Predicting bank financial failures using neural networks support vector machines and multivariate statistical methods A comparative analysis in the sample of savings deposit insurance fund SDIF transferred banks in Turkey Expert Systems with Applications 36 16851694 Brissimis SN Delis MD Papanikolaou NI 2008 Exploring the nexus between banking sector reform and performance Evidence from newly acceded EU countries Journal of Banking and Finance 32 26742683 Camanho AS Dyson RG 2005a Cost efficiency measurement with price uncertainty A DEA application to bank branch assessments European Journal of Operational Research 161 432446 Camanho AS Dyson RG 2005b Cost efficiency Production and valueadded models in the analysis of bank branch performance Journal of the Operational Research Society 56 483494 Camanho AS Dyson RG 2006 Data envelopment analysis and Malmquist indices for measuring performance Journal of Productivity Analysis 26 3549 Camanho AS Dyson RG 2008 A generalisation of the Farrell cost efficiency measure applicable to nonfully competitive settings Omega 36 147162 Canbas S Cabuk A Kilic SB 2005 Prediction of commercial bank failure via multivariate statistical analysis of financial structures The Turkish case European Journal of Operational Research 166 528546 Canhoto A Dermine J 2003 A note on banking efficiency in Portugal New vs Old banks Journal of Banking and Finance 27 20872098 Casu B Girardone C 2004 Financial conglomeration Efficiency productivity and strategic drive Applied Financial Economics 14 687696 Casu B Girardone C 2006 Bank competition concentration and efficiency in the single European Market The Manchester School 74 441468 Casu B Girardone C Molyneux P 2004 Productivity change in European banking A comparison of parametric and nonparametric approaches Journal of Banking and Finance 28 25212540 Casu B Molyneux P 2003 A comparative study of efficiency in European banking Applied Economics 35 18651876 Charnes A Cooper WW Rhodes E 1978 Evaluating program and managerial efficiency An application of data envelopment analysis to program follow through Management Science 27 668697 Charnes A Cooper WW Rhodes E 1981 Evaluating program and managerial efficiency An application of data envelopment analysis to program follow through Management Science 27 668697 Chen TY 1998 A study of bank efficiency and ownership in Taiwan Applied Economics Letters 5 613616 Chen TY 2001 An estimation of Xinefficiency in Taiwans banks Applied Financial Economics 11 237242 Chen WH Shih JY 2006 A study of Taiwans issuer credit rating systems using support vector machines Expert Systems with Applications 30 427 435 Chen X Skully M Brown K 2005 Banking efficiency in China Application of DEA to pre and postderegulation eras 19932000 China Economic Review 16 229245 Chu SF Lim GH 1998 Share performance and profit efficiency of banks in an oligopolistic market Evidence from Singapore Journal of Multinational Financial Management 8 155168 Coelli TJ Perelman SA 1996 Comparison of Parametric and Nonparametric Distance Functions With Application to European Railways CREPP Discussion Paper No 199611 University of Liege Liege Coelli TJ Prasada Rao DS ODonnell CJ Battese GE 2005 An Introduction to Efficiency and Productivity Analysis second ed Springer USA Cook WD Seiford LM 2009 Data envelopment analysis DEA Thirty years on European Journal of Operational Research 192 117 Cook WD Hababou M 2001 Sales performance measurement in bank branches Omega 29 299307 Cook WD Hababou M Tuenter HJH 2000 Multicomponent efficiency measurement and shared inputs in Data Envelopment Analysis An application to sales and service performance in bank branches Journal of Productivity Analysis 14 209224 Das A Ghosh S 2006 Financial deregulation and efficiency An empirical analysis of Indian banks during the post reform period Review of Financial Economics 15 193221 Das A Ray SC Nag A 2009 Labouruse efficiency in Indian banking A branch level analysis Omega 37 411425 Dekker D Post T 2001 A quasiconcave DEA model with an application for bank branch performance evaluation European Journal of Operational Research 132 296311 Delis MD KoutsomanoliFillipaki K Staikouras CK Gerogiannaki K 2009 Evaluating cost and profit efficiency A comparison of parametric and nonparametric methodologies Applied Financial Economics 19 191202 Delis MD Papanikolaou NI 2009 Determinants of bank efficiency Evidence from a semiparametric methodology Managerial Finance 35 260275 Dimitras AI Zanakis SH Zopounidis C 1996 A survey of business failures with an emphasis on prediction methods and industrial applications European Journal of Operational Research 90 487513 Drake L Hall MJB Simper R 2006 The impact of macroeconomic and regulatory factors on bank efficiency A nonparametric analysis of Hong Kongs banking system Journal of Banking and Finance 30 14431466 Erdem C Erdem MS 2008 Turkish banking efficiency and its relation to stock performance Applied Economics Letters 15 207211 Fare R Grosskopf S Weber WL 2004 The effect of riskbased capital requirements on profit efficiency in banking Applied Economics 36 1731 1743 Fiordelisi F 2008 Efficiency and shareholder return in banking International Journal of Banking Accounting and Finance 1 114132 Gaganis C Pasiouras F Zopounidis C 2006 A multicriteria decision framework for measuring banks soundness around the world Journal of MultiCriteria Decision Analysis 14 103111 Gaganis C Liadaki A Doumpos M Zopounidis C 2009 Estimating and analyzing the efficiency and productivity of bank branches Managerial Finance 35 202 218 GarciaCestona M Surroca J 2008 Multiple goals and ownership structure Effects on the performance of Spanish savings banks European Journal of Operational Research 187 582599 Gilbert RA Wilson PW 1998 Effects of deregulation on the productivity of Korean banks Journal of Economics and Business 50 133155 Giokas DI 2008a Assessing the efficiency in operations of a large Greek bank branch network adopting different economic behaviours Economic Modelling 25 559574 Golany B Storbeck JE 1999 A Data Envelopment Analysis of the operational efficiency of bank branches Interfaces 29 1426 Guzman I Reverte C 2008 Productivity and efficiency change and shareholder value Evidence from the Spanish banking sector Applied Economics 40 2033 2040 Hahn FR 2007a Domestic mergers in the Austrian banking sector A performance analysis Applied Financial Economics 17 185196 Halkos GE Salamouris DS 2004 Efficiency measurement of the Greek commercial banks with the use of financial ratios A data envelopment analysis approach Management Accounting Research 15 201224 Hauner D 2005 Explaining efficiency differences among large German and Austrian banks Applied Economics 37 969980 Havrylchyk O 2006 Efficiency of the Polish banking industry Foreign versus domestic banks Journal of Banking and Finance 30 19751996 Huang TH Wang MH 2002 Comparison of economic efficiency estimation methods Parametric and nonparametric techniques The Manchester School 70 682709 Huang Z Chen H Hsu CJ Chen WH Wu S 2004 Credit rating analysis with support vector machines and neural networks A market comparative study Decision Support Systems 37 543558 Isik I 2008 Productivity technology and efficiency of de novo banks A counter evidence from Turkey Journal of Multinational Financial Management 18 427 442 Isik I Hassan MK 2002 Technical scale and allocative efficiencies of Turkish banking industry Journal of Banking and Finance 26 719766 Isik I Hassan MK 2003a Efficiency ownership and market structure corporate control and governance in the Turkish banking industry Journal of Business Finance and Accounting 30 13631421 Isik I Hassan MK 2003b Financial deregulation and total factor productivity change An empirical study of Turkish commercial banks Journal of Banking and Finance 27 14551485 Kantor J Maital S 1999 Measuring efficiency by product group Integrating DEA with activitybased accounting in a large Mideast bank Interfaces 29 2736 MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 197 Kirkwood J Nahm D 2006 Australian banking efficiency and its relation to stock returns The Economic Record 82 253267 Kohers T Huang MH Kohers N 2000 Market perception of efficiency in bank holding company mergers The roles of the DEA and SFA models in capturing merger potential Review of Financial Economics 9 101120 Kolari J Glennon D Shin H Caputo M 2002 Predicting large US commercial bank failures Journal of Economics and Business 54 361387 Lanine G Vander Vennet R 2006 Failure prediction in the Russian bank sector with logit and trait recognition models Expert Systems with Applications 30 463478 LozanoVivas A Pastor JT Hasan I 2001 European bank performance beyond country borders What really matters European Finance Review 5 141165 LozanoVivas A Pastor JT Pastor JM 2002 An efficiency comparison of European banking systems operating under different environmental conditions Journal of Productivity Analysis 18 5977 Luo X 2003 Evaluating the profitability and marketability efficiency of large banks An application of data envelopment analysis Journal of Business Research 56 627635 Maudos J Pastor JM 2003 Cost and profit efficiency in the Spanish banking sector 19851996 A nonparametric approach Applied Financial Economics 13 112 Maudos J Pastor JM Perez F 2002 Competition and efficiency in the Spanish banking sector The importance of specialization Applied Financial Economics 12 505516 Mercan M Reisman A Yolalan R Emel AB 2003 The effect of scale and mode of ownership on the financial performance of the Turkish banking sector Results of a DEAbased analysis SocioEconomic Planning Sciences 37 185202 Mukherjee K Ray SC Miller SM 2001 Productivity growth in large US commercial banks The initial postderegulation experience Journal of Banking and Finance 25 913939 Noulas AG 1997 Productivity growth in the Hellenic banking industry State versus private banks Applied Financial Economics 7 223228 Paradi JC Schaffnit C 2004 Commercial branch performance evaluation and results communication in a Canadian bankA DEA application European Journal of Operational Research 156 719735 Parkan C Wu ML 1999 Measurement of the performance of an investment bank using the operational competitiveness rating procedure Omega 27 201217 Pasiouras F Gaganis C Doumpos M 2007 A multicriteria discrimination approach for the credit rating of Asian banks Annals of Finance 3 351367 Pasiouras F 2008a International evidence on the impact of regulations and supervision on banks technical efficiency An application of twostage data envelopment analysis Review of Quantitative Finance and Accounting 30 187 223 Pasiouras F 2008b Estimating the technical and scale efficiency of Greek commercial banks The impact of credit risk offbalance sheet activities and international operations Research in International Business and Finance 22 301318 Pasiouras F Liadaki A Zopounidis C 2008 Bank efficiency and share performance Evidence from Greece Applied Financial Economics 18 1121 1130 Pastor JM 1999 Efficiency and risk management in Spanish banking A method to decompose risk Applied Financial Economics 9 371384 Pastor JM 2002 Credit risk and efficiency in the European banking system A threestage analysis Applied Financial Economics 12 895911 Pastor JM Serrano L 2006 The effect of specialisation on banks efficiency An international comparison International Review of Applied Economics 20 125 149 Portela MCAS Borges PC Thanassoulis E 2003 Finding closest targets in non oriented DEA models The case of convex and nonconvex technologies Journal of Productivity Analysis 19 251269 Portela MCAS Thanassoulis E 2005 Profitability of a sample of Portuguese bank branches and its decomposition into technical and allocative components European Journal of Operational Research 162 850866 Portela MCAS Thanassoulis E 2007 Comparative efficiency analysis of Portuguese bank branches European Journal of Operational Research 177 12751288 Prior D 2003 Long and shortrun nonparametric cost frontier efficiency An application to Spanish savings banks Journal of Banking and Finance 27 655 671 Ravi Kumar P Ravi V 2007 Bankruptcy prediction in banks and firms via statistical and intelligent techniques A review European Journal of Operational Research 180 128 Ravi V Kurniawan H Thai PNK Ravi Kumar P 2008 Soft computing system for bank performance prediction Applied Soft Computing 8 305315 Rezitis AN 2006 Productivity growth in the Greek banking industry A non parametric approach Journal of Applied Economics IX 119138 Sahoo BK Tone K 2009 Decomposing capacity utilization in data envelopment analysis An application to banks in India European Journal of Operational Research 195 575594 Sathye M 2003 Efficiency of banks in a developing economy The case of India European Journal of Operational Research 148 662671 Sathye M 2002 Measuring productivity changes in Australian banking An application of Malmquist indices Managerial Finance 28 4859 Sealey C Lindley J 1977 Inputs outputs and a theory of production and cost at depository financial institutions Journal of Finance 32 12511266 Sherman HD Rupert TJ 2006 Do bank mergers have hidden or foregone value Realized and unrealized operating synergies in one bank merger European Journal of Operational Research 168 253268 Simar L Wilson PW 2007 Estimation and inference in twostage semi parametric models of production processes Journal of Econometrics 136 31 64 Smith KA Gupta JND 2000 Neural networks in business Techniques and applications for the operations research Computers and Operations Research 27 10231044 Soteriou AC Zenios SA 1999a Using data envelopment analysis for costing bank products European Journal of Operational Research 114 224234 Soteriou A Zenios SA 1999b Operations quality and profitability in the provision of banking services Management Science 45 12211238 Sturm JE Williams B 2004 Foreign bank entry deregulation and bank efficiency Lessons from the Australian experience Journal of Banking and Finance 28 17751799 Sufian F Majid MZA 2006 Banks efficiency and stock prices in emerging markets Evidence from Malaysia Journal of AsiaPacific Business 7 3553 Swicegood P Clark JA 2001 Offsite monitoring systems for predicting bank underperformance A comparison of neural networks discriminant analysis and professional human judgment International Journal of Intelligent Systems in Accounting Finance and Management 10 169186 Tanna S 2009 The impact of foreign direct investment on total factor productivity growth International evidence from the banking industry Managerial Finance 35 297311 Tay FEH Cao L 2001 Application of support vector machines in financial time series forecasting Omega 29 309317 TortosaAusina E 2002a Bank cost efficiency and output specification Journal of Productivity Analysis 18 199222 TortosaAusina E 2003 Nontraditional activities and bank efficiency revisited A distributional analysis for Spanish financial institutions Journal of Economics and Business 55 371395 Tsionas EG Lolos SEG Christopoulos DK 2003 The performance of the Greek banking system in view of the EMU Results from a nonparametric approach Economic Modelling 20 571592 Thoraneenitiyan N Avkiran NK 2009 Measuring the impact of restructuring and countryspecific factors on the efficiency of postcrisis East Asian banking systems Integrating DEA with SFA SocioEconomic Planning Sciences 43 240 252 Tung WL Queka C Cheng P 2004 GenSoEWS A novel neuralfuzzy based early warning system for predicting bank failures Neural Networks 17 567587 Vapnik VN 1995 The Nature of Statistical Learning Theory SpringerVerlag New York Wang MH Huang TH 2007 A study on the persistence of Farrells efficiency measure under a dynamic framework European Journal of Operational Research 180 13021316 Weill L 2004 Measuring cost efficiency in European banking A comparison of frontier techniques Journal of Productivity Analysis 21 133152 Wheelock DC Wilson PW 2000 Why do banks disappear The determinants of US bank failures and acquisitions Review of Economics and Statistics 82 127 138 Wu D Yang Z Liang L 2006 Using DEAneural network approach to evaluate branch efficiency of a large Canadian bank Expert Systems with Applications 31 108115 Zen F Baldan C 2008 The strategic paths and performances of Italian mutual banks A nonparametric analysis International Journal of Banking Accounting and Finance 1 189214 Zhao H Sinha AP Ge W 2009 Effects of feature construction on classification performance An empirical study in bank failure prediction Expert Systems with Applications 36 26332644 Zhou P Ang BW Poh KL 2008 A survey of data envelopment analysis in energy and environmental studies European Journal of Operational Research 189 1 18 198 MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 UNIVERSIDADE DE SÃO PAULO FACULDADE DE ECONOMIA ADMINISTRAÇÃO E CONTABILIDADE DE RIBEIRÃO PRETO DEPARTAMENTO DE ADMINISTRAÇÃO PROGRAMA DE PÓSGRADUAÇÃO EM ADMINISTRAÇÃO DE ORGANIZAÇÕES CARLOS ROBERTO BONARETTI FILHO A eficiência bancária brasileira de 2000 a 2018 uma avaliação a partir da Fronteira Estocástica e do DEA de dois estágios RIBEIRÃO PRETO 2021 Prof Dr Vahan Agopyan Reitor da Universidade de São Paulo Prof Dr André Lucirton Costa Diretor da Faculdade de Economia Administração e Contabilidade de Ribeirão Preto Prof Dr Jorge Henrique Caldeira de Oliveira Chefe do Departamento de Administração Prof Dr João Luiz Passador Coordenador do Programa de PósGraduação em Administração de Organizações CARLOS ROBERTO BONARETTI FILHO A eficiência bancária brasileira de 2000 a 2018 uma avaliação a partir da Fronteira Estocástica e do DEA de dois estágios Versão Corrigida A original encontrase disponível na FEARPUSP Dissertação de Mestrado apresentada ao Programa de PósGraduação em Administração de Organizações da Faculdade de Economia Administração e Contabilidade de Ribeirão Preto da Universidade de São Paulo para a obtenção do título de Mestre em Ciências Orientador Prof Dr Alexandre Pereira Salgado Junior Ribeirão Preto 2021 metropolitan regions in India Gaganis et al 2009 examine the impact of regional conditions on the efficiency and productivity of bank branches in Greece Some interesting observations can be summarised as follows In general the existing studies of bank branch efficiency use data on a small number of branches of nonUS banks and mostly employ DEA There are studies however which evaluate bank branch performance with DEA and compare the results with other methodologies ie FDH Portela et al 2003 Malmquist indices Camanho and Dyson 2006 loglinear frontier method Giokas 2008a ABC Plus Kantor and Maital 1999 OCRA Parkan and Wu 1999 Neural Network Model Wu et al 2006 Quasiconcave DEA FDH and Parametric Frontier models Dekker and Post 2001 Most of the studies which were reviewed use technical and operational efficiencies as estimated measures Recent studies examine the allocative profit efficiency see Portela and Thanassoulis 2005 2007 and cost efficiency Camanho and Dyson 2005ab 2008 aspects of bank branches 4 Other OR and AI techniques and bank performance In this section we review 15 recent studies that develop classification models used in the prediction of bank failure 9 bank underperformance 2 and credit ratings 4 Canbas et al 2005 point out that the study of bank failure is important for at least two reasons First an understanding of the reasons beyond the failure allows regulatory authorities to manage and supervise banks more efficiently Second the ability to differentiate between healthy and troubled banks can reduce the expected cost of bank failure either by taking actions to prevent failure or by minimizing the costs to the public As discussed in Gaganis et al 2006 while many of the failure prediction models achieve very promising classification accuracies one common drawback is that they concentrate on the assignment of banks in two groups failed and nonfailed However the classification of banks as bad or good reduces the usefulness of the model Obviously the classification into more groups relates to another strand of the literature that deals with credit risk modelling and attempts to replicate the ratings assigned by the credit agencies In the following subsections we provide a brief description of the methodologies the employed variables and the classification results18 A summary of the main characteristics of these studies is available in Table 4 of the Supplementary material Appendix A 41 Methodologies Neural network NN modelling is an intelligence technique that follows a process similar to the human brain As in other classification models the parameters of an NN model need to be estimated before the network can be used for prediction purposes There are numerous NN architectures learning methods and parameters Generally speaking NN architectures can be either feedback or feedforward In a feedback network nodes can receive inputs from nodes in any other layer in the network whereas in a feedforward network inputs are only received from previous layers The multilayer preceptor MLPNN competitive learning neural networks CLNN selforganizing map neural networks SOMNN backpropagation neural networks BPNN and probabilistic neural networks PNN are alternative approaches employed in the studies under review eg Chen and Shih 2006 Boyacioglu et al 2009 FOLHA DE APROVAÇÃO Nome BONARETTIFILHO Carlos Roberto Título A eficiência bancária brasileira de 2000 a 2018 uma avaliação a partir da Fronteira Estocástica e do DEA de dois estágios Dissertação de Mestrado apresentada ao Programa de PósGraduação em Administração de Organizações da Faculdade de Economia Administração e Contabilidade de Ribeirão Preto da Universidade de São Paulo para a obtenção do título de Mestre em Ciências Aprovado em Banca examinadora Profa Dra Perla Calil Pongeluppe Wadhy Rebehy Instituição Faculdade de Administração Economia e Contabilidade de Ribeirão Preto FEARPUSP Assinatura Prof Dr Fabiano Guasti Lima Instituição Faculdade de Administração Economia e Contabilidade de Ribeirão Preto FEARPUSP Assinatura Prof Dr Andrei Aparecido de Albuquerque Instituição Universidade Federal de São Carlos Assinatura The approach of support vector machines SVMs used in Boyacioglu et al 2009 Chen and Shih 2006 and Huang et al 2004 was introduced by Vapnik 1995 It is based on the Structural Risk Minimization SRM principle from computational learning theory This seeks to minimise an upper bound of the generalisation error rather than minimise the training error SVMs use a linear structure to implement nonlinear class boundaries through extremely nonlinear mapping of the input vectors into the highdimensional space If the data are linearly separated SVM uses a special kind of linear model the optimal separating hyperplane that provides the maximum separation between the classes The training points that are closest to the maximum margin hyperplane are called support vectors All other training examples are irrelevant for determining the binary class boundaries From the implementation point of view Vapnik 1995 showed how training a SVM and finding the parameters leads to a quadratic optimisation problem with bound constraints and one linear equality constraint This means that the solution of SVMs is unique optimal and absent from local minima Tay and Cao 2001 Nearest neighbours is a nonparametric estimation method that has been applied in various problems in finance The nearest neighbour rule classifies an object ie bank to the class of its nearest neighbour in the measurement space using some kind of distance measure like the local metrics the global metrics the Mahalanobis or the Euclidean distance The modification of the nearest neighbour rule the knearest neighbour kNN method that is employed in Zhao et al 2009 classifies an object ie bank to the class ie failed or nonfailed more heavily represented among its knearest neighbours In the case of decision trees that originate from machine learning instead of developing classification functions or network architectures a binary decision tree is developed This can be accomplished through a set of ifthen split conditions that lead to the accurate classification of cases eg banks Commonly used algorithms are CART eg Ravi et al 2008 and C45 eg Zhao et al 2009 The UTilités Additives DIScriminantes UTADIS multicriteria decision aid MCDA method used in Gaganis et al 2006 employs the framework of preference disaggregation analysis for the development of an additive utility function that is used to score the firms and decide upon their classification The estimation of the additive utility model is performed through mathematical programming techniques The MultiGroup Hierarchical Discrimination MHDIS method used in Pasiouras et al 2007 is another MCDA approach that uses utility functions for discrimination purposes MHDIS distinguishes the groups progressively starting by discriminating the first group from all the others and then proceeds to the discrimination between the alternatives belonging to the other groups To accomplish this task instead of developing a single additive utility function that describes all alternatives as in UTADIS two additive utility functions are developed in each one of the n 1 steps where n is the number of groups At each stage of the hierarchical discrimination procedure two linear and a mixedinteger programming problems are solved to estimate the utility thresholds and the two additive utility functions in order to minimise the classification error As discussed in Kolari et al 2002 and Lanine and Vander Vennet 2006 trait recognition is another nonparametric approach under which individual traits are initially developed from different segments of the distribution of each variable and the interactions of these segments with one or more other variables segmented distributions When all possible traits of the variables are tabulated for all banks trait recognition uses a search routine to cull traits that do not discriminate between failed and nonfailed banks Trait recognition uses two sets of discriminators the safe traits AGRADECIMENTOS Primeiramente à minha mãe Irene exemplo de pessoa a ser seguido meu orgulho minha inspiração e meu alicerce sempre presente no meu desenvolvimento e em minhas conquistas Mãe e pai ao mesmo tempo Mulher forte guerreira e dedicada A melhor mãe que poderia ter Ao meu pai Carlos Bonaretti que mesmo não estando presente participa de certa forma de todas as minhas conquistas Meu exemplo a ser seguido pois se eu for metade da pessoa que ele foi serei um grande homem Um exemplo de pai de marido de filho de amigo de profissional enfim um exemplo de homem Dedico esse trabalho à ele meu pai que deve estar vibrando e todo orgulhoso com mais essa conquista Aos meus irmãos Luiz Otavio e André Luis pessoas essenciais em minha vida companheiros e cúmplices que com minha mãe são meus maiores orgulhos Presentes em toda a minha vida e que estarão sempre ao meu lado Aos grandes amigos com os quais a vida me presenteou Gabriel Riguetti Cesar Zambrano Murilo Kazon Paulo Camperoni Otávio Falconi e Giovani Magri Todos eles sem exceções companheiros a qualquer hora e amigos sempre presentes Amigos que levarei sempre comigo Ao meu orientador Prof Dr Alexandre Salgado que tanto me apoiou no desenvolvimento deste estudo e que compartilhou comigo toda sua experiência de vida acadêmica pessoa que muito admiro por todas suas conquistas e todo o seu conhecimento acumulado Também agradeço ao GREFIC grupo de estudos que sempre me apoiou À Profa Dra Perla pelo conhecimento e paciência Ao Marco Silva Allan Angelli e José Sardelari pela parceria Gostaria de agradecer também aos grandes amigos que a academia me trouxe Yago Marinzeck e Fabio Carlucci Para finalizar gostaria de agradecer à Deus que colocou todas essas pessoas maravilhosas em meu caminho e que me proporcionou experiências e oportunidades maravilhosas me deu o dom da vida e saúde para chegar aonde cheguei O presente trabalho foi realizado com apoio da Coordenação de Aperfeiçoamento de Pessoal de Nível Superior Brasil CAPES Código de Financiamento 001 RESUMO BONARETTI FILHO Carlos Roberto A eficiência bancária brasileira de 2000 a 2018 uma avaliação a partir da Fronteira Estocástica e do DEA de dois estágios 129 f Dissertação Mestrado em Administração de Organizações Faculdade de Economia Administração e Contabilidade de Ribeirão Preto da Universidade de São Paulo 2021 As mudanças no ambiente competitivo as políticas regulatórias e a inovação forma responsáveis nas últimas décadas por incentivarem uma significativa mudança no modelo de negócio das instituições bancárias Os bancos têm acrescido ao seu portifólio tradicional serviços como subscrição de títulos corretagem de seguros serviços de fundos mútuos serviços fiduciários e outras atividades geradoras de taxas o que incrementa sua receita além de estenderem suas fontes de passivos ao financiamento de atacado Devido à sua estrutura dinâmica e à complexidade do ambiente econômico em que operam os bancos estão sujeitos a muitos riscos dentre os quais destacase em relevância o risco de crédito responsável pelo aumento do custo marginal da dívida e do patrimônio líquido Assim à medida em que aumenta a exposição de um banco ao risco de crédito aumenta sua tendência de vivenciar uma crise financeira Daí a importância das instituições bancárias serem eficientes em suas atividades de intermediação e rentabilidade dois aspectos considerados pela academia como principais abordagens de análise de eficiência Nesse contexto o presente trabalho analisa a eficiência por meio de duas técnicas de mensuração a Análise Envoltória de Dados e a Análise da Fronteira Estocástica das instituições bancárias que compõem o sistema financeiro brasileiro São portanto analisados os principais fatores que influenciaram a eficiência bancária nacional entre os anos de 2000 e 2018 Foram considerados para análise os indicadores CAMELS das instituições financeiras Como resultado este estudo identificou as variáveis CAMELS que explicam a eficiência bancárias das instituições analisadas o que sugere que existem práticas relacionadas a referidas variáveis que influenciam negativamente os escores de eficiência Palavraschave eficiência bancária CAMELS DEA SFA Attract visitors to your showroom and get more customers Engaging content for your showroom with NFC With exhibition and showroom NFC capability customers can interact and access product information video or presentation without needing to handle anything Cover your video presentation or interesting pages with Hall of Posters Tags and get visitors to engage with your products You can use as many tags as you want ABSTRACT BONARETTI FILHO Carlos Roberto Brazilian banking efficiency from 2000 to 2018 an evaluation using Stochastic Frontier Analysis and twostage DEA 133 f Dissertation Masters Degree in Business Administration School of Economics Business Administration and Accounting at Ribeirão Preto University of São Paulo 2021 In the last decades changes in the competitive environment regulatory policies and innovation have been responsible for encouraging a significant change in banks business models Banks have added to their traditional portfolio services such as securities underwriting insurance brokerage mutual fund services fiduciary services and other feegenerating activities which increase their revenue in addition to extending their sources of liabilities to finance wholesale Due to their dynamic structure and the complexity of the economic environment in which they operate banks are subject to many risks among which stands out the credit risk responsible for the increase in the marginal cost of debt and equity Thus as the banks exposure to credit risk increases so does its tendency to experience a financial crisis Hence the importance of banking institutions being efficient in their intermediation and profitability activities two aspects considered by academia as the main approaches to analyzing efficiency In this context the present study analyzes the efficiency through two measurement techniques Data Envelopment Analysis and Stochastic Frontier Analysis of the banking institutions that make up the Brazilian financial system Therefore the main factors that influenced the national banking efficiency between 2000 and 2018 are analyzed The CAMELS indicators of financial institutions were considered for analysis As a result this study identified the CAMELS variables that explain the banking efficiency of the analyzed institutions which suggests that there are practices related to these variables that negatively influence the efficiency scores Keywords bank efficency CAMELS DEA SFA Share social media Share your Instagram TikTok Twitter to your visitors and get more followers Turn followers into buyers Share your Shop link on your favorite marketplace to get more regular customers Share music Share and play music in your shop or at your event Share QR codes Share several websites and QR codes from a single Meko Tag LISTA DE ABREVIATURAS E SIGLAS BACEN Banco Central do Brasil BCC Banker Charnes e Cooper CAMELS Capital Assets Management Earnings Liquidity Sensibility to market risk CCR Charnes Cooper e Rhodes CRS Constant Returns to Scale DEA Data Envelopment Analysis DMU Decision Making Units FFIEC Federal Financial Institutions Examination Council FMI Fundo Monetário Internacional FSI Financial Soundness Indicators iid independente e identicamente distribuída IMF International Monetary Fund OCC Office of the Comptroller of the Currency ROA Return on Assets ROE Return on Equity SFA Stochastic Frontier Analysis UFIRS Uniform Financial Institutions Rating System VRS Variable Returns to Scale Make it easy to connect Your visitors can scan a single tag with their phone to link to all your social media and clicktocall number Get more followers and share updates Motivation your customers to follow you on social media realtime with a single tap Link with all your customers Share your social media websites phone numbers product information and location from a single tag If You want to order single or bulk tags for your business please contact us Your logo your choice For bulk order you can select your logos for custom tags Or you can choose our design NFC Tag MekoTag contactmekotagcom 66 0 913460561 wwwmekotagcom mekotag mekotag showroom tags LISTA DE QUADROS Quadro 1 Principais funções determinísticas 42 Quadro 2 Quadro sinótico de revisão bibliográfica sobre utilização de DEA na avaliação de eficiência do sistema bancário 48 Quadro 3 Quadro sinótico de revisão bibliográfica sobre utilização da Fronteira Estocástica para a análise da eficiência bancária 50 Quadro 4 Literatura sobre variáveis representativas do acrônimo CAMELS C A M 63 Quadro 5 Literatura sobre variáveis representativas do acrônimo CAMELS E L S 67 Quadro 6 Variáveis independentes advindas do CAMELS 80 Quadro 7 Adequação à Teoria Financeira Abordagem de Intermediação e Rentabilidade 104 YOUR JOURNEY STARTS HERE A DIVERSE ABUNDANCE OF OUR SELECTED INDEPENDENT FILMMAKERS ANIMATION ART DOCUMENTARY DRAMA EXPERIMENTAL FICTION FRISCO IS A PROGRAM OF EXPERIMENTAL AND INDEPENDENT FILMMAKERS BASED IN SAN FRANCISCO THE PROGRAM PRESENTS LOCAL FILMMAKERS WORK IN CONTINUOUS MONTHLY PROGRAMS OFFERING A TRUE WAVER OF SOLID ART DIVERSE PERSPECTIVES AND A SENSE OF COMMUNITY FOR THE FILMMAKER AND FILMLOVING AUDIENCE chal chill 8 h mafiomaw hale lanoelemaris lONstiNVA S RWINKEYP Ty UDDATTI YAAAA 101 HOST 202 Installation Liberty SCIENCE FANTASY STARS CE LE RACY 2 SAYS LISTA DE TABELAS Tabela 1 Inputs e Outputs de Macoris Salgado Junior e Falsarella Junior 2015 75 Tabela 2 Inputs e Outputs utilizados em DEA e SFA com múltiplos inputs e outputs 75 Tabela 3 Inputs e Outputs utilizados em DEA e SFA com um único output 76 Tabela 4 Estatísticas descritivas dos escores de DEA e SFA 88 Tabela 5 Variáveis significantes para Intermediação model summary 89 Tabela 6 Variáveis significantes para Intermediação coefficients a 89 Tabela 7 Variáveis significantes para Rentabilidade model summary 90 Tabela 8 Variáveis significantes para Rentabilidade coefficients a 90 Tabela 9 Regressão para efeitos fixos de Intermediação 91 Tabela 10 Regressão para efeitos fixos de Rentabilidade 92 Tabela 11 Efeitos na eficiência SFA e DEA para Abordagem de Intermediação 99 Tabela 12 Efeitos na eficiência SFA e DEA para Abordagem de Rentabilidade 102 LISTA DE FIGURAS Figura 1 Fronteira de Eficiência 34 Figura 2 Minimização de Inputs e Maximização de Output CCR 35 Figura 3 Representação gráfica CCR Orientado a Inputs 1 e Outputs 2 36 Figura 4 Minimização de Inputs e Maximização de Output BCC 37 Figura 5 Representação gráfica BCC Orientado a Inputs 1 e Outputs 2 37 Figura 6 Classificação entre Ganhos de Escala e Orientação 38 Figura 7 Fronteira Estocástica de Produção 40 Figura 8 DEA e SFA dois estágios com regressão linear múltipla 79 Santo A Rookie But On His Way VIDEO MUSIC FEATURING J SEEK ON STAGE COLORADO SPRINGS CHRISTIAN ARTISTS SPREAD THE WORD OF CHRIST THROUGH MUSICAL EVENTS DUBLIN VIDEO JESSE SANTO TALKS ABOUT LEADING A CHRISTIAN HIP HOP CONCERT SERIES IN COLORADO SPRINGS VIDEOGRAPHY BY MIRIAM BRADY BY KENDRACOLE AMERICA IS KNOWN FOR ITS MANY EXTREMES FROM THE HOT SUMMERS TO THE COLD WINTERS FROM THE BUSINESS WORLD TO THE MUSICAL ARENA AND FROM BASKETBALL STARS TO SOLO SINGERSONGWRITERS GIVEN THAT OUR PLUTONIC MISSION IS TO HELP AND APPRECIATE BLACK AND LATINO CHRISTIAN ARTISTS ESPECIALLY THOSE WHO LIVE IN UNDERRESOURCED COMMUNITIES IT CAME AS NO SURPRISE THAT JESSE SANTO AN UP AND COMING CHRISTIAN HIPHOP ARTIST FROM COLORADO SPRINGS COLORADO STRUGGLES TO GET THE ATTENTION OF THE HIGHER UPS AT THE NATIONAL CHRISTIAN MUSIC INDUSTRY LEADERSHIP LEVEL FOR THE MOST PART HIS MUSIC AND ARTISTIC STYLES HAVE BEEN BRUSHED ASIDE FOR HAVING TOO MUCH URBAN CHRISTIAN SOUND INSTEAD OF THE TRADITIONAL FULL BAND SOUND THEY DONT WANT RAP OR HIPHOP AS CHRISTIAN MUSIC A LOT OF PEOPLE DONT GET IT SANTO SAID SANTO WANTED TO FIX THAT HE COULD NOT LET OTHER PEOPLE TELL HIM TO EMBRACE THE LEGACY OF ITS FORMALITY OF HOW TO BE A CHRISTIAN ARTIST THE PHILOSOPHY OF IF IT DOESNT FIT THE FORMULA WE DONT WANT IT IS EXACTLY WHY SANTOS MESSAGE TO YOUNG ARTISTS IS TO SEARCH FOR YOUR AUTHENTICITY IS THERE ROOM FOR US SANTO ASKED THERES ALWAYS ROOM FOR YOUR AUTHENTICITY IF THE FACT THAT YOURE BLACK IF THE FACT THAT YOURE LATINO IF THE FACT THAT YOURE YOUNG IF THE FACT THAT YOURE OLD IF THE FACT BE THAT YOU WERE RAISED IN A SINGLE MOTHER HOME IT DOESNT MATTER WE ALL HAVE THE WARRIORS IN OUR CIRCLES THAT HAVE BEEN CALLED TO PUSH OUR CULTURE FORWARD WE HAVE TO DO THAT WE HAVE TO PUSH FORWARD THE POWER OF WHATEVER GIFT YOU HAVE USE IT FOR THE GLORY OF GOD ON STAGE SANTOS PROJECT IS CALLED WARRIORS FOR LIGHT RECORDS AND ITS CONCEPT RPUSHES ARTISTS TO EMBRACE THE UNIQUE SOUND AND SOUNDS OF OUR CITIES AND COMMUNITIES SANTO IS INITIATING A CHRISTIAN HIPHOP CONCERT SERIES CALLED THE GOSPEL MOB TOUR THAT IS CURRENTLY DOING A WEST COAST RUN STARTING IN FEBRUARY 2024 HE PARTNERED WITH LOCAL COLORADO SPRINGS PROMOTERS TO BRING THE MUSIC AND DANCE CULTURE TO LOCAL CHURCHES AND VENUES THAT TYPICALLY SERVE BLACK AND LATINO COMMUNITIES THESE SHOWS ARE ABOUT CELEBRATING CHRISTIAN URBAN ARTISTS MUSIC AND DANCE CULTURE THERE IS A NEED IN THE URBAN CHURCHES FOR THIS SHOWCASE OF TALENTS HE SAID LISTA DE GRÁFICOS Gráfico 1 Abordagem Intermediação 84 Gráfico 2 Abordagem Rentabilidade 86 VIDEO MUSIC JESSE SANTO TALKS ABOUT LEADING A CHRISTIAN HIP HOP CONCERT SERIES IN COLORADO SPRINGS COLORADO CHRISTIAN ARTISTS SPREAD THE WORD OF CHRIST THROUGH MUSICAL EVENTS THATS WHERE MY PASSION COMES FROM I AM EQUIPPED TO SUPPORT THIS ART FORM AND TO RAISE UP ARTISTS TO THIS PLATFORM SANTO SAID I FEEL LIKE I AM DOING MY PART WITH THIS CONCERT SERIES THIS TYPE OF CONCERT SERIES ALSO HELPS LOCAL ARTISTS TO RECOGNIZE THAT THERE IS ROOM FOR THEM AS CHRISTIAN MUSICIANS IN THEIR COMMUNITIES ITS NOT JUST ABOUT MUSIC FOR ME ITS ABOUT POSITIVE COMMUNITY MENTAL HEALTH SPIRITUALITY SO THATS THE BASE OF WHAT I AM WORKING TO DO WITH THESE EVENTS SANTO SAID MORE INFO JESSESANTOJESSE IMINSTAGRAM JESSESANTOMUSICGMAILCOM 23 ÍNDICE 1 INTRODUÇÃO 25 11 JUSTIFICATIVA 27 12 PROBLEMATIZAÇÃO E OBJETIVOS DA PESQUISA 28 2 REFERENCIAL TEÓRICO 29 21 CONCEITO DE EFICIÊNCIA 29 22 TÉCNICAS DE ANÁLISE DA EFICIÊNCIA 30 221 Análise Envoltória de Dados 31 222 Análise da Fronteira Estocástica 37 23 EFICIÊNCIA BANCÁRIA 43 24 FERRAMENTAS DE AVALIAÇÃO DA EFICIÊNCIA BANCÁRIA OU MENSURAÇÃO DA EFICIÊNCIA BANCÁRIA 46 25 FATORES EXPLICATIVOS DO DESEMPENHO DE INSTITUIÇÕES FINANCEIRAS 49 26 INDICADORES DE RISCO DE INSTITUIÇÃO FINANCEIRA CAMELS 50 261 Adequação de capital capital adequacy 51 262 Qualidade dos ativos asset quality 53 263 Gestão administração management 54 264 Lucro earnings 54 265 Liquidez liquidity 56 266 Sensibilidade ao risco de mercado sensitivity to market risk 57 267 Variáveis CAMELS 58 3 PROCEDIMENTOS METODOLÓGICOS 70 31 CLASSIFICAÇÃO DA PESQUISA 70 32 ETAPAS DA PESQUISA 71 4 DESCRIÇÃO DOS RESULTADOS 80 41 DEA Segundo Estágio resultados da regressão linear múltipla 85 411 DEA Segundo Estágio para abordagem Intermediação 85 412 DEA Segundo Estágio para abordagem Rentabilidade 86 42 SFA e os fatores explicativos para eficiência bancária 87 43 Análise consolidada DEA e SFA nas abordagens Intermediação e Rentabilidade 89 431 Pressupostos dos indicadores CAMELS para as Abordagens de Intermediação e Rentabilidade 92 432 Intermediação 94 433 Rentabilidade 97 5 DISCUSSÃO DOS RESULTADOS 100 6 CONCLUSÕES 110 REFERÊNCIAS BIBLIOGRÁFICAS 112 24 25 1 INTRODUÇÃO Houve nas últimas décadas uma mudança significativa dos negócios dos bancos para áreas não tradicionais em resposta às mudanças no ambiente competitivo às políticas regulatórias e à inovação nas habilidades e tecnologias de gestão WU HSIA HENG 2006 Além dos serviços tradicionais de empréstimo e recebimento de depósitos os bancos têm se empenhado cada vez mais na subscrição de títulos corretagem de seguros serviços de fundos mútuos serviços fiduciários e muitas outras atividades geradoras de taxas para renda extra além de estenderem suas fontes de passivos ao financiamento de atacado FRIOSI et al 2017 Embora o grau de diversificação tenha diminuído relativamente em alguns países após a crise financeira global de 2008 e 2009 MARTEL VAN RIXTEL MOTA 2012 ROENGPITYA TARASHEV TSATSARONIS 2014 os passivos de fontes não depositárias ainda são amplamente consideradas como determinantes importantes de seu desempenho MERGAERTS VENNET 2016 A principal fonte de receita do setor bancário consiste em empréstimos concedidos por bancos comerciais Portanto o risco de crédito é um dos riscos mais importantes enfrentados pelos bancos O risco de crédito é definido pelo Comitê de Supervisão Bancária da Basiléia BASEL COMMITTEE ON BANKING SUPERVISION 2000 como a probabilidade de perda parcial ou total do empréstimo em aberto devido ao não pagamento do empréstimo no prazo Um aumento no risco de crédito aumenta o custo marginal da dívida e do patrimônio líquido Correspondentemente o custo de captação do banco aumenta À medida que aumenta a exposição de um banco ao risco de crédito aumenta a tendência dos bancos de vivenciar uma crise financeira À medida em que as transações e produtos financeiros se tornam mais complexos e a integração e liberalização bancária tornam o setor mais competitivo há sobre os bancos uma forte pressão para administrar os riscos de forma eficaz sem negociar por menor retorno ou eficiência Muitos estudos examinaram a relação entre risco bancário e eficiência usando diferentes medidas ALTUNBAS et al 2007 BERGER MESTER 1997 KWAN EISENBEIS 1997 PRUTEANUPODPIERA WEILL SCHOBERT 2008 Também está em questão se qualquer relação causal entre risco e eficiência é temporária ou de longo prazo Staikouras Mamatzakis e KoutsomanoliFillippaki 2009 descobriram que para os bancos europeus o impacto da ineficiência no risco é pequeno e breve enquanto o efeito reverso do risco na ineficiência é negativo e significativo 26 O impacto da competição sobre a lucratividade no setor bancário foi documentado no paradigma tradicional de estruturacondutadesempenho que argumenta que em uma indústria mais concentrada com um nível mais baixo de competição as empresas tendem a conspirar entre si para obter lucros mais elevados Além disso há uma grande quantidade de literatura investigando o impacto da concorrência na lucratividade do setor bancário MAUDOS FERNANDEZ DE GUEVARA 2004 TAN FLOROS 2014 TAN 2016 A compreensão das razões que impactam no fator de produção do setor bancário tem um papel de destaque no contexto empresarial e consiste numa ferramenta eficaz de obtenção de lucros maiores índices de eficiência nos processos e flexibilidade estrutural aspectos de grande relevância à análise da competitividade e consolidação das instituições no mercado BRAGA JUNIOR SILVA SILVA 2014 Para Macoris Salgado Junior e Falsarella Junior 2015 ao avaliarse a eficiência das instituições financeiras é necessário realizar a comparação do desempenho dos bancos que executam eficientemente determinada atividade em relação aos demais Sendo assim é de grande importância utilizar os bancos considerados eficientes como benchmark para as demais instituições bancárias nacionais e internacionais Portanto compreender os fatores que determinam a eficiência bancária e considerado um fator chave para o desenvolvimento de melhores estratégias de gestão e políticas públicas STAUB SOUZA TABAK 2010 Verificase desta forma que a mensuração da eficiência bancária é um índice de desempenho relevante e que a relação entre as entradas ou seja os recursos empregados com as saídas que são os resultados obtidos pode ser considerada uma ferramenta eficiente de gestão para decisões de investimento avaliação de recursos humanos redução de custos e despesas entre outros fatores BATISTA 2009 São diversas as abordagens que podem realizar mensuração e a avaliação da eficiência bancária porém há destaque para duas abordagens distintas intermediação e rentabilidade MACORIS SALGADO JUNIOR FALSARELLA JUNIOR 2015 Conhecer os impactos de variáveis sobre as óticas de intermediação e rentabilidade é a base para se entender o atual momento do setor bancário Na intermediação são avaliadas e mensuradas as instituições com foco na capacidade de atuarem como intermediadores financeiros captando fundos de agentes superavitários e repassandoos na forma de crédito para agentes deficitários YANG 2009 Para definir entradas e saídas bancárias sob a ótica da abordagem de intermediação baseada em Berger e Mester 1997 2003 e muitos outros LOZANOVIVAS PASIOURAS 27 2010 GAGANIS PASIOURAS 2013 MALIKOV HUMBHAKAR SUN 2016 KHAN et al 2018 Essa abordagem assume que os bancos coletam fundos principalmente depósitos como entradas e os transformam em empréstimos ou outros ativos Na abordagem da rentabilidade são levadas em consideração a capacidade de a instituição maximizar as riquezas frente à redução de custos como forma de ampliar a sua rentabilidade e a sua lucratividade DRAKE HALL SIMPER 2006 FETHI PASIOURAS 2010 As abordagens de intermediação e rentabilidade foram utilizadas no presente estudo por constituíremse nas principais abordagens utilizadas para avaliar a eficiência MACORIS SALGADO JÚNIOR FALSARELLA JÚNIOR 2015 A maior parte da literatura reconhece a relevância das técnicas não paramétrica da Análise Envoltória de Dados ou Data Envelopment Analysis DEA e paramétrica da Análise da Fronteira Estocástica ou Stochastic Frontier Analysis SFA como principais métodos de análise de eficiência TABACK TECLES 2010 STAIKOURAS MAMATZAKIS KOUTSOMANOLIFILIPPAKI 2008 Assim sendo contribuise com a literatura ao analisar a eficiência em duas principais abordagens por meio da comparação de duas técnicas de mensuração de eficiência das instituições bancárias que compõem o sistema financeiro nacional composto por um sistema bancário reconhecidamente consolidado e bastante competitivo Esta dissertação está organizada em seis capítulos sendo o primeiro a introdução o segundo o referencial teórico seguidos pelos capítulos da metodologia descrição dos resultados discussão dos resultados e conclusão 11 JUSTIFICATIVA A eficiência do setor bancário tem sido amplamente estudada e discutida em diversos trabalhos nacionais e internacionais Tratase de um setor de elevada importância e com impactos no crescimento e estabilidade das economias nacionais e mundial Frente a análise de diferentes países identificouse um alto grau de ineficiência que afetou principalmente os bancos pertencentes a economias em desenvolvimento TABAK TECLES 2010 A crise iniciada em 2007 nos Estados Unidos e que se espalhou para as demais economias do mundo demonstrou o impacto que uma instituição financeira pode devido a sua característica de organização complexa e interconectada causar na estabilidade do sistema financeiro global com consequências negativas para a economia real Diante disto órgãos reguladores e formuladores de políticas definiram a necessidade de estipular medidas e modelos 28 para medir o risco destas instituições e identificar as instituições que poderiam prejudicar os mercados financeiros e levar a um colapso de todo o sistema financeiro BONGINI NIERI PELAGATTI 2015 Levando em consideração o papel de fundamental importância dos bancos nas economias nacionais as análises e mensurações de seu desempenho são extremamente relevantes FETHI PASIOURAS 2010 Frente a isto esta dissertação auxilia na resolução deste problema ao analisar medidas de eficiência bancária que funcionam como uma ferramenta de monitoramento destas instituições O Brasil possui relevante importância na América Latina devido a sua economia e consequentemente devido ao seu sistema bancário Frente a relevância do Brasil seu mercado de capitais ou mercado de títulos ainda é pouco desenvolvido ficando o mercado de crédito responsável por desempenhar um papel principal na economia brasileira Este fato é comum em economias em desenvolvimento e o entendimento do sistema bancário se faz necessário para uma efetiva condução de estratégias por parte dos gestores bem como de órgãos públicos STAUB SOUZA TABAK 2010 Este estudo compara todos os bancos que constam no segmento B1 do Banco Central do Brasil Bacen ao longo de 18 anos 12 PROBLEMATIZAÇÃO E OBJETIVOS DA PESQUISA Com a finalidade de analisar a eficiência bancária brasileira e os seus fatores explicativos este estudo objetivou investigar o problema de pesquisa que se segue Quais foram os fatores que influenciaram a eficiência dos bancos brasileiros entre os anos de 2000 e 2018 Portanto o objetivo deste trabalho é identificar os fatores que influenciaram a eficiência dos bancos brasileiros nas abordagens de intermediação e rentabilidade Os objetivos específicos são i propor um modelo de inputs e outputs para a mensuração de eficiência bancária ii avaliar a eficiência e os fatores explicativos sob a ótica de duas técnicas de mensuração de eficiência DEA e SFA 29 2 REFERENCIAL TEÓRICO Considerando a eficiência um aspecto intrínseco para a competitividade do setor bancário a revisão da literatura realizada foi de fundamental importância para embasar este estudo e melhor compreender os conceitos de eficiência que impactam no setor bem como compreender a utilização de duas técnicas uma paramétrica Análise Envoltória de Dados e outra não paramétrica Análise da Fronteira Estocástica no setor bancário brasileiro O objetivo deste referencial teórico é o de apresentar os conceitos de eficiência no setor bancário analisados através das técnicas DEA e SFA em concordância com a proposta de avaliar quais as variáveis impactam nos escores de eficiência dentre os bancos analisados 21 CONCEITO DE EFICIÊNCIA Eficiência é objeto de estudo de diversas áreas do conhecimento que visam extrair o máximo de produtos ou outputs através da utilização mínima possível de recursos ou inputs Tal fato indica se uma instituição é econômica ao utilizar os seus recursos FARRELL 1957 A apresentação deste conceito pode ser relacionada com a condição de equilíbrio que representa a existência da maximização da capacidade econômica sem o efeito de prejuízo de outros fatores e que pode ser representada e calculada através do uso de expressões matemáticas PARETO 1897 Através desta análise Pareto 1897 descreve ser possível comparar e mensurar a eficiência de um sistema econômico considerando a divisão em três grupos O primeiro considera a eficiência na produção como sendo a capacidade de se maximizar a produção sem que haja alterações na produção de demais produtos deste sistema A segundo especifica a eficiência na cesta de produtos como a capacidade de produção para atendimento das preferências dos agentes econômicos Por fim o terceiro considerado como eficiência nas trocas é determinado como a capacidade de realizar a distribuição dos produtos produzidos da forma mais eficiente Ao conceituar eficiência Farrel 1957 considera que em ambientes produtivos a eficiência é obtida ao analisar os dados da produção de acordo com o coeficiente de utilização dos recursos Dentre as formas de verificar a eficiência a razão entre quantidade de insumos e produtos é uma das mais utilizadas e é matematicamente expressa por 30 𝐸𝑓𝑖𝑐𝑖ê𝑛𝑐𝑖𝑎 ƒ𝑚𝑎𝑥𝐷𝑎𝑑𝑜𝑠𝑃𝑟𝑜𝑑𝑢𝑧𝑖𝑑𝑜𝑠 ƒ𝑚𝑖𝑛𝐼𝑛𝑠𝑢𝑚𝑜𝑠 𝑈𝑡𝑖𝑙𝑖𝑧𝑎𝑑𝑜𝑠 Ainda segundo Farrel 1957 produzir o máximo de produtos através de uma certa quantidade de entradas ou da mesma forma produzir determinado produto utilizando quantidades mínimas de insumo é considerado eficiência técnica Sendo assim quando uma unidade é analisada e considerada eficiente dentro dos parâmetros anteriores ela é considerada eficiente e presente na fronteira de produção Dentre as diversas técnicas utilizadas para se medir a eficiência duas delas possuem reconhecimento comprovado em diversas áreas de estudo uma paramétrica DEA e outra não paramétrica SFA Os modelos não paramétricos baseiamse na construção de fronteiras determinísticas onde encontramse as unidades com as melhores práticas dentre todas as demais unidades presentes na amostra e devido a este fato sua vantagem decorre do fato da não necessidade de apresentar hipóteses sobre a função de produção desenvolvida por Farrel 1957 bem como a não necessidade de identificar a ineficiência ODECK BRATHEN 2012 Os modelos paramétricos por sua vez procedem com a construção da fronteira considerando os impactos que diferentes cenários podem causar na função da produção de cada unidade ou seja a possibilidade de um mesmo modelo ter resultados diferentes quando analisado em cenários diferentes Além disso o fato desta técnica considerar o termo de erro aleatório e o termo de ineficiência em conjunto bem como o impacto de cenários diferentes na função produção traduzemse na vantagem desta técnica KUMBHAKAR LOVELL 2003 22 TÉCNICAS DE ANÁLISE DA EFICIÊNCIA É de grande importância após a definição da técnica de mensuração identificar a ótica e abordagem de eficiência de acordo basicamente com o objeto do estudo bem como os seus inputs e outputs Este estudo apresentará a partir deste momento a revisão sobre as duas principais técnicas utilizadas nesta pesquisa e que foram empregadas para a geração dos resultados preliminares que se encontram aqui demonstrados Esperase frente ao conteúdo deste referencial teórico sobre as técnicas não paramétricas e paramétricas que os resultados obtidos a partir da Análise Envoltória de Dados 31 apresentem médias menores quando comparadas as médias obtidas pela Análise da Fronteira Estocástica e por outro lado uma maior volatilidade BERGER MESTER 1997 221 Análise Envoltória de Dados Ao realizar um estudo que analisava a produção agrícola norte americana Farrel 1957 preconizou a mensuração da eficiência da produtividade a fronteira de eficiência e a DEA ao no desenvolvimento do estudo medir a produtividade a partir da observação da relação existente entre produto e insumo dentro de um processo produtivo específico Considerase uma unidade produtiva eficiente quando se obtém a máxima produção quando aplicado determinada quantidade de insumo para o processo O termo Data Envelopment Analysis DEA ficou conhecido oficialmente quando Charnes Cooper e Rhodes 1978 apresentaram a técnica em um estudo Devido a inicial dos nomes dos autores esta metodologia passou a ser denominada CCR Este modelo também é conhecido como CRS sigla em inglês para Constant Returns to Scale ou retorno constante de escalas Tratase de uma técnica de programação matemática que possibilita avaliar o grau de eficiência produtiva entre várias unidades tomadoras de decisão ou Decision Making Units DMUs ao se considerar os insumos disponíveis inputs e os produtos produzidos outputs Para Fethi e Pasiouras 2010 esta técnica foi identificada em 181 do total de 196 artigos revisados consolidandose desta forma como a técnica mais utilizada para avaliação deste segmento econômico Esta é uma técnica não paramétrica e por este motivo não objetiva otimizar um plano de regressão simples Ou seja a DEA visa a otimização individual de cada uma das observações resultando desta forma uma fronteira de eficiência que é definida como a representação do nível máximo de produção para determinado nível de insumos e baseiase no conceito de ParetoKoopmans GUERREIRO 2006 GRIGORIAN MANOLE 2006 A Figura 1 demonstra a representação gráfica da relação entre a quantidade de insumos e a quantidade produzida por DMU e é útil para ilustrar e exemplificar o conceito de fronteira 32 Figura 1 Fronteira de Eficiência Fonte Guerreiro 2006 Percebese por meio da figura 1 que uma das características da técnica DEA é a união das unidades eficientes por uma linha que resulta em uma superfície côncava onde as unidades ineficientes se encontram abaixo desta superfície Segundo Guerreiro 2006 o índice de eficiência das DMUs ineficientes é calculado através da projeção destas até a linha da fronteira Portanto é a representação numérica da distância da DMU ineficiente até a fronteira Esta projeção determina a utilização dos dois modelos existentes na literatura sendo o modelo orientado ao output e o modelo orientado ao input Becker Lunardi e Maçada 2003 descreveram estas duas vertentes sendo o modelo orientado ao output o que apresenta a orientação que mantem os inputs constantes e busca maximizar os resultados obtidos portanto o fato de produzir mais outputs através da mesma quantidade de inputs Já o modelo orientado ao input foi classificado como sendo o que apresenta a orientação onde há como foco a redução dos insumos mantendo os outputs constantes portanto produzir a mesma quantidade de outputs com a menor utilização de inputs Este estudo apresentou até o momento o conceito inicial da técnica DEA e sua evolução a partir do estudo de Farrel 1957 e a apresentação do conceito de DEA por Charnes 33 Cooper e Rhodes 1978 como uma técnica que constrói fronteiras de eficiência através dos inputs e outputs das DMUs analisadas Porém essa técnica de fronteiras pode apresentar retornos constantes ou variáveis de escala O modelo CCR já apresentado traça fronteiras que apresentam retornos de escala constantes e há também um segundo modelo a ser abordado onde a técnica DEA foi aprimorada pelo estudo de Banker Charnes e Cooper 1984 recebendo o nome de modelo BCC também devido as iniciais dos nomes dos autores O modelo BCC apresenta a construção das fronteiras através de retornos variáveis de escala O modelo CCR que surgiu em 1978 tem a representação matemática para as duas formas de orientação apresentadas na Figura 2 Figura 2 Minimização de Inputs e Maximização de Output CCR Fonte Adaptado de Guerreiro 2006 Onde Eƒƒ0 eficiência da DMU0 Uj vi pesos de outputs e inputs respectivamente Xik yjk inputs i e outputs j da DMUk Xi0 yj0 inputs i e outputs j da DMU0 34 A representação gráfica para a técnica DEA CCR para as duas formas de orientação está apresentada na Figura 3 Figura 3 Representação gráfica CCR Orientado a Inputs e Outputs Fonte Adaptado de Guerreiro 2006 Banker Charnes e Cooper 1984 alteraram a o modelo CCR em duas óticas principais sendo uma a eficiência técnica e a outra a eficiência de escala Ou seja passaram a considerar que as DMUs estão sujeitas a variações de escalas resultando em ganhos ou perdas ou seja retornos variáveis de escala sendo em inglês Variable Returns to Scale que dá origem a sigla VRS Para Freaza Madeiro e Gomes 2004 o modelo BCC identifica a correta utilização dos recursos de acordo com o grau de escala de operação da DMU observada Ou seja a eficiência da escala representa o quociente da eficiência do modelo BCC com a eficiência do modelo CCR O resultado é a medida da distância de uma DMU em análise para uma DMU modelo que é ficticiamente operada com uma escala de tamanho mais produtivo O cálculo de minimização de inputs e maximização de outputs está estampado na Figura 4 35 Figura 4 Minimização de Inputs e Maximização de Output BCC Fonte Adaptado Guerreiro 2006 Onde Eƒƒ0 eficiência da DMU0 Uj vi pesos de outputs e inputs respectivamente Xik yjk inputs i e outputs j da DMUk Xi0 yj0 inputs i e outputs j da DMU0 A representação gráfica para a técnica DEA BCC para as duas formas de orientação está apresentada na Figura 5 Figura 5 Representação gráfica BCC Orientado a Inputs e Outputs Fonte Adaptado Guerreiro 2006 36 Para consolidar a apresentação do conceito de DEA e seus modelos bem como suas possíveis orientações podese apresentar a Figura 6 que sintetiza todo o conteúdo analisado nesta seção O modelo representa as possibilidades de análise e utilização da técnica Ganhos de escala constantes direcionando para a técnica DEA CCR e as suas duas orientações ou seja orientado ao input ou orientado ao output Ganhos de escala variáveis direcionando para a técnica DEA BCC e suas duas orientações ou seja orientado ao input ou orientado ao output Figura 6 Classificação entre Ganhos de Escala e Orientação Fonte Jubran 2006 Uma das técnicas que se destacam para a avaliação bancária é a análise envoltória de dados ou DEA Para Fethi e Pasiouras 2010 esta técnica foi identificada em 181 do total de 196 artigos revisados consolidandose desta forma como a técnica mais utilizada para avaliação deste segmento econômico A técnica DEA possui certa vantagem sobre os tradicionais índices contábeis e financeiros pois pode considerar múltiplos inputs e múltiplos outputs o que é bastante válido 37 no setor bancário NIGMONOVI 2010 GRIGORIAM MANOLE 2006 THANASSOULIS BOUSSOFIANE DYSON 1996 O fato de se considerar múltiplos inputs e outputs é essencial quando se pretende observar estratégias que podem influenciar os resultados em diversos setores ou níveis de organizações Para avaliar o impacto de variáveis independentes na eficiência das DMUs mensurada ao utilizar a DEA aplicase uma Análise de Regressão para modelos que apresentam múltiplos estágios conforme estudos realizados por Despotis e Smirlis 2002 e Zhu 2003 Wanke 2012 discorre sobre as abordagens de DEA em dois estágios que demandam a análise de fatores externos que impactem nos resultados obtidos como escores de eficiência A Análise de Regressão neste estágio possui a mesma eficácia de outros métodos paramétricos aplicados com o objetivo de identificar o impacto das variáveis independentes na variável dependente WANKE 2012 222 Análise da Fronteira Estocástica A SFA surgiu dos estudos feitos por Aigner Lovell e Schimidt 1977 e dos estudos dos pesquisadores Meeusen e van de Broeck 1977 Tratase de uma técnica capaz de reduzir alguns gargalos deixados por técnicas determinísticas como por exemplo o estudo da ineficiência REIFSCHNEIDER STEVENSON 1991 Estes estudos surgiram como contraponto às técnicas não paramétricas com o objetivo de analisar os desvios da fronteira de produção e são capazes de estimar a ineficiência global das unidades ou firmas através da decomposição do termo de erro em dois elementos O primeiro destes elementos referese ao termo que contabiliza os efeitos aleatórios e o segundo referese à ineficiência tecnológica da firma Battese e Corra 1977 também realizaram estudos a fim de identificar estes termos A principal vantagem dos modelos de SFA reside no fato deste tipo de modelo conseguir identificar que os desvios em relação a fronteira de produção ou de custo podem ocorrer devido a ineficiência técnica dos produtores ou seja pela incapacidade de transformar uma quantidade x de insumos em uma quantidade y de produto de maneira tecnologicamente viável fazendo com que este produtor desviese da fronteira ou no reconhecimento de que estes desvios podem ocorrer devido a choques aleatórios ou interferências externas fora do controle dos produtores unidades ou firmas ZANINI 2004 38 Dessa forma podese definir que as firmas localizadas na fronteira determinística são classificadas como tecnicamente eficientes enquanto as firmas que estão abaixo desta fronteira de produção são consideradas tecnicamente ineficientes Greene 2002 sintetiza a proposta do modelo de SFA como um modelo de regressão estimado por verossimilhança que apresenta um distúrbio assimétrico e não normal Kumbhakar e Lovell 2003 classificam as fronteiras de produção as fronteiras de custo as fronteiras de receita e as fronteiras de lucro como estocásticas por apresentarem uma variação aleatória em seu ambiente operacional ou seja a ineficiência causa desvios unilaterais nas fronteiras estocásticas A fim de capturar os efeitos da variação aleatória que ocorre no ambiente operacional e identificar os efeitos das ineficiências técnicas ou alocativas utilizase o erro simétrico e erro unilaterais respectivamente A Figura 7 representa o padrão da SFA de Produção onde as firmas i e j estão inseridas A firma i está apresentada acima da fronteira determinística e a firma j abaixo da fronteira determinística Podese observar que o efeito da eficiência encontrase partindo da quantidade de produtos produzidos até o ponto em que a firma se encontra Também se observa na figura 7 que o efeito do ruído ou seja fatores que estão fora do controle das firmas estão representados e representam a distância do ponto representado pela firma até a fronteira determinística onde encontrase as firmas consideradas eficientes tecnicamente Figura 7 Fronteira Estocástica de Produção Fonte Pontes 2017 39 Segundo estudos de Aigner Lovell e Schimidt 1977 os modelos de SFA por considerarem a ineficiência técnica e os choques aleatórios também chamados de ruído podem ser representados da seguinte maneira 𝑦𝑖 𝑇𝐸𝑖 𝑓𝑥𝑖 𝛽 𝑒𝑣𝑖 Onde 𝑓𝑥𝑖 𝛽 𝑒𝑣𝑖 fronteira de produção estocástica 𝑣𝑖 variável aleatória irrestrita de sinal 𝑇𝐸𝑖 estimativa de eficiência da iésima firma produtora 𝛽 vetor de parâmetros da função de produção a ser estimada 0 𝑇𝐸𝑖 1 Em que a parte que representa a fronteira determinística ou seja 𝑓𝑥𝑖 𝛽 que é comum a todas as firmas e a parte que representada por 𝑒𝑣𝑖 a captura os efeitos de choques aleatórios que afetam a iésima firma produtora Sabese ainda que 𝑇𝐸𝑖 é uma medida de eficiência técnica representada por um valor entre zero e um e que se refere a medida entre a saída da iésima firma em relação à saída que poderia ser produzida por uma firma totalmente eficiente usando o mesmo vetor de entrada de modo que 𝑇𝐸𝑖 𝑦𝑖 𝑓𝑥𝑖 𝛽 𝑒𝑣𝑖 Relativo à forma da função 𝑓𝑥𝑖 𝛽 conforme estudo realizado por Coelli Rao e Battese 2005 esta deve preferencialmente possuir as seguintes características 1 Flexível permitindo aproximações diferenciais de primeira ordem no caso das formas flexíveis de primeiraordem ou de segunda ordem no caso das flexíveis de segundaordem sabendo que essas últimas podem aumentar o número de parâmetros a estimar e a possibilidade de multicolinearidade por exemplo 40 2 Linear em parâmetros tornandoas favoráveis às estimativas usando técnicas de regressão 3 Regular satisfazendo as condições de regularidade econométrica a depender se tratase de uma função produção custo etc Exemplo não negatividade essencialidade fraca convexidade e nãodecrescente em relação aos inputs para uma função de produção 4 Parcimoniosa sendo nas palavras de Coelli Rao e Battese 2005 a forma funcional mais simples que realize o trabalho adequadamente As principais funções determinísticas são retratadas pelo Quadro 1 Quadro 1 Principais funções determinísticas Fonte Adaptado de Coelli Rao e Battese 2005 Com a finalidade de desenvolver a última equação apresentada de uma maneira mais prática diversos autores em seus estudos sobre SFA consideraram a função 𝑓𝑥𝑖 𝛽 como 41 linear e aplicaram uma transformação logarítmica com a finalidade de isolar o termo 𝜇𝑖 como o erro que captura os efeitos da ineficiência técnica AIGNER LOVELL SCHMIDT 1977 MEEUSEN VAN DEN BROECK 1977 KUMBHAKAR GHOSH MCGUKIN 1991 REIFSCHNEIDER STEVENSON 1991 BATTESE COELLI 1995 KUMBHAKAR LOVELL 2003 COELLI RAO BATTESE 2005 Desta forma ao aplicarse uma transformação logarítmica a equação 𝑦𝑖 𝑇𝐸𝑖 𝑓𝑥𝑖 𝛽 temos o seguinte desenvolvimento 𝐿𝑛𝑦𝑖 𝐿𝑛𝑓𝑥𝑖 𝛽 𝐿𝑛𝑇𝐸𝑖 Sendo que 𝐿𝑛𝑦𝑖 𝐿𝑛𝑓𝑥𝑖 𝛽 𝑢𝑖 Onde 𝑢𝑖 é uma medida de ineficiência técnica 𝑢 𝐿𝑛𝑇𝐸𝑖 0 𝑢 1 𝑇𝐸𝑖 Então 𝑇𝐸𝑖 𝑒𝑢 Desta forma a equação acima pode ser reescrita conforme abaixo 𝑦𝑖 𝑓𝑥𝑖 𝛽 𝑒𝑣𝑖𝑢𝑖 Segundo Coelli Rao e Battese 2005 a análise de SFA em sua maioria é direcionada para a previsão dos efeitos da ineficiência A medida mais comum de eficiência técnica orientada para o produto é a relação entre a produção observada e a SFA correspondente 𝑇𝐸𝑖 𝑦𝑖 𝑦𝑖 𝑒𝑥𝑝𝛽0 𝛽𝑖𝑥𝑖 𝑣𝑖 𝑢𝑖 𝑒𝑥𝑝𝛽0 𝛽𝑖𝑥𝑖 𝑣𝑖 exp 𝑢𝑖 42 Onde 𝑇𝐸𝑖 estimativa de eficiência da iésima firma produtora 𝑦𝑖 quantidade de um único produto 𝛽 vetor de parâmetros da função de produção a ser estimada 𝑣𝑖 variável aleatória irrestrita de sinal 𝑢𝑖 é uma medida de ineficiência técnica Ainda de acordo com as contribuições de Coelli Rao e Battese 2005 o resultado obtido da fórmula acima mensura a eficiência técnica com seu valor entre zero e um e a relação entre a produção que poderia ser produzida por uma empresa totalmente eficiente e utilizando o mesmo input em relação a iésima empresa Devido à característica do modelo que apresenta um erro simétrico 𝑢𝑖 e uma variável aleatória não negativa 𝑣𝑖 há a necessidade de se estimar os parâmetros de análise com atenção a estes dois termos Devese levar em consideração que cada 𝑣𝑖 é distribuído independentemente de cada 𝑢𝑖 COELLI RAO BATTESE 2005 Aigner Lovell Schimidt 1977 através do modelo HalfNormal utilizaramse de duas premissas A premissa de que a variável aleatória 𝑣𝑖 é considerada uma iid independente e identicamente distribuída com distribuição normal que assume média zero e variância 𝜎𝑣 2 E a premissa que considera que a medida de ineficiência técnica 𝜇𝑖 também considerada uma iid com distribuição halfnormal e variância 𝜎𝑢 2 Segundo Coelli Rao e Battese 2005 outros modelos também podem ser utilizados em substituição ao modelo halfnormal como o modelo de fronteira normal truncada e o modelo gamma No contexto desta dissertação e considerando a geração de fronteiras por meio da técnica de SFA é importante identificar os impactos das variáveis independentes nos escores ou variável dependente Com isso a heterocedasticidade que ocorre quando a variância dos resíduos não é constante atua como ferramenta na identificação da ineficiência das firmas Brito 2016 aborda o comportamento heterocedástico do fator de ineficiência e o comportamento homocedástico do termo aleatório afirmando com isto que não considerar este fato na análise da fronteira de produção ocasiona em uma perda de eficiência dos estimadores Para o autor o modelo heterocedástico permite identificar quais fatores ou características de cada firma estimulam a eficiência ou a ineficiência 43 Para considerar esse aspecto conforme indicado por Danelon 2018 detalhase o termo 𝑢𝑖 das fórmulas anteriores de modo que 𝑢𝑖 𝛿𝑧𝑖 𝑤 Em que 𝑧𝑖 referese a cada uma das varíaveis que afetam a ineficiência técnica e 𝛿 é um vetor de parâmetros a ser determinado A utilização da heterocedasticidade foi considerada neste estudo a fim de mensurar os impactos que as variáveis independentes podem ocasionar nos escores gerados através da técnica de SFA Segundo Danelon 2018 outros fatores do ambiente de produção podem afetar os resultados da eficiência técnica tornandoos componentes importantes para a análise da produtividade 23 EFICIÊNCIA BANCÁRIA Os estudos de eficiência bancária podem ser divididos em dois níveis de acordo com as partes interessadas stakeholders Em um nível macro há evidências de que o crescimento econômico está positivamente relacionado à eficiência do setor bancário ABEDIFAR HASAN TARAZI 2016 BERGER et al 2004 Em um nível micro os estudos de eficiência fornecem informações de benchmarking que serão do interesse de gerentes de bancos e formuladores de políticas a fim de melhorar o desempenho dos bancos A eficiência bancária e a dinâmica relacionada estão intimamente ligadas à tomada de risco e capitalização do banco enquanto a competição cada vez maior no setor de serviços financeiros torna a eficiência uma consideração primária para os gestores bancários reguladores e formuladores de políticas FIORDELISI MARQUESIBANEZ MOLYNEUX 2011 Altunbas et al 2007 encontravam evidências em bancos europeus que analisaram que mostram relação positiva entre risco no nível de capital e liquidez e que a solidez financeira do setor influencia positivamente na redução da assunção de riscos bancários e dos níveis de capital De igual forma um efeito positivo da ineficiência na assunção de riscos foi encontrado em trabalho de Kwan e Eisenbeis 1997 que identificaram que o efeito positivo verificado da ineficiência sobre o nível de capital pode ser atribuído à pressão regulatória sobre as 44 instituições de baixo desempenho Os autores também constataram a relação direta entre a ineficiência e o crescimento dos empréstimos indicativo de que a eficiência operacional melhora a uma taxa decrescente à medida em que a taxa de crescimento dos empréstimos aumenta o que apoia a hipótese de que os gestores que buscam crescimento tendem a operar de forma ineficiente KWAN EISENBEIS 1997 Por meio da SFA Lensink Meesters e Naaborg 2008 analisaram 2095 bancos comerciais em 105 países no período de 1998 a 2003 e também descobriram que a propriedade estrangeira afeta negativamente a eficiência do banco efeito menos pronunciado em países com boa governança Seu trabalho também apurou que a qualidade das instituições no país de origem e a maior similaridade entre a qualidade institucional do país de origem e a do país anfitrião reduzem a ineficiência dos bancos estrangeiros LENSINK MEESTERS NAABORG 2008 Já Nigmonov 2010 utilizandose da DEA concluiu que houve diminuição dos níveis gerais de eficiência de bancos do Uzbequistão analisados no período de 2004 a 2006 tendo como principal fonte de ineficiência se a eficiência técnica que varia conforme o porte das instituições bancárias Yin 2021 em seu estudo que contempla um conjunto de dados abrangente cobrindo 148 países no período de 19952015 investigou a relação entre concorrência e eficiência no setor bancário As evidências mostram que a competição bancária é prejudicial à eficiência de custos O ambiente regulatório e institucional do banco no qual os bancos operam não apenas influencia a eficiência do banco mas também afeta o vínculo entre competição e eficiência Especificamente regulamentações restritivas sobre atividades bancárias e requisitos de capital rigorosos levam à ineficiência de custos mas supervisões e compartilhamento de informações eficazes dos registros de crédito são mais propícios à eficiência do banco Além disso este estudo conclui que o impacto adverso da concorrência sobre a eficiência pode ser mitigado ou mesmo revertido por regulamentações rigorosas supervisões bem implementadas eou mecanismos eficazes de compartilhamento de informações Para Duan e Zhang 2001 a pesquisa sobre a eficiência dos bancos está crescendo naturalmente e a literatura anterior enfatiza dois conceitos diferentes de eficiência econômica para medir a eficiência do banco eficiência de custo e eficiência de lucro A eficiência de custos é o critério mais utilizado em relação ao outro eficiência de lucro na literatura Berger 1995 encontra uma relação negativa entre eficiência de custos e empréstimos inadimplentes e ele também revela que a eficiência de custos pode ser um indicador importante do problema futuro dos bancos BERGER DEYOUNG 1997 FRIES TACI 2005 Recentemente Tabak Fazio e Cajueiro 2012 e Silva et al 2016 descobriram que os bancos grandes parecem superar os 45 bancos não grandes em termos de eficiência de custo e lucro Assaf et al 2019 examinam os efeitos da eficiência do banco em tempos normais sobre o desempenho do banco eles descobriram que a eficiência de custo durante tempos normais ajuda os bancos a reduzir o risco e aumentar a lucratividade durante as crises financeiras institucionais subsequentes enquanto a eficiência do lucro tem benefícios limitados Yin Yang e Lu 2020 discutem as associações entre globalização bancária e eficiência Eles descobriram que a entrada de bancos estrangeiros está associada a uma menor eficiência nos países anfitriões As instituições financeiras possuem elevada importância e geram diversos impactos a economia real YANG et al 2020 Por isso a mensuração da eficiência bancária deve assumir uma parcela das análises destas instituições tanto em tempos de crise financeira onde auxiliam na avaliação dos prováveis impactos do sofrimento de uma determinada instituição financeira na estabilidade do sistema financeiro quanto em tempos normais onde a utilização dessas ferramentas é crucial para calibrar instrumentos prudenciais como capital requisitos e prêmios de seguro de acordo com a contribuição relativa de diferentes instituições ao risco sistêmico GRAVELLE LI 2013 Para Shamshur e Weill 2019 a eficiência de custos dos bancos é uma medida ampla de desempenho bancário que tem sido frequentemente utilizada na literatura bancária empírica nas últimas duas décadas Ele mede a capacidade de um banco de operar com custos mais baixos comparando sua estrutura de custos com a de um banco com as melhores práticas Uma vasta literatura sobre eficiência bancária tem se concentrado em medir o nível e os determinantes da eficiência bancária em todo o mundo BERGER HASAN ZHOU 2009 FUJI MANAGI MATOUSEK 2014 GODDARD MOLYNEUX WILSON 2004 com o objetivo de melhorar o desempenho das instituições financeiras ALTUNBAS et al 2007 BONIN HASAN WACHTEL 2005 BERGER BONACCORSI DI PATTI 2006 BARTH CAPRIO LEVINE 2013 Para Thanassoulis 2003 existem motivações que levam ao estudo e a análise da eficiência Tais motivações podem ser descritas na forma de incentivos que levam a melhorias no processo produtivo Alguns dos principais incentivos descritos são focados na produção e em seu resultado considerando os ganhos com as taxas marginais de substituição de fatores de produção a redução de custos na utilização dos recursos avaliação da variação de produtividade entre outros 46 24 FERRAMENTAS DE AVALIAÇÃO DA EFICIÊNCIA BANCÁRIA OU MENSURAÇÃO DA EFICIÊNCIA BANCÁRIA A eficiência bancária é objeto de estudo relevante e uma das formas de mensurála baseiase na utilização de técnicas estatísticas que frente a cálculos predefinidos geram fronteiras de eficiência BERGER HUMPHREY 1997 Sabese que os escores de eficiência gerados por ambas as técnicas oferecem a possibilidade de críticas frente a técnica escolhida e por este motivo este trabalho utiliza conjuntamente ambas as técnicas para a análise de uma mesma base de dados com o intuito de comparar os resultados geradas e auxiliar no entendimento da eficiência bancária brasileira de 2000 a 2018 Através da definição de Cooper Seiford e Zhu 2011 a técnica de DEA desenvolvida por Charnes Cooper e Rhodes 1978 é uma das técnicas de maior relevância e reconhecimento para a mensuração de eficiência e aceita pela comunidade científica como modelo de processos operacionais A utilização desta técnica é amplamente empregada e encontrada na literatura como ferramenta para a mensuração de eficiência bancária devido ao seu caráter de técnica não paramétrica Esta característica a classifica como uma técnica mais flexível em relação as variáveis utilizadas PERICO REBELATTO SANTANA 2008 Outro fator de grande relevância é de que a utilização da ferramenta DEA ocorreu em praticamente todos os sistemas bancários internacionais FETHI PASIOURAS 2010 É possível identificar que a metodologia de análise empregada pela técnica DEA possui ampla utilização quando se trata de estudos que avaliam o sistema bancário O Quadro 2 sintetiza relevante literatura seminal acerca da utilização da DEA na avaliação do sistema bancário Quadro 2 Quadro sinótico de revisão bibliográfica sobre utilização de DEA na avaliação de eficiência do sistema bancário Autores Resumo do trabalho Paradi e Schaffnit 2004 Os autores avaliaram o desempenho de filiais de uma grande rede canadense de bancos utilizandose de DEA Dois modelos foram considerados no estudo um direcionado à observação do uso dos recursos e o outro direcionado à análise da gestão sênior da instituição 47 Mostafa 2009 O estudo investigou a eficiência de bancos árabes utilizando duas metodologias quantitativas DEA e redes neurais Os resultados indicam que a precisão da análise dos modelos preditivos é similar à de métodos estatísticos tradicionais O estudo destacou a importância econômica de se incentivar o aumento da eficiência no setor bancário árabe Yang et al 2010 O trabalho utilizouse de uma análise híbrida minimax e DEA aplicada a um estudo de caso realizado em quatorze filiais de um banco internacional de Manchester a fim de avaliar a eficiência de instituições bancárias Os resultados serviram ao apoio à tomada de decisão sobre definição de metas e alocação de recursos Siriopoulos e Tziogkidis 2010 O trabalho analisou a eficiência de bancos comerciais gregos no período de 1995 a 2003 por meio da análise DEA a duas abordagens utilizadas para medir eficiência Os resultados da pesquisa sugerem que o setor bancário grego operou com eficiência na média durante períodos de desestabilização Avkiran 2011 Este trabalho investiga a extensão a análise DEA de eficiência bancária está associada aos principais índices financeiros A utilização da DEA para análise de eficiência surgiu como combinação mais significativa e explica a variação NOPAT e ROE Luo Bi e Liang 2012 Neste trabalho foi proposto um novo método de escolha de variáveis DEA baseado na concepção de valor adicionado ao dinheiro Este método trouxe como vantagens maior objetividade por evitar a influência de fatores subjetivos subsequente aos cálculos e o fato de proporcionar aos gestores e pesquisadores a medida de variáveis e sua exata classificação O estudo analisou quatorze bancos comerciais chineses com testes estatísticos e de regressão satisfatórios Paradi Zhu e Edelstein 2012 O trabalho desenvolveu uma abordagem que utiliza a DEA como framework para identificar grupos gerenciais e grupos líderes em desempenho em cerca de mil filiais de um grande banco canadense As vantagens desta nova abordagem foram demonstradas por meio da comparação dos resultados obtidos com algoritmos tradicionais de clustering Chortareas Girardoni e Ventouri 2012 O trabalho investigou a dinâmica entre políticas regulatórias e de supervisão e vários aspectos de eficiência e desempenho de bancos comerciais de 22 países da União Europeia entre 2000 e 2008 O primeiro estágio do trabalho utilizou a DEA para análise de eficiência Foi ainda conduzida uma análise de sensibilidade sobre a robustez do modelo Os resultados demonstraram que o fortalecimento das restrições de capital e do poder de supervisão das operações melhora a eficiência bancária Fonte elaborado pelo autor 48 O método SFA desenvolveuse a partir dos estudos simultâneos de Aigner Lovell e Schmidt 1977 e dos estudos de Meeusene e Van Den Broeck 1977 Recentemente a quantidade de estudos sobre eficiência bancária através da utilização da técnica de SFA vem ganhando força dentre os estudos existentes e frente aos estudos que envolvem a DEA Trabalhos envolvendo eficiência bancária e SFA têm demonstrado a importância da técnica e do fato de se levar em consideração o termo de ineficiência como se observa na compilação de trabalhos apresentada no Quadro 3 Quadro 3 Quadro sinótico de revisão bibliográfica sobre utilização da Fronteira Estocástica para a análise da eficiência bancária Autores Resumo do trabalho Dong et al 2016 O trabalho examinou a eficiência de custo e lucratividade de quatro tipos de bancos comerciais chineses no período de 2002 a 2013 Os autores descobriram que as eficiências de custo e lucratividade dos bancos melhoraram ao longo dos anos e que as instituições são mais eficientes em custo que em lucratividade O gap de eficiência de lucro entre os bancos estrangeiros e os bancos domésticos aumentou no período de 2007 a 2013 Estrutura de ativos competição de mercado e porte são os principais determinantes da eficiência dos bancos chineses Boucinha Ribeiro e WeymanJones 2013 O artigo analisou analisa a tecnologia de produção dos bancos portugueses entre 1992 e 2006 período que engloba a adesão de Portugal à área do Euro Como resultado o estudo identificou que o avanço tecnológico deslocou a fronteira de custos para baixo ao longo do período considerado enquanto a distância em que os bancos operaram parece ter permanecido constante Notouse ainda que aumentos na produção em economias de escala também contribuíram para o aumento registrado na produtividade Jiang Yao e Feng 2013 O artigo combinou o efeito estático da propriedade e o efeito dinâmico da privatização com o desempenho de bancos chineses no período de 1995 a 2010 Foi observado um desempenho significativamente maior por intermediários privados bancos comerciais com ação conjunta com bancos comerciais municipais em relação aos bancos estatais Os bancos de capital aberto sujeitos a monitoramento do mercado apresentam melhor desempenho independentemente do status de propriedade A privatização de bancos melhorou o desempenho no que diz respeito ao ingresso de receitas e ganhos de eficiência no curto ou longo prazo Os resultados sugerem que os bancos chineses são muito mais eficientes na geração de receita de juros do que na captação de receita não financeira Tabak Fazio e Cajueiro 2012 Este artigo estudou o efeito da competição de bancos latino americanos na assunção de riscos e analisou se a capitalização e o porte da instituição interferem nessa relação Concluiuse que a competição afeta o risco de forma não linear a intensidade da 49 competição está relacionada à estabilidade o porte do banco explica a vantagem da competição enquanto a capitalização só é positiva para bancos maiores o índice de capital explica a vantagem da concorrência mais baixa Fiordelisi MarquesIbanez e Molyneux 2011 Foi avaliada a relação intertemporal entre eficiência capital e risco bancário em uma amostra de bancos comerciais europeus empregando várias definições de eficiência risco e capital Os resultados sugerem que a menor eficiência do banco em relação aos custos e receitas causa maior risco e que aumentos no capital do banco precedem melhorias na eficiência de custo Também descobriuse que bancos mais eficientes eventualmente se tornam mais capitalizados e que níveis de capital mais altos tendem a ter um efeito positivo sobre os níveis de eficiência Rossi Schwaiger e Winkler 2009 O artigo analisou como a diversificação dos bancos de acordo com o tamanho e o ramo de atividade afeta o risco a eficiência de custo e lucro e a capitalização bancária Foram analisados os grandes bancos comerciais austríacos durante os anos de 1997 a 2003 Foram testados vários tipos diferentes de hipóteses gerenciais formalizadas de acordo com uma versão modificada do modelo de Berger e DeYoung Descobriuse que embora a diversificação afete negativamente a eficiência de custos ela aumenta a eficiência dos lucros e reduz o risco realizado dos bancos Por fim a diversificação parece ter um impacto positivo na capitalização dos bancos Berger Hasan e Zhou 2009 O artigo analisou a eficiência dos bancos chineses no período de 1994 a 2003 As descobertas sugerem que os quatro grandes bancos estudados são de longe os menos eficientes os bancos estrangeiros são mais eficientes e a propriedade estrangeira minoritária está associada a uma eficiência significativamente melhorada Foram apresentadas verificações de robustez que corroboram os resultados Essas descobertas sugerem que a propriedade estrangeira minoritária dos quatro grandes bancos em via de privatização provavelmente melhorará seu desempenho de forma significativa Fu e Heffernan 2009 Este artigo investigou a relação entre a estrutura de mercado e o desempenho do sistema bancário chinês de 1985 a 2002 O modelo foi estendido para considerar questões como o impacto do tamanho propriedade do banco e se os quatro grandes bancos desfrutam de uma vida tranquila Em média a eficiência diminuiu significativamente e a maioria dos bancos estava operando abaixo dos níveis de eficiência de escala Não houve evidência para apoiar a hipótese da vida tranquila provavelmente porque os controles rígidos das taxas de juros impediram os bancos estaduais de obter lucros de monopólio Fonte Elaborado pelo autor 25 FATORES EXPLICATIVOS DO DESEMPENHO DE INSTITUIÇÕES FINANCEIRAS Algumas peculiaridades do sistema financeiro tornamno suscetível à uma combinação de riscos incomum aos demais players econômicos 50 Especialmente face à sua estrutura dinâmica e à complexa natureza do ambiente econômico em que operam os bancos enfrentam uma diversidade de riscos EKINCI POYRAZ 2019 Koch e Macdonald 2014 argumentam que os riscos enfrentados pelos bancos podem ser categorizados em seis risco de crédito risco de liquidez risco de mercado risco operacional risco nominal e riscos legais Cada um tem potencial de impactar negativamente a lucratividade o valor de mercado o passivo e o patrimônio líquido das instituições financeiras KOCH MACDONALD 2014 Como a principal fonte de receita do setor bancário consiste em empréstimos concedidos por bancos comerciais o risco de crédito defendem Ekinci e Poyraz 2019 é um dos mais importantes riscos enfrentados pelos bancos Risco de crédito assim definido pelo Comitê de Supervisão Bancária da Basiléia BASEL COMMITTEE ON BANKING SUPERVISION 2000 é a probabilidade de perda parcial ou total do empréstimo em aberto devido ao não pagamento do empréstimo no prazo Em tempos de crise a reação dos bancos menos capitalizados que compram títulos com rendimento mais baixo consiste no aumento dos títulos sobre a oferta de crédito em busca de mais liquidez e portanto segurança PEYDRÓ POLO SETTE 2021 O aumento do risco de crédito impacta no aumento do custo marginal da dívida e do patrimônio líquido e consequentemente no aumento do custo de captação da instituição e na tendência dos bancos de vivenciarem uma crise EKINCI POYRAZ 2019 A propósito disso De Brandt Lecarpentier e Pouvelle 2020 encontraram forte interação entre a liquidez do mercado e a liquidez do financiamento bancário durante os períodos de crise resultado impulsionado pelos bancos menos líquidos 26 INDICADORES DE RISCO DE INSTITUIÇÃO FINANCEIRA CAMELS Um sistema financeiro equilibrado e fortalecido deve ser obtido em conjunto a uma economia saudável e um setor bancário sólido e eficiente O sistema bancário desempenha um papel essencial na criação de ativos produtivos e geração de superávits As instituições bancárias utilizam estratégias e procedimentos para prevenir possíveis deteriorações early warning system que são projetados para gerar os ratings de solidez individual daquele banco JEELANBASA JEELANBASHA 2018 PIMENTA 2014 Diversos modelos de classificação de risco foram desenvolvidos a fim de funcionar como preditores de crises bancárias auxiliando na identificação das instituições cuja condição 51 merece atenção especial na supervisão de seus regimes de riscos Essas classificações são baseadas em avaliações subjetivas de vários aspectos do funcionamento de uma instituição bancária onde o examinador pode incluir outros fatores pertinentes à sua avaliação Geralmente resultados obtidos são compartilhados apenas com a administração do banco em questão e não são divulgadas publicamente SAHAJWALA VAN DEN BERGH 2000 Em 1979 o Federal Financial Institutions Examination Council FFIEC um comitê interno do Federal Reserve System criou o UFIRS Uniform Financial Institutions Rating System cujo objetivo era avaliar de maneira uniforme os riscos e a solidez das instituições bancárias por meio de análise dos critérios de supervisão e previsão de riscos das instituições financeiras de maneira individual A estratégia utilizada foi conhecida como rating CAMEL SAHAJWALA VAN DEN BERGH 2000 Inicialmente o acrônimo CAMEL possuía cinco componentes de avaliação porém com a mudança nos mercados globais políticas e procedimentos de supervisão em 1997 os órgãos norteamericanos viram a necessidade de incluir mais um indicador dessa forma surgiu o termo CAMELS O rating CAMELS avalia seis medidas de desempenho sendo C Adequação de capital A Qualidade dos ativos M Gestão Administração E Lucro e L Liquidez e S Sensibilidade ao risco de mercado SAHAJWALA VAN DEN BERGH 2000 261 Adequação de capital capital adequacy Conforme citado anteriormente o CAMELS avalia seis indicadores de maneira distinta O primeiro item avaliado é a adequação de capital representada pela letra C Capital adequacy Esse item envolve pontos essenciais de análise do capital como a relação entre o nível e a qualidade do capital com a condição financeira geral da instituição necessidade de capital emergente adicional característica e volume dos ativos problemáticos e sua adequação para as possíveis perdas com os empréstimos e arrendamentos balanço patrimonial detalhado contendo os ativos intangíveis riscos de mercado riscos de concentração e riscos associados à atividades não tradicionais riscos por atividades extrapatrimoniais perspectivas e planos de crescimento acessibilidade a mercados de capitais e outras fontes de capital FEDERAL RESERVE SYSTEM 1996 Para Dincer et al 2011 índices que compõem a adequação de capital demonstram a capacidade financeira de uma instituição em cumprir suas funções frente aos e riscos 52 envolvidos como o risco de crédito de mercado e risco operacional assim sendo estabelecem qual a parcela do capital é utilizado para suportar os ativos de risco dos bancos Pile e Paradi 2002 afirmam tratarse de um índice que pode revelar a possibilidade de falhas de uma instituição financeira em lidar com o risco Canbas et al 2005 afirmam ser um importante índice para mensurar a possibilidade de risco de default de uma instituição Mili et al 2017 defendem que para as autoridades reguladoras o problema é definir as ferramentas para controlar e fiscalizar a saúde financeira de um banco isso gerou o conceito de adequação de capital em meados da década de 1970 A expansão dos empréstimos dos bancos sem aumento paralelo de seu capital obrigou os reguladores a definir diversos procedimentos de controle e a lançar novas reformas para evitar a insolvência do setor bancário No contexto de proteção contra choques os requisitos de capital são a regulamentação prudencial mais frequentemente citada A importância do capital regulatório para garantir a estabilidade bancária tem motivado diversos pesquisadores a estudar os determinantes dos bancos de capital regulatório O exame das relações entre as classificações de crédito e a estrutura de capital dos bancos em um contexto de vários países é oportuno e importante para a lucratividade estabilidade e regulamentação dos bancos pelas seguintes razões Em primeiro lugar os ativos e passivos dos bancos são fundamentalmente diferentes daqueles das empresas não financeiras FAN et al 2012 e as informações embutidas em uma classificação de crédito constituem uma base de confiança e segurança dos clientes na qualidade dos ativos de um banco o que é importante para a estabilidade do banco MILLON THAKOR 1985 Em segundo lugar várias regulamentações relacionadas ao capital dos bancos e investimentos em títulos estão diretamente vinculadas às classificações de crédito ressaltando assim a importância das classificações para a lucratividade e estabilidade dos bancos KISGEN 2019 De fato Kisgen 2019 apropriadamente observa que os reguladores financeiros contam com classificações de crédito e níveis de classificação para decidir se permitem que determinados grupos de investidores como bancos invistam em títulos de outros bancos e em que medida os grupos de investidores atendem aos requisitos de capital específicos ao emprestar para outros bancos Terceiro a nova regulamentação de capital Basileia III impõe que os bancos tenham pelo menos 3 do capital em relação ao total de seus ativos Este novo requisito de índice de alavancagem coincide com o índice do Lehman Brothers em sua última demonstração de posição financeira antes de ir à falência ADMATI HELLWIG 2019 53 262 Qualidade dos ativos asset quality O segundo item qualidade dos ativos Asset quality avalia se há potencial risco de crédito associados às carteiras de empréstimos e investimentos tanto imóveis quanto outros ativos Nesse item são avaliados todos os riscos que afetam o valor ou a comercialização dos ativos incluindo os riscos de mercado de imagem operacionais e estratégicos FEDERAL RESERVE SYSTEM 1996 Este item demonstra o nível de risco dos ativos e a capacidade financeira bancária sendo utilizado na avaliação da condição financeira atual e futura ao auxiliar com a previsão de ameaças a lucratividade e a viabilidade da instituição financeira DINCER et al 2011 Para Curry Elmer e Fissel 2003 os índices de qualidade dos ativos são utilizados para identificar os ativos que podem gerar dificuldades financeiras as instituições bancárias Já Affes e HentatiKaffel 2017 indicam que estes índices demonstram quanto capital é utilizado para apoiar os ativos de risco de uma instituição bancária mensurando a alavancagem e a resistência a choques financeiros Para Beltrame Previtali e Sclip 2018 o aumento dos requisitos de capital impostos pelas autoridades reguladoras no âmbito da estrutura de Basileia ou adicionalmente exigidos para situações específicas contribuem para apoiar os bancos na absorção de perdas em empréstimos inadimplentes e de forma mais geral promover a resiliência do sistema bancário a potenciais choques negativos Apesar de sob o ponto de vista dos banqueiros levantar novo capital social ser caro a evidência empírica sugere uma relação inversa entre capital próprio e risco sistemático em todos os principais mercados bancários Ahamed 2017 afirma que a qualidade dos ativos é considerada uma das principais questões no caminho da solidez e saúde financeira de um banco a inadimplência em empréstimos e adiantamentos e a baixa qualidade dos ativos tem repercussão significativa para toda a economia de um país Embora haja evidências do efeito adverso da baixa qualidade dos ativos sobre a lucratividade há evidências reconhecidamente limitadas ou inexistentes sobre a ligação entre a diversificação da receita e a lucratividade em bancos com diferentes qualidades de ativos Na categoria de indicadores de qualidade de ativos considerase o índice de Nonperforming Loan NPL o índice de cobertura abaixo do padrão e os empréstimos depreciados não reservados Este grupo inclui informações úteis para analisar a credibilidade financeira dos bancos e suas exposições ao risco que são derivadas do negócio principal dos bancos ou seja suas atividades de empréstimo O índice de inadimplência é a proporção de 54 empréstimos inadimplentes brutos pelo para o total de empréstimos e contas a receber brutos com clientes BERGER KLAPPER TURKARISS 2009 Para Forgione e Migliardo 2018 o índice define empréstimos inadimplentes como empréstimos que estão vencidos por mutuários em dificuldades Como esse índice é uma medida do valor dos empréstimos duvidosos uma parte do índice de inadimplência provavelmente se tornará perdas no período subsequente Quanto menor a proporção melhor é a qualidade do ativo A este respeito a recente crise financeira degradou o desempenho financeiro das empresas e por sua vez expandiu a quantidade de empréstimos inadimplentes 263 Gestão administração management No item gestãoadministração representado pela letra M Management mensurase a capacidade dos gestores em identificar monitorar e controlar os riscos de atividades gerais da instituição a fim de garantir operação segura sólida eficiente e respeitando as leis e regimentos do país Os riscos de crédito de mercado de imagem conformidade e liquidez são itens avaliados nesse tópico mediante a postura e tomada de decisão do gestor FEDERAL RESERVE SYSTEM 1996 Para Dincer et al 2011 é um componente que abrange um grande número de variáveis como por exemplo a experiência da gerência o que torna difícil a mensuração quando comparado aos demais índices Assim como para Affes e HentatiKaffel 2017 que complementam o abordado anteriormente ao afirmar que este indicador abrange a qualidade dos gestores bem como da estrutura de gerenciamento das instituições financeiras Para DeYoung 1998 os resultados obtidos em seus estudos demonstram que a comum afirmação de que uma eficiência nos custos reflete a qualidade da gestão é verdadeira e traduz o fato de que uma instituição bem gerida utiliza os seus recursos de maneira mais eficiente 264 Lucro earnings O lucro earnings é medido não somente pela quantidade dos ganhos propriamente ditos mas também pelos fatores que podem influenciar a sustentabilidade dos ganhos como os riscos de créditos excessivos ou inadequados Os lucros podem ser influenciados pela dependência indevida de ganhos extraordinários eventos não recorrentes ou efeitos fiscais favoráveis FEDERAL RESERVE SYSTEM 1996 55 Tratase do índice mais utilizado nos estudos nacionais e internacionais e por retratar a rentabilidade das instituições financeiras possui uma relação inversamente proporcional a falhas e ineficiência AVKIRAN CAI 2012 Pennacchi e Santos 2021 citam trabalhos que reconheceram o foco dos bancos no ROE mas principalmente para criticar o ROE como uma meta de desempenho devido ao seu fracasso em contabilizar o risco de patrimônio Alessandri e Haldane 2009 e Drake 2001 mostram que para os bancos do Reino Unido a década de 1970 sinalizou um mar de mudança à medida que o ROE médio dos bancos saltou de cerca de 7 para cerca de 20 e atribuem isso ao menor capital e maior risco de ativos bancários Begenau e Stafford 2016 Eles mostram empiricamente que os bancos com baixo retorno sobre ativos ROA tentam manter um alto ROE usando maior alavancagem Além disso as medidas do valor contábil do mercado de ações do patrimônio do banco estão altamente correlacionadas com o ROE Juntando esses dois fatos eles sugerem que os bancos manipulam o ROE para cima por meio da alavancagem porque os investidores do mercado de ações ineficientemente se concentram no ROE Pennacchi e Santos 2021 argumentam que escolhida uma meta de ROE o modelo criado prevê que os bancos seriam especialmente resistentes à regulamentação póscrise financeira como Basileia III que gradualmente os forçou a aumentar seu capital Quando o valor do contrato é baixo padrões de capital mais elevados reduzem o excesso de patrimônio líquido de um banco levandoo a escolher o padrão de capital mínimo Assim um subproduto dos requisitos de capital mais elevados é a pressão para baixo no ROE do banco o que agora pode tornálo uma métrica de desempenho pior do que o EPS Consequentemente podemos esperar que os bancos diminuam a ênfase do ROE em favor do EPS Finalmente embora nosso artigo forneça uma teoria positiva para o foco dos bancos no ROE pesquisas futuras podem considerar se há uma justificativa normativa para que a regulamentação imponha uma métrica de desempenho diferente Ekinci e Poyraz 2019 analisam o impacto do risco de crédito no desempenho dos bancos estatais bancos privados e bancos estrangeiros que operavam na Turquia entre 2005 e 2017 a fim de comparar os bancos de acordo com sua estrutura de propriedade O Retorno sobre o Ativo ROA e o Retorno sobre o Patrimônio Líquido ROE foram usados como indicadores de desempenho financeiro enquanto o NonPerforming Loans NPLs foi usado como indicadores de risco de crédito Os resultados da estimativa mostraram que existe uma relação negativa entre o risco de crédito e o ROA bem como entre o risco de crédito e o ROE sugerindo então que existe uma relação entre a gestão do risco de crédito e a rentabilidade bancária 56 265 Liquidez liquidity No item liquidez liquidity devemse comparar os níveis atuais e as principais fontes de liquidez da instituição com as necessidades de financiamento levando em consideração a gestão de fundos em relação ao tamanho complexidade e perfil de risco da instituição Ou seja é importante garantir que o nível de liquidez seja suficiente para cumprir as obrigações financeiras e atender as necessidades bancárias da comunidade FEDERAL RESERVE SYSTEM 1996 Conforme observado pelo Comitê de Basileia BASEL COMMITTEE ON BANKING SUPERVISION 2000 alguns bancos faliram durante a fase inicial de estresse de liquidez da última crise financeira apesar de terem níveis de patrimônio robustos De acordo com o Comitê um dos motivos para essas falhas parece ter sido o fato de os bancos não administrarem sua liquidez de forma prudente Para Meriläinena e Junttila 2020 uma solução proeminente dos problemas de liquidez no setor bancário deflagrados pela crise financeira de 20082009 a estrutura de Basileia III introduz duas melhorias regulatórias o índice de financiamento estável líquido NSFR e o índice de cobertura de liquidez ICL O ICL visa garantir que os bancos tenham um estoque adequado de ativos líquidos para atender às suas necessidades de liquidez de curto prazo O NSFR por sua vez é um índice de liquidez estrutural que trata dos descasamentos de prazos entre ativos e passivos O novo regulamento visa reduzir os riscos nestas funções estabelecendo requisitos mínimos para os dois índices O regulamento tenta evitar problemas que possam ocorrer devido a choques de liquidez Esta visão prevê que a criação de liquidez deixa os bancos vulneráveis à insolvência quanto maior a criação de liquidez maior a probabilidade de falência do banco CANBAS CABUK KILIC 2005 Mas outra visão prevê que a criação de liquidez está associada negativamente à falência de bancos por dois motivos Primeiro a criação de liquidez é uma função primária e crucial das instituições bancárias Uma vez que a criação de liquidez como uma medida chave da produção bancária total contém informações sobre a capacidade de um banco de apoiar a macroeconomia e facilitar as transações entre os agentes econômicos por meio da transformação de maturidade a incapacidade do banco de gerenciar seu balanço e realizar uma criação de liquidez é provável que a função seja um sinal de alerta precoce de problemas FUNGACOVA PESSAROSSI WEILL 2013 No novo regime de Basileia III foi acordado pela primeira vez um quadro regulamentar de liquidez a nível internacional com a introdução de dois rácios de liquidez O 57 Índice de Cobertura de Liquidez e o Índice de Financiamento Estável Líquido NSFR buscam objetivos complementares que são promover a resiliência de curto prazo do perfil de liquidez dos bancos e manter um perfil de financiamento estável respectivamente Embora o último ainda não tenha entrado em vigor o primeiro já foi implementado progressivamente desde 2015 O ICL é projetado para garantir que os bancos resistam a um cenário de estresse de liquidez de 30 dias Mais recentemente o impacto da regulamentação de liquidez de Basileia III foi avaliado em termos de prevenção de risco de liquidez bem como seu impacto macroeconômico geral Teoricamente End e Kruidhof 2013 tentam simular as implicações sistêmicas do Liquidity Coverage Ratio No entanto empiricamente a maioria dos estudos enfoca proxies dos índices de liquidez regulatória como depósitos sobre empréstimos para Tabak e Tecles 2010 266 Sensibilidade ao risco de mercado sensitivity to market risk Por último localizase a sensibilidade ao risco de mercado sensitivity to market risk nesse item são avaliadas as alterações nas taxas de juros taxas de câmbio preços de commodities ou preços de ações Esses itens podem afetar diretamente os lucros eou o capital econômico bancário FEDERAL RESERVE SYSTEM 1996 Nesse modelo Dincer et al 2011 avaliam o nível de liquidez dos bancos sob a ótica da capacidade da instituição financeira em pagar seus ativos e de cobrir retiradas de dinheiro inesperadas Peydró Polo e Sette 2021 demonstram a importância dos títulos de valores mobiliários principalmente em tempos de crise com condições de política monetária mais amenas os bancos menos capitalizados reagem aumentando os títulos sobre a oferta de crédito com efeitos reais associados ao nível da empresa É importante ressaltar que os bancos menos capitalizados compram títulos com rendimento mais baixo mesmo dentro de títulos com pesos de risco de capital regulatório idênticos alcançando assim segurança e liquidez Os resultados sugerem que a liquidez e a capacidade de assumir riscos são os principais motores do comportamento dos bancos devido às condições de política monetária Ao contrário de tempos de crise em tempos de précrise normais quando os atritos financeiros são limitados à medida que as condições de política monetária se tornam mais suaves os bancos menos capitalizados não expandem os títulos em vez da oferta de crédito Além disso em tempos de 58 crise com condições monetárias mais amolecidas a substituição do crédito por títulos ajuda a restaurar a rentabilidade e o capital dos bancos menos capitalizados e esta melhoria no balanço ajuda a reiniciar a oferta de crédito com defasagem de um ano Affinito Albareto e Santioni 2016 exploram os determinantes do aumento dos títulos de dívida soberana nas carteiras dos bancos durante a crise geralmente adotou uma perspectiva macroeconômica Os autores adotam uma abordagem microeconômica e analisam os principais determinantes das compras banco a banco investigando as condições do balanço dos bancos italianos de 2007 a 2013 Os resultados mostram que as características específicas do balanço dos bancos são importantes e os bancos compram títulos do governo para sustentar suas condições financeiras A alta liquidez dos títulos do governo os altos rendimentos e a conveniência em termos de encargos de capital os tornam adequados para satisfazer as necessidades dos bancos em períodos de intensa demanda de liquidez declínio da lucratividade dos bancos e da qualidade dos empréstimos e aumento das restrições de capital 267 Variáveis CAMELS Os escores do rating CAMELS são avaliados separadamente onde escores mais altos representam desempenho mais fraco práticas inadequadas de gerenciamento de risco em relação ao tamanho complexidade e perfil de risco da instituição portanto maior grau de preocupação de supervisão SAHAJWALA VAN DEN BERGH 2000 Após a identificação dos riscos cabe à gestão conduzir e implementar mudanças no setor como a criação de produtos ou início de novas atividades Com a ampliação na gama de produtos financeiros oferecidos pelos bancos as instituições perceberam a necessidade de utilizar sistemas sólidos para o gerenciamento de riscos devido à aceleração no ritmo das transações que afetam o risco de mercado Dessa forma o sistema CAMELS revisado auxilia a administração a identificar mensurar monitorar e controlar os possíveis riscos financeiros PIMENTA 2014 Para avaliar as práticas de gerenciamento de riscos devese levar em consideração o tamanho a complexidade e o perfil de risco de uma instituição financeira Por isso é esperado que bancos menos complexos não utilizem sistemas elaborados e altamente formalizados de gerenciamento de riscos para receber classificações fortes ou satisfatórias de componentes ou compostos pois é esperado que eles gerenciem seus riscos de maneira eficaz SAHAJWALA VAN DEN BERGH 2000 PIMENTA 2014 59 Conforme citado anteriormente a literatura nos traz diferentes cálculos para mensurar individualmente os seis componentes do CAMELS SAHAJWALA VAN DEN BERGH 2000 PIMENTA 2014 podendo variar de acordo com o ponto de vista do examinador eou do investigador dos órgãos reguladores como a Office of the Comptroller of the Currency OCC Gabinete de Controladoria da Moeda e o FSI Financial Soundness Indicators Indicadores de Solidez Financeira criado pelo IMF International Monetary Fund Fundo Monetário Internacional De acordo com International Monetary Fund 2006 os seis itens de avaliação do CAMELS possuem suas particularidades individuais para cada categoria O indicador para a Adequação de Capital C deve ser calculado por meio dos valores de empréstimos inadimplentes Nonperforming loans menos os valores das provisões específicas dividido pelo o capital O resultado permite identificar precocemente os possíveis ativos problemáticos INTERNATIONAL MONETARY FUND 2006 Para Qualidade dos ativos A o cálculo do indicador envolve o valor de inadimplência da carteira dividido pelo valor total da carteira de crédito que deve incluir a inadimplência antes da dedução das provisões de liquidações duvidosas específicas Dessa forma é possível prever problemas relacionados com a qualidade dos ativos da carteira de crédito INTERNATIONAL MONETARY FUND 2006 O indicador de gestãoadministração M mensura a qualidade gerencial da instituição financeira ou seja é esperado que bancos com maior eficiência em sua gestão estejam menos propensos a riscos financeiros Para avaliar essa variável devemos utilizar o resultado de intermediação financeira e somar com as receitas com serviços esse valor deve ser dividido pela soma das despesas pessoais com as despesas administrativas INTERNATIONAL MONETARY FUND 2006 Os lucros E devem ser calculados por meio da soma entre o maior retorno de ativos ROA com os resultados de intermediação financeira e divididos pela receita bruta da intermediação financeira Os lucros indicam um bom desempenho financeiro para a instituição que consegue dessa forma investir na expansão se mantenha competitiva no mercado e aumente seu capital INTERNATIONAL MONETARY FUND 2006 O indicar para a liquidez L avalia a capacidade da gestão em balancear a lucratividade com os riscos de perda ou seja para calcular a liquidez é necessário dividir o total de ativos líquidos pelo total de ativos INTERNATIONAL MONETARY FUND 2006 Para a sensibilidade ao risco de mercado S último indicador a ser calculado há que se considerar a posição aberta de câmbio líquida dividida pelo capital O resultado obtido 60 permite com que o avaliador análise o poder da gestão em controlar o nível de exposição da instituição frente ao risco das taxas de juros flutuação do preço patrimonial e da volatilidade cambial INTERNATIONAL MONETARY FUND 2006 Diante da literatura nacional e internacional e a fim de embasar a escolha das fórmulas para cada uma das letras do acrônimo CAMELS utilizadas neste estudo e apresentadas na seção do método deste trabalho e ampliar o que foi posto anteriormente tendo apenas o FMI como base os Quadros 4 e 5 com os autores seus trabalhos e as fórmulas utilizadas por cada um deles apresenta o resultado da revisão bibliográfica deste tema e elucida a quantidade de índices que podem ser utilizados para cada novo autor A realização desta revisão sustenta a escolha das fórmulas ou métodos de mensuração que são utilizados neste trabalho e apresenta contribuição para a literatura e para futuros estudos 61 Quadro 4 Literatura sobre variáveis representativas do acrônimo CAMELS C A M C Capital A Asset M Management Literatura Empréstimos Vencidos NPLs Provisão para Créditos de Liquidação Duvidosa Capital Empréstimos Vencidos NPLs Crédito Total Resultado de Intermediação Financeira Receitas com Serviços Despesas Pessoais Despesas Administrativas IMF 2006 Empréstimos Vencidos NPLs Provisão para Créditos de Liquidação Duvidosa Capital Empréstimos Vencidos NPLs Crédito Total Resultado de Intermediação Financeira Receitas com Serviços Despesas Pessoais Despesas Administrativas PIMENTA 2014 Capital Próprio Ativo Total Empréstimos Vencidos NPLs Crédito Total Despesas Operacionais Total de Receitas GOMES 2012 Ativo Total Capital Próprio ROA Lucro Líquido Ativo Total x 100 Capital Tier 12 e3 Ativos Ponderados pelo Risco Provisões para Empréstimos Vencidos Empréstimos Vencidos NPLs ROE Lucro Líquido Patrimônio Líquido x 100 Patrimônio Líquido Títulos e Valores Mobiliários e Instrumentos Financeiros Derivativos Relações Interfinanceiras Relações Interdependências Operações de Crédito Operações de Arrendamento Mercantil Outros Créditos Outros Valores e Bens Provisões para Operações de Crédito Total da Carteira Despesas de CaptaçãoCirculante e Exigível de Longo Prazo ROSA GARTNER 2018 Créditos vencidos total das operações de crédito X KORONTAI 2016 62 C Capital A Asset M Management Literatura Capital total regulamentar para cobertura dos ativos ponderados pelo risco Créditos vencidos líquidos de provisão capital Capital nível 1 para cobertura de ativos ponderados pelo risco Distribuição Setorial dos créditos total das operações de crédito X X X GODOI et al 2016 Operações de Crédito e Arrendamento Mercantil Outros Créditos Patrimônio Líquido Imobilizado de Arrendamento Permanente Ajustado x 100 Patrimônio Líquido X MARCUSSO 2017 Capital de Terceiros Patrimônio Líquido Operações de Crédito e Arrendamento Mercantil Outros Créditos Ativo Total Relações Interdependência Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos Patrimônio Líquido Tier I Tier II Ativos Ponderados pelo Risco Total de Empréstimos Vencidos Provisões Empréstimos Total Despesas Administrativas Vendas CHRISTOPOULOS MYLONAKIS DIKTAPANIDIS 2011 Patrimônio Líquido Ativo Total Empréstimos em Inadimplência Empréstimos Bruto Despesas Operacionais Ativo Total ROMAN SARGU 2013 63 C Capital A Asset M Management Literatura Loan Loss Provisions Net Interest Revenues Despesas com Juros Depósitos Total loans total asset Patrimônio líquido Ativos Sujeito a Risco de Crédito Risco de Mercado Risco Operacional Ativos Financeiros Líquidos Ativo Total Despesa de Juros Despesa Total DINCER et al 2011 Patrimônio Líquido Ativo Total Empréstimo Total e Recebíveis Ativo Total Receita de Juros Receita Total Patrimônio Líquido Depósitos Fundos Não Depósitos Ativo Permanente Ativo Total Renda Total Despesa Total Patrimônio Líquido Ativo Total Empréstimo Não RecebidoAtivo Total Despesa com PessoalDespesa Operacional Total MAGRO MICHELS SILVA 2017 Empréstimo Não Recebido Crédito Bruto Despesas GeraisAtivo Total Patrimônio Líquido Empréstimo Total Credito LíquidoAtivo Total Receita totalAtivo Total Provisão de PerdasAtivo Total Total de Despesas OperacionaisReceita Total Crédito TotalPatrimônio Líquido Total Equity Total Assets Nonperforming loanstotal assets X AFFES HENTATI KAFFEL 2017 Total Equity Total Loans Nonperforming loansgross loans 64 C Capital A Asset M Management Literatura Loan loss reservestotal assets Loan loss reservesgross loans Patrimônio Líquido Ativo Total Empréstimos Vencidos NPLs Empréstimo Total Despesa não Financeira Receita Financeira Líquida Receita Financeira AVKIRAN LIN 2012 Empréstimos Vencidos NPLs Ativo Total Despesa com PessoalMédia dos Ativos Provisões para Empréstimos Ativo Total Custos Receita Provisões para Empréstimos Empréstimo Total Fonte Elaborado pelo autor 65 Quadro 5 Literatura sobre variáveis representativas do acrônimo CAMELS E L S E Earning L Liquidity S Sensitivity to Market Risk Literatura ROA Lucro Líquido Ativo Total x 100 Total de Ativos Líquidos Total de Ativos Posição Aberta de Câmbio Líquida Capital IMF 2006 Resultado de Intermediação Financeira Receita Bruta de Intermediação Financeira ROA Lucro Líquido Ativo Total x 100 Total de Ativos Líquidos Total de Ativos Posição Aberta de Câmbio Líquida Capital PIMENTA 2014 Resultado de Intermediação Financeira Receita Bruta de Intermediação Financeira X Crédito a Clientes Líquido Depósitos de Clientes Value at Risk VaR GOMES 2012 Disponibilidades Crédito a Instituições Financeiras Total de Ativos Receitas Operacionais Despesas Operacionais Circulante e Realizável de Longo Prazo Permanente Disponibilidades Aplicações Interfinanceiras de Liquidez Índice do Segmento Financeiro de Valores Mobiliários Livres Depósitos Obrigações por Operações Compromissadas X ROSA GARTNER 2018 ROA Lucro Líquido Ativo Total x 100 Ativos Líquidos Ativo Total Posição Líquida em Moeda Estrangeira Capital KORONTAI 2016 ROE Lucro Líquido Patrimônio Líquido x 100 Margem Financeira Resultado de Intermediação Receitas Ativos Líquidos Dívidas de Curto Prazo Despesas Administrativas Receitas ROE Lucro Líquido Patrimônio Líquido x 100 X Passivo Circulante Passivo Exigível à Longo Prazo Patrimônio Líquido GODOI et al 2016 66 E Earning L Liquidity S Sensitivity to Market Risk Literatura Lucro Líquido Receita Líquida da Intermediação Financeira Dívidas com Vencimento a Longo Prazo superiores há 1 ano Patrimônio Líquido Número de ações X Preço das ações Preço de Mercado da Ação Lucro Contábil por Ação Despesas de Intermediação Financeira de Captações no Mercado Despesas Empréstimos e Repasses Despesas Arrendamento Mercantil Despesas de Operações de Vendas ou Transferências de Ativos Financeiros Previsão para CL Ativo Total Ativo Permanente Relações Interdependências Imobilizado de Arrendamento Disponibilidades Aplicações Interfinanceiras Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos Outras Obrigações e Exercícios Futuros Obrigações por Empréstimos e Repasses X MARCUSSO 2017 Atividade Bancária x 100 Ativo Total Ativo Permanente Relações Interdependência Imobilizado de Arrendamento Aplicações Financeiras Líquidas Depósitos Interfinanceiros Depósitos a Prazo ROA Lucro Líquido x 100 Ativo Total Depósitos à Vista Depósitos de Poupança Passivo Exigível x100 ROE Lucro Líquido x 100 Patrimônio Líquido Depósito Total Depósito Total Recursos de Aceites Total Obrigações por Empréstimos e Repasses Patrimônio Líquido Receita Operações Arrendamento Mercantil Receitas Operações Vendas ou Transferência de Ativos Financeiros Receita de Intermediação financeira Operação de Crédito e Arrendamento Mercantil Receitas Operações com TVM Receitas Operações com Instrumentos Financeiros Derivativos Receitas Operações de Câmbio Receitas Aplicações Compulsórias Outras Receitas ou Despesas Operacionais Receitas de Prestação de Serviços Outras Receitas ou Despesas Operacionais de Renda e Rendas de Tarifas Bancárias Ativo Total Relações de Interdependência Permanente 67 E Earning L Liquidity S Sensitivity to Market Risk Literatura Imobilizado de Arrendamento Despesas Intermediação Financeira e Captações no Mercado Despesas Empréstimos e Repasses Despesas Arrendamento Mercantil Despesas Operações Câmbio Provisão para CL Despesas Operações de Vendas ou Transferências de Ativos Financeiros Ativo Total Permanente Imobilizado de Arrendamento Relações Interdependências x100 ROA Lucro Líquido Ativo Total x 100 Empréstimos Total Depósitos Total Total Títulos Valores Mobiliários Ativo Total CHRISTOPOULOS MYLONAKIS DIKTAPANIDIS 2011 ROE Lucro Líquido Patrimônio Líquido x 100 Cativo Circulante Ativo Total ROA Lucro Líquido Ativo Total x 100 Ativo Total Depósitos Líquidos e Financiamento de Curto Prazo Ativo Total Ativo Total do Setor ROMAN SARGU 2013 ROE Lucro Líquido Patrimônio Líquido x 100 Empréstimos Depósitos Líquidos e Financiamento de Curto Prazo Lucro Líquido Ativo Total Ativo LíquidoAtivo Total Ativo Total Ativo do Setor DINCER et al 2011 Lucro Líquido Patrimônio Líquido Ativo Líquido Passivo Curto Prazo Total de Empréstimos e Recebíveis Setor Empréstimos e Recebíveis Ativo Líquido Depósito Total Fundos Não Depósito Depósito Total Depósitos Setoriais ROE Sobra LíquidaPatrimônio Líquido Ativo LíquidoAtivo Total Valores Mobiliários em CarteiraAtivo Total MAGRO MICHELS SILVA 2017 ROA Sobra LíquidaAtivo Total Deposito TotalAtivo Total Receita TotalReceita Operacional Depósito TotalPatrimônio Líquido Valores Mobiliários em CarteiraEmpréstimos Total Provisão para Credito de Difícil LiquidaçãoAtivo Total Crédito TotalDepósito total 68 E Earning L Liquidity S Sensitivity to Market Risk Literatura Financiamento a Curto PrazoPassivo Total ROA Net Incometotal assets Total loanstotal customer deposits X AFFES HENTATI KAFFEL 2017 ROE Net IncomeTotal Equity Total customer depositstotal assets Lucro Líquido Ativo Total Compras Líquidas de Fundos Federais Ativo total Títulos Negociados Ativo Total AVKIRAN LIN 2012 Ativos Estrangeiros Passivos Estrangeiros Ativos Líquidos Ativo Total Receita Financeira Líquida Ativo Médio Fonte Elaborado pelo autor 69 Baseado no conceito de eficiência e sua aplicação no setor bancário bem como no embasamento das técnicas de mensuração de eficiência sendo uma paramétrica Fronteira Estocástica e outra não paramétrica Análise Envoltória de Dados além do estudo do acrônimo CAMELS para a mensuração de riscos este estudo foi estruturado de modo a analisar o impacto de cada um destes itens no setor bancário brasileiro ao longo dos anos 2000 a 2018 trazendo novo conhecimento a respeito das técnicas aplicadas em conjunto e do sistema de early warnings CAMELS aplicado na identificação de impactos nas variáveis dependentes deste estudo 70 3 PROCEDIMENTOS METODOLÓGICOS Nesta seção serão apresentados a caracterização da pesquisa e os procedimentos metodológicos com as etapas necessárias para se atingir os objetivos propostos 31 CLASSIFICAÇÃO DA PESQUISA Este estudo utilizou a abordagem metodológica quantitativa com o intuito de empregar métodos estatísticos e não estatísticos para verificar o relacionamento entre as variáveis analisadas após o levantamento da base de dados utilizada A abordagem quantitativa procura quantificar os dados e busca encontrar uma evidência conclusiva É baseada em amostras grandes e representativas e realiza a aplicação e análise estatística Os resultados obtidos de uma pesquisa qualitativa podem ser tratados como conclusivos e utilizados para recomendar um curso de ação final MALHOTRA et al 2005 A pesquisa quantitativa se centra na objetividade Influenciada pelo positivismo considera que a realidade só pode ser compreendida com base na análise de dados brutos recolhidos com o auxílio de instrumentos padronizados e neutros A pesquisa quantitativa recorre à linguagem matemática para descrever as causas de um fenômeno as relações entre variáveis etc FONSECA 2002 p 36 Uma pesquisa além da definição de sua abordagem metodológica também deve ser classificada quanto aos seus objetivos e procedimentos de investigação Este projeto sob a ótica de seus objetivos pode ser classificado como descritivo uma vez que procura especificar as propriedades as características e os perfis importantes de pessoas grupos comunidades ou qualquer outro fenômeno que se submeta a análise Eles medem avaliam ou coletam dados sobre diversos aspectos dimensões ou componentes do fenômeno a ser pesquisado Do ponto de vista científico descrever é coletar dados para pesquisadores quantitativos medir para os qualitativos coletar informações SAMPIERI COLLADO LUCIO 2006 p 101 Vergara 2006 considera que a pesquisa descritiva possui o objetivo de descrição das características de populações ou fenômenos e o estabelecimento de relações entre variáveis 71 Finalmente quanto aos procedimentos de investigação este projeto pode ser classificado como bibliográfico e documental pois as informações necessárias para sua elaboração são obtidas de fontes secundárias desde periódicos e artigos até a base de dados e relatórios divulgados em órgãos de regulação e controle Segundo Vergara 2006 a investigação documental é realizada em documentos mantidos no interior de órgãos públicos ou privados Para Gil 1999 a pesquisa bibliográfica utilizase principalmente das contribuições da comunidade científica sobre determinado tema de estudo já a pesquisa documental utiliza se de materiais que ainda não receberam um tratamento analítico ou que podem ser utilizados ou moldados de acordo com os objetivos da pesquisa A partir deste delineamento tornase possível atingir os objetivos propostos nesta dissertação de maneira a garantir que as normas de pesquisa sejam respeitadas 32 ETAPAS DA PESQUISA Para a realização da presente pesquisa foram utilizados os Balanços Patrimoniais e os Demonstrativos de Resultado do Exercício de cada banco consolidados e referentes aos anos de interesse desta pesquisa coletados no site do Banco Central do Brasil A base de dados disponível no site do BACEN iniciase em 2000 e possui dados até os fechamentos anteriores a esta pesquisa Esta base de dados de caráter público é amplamente utilizada no sistema financeiro e por pesquisadores Após o levantamento das informações financeiras através do Balanço Patrimonial e Demonstrativo de Resultado de cada banco foram selecionadas as variáveis analisadas em cada uma das três abordagens que orientam este estudo A metaanálise realizada por Macoris Salgado Junior e Falsarella Junior 2015 obteve como resultado as variáveis mais utilizadas pelas abordagens nos estudos envolvendo DEA e serviu de base para a escolha dos inputs e outputs a serem utilizados neste projeto de pesquisa tanto para a técnica DEA quanto para a técnica SFA conforme exposto na Tabela 1 72 Tabela 1 Inputs e Outputs de Macoris Salgado Junior e Falsarella Junior 2015 INTERMEDIAÇÃO RENTABILIDADE Inputs Depósitos Número de Pessoal Despesas com Pessoal Despesas Operacionais Despesas Operacionais não relacionadas a Pessoal Ativos Patrimônio Líquido Passivo Outputs Empréstimos Investimentos e Outros Ativos Realizáveis ROE ROI Lucro Fonte Adaptado Macoris Salgado Junior e Falsarella Junior 2015 O trabalho de Macoris Salgado Junior e Falsarella Junior 2015 foi adaptado à realidade atual das instituições financeiras a base de dados do Banco Central do Brasil e as necessidades de obtenção de escores que demonstrem de maneira mais eficaz a eficiência em cada uma das abordagens para as instituições financeiras nacionais O resultado desta adaptação e aprimoramento está demonstrado na Tabela 2 Tabela 2 Inputs e Outputs utilizados em DEA e SFA com múltiplos inputs e outputs INTERMEDIAÇÃO RENTABILIDADE Input Captações Ativo Total Input Despesas de Pessoal Patrimônio Líquido Output Aplicações Interfinanceiras Lucro Líquido Output Operações de Crédito e Arrendamento Mercantil Total Fonte Elaborado pelo autor A tabela 2 demonstra as variáveis utilizadas como inputs e outputs para a técnica DEA e SFA quando considerada a utilização de múltiplos inputs e múltiplos outputs e que foi desenvolvida em estudos anteriores a este que se apresenta A fim de se utilizar as duas técnicas em conjunto DEA e SFA este estudo realizou uma adaptação aos inputs e outputs apresentados na tabela anterior com o objetivo de atender também trabalhos que utilizem um único output face à necessidade do grupo de trabalho de rodar um modelo mais simples posto que se trata do primeiro estudo do grupo com a técnica SFA De fato está sendo apresentado modelo simplificado de análise de eficiência com um único output que considerou 73 i a não utilização de Lucro Líquido ii a utilização de Despesas Administrativas nos inputs da Abordagem de Intermediação em virtude da facilidade tecnológica iii a revisão dos inputs e outputs da Abordagem de Produção vez que tal indicador contempla praticamente todas as despesas fixas da instituição porém apenas uma parte da receita vez que as receitas não relacionadas a juros abrangem principalmente a receita com títulos e câmbio iv a revisão da Abordagem de Rentabilidade vez que Patrimônio Líquido contempla o patrimônio líquido efetivamente estampado no passivo o que resulta numa sobreposição de valores entre esses inputs por isso a utilização apenas de Ativo como input e Receita como output v que Receita de Serviços é muito similar a Receita com Juros vez que existem reciprocidades nas operações de crédito o que deve ser considerado como custo ou seja grande parte da Receita de Serviços advém diretamente dos juros O resultado de referidas adaptações é apresentado na Tabela 3 Tabela 3 Inputs e Outputs utilizados em DEA e SFA com um único output INTERMEDIAÇÃO RENTABILIDADE Input Captações Ativo Total Input Despesas Administrativas e Despesa de Pessoal Soma Despesas de Intermediação Financeira Despesas de Pessoal Despesas Administrativas Despesas Tributárias e Outras Despesas Operacionais Output Soma de Aplicações Interfinanceiras de Liquidez e Operação de Crédito e Arrendamento Mercantil Receitas Fonte Elaborada pelo autor Como primeiro estágio desta etapa os inputs e outputs acima foram aplicados aos bancos selecionados na categoria B1 A categoria B1 de acordo com a classificação de Tipo de Consolidado Bancário do Banco Central do Brasil compreende os bancos comerciais bancos múltiplos com carteira comercial ou caixas econômicas 74 Os bancos que apresentaram prejuízos no ano ou inputs e outputs sem informações foram excluídos chegandose dessa forma a amostra final analisada Os números negativos não foram considerados devido a uma restrição do modelo DEA Além disso os outliers que são unidades que apresentam desvios significativos em relação ao comportamento médio frente as demais DMUs ou unidades foram retirados da base de dados antes da geração da fronteira através dos softwares para as técnicas DEA e SFA Posteriormente os bancos foram divididos pelo porte pela estrutura de capital e pelo país de origem do controlador Para a classificação de porte que considerou os ativos totais de cada banco os bancos foram classificados em grande porte médio porte e pequeno porte Na sequência os bancos foram analisados frente aos escores obtidos por ambas as técnicas DEA e SFA para as três abordagens já apresentadas nesta seção As séries temporais contendo o Balanço Patrimonial e Demonstrativo de Resultado dos bancos analisados foram corrigidas pela inflação do período para representar fielmente as evoluções nos escores frente ao tempo Dessa forma tornouse possível a comparação efetiva dos bancos através da homogeneidade dos valores financeiros e consequentemente dos escores da técnica DEA e SFA Após a geração dos escores para ambas as técnicas gráficos contendo os escores para cada uma das abordagens bem como para cada uma das técnicas foram formulados para que movimentos relevantes fossem observados de forma visual Utilizouse para isto as medianas dos escores Após o primeiro estágio caracterizado pela aplicação das técnicas DEA e SFA e geração dos escores bem como a fronteira de eficiência a fim de buscar um maior entendimento e profundidade sobre os resultados obtidos aplicouse o segundo estágio baseado em um modelo de regressão linear múltipla Hair et al 2005 define a regressão linear múltipla como a técnica mais utilizada e mais versátil e que deve ser aplicada quando há apenas uma única variável dependente e duas ou mais variáveis independentes relacionadas a ela Esta dissertação utiliza a análise de regressão múltipla cujo objetivo é utilizar técnicas estatísticas para construir modelos que descrevem as relações entre várias variáveis explicativas de um determinado processo Proposto por Galton 1885 o conceito de regressão teve seu desenvolvimento através dos anos e por diversos autores Neter e Wasserman 1974 seguidos por Draper e Smith 1981 desenvolveram os estudos anteriores e auxiliaram na formulação que representa a regressão ao considerar novas aplicabilidades e conceitos Bowerman e OConnel 1990 75 também foram autores importantes neste desenvolvimento a partir do estudo realizado sobre modelos lineares de estatística A fórmula utilizada da equação de Regressão Linear Múltipla é comumente apresentada como 𝑌 𝛽0 𝛽1𝑋1 𝛽𝑝𝑋𝑝 𝑒 Onde Y Variável dependente X Variável independente 𝛽 coeficiente associado a variável e erro que apresenta distribuição normal com média zero e variância 𝜎2 A apresentação dos conceitos de Regressão Linear Múltipla se faz necessária neste estudo uma vez que será utilizada como o segundo estágio da DEA para verificar os efeitos das variáveis independentes em seus escores Para tanto foi realizado teste de heterocedasticidade para identificar os impactos das variáveis independentes nas variáveis dependentes Para o segundo estágio os escores obtidos anteriormente serão usados como variável dependente e as variáveis CAMELS utilizadas como as variáveis independentes conforme Figura 8 76 Figura 8 DEA e SFA dois estágios com regressão linear múltipla Fonte Elaborado pelo autor 77 O segundo estágio é utilizado para verificar quais variáveis podem influenciar significativamente nos escores de eficiência gerados no primeiro estágio Para o segundo estágio as variáveis CAMELS serão utilizadas como as variáveis independentes e os escores gerados pela técnica DEA e SFA serão para cada caso utilizados como as variáveis dependentes A escolha do acrônimo CAMELS já embasada pelo referencial teórico deste estudo após pesquisa na literatura nacional e internacional possibilitou a identificação de diversas fórmulas e índices financeiros Destas fórmulas para fins de adequação ao mercado brasileiro e a forma de publicação das demonstrações financeiras bem como a base utilizada IFBACEN as fórmulas que possuem contas tanto de balanço patrimonial quanto de demonstrativo de resultado representadas nas demonstrações financeiras das instituições financeiras brasileiras foram consideradas e possibilitaram a formulação do Quadro 6 Após avaliação das disponibilidades de informações utilizouse o seguinte critério utilização das equações apresentadas pelo Fundo Monetário Internacional presentes no banco de dados do Bacen e as que complementam a representação da definição de capital ativos gestão lucro liquidez e sensibilidade ao risco de mercado com outro conceito O Quadro 6 apresenta as equações a serem utilizadas por este estudo para os seis componentes do acrônimo CAMELS essas serão utilizados como variáveis independentes da regressão linear realizada no segundo estágio Quadro 6 Variáveis independentes advindas do CAMELS Inicial Equação Autores C Patrimônio Líquido Ativo Total Roman Sargu 2013 Dincer et al 2011 Magro Michels e Silva 2017 A Provisões para Operações de Crédito Total da Carteira Rosa e Gartner 2018 Ativo Circulante Ativo Total Dincer et al 2011 M Resultado de Intermediação Financeira Receitas com Serviços Despesas Pessoais Despesas Administrativas International Monetary Fund 2006 Pimenta 2014 E ROA Lucro Líquido Ativo Total x 100 International Monetary Fund 2006 Christopoulos Mylonakis e Diktapanidis 2011 Dincer et al 2011 Roman Sargu 2013 Pimenta 2014 Korontai 2016 Magro Michels Silva 2017 Marcusso 2017 ROE Lucro Líquido Patrimônio Líquido x 100 Korontai 2016 Christopoulos Mylonakis e Diktapanidis 2011 Dincer et al 2011 78 Roman Sargu 2013 Magro Michels Silva 2017 Marcusso 2017 L Ativo Circulante Ativo Total International Monetary Fund 2006 Christopoulos Mylonakis e Diktapanidis 2011 Dincer et al 2011 Pimenta 2014 Korontai 2016 Magro Michels Silva 2017 Total Créditos Total Depósitos Christopoulos Mylonakis e Diktapanidis 2011 Magro Michels Silva 2017 S Valores Mobiliários em Carteira Ativo Total Magro Michels Silva 2017 Valores Mobiliários em Carteira Empréstimos Total Fonte Elaborado pelo autor A análise de adequação de capital com o objetivo de mensurar a qualidade do capital da instituição frente ao seu cenário financeiro teve a fórmula selecionada que utiliza as contas de Patrimônio Líquido e Ativo Total com a finalidade de demonstrar quanto do Ativo da instituição financeira é financiado pelo seu Patrimônio Líquido assim sendo qual a dependência a instituição financeira possui frente aos recursos de terceiros para a continuidade de suas operações Para qualidade dos ativos a fim de verificar o risco da instituição de não recebimento de seus ativos duas fórmulas foram selecionadas e consideram as contas de Provisão sobre Operações de Crédito que considera a provisão para perdas com as operações de acordo com o nível de risco destas Operações de Crédito que contabiliza todas as operações sendo elas classificadas como normais em atraso ou em liquidação e Ativo Total a fim de mensurar quanto da provisão sobre estas operações impacta em sua carteira de crédito especificamente e de forma mais ampla em seu ativo total O componente de gestão que visa identificar a capacidade dos gestores em acompanhar e mitigar os riscos aos quais a instituição financeira está exposta teve a equação que considera as contas de Resultado de Intermediação Financeira Rendas com Prestação de Serviço Despesas de Pessoal e Despesas Administrativas selecionada com a finalidade de identificar o impacto das despesas com colaboradores e administrativas em geral na geração de resultados e rendas oriundas de sua atividade principal de intermediação de recursos e prestação de serviços Cabe destacar que a geração de resultados e rendas demonstra eficiência na mitigação de riscos e contenção de perdas sendo um bom indicativo de gestão de riscos Para o componente de lucro os índices financeiros de lucratividade ROA e ROE foram selecionados Estes dois índices possuem uma aplicabilidade comprovada cientificamente em diversos setores e organizações As instituições financeiras medem os retornos obtidos lucro 79 líquido sobre a sua carteira de ativos e sobre o seu patrimônio como forma de mensurar a sua eficiência Para liquidez foram consideradas as contas de Ativo Circulante Ativo Total Operações de Crédito e Depósito Total com o objetivo de quantificar o nível de liquidez da instituição financeira e a sua capacidade de cumprir suas obrigações financeiras e de gestão de ativos de terceiros e clientes 80 4 DESCRIÇÃO DOS RESULTADOS Este capítulo apresenta os resultados encontrados através da utilização das técnicas estatísticas DEA e SFA diante de uma mesma base de dados Alinhado ao objetivo deste trabalho as abordagens de intermediação e rentabilidade foram analisadas através dos gráficos gerados após a obtenção dos escores de eficiência Para que as análises e comparações entre as duas técnicas pudessem ocorrer considerouse a mesma base de dados contendo os mesmos bancos e períodos Todos os gráficos apresentam dados pontuais porém em cada uma das apresentações as linhas são contínuas apenas para melhor visualização dos pontos e de suas tendências Os gráficos contendo as medianas dos escores das técnicas DEA e SFA para as instituições financeiras brasileiras para as abordagens de intermediação e rentabilidade estão representados abaixo O Gráfico 1 apresenta a mediana dos escores obtidos pelas técnicas DEA e SFA para a abordagem de intermediação onde os inputs e o output utilizados estão representados na Tabela 3 81 Gráfico 1 Abordagem Intermediação Fonte Elaborado pelo autor 82 Percebese que a mediana dos escores obtidos a partir da técnica SFA apresentou uma menor volatilidade em comparação à técnica DEA Da mesma forma seus valores absolutos são maiores e concentrados na faixa de escores de 07 enquanto os resultados DEA são menores e encontramse distribuídos entre as faixas de 03 e 05 Mesmo com a menor volatilidade analisada no gráfico da técnica de SFA é possível identificar que ambas as técnicas apresentam momentos de elevações e diminuições dos índices porém não se pode afirmar que há uma relação entre estes movimentos das duas técnicas ou seja existem momentos em que os movimentos são opostos ou momentos em que apenas a técnica DEA demonstra variações O Gráfico 2 representa a mediana dos escores obtidos pelas técnicas DEA e SFA para a abordagem de rentabilidade onde os inputs e o output utilizados estão representados na Tabela 3 83 Gráfico 2 Abordagem Rentabilidade Fonte Elaborado pelo autor 84 É possível identificar que assim como a abordagem anterior já analisada a abordagem de rentabilidade apresenta os menores escores para a técnica DEA enquanto a técnica SFA apresenta a menor variabilidade no decorrer do período analisado Foi possível encontrar padrões de altas e baixas com o mesmo movimento nos anos 2002 2004 2005 2008 2011 e 2013 Observase neste trabalho ao analisarse os Gráficos 1 e 2 que as técnicas DEA e SFA apresentam compatibilidade contudo é importante salientar que as duas técnicas possuem vieses e a combinação das técnicas permite uma melhor reflexão sobre o desempenho das variáveis conforme será apresentado na seção discussão dos resultados onde ao considerar bases de dados diferentes e países diferentes os resultados são diversos devido aos modelos matemáticos que suportam cada um dos estudos analisados Ao analisarse as abordagens de intermediação e rentabilidade é possível identificar que a técnica SFA foi a que acusou menor volatilidade representada nos gráficos 1 e 2 A abordagem de intermediação apresentou maior oscilação comparada com a abordagem de rentabilidade Para ambas as técnicas e ambas as abordagens os gráficos gerados através dos escores obtidos pela técnica SFA foram os que apresentaram os maiores valores para os escores Nesta visão destacase a abordagem de rentabilidade sendo bastante superior e consistente quando comparada a abordagem de intermediação Os gráficos gerados através da técnica DEA para ambas as abordagens apresentaram maior volatilidade mas ao mesmo tempo mantiveramse sempre nos menores patamares e muito próximos aos escores centrais de eficiência ou seja próximos a faixa de 05 É importante salientar que ambas as técnicas possuem modelos matemáticos diferentes e por este motivo são esperados resultados diferentes entre as técnicas e sua aplicação para as abordagens deste estudo Com o objetivo de embasar ainda mais esta análise na Tabela 4 encontramse apresentadas as estatísticas descritivas em relação as duas técnicas e as abordagens deste estudo Assim como exposto no referencial teórico de acordo com Berger e Mester 1997 comprovouse o cenário de maior volatilidade e médias menores obtidas pela técnica de Análise Envoltória de Dados frente a técnica de Análise de Fronteira Estocástica 85 Tabela 4 Estatísticas descritivas dos escores de DEA e SFA DEA SFA Intermediação Rentabilidade Intermediação Rentabilidade Máximo 10000 10000 09799 09694 Mínimo 00007 00263 01065 00346 Média 04302 05273 08434 07295 Mediana 04186 05064 08546 07465 Desvio Padrão 02287 01698 00706 01019 Variância 00523 00288 00050 00104 Fonte Elaborada pelo autor Através das estatísticas descritivas é possível confirmar que a técnica SFA é menos sensível aos outliers e por este motivo apresenta valores absolutos maiores para as médias quando comparada aos valores gerados para as médias da técnica DEA O mesmo ocorre para a análise da mediana Seguindo o mesmo racional da análise de média e mediana podese verificar que os desviopadrão para a técnica SFA são menores bem como os valores de variância confirmando a menor sensibilidade desta em relação aos outliers 41 DEA Segundo Estágio resultados da regressão linear múltipla A seguir são transcritos os resultados das análises realizadas em software IBM SPSS com destaque para as variáveis que deram significantes para 411 DEA Segundo Estágio para abordagem Intermediação As variáveis significantes para Intermediação estão presentadas nas Tabelas 5 e 6 86 Tabela 5 Modelo para Intermediação model summary R Square Adjusted R Square Std Error of the Estimate DurbinWatson 461 459 157362111 466 Fonte elaborada pelo autor Tabela 6 Variáveis significantes para Intermediação coefficients a Constantes Beta Sig Tolerance VIF Patrimônio Líquido Ativo Total 570 000 767 1303 Provisão sobre Operações de Crédito Operações de Crédito 099 000 872 1147 Operações de Crédito Ativo Total 413 000 387 2583 Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas 253 000 631 1585 ROE Lucro Líquido Patrimônio Líquido x 100 042 027 916 1092 Ativo Circulante Ativo Total 393 000 441 2266 TVM e instrumentos Financeiros Derivativos Ativo Total 094 001 395 2530 TVM e Instrumentos Financeiros Derivativos Operações de Crédito 086 000 670 1492 Fonte elaborada pelo autor Onde Dependent Variable DEAIntermPredictors Constant Patrimônio Líquido Ativo Total Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas TVM e instrumentos Financeiros Derivativos Ativo Total Ativo Circulante Ativo Total Operações de Crédito Ativo Total Provisão sobre Operações de Crédito Operações de Crédito TVM e Instrumentos Financeiros Derivativos Operações de Crédito ROE Lucro Líquido Patrimônio Líquido x 100 412 DEA Segundo Estágio para abordagem Rentabilidade As variáveis que deram significativas para Rentabilidade na análise via SPSS estão especificadas nas Tabelas 7 e 8 87 Tabela 7 Modelo para Rentabilidade model summary R Square Adjusted R Square Std Error of the Estimate DurbinWatson 286 285 136337268 755 Fonte elaborada pelo autor Onde Predictors Constant ROE Lucro Líquido Patrimônio Líquido x 100 Patrimônio Líquido Ativo Total Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas Provisão sobre Operações de Crédito Operações de Crédito ROA Lucro Líquido Ativo Total x 100 TVM e Instrumentos Financeiros Derivativos Operações de Crédito Tabela 8 Variáveis significantes para Rentabilidade coefficients a Constantes Beta Sig Tolerance VIF Patrimônio Líquido Ativo Total 327 000 879 1137 Provisão sobre Operações de Crédito Operações de Crédito 156 000 904 1107 Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas 201 000 691 1448 ROA Lucro Líquido Ativo Total x 100 155 000 342 2927 ROE Lucro Líquido Patrimônio Líquido x 100 275 000 339 2947 TVM e Instrumentos Financeiros Derivativos Operações de Crédito 065 000 720 1388 Fonte elaborada pelo autor Onde Dependent Variable DEARent 42 SFA e os fatores explicativos para eficiência bancária O modelo SFA CobbDouglas halfnormal com heterocedasticidade que separa ineficiência persistente e ineficiência variável no tempo KUMBHAKAR LIEN HARDAKER 2012 é apresentado a seguir 88 No primeiro passo o Teste de Especificação de Hausman para Intermediação e Rentabilidade rejeitou a hipótese nula de que o modelo de painel com efeitos aleatórios seja mais apropriado considerandose desta forma efeitos fixos confirmados por Prob chi2 00000 Também no primeiro passo foi executada a regressão para efeitos fixos resultando nos dados estampados nas Tabelas 9 e 10 para Intermediação e Rentabilidade respectivamente Tabela 9 Regressão para efeitos fixos de Intermediação Variável Coef Std Err Pt Patrimônio Líquido Ativo Total 112686 0173637 0000 Provisão sobre Operações de Crédito Operações de Crédito 0315239 020155 0118 Operações de Crédito Ativo Total 3880897 0172254 0000 Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas 0013102 0004834 0007 ROA Lucro Líquido Ativo Total x 100 0021496 0008114 0008 ROE Lucro Líquido Patrimônio Líquido x 100 0003349 0001423 0019 Ativo Circulante Ativo Total 406171 0174784 0000 Operações de CréditoDepósito Total 288e06 209e06 0168 TVM e instrumentos Financeiros DerivativosAtivo Total 2275202 0163384 0000 TVM e Instrumentos Financeiros DerivativosOperações de Crédito 615e06 226e06 0007 Fonte elaborada pelo autor O teste retornou u 11474461 e 07553889 P 69764795 89 Tabela 10 Regressão para efeitos fixos de Rentabilidade Variável Coef Std Err Pt Patrimônio Líquido Ativo Total 0014794 013121 0910 Provisão sobre Operações de Crédito Operações de Crédito 047714 0152768 0002 Operações de Crédito Ativo Total 1222386 013028 0000 Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas 0045969 0003675 0000 ROA Lucro Líquido Ativo Total x 100 0113539 0006581 0000 ROE Lucro Líquido Patrimônio Líquido x 100 0007362 0001194 0000 Ativo Circulante Ativo Total 0641651 013215 0000 Operações de CréditoDepósito Total 220e06 158e06 0164 TVM e instrumentos Financeiros DerivativosAtivo Total 0665468 0123492 0000 TVM e Instrumentos Financeiros DerivativosOperações de Crédito 628e06 171e06 0000 Constante 7555242 0088925 0000 Fonte elaborada pelo autor O teste retornou u 05054876 e 05692987 P 44083762 43 Análise consolidada DEA e SFA nas abordagens Intermediação e Rentabilidade Os resultados deste estudo trazem os efeitos das variáveis independentes CAMELS na variável dependente escores DEA e escores SFA encontrados através da regressão linear 90 múltipla aplicada Essas avaliações são realizadas em duas principais abordagens utilizadas para o estudo da eficiência bancária intermediação e rentabilidade Ao considerar a base de dados de bancos brasileiros no período que contempla os anos de 2000 a 2018 podemos identificar as variáveis que apresentaram resultados estatisticamente significantes e seus impactos Desta forma para este estudo entre as variáveis estudadas observaramse cinco situações i variáveis estatisticamente relevantes e com impacto positivo sobre a eficiência bancária ii variáveis estatisticamente relevantes e com impacto negativo sobre a eficiência bancária iii variáveis não estatisticamente relevantes e com impacto positivo sobre a eficiência bancária iv variáveis não estatisticamente relevantes e com impacto negativo sobre a eficiência bancária e por último v variáveis sem relação com a eficiência bancária Alinhado ao objetivo deste estudo que busca identificar o impacto de variáveis CAMELS na eficiência bancária considerouse para análise apenas as variáveis estatisticamente relevantes e com impactos positivos ou negativos sobre a eficiência No contexto brasileiro abrangido pelo período deste estudo podese afirmar que as variáveis que apresentam impactos positivos podem ser interpretadas como quanto maior o valor da variável analisada maior a eficiência DEA e SFA para a abordagem analisada Da mesma forma ao analisar as variáveis que apresentam impacto negativo a interpretação é quanto maior o valor da variável menor a eficiência DEA e SFA para a abordagem analisada Entendese que a variável Patrimônio Líquido sobre Ativo Total demonstra quanto de recursos próprios é utilizado no financiamento e manutenção de sua atividade Assim sendo ao considerar o setor bancário tanto sob a ótica da abordagem de intermediação quanto sob a ótica da abordagem de rentabilidade entendese que quanto menor o resultado deste índice melhor podemos considerar seu impacto para as instituições bancárias O índice representado pela razão entre Provisões sobre Operações de Crédito sobre Operações de Crédito e o índice representado por Operações de Crédito sobre Ativo Total buscam mensurar o risco de a instituição bancária não receber seus ativos Esperase para o primeiro índice abordado que quanto menor este índice melhor a eficiência da instituição 91 financeira para ambas as abordagens aplicadas nesta dissertação Já para o segundo índice abordado esperase para a abordagem de intermediação que quanto maior melhor a eficiência bancária Por outro lado sob a ótica de intermediação esperase que quanto maior o índice melhor a sua eficiência e esperase o contrário sob a ótica de rentabilidade ou seja quanto menor este índice melhor a eficiência da instituição analisada Para estudar o impacto do indicador de gestão na eficiência bancária este estudo utilizouse do índice da soma do Resultado de Intermediação Financeira com Renda de Prestação de Serviços sobre a soma de Despesas com Pessoal com Despesas Administrativas Através deste índice mensurase o impacto das despesas com colaboradores e administrativas na geração de renda na atividade bancária Para isso ao considerarse a abordagem de intermediação quanto maior o valor encontrado neste índice melhor a eficiência bancária da instituição Já para a abordagem de rentabilidade podese considerar que não há uma relação conclusiva em relação a este indicador e a variável independente de gestão do acrônimo CAMELS utilizada neste estudo Um dos índices mais utilizados para justificar a eficiência bancária encontrase nos indicadores de rentabilidade Dentre eles o presente estudo utilizou os mais encontrados na literatura nacional e internacional o ROA e o ROE A utilização dos retornos obtidos por instituições financeiras sobre sua carteira de ativos e patrimônio são fundamentais para identificar a eficiência bancária Esperase portanto tanto para ROA quanto para ROE que quanto maiores os índices encontrados melhor para ambas as abordagens Para mensurar o impacto da liquidez na eficiência bancária utilizouse a razão encontrada entre o Ativo Circulante sobre o Ativo Total sendo utilizado para quantificar o nível de liquidez da instituição bancária ou seja a sua capacidade de pagamento de suas obrigações junto a terceiros e clientes Frente ao exposto considerase que quanto maior este índice mais eficiente a instituição financeira em ambas as abordagens A sensibilidade ao risco de mercado foi mensurada através de duas variáveis a variável Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos sobre o Ativo Total e a variável TVM e Instrumentos Financeiros Derivativos sobre Operações de Crédito Essas variáveis identificam a exposição da instituição financeira ao mercado de crédito e o risco de mercado de suas operações mantidas pelas instituições financeiras Para a abordagem de intermediação esperase para ambas as variáveis analisadas que quanto menor melhor a sua eficiência Em contrapartida para a abordagem de rentabilidade esperase que quanto maior os índices obtidos melhor a eficiência bancária das instituições devido ao maior retorno que estas operações podem desenvolver para compensar o risco 92 Com o objetivo de entender os impactos gerados pelas variáveis independentes sobre a eficiência bancária brasileira dos anos de 2000 a 2018 sob a ótica das abordagens de produção e rentabilidade este estudo realizou a análise das regressões retro item 42 segundo estágio DEA e heterocedasticidade SFA para identificar se o impacto das variáveis independentes sobre a eficiência bancária ocorreu neste estudo de acordo com a teoria financeira apresentada ou de forma contrária Além disso esta análise possibilita entender o comportamento de ambas as técnicas DEA e SFA bem como identificar semelhanças e diferenças entre elas A fim de atingir este objetivo as variáveis são analisadas uma a uma para cada uma das técnicas e comparadas com os achados sobre a teoria financeira 431 Pressupostos dos indicadores CAMELS para as Abordagens de Intermediação e Rentabilidade A abordagem de intermediação trata sobre a capacidade que a instituição financeira possui em intermediar a transação de recursos financeiros entre agentes superavitários e agentes deficitários A eficiência bancária sob a ótica de intermediação representa a quantidade de recursos que a instituição financeira consegue intermediar de maneira a receber o maior retorno sobre o maior volume de transações A abordagem de rentabilidade mensura a capacidade da instituição financeira de gerar receita orientada ao lucro Dessa forma analisase a eficiência sob a ótica de rentabilidade de uma instituição financeira pela sua capacidade de minimizar os custos resultantes da geração de receitas Buscase portanto lucratividade Diante destes fatos ao analisarmos as afirmações da teoria financeira podemos considerar sob a ótica da abordagem de intermediação e rentabilidade frente as variáveis independentes CAMELS que apresentaram impactos significativos nas variáveis dependentes tanto positivos quanto negativos o que se segue Entendese que a variável Patrimônio Líquido sobre Ativo Total demonstra quanto de recursos próprios é utilizado no financiamento e manutenção de sua atividade Assim sendo ao considerar o setor bancário tanto sob a ótica da abordagem de intermediação quanto sob a ótica da abordagem de rentabilidade entendese que quanto menor o resultado deste índice melhor podemos considerar seu impacto para as instituições bancárias O índice representado pela razão entre Provisões sobre Operações de Crédito sobre Operações de Crédito e o índice representado por Operações de Crédito sobre Ativo Total 93 buscam mensurar o risco de a instituição bancária não receber seus ativos Esperase para o primeiro índice abordado que quanto menor este índice melhor a eficiência da instituição financeira para ambas as abordagens aplicadas nesta dissertação Já para o segundo índice abordado esperase para a abordagem de intermediação que quanto maior melhor a eficiência bancária Por outro lado sob a ótica de intermediação esperase que quanto maior o índice melhor a sua eficiência e esperase o contrário sob a ótica de rentabilidade ou seja quanto menor este índice melhor a eficiência da instituição analisada Para estudar o impacto do indicador de gestão na eficiência bancária este estudo utilizouse do índice da soma do Resultado de Intermediação Financeira com Renda de Prestação de Serviços sobre a soma de Despesas com Pessoal com Despesas Administrativas Através deste índice mensurase o impacto das despesas com colaboradores e administrativas na geração de renda na atividade bancária Para isso ao considerarse a abordagem de intermediação quanto maior o valor encontrado neste índice melhor a eficiência bancária da instituição Já para a abordagem de rentabilidade podese considerar que não há uma relação conclusiva em relação a este indicador e a variável independente de gestão do acrônimo CAMELS utilizada neste estudo Um dos índices mais utilizados para justificar a eficiência bancária encontrase nos indicadores de rentabilidade Dentre eles o presente estudo utilizou os mais encontrados na literatura nacional e internacional o ROA e o ROE A utilização dos retornos obtidos por instituições financeiras sobre sua carteira de ativos e patrimônio são fundamentais para identificar a eficiência bancária Esperase portanto tanto para ROA quanto para ROE que quanto maiores os índices encontrados melhor para ambas as abordagens Para mensurar o impacto da liquidez na eficiência bancária utilizouse a razão encontrada entre o Ativo Circulante sobre o Ativo Total sendo utilizado para quantificar o nível de liquidez da instituição bancária ou seja a sua capacidade de pagamento de suas obrigações junto a terceiros e clientes Frente ao exposto considerase que quanto maior este índice mais eficiente a instituição financeira em ambas as abordagens A sensibilidade ao risco de mercado foi mensurada através de duas variáveis a variável Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos sobre o Ativo Total e a variável TVM e Instrumentos Financeiros Derivativos sobre Operações de Crédito Essas variáveis identificam a exposição da instituição financeira ao mercado de crédito e o risco de mercado de suas operações mantidas pelas instituições financeiras Para a abordagem de intermediação esperase para ambas as variáveis analisadas que quanto menor melhor a sua eficiência Em contrapartida para a abordagem de rentabilidade esperase que quanto maior os 94 índices obtidos melhor a eficiência bancária das instituições devido ao maior retorno que estas operações podem desenvolver para compensar o risco Com o objetivo de entender os impactos gerados pelas variáveis independentes sobre a eficiência bancária brasileira dos anos de 2000 a 2018 sob a ótica das abordagens de produção e rentabilidade este estudo realizou a análise das regressões segundo estágio DEA e heterocedasticidade SFA para identificar se o impacto das variáveis independentes sobre a eficiência bancária ocorreu neste estudo de acordo com a teoria financeira apresentada ou de forma contrária Além disso esta análise possibilita entender o comportamento de ambas as técnicas DEA e SFA bem como identificar semelhanças e diferenças entre elas A fim de atingir este objetivo as variáveis serão analisadas uma a uma para cada uma das técnicas e comparadas com os achados sobre a teoria financeira 432 Intermediação Neste estudo ao analisar o item de adequação do capital a variável Patrimônio Líquido sobre Ativo Total apresentou efeito negativo sobre a eficiência para a técnica DEA Desta forma considerase adequada à teoria utilizada como base deste estudo Ou seja a técnica DEA para este estudo para a base de dados deste estudo e para o período abrangido por ele demonstra suportar as teorias de finanças utilizadas para embasar este estudo Por outro lado o índice apresentou impacto positivo sobre a eficiência para a técnica SFA e portanto podese identificar uma não conformidade com a teoria apresentada anteriormente Para analisar a qualidade dos ativos o índice representado por Operações de Crédito sobre Ativo Total apresentou resultado positivo para a técnica DEA e para a técnica SFA por este motivo pode ser considerado como uma das variáveis que estão alinhadas a teoria financeira Além desse índice a razão entre Provisões sobre Operações de Crédito e Operações de Crédito também apresentou resultado compatível com a teoria financeira aplicada nesta dissertação Este fato para a variável em questão ocorreu apenas para a técnica DEA Ao estudar o impacto do indicador de gestão na eficiência bancária este estudo utilizouse do índice da soma do Resultado de Intermediação Financeira com Renda de Prestação de Serviços sobre a soma de Despesas com Pessoal com Despesas Administrativas a técnica DEA e SFA apresentaram um impacto negativo deste índice sobre a eficiência Neste caso sugerese sugere não adequação a teoria financeira 95 Um dos índices mais utilizados para justificar a eficiência bancária encontramse nos indicadores de rentabilidade Dentre eles o presente estudo utilizou os mais encontrados na literatura nacional e internacional o ROA e o ROE Os resultados obtidos demonstram que para a técnica DEA somente o ROE apresentou impacto a ser considerado onde encontrouse um efeito positivo na eficiência bancária e adequação a teoria financeira já abordada Os resultados obtidos nos indicadores de rentabilidade demonstram que o ROA apresentou impactos negativos para a técnica SFA resultado que indica neste estudo a não adequação a teoria fato que deve ser comprovado através de estudos futuros sobre o tema Já para os resultados do índice de lucratividade ROE encontrouse uma relação positiva para a técnica SFA e consequentemente o status de adequada a teoria financeira O índice utilizado para mensurar o impacto da liquidez na eficiência bancária e que apresentou efeito positivo tanto para a técnica DEA quanto para SFA e adequação à teoria financeira foi a razão encontrada entre o Ativo Circulante sobre o Ativo Total Com o objetivo de mensurar a sensibilidade ao risco de mercado duas variáveis foram analisadas A variável Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos sobre o Ativo Total que apresentou efeito positivo para DEA sendo classificado como adequado ao estipulado pela teoria financeira e que apresentou impacto negativo para SFA e foi classificada como não adequada à teoria financeira aplicada A variável Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos sobre Operações de Crédito apresentou efeito positivo para ambas as técnicas e portanto foi classificada como adequada a teoria financeira A Tabela 11 compila os resultados obtidos para a abordagem de intermediação para ambas as técnicas Análise Envoltória de Dados e Fronteira Estocástica 96 Tabela 11 Efeitos na eficiência SFA e DEA para Abordagem de Intermediação CAMELS Variável SFA Efeitos fixos e aleatórios R2 048 DEA R2 046 Efeito na eficiência Significância Coef Efeito na eficiência Significância Coef C Patrimônio Líquido Ativo Total POSITIVO 0000 1 13E01 NEGATIVO 0000 570E01 A Provisão sobre Operações de Crédito Operações de Crédito Sem relação Sem relação Sem relação POSITIVO 0000 990E02 Operações de Crédito Ativo Total POSITIVO 0000 388E01 POSITIVO 0000 413E01 M Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas NEGATIVO 0007 131E03 NEGATIVO 0000 25E01 E ROA Lucro Líquido Ativo Total x 100 NEGATIVO 0008 215E03 Sem relação Sem relação Sem relação ROE Lucro Líquido Patrimônio Líquido x 100 POSITIVO 0019 335E04 POSITIVO 0027 420E02 L Ativo Circulante Ativo Total POSITIVO 0000 406E01 POSITIVO 0000 393E01 S TVM e instrumentos Financeiros Derivativos Ativo Total NEGATIVO 0000 228E01 POSITIVO 0001 940E02 TVM e instrumentos Financeiros Derivativos Operações de Crédito POSITIVO 0007 615E06 POSITIVO 0000 860E02 Fonte Elaborada pelo autor 97 433 Rentabilidade Da mesma forma que a abordagem de Intermediação a abordagem de Rentabilidade também considerou as variáveis estatisticamente relevantes e com impactos positivos ou negativos sobre a eficiência É importante ressaltar que de acordo com o período base de dados e modelos matemáticos das técnicas DEA e SFA utilizados neste estudo podemos considerar que as variáveis que apresentam impactos positivos podem ser interpretadas como quanto maior o valor da variável analisada maior a eficiência DEA E SFA para a abordagem analisada Da mesma forma ao analisar as variáveis que apresentam impacto negativo a interpretação é quanto menor o valor da variável maior a eficiência DEA e SFA para a abordagem analisada Para a abordagem de Rentabilidade a variável Patrimônio Líquido sobre Ativo Total apresentou significância apenas para a técnica DEA identificando um impacto negativo desta variável na eficiência e consequentemente sendo classificada como adequada a teoria financeira abordada neste estudo Neste estudo a razão entre Provisões sobre Operações de Crédito e Operações de Crédito apresentou um impacto negativo para a técnica DEA o que condiz com a teoria de finanças apresentada Já para a técnica SFA o efeito observado foi positivo desenquadrandoa do embasamento teórico utilizado Já a variável Operações de Crédito sobre Ativo Total apresentou um efeito positivo sobre a eficiência apenas para a técnica SFA e portanto desenquadrandoa do grupo de variáveis que se adequam a teoria financeira deste estudo Ao analisar a soma do Resultado de Intermediação Financeira com Renda de Prestação de Serviços sobre a soma de Despesas com Pessoal com Despesas Administrativas encontrou se um impacto negativo deste índice sobre a eficiência para ambas as técnicas utilizadas neste estudo Este fato diante a teoria financeira utilizada por este estudo para a abordagem de rentabilidade evidenciou a impossibilidade de se concluir se um efeito negativo ou positivo no efeito da eficiência seria o mais adequado Os índices mais utilizados ao se mensurar eficiência bancária ROA e ROE são índices indispensáveis para a análise da eficiência bancária sob a ótica de rentabilidade Ambos os índices apresentaram efeito positivo na eficiência bancária para as técnicas DEA e SFA Desta forma considerase ambos os índices adequados a teoria financeira presente nesta dissertação O índice Ativo Circulante sobre o Ativo Total apresentou apenas para a técnica SFA 98 impacto positivo sobre a eficiência bancária alinhandose desta forma aos indicadores que se relacionam adequadamente a teoria financeira A variável Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos sobre o Ativo Total apresentou impacto positivo e adequação a teoria financeira apenas para a técnica SFA Já a variável Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos sobre Operações de Crédito apresentou impacto negativo para a técnica DEA e positivo para a técnica SFA Desta forma considerase adequado a teoria financeira apenas para a técnica SFA A Tabela 12 compila os resultados obtidos para a abordagem de intermediação para ambas as técnicas Análise Envoltória de Dados e Fronteira Estocástica 99 Tabela 12 Efeitos na eficiência SFA e DEA para Abordagem de Rentabilidade CAMELS Variável SFA Efeitos fixos e aleatórios R2 038 DEA R2 028 Efeito na eficiência Significância Coef Efeito na eficiência Significância Coef C Patrimônio Líquido Ativo Total Sem relação Sem relação Sem relação NEGATIVO 0000 327E01 A Provisão sobre Operações de Crédito Operações de Crédito POSITIVO 0002 477E02 NEGATIVO 0000 156E01 Operações de Crédito Ativo Total POSITIVO 0000 122E01 Sem relação Sem relação Sem relação M Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas NEGATIVO 0000 460E03 NEGATIVO 0000 201E01 E ROA Lucro Líquido Ativo Total x 100 POSITIVO 0000 114E02 POSITIVO 0000 155E01 ROE Lucro Líquido Patrimônio Líquido x 100 POSITIVO 0000 736E04 POSITIVO 0000 275E01 L Ativo Circulante Ativo Total POSITIVO 0000 642E02 Sem relação Sem relação Sem relação S TVM e instrumentos Financeiros Derivativos Ativo Total POSITIVO 0000 665E02 Sem relação Sem relação Sem relação TVM e instrumentos Financeiros Derivativos Operações de Crédito POSITIVO 0000 628E06 NEGATIVO 0000 65E02 Fonte Elaborada pelo autor 100 5 DISCUSSÃO DOS RESULTADOS A utilização combinada das técnicas DEA e SFA pode ser justificada pela diferença em seus modelos matemáticos que são mais sensíveis cada um com suas peculiaridades a determinadas alterações das variáveis Desta forma criase uma harmonia entre a utilização combinada das mesmas o que fortalece a análise dos fatores que influenciaram a eficiência bancária brasileira entre os anos de 2000 e 2018 ponto central desta dissertação A fim de aprofundar a análise dos resultados estudos internacionais que possuem resultados sobre a influência de índices CAMELS na eficiência bancária foram analisados e relacionados É importante ressaltar que além de modelos matemáticos diferentes estes estudos também possuem diferenças entre locais onde foram aplicados bem como os períodos de aplicação O quadro sinótico dos pressupostos teóricos relativos à abordagem de Intermediação e de Rentabilidade está estampado no Quadro 7 101 Quadro 7 Adequação à Teoria Financeira Abordagem de Intermediação e Rentabilidade CAMELS Variável Abordagem de intermediação Abordagem de rentabilidade DEA SFA DEA SFA Adequação à Teoria Financeira Adequação à Teoria Financeira Adequação à Teoria Financeira Adequação à Teoria Financeira C Patrimônio Líquido Ativo Total CORROBORA DIVERGE CORROBORA X A Provisão sobre Operações de Crédito Operações de Crédito X CORROBORA CORROBORA DIVERGE Operações de Crédito Ativo Total CORROBORA CORROBORA X DIVERGE M Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas DIVERGE DIVERGE INCONCLUSIVO INCONCLUSIVO E ROA Lucro Líquido Ativo Total x 100 CORROBORA X CORROBORA CORROBORA ROE Lucro Líquido Patrimônio Líquido x 100 CORROBORA CORROBORA CORROBORA CORROBORA L Ativo Circulante Ativo Total CORROBORA CORROBORA X CORROBORA S TVM e instrumentos Financeiros Derivativos Ativo Total DIVERGE CORROBORA X CORROBORA TVM e instrumentos Financeiros Derivativos Operações de Crédito DIVERGE DIVERGE DIVERGE CORROBORA Fonte Elaborado pelo autor 102 Para a análise da Adequação de Capital através da variável Patrimônio Líquido sobre Ativo Total ao considerar a aplicação da técnica DEA para a abordagem de Intermediação e realização do segundo estágio Adesina 2019 buscou preencher uma lacuna importante na literatura de eficiência bancária investigando a influência do IC Capital Intelectual nas eficiências técnicas alocativas e de custo dos bancos na África A amostra consistiu de 339 bancos comerciais em 31 países africanos no período de 2005 a 2015 Rosman Wahab e Zainol 2014 e Sufiana e Habibullah 2012 evidenciaram o impacto positivo que esta variável possui sobre a eficiência bancária para bancos islâmicos No caso dos bancos em países do Oriente Médio e da Ásia em média podem ser caracterizados como tecnicamente eficientes ao longo dos períodos Isso indica que o a gestão dos bancos foi capaz de controlar de forma eficiente os custos e usar a diversidade de entradas para produzir saídas independentemente dos efeitos de escala Ainda para a abordagem de Intermediação Bittar Pukthuanthong e Walker 2019 através da utilização da técnica DEA ao analisarem a adequação de capital encontraram efeitos positivos da capitalização sobre a eficiência Os autores exploraram o efeito de manter maiores índices de capital e liquidez sobre a eficiência dos bancos convencionais e islâmicos bem como se esse efeito difere entre bancos menos eficientes e mais eficientes usando regressões de quantis condicionais Os resultados sugerem que os índices de capital e liquidez estão positivamente associados à eficiência existente dos dois tipos de banco No entanto este efeito positivo parece ser menos pronunciado para os bancos islâmicos indicando que maiores índices de capital e liquidez ampliam a eficiência diferença entre os dois tipos de banco em favor dos bancos convencionais Por fim os resultados sugerem que bancos mais capitalizados e mais líquidos também foram mais eficientes durante a crise financeira global e a Primavera Árabe Da mesma forma porém com a aplicação da técnica SFA Otero et al 2020 evidenciaram impacto positivo desta variável sobre a eficiência das instituições financeiras Este estudo investigou a eficiência de custos do sistema bancário MENA indústria para o período de 20052012 usando um modelo de fronteira estocástica com variáveis específicas do país Os resultados da análise mostraram que a pontuação média de eficiência de custo foi de 77 e que houve uma melhoria nas pontuações de eficiência durante o período coberto A variação em termos de eficiência de custos é de 19 por cento entre os países Geograficamente Israel 86 é o mais econômico enquanto os bancos do Kuwait 67 são os menos 103 Os resultados também mostraram que os bancos em Jordânia 82 Tunísia 82 e Omã 80 em média tiveram maior pontuações comparadas com outros países Diante desses resultados parece que ainda há espaço para melhorias quando se trata de banco eficiência nesta região Em relação aos determinantes da eficiência o artigo mostra a importância da estrutura de mercado como principal determinante da eficiência de custos para os bancos que operam na região MENA Em particular o nível de concentração e participação de mercado tiveram uma influência negativa na técnica eficiência apoiando a hipótese de vida silenciosa Também encontrouse suporte para a hipótese de competiçãoineficiência que pode ser devido ao relação menos estável entre clientes e bancos e quanto maior probabilidade de buscarem melhores condições em outro lugar No nível individual o tamanho do banco mostra a importância das economias de escala e o nível de capital também produz um efeito positivo Finalmente nenhuma diferença foi encontrada em termos de eficiência entre bancos convencionais e islâmicos Os efeitos na estrutura do mercado são mais importantes no pós crise período do que o précrise Na verdade o efeito do desempenho na eficiência é só positivo no período de crise mostrando que bancos em expansão fase têm outras fontes disponíveis não dependem da eficiência para ser rentáveis e menos preocupados com os custos O mesmo acontece com competição e é em tempos de crise que a rivalidade tem um impacto negativo na eficiência de custos Crise pode forçar bancos a serem mais agressivos para tentar capturar depósitos e este comportamento aumenta a instabilidade do a relação entre os bancos e os clientes A crise também pode afetar negativamente a solvência de certos bancos incentivando os clientes a mudar para outra instituição financeira devido à amplificação da informação assimétrica Portanto os resultados mostram a importância de se considerar a situação econômica na análise da eficiência de custos Nesse sentido as políticas bancárias devem promover o desempenho do banco porque como apontamos está positivamente associado ao custo eficiência Além disso as autoridades monetárias e os decisores políticos devem adotar políticas que podem aumentar o tamanho dos bancos MENA mas ao mesmo o tempo controla o nível de concentração e competição isto é porque de acordo com nossa análise altos níveis de ambas as características de mercado conduzem a um efeito negativo na eficiência Ademais as autoridades devem aumentar a eficiência em uma fase de expansão do mercado porque é quando os bancos estão menos comprometidos com a eficiência de custos Finalmente capital 104 requisitos bem como contribuindo para a solvência dos bancos têm mostrou um impacto positivo na eficiência apoiando regras e regulamentos como Basileia III Saeed et al 2020 encontraram em seu estudo resultado importante ao concluírem que para a abordagem de Intermediação geralmente um aumento da capitalização apresenta impacto negativo na eficiência contudo para os bancos islâmicos da amostra utilizada o resultado é inverso e o aumento da capitalização possui impacto positivo sobre a eficiência e resulta em um aumento da eficiência bancária onde a capitalização é representada pela relação entre o patrimônio líquido e os ativos O que evidencia que a capitalização mais alta reduz a exposição ao risco e oferece uma afirmação da hipótese regulatória segundo a qual o capital deve aumentar proporcionalmente com a exposição ao risco do banco O efeito da capitalização para o risco representa as consequências não intencionais de ações regulatórias onde os bancos respondem aos requisitos de capitalização aumentados assumindo mais risco para compensar os benefícios perdidos da alavancagem Sarmiento e Galán 2017 através da aplicação da técnica SFA também evidenciaram a relação positiva desta variável sobre a eficiência Em geral os bancos colombianos exibem uma tendência de queda no risco de crédito e de mercado juntamente com níveis estáveis de capitalização e liquidez crescente No entanto os autores observaram diferenças importantes no nível de exposição ao risco de bancos com diferentes características de porte e propriedade que coincidem com as alterações regulatórias mencionadas incluindo aquelas adotadas em 20072008 para atender aos padrões de Basileia II Os autores consideram a medida de capitalização como baseada em duas características importantes Em primeiro lugar a regulamentação colombiana que estabelece que os bancos estrangeiros devem manter o mesmo capital mínimo que os bancos locais para operar Isso ocorre porque os bancos estrangeiros operam como subsidiárias em vez de agências na Colômbia e por sua vez precisam manter seu próprio capital Desta forma a medida de capitalização apresentada é comparável entre bancos com diferentes propriedades Em segundo lugar os autores argumentam que as diferenças nos níveis de capitalização podem sinalizar o apetite de risco dos bancos e influenciar seu desempenho Foi identificado no estudo realizado que níveis mais altos de capitalização levam a uma maior eficiência de custo e lucro As razões para esses resultados podem ser derivadas dos problemas de agência entre acionistas e administradores Os acionistas de bancos altamente capitalizados têm mais incentivos para controlar melhores custos e alocação de capital do que os de bancos 105 pouco capitalizados Isso incentiva melhores mecanismos de governança corporativa que podem levar a melhorias de eficiência Os resultados indicam que o efeito da capitalização sobre a eficiência difere entre bancos com diferentes tamanhos e participações e que os bancos pequenos e domésticos se beneficiam mais de índices de capital mais elevados em termos de eficiência de custo e lucro Para Djalilov e Piesse 2019 ao utilizarem a técnica SFA para a análise de adequação de capital encontraram impacto negativo sobre a eficiência Para os autores o nível de requisitos de capital impostos pelos reguladores também é controverso na literatura De acordo com a visão do interesse público a política de requisitos de capital reduz significativamente o nível de risco moral quando os proprietários dos bancos são obrigados a ter mais capital em risco Isso eventualmente leva a empréstimos mais cuidadosos e melhor desempenho O efeito dos requisitos de capital prudencial sobre a estabilidade bancária parece ser positivo nos setores bancários com supervisão e monitoramento relativamente fracos e instituições subdesenvolvidas características presentes nos setores bancários de países em transição Por outro lado a maioria desses países experimentou um crescimento sustentável e baixas taxas de inflação nas últimas duas décadas Isso causou aumentos na demanda por empréstimos bancários e portanto a presença de requisitos de capital prudencial pode limitar as oportunidades de crescimento dos bancos Sufian e Habibullah 2009 ao aplicar a técnica DEA corrobora com este resultado e considera o efeito negativo desta variável sobre a eficiência bancária Os autores demonstram que os bancos nacionais foram atingidos nas últimas décadas em simultâneo em várias frentes nomeadamente no levantamento de fundos estrangeiros perdas cambiais um aumento acentuado dos NPLs e perdas em participações acionárias A erosão da base de capital do sistema bancário devido a todos esses fatores restringiu severamente sua capacidade de emprestar até mesmo para empresas solventes em meio à crise devido à necessidade de cumprir as regras internacionais de adequação de capital Othman AbdulMajid e AbdulRahman 2017 ao aplicarem a técnica SFA também encontraram impacto negativo desta variável sobre a eficiência bancária na abordagem de Intermediação De acordo com a hipótese de risco bancos altamente capitalizados podem ter menos incentivos de risco e portanto são mais propensos a adotar práticas de redução de custos Por exemplo os acionistas podem ser mais ativos no controle dos custos dos bancos ou na alocação de capital por meio da disciplina de mercado 106 Em outras palavras esperase que os bancos altamente capitalizados que oferecem grande quantidade de financiamento de parceria sejam mais eficientes Para os autores o estudo apoia a ideia de que o financiamento de parceria pode ser usado como uma estratégia para melhorar a eficiência bancária especialmente para bancos com alta capitalização portanto baixo risco de capital Portanto ao monitorar os bancos que oferecem financiamento de parceria as autoridades podem precisar examinar seu nível de financiamento de parceria bem como seu risco de capital O estudo de Silva et al 2017 apesar de não ter realizado o segundo estágio para verificar o impacto de CAMELS na eficiência considerou positivamente mensurar a eficiência bancária através dos índices escolhidos e objetivou analisar índices que representam a possibilidade de aumento do comprometimento da gestão de crédito das cooperativas e a relação com a necessidade de mudanças nos instrumentos de controle gerencial Já para a abordagem de Rentabilidade ao estudar a variável Patrimônio Líquido sobre Ativo Total Fernandes Stasinakis e Bardarova 2018 encontraram impactos positivos gerados por este indicador CAMELS na eficiencia bancária de países europeus através da técnica DEA Da mesma forma Chen et al 2019 ao estudar os bancos Taiwaneses através da técnica DEA também encontrou impactos positivos sobre a eficiência bancária Ao considerar a razão entre Provisões sobre Operações de Crédito e Operações de Crédito Sufian e Habibullah 2009 ao considerar a técnica DEA e a abordagem de Intermediação encontrou que a eficiência bancária está negativamente relacionada ao comportamento de preferência de despesa e às condições econômicas enquanto a eficiência do banco está positivamente relacionada à intensidade de empréstimos em bancos da Malásia Sufiana e Habibullah 2012 chegaram ao mesmo resultado ao utilizar a técnica DEA em bancos da Indonésia onde os bancos relativamente melhor capitalizados com níveis de liquidez mais baixos exibem níveis mais elevados de eficiência enquanto os bancos com alto risco de crédito e despesas gerais tendem a ser relativamente ineficientes em sua função de intermediação O resultado sobre o impacto do crescimento do PIB dá suporte ao argumento da associação positiva entre crescimento econômico e desempenho do setor financeiro Ao estudar bancos chineses através da técnica DEA Avkiran 2011 ao analisar o impacto da variável da razão entre Provisões sobre Operações de Crédito e Operações de Crédito não encontrou relação de significância na abordagem de Rentabilidade Nesse estudo a rentabilidade teve associação com dois índices de lucratividade lucros pósimpostos sobre 107 ativos totais médios e retorno sobre o patrimônio líquido médio indicados como significativos em associação com estimativas de eficiência AVKIRAN 2011 DONG et al 2016 Ainda em relação a Qualidade dos Ativos sob a ótica da abordagem de Intermediação Saeed et al 2020 não encontraram resultados estatisticamente significativos para mensurar o impacto das Operações de Crédito sobre o Ativo Total dos bancos estudados ao utilizarem a técnica SFA Sufian e Habibullah 2009 através da aplicação da técnica DEA encontrou a relação positiva entre esta variável e o seu impacto sobre a eficiência bancária Da mesma forma Rosman Wahab e Zainol 2014 também chegaram a mesma conclusão ao aplicar a técnica DEA para o estudo de bancos islâmicos Por outro lado ao estudar as instituições bancárias da Indonésia Sufian e Habibullah 2012 encontram impactos negativos sobre a eficiência de instituições financeiras Sob a ótica da abordagem de Rentabilidade não houve estudos no levantamento realizado neste trabalho que analisaram os impactos da variável Operações de Crédito sobre Ativo Total a fim de analisar a qualidade dos ativos Ainda sobre a análise das variáveis CAMELS sob a ótica de Intermediação para Gestão a variável que traz a soma do Resultado de Intermediação Financeira com Renda de Prestação de Serviços sobre a soma de Despesa com Pessoal com Despesa Administrativa não foi encontrada nos estudos analisados Contudo o estudo de Shaddady e Moore 2018 identifica através do estudo das variáveis CAMELS e utilização da técnica DEA a relevância da supervisão sobre a gestão com um impacto positivo para economias avançadas e negativo para economias emergentes Também não houve estudos sob a ótica de Rentabilidade que analisaram o impacto desta variável a fim de analisar a Gestão das instituições bancárias As variáveis ROA e ROE são as variáveis mais encontradas em estudos sobre rentabilidade e eficiência bancária Nos estudos utilizados nesta pesquisa a variável ROA apresentou a totalidade dos casos tanto para a abordagem de Intermediação quanto para a abordagem de Rentabilidade Para os estudos analisados neste trabalho Djalilov e Piesse 2019 através da aplicação da técnica SFA identificaram impactos positivos da variável ROA sobre a eficiência para a abordagem de Intermediação ao considerar que a lucratividade é importante porque contribui para os níveis de eficiência dos bancos Portanto o Retorno sobre Ativos ROA é usado como uma variável de lucratividade Isso pode significar que os bancos com maior ROA e maiores em tamanho são mais eficientes Além disso também pode indicar que o crescimento 108 positivo do PIB e as restrições de atividade melhoram a eficiência bancária na amostra utilizada no estudo Da mesma forma Otero et al 2020 também encontraram impactos positivos da variável ROA sobre a eficiência bancária através da utilização da técnica SFA Para os autores o efeito do ROA é estatisticamente significativo e confirma a noção geral de que a lucratividade está positivamente relacionada à eficiência de custos Consequentemente os bancos com maior lucro tendem a ser mais eficientes e afirmam que resultado encontrado demonstra que os bancos com melhores práticas de gestão e controle de custos obtiveram maior lucratividade destacando a importância da eficiência de custos para o desempenho dos estabelecimentos nos países estudados Além disso os bancos mais rentáveis dispõem de recursos para investir em tecnologia processos e recursos humanos para aumentar sua eficiência de custos Ao analisar o impacto da variável ROA sobre a eficiência bancária através da técnica DEA Rosman Wahab e Zainol 2014 encontraram efeito também positivo e demonstraram que as instituições financeiras se tornam mais eficientes de acordo com o aumento de sua rentabilidade Isso ocorre pois a variável ROA incluída no modelo de regressão como proxy da lucratividade auxilia na comprovação de que os bancos que relatam índices de lucratividade mais altos são geralmente preferidos pelos clientes e portanto atraem a maior parte dos depósitos e os melhores tomadores de crédito potenciais que por sua vez criam um ambiente favorável para os bancos lucrativos serem mais eficientes Assim como para a abordagem de Intermediação para a abordagem de Rentabilidade foram encontrados estudos que analisaram apenas a variável ROA Nestes estudos Theodoridis e Anwar 2011 através da aplicação da técnica SFA encontrou um efeito positivo sobre a razão entre Lucro Líquido e Ativo Total sobre a eficiência bancária O ROA como proxy para a lucratividade do banco tem um impacto positivo na eficiência de custos dos bancos indonésios sobre os três modelos utilizados e com um nível de confiança de 99 Estes resultados implicam que a rentabilidade obtida pelos bancos indonésios impactam significativamente na melhoria do nível de eficiência de custos dos bancos comerciais durante o período Da mesma forma porém através da aplicação da técnica DEA Fernandes Stasinakis e Bardarova 2018 também encontraram efeitos positivos desta variável sobre as instituições bancárias estudadas Para os autores o ROA ainda exerce um efeito positivo e estatisticamente significativo em duas medidas de eficiência dos bancos No entanto o efeito é menor durante a crise financeira do que em períodos sem crises Este é um resultado esperado uma vez que os 109 bancos tendem a ser mais produtivos quando conseguem atrair níveis mais elevados de depósitos e encontrar tomadores de crédito com boa capacidade de crédito Por outro lado Chen et al 2019 ao estudar bancos taiwaneses encontrou impacto negativo desta variável sobre a eficiência bancária diferindo das expectativas gerais o que reflete a circunstância em que os bancos negligenciam a eficiência operacional ao tentar expandir suas operações e itens de negócios para criar um ROA mais alto Para a abordagem de Intermediação Sarmiento e Galán 2017 analisaram através da aplicação da técnica SFA que a razão entre o Ativo Circulante e Ativo Total tem um efeito negativo na eficiência bancária A proporção de ativos líquidos sobre ativos totais tem aumentado gradualmente ao longo do tempo especialmente para bancos grandes e estrangeiros A liquidez é medida como a razão entre os ativos líquidos sobre os ativos totais em que os ativos líquidos incluem disponibilidades de caixa negociáveis e disponíveis para vender instrumentos de dívida pública e privada e garantias dadas em operações compromissadas Ativos com maior liquidez impedem os bancos de descasamentos de vencimento embora mantêlos não tenha custo pois eles têm vencimentos mais curtos e portanto retornos mais baixos Por outro lado Bitar Pukthuanthong e Walker 2019 identificaram que os índices de liquidez estão positivamente associados à eficiência bancária analisada Para os autores tanto os depositantes quanto os investidores preferem negociar com bancos que tenham uma proporção saudável de ativos líquidos em relação aos depósitos e portanto os bancos mais eficientes tendem a manter índices de liquidez mais altos que podem servir como um mecanismo de segurança para proteger contra baixas de capital vendas dispendiosas e necessidades repentinas de aumentar o patrimônio líquido caro o que poderia afetar negativamente suas pontuações de eficiência Para a abordagem de Rentabilidade não houve estudos no levantamento realizado neste trabalho que analisaram os impactos da variável Ativo circulante sobre Ativo Total a fim de analisar a liquidez Também não houver estudos para ambas as abordagens que analisaram os impactos das variáveis de Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos sobre Ativo Total ou sobre Operações de Crédito a fim de analisar a sensibilidade ao risco de mercado 110 6 CONCLUSÕES Buscouse nesta pesquisa identificar os fatores que influenciaram a eficiência dos bancos brasileiros nas abordagens de intermediação e rentabilidade no período abordado sob a ótica de duas principais técnicas de mensuração de eficiência DEA e SFA Para isto foram tratados os dados de 221 instituições financeiras pertencentes ao grupo B1 extraídas do endereço eletrônico do BACEN atraves do relatório denominado IFdata Através da avaliação dos indicadores de eficiência pelo modelo proposto para as técnicas DEA e SFA em conjunto com a Análise de Regressão Linear Múltipla e do teste de heterocedasticidade foi possível identificar as variáveis CAMELS que explicam a eficiência bancária das instituições analisadas Para a abordagem de intermediação dos dez indicadores CAMELS nove apresentaram resultados significativos dentre os quais apenas dois deles do grupo Management e Sensitivity to Market Risk apesar de significativos divergem para ambas as técnicas da teoria financeira que suporta este estudo Da mesma forma para a abordagem de rentabilidade dos dez indicadores CAMELS nove apresentaram resultados significativos dentre os quais apenas um deles do grupo Management apesar de significativo foi inconclusivo para ambas as técnicas em relação a teoria financeira que suporta este estudo Notase uma relação entre os indicadores deAsset Quality Earnings eLiquidity com a eficiência na abordagem de intermediação Por outro lado uma relação inversa entre os indicadores de Management e Sensibility to Market Risk relacionados a mesma abordagem O indicador de Capital Adequacy ficou inconclusivo Na abordagem de rentabilidade percebeu se uma relação positiva entre os indicadores de Capital Adequacy Earnings e Liquidity com a eficiência e resultados inconclusivos nos demais Os resultados sugerem que existem práticas relacionadas a essas variáveis que influenciam positivamente os escores de eficiência bem como práticas que influenciam negativamente os escores de eficiência o que também foi observado na revisão bibliográfica realizada Diversos pontos deste estudo cumpriram os objetivos propostos e permitiram concluir que os modelos escolhidos para suportar as técnicas DEA e SFA apoiados na adaptação do modelo de inputs e outputs realizado bem como os indicadores CAMELS aplicados na regressão são suficientes para explicarem a eficiência bancária brasileira no período deste estudo 111 Esta pesquisa tem como limitações a utilização conjunta das técnicas vez que ambas possuem modelos matemáticos diferentes Entretanto aprofundarse a essa análise seria adentrar em uma discussão de modelos matemáticos e não na análise da eficiência bancária foco primário deste estudo Embora o objetivo não tenha sido comparar as técnicas notase uma similaridade nos resultados obtidos por ambas técnicas Para este estudo considerando o país a base de dados disponível e o período analisado não se identificou diferenças significativas entre as técnicas Por isso optouse por focarse no resultado da eficiência e não nos modelos matemáticos possíveis Foi utilizado o modelo que mais se adequava à base de dados disponível Outro fator limitante concentrase nas características individuais de cada país e de seu sistema financeiro ou seja houve índices CAMELS utilizados em trabalhos internacionais cuja aplicação não foi possível no presente estudo face à divergência de padrão de informações financeiras disponível no site do Bacen O mesmo se deu para os inputs e outputs utilizados nas abordagens de eficiência Futuros trabalhos poderão aprofundarse nas análises de eficiência por meio da revisão dos modelos matemáticos utilizados para embasamento das técnicas DEA e SFA e dos inputs e outputs utilizados em cada uma das abordagens Ademais a base de dados poderá ser segmentada entre instituições públicas ou privadas em combinação entre elas ou ainda entre instituições de diferentes portes a fim de que se possam inferir conclusões sobre suas comparações ao contrário do que foi feito no presente trabalho que analisou a base que engloba todas as instituições nacionais Sugerese também para próximos estudos a revisão das variáveis CAMELS utilizadas vez que não existe um padrão para tanto ou seja diversos outros índices que se encaixam em CAMELS poderiam ser incluídos na análise de seu impacto nas variáveis dependentes que no caso consistem nos escores de eficiência calculados para DEA e SFA 112 REFERÊNCIAS BIBLIOGRÁFICAS ABEDIFAR P HASAN I TARAZI A Financegrowth nexus and dualbanking systems Relative importance of Islamic banks Journal of Economic Behavior Organization v 132 p 198215 2016 ADESINA K S Bank technical allocative and cost efficiencies in Africa The influence of intellectual capital The North American Journal of Economics and Finance v 48 p 419 433 2019 ADMATI A R HELLWIG M F Bank Leverage Welfare and Regulation Rock Center for Corporate Governance at Stanford University Working Paper No 235 Stanford University Graduate School of Business Research Paper No 3729 MPI Collective Goods Discussion Paper No 201813 httpdxdoiorg102139ssrn3257957 2019 AFFES Z HENTATIKAFFEF R Predicting US Banks Bankruptcy Logit Versus Canonical Discriminant Analysis Computational Economics v54 n1 p199244 2017 AFFINITO M ALBARETO G SANTIONI R Purchases of sovereign debt securities by Italian banks during the crisis the role of balancesheet conditions Bank of Italy Economic Research and International Relations Area n 330 p 541 2016 AHAMED M M Asset quality noninterest income and bank profitability Evidence from Indian banks Economic Modelling v 63 june 2017 p 114 2017 AIGNER D J LOVELL A K SCHMIDT P Formulation and estimation of stochastic frontier production function models Journal of Econometrics v6 n1 p2137 1977 ALESSANDRI P HALDANE A G Banking on the State Bank of England working paper Disponível em httpswwwbankofenglandcoukspeech2009bankingonthestate nov 2009 ALTUNBAS Y et al Examining the Relationships between Capital Risk and Efficiency in European Banking European Financial Management v 13 n 1 4970 2007 ASSAF NETO A Mercado Financeiro 4ed São Paulo Atlas 2011 AVKIRAN N K Association of DEA SuperEfficiency Estimates with Financial Ratios Investigating the case for Chinese banks Omega v39 n3 p323334 2011 AVKIRAN N K CAI L C Predicting bank financial distress prior to crises New Zealand Finance Colloquium 2012 AVKIRAN N LIN C Predicting bank financial distress prior to crises UQ Business School The University of Queensland Australia Working paper Disponível em httpwwwnzfcacnzarchives2012papersupdated24pdf 2012 BANKER R CHARNES A COOPER W Some Models for Estimating Technical and Scale Inefficiencies in Data Envelopment Analysis Management Science v30 n9 p10781092 1984 113 BARTH J R CAPRIO G LEVINE R Bank Regulation and Supervision in 180 Countries from 1999 to 2011 httpsdxdoiorg102139ssrn2203516 2013 BASEL COMMITTEE ON BANKING SUPERVISION Principles for the management of credit risk Basel Switzerland Bank for International Settlements 2000 Disponível em httpswwwbisorgpublbcbs75pdf Acesso em 14 fev 2021 BATISTA F D Metodologia para o Uso da Análise por Envoltória de Dados no Auxilio a Decisão Dissertação Mestrado em Engenharia de Produção Universidade Federal de Itajubá Itajubá 2009 BATTESE G E COELLI T J A Model for Technical Inefficiency Effects in a Stochastic Frontier Production Function for Panel Data Empirical Economics v20 p325332 1995 BATTESE G E CORRA G S Estimation of a production frontier model With application to the pastoral zone of Estern Australia Australian Journal of Agricultural Economics v21 n3 p169179 1977 BECKER J L LUNARDI G L MACADA A C G Análise de Eficiência dos Bancos Brasileiros Um enfoque nos investimentos realizados em Tecnologia de Informação TI Producao v13 n2 2003 BEGENAU J STAFFORD E Inefficient Banking Harvard Business School working paper p 49 dec 2016 BELTRAME F PREVITALI D SCLIP A Systematic risk and banks leverage The role of asset quality Finance Research Letters v 27 dec 2018 p 113117 2018 BERGER A N et al Bank Concentration and Competition An Evolution in the Making Journal of Money Credit and Banking v 36 n 3 pt 2 p 433451 jun 2004 BERGER A N The relationship between capital and earnings in banking Journal of Money Credit and Banking v 27 p 432456 1995 BERGER A N DEYOUNG R Problem loans and cost efficiency in commercial banks Journal of Banking Finance v 21 n 6 jun 1997 p 849870 1997 BERGER A N HUMPHREY D B Efficiency of financial institutions International survey and directions for future research European Journal of Operational Research v98 n2 p175212 1997 BERGER A N MESTER L J Inside the Black Box What Explains Differences in the Efficiencies of Financial Institutions Journal of Banking and Finance v21 p895947 1997 BERGER A BONACCORSI DI PATTI E Capital structure and firm performance A new approach to testing agency theory and an application to the banking industry Journal of Banking Finance v 30 n 4 p 10651102 2006 BERGER A HASAN I ZHOU M Bank ownership and efficiency in China What will happen in the worlds largest nation Journal of Banking and Finance v33 p113130 2009 114 BERGER A KLAPPER L TURKARISS R Bank Competition and Financial Stability Journal of Financial Services Research v 35 n 2 p 99118 apr 2009 BITTAR M PUKTHUANTHONG K WALKER T J Efficiency in Islamic vs Conventional Banking The Role of Capital and Liquidity jul 58 p httpdxdoiorg102139ssrn3423228 2019 BONGINI P NERI L PELEGATTI M The importance of being systemically important financial institutions Journal of Banking and Finance v50 n1 p562574 2015 BONIN J P HASAN I WACHTEL P Bank performance efficiency and ownership in transition countries Journal of Banking Finance v 29 n 1 p 3153 2005 BOUCINHA M RIBEIRO N WEYMANJONES T An assessment of Portuguese banks efficiency and productivity towards euro area participation Journal of Productivity Analysis v39 n2 177190 2013 BOWERMAN B L OCONNEL R T Linear statistical models An applied approach 2ed Boston PWSKent 1990 BRAGA JUNIOR J SILVA C A SILVA D E P Auditoria da Gestão da Inovação no Setor Bancário O caso do BANESE Navus Revista de Gestão e Tecnologia v5 n1 2014 BRITO R A Há espaços para melhora do setor leiteiro Uma análise de fronteira estocástica de produção e regressão quantílica utilizando dados do Censo Agropecuário 2006 IBGE Tese Doutorado em Ciências Economia Aplicada Escola Superior de Agricultura Luiz de Queiroz Universidade de São Paulo Piracicaba 2016 CANBAS S CABUK A KILIC S B Predicting of commercial bank failure via multivariate statistical analysis of financial structures The Turkish case European Journal of Operational Research v166 n2 p528546 2005 CHARNES A COOPER W W RHODES E Measuring the Efficiency of Decision Making Units European Journal of Operational Research v2 p429444 1978 CHEN S et al Bank Competition Foreign Bank Entry and RiskTaking Behavior Cross Country Evidence Journal of Risk and Financial Management v 12 n 106 p 126 2019 CHORTAREAS G E GIRARDONE C VENTOURI A Bank supervision regulation and efficiency Evidence from the European Union Journal of Financial Stability v8 n4 p292 302 2012 CHRISTOPOULOS A G MYLONAKIS J DIKTAPANIDIS P Could Lehman Brothers Collapse Be Anticipated An Examination Using CAMELS Rating System International Business Research v4 n2 p 1119 2011 COELLI T J RAO D S P BATTESE G E An introduction to efficiency and productivity analysis 3ed London Kluwer Academic Publishers 2005 COOPER W W SEIFORD L M ZHU J Handbook on Data Envelopment Analysis 2ed New York Springer US p498 2011 115 CURRY T J ELMER P J FISSEL G S Using Market information to help identify distressed institutions A regulatory perspective FDIC Bank Review v15 n3 p116 2003 DANELON A F Determinantes da eficiência técnica na agroindústria sucroenergética brasileira Uma abordagem por fronteira estocástica Dissertação Mestrado em Ciências Economia Aplicada Escola Superior de Agricultura Luiz de Queiroz Universidade de São Paulo Piracicaba 2018 DE BRANDT O LECARPENTIER S POUVELLE C Determinants of banks liquidity a French perspective on interactions between market and regulatory requirements Journal of Banking Finance v 124 working papers 782 2020 DESPOTIS D K SMIRLIS Y G Data envelopment analysis with imprecise data European Journal of Operational Research v140 n1 p2436 2002 DeYOUNG R Management quality and Xinefficiency in National Banks Journal of Financial Services Research v13 n1 p522 1998 DINCER H et al A performance evaluation of the Turkish banking sector after the global crisis via CAMELS ratios Procedia Social and Behavioral Sciences v24 n1 p 15301545 2011 DJALILOV K PIESSE J Bank regulation and efficiency Evidence from transition countries International Review of Economics Finance v 64 p 308322 2019 DONG Y et al Evaluating the performance of Chinese commercial banks A comparative analysis of different types of banks European Journal of Operational Research v252 n1 p280295 2016 DRAKE L Efficiency in UK Banking Applied Financial Economics v11 pp557571 2001 DRAKE L HALL M J B SIMPER R The Impact of Macroeconomic and Regulatory Factors on Bank Efficiency A nonparametric analysis of Hong Kongs banking system Journal of Banking and Finance v30 n5 p14431466 2006 DRAPER N R SMITH H Applied regression analysis 2ed New York John Wiley and Sons 1981 DUAN J ZHANG H Pricing Hang Sang Index Options Around The Asian Financial Crises A GARCH Approach Journal of Banking and Finance v 25 n 11 p 19892014 2001 EKINCI R POYRAZ G The Effect of Credit Risk on Financial Performance of Deposit Banks In Turkey Procedia Computer Science v 158 p 979987 2019 END J W KRUIDHOF M Modelling the liquidity ratio as macroprudential instrument Journal of Banking Regulation v 14 n 2 p 106 2013 FARRELL M The Measurement of Productive Efficiency Journal of the Royal Statistical Society v120 n3 p253290 1957 116 FEDERAL RESERVE SYSTEM Board of Governors of the Federal Reserve System Division of Banking Supervision and Regulation Uniform Financial Institutions Rating System SR 9638 SUP 26 de dezembro 1996 FERNANDES F D S STASINAKIS C BARDAROVA V Twostage DEATruncated Regression Application in Banking Efficiency and Financial Development Expert Systems with Applications v 96 n 15 apr p 284301 2018 FETHI M D PASIOURAS F Assessing Bank Efficiency and Performance with Operational Research and Artificial Intelligence Techniques A survey European Journal of Operational Research v204 n2 p189198 2010 FIORDELISI F MARQUESIBANEZ D MOLYNEUX P Efficiency and risk in European banking Journal of Banking and Finance v35 n5 p13151326 2011 FONSECA J J S Metodologia da Pesquisa Científica Fortaleza UEC 2002 FORGIONE A F MIGLIARDO C Forecasting distress in cooperative banks The role of asset quality International Journal of Forecasting v 34 n 4 p 678695 2018 FREAZA F P MADEIRO G L E GOMES L F A M Analise de Eficiência do Mercado Bancário Brasileiro Utilizando a Metodologia da Analise Envoltoria de Dados XXXVIII Simpósio Brasileiro Pesquisa Operacional Anais 2004 FRIES S TACI A Cost efficiency of banks in transition Evidence from 289 banks in 15 postcommunist countries Journal of Banking Finance v 29 n 1 p 5581 2005 FRIOSI J et al Análise Exploratória da Inovação Bancária Brasileira e as Tendências Para o Setor Revista de Gestão Empresarial v1 n1 2017 FU X M HEFFERNAN S The effects of reform on Chinas bank structure and performance Journal of Banking and Finance v33 n1 p3952 2009 FUJI H MANAGI S MATOUSEK R Indian Bank Efficiency and Productivity Changes with Undesirable Outputs A Disaggregated Approach Journal of Banking and Finance v 38 p 4150 2014 FUNGACOVA Z PESSAROSSI P WEILL L Is Bank Competition Detrimental to Efficiency Evidence from China China Economic Review v 27 dec BOFIT Discussion Paper No 312012 httpdxdoiorg102139ssrn2206860 2013 GAGANIS C PASIOURAS F Financial supervision regimes and bank efficiency International evidence Journal of Banking Finance v 37 n 12 p 54635475 2013 GALTON F Regression toward mediocrity in heredity stature Journal of Anthropological Institute v15 p246263 1885 GIL A C Métodos e Técnicas de Pesquisa Social 5 ed São Paulo Atlas 1999 GODDARD J MOLYNEUX P WILSON J O S The profitability of european banks a crosssectional and dynamic panel analysis The Manchester School v 72 p 363381 2004 117 GODOI A F et al Análise das Variáveis Macroeconômicas e do acrônimo CAMELS sobre o retorno das ações ordinárias nas Instituições Financeiras nacionais de grande porte Revista de Administração de Roraima v6 n1 p4770 2016 GOMES R M M P Análise de Crédito a Instituições Financeiras A metodologia CAMELS Dissertação Mestrado em Economia e Gestão Instituto Superior de Economia e Gestão Universidade Técnica de Lisboa Lisboa 2012 GRAVELLE T LI F Measuring systemic importance of financial institutions An extreme value theory approach Journal of Banking and Finance v 37 n7 p21962209 2013 GREENE W H Econometric analysis 5ed New Jersey Prentice Hall 2002 959p GRIGORIAN D A MANOLE V Determinants of Commercial Bank Performance in Transition An application of Data envelopment analysis Comparative Economic Studies v48 n3 p497522 2006 GUERREIRO A S Analise da Eficiência de Empresas de Comercio Eletronico Usando Tecnicas da Analise Envoltoria de Dados Dissertação Mestrado em Engenharia Industrial Pontifica Universidade Católica do Rio de Janeiro Rio de Janeiro 2006 HAIR J F et al Análise Multivariada de Dados 5 ed Porto Alegre Bookman 2005 INTERNATIONAL MONETARY FUND Guia de Compilación dos Indicadores de Solidez Financiera Washington FMI 2006 JEELANBASHA V JEELANBASHA V Comparative performance analysis of selected banks using CAMEL model Journal of Business and Management v20 n6 p8793 2018 JIANG C YAO S FENG G Bank ownership privatization and performance Evidence from a transition country Journal of Banking and Finance v37 n9 p33643372 2013 JUBRAN A J Modelo de Analise de Eficiência na Administracao Publica Estudo aplicado as prefeituras Brasileiras usando a analise envoltoria de dados Tese Doutorado em Engenharia Escola Politécnica Universidade São Paulo São Paulo 2006 KHAN M N et al Efficiency measurement of Islamic and conventional banks in Saudi Arabia an empirical and comparative analysis AlShajarah Journal of the International Institute of Islamic Thought and Civilization ISTAC dec p 11134 2018 KISGEN D J The impact of credit ratings on corporate behavior Evidence from Moodys adjustments Journal of Corporate Finance v 58 n C p 567582 2019 KOCH T W MACDONALD S S Bank Management 7 ed Boston SouthWestern Cengage Learning 2014 KORONTAI J N Governança corporativa dos bancos e sua relação com indicadores de desempenho e risco Dissertação Mestrado em Contabilidade Universidade do Paraná Curitiba 2016 118 KUMBHAKAR S C GHOSH S MCGUCKIN J T A generalized production frontier approach for estimating determinants of inefficiency in US Dairy farms Journal of Business and Economic Statitics v9 n1 p279286 1991 KUMBHAKAR S C LIEN G HARDAKER J B Technical efficiency in competing panel data models a study of Norwegian grain farming Journal of Productivity Analysis v 41 p 321337 2012 KUMBHAKAR S C LOVELL C A K Stochastic Frontier Analysis 2ed Cambridge University Press Cambridge 2003 KWAN S EISENBEIS R Bank Risk Capitalization and Operating Efficiency Journal of Financial Services Research v 12 n 2 p 117131 1997 LENSINK R MEESTERS A NAABORG I Bank efficiency and foreign ownership Do good institutions matter Journal of Banking and Finance v32 n5 p834844 2008 LOZANOVIVAS A PASIOURAS F The impact of nontraditional activities on the estimation of bank efficiency International evidence Journal of Banking Finance v 34 n 7 p 14361449 2010 LUO Y BI G LIANG L Inputoutput indicator selection for DEA efficiency evaluation An empirical study of Chinese commercial banks Expert Systems With Applications v39 n1 p11181123 2012 MACORIS L S SALGADO JUNIOR A P FALSARELLA JUNIOR E The different approaches of banking efficiency a MetaAnalysis International Conference on Data Envelopment Analysis 13th edition 2015 Braunschweig Germany MAGRO C B D MICHELS A SILVA T P Análise da eficiência no desempenho financeiro das cooperativas de crédito brasileiras Revista Brasileira de Gestão e Desenvolvimento Regional v13 n2 p73102 2017 MALHOTRA N K ROCHA I LAUDISIO M C ALTHEMAN E BORGES F M Introdução a Pesquisa de Marketing São Paulo Pearson 2005 MALIKOV E HUMBHAKAR S C SUN Y Varying coefficient panel data model in the presence ofendogenous selectivity and fixed effects Journal of Econometrics v 190 p 233 251 2016 MARCUSSO M A Conflito entre as abordagens de rentabilidade intermediação e produção em bancos corporate brasileiros uma análise DEA dois estágios entre 1996 e 2015 Dissertação Mestrado em Administração das Organizações Faculdade de Economia Administração e Contabilidade de Ribeirão Preto Universidade de São Paulo Ribeirão Preto 2017 MARTEL M VAN RIXTEL A MOTA E G Business models of international banks in the wake of the 20072009 global financial crisis Banco de España Revista de Estabilidad Financiera n 22 p 99121 2012 119 MAUDOS J FERNANDEZ DE GUEVARA J Factors explaining the interest margin in the banking sectors of the European Union Journal of Banking Finance v 28 N 9 p 2259 2281 sept 2004 MEEUSEN W VAN DEN BROECK J Efficiency Estimation from CobbDouglas Production Functions with Composed Error International Economic Review v18 n2 p435444 1977 MERGAERTS F VENNET R V Business models and bank performance A longterm perspective Journal of Financial Stability v 22 n C p 5775 2016 MERILÄINEN JM JUNTTILA J The relationship between credit ratings and asset liquidity Evidence from Western European banks Journal of International Money and Finance v 108 november 2020 p 121 2020 MILI M et al Determinants of the capital adequacy ratio of foreign banks subsidiaries The role of interbank market and regulation Research in International Business and Finance v 42 p 442453 2017 MILLON M H THAKOR A V Moral Hazard and Information Sharing A Model of Financial Information Gathering Agencies The Journal of Finance v 40 p 14031422 1985 MOSTAFA M Modeling the Efficiency of Top Arab Banks A DEAneural network approach Expert Systems with Applications v 6 n1 p 309320 2009 NETER J WASSERMAN W Applied linear statistical models1ed Homewood III Richard D Irwin 1974 NIGMONOV A Bank Performance and Efficiency in Uzbekistan Eurasian Journal of Business and Economics v3 n5 p125 2010 ODECK J MOLDE S B A metaanalysis of DEA and SFA studies of the technical efficiency of seaports A comparison of fixed and randomeffects regression models Transportation Research Part A Policy and Practice v46 n10 p15741585 2012 OTERO L et al What determines efficiency in MENA banks Journal of Business Research v 112 p 331341 2020 OTHMAN N ABDULMAJID M ABDULRAHMAN A Partnership financing and bank efficiency PacificBasin Finance Journal v 46 n PA p 113 2017 PARADI J C SCHAFFNIT C Commercial Branch Performance Evaluation and Results Communication in a Canadian Bank A DEA application European Journal of Operational Research v156 n3 p719735 2004 PARADI J C ZHU H EDELSTEIN B Identifying Managerial Groups in a Large Canadian Bank Branch Network with a DEA Approach European Journal of Operational Research v219 n1 p178187 2012 PARETO V The New Theories of Economics Journal of Political Economy v5 n 4 p 485502 1897 120 PENNACCHI G G SANTOS J A C Why do banks target ROE Journal of Financial Stability v 54 june 2021 in press 2021 PERICO A E REBELATTO D A DO N SANTANA N B Eficiência Bancária Os maiores bancos são os mais eficientes Uma análise por envoltória de dados Gestão Produção v15 n2 p421431 2008 PEYDRÓ JL Ç POLO A SETTE E Monetary policy at work Security and credit application registers evidence Journal of Financial Economics in press doiorg101016jjfineco202101008 2021 PILLE P PARADI J C Financial performance analysis of Ontario Canada Credit Unions An application of DEA in the regulatory environment European Journal of Operational Research v139 p339350 2002 PIMENTA A C Contribuição para a análise do desempenho das instituições financeiras a partir dos indicadores da carteira de crédito Dissertação Mestrado em Administração de Empresas Pontifícia Universidade Católica São Paulo 2014 PONTES T T S Determinantes da eficiência técnica no setor de supermercados brasileiro uma análise utilizando métodos de fronteira de produção Tese Doutorado em Administração das Organizações Faculdade de Economia Administração e Contabilidade de Ribeirão Preto Universidade de São Paulo Ribeirão Preto 2017 PRUTEANUPODPIERA A WEILL L SCHOBERT F Banking Competition and Efficiency A MicroData Analysis on the Czech Banking Industry Comparative Economic Studies v 50 n 2 p 253273 2008 REIFSCHNEIDER D STEVENSON R Systematic departures from the frontier A framework for the analysis of firm inefficiency International Economic Review v32 n3 p715723 1991 ROENGPITYA R TARASHEV N TSATSARONIS K Bank business models BIS Quarterly Review dec 2014 p 5565 2014 ROMAN A SARGU A C Analysing the financial soundness of the commercial banks in Romania An approach based on the CAMELS framework Procedia Economics and Finance v6 n1 p703712 2013 ROSA P GARTNER I Financial distress em bancos brasileiros Um modelo de alerta antecipado Revista Contabilidade Finanças v29 n77 p312331 2018 ROSMAN R WAHAB N A ZAINOL Z Efficiency of Islamic banks during the financial crisis An analysis of Middle Eastern and Asian countries PacificBasin Finance Journal v 28 p 7690 2014 ROSSI S P SCHWAIGER M S WINKLER G How loan portfolio diversification affects risk efficiency and capitalization A managerial behavior model for Austrian banks Journal of Banking and Finance v33 n12 p22182226 2009 SAEED M et al The intertemporal relationship between risk capital and efficiency The case of Islamic and conventional banks PacificBasin Finance Journal v 62 n C p 122 2020 121 SAHAJWALA VAN DEN BERGH P Supervisory Risk Assessment and Early Warning Systems Bank for International Settlements Basel Switzerland BIS Press Office 2000 SAMPIERI R H COLLADO C F LUCIO P B Metodologia de pesquisa 3 ed São Paulo MacGrawHill 2006 SARMIENTO M GALÁN J E The influence of risktaking on bank efficiency Evidence from Colombia Emerging Markets Review v 32 n C p 5273 2017 SHADDADY A MOORE T Investigation of the effects of financial regulation and supervision on bank stability The application of CAMELSDEA to quantile regressions Journal of International Financial Markets Institutions and Money v 58 2018 SHAMSHUR A WEILL L Does bank efficiency influence the cost of credit Journal of Banking Finance v 105 n C p 6273 2019 SILVA T P et al Financial and economic performance of major Brazilian credit cooperatives Contaduría y Administración v 62 n 5 p 14421459 2017 SIRIOPOULOS C TZIOGKIDIS P How do Greek banking institutions react after significant events A DEA approach Omega v38 n5 p294308 2010 STAIKOURAS C MAMATZAKIS E KOUTSOMANOLIFILIPPAKI A Cost efficiency of the banking industry in the South Eastern European region Journal of International Financial Markets v18 n5 p483497 2008 STAUB R B SOUZA G TABAK B M Evolution of Bank Efficiency in Brazil A DEA approach European Journal of Operational Research v202 n1 p204213 2010 SUFIAN F HABIBULLAH M S Bank specific and macroeconomic determinants of bank profitability Empirical evidence from the China banking sector Frontiers of Economics in China v 4 p 274291 2009 SUFIAN F HABIBULLAH M S Globalizations and bank performance in China Research in International Business and Finance v 26 n 2 may p 221239 2012 TABAK B M FAZIO D M CAJUEIRO D O The relationship between banking market competition and risktaking Do size and capitalization matter Journal of Banking and Finance v36 n12 p33663381 2012 TABAK B TECLES P Estimating a bayesian stochastic frontier for the Indian banking system International Journal of Production Economics v125 p96110 2010 TAN Y The impacts of risk and competition on bank profitability in China Journal of International Financial Markets Institutions and Money v 40 p 85110 2016 TAN Y FLOROS C Risk profitability and competition evidence from the Chinese banking industry Journal of Developing Areas v 48 n 3 p 303319 2014 THANASSOULIS E Introduction to the Theory and Application of Data Envelopment Analysis A Foundation Text with Integrated Software Springer 2003 122 THANASSOULIS E BOUSSOFIANE A DYSON R G A Comparison of Data Envelopment Analysis and Ratio Analysis as Tools for Performance Assessment Omega International Journal of Management Science v24 n3 p229244 1996 THEODORIDIS A M ANWAR M A comparison of DEA and SFA methods a case study of farm households in Bangladesh Journal of Developing Areas v 45 N 1 p 95110 2011 VERGARA S C Projetos e Relatórios de Pesquisa em Administração São Paulo Atlas 2006 WANKE P Determinants of scale efficiency in the Brazilian thirdparty logistics industry from 2001 to 2009 Brazilian Administration Review v9 n1 p 6687 2012 WU JH HSIA TL HENG M S H Core capabilities for exploiting electronic banking Journal of Electronic Commerce Research v 7 n 2 p 112122 2006 YANG J B et al Integrated Bank Performance Assessment and Management Planning Using Hybrid Minimax Reference Point DEA approach European Journal of Operational Research v207 n3 p15061518 2010 YANG X et al Systemic importance of financial institutions A complex network perspective Physica A Statistical Mechanics and its Applications v545 n1 p111 2020 YANG Z Bank Branch Operating Efficiency A DEA Approach International Multi Conference of engineers and Computer Scientists Vols I and II Anais 2009 YIN H The impact of competition and bank market regulation on banks cost efficiency Journal of Multinational Financial Management 100677 in press 2021 YIN H YANG J LU X Bank globalization and efficiency Host and homecountry effects Research in International Business and Finance v 54 dec 2020 ZANINI A Regulação econômica no setor elétrico brasileiro uma metodologia para definição de fronteiras de eficiência e cálculo do fator x para empresas distribuidoras de energia elétrica Tese Doutorado em Engenharia Elétrica Pontifícia Universidade Católica do Rio de Janeiro Rio de Janeiro 2004 ZHU J Imprecise data envelopment analysis IDEA A review and improvement with an application European Journal of Operational Research v144 n3 p513529 2003 Electronic copy available at httpssrncomabstract1350544 Electronic copy available at httpssrncomabstract1350544 Assessing Bank Performance with Operational Research and Artificial Intelligence Techniques A Survey Meryem DuygunFethi School of Management University of Leicester Fotios Pasiouras School of Management University of Bath University of Bath School of Management Working Paper Series 200902 This working paper is produced for discussion purposes only The papers are expected to be published in due course in revised form and should not be quoted without the authors permission Electronic copy available at httpsssrncomabstract1350544 Electronic copy available at httpssrncomabstract1350544 Electronic copy available at httpssrncomabstract1350544 2 University of Bath School of Management Working Paper Series School of Management Claverton Down Bath BA2 7AY United Kingdom Tel 44 1225 826742 Fax 44 1225 826473 httpwwwbathacukmanagementresearchpapershtm 2009 200901 Androniki Apostolakou Gregory Jackson Corporate Social Responsibility in Western Europe An Institutional Mirror or Substitute 200902 Meryem Duygun Fethi Fotios Pasiouras Assessing Bank Performance with Operational Research and Artificial Intelligence Techniques A Survey Electronic copy available at httpsssrncomabstract1350544 3 Assessing Bank Performance with Operational Research and Artificial Intelligence Techniques A Survey Meryem DuygunFethi1 Fotios Pasiouras2 1School of Management University of Leicester UK 2School of Management University of Bath UK Abstract This paper presents a comprehensive review of 179 studies which employ operational research OR and Artificial Intelligence AI techniques in the assessment of bank performance We first discuss numerous applications of data envelopment analysis which is the most widely applied OR technique in the field Then we discuss applications of other techniques such as neural networks support vector machines and multicriteria decision aid that have also been used in recent years in bank failure prediction studies and the assessment of bank creditworthiness and underperformance Keywords Artificial Intelligence Banks Data Envelopment Analysis Operational Research Literature review Copyright The Authors 2009 Author for correspondence Tel 44 0 116 252 5328 Fax 44 0 116 252 5515 Emails mfethileacuk M DuygunFethi fpasiourasbathacuk F Pasiouras Electronic copy available at httpsssrncomabstract1350544 4 1 Introduction Banks play a central role in the economy They keep the savings of the public and finance the development of business and trade Furthermore numerous studies argue that the efficiency of financial intermediation affects economic growth while others indicate that bank insolvencies can result in systemic crises which have adverse consequences for the economy as a whole1 Thus the performance of banks has been an issue of major interest for various stakeholders such as depositors regulators customers and investors While bank performance has been traditionally evaluated on the basis of financial ratios advances in operational research OR and artificial intelligence AI have resulted in a shift towards the use of such stateoftheart techniques Of course this is not surprising since OR has been extensively applied to finance during the last half century Board et al 2003 This paper presents a comprehensive review of the use of OR and AI techniques in the assessment of bank performance The rest of the paper is structured as follows Section 2 positions the survey within the existing literature and discusses our framework Section 3 discusses applications of data envelopment analysis DEA in the estimation of bank efficiency and productivity growth Section 4 presents applications of other OR and AI techniques in the prediction of bank failure and the assessment of bank creditworthiness and underperformance Section 5 summarizes our conclusions 2 Scopus and framework There are several interesting reviews that are related to our survey For example Cook and Seiford 2009 review the methodological developments of DEA over the last thirty years However they do not discuss applications of DEA Zhou and Poh 2008 provide a recent survey of DEA applications but they focus on energy and environmental studies Dimitras et al 1996 discuss applications of various techniques in the prediction of business failures but they focus on industrial firms Smith and Gupta 2000 provide a discussion of the application of neural networks in business problems Board et al 2003 survey OR applications in the financial markets Thus the above surveys are either quite general or they do not focus on applications in banking 1 See Levine 2005 for a discussion of the theoretical and empirical discussion of the literature on finance and growth See Caprio and Klingebiel 2003 for banking crises and the associated costs Electronic copy available at httpsssrncomabstract1350544 5 Berger and Humphrey 1997 review studies that examine the efficiency of financial institutions However their coverage is limited to efficient frontier techniques eg DEA stochastic frontier analysis Furthermore the survey is now more than ten years old and since that time numerous papers have been published Berger 2007 discusses more recent applications of frontier techniques but his survey focuses only on studies that provide international comparisons of bank efficiency Finally Ravi Kumar and Ravi 2008 discuss applications of statistical and AI techniques in bankruptcy prediction The applications that they survey were published until 2005 and although a few of them focus on the banking sector most of the studies deal with nonfinancial firms We differentiate our review from the above surveys by discussing applications of OR and AI techniques over the period 19982008 while focusing on bank performance2 We searched for papers in Scopus which is considered to be one of the largest abstract and citation databases We consider only journal articles and we do not include working papers monographs dissertations or other publication outcomes Furthermore our search is limited to articles written in English We use a combination of various keywords such as bank efficiency bank and data envelopment analysis bank performance bank and neural networks bank and artificial intelligence bank and operational or operations research A few additional studies were identified from crossreferencing and were manually collected We reviewed a total of 179 studies DEA is by far the most commonly used ORAI technique in assessing bank performance and we identified 136 studies that use DEAlike techniques to estimate various measures of bank efficiency and productivity growth and 28 studies that provide similar estimates at the branch level3 We also identified 15 studies that use classification techniques such as neural networks support vector machines multicriteria decision aid decision trees nearest neighbours to predict bank failure or assess bank creditworthiness and bank underperformance These studies were published in a total of 67 journals however around 56 of them appeared in just eleven journals As shown in Table 1 the most 2 Some of the studies were still in press at the time of the writing of this review but we included them in our discussion as they were already publicly available since 2008 ie 2008 DOI 3 Our survey focuses on studies that examine banking institutions as a whole however we also discuss studies on branch efficiency in section 327 as one could argue that the efficiency of individual branches can influence the performance of banks as a whole Electronic copy available at httpsssrncomabstract1350544 6 frequent sources of publication are the European Journal of Operational Research 18 and the Journal of Banking and Finance 15 followed by Applied Financial Economics 13 Applied Economics 9 Expert Systems with Applications 9 and the Journal of Productivity Analysis 9 In the sections that follow we discuss various issues surrounding these studies while additional information is available in Appendices I to III Insert Table 1 Around Here 3 DEA and bankefficiency DEA is a mathematical programming technique for the development of production frontiers and the measurement of efficiency relative to these frontiers Charnes et al 1978 The bestpractice production frontier for a sample of banks is constructed through a piecewise linear combination of actual inputoutput correspondence set that envelops the inputoutput correspondence of all banks in the sample Thanassoulis 2001 Each bank is assigned an efficiency score between 0 and 1 The scores are only relative to the banks in the sample with higher scores indicating a more efficient bank One of the wellknown advantages of DEA is that it works relatively well with small samples Other advantages of DEA are that it does not require any assumptions to be made about the distribution of inefficiency and it does not require a particular functional form on the data in determining the most efficiency banks However DEA is also subject to few limitations Two of the best known shortcomings are that DEA assumes data to be free of measurement error and that it is sensitive to outliers Coelli et al 2005 also point out that i having few observations and many inputs andor outputs will result in many firms appearing on the DEA frontier ii treating inputsoutputs as homogenous commodities when they are heterogeneous may bias the results iii not accounting for differences in the environment may give misleading results iv standard DEA does not control for multiperiod optimization or risk managerial decision making Our survey shows that recent DEA studies have examined almost all the banking sectors around the world A few recent studies provide crosscountry evidence Most of them examine banks from the large EU banking sectors Pastor 2002 Casu and Molyneux 2003 Beccalli et al 2006 LozanoVivas et al 2002 Electronic copy available at httpsssrncomabstract1350544 7 examine ten EU countries Bergendahl 1998 focuses on Nordic countries while Pasiouras 2008a examines an international dataset 31 Methodological issues 311 Efficiency measures Most of the studies focus on the technical efficiency of banks eg LozanoVivas et al 2002 Drake et al 2006 Pasiouras 2008ab This efficiency measure indicates whether a bank uses the minimum quantity of inputs to produce a given quantity of outputs or maximizes the output quantity given a certain quantity of inputs However when price data for the inputs andor outputs are available one can also estimate cost andor profit efficiency measures4 Cost efficiency is the product of technical efficiency and allocative efficiency The latter refers to the ability of a bank to use the optimum mix of inputs given their respective prices Consequently cost efficiency shows the ability of a bank to provide services without wasting resources as a result of technical or allocative inefficiency Appendix I shows that around 35 studies present measures of DEA cost efficiency eg TortosaAusina 2002abc Isik and Hassan 2002 2003a Pastor and Serrano 2006 propose the decomposition of cost inefficiency into composition inefficiency and intraspecialisation inefficiency The first component indicates the part of inefficiency due to the composition of specialisations of the banks in each banking sector The second component reveals the inefficient use of resources within each of the specialisation selected Prior 2003 also deviates from the above studies by calculating measures of short and longrun cost inefficiency as well as capacity inefficiency for Spanish banks The first refers to the case that a subset of inputs are fixed and impossible to modify in the shortrun Longrun inefficiency estimates are obtained under the assumption that inputs are variable and under the control of the company Finally capacity inefficiency obtained by the ratio of longrun to shortrun inefficiency refers to excess in costs as a result of inappropriate level in fixed inputs Similar concepts along with an application in the Indian banking sector are discussed in Sahoo and Tone 2008 4 One can also estimate revenue efficiency which is similar to profit efficiency In both cases both inputs and output prices are required The difference is that in the former measure the aim is to maximize revenues rather than profits ie revenues minus costs We are no aware of DEA studies focusing on revenue efficiency so we do not discuss this issue further Readers interesting in revenue efficiency could see Coelli et al 2005 for further details Electronic copy available at httpsssrncomabstract1350544 8 Estimations of profit efficiency with DEA are rather limited in the literature One potential reason is the difficulty in collecting reliable and transparent information for output prices Furthermore the decomposition of profit efficiency into technical and allocative efficiency is not straightforward Coelli et al 2005 Fare et al 2004 propose the solution of two sets of linear programmes In the first a profit maximising DEA is solved to measure profit efficiency In the second DEA problem technical efficiency is measured on the basis of a directional distance function that allows the simultaneous adjustment of inputs and outputs Kirkwood and Nahm 2006 also estimate profit efficiency although they use input prices only Therefore in a sense they calculate a measure of efficiency that is similar to Berger and Mesters 1997 alternative profit efficiency which is commonly used in the stochastic frontier analysis literature The studies of Maudos and Pastor 2003 and Ariff and Can 2008 provide estimates of both standard and alternative profit efficiency Finally around 30 of the studies obtain estimates of total factor productivity TFP growth eg Sathye 2002 Casu et al 20045 This measure is usually decomposed further into technological change ie shift in the best practice frontier and technical efficiency change Furthermore in most cases technical efficiency change is disaggregated into pure technical efficiency change and scale efficiency change under the variable returns to scale assumption discussed below 312 Constant vs Variable Returns to scale DEA can be implemented by assuming either constant returns to scale CRS or variable returns to scale VRS In their seminal study Charnes et al 1978 proposed a model that had an input orientation and assumed CRS This model returns a score that indicates the overall technical efficiency OTE of each bank Banker et al 1984 suggested the use of variable returns to scale VRS that decomposes OTE into product of two components pure technical efficiency PTE and scale efficiency SE The former relates to the ability of managers to utilize firms given resources while the latter refers to exploiting scale economies by operating at a point where the production frontier exhibits CRS 5 Almost all the studies use the DEAlike Malmquist index Zen and Baldan 2008 use the Luenberger Indicator that is a generalization of the Malmquist Index Electronic copy available at httpsssrncomabstract1350544 9 In most of the recent papers DEA models are estimated using the assumption of VRS while arguing that CRS is only appropriate when all firms are operating at an optimal scale6 Nevertheless other studies argue in favour of CRS rather than VRS For example Noulas 1997 points out that the assumption of CRS allows the comparison between small and large banks He claims that in a sample where a few large banks are present the use of VRS framework raises the possibility that these large banks will appear as being efficient for the simple reason that there are no truly efficient banks Berg et al 1991 Avkiran 1999 also mentions that under VRS each unit is compared only against other units of similar size instead of against all units Hence the assumption of VRS may be more suitable for large samples Soteriou and Zenios 1999a argue that caution is necessary when using the VRS formulation First because the model orientation ie input minimization or output maximization becomes important Second because the use of the weights restriction in the VRS assessment may lead to some other problematic results Allen 1997 Consequently many studies report the results obtained under both CRS and VRS assumptions eg Canhoto and Dermine 2003 Casu and Molyneux 2003 313 Outputinput orientation Technical Efficiency can be estimated under either an inputoriented or output oriented approach7 As Coelli et al 2005 point out the inputoriented technical efficiency measures address the question By how much can input quantities be proportionally reduced without changing the output quantities produced p 137 In contrast the outputoriented measures of technical efficiency address the question By how much can output quantities be proportionally expanded without altering the input quantities used p 137 By far studies in banking obtain efficiency estimates under the inputoriented approach8 This is most likely due to the assumption that bank managers have higher 6 Reasons that may not allow a firm to operate at optimal scale include among others imperfect competition government regulations constrains on finance etc Coelli et al 2005 7 If price data are available then under the under cost minimization objective ie cost efficiency one obtains inputoriented measures of technical efficiency and inputmix allocate efficiency Revenue maximization ie revenue efficiency results is outputoriented technical efficiency and outputmix allocative efficiency measures In the case of profit maximization profit efficiency technical efficiency can be obtained under either the input or outputoriented assumption Coelli et al 2005 8 Additional information is available in Appendix I In contrast to efficiency estimates productivity measures are in several cases obtained using an outputoriented Malmquist index However as discussed in Coelli et al 2005 p 80 although the values of the components and consequently their Electronic copy available at httpsssrncomabstract1350544 10 control over inputs eg personnel expenses rather than outputs eg loans income etc However there are also some studies that adopt the outputoriented approach eg Ataullah et al 2004 Ataullah and Le 2006 or report the results from both eg Casu and Molyneux 2003 Beccalli et al 2006 It should be mentioned that the inputoriented and outputoriented measures always provide the same value under CRS but they are unequal when VRS is assumed However Coelli et al 2005 mention that since linear programming does not suffer from statistical problems the choice of an appropriate orientation is not as important as in the case of econometric approaches Furthermore in many instances the choice of orientation has only a minor influence upon the scores obtained Coelli and Perelman 1996 314 Selection of Inputs and Outputs There is an ongoing discussion in the banking literature regarding the proper definition of inputs and outputs In the words of Bergendahl 1998 There have been almost as many assumptions of inputs and outputs as there have been applications of DEA p 235 Berger and Humphrey 1997 identify two main approaches for the selection of inputs and outputs These are the production approach and the intermediation approach The first assumes that banks produce loans and deposits account services using labour and capital as inputs and that the number and type of transactions or documents processed measure outputs The second approach perceives banks as financial intermediaries between savers and investors Berger and Humphrey 1997 argue that neither of these two approaches is perfect because they cannot fully capture the dual role of financial institutions as providers of transactionsdocument processing services and also being financial intermediaries They point out that the production approach may be somewhat better for evaluating the efficiencies of bank branches and the intermediation approach may be more appropriate for evaluating financial institutions as a whole Furthermore there are difficulties in collecting the detailed transaction flow information required in the production approach As a result the intermediation approach is the one favoured in the literature However there is a controversy even within this approach concerning the role of deposits Berger and Humphrey 1997 Consequently some studies use only contribution to overall productivity change may differ the overall TFP change measure will be the same regardless of the orientation ie input or output that is imposed Electronic copy available at httpsssrncomabstract1350544 11 earning assets as outputs a selection that is in line with the asset approach of Sealey and Lindley 1977 while others consider deposits as an additional output a selection that is more closely related to the socalled valueadded approach We find around eighty applications in bank efficiency where the monetary value of deposits is part of the input vector and twenty applications where deposits are part of the output vector9 Around thirty studies use interest expenses as an input without using the stock of deposits eg Sathey 2002 Weill 2004 In another eight applications the stock of deposits is used as an output and the interest expense paid on deposits constitutes an input eg Maudos et al 2002 Saha and Ravisankar 2000 Chen et al 2005 Furthermore around five studies use time deposits and saving deposits as input and demand deposits as output eg Bauer et al 1998 Gilbert and Wilson 1998 Sathye 2001 Finally in a few applications the deposits are included as both an input and an output eg TortosaAusina 2002a More recently some studies have adopted another variation of the intermediation approach This is the socalled profitoriented or operating approach which defines revenue components eg interest income noninterest income etc as outputs and cost components eg personnel expenses interest expenses etc as inputs10 Drake et al 2006 mention that from the perspective of an inputoriented DEA relative efficiency analysis the more efficient units will be better at minimizing the various costs incurred in generating the various revenue streams and consequently better at maximizing profits p 1451 They also argue that this approach can be more appropriate in capturing the diversity of strategic responses by financial firms in the face of dynamic changes in competitive and environmental conditions Furthermore Luo 2003 calculates a measure of marketability efficiency in an attempt to capture the value of the bank in the stock market In this case revenue and profits are considered inputs whereas market value earnings per share and stock price are outputs As discussed before with the exception of deposits there is a general agreement about the main categories of inputs and outputs however this does not 9 This discussion refers to banks as a whole and not branches The number of applications does not match the number of studies as in several cases there are numerous models developed in each study Furthermore in some cases deposits or non interest expenses are not considered in the analysis In one case Prior 2003 the output was the number of current and saving accounts rather than the monetary value of deposits ie stock 10 See Chu and Lim 1998 Avkiran 1999 Sturm and Williams 2004 Das and Ghosh 2006 Drake et al 2006 Ataullah and Le 2006 Pasiouras 2008b Pasiouras et al 2008 Electronic copy available at httpsssrncomabstract1350544 12 necessarily imply that there is consistency with respect to the specific inputsoutputs used in various studies For instance the traditional inputs are fixed assets personnel11 and in many cases deposits eg Isik and Hassan 2002 Maudos and Pastor 2003 Casu and Girardone 2004 Havrylchyk 2006 However some studies use branches eg Chen 2001 loan loss provisions eg Drake 2001 Drake et al 2006 Pasiouras 2008b and equity eg Chu and Lim 1998 Mukherjee et al 2001 Sturm and Williams 2004 Pasiouras 2008a as additional or alternative inputs Chen 2001 disaggregates deposits into current deposits and time deposits while Das and Ghosh 2006 use demand savings and fixed deposits Casu and Girardone 2006 and Beccalli et al 2006 use total costs as a single input while Casu and Molyneux 2003 use two inputs namely total costs and total deposits ie customers and short term funding Several studies use two outputs usually loans and other earning assets eg Casu and Molyneux 2003 Casu and Girardone 2004 200612 However some studies disaggregate loans into various categories such as housing loans and other loans eg Sturm and Williams 2004 real estate loans commercial loans and personal loans eg Mukherjee et al 2001 Fare et al 2004 or shortterm and long term loans Isik and Hassan 2002 Others disaggregate other earning assets into investments and liquid assets Tsionas et al 2003 or investment in government securities and investments in public and private enterprises eg Chen 2001 Canhoto and Dermine 2003 use the number of branches as an additional output under the assumption that it represents an additional value for retail customers Finally recent studies include noninterest income or offbalancesheet items as additional outputs eg Isik and Hassan 2002 2003a Sturm and Williams 2004 TortosaAusina 2003 Havrylchyk 2006 Pasiouras 2008b13 Finally Halkos and Salamouris 2004 propose an approach that deviates from the above literature They use an output vector that consists of five financial ratios and no inputs Their underlying hypothesis is that inputs are considered similar and equal 11 Some studies use the number of personnel eg Drake 1999 Sathye 2001 Maudos and Pastor 2003 Pasiouras 2008b However due to data unavailability others rely on personnel expenses eg Bergendahl 1998 TortosaAusina 2002 LozanoVivas et al 2002 Drake et al 2006 12 Other earning assets normally include various items such as government securities investment securities trading securities other securities equity investments and other investments 13 Isik and Hassan 2003a mention among others that ignoring such nontraditional outputs may penalize banks that are heavily involved in these activities The reason is that while the resources used to produce these nontraditional outputs are part of the input vector the outputs generated using these inputs are not included in the output vector Electronic copy available at httpsssrncomabstract1350544 13 for all banks as they operate in the same markets for money and service The comparison with standard DEA models developed under either CRS or VRS assumptions shows that the DEA ratio model has the highest discrimination power 315 Adjusting for the environment Coelli et al 2005 discuss four approaches that can be used to incorporate environmental variables in DEA applications14 The first method by Banker and Morey 1986 requires the environmental variables to be ordered from the least to the most harmful ones for efficiency Then the efficiency of a given firm is compared with those firms in the sample that have a value of the environmental variable which is less than or equal to the given firm This ensures that banks are not compared with peers operating in a more favourable environment The second method by Charnes et al 1981 requires the investigator to i divide the sample into subsamples and solve DEA problems for each subsample ii project all observed data points into their prospective frontiers and iii solve a single DEA using the projected points and assess any difference in the mean efficiency of the two subsamples According to Coelli et al 2005 the following two problems are common in both methods i by splitting up the sample they reduce the comparison set and ii only one environmental variable can be considered in each case thereby limiting the scope of the analysis Under the third method the environmental variables are included directly into the DEA problem as nondiscretionary inputs if it is believed to have a positive effect on efficiency or outputs if they have a negative effect on efficiency The disadvantage of this approach is that one must know a priori the direction of the influence a shortcoming that is also applicable in the case of the first method Alternatively the environmental variables can be included as nondiscretionary neutral variables using an equality form The shortcoming of this approach is that it can reduce the reference set for each firm Recent applications of the above approaches in banking can be found in Pastor 1999 and LozanoVivas et al 2001 2002 14 According to Coelli et al 2005 environmental variables can include ownership differences location characteristics labour union power and government regulations and in a sense any factors that can influence the efficiency of the firm without being traditional inputs or under the control of managers Electronic copy available at httpsssrncomabstract1350544 14 The fourth method that is discussed in Coelli et al 2005 is the twostage approach This involves a DEA problem with traditional inputs and outputs in the first stage In the second stage the efficiency scores obtained are regressed on the environmental variables While this approach has been frequently used in the banking literature with numerous applications it fails to adjust the efficiency measures Therefore it is more suitable when the objective is to examine the correlations of efficiency with various factors see Section 321 rather than provide the basis for absolute comparisons across different environments Finally Pastor 2002 and Drake et al 2006 adjust the bank efficiency scores for risk and external environmental factors respectively using a multistage DEA see Fried et al 1999 2002 In this case the application starts with the estimation of a DEA model with traditional inputs and outputs Then using the slacks from the DEA model they quantify the effect of the operating environment and adjust the initial dataset inputs andor outputs Finally they rerun the initial DEA model using the adjusted data 32 Topics of interest 321 Determinants of efficiency Several studies attempt to investigate the factors that influence the efficiency of banks Some studies examine only bankspecific factors and others examine both bankspecific attributes and environmental determinants Commonly found bankspecific factors are size profitability capitalisation loans to assets Casu and Molyneux 2003 Casu and Girardone 2004 Ataullah and Le 2006 Ariff and Can 2008 Isik and Hassan 2003a examine additional characteristics such as the educational profile of bank personnel and CEOChairman affiliation Some studies examine whether age is related to efficiency eg Isik and Hassan 2003a Canhoto and Dermine 2003 Isik 2008 Furthermore Isik and Hassan 2002 Casu and Girardone 2004 and Pasiouras 2008b examine the international presence of Turkish Italian and Greek banks respectively Countryspecific factors include market concentration presence of foreign banks ratio of private investments to GDP fiscal deficits to GDP GDP growth Hauner 2005 Ataullah and Le 2006 More recently Pasiouras 2008a examined the relationship between technical efficiency and regulations related to capital Electronic copy available at httpsssrncomabstract1350544 15 adequacy private monitoring banks activities deposit insurance schemes supervisory power and bank entry into the industry These studies use the twostage approach discussed above First they use DEA to obtain efficiency estimates Then in a second stage the DEA scores are regressed on a number of explanatory variables using Tobit eg Hauner 2005 OLS Ataullah and Le 2006 GMM Ataullah and Le 2006 or GLS Isik and Hassan 2003a regression The rationale to use Tobit lies on the fact that the efficiency scores are bounded between 0 and 1 and hence noncensored estimates will be biased Ataullah and Le 2006 among others mention that it is not necessary to use Tobit as long as the efficiency scores can be transformed by taking the natural logarithm of efficiency score 1efficiency score Whether or not censored regression is used the most important argument comes from Simar and Wilson 2007 who pont out that the covariates in the second step regression are obviously correlated with the one side error terms from the first step as otherwise there would be no need for the second stepregression Furthermore the covariates in the secondstep are likely to be highly correlated with the covariates in the firststep This means that the errors and the covariates in the first step cannot be independent Thus Simar and Wilson conclude that the likelihood that is maximized is not the correct one unless one takes account of the correlation structure Casu and Molyneux 2003 present an early attempt to account for the problems that can emerge in the twostage DEA method using a bootstrap approach However Simar and Wilson 2007 mention that the simple bootstrap is not enough to overcome the drawbacks pointed out above They propose an algorithm that uses a double bootstrap procedure and they present an application to US banks Brissimis et al 2008 adopt this approach to examine the determinants of efficiency in the new EU banking sectors 322 Stock returns and efficiency Following the work of Ball and Brown 1968 the relationship between stock returns and publicly available information has attracted considerable attention in the accounting and finance literature Whilst most of the studies examine whether earnings reflect some of the information in stock prices recent research has however shifted towards the use of additional data such as accruals revenues economic value added and efficiency to understand how they affect stock prices and returns DEA Electronic copy available at httpsssrncomabstract1350544 16 applications that were published over the period of our survey and investigate the relationship between bank efficiency and stock returns examine Australia Kirkwood and Nahm 2006 Greece Pasiouras et al 2008 Malaysia Sufian and Majid 2006 Singapore Chu and Lim 1998 Sufian and Majid 2007 Turkey Erdem and Erdem 2008 and Spain Guzman et al 2008 Furthermore in a crosscountry setting Beccalli et al 2006 provide evidence from France Germany Italy Spain UK that is the five principal EU banking sectors In most cases the results of these studies indicate a positive relationship between stock returns and efficiency changes Furthermore Becalli et al 2006 reveal that the explanatory power of the model with DEA scores is higher than that of a model that uses the return on equity ROE as a measure of performance 323 Bank ownership A number of studies compare the efficiency of banks across different ownership types One approach used in the literature is to split the sample and compare the means of the different ownership groups Another approach is to incorporate dummy variables in a second stage analysis as the one described in Section 321 Some studies compare domestic and foreign banks Havrylchyk 2006 finds that greenfield banks are more efficient than domestic banks in Poland whereas foreign banks that acquired domestic ones have not successfully increased their efficiency Sturm and Williams 2004 report that foreign banks in Australia are more efficient than domestic ones Isik and Hassan 2003a find that foreign banks are more efficient than private domestic Turkish banks while Isik 2008 reports similar results for TFP growth estimates However Ataullah and Le 2004 find that prior to the financial liberalization of 19911992 foreign banks where less efficient than domestic ones in India and Pakistan nevertheless the opposite picture emerged after this period Other studies examine the efficiency of stateowned banks Many of them find that stateowned banks are less efficient than other banks For instance Garcia Cestona and Surroca 2008 find that Spanish banks controlled by insiders ie managers and workers are more efficient than the ones controlled by public administrations Ariff and Can 2008 find that jointstock banks in China are more cost and profitefficient than stateowned banks Chen 1998 in Taiwan and Mercan et al 2003 in Turkey also report that the efficiency of privatelyowned banks is Electronic copy available at httpsssrncomabstract1350544 17 higher than that of stateowned banks Similarly Isik 2008 focuses on TFP growth and finds that the growth of private banks is more than double that of state banks In contrast other studies report that the efficiency of private banks is lower than that of stateowned banks in Turkey Isik and Hassan 2003a India Sathye 2003 and Austria and Germany Hauner 2005 324 Corporate events and efficiency Another part of the literature relates DEA efficiency estimates to corporate events such as mergers and acquisitions andor bankruptcy Some of the studies compare the efficiency of acquirers and their targets and examine whether mergers improve efficiency Avkiran 1999 reports that acquiring banks are more efficient than acquired Australian banks However the results of this study also indicate that acquiring banks do not always maintain their premerger efficiency and there is mixed evidence on the extent to which the benefits of efficiency are passed to the public The results of AlSharkhas et al 2008 indicate that merged US banks are on average more technically efficient and that they experience higher productivity growth than nonmerged banks Hahn 2007a finds evidence that Austrian banks which engaged in domestic merger deals achieved a higher productive efficiency level than banks which did not participate in such deals The results also show that merger gains remain significant over a longer period of time more than five years but there is slight tendency to level off Sherman and Rupert 2006 also focus on the efficiency gains from bank mergers but they concentrate on the branch rather than the bank level They find that there are opportunities to reduce operating costs Such benefits however are not realised until four years after the merger Kohers et al 2000 follow a different approach to test whether Xefficiencies influence the markets assessment of bank mergers They find that that the abnormal returns of bidders around the announcement period were associated with both the target banks profit efficiency and cost efficiency as well as the difference between the average peer cost efficiency score and the corresponding targets cost15 Wheelock and Wilson 2000 also relate efficiency with bank acquisitions However in their 15 Only cost efficiency was estimated with DEA in this study Profit efficiency was estimated using stochastic frontier analysis As in previous studies Kohers et al 2000 also provide comparisons of efficiency across different groups They find that acquiring US bank holding companies BHCs are less costefficient compared to their targets Also industry peers operate with greater cost efficiencies that then bidder and target BHCs Electronic copy available at httpsssrncomabstract1350544 18 case DEA efficiency scores are used as an input in hazard models that capture the acquisition and failure likelihood They find that inefficiency increases the risk of failure while reducing the probability of a US banks being acquired Using a sample of 245 US banks Luo 2003 also reports that overall technical efficiency can be useful in predicting bank failures 325 Regulatory reformliberalization and efficiency A number of studies examine the impact of regulatory reform and liberalization initiatives on bank efficiency and productivity The studies that we review seem to indicate a positive relationship in most countries including India and Pakistan Ataullah et al 2004 Australia Avkiran 1999 Sturn and Williams 2004 US Mukherjee et al 2001 China Chen et al 2005 Greece Tsionas et al 2003 Rezitis 2006 Taiwan Wang and Huang 2007 Korea Gilbert and Wilson 1998 Turkey Isik and Hassan 2003b and the new EU countries Brissimis et al 2008 However Hauner 2005 finds no evidence that productivity is related to deregulation in Germany and Austria while Sathye 2002 argues that there can be a limit of deregulation after which no further productivity gains could be possible due to deregulation Some other interesting observations can be summarized as follows The banking sector reform in the newly acceded EU countries had a positive impact on bank efficiency while the effect of reform on TFP growth was significantly only toward the end of the reform process Brissimis et al 2008 The implementation of financial liberalization programmes may enable foreign banks to overcome the liability of foreignness and enhance their resource utilization Ataullah and Le 2004 The impact of deregulation will not necessarily be the same across different bank ownerships such as state domestic foreign Isik and Hassan 2003b 326 Comparison of frontier techniques A few studies compare alternative frontier techniques over the period of our survey Bauer et al 1998 use a sample of US banks to compare DEA SFA TFA thick frontier approach and DFA distributionfree approach over six consistency criteria They find that DEA yields much lower average efficiencies ranks the banks differently and identifies the best and worst banks differently from parametric methods Furthermore compared to DEA the parametric measures were more highly Electronic copy available at httpsssrncomabstract1350544 19 correlated with the traditional nonfrontier performance measures Huang and Wang 2002 provide another comparison of SFA DFA and DEA using data from Taiwanese banks They also argue that the choice of the frontier approach can result in different conclusions Weill 2004 provides more recent evidence using a sample from five European countries However this crosscountry setting does not alter the main conclusion of lack of robustness across different approaches confirming the results of the previous two studies Delis et al 2008 provide further support to these findings using a dataset of Greek banks The study of Beccalli et al 2006 also provides useful insights for the differences of SFA and DEA While this study does not focus on the comparison between these two methods it finds that while changes in the stock prices of banks reflect percentage changes in DEA cost efficiency scores this trend is less clear with SFA efficiency estimates In contrast Fiordelisi 2008 reports that SFA cost efficiency estimates explain better the variations in shareholder value creation than those derived from DEA Casu et al 2004 compare productivity growth estimates obtained through parametric and nonparametric approaches Their study seems to indicate that in this case the differences are not as large as in the efficiency studies They conclude that although the alternative methodologies produce in some cases opposing findings as for the sources of productivity for individual years in general they do not yield noticeably different results in terms of identifying the components of the productivity growth of EU banks during the period of their study 327 Efficiency of bank branches While most of the above studies focus on the efficiency of banking institutions as a whole a related strand of the literature examines the efficiency of bank branches16 In general branches are predominantly regarded as production units see Camanho and Dyson 1999 2005 2006 Zenios et al 1999 Soteriou and Zenios 1999b Golany and Storbeck 1999 Athanassopoulos and Giokas 2000 and Hartman et al 2001 where output is measured by the number of various accounts andor transactions as contrast to the intermediation approach employed by Portela and Thanassoulis 2005 2007 Portela et al 2003 Giokas 2008a where branches are seen as intermediaries and thus outputs are measured in monetary terms 16 See Appendix II for information on branch efficiency studies Electronic copy available at httpsssrncomabstract1350544 20 There are studies which consider the changing role of bank branches from a predominantly transactionbased one to a salesoriented role Cook et al 2000 and Cook and Hababou 2001 distinguish sales and services transactions functions of bank branches in a Canadian Bank where inputs are usually shared between these two functions To model the shared resources Cook et al 2000 extend the usual methodology to develop a model which incorporates the best resource split and optimises the aggregate efficiency score Further Cook and Hababou 2001 model the shared resources concept by adopting the Additive DEA model In a recent novel application Portela and Thannasoulis 2007 assess the branches of a Portuguese bank in terms of their performance in their new roles by specifying efficiency measures which consider sales transactional activities as well as bank branch profit efficiency Service quality is also a significant dimension of bank branches performance Two different ways are developed in empirical applications to consider service quality For instance Soteriou and Zenios 1999b Golany and Storbeck 1999 Athanassopoulos and Giokas 2000 include quality variables in efficiency analysis whereas Portela and Thannasoulis 2007 use a posthoc analysis to compare service quality index with efficiency measures They find that service quality is positively related with operational and profit efficiency The diversity of environments in which the branches are operating is also considered in the literature Paradi and Schaffnit 2004 considered the role of the environmental parameters that are outside the management control in the commercial branches of a large Canadian bank and incorporated risk and economic growth rate of the region as two nondiscretionary factors On the other hand Camanho and Dyson 2006 assess the impact of environmental factors and regional managerial policies on branches productivity through the construction of an index that reflects the relative performance of bank branches operating in four different regions in Portugal Das et al 2007 introduce the concept of spatial efficiency for each region relative to the nation and thus measure the effects of differences in the regional characteristics on the efficiency of bank branches across four metropolitan regions in India Some interesting observations can be summarised as follows In general the existing studies of bank branch efficiency use data on a small number of branches of nonUS banks and mostly employ DEA There are studies however which evaluate bank branch performance with DEA and compare the results with other methodologies ie FDH Portela et al 2003 Malmquist indices Camanho and Electronic copy available at httpsssrncomabstract1350544 21 Dyson 2006 loglinear frontier method Giokas 2008b ABC Plus Kantor and Maital 1999 OCRA Parkan and Wu 1999 Neural Network Model Wu et al 2006 Quasiconcave DEA FDH and Parametric Frontier models Dekker and Post 2001 Most of the studies which were reviewed use technical and operational efficiencies as estimated measures Recent studies examine the allocative profit efficiency see Portela and Thanassoulis 2005 2008 and cost efficiency Camanho and Dyson 2005 2008 aspects of bank branches 4 Other OR and AI techniques and bank performance In this section we review 15 recent studies that develop classification models used in the prediction of bank failure 9 bank underperformance 2 and credit ratings 4 Canbas et al 2005 point out that the study of bank failure is important for at least two reasons First an understanding of the reasons beyond the failure allows regulatory authorities to manage and supervise banks more efficiently Second the ability to differentiate between healthy and troubled banks can reduce the expected cost of bank failure either by taking actions to prevent failure or by minimizing the costs to the public Thomson 1991 As discussed in Gaganis et al 2006 while many of the failure prediction models achieve very promising classification accuracies one common drawback is that they concentrate on the assignment of banks in two groups failed and nonfailed However the classification of banks as bad or good reduces the usefulness of the model Obviously the classification into more groups relates to another strand of the literature that deals with credit risk modelling and attempts to replicate the ratings assigned by the credit agencies In the following subsections we provide a brief description of the methodologies the employed variables and the classification results17 A summary of the main characteristics of these studies is available in Appendix III 41 Methodologies Neural network NN modelling is an intelligence technique that follows a process similar to the human brain As in other classification models the parameters of a NN model need to be estimated before the network can be used for prediction purposes There are numerous NN architectures learning methods and parameters 17 To conserve space we keep our discussion on methodologies short Detailed information is available in the corresponding applications andor devoted textbooks Electronic copy available at httpsssrncomabstract1350544 22 Generally speaking NN architectures can be either feedback or feedforward In a feedback network nodes can receive inputs from nodes in any other layer in the network whereas in a feedforward network inputs are only received from previous layers The multilayer preceptor MLPNN competitive learning neural networks CLNN selforganizing map neural networks SOMNN backpropagation neural networks BPNN and probabilistic neural networks PNN are alternative approaches employed in the studies under review eg Chen and Shih 2006 Boyacioglu et al 2008 The approach of Support Vector Machines SVMs used in Boyacioglu et al 2008 Chen and Shih 2006 and Huang et al 2004 was introduced by Vapnik 1995 It is based on the Structural Risk Minimization SRM principle from computational learning theory This seeks to minimise an upper bound of the generalisation error rather than minimise the training error SVMs use a linear structure to implement nonlinear class boundaries through extremely nonlinear mapping of the input vectors into the highdimensional space If the data are linearly separated SVM uses a special kind of linear model the optimal separating hyperplane that provides the maximum separation between the classes The training points that are closest to the maximum margin hyperplane are called support vectors All other training examples are irrelevant for determining the binary class boundaries From the implementation point of view Vapnik 1995 showed how training a SVM and finding the parameters leads to a quadratic optimisation problem with bound constraints and one linear equality constraint This means that the solution of SVMs is unique optimal and absent from local minima Tay and Cao 2001 Nearest Neighbours is a nonparametric estimation method that has been applied in various problems in finance The nearest neighbour rule classifies an object ie bank to the class of its nearest neighbour in the measurement space using some kind of distance measure like the local metrics the global metrics the Mahalanobis or the Euclidean distance The modification of the nearest neighbour rule the knearest neighbour kNN method that is employed in Zhao et al 2008 classifies an object ie bank to the class ie failed or nonfailed more heavily represented among its k nearest neighbours In the case of decision trees that originate from machine learning instead of developing classification functions or network architectures a binary decision tree is developed This can be accomplished through a set of ifthen split conditions that lead Electronic copy available at httpsssrncomabstract1350544 23 to the accurate classification of cases eg banks Commonly used algorithms are CART eg Ravi et al 2008 and C45 eg Zhao et al 2008 The UTilités Additives DIScriminantes UTADIS multicriteria decision aid MCDA method used in Gaganis et al 2006 employs the framework of preference disaggregation analysis for the development of an additive utility function that is used to score the firms and decide upon their classification The estimation of the additive utility model is performed through mathematical programming techniques The MultiGroup Hierarchical Discrimination MHDIS method used in Pasiouras et al 2007 is another MCDA approach that uses utility functions for discrimination purposes MHDIS distinguishes the groups progressively starting by discriminating the first group from all the others and then proceeds to the discrimination between the alternatives belonging to the other groups To accomplish this task instead of developing a single additive utility function that describes all alternatives as in UTADIS two additive utility functions are developed in each one of the n1 steps where n is the number of groups At each stage of the hierarchical discrimination procedure two linear and a mixedinteger programming problems are solved to estimate the utility thresholds and the two additive utility functions in order to minimise the classification error As discussed in Kolari et al 2002 and Lanine and Vander Vennet 2006 trait recognition is another nonparametric approach under which individual traits are initially developed from different segments of the distribution of each variable and the interactions of these segments with one or more other variables segmented distributions When all possible traits of the variables are tabulated for all banks trait recognition uses a search routine to cull traits that do not discriminate between failed and nonfailed banks Trait recognition uses two sets of discriminators the safe traits associated with nonfailed banks and the unsafe traits failed banks known as features Then these features are used to vote on each bank and classify it as failed or nonfailed Lanine and Vander Vennet 2006 argue that the general rule of classifying safe and unsafe cells and assigning default probabilities works poorly when the number of banks falling into mixed cells is high To avoid this problem they suggest a formula to compute the probability of default directly without preliminary classification of cells as safe or unsafe Some researchers use soft computing approaches with the aim of synthesizing the human ability to tolerate and process uncertain imprecise and incomplete Electronic copy available at httpsssrncomabstract1350544 24 information during the decisionmaking Tung et al 2004 Soft computing usually integrates various techniques that may originate from different disciplines such as fuzzy logic18 neural networks machine learning etc in various combinations that will allow the researcher to exploit the strengths of individual techniques Ravi Kumar and Ravi 2008 A popular approach is the integration of neural network and fuzzy sets to develop a neural fuzzy network For example Tung et al 2004 propose a Generic Selforganizing Fuzzy Neural Network GenSoFNN based on the compositional rule of inference CRI to predict bank failure Multiple discriminant analysis MDA and logistic regression analysis LRA are two traditional techniques that are commonly used as benchmarks in classification studies eg Swicegood and Clark 2001 Pasiouras et al 2007 Kosmidou and Zopounidis 2008 In contrast Tung et al 2004 benchmark their models against Cox regression which estimates the dependence of the risk of failure on banks characteristics and of the evolution of that risk over time 42 Variables In most of the cases the variables are selected on the basis of the CAMEL model and include measures for Capital strength Asset quality Earnings and Liquidity19 The number of variables differs significantly among studies Some researchers start from a large list of variables and then use statistical screening eg KruskalWallis or dimension reduction eg factor analysis to end up with a reduced set of variables In addition to the financial variables some studies include additional characteristics related to ownership eg of shares held by the government market information eg stock price in previous year auditing and country eg Heritage index 43 Classification results Many of the studies in Appendix III report quite satisfactory classification accuracies and in several cases the proposed OR and AI methods outperform the traditional techniques ie MDA LRA Cox regression Furthermore the studies that focus on bankruptcy prediction in general perform more accurately than those that deal with 18 The theory of fuzzy sets proposed by Zadeh 1965 provides a mathematical approach to emulate human cognitive process The fuzzy set theory aims to classify subjective reasoning and assign degrees of possibilities in reaching conclusions Fuzzy logic can also be used to obtain fuzzy ifthen rules 19 M in CAMEL refers to management quality and it is rarely used in empirical studies due to difficulties in assessing management quality Electronic copy available at httpsssrncomabstract1350544 25 credit ratings However this is not surprising for two reasons First while classifying the banks in more groups is more informative it becomes more difficult to discriminate among groups in the intermediate categories Second credit ratings may incorporate additional qualitative information and human judgment that cannot be captured by the conventional variables used in most studies In any case comparisons across different studies should be treated with extreme caution One reason is that different studies make different assumptions about the prior probabilities of group membership as well as misclassification costs A second reason is that different studies use different approaches to validate the developed models For instance some researchers simply split the total sample in training and holdout datasets without accounting for the stability of models over time others use resampling techniques eg kfold crossvalidation and others use a holdout sample from a future period However it should kept in mind that classification ability is likely to overstate predictive ability Thus a superior approach would require that the model be validated against a future period as this approach more closely reflects a real world setting 5 Concluding remarks We have presented a comprehensive review of applications of operational research and artificial intelligence techniques in the assessment of bank performance by discussing a total of 179 studies published between 1998 and 2008 We have classified the studies in two main categories The first uses DEA and the DEAlike Malmquist index to estimate the efficiency and productivity growth of banks and bank branches We have discussed various methodological issues such as the estimated measures of efficiency the underlying assumptions of the estimated models and the selection of inputs and outputs We also discussed the main topics of interest including the relationship between ownership and efficiency stock returns and efficiency the determinants of efficiency the efficiency of bank branches amongst others Three of the main conclusions are that i profit efficiency and capacity efficiency have received quite limited attention in DEA studies in banking ii most studies that use a twostage DEA do not employ appropriate bootstrapping techniques and their results may be biased iii there is much diversity among studies with respect to the selection of input and Electronic copy available at httpsssrncomabstract1350544 26 outputs As in the case banks as a whole cost and profit efficiency has received considerably less attention in branch efficiency studies Furthermore an area of research deserving attention would be the estimation of bank branches efficiency over successive time periods Studies falling into the second category attempt to develop classification models to predict failure the credit ratings of banks and identify underperformers We have found both countryspecific and crosscountry studies and applications of techniques that originate from various disciplines Most of the studies rely heavily on financial information although in some cases nonfinancial variables are also used Given the differences in the approaches employed to validate the models comparisons of classification accuracies across studies should be treated with caution We find only a few studies that propose the combination of the predictions of individual models into integrated metaclassifiers and we believe that this is an area of research that is worthy of further attention References Akhtar MH 2002 Xefficiency analysis of commercial banks in Pakistan a preliminary investigation Pakistan Development Review 41 4 567580 Alam P Booth D Lee K Thordarson T 2000 The use of fuzzy clustering algorithm and selforganizing neural networks for identifying potentially failing banks an experimental study Expert Systems with Applications 18 2000 185 199 Allen R 1997 Incorporating value judgements in data envelopment analysis PhD Dissertation Warwick Business School University of Warwick UK AlSharkas AA Hassan MK Lawrence S 2008 The Impact of Mergers and Acquisitions on the Efficiency of the US Banking Industry Further Evidence Journal of Business Finance and Accounting 35 5070 Ariff M Can L 2008 Cost and profit efficiency of Chinese banks A non parametric analysis China Economic Review 19 260273 Asmild M Paradi JC Aggarwall V Schaffnit C 2004 Combining DEA Window Analyis with the Malmquist Index Approach in a Study of the Canadian Banking Industry Journal of Productivity Analysis 21 6789 Electronic copy available at httpsssrncomabstract1350544 27 Ataullah A Cockerill T Le H 2004 Financial liberalization and bank efficiency a comparative analysis of India and Pakistan Applied Economics 36 1915 1924 Ataullah A Le H 2004 Financial repression and liability of foreignness in developing countries Applied Economics Letters 11 545549 Ataullah A Le H 2006 Economic reforms and bank efficiency in developing countries the case of the Indian banking industry Applied Financial Economics 16 653663 Athanassopoulos AD Giokas D 2000 The use of data envelopment analysis in banking institutions evidence from the commercial bank of Greece Interfaces 30 2 8195 Avkiran NK 1999 The evidence on efficiency gains The role of mergers and the benefits to the public Journal of Banking and Finance 23 9911013 Aysan AF Ceyhan SP 2008 What determines the banking sector performance in globalized financial markets The case of Turkey Physica A 387 15931602 Bala K Cook WD 2003 Performance measurement with classification information an enhanced additive DEA model Omega 31 439450 Ball R Brown P 1968 An Empirical Evaluation of Accounting Income Numbers Journal of Accounting Research 6 159178 Banker RD Morey RC 1986 Efficiency Analysis for Exogenously Fixed Inputs and Outputs Operations Research 34 513521 Banker RD Morey RC 1986 Efficiency Analysis for Exogenously Fixed Inputs and outputs Operations Research 34 513521 Barr RS 2002 Evaluating the productive efficiency and performance of US commercial banks Managerial Finance 28 8 325 Bauer PW Berger AN Ferrier GD Humphrey DB 1998 Consistency Conditions for Regulatory Analysis of Financial Institutions A Comparison of Frontier Efficiency Methods Journal of Economics and Business 50 85114 Beccalli E Casu B Girardone C 2006 Efficiency and Stock Performance in European Banking Journal of Business Finance and Accounting 33 245 262 Berg S Forsund F Jansen E 1991 Malmquist indices of productivity growth during the deregulation of Norwegian banking 198089 Scandinavian Journal of Economics 94 211228 Electronic copy available at httpsssrncomabstract1350544 28 Bergendahl G 1998 DEA and benchmarks an application to Nordic banks Annals of Operations Research 82 233249 Bergendahl G Lindblom T 2008 Evaluating the performance of Swedish savings banks according to service efficiency European Journal of Operational Research 185 16631673 Berger AN 2007 International Comparisons of Banking Efficiency Financial Markets Institutions Instruments 16 119 144 Berger AN Humphrey DB 1997 Efficiency of financial institutions International survey and directions for future research European Journal of Operational Research 98175212 Berger AN Mester LJ 1997 Inside the black box What explains differences in the efficiencies of financial institutions Journal of Banking and Finance 21 895 947 Board J Sutcliffe C Ziemba WT 2003 Applying Operations Research Techniques to Financial Markets INTERFACES 33 1224 Boyacioglu MA Kara Y Baykan OmK 2008 Predicting bank financial failures using neural networks support vector machines and multivariate statistical methods A comparative analysis in the sample of savings deposit insurance fund SDIF transferred banks in Turkey Expert Systems with Applications doi101016jeswa200801003 Brissimis SN Delis MD Papanikolaou NI 2008 Exploring the nexus between banking sector reform and performance Evidence from newly acceded EU countries Journal of Banking and Finance 32 26742683 Camanho AS Dyson RG 1999 Efficiency size benchmarks and targets for bank branches An application of data envelopment analysis Journal of the Operational Research Society 50 9 903915 Camanho AS Dyson RG 2005 Cost efficiency measurement with price uncertainty a DEA application to bank branch assessments European Journal of Operational Research 161 432446 Camanho AS Dyson RG 2006 Data envelopment analysis and Malmquist indices for measuring performance Journal of Productivity Analysis 26 3549 Camanho AS Dyson RG 2008 A generalisation of the Farrell cost efficiency measure applicable to nonfully competitive settings Omega 36 147162 Electronic copy available at httpsssrncomabstract1350544 29 Canbas S Cabuk A Kilic SB 2005 Prediction of commercial bank failure via multivariate statistical analysis of financial structures the Turkish case European Journal of Operational Research 166 528546 Canhoto A Dermine J 2003 A note on banking efficiency in Portugal New vs Old banks Journal of Banking and Finance 27 20872098 Caprio G Klingebiel D 2003 Episodes of Systemic and Borderline Financial Crises World Bank January Available at httpgoworldbankorg5DYGICS7B0 Casu B Girardone C 2004 Financial conglomeration efficiency productivity and strategic drive Applied Financial Economics 14 687696 Casu B Girardone C 2005 An analysis of the relevance of offbalance sheet items in explaining productivity change in European banking Applied Financial Economics 15 10531061 Casu B Girardone C 2006 Bank competition concentration and efficiency in the single European Market The Manchester School 74 441468 Casu B Girardone C Molyneux P 2004 Productivity change in European banking A comparison of parametric and nonparametric approaches Journal of Banking and Finance 28 25212540 Casu B Molyneux P 2003 A comparative study of efficiency in European banking Applied Economics 35 18651876 Chang TC Chiu YH 2006 Affecting factors on riskadjusted efficiency in Taiwans banking industry Contemporary Economic Policy 24 4 634648 Charnes A Cooper WW Rhodes E 1978 Evaluating Program and Managerial Efficiency An Application of Data Envelopment Analysis to Program Follow Through Management Science 27 668697 Charnes A Cooper WW Rhodes E 1981 Evaluating Program and Managerial Efficiency An Application of Data Envelopment Analysis to Program Follow Through Management Science 27 668697 Chen TY 1998 A study of bank efficiency and ownership in Taiwan Applied Economics Letters 5 613616 Chen TY 2001 An estimation of Xinefficiency in Taiwans banks Applied Financial Economics 11 237242 Chen TY 2002a Measuring firm performance with DEA and prior information in Taiwans banks Applied Economics Letters 9 201204 Electronic copy available at httpsssrncomabstract1350544 30 Chen WH Shih JY 2006 A study of Taiwans issuer credit rating systems using support vector machines Expert Systems with Applications 30 427435 Chen X Skully M Brown K 2005 Banking efficiency in China Application of DEA to pre and postderegulation eras 19932000 Chen CW Liou TS Wu YS 2004 The performance measurement of local banks in the era of the financial liberalization on Taiwan Information and Management Sciences 15 4 89104 Chen TY 2002b A comparison of chanceconstrained DEA and stochastic frontier analysis bank efficiency in Taiwan Journal of the Operations Research Society 53 5 492500 Chen TY 2004 A study of cost efficiency and privatisation in Taiwans banks the impact of the Asian financial crisis Service Industries Journal 24 5 137 151 Chen TY Yeh TL 2000 A measurement of bank efficiency ownership and productivity changes in Taiwan Service Industries Journal 20 1 95109 Chiu YH Jan C Shen DB Wang PC 2008 Efficiency and capital adequacy in Taiwan banking BCC and superDEA estimation Service Industries Journal 28 4 479496 Chu SF Lim GH 1998 Share performance and profit efficiency of banks in an oligopolistic market evidence from Singapore Journal of Multinational Financial Management 8 155168 Coelli TJ Perelman SA 1996 Comparison of Parametric and Nonparametric Distance Functions With Application to European Railways CREPP Discussion Paper no 199611 University of Liege Liege Coelli TJ Prasada Rao DS ODonnell CJ Battese GE 2005 An Introduction to Efficiency and Productivity Analysis 2nd Ed Springer USA Cook WD Seiford LM 2009 Data envelopment analysis DEA Thirty years on European Journal of Operational Research 192 117 Cook WD Seiford LM Zhu J 2004 Models for performance benchmarking measuring the effect of ebusiness activities on banking performance Omega 32 313322 Cook WD Hababou M 2001 Sales performance measurement in bank branches Omega 29 299307 Electronic copy available at httpsssrncomabstract1350544 31 Cook WD Hababou M Liang L 2005 Financial liberalization and efficiency in Tunisian banking industry DEA test International Journal of Information Technology Decision Making 4 3 455475 Cook WD Hababou M Tuenter HJH 2000 Multicomponent efficiency measurement and shared inputs in Data Envelopment Analysis an application to sales and service performance in bank branches Journal of Productivity Analysis 14 209224 Damar HE 2006 The effects of shared ATM networks on the efficiency of Turkish banks Applied Economics 38 683697 Das A Ghosh S 2006 Financial deregulation and efficiency An empirical analyis of Indian banks during the post reform period Review of Financial Economics 15 193221 Das A Ray SC Nag A 2007 Labouruse efficiency in Indian banking a branchlevel analysis Omega doi101016jomega200705002 Debasish SS Mishra B 2007 Performance Evaluation of Indian BanksA Study using Data Envelopment Analysis European Journal of Economics Finance and Administrative Sciences9 8694 Dekker D Post T 2001 A quasiconcave DEA model with an application for bank branch performance evaluation European Journal of Operational Research 132 296311 Delis MD KoutsomanoliFillipaki K Staikouras CK Gerogiannaki K 2008 Evaluating cost and profit efficiency a comparison of parametric and nonparametric methodologies Applied Financial Economics doi 10108009603100801935370 Devaney M Weber WL 2000 Productivity growth market structure and technological change evidence from the rural banking sector Applied Financial Economics 10 587595 Dimitras AI Zanakis SH Zopounidis C 1996 A survey of business failures with an emphasis on prediction methods and industrial applications European Journal of Operational Research 90 487513 Dogan E Fausten DK 2003 Productivity and Technical Change in Malaysian Banking 19891998 AsiaPacific Financial Markets 10 205237 Drake L 2001 Efficiency and productivity change in UK banking Applied Financial Economics 11 557571 Electronic copy available at httpsssrncomabstract1350544 32 Drake L Hall MJB 2003 Efficiency in Japanese banking An empirical analysis Journal of Banking and Finance 27 891917 Drake L Hall MJB Simper R 2006 The impact of macroeconomic and regulatory factors on bank efficiency A nonparametric analysis of Hong Kongs banking system Journal of Banking and Finance 30 14431466 Erdem C Erdem MS 2007 Turkish banking efficiency and its relation to stock performance Applied Economics Letters 15 207211 Fare R Grosskopf S Weber WL 2004 The effect of riskbased capital requirements on profit efficiency in banking Applied Economics 36 1731 1743 Fiordelisi F 2008 Efficiency and shareholder return in banking International Journal of Banking Accounting and Finance 1 114 132 Fried HO Lovell CAK Schmidt SS Yaisawarng S 2002 Accounting for Environmental Effects and Statistical Noise in Data Envelopment Analysis Journal of Productivity Analysis 17 157174 Fried HO Schmidt SS Yaisawarng S 1999 Incorporating the operating environment into a nonparametric measure of technical efficiency Journal of Productivity Analysis 12 249267 Fukuyama H Weber WL 2005 Estimating output gains by means of Luenberger efficiency measures European Journal of Operational Research 164 535547 Fukuyama H Weber WL 2002 Estimating output allocative efficiency and productivity change Application to Japanese banks European Journal of Operational Research 137 177190 Fung MK 2006 Scale economies Xefficiency and convergence of productivy among bank holding companies Journal of Banking and Finance 30 2857 2874 Gaganis C Pasiouras F Zopounidis C 2006 A Multicriteria Decision Framework for Measuring Banks Soundness Around the World Journal of MultiCriteria Decision Analysis 14 103111 GarciaCestona M Surroca J 2008 Multiple goals and ownership structure Effects on the performance of Spanish savings banks European Journal of Operational Research 187 582599 Gilbert RA Wilson PW 1998 Effects of Deregulation on the Productivity of Korean Banks Journal of Economics and Business 50 133155 Electronic copy available at httpsssrncomabstract1350544 33 Giokas DI 2008a Cost efficiency impact of bank branch characteristics and location Managerial Finance 34 3 172185 Giokas DI 2008b Assessing the efficiency in operations of a large Greek bank branch network adopting different economic behaviors Economic Modelling 25 559574 Golany B Storbeck JE 1999 A Data Envelopment Analysis of the operational efficiency of bank branches Interfaces 29 3 1426 Griffiths B Beynon MJ 2005 Expositing stages of VPRS analysis in an expert system Application with bank credit ratings Expert Systems with Applications 29 879888 Guzman I Reverte C 2008 Productivity and efficiency change and shareholder value evidence from the Spanish banking sector Applied Economics 40 2033 2040 Habibullah M S Muhd Iqbal Makmur M I AzmanSaini WNW Radam A Boon OH 2005 Bank Efficiency and the Efficient Market Hypothesis The Case for Bank Stock Prices in KLSE Savings and Development 29 4 263290 Hahn FR 2007a Domestic mergers in the Austrian banking sector a performance analysis Applied Financial Economics 17 185196 Hahn FR 2007b Environmental determinants of banking efficiency in Austria Empirica 34 231245 Halkos GE Salamouris DS 2004 Efficiency measurement of the Greek commercial banks with the use of financial ratios a data envelopment analysis approach Management Accounting Research 15 201224 Hartman TE Storbeck JE Byrnes P 2001 Allocative efficiency in branch banking European Journal of Operational Research 134 232242 Hauner D 2005 Explaining efficiency differences among large German and Austrian banks Applied Economics 37 969980 Havrylchyk O 2006 Efficiency of the Polish banking industry Foreign versus domestic banks Journal of Banking and Finance 30 19751996 Howland M Rowse J 2006 Measuring bank branch efficiency using data envelopment analysis managerial and implementation issues INFOR 44 1 49 63 Electronic copy available at httpsssrncomabstract1350544 34 Hu WC Lai MC Huang HC 2008 Rating the relative efficiency of financial holding companies in an emerging economy A multiple DEA approach Expert Systems with Applications doi 10 1016jeswa200806080 Huang TH Wang MH 2002 Comparison of economic efficiency estimation methods parametric and nonparametric techniques The Manchester School 70 682709 Huang LH Hsiao HC Cheng MA Chang SJ 2008 Effects of financial reform on productivity change Industrial Management Data Systems 108 7 867886 Huanga Z Chena H Hsua CJ Chenb WH Wu S 2004 Credit rating analysis with support vector machines and neural networks a market comparative study Decision Support Systems 37 543 558 Isik I 2007 Bank ownership and productivity developments evidence from Turkey Studies in Economics and Finance 24 115139 Isik I 2008 Productivity technology and efficiency of de novo banks A counter evidence from Turkey Journal of Multinational Financial Management 18 427442 Isik I Akcaoglu E 2006 An Empirical Analysis of Productivity Developments in Traditional Banks The Initial PostLiberalization Experience Central Bank Review 1 135 Isik I Hassan MK 2002 Technical scale and allocative efficiencies of Turkish banking industry Journal of Banking and Finance 26 719766 Isik I Hassan MK 2003a Efficiency Ownership and Market Structure Corporate Control and Governance in the Turkish Banking Industry Journal of Business Finance and Accounting 30 1363 1421 Isik I Hassan MK 2003b Financial deregulation and total factor productivity change An empirical study of Turkish commercial banks Journal of Banking and Finance 27 14551485 Isik I Hassan MK 2003c Financial disruption and bank productivity The 1994 experience of Turkish banks The Quarterly Review of Economics and Finance 43 291320 Isik I Uysai D 2006 The Sources of Productivity Growth in Turkish Banks by Ownership The Initial Responses of Public Private and Foreign Banks to Liberalization Turkish Studies 7 365403 Electronic copy available at httpsssrncomabstract1350544 35 Kantor J Maital S 1999 Measuring efficiency by product group integrating DEA with activitybased accounting in a large Mideast bank Interfaces 29 3 2736 Kao C Liu ST 2004 Predicting bank performance with financial forecasts A case of Taiwan commercial banks Journal of Banking and Finance 28 2353 2368 Kirkwood J Nahm D 2006 Australian Banking Efficiency and Its Relation to Stock Returns The Economic Record 82 253267 Kohers T Huang Mh Kohers N 2000 Market perception of efficiency in bank holding company mergers the roles of the DEA and SFA models in capturing merger potential Review of Financial Economics 9 101120 Kolari J Glennon D Shin H Caputo M 2002 Predicting large US commercial bank failures Journal of Economics and Business 54 361387 Kosmidou K Zopounidis C 2008 Predicting US commercial bank failures via a multicriteria approach International Journal of Risk Assessment and Management 9 2643 Krishnasamy G Ridzwa AH Perumal V 2003 Malaysian Post Merger Banks Productivity Application of Malmquist Productivity Index Managerial Finance 30 63 74 Kumar S Gulati R 2008 Evaluation of technical efficiency and ranking of public sector banks in India International Journal of Productivity Performance Management 57 7 540568 Kuosmanen T Post T 2001 Measuring economic efficiency with incomplete price information With an application to European commercial banks European Journal of Operational Research 134 4358 Kyj L Isik I 2008 Bank xefficiency in Ukraine An analysis of service characteristics and ownership Journal of Economics and Business 60 369393 Lanine G Vander Vennet R 2006 Failure prediction in the Russian bank sector with logit and trait recognition models Expert Systems with Applications 30 463478 Laurenceson J Qin F 2008 Has minority foreign investment in Chinas banks improved their cost efficiency China and World Economy 16 3 5774 Leightner JE Knox Lovell CA 1998 The impact of financial liberalization on the performance of Thai banks Journal of Economics and Business 50 115 131 Electronic copy available at httpsssrncomabstract1350544 36 Levine R 2005 Finance and Growth Theory and Evidence Chapter 12 in Handbook of Economic Growth vol 1 Part A Springer pp 865934 Lim GH Randhawa DS 2005 Competition liberalization and efficiency Evidence from a twostage banking model on banks in Hong Kong and Singapore Managerial Finance 31 1 5277 Lin YH Hsu GJY Hsiao CK 2007 Measuring efficiency of domestic banks in Taiwan application of data envelopment analysis and Malmquist index Applied Economics Letters 14 821827 Liu ST 2008 Slacksbased efficiency measures for predicting bank performance Expert Systems with Applications doi 101016jeswa200801032 Liu ST Kao C 2008 Stochastic data envelopment analysis in measuring the efficiency of Taiwan commercial banks European Journal of Operational Research doi 101016jejor200802023 LozanoVivas A Pastor JT Hasan I 2001 European Bank Performance Beyond Country Borders What Really Matters European Finance Review 5 141165 LozanoVivas A Pastor JT Pastor JM 2002 An Efficiency Comparison of European Banking Systems Operating under Different Environmental Conditions Journal of Productivity Analysis 18 5977 Luo X 2003 Evaluating the profitability and marketability efficiency of large banks An application of data envelopment analysis Journal of Business Research 56 627635 Maghyereh A 2004 The effect of financial liberalization on the efficiency of financial institutions the case of Jordanian commercial banks Savings and Development 124 Maudos J Pastor JM 2003 Cost and profit efficiency in the Spanish banking sector 19851996 a nonparametric approach Applied Financial Economics 13 112 Maudos J Pastor JM Perez F 2002 Competition and efficiency in the Spanish banking sector the importante of specialization Applied Financial Economics 12 505516 Mercan M Reisman A Yolalan R Emel AB 2003 The effect of scale and mode of ownership on the financial performance of the Turkish banking sector results of a DEAbased analysis SocioEconomic Planning Sciences 37 185202 Electronic copy available at httpsssrncomabstract1350544 37 Mester LJ 1996 A Study of Bank Efficiency Taking into Account RiskPreferences Journal of Banking and Finance 20 102545 Mostafa M 2007a Benchmarking top Arab banks efficiency through efficient frontier analysis Industrial Management Data Systems 107 6 802823 Mostafa M 2007b Modeling the efficiency of GCC banks a data envelopment analysis approach International Journal of Productivity and Performance Management 56 7 623643 Mukherjee K Ray SC Miller SM 2001 Productivity growth in large US commercial banks The initial postderegulation experience Journal of Banking and Finance 25 913939 Mukherjee A Nath P Pal M 2003 Resource service quality and performance triad a framework for measuring efficiency of banking services Journal of Operations Research Society 54 723735 Neal P 2004 Xefficiency and productivity change in Australian banking Australian Economic Papers June 174191 Ng GS Quek C Jiang H 2008 FCMACEWS A bank failure early warning system based on a novel localized pattern learning and semantically associative fuzzy neural network Expert Systems with Applications 34 9891003 Noulas AG 2001 Deregulation and operating efficiency the case of the Greek banks Managerial Finance 27 8 3547 Noulas AG 1997 Productivity growth in the Hellenic banking industry state versus private banks Applied Financial Economics 7 223228 Noulas AG Glaveli N Kiriakopoulos I 2008 Investigating cost efficiency in the branch network of a Greek bank empirical study Managerial Finance 34 3 160171 Olgu O WeymanJones TG 2008 Parametric and nonparametric Malmquist productivity decomposition a case study of European commercial banks International Journal of Business Performance Management 10 4 344365 Paradi JC Schaffnit C 2004 Commercial branch performance evaluation and results communication in a Canadian banka DEA application European Journal of Operational Research 156 719735 Parkan C Wu ML 1999 Measurement of the performance of an investment bank using the operational competitiveness rating procedure Omega 27 201217 Electronic copy available at httpsssrncomabstract1350544 38 Pasiouras F Gaganis Ch Doumpos M 2007 A multicriteria discrimination approach for the credit rating of Asian banks Annals of Finance 3 351367 Pasiouras F 2008b Estimating the technical and scale efficiency of Greek commercial banks The impact of credit risk offbalance sheet activities and international operations Research in International Business and Finance 22 301318 Pasiouras F 2008a International evidence on the impact of regulations and supervision on banks technical efficiency an application of twostage data envelopment analysis Review of Quantitative Finance and Accounting 30 187 223 Pasiouras F Liadaki A Zopounidis C 2008 Bank efficeincy and share performance evidence from Greece Applied Financial Economics 18 1121 1130 Pastor JM 1999 Efficiency and risk management in Spanish banking a method to decompose risk Applied Financial Economics 9 371384 Pastor JM 2002 Credit risk and efficiency in the European banking system A threestage analysis Applied Financial Economics 12 895911 Pastor JM Serrano L 2006 The Effect of Specialisation on Banks Efficiency An International Comparison International Review of Applied Economics 20 125 149 Paul S Kourouche K 2008 Regulatory policy and the efficiency of the banking sector in Australia Australian Economic Review 41 3 260271 Porembski M Breitenstein K Alpar P 2005 Visualising efficiency and reference relations in data envelopment analysis with an application to the branches of a German bank Journal of Productivity Analysis 23 203221 Portela MCAS Borges PC Thanassoulis E 2003 Finding closest targets in nonoriented DEA models the case of convex and nonconvex technologies Journal of Productivity Analysis 19 251269 Portela MCAS Thanassoulis E 2005 Profitability of a sample of Portuguese bank branches and its decomposition into technical and allocative components European Journal of Operational Research 162 850866 Portela MCAS Thanassoulis E 2007 Comparative efficiency analysis of Portuguese bank branches European Journal of Operational Research 177 12751288 Electronic copy available at httpsssrncomabstract1350544 39 Post T Nonparametric efficiency estimation in stochastic environments Noiseto signal estimation finite sample performance and hypothesis testing Journal of Banking and Finance 20652080 Prior D 2003 Long and shortrun nonparametric cost frontier efficiency An application to Spanish savings banks Journal of Banking and Finance 27 655 671 Ramanathan R 2007 Performance of banks in countries of the Gulf Cooperation Council International Journal of Productivity and Performance Management 56 2 137154 Ravi Kumar P Ravi V 2007 Bankruptcy prediction in banks and firms via statistical and intelligent techniques A review European Journal of Operational Research 180 128 Ravi V Kurniawan H Thai PNK Ravi Kumar P 2008 Soft computing system for bank performance prediction Applied Soft Computing 8 305315 Ravi V Pramodh C 2008 Threshold accepting trained principal component neural network and feature subset selection Application to bankruptcy prediction in banks Applied Soft Computing doi101016jasoc200712003 Ray SC Mukherjee K 1998 A study of size efficiency in US banking identifying banks that are too large International Journal of systems Science 29 11 12811294 Resti A 1998 Regulation can foster mergers can mergers foster efficiency The Italian case Journal of Economics and Business 50 157169 Rezitis AN 2006 Productivity growth in the Greek banking industry A non parametric approach Journal of Applied Economics IX 119138 Rizvi SFA 2001 Postliberalisation efficiency and productivity of the banking sector in Pakistan Pakistan Development Review 40 4 605632 Saha A Ravisankar TS 2000 Rating of Indian commercial banks A DEA approach European Journal of Operational Research 124 187203 Sahoo BK Tone K 2008 Decomposing capacity utilization in data envelopment analysis An application to banks in India European Journal of Operational Research doi 101016jejor200802017 Sathye M 2003 Efficiency of banks in a developing economy The case of India European Journal of Operational Research 148 662671 Electronic copy available at httpsssrncomabstract1350544 40 Sathye M 2001 Xefficiency in Australian banking An empirical investigation Journal of Banking and Finance 25 613630 Sathye M 2002 Measuring Productivity Changes in Australian Banking An Application of Malmquist Indices Managerial Finance 28 48 59 Sealey C Lindley J 1977 Inputs outputs and a theory of production and cost at depository financial institutions Journal of Finance 32 12511266 Seiford LM Zhu J 1999 Profitability and Marketability of the Top 55 US Commercial Banks Management Science 45 12701288 Sherman HD Rupert TJ 2006 Do bank mergers have hidden or foregone value Realized and unrealized operating synergies in one bank merger European Journal of Operational Research 168 253268 Sherman HD Zhu J 2006 Benchmarking with qualityadjusted DEA QDEA to seek lowercost high quality service Evidence from a US bank application Annals of Operations Research 145 301309 Simar L Wilson PW 2007 Estimation and inference in twostage semi parametric models of production processes Journal of Econometrics 136 31 64 Smith KA Gupta JND 2000 Neural networks in business techniques and applications for the operations research Computers and Operations Research 27 10231044 Soteriou AC Zenios SA 1999a Using data envelopment analysis for costing bank products European Journal of Operational Research 114 23424 Soteriou A Zenios SA 1999b Operations quality and profitability in the provision of banking services Management Science 45 9 12211238 Sowlati T Paradi JC 2004 Establishing the practical frontier in data envelopment analysis Omega 32 261272 Stavárek D 2006 Banking efficiency in the context of European integration Eastern European Economics 44 4 531 Sturm JE Williams B 2004 Foreign bank entry deregulation and bank efficiency Lessons from the Australian experience Journal of Banking and Finance 28 17751799 Sueyoshi T Kirihara Y 1998 Efficiency measurement and strategic classification of Japanese banking institutions International Journal of Systems Science 29 11 12491263 Electronic copy available at httpsssrncomabstract1350544 41 Sufian F 2008 Determinants of bank efficiency during unstable macroeconomic environment Empirical evidence from Malaysia Research in International Business and Finance doi 101016jribaf200807002 Sufian F Majid MZ A 2006 Banks Efficiency and Stock Prices in Emerging Markets Evidence from Malaysia Journal of AsiaPacific Business 7 3553 Sufian F Majid MZ A 2007 Xefficiency and share prices in the Singaporean banking sector A DEA window analysis approach Investment Management and Financial Innovations 4 73 90 Sufian F 2007 Trends in the efficiency of Singapores commercial banking groups a nonstochastic frontier DEA window analysis approach International Journal of Productivity Performance Management 56 2 99136 Swicegood P Clark JA 2001 Offsite Monitoring Systems for Predicting Bank Underperformance A Comparison of Neural Networks Discriminant Analysis and Professional Human Judgment International Journal of Intelligent Sytems in Accounting Finance and Management 10 169186 Tay FEH Cao L 2001 Application of support vector machines in financial time series forecasting Omega The International Journal of Management Science 29 309317 Thanassoulis E 2001 Introduction to the Theory and Application of Data Envelopment Analysis A Foundation Text with Integrated Software Kluwer Academic Publishers USA Thomson JB 1991 Predicting bank failure in the 1980s Economic Review Federal Reserve Bank of Cleveland Q1 920 TortosaAusina E 2002a Bank Cost Efficiency and Output Specification Journal of Productivity Analyis 18 199222 TortosaAusina E 2002b Cost Efficiency and Product Mix Clusters across the Spanish Banking Industry Review of Industrial Organization 20 163181 TortosaAusina E 2002c Exploring efficiency differences over time in the Spanish banking industry European Journal of Operational Research 139 643664 TortosaAusina E 2003 Nontraditional activities and bank efficiency revisited a distributional analysis for Spanish financial institutions Journal of Economics and Business 55 371395 TortosaAusina E GrifellTatje E Armero C Conesa D 2008 Sensitivity analysis of efficiency and Malmquist productivity indices An application to Electronic copy available at httpsssrncomabstract1350544 42 Spanish savings banks European Journal of Operational Research 184 1062 1084 Tsionas EG Lolos SEG Christopoulos DK 2003 The performance of the Greek banking system in view of the EMU results from a nonparametric approach Economic Modelling 20 571592 Tunga WL Queka C Cheng P 2004 GenSoEWS a novel neuralfuzzy based early warning system for predicting bank failures Neural Networks 17 567587 Valverde SC Humphrey DB Paso RL 2007 Opening the black box Finding the source of cost inefficiency Journal of Productivity Analysis 27 209220 Vapnik VN 1995 The nature of statistical learning theory SpringerVerlag New York Wang MH Huang TH 2007 A study on the persistence of Farrells efficiency measure under a dynamic framework European Journal of Operational Research 180 13021316 Webb RM 2003 Levels of efficiency in UK retail banks a DEA window analysis International Journal of the Economics of Business 10 305322 Weill L 2004 Measuring Cost Efficiency in European Banking A Comparison of Frontier Techniques Journal of Productivity Analyis 21 133152 Wheelock DC Wilson PW 2000 Why do banks disappear The determinants of US bank failures and acquisitions The Review of Economics and Statistics 82 127138 Wu D Yang Z Liang L 2006 Using DEAneural network approach to evaluate branch efficiency of a large Canadian bank Expert Systems with Applications 31 108115 Yildirim C 2002 Evolution of banking efficiency within an unstable macroeconomic environment the case of Turkish commercial banks Applied Economics 34 22892301 Zadeh L A 1965 Fuzzy sets Information and Control 8 338353 Zen F Baldan C 2008 The strategic paths and performances of Italian mutual banks a nonparametric analysis International Journal of Banking Accounting and Finance 1 189 214 Zenios CV Zenios SA Agathocleous K Soteriou AC 1999 Benchmarks of the efficiency of bank branches Interfaces 29 3 3751 Electronic copy available at httpsssrncomabstract1350544 43 Zhao H Sinha AP Ge W 2008 Effects of feature construction on classification performance An empirical study in bank failure prediction Expert Systems with Applications doi101016jeswa200801053 Zhao T Casu B Ferrari A 2008 Deregulation and productivity growth a study of the Indian commercial banking industry International Journal of Business Performance Management 10 4 318343 Zhou p Ang BW Poh KL 2008 A survey of data envelopment analysis in energy and environmental studies European Journal of Operational Research 189 118 Electronic copy available at httpsssrncomabstract1350544 44 Table 1 Surveyed studies by publication outlet Bank Efficiency productivity Branch Efficiency productivity Bank Bankruptcy ratings underperformance Total European Journal of Operational Research 12 7 0 19 Journal of Banking and Finance 15 0 0 15 Applied Financial Economics 13 0 0 13 Applied Economics 9 0 0 9 Journal of Productivity Analysis 5 4 0 9 Expert Systems with Applications 2 1 6 9 Journal of Economics and Business 6 0 1 7 Managerial Finance 5 2 0 7 Omega 0 5 0 5 Applied Economics Letters 4 0 0 4 Interfaces 0 4 0 4 Other 65 5 8 78 Total 136 28 15 179 Electronic copy available at httpsssrncomabstract1350544 45 Appendix I Studies in bank efficiency and productivity change Authors Publication year Country Sample Period InputOutput Oriented Treatment of deposits Estimated measures 2nd Stage regression Akhtar 2002 Pakistan 40 banks 1998 Input Input OTE AE OE na AlSharkas et al 2008 US 440 bank mergers peer group of non merged banks 19862002 Input Input OTE PTE SE AE CE Productivity and APE using SFA na Ariff Can 2008 China 28 banks 230 obs 19952004 Input CE Output PE Input CE APE SPE Tobit Asmild et al 2004 Canada 5 banks 19812000 Input Output OTE Productivity na Ataullah et al 2004 India Pakistan na 19881998 Output Interest expense input OTE PTE SE na Ataullah Le 2006 India 566 Obs 19921998 Output Interest expense input PTE OLS GMM Ataullah Le 2004 India Pakistan na 19871998 Input Deposits Output Financial expenses input PTE na Avkiran 1999 Australia 1619 banks 19861995 Input Interest expense input in Model 1 Deposits input in Model 2 OTE na Aysan Ceyhan 2008 Turkey 466 obs 19902006 Input Input PTE Productivity FE regression Barr et al 2002 US na 19841998 Input Interest expense input OTE na Bauer et al 1998 US 683 banks 19771988 Input Demand deposits output Time saving deposits input CE na Beccalli et al 2006 France Germany Italy Spain UK 1129 banks per country 90 banks in total 19992000 Both Interest expenses part of single input ie total cost PTE and CE with SFA na Bergendahl 1998 Denmark Finland Norway Sweden 48 banks 19921993 Input Output OTE PTE na Electronic copy available at httpsssrncomabstract1350544 46 Bergendahl Lindblom 2008 Sweden 8588 banks 19972001 Input Output OTE na Brissimis et al 2008 10 new EU countries 364 banks 4368 obs 19942005 Input Input PTE Productivity double bootstrap two stage least squares truncated Canhoto Dermine 2003 Portugal 20 banks 19901995 Input Output OTE PTE SE Productivity na Casu Girardone 2004 Italy 3648 banks 168 obs 19961999 Input Ef Output Prod Input PTE AE CE Productivity Logistic Casu Molyneux 2003 France Germany Italy Spain UK 530 banks 19931997 Both Input OTE PTE BootstrapTobit Casu et al 2004 France Germany Italy Spain UK 2086 banks 19942000 Output Cost of deposits as input Productivity na Casu Girardone 2006 EU15 11000 obs 19972003 Input Interest expenses part of a single input ie total cost PTE na Casu Girardone 2005 France Germany Italy Spain UK 2086 banks 19942000 Output Cost of deposits input Productivity na Chang Chiu 2006 Taiwan 26 banks 19962000 Input Input OTE AE CE Tobit Chen 2001 Taiwan 41 banks 19881997 Both Input OTE na Chen 1998 Taiwan 34 banks 1996 Input Deposits input in 2 Models Interest expenses input in 6 models OTE PTE SE Type of regression not specified Chen 2002a Taiwan 41 banks 1998 Output na OTE na Chen 2002b Taiwan 39 banks 273 obs 19942000 Input Input OTE Regression Electronic copy available at httpsssrncomabstract1350544 47 Chen 2004 Taiwan 44 banks 19942000 Input Input CEOTEPTESEAE na Chen Yeh 2000 Taiwan 34 banks 19951996 Input Ef Output Prod Input OTEPTESE na Chen et al 2005 China 43 banks 19932000 Input Price of deposits input Deposits output PTE AE CE na Chen et al 2004 Taiwan 49 banks 20002002 Input Ef Output Prod Input OTE PTESE Productivity na Chiu et al 2008 Taiwan 46 banks 20002002 Input Ef Output Prod Input PTE productivity Chu Lim 1998 Singapore 6 banks 19921996 Both Interest expenses input OTE PTE SE na Cook et al 2005 Tunisia 10 banks 69 obs 19921997 Input Interm App Interest expenses Input Product App Deposits Input OTE Regression Damar 2006 Turkey 3538 banks 20002003 Input Interest expense input Deposits output OTE PTE SE Type of regression not specified Das Ghosh 2006 India 7498 banks 19922002 Input 3 models in total Interest expense input in 2 models Deposits output in 1 model OTE PTE SE Tobit Debasish Mishra 2007 India 93 banks 20002007 Output Input OTE na Delis et al 2008 Greece 1423 banks 244 obs 19932005 Input Input CE and PE using SFA na Devaney Weber 2000 US 3391 4146 banks 19901993 Output Input OTE PTE SE Productivity SUR Dogan Fauesten 2003 Malaysia 1618 banks 19891998 Output Deposits output Interest expenses input OTE Productivity FE linear model Drake 2001 UK 9 banks 19841995 Input Input in Models 1 and 1a Output in Models 2 and 2b OTE PTE SE Productivity na Drake et al 2006 Hong Kong 4766 banks 413 observations in total 19952001 Input na in model 1 output in model 2 PTE Tobit Drake Hall 2003 Japan 149 banks 1997 Input Input OTE PTE SE na Electronic copy available at httpsssrncomabstract1350544 48 Erdem Erdem 2008 Turkey 10 banks 19982004 Input Input OTE AE CE na Fare et al 2004 US 858938 banks 19901994 Both Transaction deposits Output Non transaction deposits input TE AE SPE na Fiordelisi 2008 France Germany Italy UK Varies from 68 UK 1997 to 2183 Germany 2002 19972002 Input Output PTE SE AE CE and PE using SFA na Fukuyama Weber 2005 Japan 110141 banks 1073 obs 19921999 Output TE Input AE Input Luenberger Farrell PTE AE na Fukuyama Weber 2002 Japan 138141 banks 19921996 Both Input TE Output AE Output TE Input AE Input part of funds from customers Productivity na Fung 2006 US 135 banks 1080 obs 19962003 Input Input PTE SE Productivity OLS GarciaCestona Surroca 2008 Spain 226 Obs 19982002 Output Output IP AE IPR na Gilbert Wilson 1998 Korea 1525 banks 19801994 Output Demand deposits output Time savings deposits input TE Productivity na Guzman Reverte 2008 Spain 14 banks 20002004 Both Input OTE PTE Productivity na Habibullah et al 2005 Malaysia 37 banks 19881993 Input Input OTESEPTE Congestion Efficiency Granger causality Hahn 2007a Austria Around 800 banks 19962002 Input na in Model 1 Input in Model 2 PTE Tobit Bootstrapped Hahn 2007b Austria More than 800 banks 19952002 Input Profit oriented na Int App Input PTE Tobit Halkos Salamouris 2004 Greece 1518 banks 19971999 Output na in basic model Interest expense input in benchmark model OTE PTE SE and superefficiency measures na Hauner 2005 Germany Austria 97 banks 485 obs 19951999 Input Input PTE SE AE CE Productivity Tobit Havrylchyk 2006 Poland 247 Obs 19972001 Input Input OTE PTE SE AE CE Tobit Hu et al 2008 Taiwan 14 banks 2007 Input na OTE PTE SE na Electronic copy available at httpsssrncomabstract1350544 49 Huang et al 2008 Taiwan 42 banks 20012004 Output Input Productivity Regression Huang Wang 2002 Taiwan 22 banks 19821997 Input Input CE na Isik 2008 Turkey 794 obs 19811996 Input Input OTE PTE SE Productivity na Isik 2007 Turkey 51 banks 439 obs 19811990 Output Input Productivity GLS Isik Hassan 2003c Turkey 54 banks 19921996 Input Ef Output Prod Input part of loanable funds OTE PTE SE Productivity na Isik Akcaoglu 2006 Turkey 28 banks 19811990 Output Input part of loanable funds Productivity na Isik Hassan 2002 Turkey 36 banks in 1998 50 banks in 1992 53 banks in 1996 1988 1992 1996 Input Input OTE PTE SE AE CE also CE PE from SFA GLS Isik Hassan 2003a Turkey 39 54 56 banks 1988 1992 1996 Input Input OTE PTE SE AE CE GLS Tobit Isik Hassan 2003b Turkey 3856 banks 458 obs 19811990 Input Ef Output Prod Input OTE PTE SE Productivity na Isik Uysal 2006 Turkey 439 obs 19811990 Output Input part of loanable funds Productivity na Kao Liu 2004 Taiwan 24 banks 2000 Input Input OTE na Kao Liu 2008 Taiwan 25 banks 19972001 Both min max problems Purchased funds input Demand deposits output OTE na Kirkwood Nahm 2006 Australia 10 banks 19952002 Input Input PTE SE AE CE APE Productivity na Kohers et al 2000 US 94 bidder 94 targets 8960 obs peers 19901995 Input Input Output CE na Krishnasamy et al 2003 Malaysia 10 banks 20002001 Output Output Productivity na Kumar Gulati 2008 India 27 banks 20042005 Input Loanable funds input OTE Super eff Regression Kuosmanen Post 2001 EU 453 banks 1997 Input Debt capital input TE AE CE na Kyj Isik 2008 Ukraine 150 banks 883 obs 19982003 Input Input part of loanable funds OTE PTE SE GLS Laurenceson Qin 2008 China 65 banks 20012006 Input Interest expenses input Deposits output CE Tobit Leightner Lovell 1998 Thailand 31 banks 19891994 Input na Productivity na Electronic copy available at httpsssrncomabstract1350544 50 Lim Randhawa 2005 Hong Kong Singapore 19 banks 19951999 Input Production output Intermediation input OTE PTESE na Lin et al 2007 Taiwan 37 banks 20022003 Output Interest expenses input OTE Productivity na Liu 2008 Taiwan 24 banks 2000 Input Input OTE na LozanoVivas et al 2002 10 European 612 banks 1993 Input Output PTE na LozanoVivas et al 2001 10 EU countries 612 banks 1993 Input Output PTE na Luo 2003 US 245 banks 2000 Input na OTE PTE SE na Maghyereh 2004 Jordan 14 banks 19842001 Input Input OTE PTE SE Productivity Tobit Maudos et al 2002 Spain 1666 Obs 19851996 Input Price of loanable funds Input Loanable funds output CE Tobit Maudos Pastor 2003 Spain 5077 savings 7598 national 19851996 Input CE Output PE Input CE APE SPE na Mercan et al 2003 Turkey 545 obs 19891999 Input na OTE na Mostafa 2007b GCCGulf Coopn Council 43 banks 2005 Output na PTEOTESE na Mostafa 2007a Arab banks 85 banks 2005 Output na PTEOTESE na Mukherjee et al 2003 India 27 banks 19972000 Output Output PTE na Mukherjee et al 2001 US 201 banks 19841990 Input Input OTE PTE GLS Neal 2004 Australia 1226 banks 19951999 Input Ef Output Prod Demand deposits output Term deposits certificates of deposits other deposits input PTE SE AE CE Productivity na Noulas 2001 Greece 19 banks 19931998 Input Interest expense input OTE na Olgu WeymanJones 2008 EU12 and EU10 164 banks 19972001 Output Input Productivity na Pasiouras 2008b Greece 1218 banks 78 obs 20002004 Input Deposits Input in Models 14 Interest expense input in Model 5 PTE SE Tobit Electronic copy available at httpsssrncomabstract1350544 51 Pasiouras 2008a 95 countries 715 bank 2003 Input Input OTE PTE SE Tobit Pasiouras et al 2008 Greece 10 banks 2002005 Input Interest expense input OTE PTE SE na Pastor 1999 Spain na 19851995 Input na OTE PTE SE na Pastor 2002 Spain Italy Frane Germany 2598 Obs 19881994 Input Output CE Tobit Logistic Pastor Serrano 2006 10 European 540 banks 3780 obs 19921998 Input Input CE COMPE ISPE na Paul Kourouche 2008 Australia 10 banks 90 obs 19972005 Input Interest expense input OTEPTESE na Prior 2003 Spain na 1986 1990 1995 Input Number of current savings accounts output SRCE LRCECAPE na Ramanathan 2007 GCC 55 banks 20002004 Input Input OTE PTESE Productivity na Ray Mukherjee 1998 US 201 banks 19841990 Input Transaction deposits Input PTESESZE Tobit Resti 1998 Italy 67 merger deals 19871995 Input Output PTE Extra efficiency na Rezitis 2006 Greece 6 banks 19821997 Output Input OTE PTE SE Productivity Tobit Rizvi 2001 Pakistan 37 banks 19931998 Both Output OTEPTESE Productivity na Saha Ravisankar 2000 India 25 banks 19921995 Input Interest expense input Deposits output OTE na Sahoo Tone 2008 India 78 banks 19972001 Both Input PCU TE RECU OCI na Sathye 2003 India 94 banks 19971998 Input Interest expense input in Model 1 Deposits input in Model 2 PTE na Sathye 2001 Australia 29 banks 1996 Input Time saving deposits input Demand deposits output OTE AE CE Type of regression not specified Sathye 2002 Australia 17 banks 19951999 Output Interest expenses input Productivity Type of regression not specified Seiford Zhu 1999 US 55 banks 1995 Output oriented Input na OTE PTE Electronic copy available at httpsssrncomabstract1350544 52 congestion Simar Wilson 2007 US 322 banks 6955 banks Q4 2002 na Input TE Double Bootstrap truncated regression Stavarek 2006 EU11 countries 122126 banks 20012003 Input Input PTE SE na Sturm Williams 2004 Australia 1026 banks 273 obs In total 19882001 Input Input OTE PTE SE Productivity na Sueyoshi Kirihara 1998 Japan 136 banks na Input na OTE na Sufian 2008 Malaysia 3336 banks 171observations 19951999 Input Model 1 input Model 2 output with interest expense as input Model 3 interest expense input OTE PTE SE Tobit Sufian 20007 Singapore 141 banks 19932003 Input Input OTEPTESE na Sufian Majid 2007 Singapore 6 banks 19932003 Input Input PTE na Sufian Majid 2006 Malaysia 9 banks 20022003 Both Interest expenses input OTE PTE SE na TortosaAusina 2002c Spain 116121 banks 19861995 Input Model 1 Input Model 2 savings time deposits output savings deposits other deposits and interbank deposits input CE na TortosaAusina 2002a Spain 104 banks 19851997 Input Model 1 Input Model 2 Input Output CE na TortosaAusina 2002b Spain 104 banks 19851997 Input Input CE na TortosaAusina 2003 Spain 120162 banks 19861997 Input Input CE na TortosaAusina et al 2008 Spain na 19921998 Output Both OTE Productivity na Tsionas 2003 Greece 1719 banks 19931998 Input Ef Output Prod Input OTE PTE SE AE CE Productivity na Valverde et al 2007 Spain 77 banks 1540 obs 19922001 Input Input OCE ICE na Electronic copy available at httpsssrncomabstract1350544 53 Wang Huang 2007 Taiwan 22 banks 19822001 Input Input OTE OTEQFI AE CE AR GMM Correlation Markov Webb 2003 UK 68 banks 19821995 Input Input OTE PTE SE na Weill 2004 France Germany Italy Spain Switzerland 688 banks 19921998 Input Interest expense input CE Correlation Wheelock Wilson 2000 US 13162555 banks 19841993 Both Input PTE and CE with SFA na Yildirim 2002 Turkey 3859 banks 594 obs 19881999 Input Input OTE PTE SE na Zen Baldan 2008 Italy 121 banks 20012005 Input Output PTE Productivity na Zhao et al 2008 India 65 banks 845 obs 19922004 Input Input part of loanable funds OTE PTESE Productivity na Notes Classified in 2008 on the basis of doi OTE Overall Technical Efficiency ie CRS PTE Pure Technical Efficiency ie VRS SE Scale Efficiency AE Allocative Efficiency CE Cost Efficiency APE Alternative Profit Efficiency SPE Standard Profit Efficiency Productivity Various Productivity Growth Measures SRCE Shortrun cost efficiency LRCE Longrun cost efficiency CAPE Capacity efficiency PCU Physical capacity utilization RECU Ray economic capacity utilization OCI Optimal capacity idleness IP Performance index of PTE result of private negotiations among stakeholders AE A proxy of Allocative Efficiency calculated without including input prices IPR Index of Efficiency with preferences revealed by the legislator IPR AE x IP COMPE Composition efficiency ISPE Intraspecialization efficiency OCE Operating cost efficiency ICE Interest cost efficiency OTEQFI OTE in the presence of quasifixed inputs SZE Size efficiency PTE OTE these studies report measures which they label as Pefficiency or cost efficiency simply on the basis of some inputsoutputs that they use but they do not include inputsoutput price inputs so in a sense they present measures of technical efficiency Electronic copy available at httpsssrncomabstract1350544 54 Appendix II Studies in branch efficiency and productivity growth Authors Inputs Outputs InputOutput Oriented Approach for inputoutput selection Estimated measures Sample size Country Portela Thannasoulis 2007 INPUT SET1 TRANSCT EFF 1 Number ETMs ATMs CATs 2 Rent 3 No clients not registered INPUT SET2OPRTNL EFF 1 Number of staff 2 Rent INPUT SET3PROFIT EFF 1 Number of staff 2 Supply costs OUTPUT SET1 TRANSCT EFF1 No new registrations for internet use 2 No transactions in CATs 3 No deposits in ETMs OUTPUT SET2 OPRTNL EFF 1number of clients 2 Value current accounts 3 Value other resources 4 Value titles deposited 5 Value credit by bank 6 Value credit by associates 7 Number transactions OUTPUT SET3PROFIT EFF1 Value current accounts 2 Value other resources 3 Value credit over bank 4 Value credit associates Output except profit efficiency nonoriented DEA Intermediation Transactional Operational Profit 57 Portugal Portela Thannasoulis 2005 1Number of staff 2Supply costs 1Value current accounts 2Value other resources 3Value credit by bank 4Value credit associates Nonoriented Intermediation Overall profit efficiency technical profit efficiency allocative profit efficiency scale and mix effects 57 Portugal Portela et al 2003 1staff costs 2other operating costs 1 Value of current accounts 2Value of credit 3 Interest revenues Nonoriented Intermediation PTE 24 Portugal Camanho Dyson 2008 1Number of branch and account managers 2Number of administrative and commercial staff 3Number of tellers 4Operational costs excluding staff costs And input prices 1 Total value of deposits 2 Total value of loans 3 Total value of off balance sheet business 4 Number of general service transactions Input DEA Production OTE Allocative efficiency and Farrell CE measures market efficiency and economic efficiency 39 Portugal Electronic copy available at httpsssrncomabstract1350544 55 Camanho Dyson 2006 1Number of branch employeesincluding branch and account managers administrative and commercial staff and tellers2 Operational costs excluding staff costsAnd input prices 1 Total value of deposits 2 Total value of loans 3 Other revenue 4 Number of general service transactions Input DEA Production Overall regional performance within region efficiency spread and frontier productivity na Portugal Camanho Dyson 2005 1 Number of branch and account managers 2 Number of administrative and commercial staff 3Number of tellers 4Operational costs excluding staff costs 1Number of general service transactions Input DEA Production OTE Farrell CE Optimistic CE and Pessimistic CE 144 Portugal Camanho Dyson 1999 1 Number of employees in the branch 2 Floor space of the branch 3 Operational costs costs of supplies and other services 4 Number of external ATMs 1Number of general service transactions performed by branch staff 2Number of transactions in external ATMs 3 Number of all types of accounts at the branch 4 Value of savings 5Value of loans Both Production PTE SE 168 Portugal Das et al 2007 1 of different categories of employees 1value of deposits 2value of credit 3noninterest income Input DEA Intermediation Labor use efficiency Regional or spatial efficiency 222 India Giokas 2008a 1Personnel costs 2Running costs 3Operating expenses 1Value of loans 2Value of deposits 3Noninterest income Input DEA Intermediation Operating efficiency 171 Greece Giokas 2008b INPUT SET1Production efficiency 1Personnel costs 2Running and other operating costs INPUT SET2 Transaction efficiency 1Personnel costs 2Running costs and other operating costs INPUT SET3 Intermediation efficiency 1Interest costs 2Non interest costs OUTPUT SET1 1Value of loan portfolio 2Value of deposits 3Non interest income OUTPUT SET 2 1Loan transactions 2Deposit transactions 3Remaining transactions OUTPUT SET 3 1Interest income 2Noninterest income Input DEA Production Intermediation Production transaction Intermediation specific efficiency 44 Greece Electronic copy available at httpsssrncomabstract1350544 56 Porembski et al 2005 1 of Employees 2 Office space 1Private demand deposits 2Business demand deposits 3Time deposits 4Saving deposits 5Credits 6Bearer securities 7Recourse guarantees 8Bonds 9Investment deposits10 Insurances contracts 11Contributions to a building society contracts Input DEA Production PTE SE 141 German Paradi Schaffnit 2004 INPUT SET 1 Model 1 Production model 1staff 2information technology 3premises 4other non interest expenses INPUT SET 2 Model 2 Strategic model 1staff 2information technology 3premises 4other non interest expenses 5nonaccrual loans Environmental factors Growth factor risk rating incorporated into Models OUTPUT SET1 1Deposits2Loans3Fee income4Maintenance activities OUTPUT SET2 1Deposits 2Loans 3Fee income4 Deposit spread 5 Loan spread Both Intermediation OTE PTE Effectiveness Cost effectiveness 90 Canada Sowlati Paradi 2004 1FTE sales 2FTE support 3FTE other 1Loans 2Mortgages 3RRSPs 4Letters of Credit Input DEA Production PTE 79 Canada Athanassopoulos Giokas 2000 INPUT SET 1 1labor hrs 2branch size 3 computer terminals 4operating expenditure INPUT SET 2 1labor costs 2operating expenses 3running costs of the building OUTPUT SET1 1st Stage1Group A transactions 2 Group B transactions 3Group C transactions 4Group D transactions 2nd stage 1Credit transactions 2 Deposit trans 3 Foreign receipts OUTPUT SET 2 1savings deposits2current deposits 3demand deposits4time deposits 5total loans 6noninterest income Input DEA Intermediation Production Intermediation market efficiency 47 Greece Cook Hababou Tuenter 2000 1FSE service staff2 FSA sales staff 3FSU support staff 4FOT other staff 1MDP counter level deposits 2 MTR transfers between accounts 3RSP retirement savings plan openings 4MOR mortgage accounts opened Production Aggregate Sales Service efficiencies 20 Canada Electronic copy available at httpsssrncomabstract1350544 57 Cook Hababou 2001 Service 1FSEtotal number of full time equivalent service staff Sales 1FSAtotal number of full time equivalent sales staff Shared 1FSU total number of full time equivalent support staff 2FSTtotal number of full time equivalent other staff Service 1TOTMENUtotal number of menu account transactions 2VISAnumber of Visa cash advances 3 CADnumber of commercial deposit transactions Sales 1RSPnumber of RSP account openings 2MORTnumber of mortgages transacted 3BPLnumber of variable rate consumer loans transacted Input DEA Production Sales service efficiency 20 Canada Golany Storbeck 1999 1Teller Labor 2NonTeller 3Labor Retail Sq Feet 4Marketing 5Employment rate 1Total Loans 2Total Deposits 3Depth 4Satisfaction Output DEA Production Operational efficiency 182 USA Hartman et al 2001 1No of staff 2No of computer terminals3square meters of premises 1Amount of deposits 2amount of loans 3Amount of house mortgages Nondiscretionary variables no of customers Input DEA Production OTE PTE AE 50 Sweden Kantor Maital 1999 INPUT SET 1 Customer services 1Labor costs MSincluding social benefits 2services 3area square meters for services only INPUT SET 2 Bank transactions 1Labor costs MSincluding social benefits2 transactions only 3area square meters OUTPUT SET 1 1number of demand deposit accounts 2weighted output indexcustomer service transactions 3queuereplacing actions OUTPUT SET 2 1Credit cards regular and gold 2weighted output index transactions3 commissions on import and export commercial accounts 4savings accounts activity DEA Production customer service efficiency transaction efficiency 250 na Parkan Wu 1999 Cost 1salaries 2 staff benefits 3 electronic data processing expenses 4 occupancy 5 temporary staff expenses6 printing and stationary Revenue 1 LC letter of credit advice 2 LC confirmation 3 LC issuance 4 guarantee 5 acceptance 6 negotiation and 7 net interest revenue Input DEA Intermediation DEA scaled inefficiency 24 Hong Kong Wu et al 2006 1Personnel 2Other general expenses 1Deposits 2 Revenues 3 Loans Input DEA Intermediation DEA efficiency DEA NN efficiency 142 Canada Electronic copy available at httpsssrncomabstract1350544 58 Dekker Post 2001 1Front office personnel 2Facilitating personnel Total revenue Input DEA Intermediation Revenue generating efficiency 314 Netherlands Soteriou Zenios 1999 INPUT SET 1 Operational efficiency 1st Set of Inputs 1Managerial personnel 2Clerical personnel 3Computer terminals 4Space 2ndset of inputs1Current accounts 2Savings accounts 3Foreign currency and commercial accounts 4Credit applications 5 Commissions INPUT SET2 Service Quality Efficiency Model 1st Set of Inputs 1Managerial personnel 2Clerical personnel 3Computer terminals 4Space 2ndset of inputs1Current accounts 2Savings accounts 3Foreign currency and commercial accounts 4Credit applications 5 Commissions INPUT SET3 Profitability efficiency 1st Set of Inputs 1Managerial personnel 2Clerical personnel 3Computer terminals 4Space 2ndset of inputs1Current accounts 2Savings accounts 3Foreign currency and commercial accounts 4Credit applications 5 Commissions OUTPUT SET 1 Total amount of work produced OUTPUT SET 2 The level of service quality achieved ie Operational Internal MarketingPerceived measures OUTPUT SET 3 Profit Both Production Operational efficiency Service Quality Efficiency Profitability Efficiency 144 Cyprus Zenios et al 1999 INPUT SET1 1Managerial personnel 2Clerical personnel 3Computer terminals 4Space INPUT SET 2 1Current accounts 2Savings accounts 3Foreign currency 4commercial accounts Credit applications Total amount of work produced Input DEA Production Operating efficiency 144 Cyprus Sherman Rupert 2006 Personnel FTEs 1 Platform 2Teller3Manager 4Operating Expenses Supplies Telephone Travel Postage 1 Teller Transactions Deposits Withdrawals Checks Cashed Bank Checks Loan Payments Bonds Sold Redeemed Coupons 2MarketingNew Accounts 3Night Deposits 4Safe Deposit Visits 5ATMs Serviced Input DEA Production OTE 217 branches from 4 merged banks USA Electronic copy available at httpsssrncomabstract1350544 59 6Loans Mortgage Consumer Referrals Applications Closings Howland Rowse 2006 1Nonsales FTE2sales FTE 3 Size in total square feet 4City employment rate 1Loan volume 2 Deposit volume 3 Average number of productscustomer 4 customer loyalty Both Production PTE 162 Canada Sherman Zhu 2006 1Platform FTSe 2 Teller FTEs 3Management FTEs 4 Postage supplies telephone travel expenses 1Deposits withdrawals checks cashed 2Bank checks 3Bond transactions 4Night deposits 5Safe deposits visits 6New accounts 7Mortgage and consumer loans 8 ATMs Quality measuremystery shopper scores Input DEA Production OTE 225 US Noulas et al 2008 Value of 1Personnel costs 2Other operating expenses Value of 1Deposits 2Financial products 3Loans 4 Other loans Input DEA Intermediation OTE 58 Greece Electronic copy available at httpsssrncomabstract1350544 60 Appendix III Studies in bank bankruptcy credit rating and underperformance Part A Authors Subject Country Training sample Period covered Techniques Used Alam et al 2000 BF US Training 3 failed 17 extreme performance 80 healthy Validation not specified although the authors mention that they repeat their analysis in a second sample of another 100 obs with qualitatively similar results 1992 FC CLNN SOMNN Boyacioglu et al 2008 BF Turkey Training 14 failed 29 nonfailed Validation 7 failed 15 nonfailed 19972003 MLPNN CLNN SOMNN LVQNN SVMs MDA CA LRA Chen Shih 2006 CR Taiwan Training twAA higher ratings 16 TwA 25 TwBBB 26 tw BB lower ratings 6 Validation twAA higher ratings 6 TwA 9 TwBBB 9 tw BB lower ratings 2 19982003 SVMs BPNN LR Gaganis et al 2006 CR 79 countries Training Fitch ratings A B 466 C 214 D E 214 Validation 10fold crossvalidation 2004 UTADIS MDA OLR Huang et al 2004 CR US Taiwan Training Taiwan AAA 8 AA 11 A 31 BBB 23 BB 1 US AA 20 A 181 BBB 56 BB 7 B 1 Validation 10fold cross validation leaveoneout crossvalidation Taiwan 19982002 US 1991 2000 SVMs BPNN LRA Kolari et al 2002 BF US Training 2 samples from 1989 Failed banks 18 Nonfailed banks 1012 1061 Validation Various samples from 1990 1991 1992 Failed banks 6 18 Nonfailed banks 1053 1178 19891992 TR LRA Kosmidou Zopounidis 2008 BF US Training 4 samples from 19931999 Failed banks 1989 Nonfailed banks 90455 Validation 20002003 Failed banks 23 Nonfailed banks 112 19932003 UTADIS MDA Lanine Vander Vennet 2006 BF Russia Training 4 samples from January 1997March or August 2000 Failed banks 5878 Nonfailed banks 290390 Validation 4 samples from April or September 2000 to November 2003 or January or June 2004 19972004 MTR TR LRA Ng et al 2008 BF US Training 20 of total sample Failed 358548 Nonfailed 25552585 Validation 80 of total sample Failed 358 548 Nonfailed 2555 2585 19802000 FCMACCRISS GenSoFNNCRISS Coxs hazard model Pasiouras et al 2007 CR 9 Asian countries Training Fitch ratings A AB 4 B BC 31 C CD 69 D DE 77 E 34 Validation 10fold crossvalidation 2004 MHDIS MDA LRA Electronic copy available at httpsssrncomabstract1350544 61 Ravi et al 2008 BU US Training 640 adequate performance 160 poor performance banks the 160 poor performance banks where then duplicated to produce a total of 640 poor performance cases Validation 200 banks 1993 MLFFBP PNN RBFN SVM CART FRBC PCAMLFFBP Ravi Pramodh 2008 BF Turkey Spain Training Turkey 40 Spain 66 Validation 10fold crossvalidation na PCNN PCATANN PCABPNN TANN BPNN Swicegood Clark 2001 BU US Smallcommunity banks Adequate performance 480 training 320 validation Underperformance 120 training 80 validation Large regional banks Adequate perf 356 training 237 validation Underperformance 89 training 59 training 1993 NN MDA Tung et al 2004 BF US Failed 548 Nonfailed 2555 Training 20 Validation80 19802000 GenSoFNNCRI Cox Zhao et al 2008 BF US Training 240 failed 240 nonfailed Validation 10fold crossvalidation 19911992 LRA C45 decision tree BPNN kNN Notes BF Bank Failure CR Credit ratings BU Bank underperformance FC Fuzzy clustering MLPNN Multilayer perception neural networks CL NNCompetitive learning NN SOMNN Selforganizing map NN LVQNN Learning vector quantization NN SVMs Support Vector Machines MDA Multivariate discriminant analysis CA kmeans Cluster Analysis LRA Logistic Regression Analysis TR Trait Recognition MTR Modified Trait Recognition BPNN Back propagation NN LR Linear regression GenSoFNNCRI Generic Selforganising Fuzzy Neural NetworksCRI Scheme kNN knearest neighbour UTADIS UTilités Additives DIScriminantes MHDIS MultiGroup Hierarchical Discrimination Electronic copy available at httpsssrncomabstract1350544 62 Appendix III Studies in bank bankruptcy credit rating and underperformance Part B Authors Variables considered Classification resuls Alam et al 2000 1 Net income to total assets 2 Net Loan losses to Adjusted assets 3 Net loan losses to total loans 4 nonperforming loans to total assets 5 net loan losses provisions to net income na Boyacioglu et al 2008 1 Shareholders equity total assets 2 shareholders equity total loans 3 shareholders equity net profit total assets offbalance sheet commitments 4 permanent assets total asses 5 total loanstotal assets 6 loans under followuptotal loans 7 specific provisiontotal loans 8 specific provisionloans under followup 9 personnel expenses average assets 10 net profitaverage assets 11 net profitaverage shareholders equity 12 income before taxesaverage assets 13 interest incometotal operating income 14 noninterest expensestotal operating income 15 liquid assetstotal assets 16 total loanstotal deposits 17 trading securitiestotal assets 18 FX assetsFX liabilities 19 net interest incomeaverage assets 20 net on balance sheet positiontotal shareholders equity Note reduced or normalized sets of variables were used in some models Correct accuracies MLPNN 9550 CLNN 6818 SOMNN 6363 LVQ NN 100 SVMs 9090 MDA 6818 CA 8181 LRA 8181 Chen Shih 2006 1 of shares held by the government 2 of shares held by major shareholders above 10 3 average stock price in previous year 4 total assets 5 liabilities 6 shareholders equity 7 net sales 8 operating income loss 9 liabilities to assets 10 deposits to net worth 11 fixed assets to net worth 12 current reverse ratio 13 total assets turnover times 14 deposit to loan 15 nonperforming loan ratio 16 interest payments to annual average balance of deposits ratio 17 interest income to annual average balance of credit extension ratio 18 average amount of business income per employee 19 average amount of profit per employee 20 return on total assets 21 return on shareholders equity 22 operating income to paidin capital 23 net profit to sales 24 EPS 25 cash flow adequacy ratio Note some models use subsets of these variables Correct accuracies SVMs 73088462 BPNN 506538 LR 2308 50 Gaganis et al 2006 1 EquityTotal assets 2 Loan loss provisions net interest revenue 3 Return on average assets 4 Cost to income ratio 5 Liquid assets Customer short term funding 6 Log total assets 7 Number of subsidiaries 8 Number of institutional shareholders 9 Whether the banks is listed on a stock exchange or not 10 Whether the country is developed or developing UTADIS 6891 MDA 6506 OLR 6288 Huang et al 2004 1 Total assets 2 Total liabilities 3 Longterm debts total invested capital 4 Debt ratio 5 Current ratio 6 EBITinterest 7 Operating profit margin 8 Shareholders equity longterm debt fixed assets 9 Quick ratio 10 Return on total assets 11 Return on equity 12 Operating incomereceived capitals 13 Net income before tax received capitals 14 Net profit margin 15 EPS 16 Gross profit margin 17 Nonoperating income sales 18 Net income before tax sales 19 Cash flow from operating activities current liabilities 20 cash flow from operating activities capital expenditures increase in inventory cash dividends in last 5 years 21 cash flow from operating activities cash dividends fixed assets other assets working capitals Note Some models were developed with subsets of these variables Correct accuracies SVMs Taiwan 7568 7973 US7887 8038 BP NN Taiwan 7432 7568 US 7568 8075 LRA Taiwan 7027 7568 US 7509 7698 Kolari et al 2002 1 Total assets millions 2 Net interest incometotal assets 3 Net income after taxes total assets 4 Total equitytotal assets 5 Allowance for loan lossestotal assets 6 Provisions for loan lossestotal assets 7 Net loan chargeoffs total assets Correct accuracies TR around 85 LRA slightly Electronic copy available at httpsssrncomabstract1350544 63 8 Total securities total assets 9 Nondeposit liabilities total liabilities 10 Certificates of deposit total deposits 11 Total loans leases total assets 12 Sum of key asset accounts total assets 1324 maximum change in each preceding ratio mean of corresponding ratio 25 Maximum change in loans past due at least 90 daysmean of numerator 26 Bank Holding Co total assets millions 27 Bank total assets BHC total assets 28 Maximum change in total assets mean change in total assets better than chance Kosmidou Zopounidis 2008 26 variables were initially considered After a combination of correlation analysis ANOVA KruskalWallis MannWhitney the following 9 variables were used 1 Return on current assets 2 Financing of current assets 3 Operating income current assets 4 Income before taxes exceptional outcome provisions for loan losses net depreciations 5 Asset per employee 6 Provisions for precarious loans overdue loans 7 loans internal deposits 8 Tier 1 ratio 9 Ratio of precarious capital Correct accuracies UTADIS 7575 8544 DA 75757935 Combined model 836 Lanine Vander Vennet 2006 1 Net incometotal assets 2 Liquid assetstotal assets 3 Government debt securitiestotal assets 4 Capital total assets 5 Overdue loans overdue promissory notes total loans 6 Total loanstotal assets 7 Log total assets Mean Square Error TR 0091 0167 MTR 0072 0149 LRA 0098 0138 AUCROC TR 0570 0745 MTR 0539 0906 LRA 0511 0752 Ng et al 2008 1 Average total equity capital average total assets 2 Average accumulated loan loss allowance average gross total loans and lease 3 Average accumulated loans 90 days late average gross total loans leases 4 Annual loan loss provision average gross total loans and leases 5 Noninterest expense operating income 6 total interest income interest expense average total asset 7 net after tax income applicable income taxes average total equity capital 8 average cash average federal funds sold average total deposit average fed funds purchased average banks liability on acceptance average other liabilities 9 change in Annual Gross total loans leases Note Additional models with 3 selected variables were also developed FCMACCRISS9 inputs 8207 year 3 to 9111 year 1 FCMACCRISS 3 inputs 8376 year 3 to 9528 year 1 GenSoFNNCRISS Cox vs FCMACCRISvarious scenarios as for misclassification costs of failed vs survived banks FCMAC better in minimizing Error I ie classiyfing a failed bank as survived FCMAC vs GenSoFNN no difference in year 1 error rate of GenSoFNN almost double in years 2 3 Pasiouras et al 2007 19 financial variables were initially considered Using factor analysis the following 5 financial variables were selected 1 Equity Customer short term funding 2 Net interest margin 3 Liquid assets Total deposits borrowings 4 Net loans Total deposits borrowings 5 Return on average equity In addition the following 5 nonfinancial variables were used 6 Number of institutional shareholders 7 Number of subsidiaries 8 Auditors opinion 9 Whether the bank is listed or not 10 Heritage Correct accuracies MHDIS 6603 MDA 5373 LRA 4755 Electronic copy available at httpsssrncomabstract1350544 64 bankingfinance index Ravi et al 2008 Ravi Pramodh 2008 Turkey 1 Interest expenses average profitable assets 2 Interest expenses average nonprofitable assets 3 Shareholders equity total incomedeposits nondeposit funds 4 Interest income interest expenses 5 Shareholders equity total income total assets 6 Shareholders equity total incomeTotal assets contingencies commitments 7 Networking capital total assets 8 Salary employees benefits reserve for retirement no of personnel 9 Liquid assets deposits nondeposit funds 10 Interest expense total expenses 11 Liquid assets total assets 12 Standard capital ratio Spain 1 Current assets total assets 2 Current assets cash total assets 3 Current assets loans 4 Reserves loans 5 Net income total assets 6 Net income total equity capital 7 Net income loans 8 Cost of sales sales 9 Cash flow loans Note in the case of Spain Turkey 6 7 variables are selected when feature subset selection algorithm is used Correct accuracies PCNN WOFSLTF Spain 966 Turkey 975 PCNN WFSLTF Spain 925 Turkey100 TANN Spain 916 Turkey 925 PCATANN Spain 975 Turkey 975 PCABPNN Spain 841 Turkey 85 BPNN Spain 8167 Turkey 875 PCNNWOFSSTF Spain 824 Turkey 925 PCNNWFSSTF Spain 866 Turkey 90 Swicegood Clark 2001 1 Earning assets Total assets 2 Total securities Total assets 3 Total loans Total asses 4 Nonperforming assets Total assets 5 Allowance for loan losses Total loans 6 Asset growth 7 Core deposits Total assets 8 Volatile liabilities Total liabilities 9 Total equity Total assets 9 Noninterest income Total assets 10 Interest income Total assets 11 Gains Losses from sale of securities Total assets 12 Salary Total assets 13 Noninterest expense SalaryTotal assets 14 Total interest expense Total assets 15 Provision expense Total loans 16 Off balance sheet commitments Total assets 17 Regional geographic location 18 Charter 19 Holding company affiliation 20 Federal Reserve Bank member 21 Branch or Unit bank 22 Deposit insurance Correct accuracies NN Large Regional Underperformance 661 Adequate performance 852 Small Community Underperformance 600 Adequate performance 828 MDA Large Regional Underperformance 508 Adequate performance 953 Small Community Underperformance 287 Adequate performance 921 Tung et al 2004 Average total equity capital average total assets 2 Average Accumulated loan loss allowance average gross total loans and leases 3 Average accumulated loans 90 plus days late average gross total loans and leases 4 Annual loan loss provisions Equal error rates GenSoFNNCRI 817 Electronic copy available at httpsssrncomabstract1350544 65 average gross total loans and leases 5 Noninterest expense operating income 6 Total interest income interest expense average total assets 7 Net income After tax applicable income taxes average total equity capital 8 Average cash average federal funds soldaverage total deposits average fed funds purchased average banks liability on acceptances average other liabilities 9 change in annual Gross total loans and leases 3610 Errors under different misclassification costs Cox Type I error 6 54 Cox Type II error 310 836 GenSoFNN CRI Type I error 0 069 GenSoFNNCRI Type II error 4545 100 Zhao et al 2008 93 raw accounting variables 26 fianancial ratios 1 Total equity total assets 2 Gross loans total assets 3 Commercial loans gross loans 4 Individuals loans gross loans 5 Real estate loans gross loans 6 Late loans 90 days gross loans 7 Not accruing loans gross loans 8 Loan loss provisions gross loans 9 Chargeoff loans gross loans 10 Loan allowance gross loans 11 Total loans netoff unearned total equity 12 Net income operating income 13 Net income total assets 14 Operating expenses total assets 15 Operating income total assets 16 Interest income interest expenses 17 Operating income operating expenses 18 Net income total assets 19 Net income total equity 20 Interest expenses for deposits Total deposits 21 Interest income from Loans Gross loans 22 Interest income operating income 23 Cash total assets 24 Cash Securities Total assets 25 Cash Federal funds sold US Treasury US Government Obligations total assets 26 Gross loans total deposits Misclassifications costs under different prior probabilities and cost ratios With raw accounting items LRA00870287 Tree 00760298 BPNN 0091 0417 kNN 0071 0297 With financial ratios LRA 0053 0199 Tree 0063 0257 BPNN 0057 0240 kNN 0057 0272 Electronic copy available at httpsssrncomabstract1350544 66 Electronic copy available at httpsssrncomabstract1350544 Production 263 551561 julset 2016 httpdxdoiorg10159001036513185714 1 Introduction The Brazilian banking sector has experienced vast changes over the last years This provoked a wave of banking mergers and acquisitions MAs and the penetration of some foreign banks into the Brazilian retail banking market Such processes focus on providing the institutions greater power in this competitive environment Coupled to this there are numerous innovative processes related to management which also allow banks to not only further improve customer service but also to offer a greater variety of services at lower costs The objective of this paper is to analyze by applying the bootstrap DEA the efficiency of banks in Brazil using a financial intermediation To reach the proposed goal we use the database entitled The Largest Banks from 2010 to 2013 which is periodically published by the Central Bank of Brazil The banking literature recognizes the vast progress made by banks with regard to its main functions and responsibilities Over time practically everywhere in the world the changes seen in the banks function are remarkable with new products and services increasingly available to customers However the financial intermediary role still prevails In fact all the other features incorporated into the functions of a bank are still the result of its primary function which is financial intermediation The structure of the paper is as follows Section two presents a brief literature review on bank efficiency Section three describes the data the employed methodology and the proposed DEAbased Section four discusses the results obtained from the application of the model and the final remarks are in section five Received Oct 23 2014 Accepted Nov 12 2015 Estimating the efficiency from Brazilian banks a bootstrapped Data Envelopment Analysis DEA Ana Elisa Péricoa Naja Brandão Santanab Daisy Aparecida do Nascimento Rebelattoc aDepartamento de Economia Faculdade de Ciências e Letras Universidade Estadual Paulista Araraquara SP Brasil bCentro de Ciências da Natureza Universidade Federal de São Carlos Buri SP Brasil cDepartamento de Engenharia de Produção Escola de Engenharia de São Carlos Universidade de São Paulo São Carlos SP Brasil anaelisafclarunespbr Abstract The Brazilian banking sector went through several changes in its structure over the past few years Such changes are related to fusions and acquisitions as well as the largest market opening to foreign banks The objective of this paper is to analyze by applying the bootstrap DEA the efficiency of banks in Brazil in 20102013 The methodology was applied to 30 largest banking organizations in a financial intermediation approach In that model the resources entering a bank in the form of deposits and total assets are classified as inputs and besides these manual labor is also considered as a resource capable of generating results For the output variable credit operations represent the most appropriate alternative considering the role of the bank as a financial intermediary In this work the matter of the best classification among retail banks and banks specialized in credit has little relevance The low relevance in this type of comparison is a result of analysis by segments segments were analyzed separately The results presented here point to an average level of efficiency for the large Brazilian banks in the period This scenario requires efforts to reduce expenses but also to increase revenues Keywords Bank efficiency Data Envelopment Analysis DEA Financial intermediation Bootstrap Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 552 Périco A E et al 2 Efficiency studies of the banking systems The literature offers two main approaches to evaluate the efficiency of banks financial intermediation and results The intermediation approach Sealey Lindley 1977 originates directly from the traditional role of financial institutions as financial intermediaries in which the banks main activity is related to raising funds from savings and redirecting it to borrowers deficit agents Thus in that model the resources entering a bank in the form of deposits and capital are classified as inputs and besides these manual labor is also considered as a resource capable of generating results For the output variable credit operations represent the most appropriate alternative considering the role of the bank as a financial intermediary For the resultoriented approach a set of expenditures is considered on the input side and a set of revenues on the output side Giokas 2008 This work regards the recent literature that investigates the characteristics of banks influences on credit as well as factors related to the performance of these institutions On the subject of banking efficiency Halkos Tzeremes 2013 analyzed the efficiency of fortyfive Greek banks that participated in merger or acquisition processes The results showed that in periods of crisis most of these banks were unable to generate operational efficiency gains yet in less turbulent periods the efficiency gains were observed Halkos Salamouris 2004 analyzed the Greek commercial banks with the use of financial ratios They find a wide variation in performance and show that the increase in efficiency was accompanied of a reduction in the number of small banks due to mergers and acquisitions Several authors have investigated banking efficiency taking into consideration different aspects Paradi et al 2011 for example argue that the management related aspects are crucial in determining the efficiency of a bank However by contrast much of the work conducted on bank efficiency Gaganis et al 2009 Giokas 2008 Portela Thanassoulis 2007 Pastor et al 2006 Stavarek 2005 among others does not even consider using management variables to measure efficiency Gaganis et al 2009 considering the variables used to measure efficiency and the results found point to risk nonpayment as the main element to determine the effectiveness of the Greek banks analyzed The results of Pastor et al 2006 for European banks are in line with the results of Gaganis et al 2009 Determining efficiency will depend greatly on the variables used and the efficiency approach chosen Some authors claim that the size of the bank is a determining factor for its efficiency as for instance in Macedo Barbosa 2009 which clearly defends the middlemarket segment These authors observed a relationship between bank size and performance since the results show that in this segment it is not possible for a small institution to have high performance Périco et al 2008 claim that the size of a bank was not a decisive factor to attribute efficiency given that in Brazil many medium and small banks had greater efficiency than the bigger banks The literature contains a wide range of models within different efficiency approaches always striving for greater performance by the decision makers Ultimately the driving force of this work is to contribute to this discussion 3 Data and methodology This section is divided into three parts The first part is a summary of main information concerning the data used in this research the composition of the investigated sample specification of variables used temporal delimitation as well as the data source The second and third parts represent brief descriptions and characterization of the techniques used in the manipulation of data Data Envelopment Analysis and the technique of Bootstrap resampling 31 Data In this study we used a sample of the 30 largest banks in accordance with the classification criteria used by the Central Bank for all periods It should be noted that only Commercial Bank institutions or Multiple Banks with commercial portfolios were selected We sought to work with institutions of all sizes large medium and small and service sectors retail and loan specialized specific niches to enable the analysis to also consider these aspects In our analysis we use the intermediation approach considering that financial institutions act as intermediates between depositors and borrowers Therefore we define our DEA formulation based on the adaptation of several other studies Sealey Lindley 1977 Berger Humphrey 1992 1997 Drake Hall 2003 Fukuyama Matousek 2011 by using total deposits labor and total assets as inputs and credit operations as output It is worth noting that the selection of these variables followed mainly two parameters The first one is related to the variables used in similar studies developed in other countries Sealey Lindley 1977 as Berger Humphrey 1992 1997 are important references of international banking literature the Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 553 Périco A E et al first studies on bank efficiency were carried out by them The other authors followed a similar pattern The second parameter which justifies the use of these variables is related to the recognition of the fundamental role of a bank financial intermediation The bank raises funds deposits to generate credit operations To capture the deposits are required employees and fixed investments total assets Table 1 presents details about the variables The data used were the banking institutions annual financial records from January 2010 to December 2013 obtained from the website of the Central Bank of Brazil in the report Top 50 Banks see Table 2 for the descriptive statistics of the variables used 32 Data Envelopment Analysis DEA is an operational research technique which is based on linear programming with the objective to comparatively analyze independent units in terms of their relative performance As it does not use a predefined production function identical to all organizations in the inputoutput relationship analysis it is classified as nonparametric Therefore the data envelopment analysis not require preparing a fixed weighted formula to measure the efficiency of the units under analysis because the weights of each variable are determined by the technique itself DEA can be regarded as a body of concepts and methodologies incorporated into a collection of models with different possible interpretations Charnes et al 1994 The envelopment surface will differ depending on the scale assumptions that underpin the model Two scale assumptions are generally employed constant returns to scale CRS and variable returns to scale VRS CRS reflects the fact that output will change in the same proportion as inputs are changed VRS reflects the fact that production technology may exhibit increasing constant and decreasing returns to scale The outputoriented model with variable returns to scale Banker et al 1984 is given below Equation 1 Table 1 The Description of variables Variable Description Type of variable Deposits Total funds raised by the bank R Input Total Assets The sum of current and longterm investment owned by a bank R Input Employees Number of bank employees Input Credit Operations Credit operations granted by the bank R Output Table 2 Descriptive statistics of the inputs and output Year 2010 2011 2012 2013 Deposits Mean 456628150333 521115219000 542254312000 558905980333 Median 61957005000 77399675000 80379365000 82346000000 Std 894408037244 1044387421259 1121331790300 1159606894802 Min 4328340000 8959830000 13446400000 5566110000 Max 3774464830000 4427709130000 4728728180000 4712436530000 Total Assets Mean 101324005667 114807047333 131259972000 131071639000 Median 20668655000 23710745000 27442270000 27501885000 Std 189417267802 211361958139 238948290176 237526293726 Min 2026390000 4021390000 4636810000 5739680000 Max 653224550000 725284140000 828252210000 864668980000 Employees Mean 195841000 204111667 204484667 202552667 Median 13935000 14485000 14385000 14475000 Std 382685442 397364027 402363999 398818453 Min 420000 390000 350000 150000 Max 1264260000 1312990000 1306380000 1253190000 Credit Operations Mean 444317015000 544711561333 639246160667 715029469333 Median 65432925000 72121090000 84388715000 97058130000 Std 837404182047 1032839825529 1247355762180 1427959213199 Min 1453180000 1871370000 4672510000 6393090000 Max 3341930460000 3975211610000 4905323020000 5479481140000 Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 554 Périco A E et al n DEA i ki k i 1 1 1 1 Min v x v 1 0 0 ˆ m r rk r n m i jr r jr k i r r i Subjetctto u y v x u y v u v θ 1 where y output x inputs u v weights r 1 m i 1 n j1n The variables uk and vk are introduced representing the variable returns to scale These variables should not meet the constraint of positivity and may be negative values Among the DEA models it is possible and recommended to choose the most suitable one for the sample through the testing hypotheses of returns to scale presented in Banker 1996 which verifies that scale return hypothesis constant or variable is most plausible for the data set used Banker 1996 suggests applying the nonparametric test of two KolmogorovSmirnov samples based on the maximum distance of the cumulative distributions of the efficiency scores of the DEACRS and DEAVRS models The test evaluates the null hypothesis of constant returns to scale against the alternative hypothesis of variable returns to scale This test is based on the maximum vertical distance between ˆ c ln ˆ c j F θ and ˆ v ln ˆ v j F θ the empirical distributions of ln ˆc θ j and ln ˆv θ j are used The statistic takes values between 0 and 1 Values near 1 tend to reject the null hypothesis and accept the alternative hypothesis Banker Natarajan 2004 33 Statistical Inference by bootstrap Considering that DEA is a deterministic approach a different result from the full efficiency can be interpreted as inefficiency Among other factors this inefficiency or pseudo efficiency may be due to data collection errors or factors attributed to chance compromising estimates made about the scores Dong Featherstone 2004 Aiming to correct this weakness several studies Efron 1987 Xue Harker 1999 Löthgren Tambour 1999 have suggested the use of the bootstrap for more consistent results Bootstrap procedures produce confidence limits on the efficiencies of units to capture the true efficient frontier within the specified interval Dyson Shale 2010 To correct the efficiency values in view of the inherent random data error the approach proposed by Simar Wilson 1998 2000 was used Through this proposal the Bootstrap technique is applied to the DEA methodology to proceed with the statistical inference of the efficiency results achieved by the DEA model Appendix A Thus for each DMU the confidence interval of efficiency the bias and the corrected efficiency can be estimated which will be considered for the performance evaluation of the banks The bootstrap bias estimate for the original DEA estimator 0 ˆDEA x 0 y θ can be calculated as Equation 2 B 1 B DEA 0 0 DEAb 0 0 DEA 0 0 b 1 BIAS x y B x y ˆ ˆ y ˆ x θ θ θ 2 Furthermore 0 0 ˆ DEA b x y θ are the bootstrap value and B is the number of bootstrap replications In this way a biased corrected estimator of 0 x 0 y θ can be calculated as Equation 3 0 0 0 0 0 0 1 0 0 0 0 1 ˆ ˆ ˆ ˆ ˆ 2 DEA B DEA DEA B DEA DEA b b x y x y BIAS x y x y B x y θ θ θ θ θ 3 According to Simar Wilson 2000 this bias correction may create additional noise the sample variance of bootstrap values 0 ˆDEA b x 0 y θ need to be calculated The calculation of the variance of the bootstrap values is illustrated below Equation 4 2 2 1 1 0 0 0 0 1 1 ˆ ˆ ˆ B B DEA b DEA b b b B x y B x y σ θ θ 4 The bias correction illustrated in Equation 3 needs to be avoided unless Equation 5 0 0 ˆ ˆ 1 3 B DEA BIAS x y θ σ 5 4 Financial performance of banks 41 Data envelopment analysis Table 3 summarizes the distribution of efficiency scores of Brazilian banks considering the combined period 20102013 for the two returns to scale models constant returns to scale and variable returns to scale With regards to the definition of the DEA model constant or variable scale literature suggests performing the KolmogorovSmirnov test KS since the choice of technology is a key issue and if decided arbitrarily it can produce biased results In the test procedure the value of this statistic 07197 was obtained α1 allowed accepting the assumption of variable returns to scale Accordingly Table 4 shows Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 555 Périco A E et al the efficiencies of the banks investigated considering the variable returns to scale for each year as well as for the combined period In order to fix efficiency values considering the random error inherent in the data we used the classification proposed by Simar Wilson 1998 2000 Table 5 shows the original efficiency indicators set on the bias the bias and indicators corrected by the bootstrap technique 2000 pseudosamples generated the bootstrap results From the results shown in Table 5 the significant influence of the efficiency rates on the changes in the samples was found reducing the average efficiency by 1443 Corrected average efficiency values in most cases are considerably different in magnitude when compared with the original values Note for example the case of one of the largest Brazilian banks Banco do Brasil which after the bootstrap technique was applied had its index corrected from 998 to 5731 Table 3 Distribution of banks per class Performance Class Banks Banks 0020 3 10 2 667 2040 4 1333 2 667 4060 7 2333 4 1333 6080 9 30 5 1667 8099 2 667 4 1333 100 5 1667 13 4333 Total 30 100 30 100 DEA CRS DEA VRS Minimum 668 729 Average 6197 7572 Maximum 100 100 Table 4 Efficiency indicators for banks Bank Total 2010 2011 2012 2013 Average ABCBrasil 9830 100 9860 7987 7117 8741 Alfa 100 100 9365 9047 100 9603 Bancoob 100 4776 100 100 100 8694 Banestes 4250 4147 42 3730 4108 4046 Banrisul 5249 5624 5493 5208 4510 5208 Bansicredi 100 7862 100 100 100 9465 Basa 2740 3268 3398 3236 2305 3051 Banco do Brasil 100 992 100 100 100 998 BIC 6967 7436 6566 7012 6069 677 BMG 8718 7762 6314 9556 7942 7893 BNB 6016 6784 6580 5502 4845 5927 BNP Paribas 4759 6165 5754 3819 2636 4593 Bradesco 9607 9742 9627 9542 9421 9583 BRB 6894 4675 6192 7107 7567 6385 BTG Pactual 1777 1696 1403 1572 2246 1729 CEF 100 100 994 100 100 9985 Citibank 2799 3226 3169 2689 2297 2845 Credit Suisse 100 100 100 100 5446 8861 Daycoval 6228 6583 6642 5452 5094 5942 Deutsche 100 1230 100 4017 1867 4278 Fibra 100 8703 9308 9077 100 9272 HSBC 5148 5625 5451 5246 4969 5322 Itaú 100 100 100 100 9629 9907 JP Morgan Chase 729 100 567 662 313 2885 Mercantil do Brasil 8521 6541 7238 8444 9785 8002 Panamericano 6929 100 6157 5347 5541 6761 Safra 100 9507 100 100 7977 9371 Santander 100 100 100 100 9849 9962 Societe Generale 100 100 100 993 100 9982 Votorantim 100 100 991 100 100 9977 Average 7572 7375 7438 7139 6718 7168 Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 556 Périco A E et al This enables verify that by conducting a purely deterministic analysis disregarding the statistical bias influence the results found may be overestimated 42 Analysis of results Table 6 summarizes the descriptive efficiency statistics of 30 banks In the analysis of the disaggregated periods a decrease in average efficiency was observed of approximately 55 per cent 20102013 However it was concluded that analyzing the efficiency of all banks together could be a mistake because of the different types of operations performed by them Accordingly the banks were grouped into two segments retail banking and banks specialized in credit This grouping followed the criteria of the Central Bank To observe the efficiency differences by service segment the banks average efficiency indicators were calculated for two service subdivisions in 2010 2011 2012 and 2013 It should be noted that of the 30 banks analyzed 14 are retail banks large public retail banks large foreign retail banks large private domestic retail banks and regional public retail banks and the rest are banks specialized in credit focused on providing few credit modalities Table 7 shows the results Table 7 shows that the average efficiency of the sample decreased by 55 Among the banks classified as retailers an average decrease of approximately 33 was observed between 2010 and 2013 As for banks specialized in loans there was a decrease in average efficiency From 2010 to 2013 the decrease was of 7 Tables 8 and 9 show performance indicators for two bank segments It can be observed that almost all banks specialized in loans between 2011 and 2013 had their average efficiency indicators reduced In early 2012 the Federal Government concluded that Brazilian bank spreads were at much higher levels than those practiced internationally After this in April 2012 two major Brazilian public banks Banco do Brasil and Caixa Econômica Federal initiated a retraction movement in interest rates charged to loan operations In this sense although at much lower Table 5 Efficiency indicators for banks after applying Bootstrap Bank Average efficiency Bias Corrected average efficiency ABCBrasil 8741 219 8522 Alfa 9603 1360 8287 Bancoob 8694 2472 6222 Banestes 4046 240 3806 Banrisul 5208 287 4921 Bansicredi 9465 1950 7515 Basa 3051 325 2726 Banco do Brasil 998 4249 5731 BIC 677 234 6536 BMG 7893 358 7535 BNB 5927 1521 4406 BNP Paribas 4593 172 4421 Bradesco 9583 4043 5540 BRB 6385 937 5448 BTG Pactual 1729 123 1606 CEF 9985 4322 5663 Citibank 2845 295 2550 Credit Suisse 8861 2082 6779 Daycoval 5942 084 5858 Deutsche 4278 265 4013 Fibra 9272 1041 8231 HSBC 5322 908 4414 Itaú 9907 4283 5624 JP Morgan Chase 2885 873 2012 Mercantil do Brasil 8002 1541 6461 Panamericano 6761 075 6686 Safra 9371 1803 7568 Santander 9962 4364 5598 Societe Generale 9982 1102 8880 Votorantim 9977 1798 8179 Average 7167 1443 5725 Table 6 Descriptive statistics of efficiency indicators of 30 banks Descriptive Statistics 2010 2011 2012 2013 Average 5770 5988 5688 5453 Mean 5669 5833 5671 5677 StandardDeviation 2039 2172 2121 2272 Maximum 9287 9221 9003 8973 Minimum 619 723 774 583 Table 7 Efficiency average by segment 2010 2011 2012 2013 Total sample 5770 5988 5688 5453 Retail banks 5070 5113 5044 4902 Banks specialized in credit 6381 6753 6250 5935 Table 8 Efficiency indicators of retail banks Banks 2010 2011 2012 2013 Banco do Brasil 5687 5730 5698 5809 Bradesco 5689 5535 5395 5540 CEF 5619 5662 5630 5741 Itaú 5633 5676 5644 5544 Santander 5576 5619 5587 5612 Banrisul 5602 5495 4788 3798 BNB 4859 4774 4152 3839 Citibank 3007 2575 2603 2016 HSBC 4559 4486 4354 4255 Safra 7592 8022 7846 6812 Banestes 4313 3753 3450 3706 Basa 2835 2921 2820 2327 Mercantil do Brasil 5651 5936 6739 7517 BRB 4367 5397 5911 6117 Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 557 Périco A E et al intensity private banks retail and specialized in credit followed suit in this retraction movement of rates However it should be noted that Brazil has a large bank concentration in which six banks hold over 80 of the market share Departamento Intersindical de Estatística e Estudos Socioeconômicos 2013 of which two are the major Brazilian public banks Between April 2012 and January 2013 the operations for natural persons underwent significant changes in their rates Two major Brazilian public banks reduced the overdraft rate by an average of 43 while for the personal credit line the average reduction by private banks was of 375 the mean reduction in two public banks was of 26 and in the private banks the average reduction was of 145 and for the auto loan financing the average reduction was of 315 for public banks and 22 for the banks investigated Departamento Intersindical de Estatística e Estudos Socioeconômicos 2013 With regards to the credit lines for companies the mean rate reductions also differed between public and private banks For promissory note discounts the public banks offered average reductions of 345 while the average reduction by private banks was of 135 For working capital finance the average reduction of public banks was of 40 and in private banks it was of 195 Departamento Intersindical de Estatística e Estudos Socioeconômicos 2013 By monitoring the retractions in interest rates there was as expected an increase in credit operations throughout the Brazilian banking system According to the data from the Departamento Intersindical de Estatística e Estudos Socioeconômicos 2013 it can be observed that the operations of the public financial system grew by 282 in 12 months Dec2011 to Dec2012 In the same period the operations of the domestic private financial system and foreign institutions grew only 67 and 85 respectively It can then be understood why so many of the loan specialized banks underwent the greatest reduction in average efficiency during the period investigated They increased their resources the number of employees deposits and capital and the credit operation growth fell short of what was feasible This result is due to the decrease in interest rates initiated by two major Brazilian public banks which virtually dominated the expansion of loan operations within this period The loan specialized banks had higher average efficiency indicators than that of the retail banks and this is the result of their higher input productivity in relation to the loan operations generated by these institutions In other words public banks and retail banks may generate greater loan operation activities both in the number of operations as well as financial resources but their resources employees deposits and capital are less productive than the resources of banks specialized in credit The DEA analysis does not consider the size of a unit to classify it as efficient what is considered in this type of analysis is the use of resources inputs to achieve the product outputs Efficiency is the relationship between the results obtained and the resources used Accordingly managerial aspects resource allocation decisions are more relevant in the DEA analysis than the size of the decision making unit Tables 10 and 11 identify the less efficient banks considering the segment in which they operate For the Retail Bank category the least efficient bank was Citibank and for the Banks Specialized in Credit category JP Morgan was the most inefficient For Citibank the DEA suggests another bank as a benchmark the Safra bank both midsized Comparing Citibank to Safra enables identify the Table 9 Efficiency indicators for banks specialized in credit Banks 2010 2011 2012 2013 ABCBrasil 9287 9221 7974 7608 Alfa 8224 8000 7976 8949 Bancoob 4419 8366 6178 5925 BIC 7100 6484 6485 6075 BMG 7388 6109 8559 8082 BNP Paribas 5436 5312 3962 2972 BTG Pactual 1579 1422 1512 1912 Credit Suisse 6471 7418 8023 5203 Deutsche 619 8183 4514 2738 Fibra 7907 8461 7892 8665 JP Morgan Chase 5970 723 774 583 Panamericano 8457 6666 5806 5815 Votorantim 7975 7815 8275 8650 Bansicredi 6572 8485 7504 7500 Daycoval 6152 6392 5574 5313 Societe Generale 8542 9003 9003 8973 Table 10 Citibank Safra Citibank Safra Average efficiency 2550 7568 Deposits R 62579957 52290920 Net equity R 27623372 26014704 Number of employees 6190 5753 Credit operations R 51511037 156610079 Table 11 JP Morgan Alfa JP Morgan Alfa Average efficiency 2012 8287 Deposits R 6504849 8141772 Net equity R 11353675 7619346 Number of employees 737 888 Credit operations R 1439015 26979487 Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 558 Périco A E et al underutilization of some resources by the first bank Table 10 describes some of the data Using the DEA technique enabled estimate the growth potential of Loan Operations from the observed input levels of Citibank This bank could increase its output by 215 considering its resources available each year Table 11 displays the comparison between JP Morgan and its benchmark the Alpha Bank both midsized The JP Morgan bank given the resources employed had the potential to expand its lending operations by more than 1000 in the overall period Table 12 was prepared from these results in order to establish the following relationships credit operations deposits credit operationstotal assets and credit operationsemployees Table 12 enables identify the productivity of the variables of the model and verify the reason why some banks are considered more efficient than others For example if the efficiency of a bank were exclusively dependent on the relationship between deposits and operations the Alpha Bank would be more efficient since each deposit unit of this bank is converted into 331 units of loan operations Accordingly the least efficient would be JP Morgan 022 In the Credit OperationTotal Assets relation the most efficient bank was Sicredi since each total assets unit produced 1583 units of loan operations and again the least efficient was JP Morgan 013 And finally if bank efficiency were to be defined only by the productivity of employees the Credit Suisse bank would have been the most efficient with each employee having produced 390806 units of credit operations JP Morgan would again be the least efficient 1954 Of the top five retail banks operating in the country Banco do Brasil Caixa Econômica Federal Bradesco Itaú and Santander the most productive bank considering only the deposits feature was Santander considering only the total assets feature it was Caixa Econômica Federal and considering only the employees feature it was Banco do Brasil By demonstrating that a bank is more efficient in some variables Table 11 than others we do not state that the situation of the bank model is ideal since there is no consensus in the literature about Table 12 Data banks Banks Credit OperationsDeposits Credit Operations Total Assets Credit OperationsEmployees ABCBrasil 224 471 5581 Alfa 331 354 3039 Bancoob 048 1116 4515 Banestes 051 395 375 Banrisul 085 463 713 Bansicredi 093 1583 9545 Basa 083 115 220 Banco do Brasil 100 738 1378 BIC 127 526 4579 BMG 196 454 9406 BNB 116 456 346 BNP Paribas 117 224 3857 Bradesco 116 404 1022 BRB 083 651 524 BTG Pactual 044 081 2393 CEF 108 1424 1075 Citibank 082 186 832 Credit Suisse 116 118 39080 Daycoval 177 327 2675 Deutsche 043 103 1749 Fibra 137 739 4873 HSBC 073 516 631 Itaú 120 385 969 JP Morgan Chase 022 013 195 Mercantil do Brasil 097 875 726 Panamericano 131 479 5307 Safra 299 602 2722 Santander 139 264 1310 Societe Generale 242 295 2303 Votorantim 295 673 13193 Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 559 Périco A E et al the optimal capacity of such specific production factors Therefore when we declare for example that the Credit Suisse bank is the most efficient considering only the credit operations by employees we are only saying that the employees of this bank are more productive in comparison to the employees of the other 29 banks Bank efficiency is related to the balance between the input resources employees financial resources structure and equipment necessary for a banks operation and etc and output resources loan operations income from financial intermediation and etc for the quality of such services among other factors The analysis proposed in this work was purely about financial intermediation which showed that all the banks investigated can and should improve the performance of their production resources However we delimit the conclusions to the sample investigated quite aware that if the comparison undertaken encompassed large banks from different countries Brazilian banks would admittedly be more distant from the efficient frontier 5 Final Remarks This paper proposes measuring the efficiency of 30 Brazilian banks using the data envelopment analysis methodology This technique was applied to a set of banks in 2010 2011 2012 and 2013 The concept of efficiency has traditionally been related to the reason output input of investigated units The unit that generates more output with fewer resources is considered the most efficient In configurations of multiple inputs and output the allocation of weights to the inputs and output is required in order to calculate the efficiency of production units The DEA is a nonparametric technique that uses linear programming to determine optimal weights that minimize the distance between the efficient frontier and the unit investigated The main advantage of DEA is that it does not require the specification of a production function The DEA uses a set of inputs that the unit wants to minimize and a set of output that the unit wants to maximize The disadvantage of the technique is that the statistical inference is very difficult to apply in efficiency scores Therefore the DEA bootstrap procedure Simar Wilson 1998 allows extraction of the sensitivity of scores stemming from sample distribution inefficiency In this direction we seek to benefit from the advantages of using DEA and take care of what we consider the fragility of it through the use of DEA bootstrap It should be emphasized that the results obtained do not refer to absolute efficiency The banks considered efficient are only classified this way for the group analyzed In this work the matter of the best classification among retail banks and banks specialized in credit has little relevance The low relevance in this type of comparison is a result of analysis by segments segments were analyzed separately The data envelopment analysis allows to calculate the efficiency indicators of homogeneous units and of different sizes However even for banks of different sizes it was understood that they were not effectively homogeneous since they operate in different niches and therefore perform different operations Accordingly the classification of efficient retail banks is completely set apart from the classification of banks specialized in credit Therefore in this analysis the efficiency increase gains more emphasis to each segment investigated In September 2012 the Valor Econômico newspaper published an article entitled With falling interest rates banks pursue efficiency gains Mandl 2012 and reported some efficiency results of large Brazilian banks The paper discusses the results reported by Goldman Sachs declaring that compared to their peers around the world Brazilian banks were the least efficient The results presented here although only addressing Brazilian reality point to an average level of efficiency for the large Brazilian banks in the period analyzed which somehow corroborates the results reported by Goldman Sachs when referring to the efficiency of large Brazilian banks This is a new scenario which not only requires efforts to reduce expenses but also to increase revenues even taking into consideration reduced interest rates which result in lower earnings for banks References Banker R D 1996 Hypothesis tests using data envelopment analysis Journal of Productivity Analysis 723 139159 httpdxdoiorg101007BF00157038 Banker R D Natarajan R 2004 Statistical tests based on DEA efficiency scores In W Cooper L Seiford J Zhu Eds Handbook on data envelopment analysis pp 265298 New York Kluwer Academic Publishers Banker R D Charnes A Cooper W W 1984 Some models for estimating technical and scale inefficiencies in data envelopment analysis Management Science 309 10781092 httpdxdoiorg101287mnsc3091078 Berger A N Humphrey D B 1992 Measurement and efficiency issues in commercial banking In Z Griliches Output measurement in the service sectors pp 175212 Chicago University of Chicago Press Berger A N Humphrey D B 1997 Efficiency of financial institutions international survey and directions for future Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 560 Périco A E et al research European Journal of Operational Research 982 175212 httpdxdoiorg101016S0377221796003426 Charnes A Cooper W Lewin A Y Seiford L M 1994 Data envelopment analysis theory methodology and application Boston Kluwer Academic Publishers Departamento Intersindical de Estatística e Estudos Socioeconômicos DIEESE 2013 Um novo cenário para o setor financeiro no Brasil Nota Técnica 123 pp 115 São Paulo DIEESE Retrieved in 23 October 2013 from httpwwwdieeseorg brnotatecnica2013notaTec123CenarioSetorFinanceiropdf Dong F Featherstone A 2004 Technical and scale efficiencies for Chinese rural credit cooperatives a bootstrapping approach in data envelopment analysis Working Paper 04 WP366 Iowa Center for Agricultural and Rural Development 26 p Retrieved in 23 October 2013 from httpwww researchgatenetpublication24085100Technicaland ScaleEfficienciesforChineseRuralCreditCooperativesA BootstrappingApproachinDataEnvelopmentAnalysis Drake L Hall M J B 2003 Efficiency in Japanese banking an empirical analysis Journal of Banking Finance 275 891917 httpdxdoiorg101016S0378426602002406 Dyson R G Shale E A 2010 Data envelopment analysis operational research and uncertainty The Journal of the Operational Research Society 611 2534 httpdxdoi org101057jors2009145 Efron B 1987 Better bootstrap confidence intervals Journal of the American Statistical Association 82397 171185 httpdxdoiorg10108001621459198710478410 Fukuyama H Matousek R 2011 Efficiency of Turkish banking twostage network system Variable returns to scale model Journal of International Financial Markets Institutions and Money 211 7591 httpdxdoi org101016jintfin201008004 Gaganis C Liadaki A Doumpos M Zopounidis C 2009 Estimating and analyzing the efficiency and productivity of bank branches evidence from Greece Managerial Finance 352 202218 httpdxdoiorg10110803074350910923518 Giokas D 2008 Assessing the efficiency in operations of a large Greek bank branch network adopting different economic behaviors Economic Modelling 253 559 574 httpdxdoiorg101016jeconmod200710007 Halkos G E Salamouris D S 2004 Efficiency measurement of the Greek commercial banks with the use of financial ratios a data envelopment analysis approach Management Accounting Research 152 201224 httpdxdoi org101016jmar200402001 Halkos G E Tzeremes N G 2013 Estimating the degree of operating efficiency gains from a potential bank merger and acquisition A DEA bootstrapped approach Journal of Banking Finance 375 16581668 httpdxdoi org101016jjbankfin201212009 Löthgren M Tambour M 1999 Testing scale efficiency in DEA models a bootstrapping approach Applied Economics 3110 12311237 httpdxdoiorg101080000368499323445 Macedo M A S Barbosa A C M 2009 Eficiência no sistema bancário Brasileiro uma análise do desempenho de bancos de varejo atacado middlemarket e financiamento utilizando DEA Revista de Informação Contábil 3 124 Mandl C 2012 September 3 Com juros em queda bancos perseguem ganhos de eficiência Valor Econômico Caderno de Finanças Retrieved in 23 October 2013 from http wwwvalorcombrfinancas2814064comjurosemqueda bancosperseguemganhosdeeficienciacomments Paradi C P Rouatt S Zhu H 2011 Twostage evaluation of bank branch efficiency using Data Envelopment Analysis Omega 391 99109 httpdxdoiorg101016j omega201004002 Pastor J T Lovell C A K Tulkens H 2006 Evaluating the financial performance of bank branches Annals of Operations Research LouvainlaNeuve 1451 321337 httpdxdoiorg101007s1047900600383 Périco A E Rebelatto D A N Santana N B 2008 Eficiência bancária os maiores bancos são os mais eficientes Uma análise por envoltória de dados Gestão Produção 152 421431 httpdxdoiorg101590 S0104530X2008000200016 Portela M Thanassoulis E 2007 Comparative efficiency analysis of Portuguese bank branches European Journal of Operational Research 1772 12751288 httpdxdoi org101016jejor200601007 Sealey C Jr Lindley J T 1977 Inputs outputs and a theory of production and cost at depository financial Institutions The Journal of Finance 324 12511266 httpdxdoiorg101111j154062611977tb03324x Simar L Wilson P 2000 A general methodology for bootstrapping in nonparametric frontier models Journal of Applied Statistics 276 779802 httpdxdoi org10108002664760050081951 Simar L Wilson P W 1998 Sensitivity analysis of efficiency scores How to bootstrap in nonparametric frontier models Management Science 441 4961 http dxdoiorg101287mnsc44149 Stavarek D 2005 Estimation of Factors Influencing Efficiency of Banks in New Member States before Joining the European Union Ekonomicky Casopis 53 593610 Xue M Harker P T 1999 Overcoming the inherent dependency of DEA efficiency scores a bootstrap approach Working Paper 9917 The Wharton Financial Institutions Center Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 561 Périco A E et al Appendix A Bootstrap procedures This appendix illustrates the bootstrapbased algorithm introduced by Simar Wilson 1998 2000 The procedure is as follows Step 1 Transform the inputoutput vectors using the original efficiency estimates 1 ˆ ˆ ˆ l in i i i in i i n as x y x y θ θ Step 2 Generate smoothed resampled pseudoefficiencies iy as follow Step 21 Given a set of estimated efficiencies ˆin θ use the rule of thumb Silverman 1986 p 4748 to obtain the bandwidth parameter h as h 09 15 13 ˆ min 4 ˆ 1 3 σθ R where ˆˆθ σ is the standard deviation of ˆin θ and 13 R is the interquartile range of the empirical distribution of ˆin θ Step 22 Generate iδ by replacing with replacement from the empirical distribution of ˆin θ estimated efficiencies Step 23 Generate the sequence iδ using 1 2 i i i i i i i h if h h otherwise δ ε δ ε δ δ ε where iε is drawn iid from a standard normal distribution Step 24 Generate smoothed pseudoefficiencies iγ using the following formula ˆ 2 2 1 ˆ i i i i h θ δ δ δ γ σ where 1 n i i i n δ δ which is the average of the resampled original efficiencies Step 3 Let the pseudodata be given by ˆ l i i i i i x y x y γ Step 4 Estimate the bootstrap efficiencies using the pseudodata as 1 1 ˆ n SW n in z i i i i min y Yz x X z z z R θ θ θ θ Step 5 Repeat steps 24 2000 times to create a set of 2000 bankspecific bootstrapped efficiency estimates 1 1 2000 ˆSW b in i n b θ References Silverman B W 1986 Density estimation for statistics and data analysis monographs on statistics and applied probability London Chapman Hall Simar L Wilson P W 1998 Sensitivity analysis of efficiency scores How to bootstrap in nonparametric frontier models Management Science 441 4961 httpdxdoiorg101287mnsc44149 Simar L Wilson P 2000 A general methodology for bootstrapping in nonparametric frontier models Journal of Applied Statistics 276 779802 httpdxdoiorg10108002664760050081951
Send your question to AI and receive an answer instantly
Recommended for you
4
Analise de Dividendos Calculo de Media Desvio Padrao e Coeficiente de Variacao
Estatística Aplicada para Finanças
FGV
37
Harvard Case Estatística
Estatística Aplicada para Finanças
FGV
5
Questão Central
Estatística Aplicada para Finanças
FGV
7
Job Task 1 Estatistica Aplicada
Estatística Aplicada para Finanças
FGV
6
Job Task 02 Estatistica
Estatística Aplicada para Finanças
FGV
2
Exercícios Resolvidos de Probabilidade e Estatística: Defeitos em Fabricação e Medidas Estatísticas
Estatística Aplicada para Finanças
FGV
1
Estatística - Cálculo do Tamanho da Amostra e Plano Amostral
Estatística Aplicada para Finanças
UFV
2
Analise do Impacto da Pandemia no Turismo da Bahia - Estudo de Correlação e Modelagem Preditiva
Estatística Aplicada para Finanças
UFV
11
Fórmulas e Testes Estatísticos de Correlação
Estatística Aplicada para Finanças
FAAP
1
Atividade sobre Amostragem e Métodos Estatísticos
Estatística Aplicada para Finanças
UFV
Preview text
EFICIÊNCIA OPERACIONAL NOS PRINCIPAIS BANCOS PÚBLICOS DO BRASIL OPERATIONAL EFFICIENCY IN THE MAIN PUBLIC BANKS OF BRAZIL Carlos Viegas Neto Ângela Rozane Leal de Souza RESUMO O estudo analisa a eficiência operacional dos principais bancos públicos do Brasil no período de 2010 a 2018 utilizando a metodologia de Análise Envoltória de Dados DEA e o Índice de Eficiência Operacional IEO A pesquisa caracterizase como quantitativa descritiva e documental com nove bancos compondo a amostra do estudo A análise da eficiência sob o prisma do IEO revelou dois grupos de bancos com patamares distintos de eficiência Além disso sinalizou uma tendência de melhora da eficiência no período Pela análise DEA verificouse que os bancos Banco do Brasil Banco do Estado de Sergipe SA e Banco do Estado do Pará SA tiveram os melhores resultados ficando na fronteira de eficiência em todos os anos avaliados Ainda pela DEA verificouse os pesos atribuídos às variáveis e o distanciamento dos bancos ineficientes da fronteira de eficiência Na comparação entre as duas ferramentas de análise percebeuse que há resultados análogos e divergentes destacandose a divergência de resultados nos bancos Banco do Estado do Espírito Santo SA e Banco do Estado do Pará SA por conta das variáveis utilizadas O estudo contribui a partir das análises com a gestão de eficiência dos bancos envolvidos pois fornece informações que podem direcionar a atuação dos bancos na busca por melhorias e também para seus stakeholders Palavraschave Eficiência Análise Envoltória de Dados Banco ABSTRACT The study analyzes the operational efficiency of the main Brazilian public banks from 2010 to 2018 using the Data Envelopment Analysis DEA methodology and the Operational Efficiency Index IEO The research is characterized as quantitative descriptive and documentary with nine banks composing the study sample The efficiency analysis from the IEO perspective revealed two groups of banks with different levels of efficiency In addition it signaled a trend of improved efficiency in the period Through the DEA analysis it was verified that the banks Banco do Brasil Banco do Estado de Sergipe SA and Banco do Estado do Pará SA had the best results remaining on the efficiency frontier in all evaluated years Also by the DEA the weights attributed to the variables and the distancing of the inefficient banks from the efficiency frontier were verified In the comparison between the two tools of analysis it was noticed that there are similar and divergent results highlighting the divergence of results in the banks Banco do Estado do Espírito Santo SA and Banco do Estado do Pará SA due to the variables used The study contributes from the analyzes to the Trabalho de Conclusão de Curso apresentado no primeiro semestre de 2019 ao Departamento de Ciências Contábeis e Atuariais da Faculdade de Ciências Econômicas da Universidade Federal do Rio Grande do Sul UFRGS como requisito parcial para obtenção do título de Bacharel em Ciências Contábeis Graduando em Ciências Contábeis na Universidade Federal do Rio Grande do Sul carlosvnetogmailcom Mestra em Ciências Contábeis pela Universidade do Vale do Rio dos Sinos UNISINOS Doutora em Agronegócio pela Universidade Federal do Rio Grande do Sul UFRGS Professora do Departamento de Ciências Contábeis e Atuarias da UFRGS 2 efficiency management of the banks involved since it provides information that can guide the performance of the banks in the search for improvements and also for their stakeholders Keywords Efficiency Data Envelopment Analysis Bank 1 INTRODUÇÃO A intermediação financeira exercida predominantemente pelos bancos possui papel central na economia brasileira CLEMENTE KÜHL 2006 Os cinco bancos com maior participação de mercado movimentaram o total de 28 trilhões de reais em operações de crédito durante o ano de 2017 DEPARTAMENTO INTERSINDICAL DE ESTATÍSTICAS E ESTUDOS SOCIOECONÔMICOS DIEESE 2017 Esse valor representa quase a metade do PIB brasileiro no mesmo ano o que denota a relevância do segmento O mercado bancário é composto por instituições públicas e privadas que atuam no varejo Os bancos públicos no Brasil possuem variados tamanhos e alguns deles abrangência nacional Embora sejam bancos com viés social e por vezes atuem como agentes de políticas de estado precisam ser competitivos em seus processos para fazer frente à forte concorrência do segmento e garantir os resultados esperados ARAÚJO CINTRA 2011 Essas instituições possuem o nível operacional com amplo espectro de atuação bancos múltiplos Segundo Fortuna 2011 os bancos múltiplos operam e administram várias carteiras de clientes sendo que uma delas deve obrigatoriamente ser comercial ou de investimentos Nesse contexto para alcançar seus clientes esses bancos tradicionalmente utilizamse de agências e correspondentes bancários espalhados por todo o país Ocorre entretanto que os canais de atendimento que não exigem a estrutura física de um ponto de venda passaram a ser muito utilizados pelas empresas financeiras Não por acaso verificase um forte movimento de encerramento de agências físicas BANCO CENTRAL DO BRASIL BACEN 2017 Mais que isso com o advento das fintechs financeiras de tecnologia as quais fornecem atendimento e realizam negócios com baixos custos operacionais exclusivamente por plataformas virtuais tornouse imperioso que os cuidados com a eficiência operacional subissem o patamar de importância nos bancos convencionais No modelo de negócios de pagamentos por exemplo as fintechs captam clientes oriundos dos bancos tradicionais sem necessitar de grandes investimentos ARAÚJO 2018 Tratar de eficiência conforme Périco Rebelatto e Santana 2008 significa obter ganhos de produtividade sem aumentar a quantidade de dispêndios Ou seja é a capacidade de otimizar o atingimento de resultados mediante uso de tecnologia ou melhoria de processos No caso específico dos bancos um indicador importante apontado nos relatórios de estabilidade financeira é o Índice de Eficiência Operacional IEO BACEN 2016 o qual informa o quanto o banco precisa dispender para gerar receitas Assim diante do crescimento da concorrência no mercado de negócios financeiros e a evolução tecnológica do segmento a eficiência nos bancos públicos ganha relevo e a seguinte questão é levantada qual o comportamento dos resultados obtidos pelos principais bancos públicos do Brasil no tocante à eficiência operacional no período de 2010 a 2018 Ante essa indagação o estudo tem por objetivo analisar a eficiência operacional dos principais bancos públicos do Brasil no período de 2010 a 2018 utilizando a metodologia Análise Envoltória de Dados DEA e o IEO A avaliação da eficiência nas organizações é objeto de estudo recorrente na academia e no meio empresarial BERGER MESTER 1997 Isso acontece porque a melhoria contínua nos processos das empresas é condição para sua perenidade Existem contudo variados métodos para avaliação de produtividade dentre os quais está a DEA Essa técnica utilizase 3 de recursos matemáticos para avaliar a obtenção de resultados por unidades produtivas a partir de processos tecnológicos parecidos CASADO 2007 Tratandose de bancos públicos o tema ganha destaque por se tratar de interesse difuso Araújo e Cintra 2011 explicam que os bancos públicos exercem papel diferenciado e entre outras funções têm o papel de atuar na oferta anticíclica de crédito Assim o estudo contribui com a gestão de eficiência dos bancos e fornece informação útil não só para as organizações envolvidas mas também para seus stakeholders 2 REFERENCIAL TEÓRICO Nesta seção é apresentada a configuração do sistema financeiro nacional bem como o papel dos bancos nesse contexto Os conceitos associados à eficiência em todos os ambientes de modo geral e no prisma do mercado bancário são abordados na sequência Depois disso apresentase aspectos relacionados ao conceito e ao modelo de avaliação DEA os quais são utilizados como ferramenta para a pesquisa em apreço Por fim são apresentados outros estudos que buscaram investigar a eficiência bancária com enfoques semelhantes à presente pesquisa 21 SISTEMA FINANCEIRO O principal instrumento regulatório do sistema financeiro nacional é a Lei n 4595 de 31 de dezembro de 1964 Nela é estabelecido os principais atores do sistema financeiro a saber Conselho Monetário Nacional CMN Banco Central do Brasil BACEN Banco do Brasil SA Banco Nacional do Desenvolvimento Econômico e demais instituições financeiras públicas e privadas Todas essas organizações em conjunto são responsáveis pelo adequado funcionamento desse sistema o qual possibilita as diversas transações eou operações financeiras entre os participantes do mercado Fortuna 2011 p 16 define sistema financeiro como sendo Um conjunto de instituições que se dedicam de alguma forma ao trabalho de propiciar condições satisfatórias para a manutenção de um fluxo de recursos entre poupadores e investidores O mercado financeiro onde se processam essas transações permite que um agente econômico qualquerum indivíduo ou empresa sem perspectivas de aplicação em algum investimento próprio da poupança que é capaz de gerar denominado como um agente econômico superavitário seja colocado em contato com outro cujas perspectivas de investimento superam as respectivas disponibilidades de poupança denominado como agente econômico deficitário Em outras palavras podese dizer que é um sistema que permite a circulação de forma regrada confiável e dinâmica de valores financeiros entre os diversos agentes da sociedade Ele é composto por agentes reguladores supervisores e operadores O Quadro 1 ilustra essa configuração 4 Quadro 1 Organização Sistema Financeiro Nacional Moeda Crédito Capitais e Câmbio Seguros Privados Previdência Fechada Reguladores Conselho Monetário Nacional Conselho Nacional de Seguros Privados Conselho Nacional de Previdência Complementar Supervisores Banco Central Comissão de Valores Mobiliários Superintendência de Seguros Privados Superintendência Nacional de Previdência Complementar Operadores Bancos e Caixas Econômicas Administradores de Consórcios Bolsa de Valores Seguradoras e Resseguradoras Fundos de Pensão Cooperativas de Créditos Corretoras e distribuidoras Bolsa de Mercadorias e Futuro Entidades Abertas de Previdência Instituições de Pagamento Demais Instituições não Bancárias Sociedades de Capitalização Fonte adaptado BACEN 2019 Conforme podese ver no Quadro 1 há uma segregação detalhada dos agentes envolvidos no sistema Além de separar as entidades por funções também o faz por área de atuação Nessa divisão os bancos e caixas econômicas estão dentre os operadores do sistema no que tange à moeda crédito capitais e câmbio De maneira geral são essas últimas instituições que materializam o sistema financeiro para a maioria das pessoas e empresas através de seus serviços FORTUNA 2011 Embora ofereçam uma gama vasta de serviços os bancos e também as caixas econômicas que se comportam como bancos múltiplos na atualidade têm como principal papel a intermediação financeira A realização desse serviço possibilita a melhor eficiência do sistema por diversos motivos conforme explica Rosa 2014 Os bancos conseguem reduzir os custos das transações em razão da escala com que operam e com os investimentos em tecnologia para melhorar o processo o que seria inviável se os agentes atuassem individualmente Além disso concentrando as operações em bancos há uma curva de aprendizagem que possibilita reduzir a assimetria de informações Com isso a avaliação de risco tornase qualificada e os recursos captados tendem a ser direcionados de forma otimizada para os agentes deficitários Mesmo a Lei n 459564 sendo o principal instrumento regulatório ela não esgota a legislação sobre o assunto Existem diversas leis que também foram marcos na regulação como as leis 638576 Lei da CVM e 640476 Lei das SA posteriormente alterada pela Lei n 1030301 5 22 EFICIÊNCIA Do ponto de vista econômico um dos grandes desafios enfrentados pelas organizações e pela sociedade como um todo é a adequada utilização dos fatores de produção capital terra trabalho tecnologia tendo em vista a sua disponibilidade limitada e a necessidade de atender demandas insaciáveis Seguindo esse raciocínio conforme Rossetti 2003 p 242 a escassez confrontada com as necessidades ilimitáveis torna onerosos tanto o processo produtivo quanto o de escolha sobre o que e quanto produzir Nessa esteira percebese que os economistas concentram grandes esforços na direção de encontrar opções para melhorar o aproveitamento dos recursos MATTOS TERRA 2015 Tratar os recursos limitados de forma a otimizar os resultados produzidos está associado ao conceito de eficiência o qual pode ser compreendido por diferentes ramificações Algumas dessas maneiras de enxergar a eficiência são classificadas como eficiência técnica eficiência alocativa e eficiência social MATTOS TERRA 2015 No que concerne ao setor bancário é usual trabalhar com a chamada eficiência operacional a qual está associada com despesas de funcionamento da instituição financeira PRIMO et al 2013 Essa medida não é apenas um instrumento de gestão mas sim um dos principais influenciadores para lucratividade de instituições bancárias VENNET 2002 A eficiência operacional enquanto uma relação de despesas operacionais e receitas evidencia o quanto de estrutura operacional é necessária par fazer frente às operações cotidianas ASSAF NETO 2015 Ressaltase contudo conforme lembram Neves Júnior et al 2007 que a eficiência operacional não é calculada de forma uniforme pelos bancos Ainda no tocante à eficiência em bancos Spong Sulivam e DeYoung 1995 entendem que a eficiência passa pela geração de produtos e serviços que gerem renda e ao mesmo tempo sejam realizados com controle de custos Destacam como fatores determinantes também a estrutura organizacional e de pessoas da instituição SPONG SULIVAM DEYOUNG 1995 A natureza jurídica dos bancos público privado ou economia mista também pode influenciar a eficiência da instituição dadas as suas peculiaridades Apesar disso é importante destacar que Altunbas Evans e Molyneux 2001 não percebem uma supremacia dos bancos privados em relação à eficiência considerando seus estudos com bancos alemães Esse mesmo pensamento é compartilhado por Staub Souza e Tabak 2010 os quais realizaram suas pesquisas considerando bancos brasileiros 23 ANÁLISE ENVOLTÓRIA DE DADOS DEA Medidas de eficiência podem ser aferidas por diferentes caminhos Casado 2007 explica que em economia é usual valerse de regressão por mínimos quadrados ordinários sendo portanto um recurso que utiliza uma função de média Uma alternativa a esse modelo é a avaliação com técnicas não paramétricas as quais permitem a obtenção da fronteira eficiente sem a necessidade de pressupostos sobre as variáveis ou funções de produção eou de custo CAMARGO JÚNIOR MATIAS 2005 p 3 Além disso é um instrumento importante para mensurar o desempenho relativo de instituições MAÇADA 2001 Macedo Santos e Silva 2006 acrescentam também que a técnica não paramétrica faz uso de ferramentas de programação linear no lugar de estatística A DEA é uma técnica não paramétrica e é conceituada da seguinte maneira segundo Casado 2007 p 60 A Análise Envoltória de Dados DEA é uma técnica nãoparamétrica que emprega programação matemática para construir fronteiras de produção de unidades produtivas DMUs que empregam processos tecnológicos semelhantes para 6 transformar múltiplos insumos em múltiplos produtos Tais fronteiras são empregadas para avaliar a eficiência relativa dos planos de operação executados pelas DMUs e servem também como referência para o estabelecimento de metas eficientes para cada unidade produtiva É importante observar que as DMUs citadas no conceito são qualquer unidade produtiva podendo ser uma empresa um setor um grupo etc A única condição é que tenham tarefasatividades semelhantes e com os mesmos objetivos para permitir uma comparação consistente conforme esclarecem Mello et al 2005 A avaliação de DMUs utilizando a metodologia DEA pode seguir diferentes abordagens Dois modelos CCR e BCC destacamse por serem amplamente utilizados observam Périco Rebelatto e Santana 2008 O modelo pioneiro CCR nome oriundo das iniciais de seus criadores tem como característica uma relação proporcional entre as variáveis de entrada e saída Ou seja assume que as DMUs apresentam retornos constantes de escala Charnes Cooper e Rhodes 1978 Já o modelo BCC cujo nome também é proveniente das iniciais de seus criadores rompe a necessidade da premissa de retornos constantes de escala das variáveis consideradas BANKER CHARNES COOPER 1984 Permitir a comparação entre variáveis sem retornos constantes de escala ampliou a gama de aplicações e segundo LaPlante e Paradi 2015 tornou a BCC uma abordagem notável Além das DMUs outro componente para construção do modelo DEA são as variáveis de entradas e saídas que serão utilizadas como parâmetro de avaliação Conforme salientam Branco et al 2016 essas variáveis são escolhidas de acordo com o objetivo da análise A seleção desses itens entretanto não é uma tarefa trivial Périco Rebelatto e Santana 2008 verificaram que a maior parte dos estudos acerca do tema restringemse em afirmar que utilizaram as variáveis mais adequadas para avaliação do desempenho das DMUs avaliadas mas não detalham de maneira sistemática o método de seleção Existem contudo técnicas para realizar essa tarefa de escolha de variáveis Senra et al 2007 destacam alguns métodos como o de Norman e Stoker o qual alia análise de correção simples e DEA cita o método baseado em 3 estágios de Golany e Roll e menciona ainda o método de Lins e Moreira o qual seria um melhoramento do proposto por Norman e Stoker De fato independente do formato da escolha das variáveis é inequívoco que as entradas e saídas precisam ser escolhidas com cautela para não perder o aspecto discriminatório do modelo alertam Freaza Guedes e Gomes 2006 Escolhido o modelo e as variáveis a serem utilizadas ainda resta definir a orientação a ser avaliada input ou output Na orientação input buscase eficiência a partir da redução das entradas e manutenção da saída Já a orientação a output o objetivo é aumentar os resultados sem variar os insumos PÉRICO REBELATTO SANTANA 2008 A análise DEA apresenta como vantagem a dispensa de especificação funcional para a fronteira de produção portanto não requer que os inputs e outputs tenham uma relação determinada MACEDO SANTOS SILVA 2006 A ausência de especificação funcional como aspecto relevante que facilita e incentiva a utilização da DEA é reforçada por Gomes Oliveira e Matias 2017 p 657 a vantagem de ser computacionalmente simples uma vez que não requer a estimativa estocástica de uma função de produção Além disso a DEA tem como característica a possibilidade de utilizar múltiplas variáveis de entrada e saída sem exigência de qualquer suposição acerca da distribuição das variáveis segundo Freaza Guedes e Gomes 2006 7 24 ESTUDOS RELACIONADOS As pesquisas com instituições financeiras especificamente bancos dada a sua relevância na economia são vastas Da mesma forma o enfoque de análise de eficiência com a abordagem DEA também é frequente Assim na sequência são apresentados alguns estudos que evidenciam o interesse no assunto Analisar a eficiência de bancos que operam no Brasil utilizando DEA e Índice de Malmquist foi o que fizeram Martin et al 2011 Na amostra de oito bancos entre outros resultados verificaram que o retorno sobre o capital próprio é diretamente proporcional à eficiência dos bancos O estudo realizado por Pedrosa 2014 concentrou seus esforços para avaliar a eficiência de bancos que atuam no Brasil com controle nacional em relação aos bancos com controle estrangeiro A partir da análise DEA buscou hierarquizar os bancos por índice de produtividade Nesse apanhado concluiu que os bancos com controle nacional são mais eficientes que os de controle estrangeiro embora não haja um distanciamento substancial entre eles Yin Yu e Huang 2018 por seu turno tratam da eficiência de bancos chineses no período de 20052016 Além da abordagem DEA o estudo também utiliza o índice de Gini para realização das análises Os resultados do estudo sugeriram em especial que há heterogeneidades significantes na eficiência dos bancos analisados em todos os aspectos produtividade e lucratividade Outra pesquisa que trata de assuntos atinentes é a realizada por Henriques et al 2018 Novamente bancos brasileiros são alvo de análise de eficiência com a metodologia DEA Trouxeram um destaque interessante para identificar possibilidades de melhoria de eficiência a partir da escala Percebeuse que os maiores bancos possuem dificuldades de apresentar ganhos de eficiência de escala Por fim a avaliação de eficiência em relação as seções judiciárias do poder judiciário da 4ª região federal foi o que pesquisaram Venturini Souza e Bianchi 2018 As análises com ajuda do ferramental DEA percorreram o horizonte compreendido entre 2013 e 2017 e utilizaram as abordagens BCC e CCR nas perspectivas de input e output Dentre os resultados apontados verificouse que ter o maior número de processos julgados não bastava para ser a mais eficiente entre as seções O tema instiga o meio acadêmico e possui amplo espaço para novas pesquisas Além disso como o ambiente corporativo e a exigência do mercado vão sendo alterados ao longo do tempo tornase oportuno efetuar novos estudos direcionados ao assunto 3 PROCEDIMENTOS METODOLÓGICOS A fim de permitir a compreensão do modo como se elaborou a pesquisa apresentam se nesta seção a classificação do estudo e os procedimentos executados para coleta e seleção de dados Além disso constam os critérios de análises e a ferramenta utilizada 31 CLASSIFICAÇÃO DA PESQUISA A pesquisa realizada neste estudo é classificada quanto aos seguintes aspectos quantitativa pela forma de abordagem do problema descritiva de acordo com seus objetivos e documental com base nos procedimentos técnicos utilizados Uma análise quantitativa busca obter respostas através de números A característica dessa abordagem é ratificada por Richardson 1999 o qual afirma que a quantificação das informações coletadas e trabalhadas são a tônica de uma abordagem dessa natureza No estudo em questão a comparação de 8 eficiência entre bancos teve um componente matemático substancial visto que utilizou como ferramental a DEA Como fonte de dados foram utilizadas informações constantes em demonstrações contábeis publicadas por essas instituições bancárias caracterizando deste modo o estudo como documental A partir desses documentos buscouse avaliar as relações existentes entre os bancos avaliados em termos de eficiência operacional Essa forma de trabalhar está em sintonia com o que se entende por uma pesquisa descritiva com procedimento documental GIL 2008 32 AMOSTRA E COLETA DE DADOS Foram selecionados todos os bancos públicos que operam no varejo para realização do estudo em razão de atingirem uma gama maior de clientes em relação aos bancos de desenvolvimento Embora esses bancos possuam diferenças consideráveis de tamanho para fins de utilização da metodologia DEA o essencial é que as unidades sejam do mesmo ramo e que tenham características semelhantes Os bancos no Quadro 2 englobam a amostra escolhida e cumprem com os requisitos mencionados já que atuam no varejo com ampla gama de serviços e com diversas carteiras de clientes Quadro 2 Relação de bancos da amostra Instituição Financeira SiglaAcrônimo Banco da Amazônia SA BASA Banco do Brasil SA BB Banco do Estado de Sergipe SA BANESE Banco do Estado do Pará SA BANPARÁ Banco do Estado do Rio Grande do Sul SA BANRISUL Banco do Nordeste do Brasil SA BNB Banco do Estado do Espírito Santo SA BANESTES Banco de Brasília SA BRB Caixa Econômica Federal CEF Fonte elaborado a partir dos dados da pesquisa 2019 As instituições financeiras do Quadro 2 precisam fornecer suas informações contábeis periodicamente para o BACEN as quais são de acesso público para qualquer usuário Assim esse foi o canal utilizado para obtenção dos insumos que alimentaram as análises Foram capturadas as demonstrações do balanço patrimonial e do demonstrativo do resultado do exercício abrangendo o período de 9 anos 2010 2018 Embora existam estudos analisando a eficiência entre bancos em um único ano como o realizado por Branco et al 2016 optou se por avaliar um intervalo de tempo de forma semelhante ao que fizeram Gomes Oliveira e Matias 2017 O ano de 2018 entretanto foi avaliado sem considerar o BRB pois suas demonstrações não estavam publicadas até o final de abril de 2019 33 CRITÉRIOS DE ANÁLISE Para cálculo do IEO foram utilizados como numerador as despesas de pessoal e despesas administrativas Apesar de o BACEN não utilizar as despesas de pessoal nesse índice BACEN 2016 essa variável é considerada nos relatórios de demonstrações financeiras do BANRISUL BANRISUL 2019 e CEF CEF 2018 Já para o denominador foram utilizadas as receitas de intermediação financeira receita de prestação de serviços e 9 receita de tarifas bancárias Esse é um índice do tipo menormelhor e foi apurado para todos os anos do período analisado A utilização da DEA requer para a comparação de eficiência entre unidades tomadoras de decisão o estabelecimento de variáveis de entrada e saída PÉRICO REBELATTO SANTANA 2008 Não há na literatura uma uniformidade na seleção dessas variáveis pelo contrário existe uma diversidade muito grande nas escolhas desses elementos Contudo é possível observar variáveis já utilizadas em outras pesquisas e que coadunam com a proposta do estudo em questão e por isso servem de amparo para as escolhas realizadas As variáveis de entrada que foram trabalhadas são o ativo não circulante imobilizado as despesas administrativas e as despesas de pessoal Tanto as despesas administrativas quanto de pessoal já foram utilizadas nos trabalhos de Freaza Guedes e Gomes 2006 e Staub Souza e Tabak 2010 Já o ativo permanente também é utilizado de forma recorrente como variável de entrada e aparece nos trabalhos de Henriques et al 2018 e Baten Kasin e Rahman 2015 Com relação à variável de saída optouse por utilizar o resultado antes do imposto de renda utilizado nos trabalhos de Camargo Júnior e Matias 2005 e Kwon e Lee 2015 O Quadro 3 apresenta de forma organizada as variáveis escolhidas Quadro 3 Variáveis de estudo Entrada Input Saída Output Ativo Permanente AP Resultado Antes do Imposto de Renda RE Despesas de Pessoal DP Despesas Administrativas DA Fonte elaborado a partir dos dados da pesquisa 2019 O processamento dos dados coletados foi feito com o software SIAD apresentado por Meza Biondi Neto e Ribeiro 2005 Essa ferramenta permite a implementação de modelos DEA no ambiente computacional facilitando a obtenção dos resultados matemáticos Considerando que não há como garantir que as variáveis de entrada forneçam retornos constantes de escala no caso dos bancos para o estudo serão feitas avaliações na abordagem BCC semelhante ao que fizeram Périco Rebelatto e Santana 2008 Com relação à orientação optouse por avaliar pela ótica das entradas input semelhante ao que fizeram CARAFFINI SOUZA BEHR 2018 4 ANÁLISE DOS DADOS Nesta seção inicialmente é apresentada a evolução do IEO de cada banco avaliado A comparação entre os bancos com enfoque BCC é apresentada na sequência juntamente com a avaliação dos pesos atribuídos às variáveis eficientes e os decréscimos necessários nas variáveis dos bancos que apresentaram ineficiência Para finalizar é traçado um paralelo entre os resultados de eficiência observados pela análise DEA e o IEO 41 ANÁLISE DO IEO O IEO procura mostrar o esforço que o banco faz para gerar receita E isso depende da capacidade do banco de otimizar seus processos internos e também da capacidade de reduzir seus desperdícios o que envolve a eficiência técnica e alocativa MATTOS TERRA 2015 Logo redução do quadro de pessoal bem como redução do número de agências em geral contribuem para a melhoria desse índice A Tabela 1 apresenta os resultados do IEO para cada banco 10 Tabela 1 Índice de eficiência operacional no período de 2010 a 2018 BANCOANO 2010 2011 2012 2013 2014 2015 2016 2017 2018 BASA 3924 3975 4256 4783 4177 3766 3757 3847 3898 BB 2720 2370 2694 2794 2289 1841 2087 2157 2316 BANESE 3849 3169 3588 4025 4535 3779 3760 3800 4292 BANPARÁ 4627 4006 3723 3430 3332 3448 3751 3845 4297 BANRISUL 3013 2679 2825 3086 2844 2348 2593 3036 3317 BNB 4554 3934 4367 3919 3867 2983 3805 4064 4243 BANESTES 2542 2606 2915 2700 2343 2087 1817 2037 2391 BRB 3623 3585 4091 4541 4201 3907 4246 4459 CEF 3078 2777 2938 2717 2179 1862 1869 2079 2333 Fonte elaborada a partir dos dados da pesquisa 2019 Lembrando que se trata de um índice do tipo menormelhor o resultado mais expressivo no índice de eficiência operacional individual observado na Tabela 1 foi obtido pelo BANESTES no ano de 2016 chegando ao valor de 1817 Esse resultado é proveniente em especial do pico obtido nas receitas de intermediação financeira desse banco Na outra ponta ficou o BASA o qual atingiu o valor de 4783 no ano de 2013 e também teve maior influência das receitas do que das despesas Houve portanto uma amplitude de quase 30 pontos percentuais entre os resultados observados nos extremos A média do período por sua vez ficou em 3304 Em geral observase na Tabela 1 que os bancos apresentaram um comportamento de leve melhora de seus índices durante o período Apesar disso os últimos dois anos 2017 e 2018 sugerem um tendência de diminuição da eficiência logo após a fase de melhores resultados ocorrida anos de 2015 e 2016 O grupo formado pelos bancos CEF BANRISUL BB e BANESTES obtiveram os menores valores Ou seja apresentam um melhor índice de eficiência operacional Já o conjunto de bancos que está com esse indicador num patamar desfavorável é composto por BASA BENESE BANPARÁ BNB e BRB 42 ANÁLISE BCC A comparação em termos de eficiência dos 9 bancos públicos considerando os inputs e output escolhidos por meio da técnica DEA na abordagem BCC mostra os bancos que estão na fronteira de eficiência e os que precisam reduzir os seus inputs para atingir o mesmo patamar A Tabela 2 apresenta os resultados obtidos em cada ano do período compreendido entre 2010 e 2018 Tabela 2 Análise BCC BancoAno 2010 2011 2012 2013 2014 2015 2016 2017 2018 BASA 081 033 074 045 097 1 1 1 076 BB 1 1 1 1 1 1 1 1 1 BANESE 1 1 1 1 1 1 1 1 1 BANPARÁ 1 1 1 1 1 1 1 1 1 BANRISUL 1 1 1 1 074 1 1 1 098 BNB 1 1 035 1 1 045 1 1 1 BANESTES 1 055 052 058 084 071 081 084 089 BRB 075 033 061 045 044 031 046 064 CEF 056 1 1 086 1 002 093 1 1 Fonte elaborada a partir dos dados da pesquisa 2019 11 A análise BCC na Tabela 2 identificou como DMUs eficientes em todo o período avaliado os bancos BB BANESE e BANPARÁ Podese incluir ainda o BANRISUL o qual ficou com a máxima eficiência em 7 anos e bem próximo nos outros 2 anos avaliados Nesse resultado observase que os 3 bancos eficientes não são os 3 maiores BB CEF BANRISUL ou os 3 menores BRB BANPARÁ BANESE evidenciando a característica do modelo que permite comparações observando DMUs distintas em termos de tamanho conforme observado por Casado 2007 Já os bancos BASA BANESTES e BRB tiveram desempenho abaixo da fronteira de eficiência em mais da metade dos anos avaliados denotando um distanciamento contínuo Nesse grupo podese incluir também a CEF a qual ficou abaixo em 4 anos A média de eficiência do período atingiu o resultado de 8585 Esse número ficou consideravelmente abaixo do que observou Pedrosa 2014 o qual apurou o valor de 9877 no mesmo modeloorientação Cabe citar contudo que o intervalo de tempo e as variáveis utilizadas não são homogêneos e por isso as diferenças podem ser interpretadas como factíveis e naturais Outro aspecto interessante de se avaliar são os pesos que a DEA atribui às variáveis Na busca pelo melhor conjunto de pesos que resolvem o modelo matemático é possível encontrar soluções que zeram uma ou mais variáveis MELLO et al 2005 Com isso evita se que as variáveis sejam ponderadas de forma discricionária Na Tabela 3 são exibidos os pesos atribuídos pela ferramenta nas variáveis de entrada de cada banco durante o período avaliado 12 Tabela 3 Pesos variáveis de entrada Banco Inputs 2010 2011 2012 2013 2014 2015 2016 2017 2018 BASA AP 0 0 0 0 0 0 0 0 0 DP 0 000000302 0 0 0 000000066 000000062 000000034 0 DA 000000543 0 000000421 000000396 000000407 000000233 000000232 000000298 000000358 BB AP 0 0 0 0 0 0 0 0 0 DP 000000002 000000007 000000007 0 000000005 000000005 000000002 000000004 0 DA 000000006 0 0 000000007 000000001 0 000000003 000000001 000000007 BANESE AP 000000012 0 0 0 000001158 0 0 000000091 0 DP 000000184 000000889 0 0 0 0000001 00000013 0 000000573 DA 000001269 0 000001111 000001172 0 000000814 000000648 000000634 0 BANPARÁ AP 000001072 000000315 0 0 000000902 000000843 0 0 000000353 DP 0 000000679 000000697 000000596 0 0 00000033 000000304 0 DA 000000615 0 0 0 0 0 0 0 000000091 BANRISUL AP 000000004 000000003 0 000000017 0 0 000000023 0 00000001 DP 000000035 000000089 0 0 000000056 000000052 00000002 000000038 000000034 DA 000000093 0 000000128 00000009 000000015 000000012 0 000000019 0 BNB AP 000000204 000000199 000000496 000000424 00000009 000000426 000000466 000000144 000000027 DP 000000033 0 0 0 000000052 0 0 000000039 0 DA 000000042 000000081 0 0 0 0 0 0 000000076 BANESTES AP 000000011 0 0 0 0 0 0 0 0 DP 000000524 000000485 000000438 000000402 0 000000047 000000064 000000041 0 DA 0 0 0 0 000000475 000000382 00000032 00000034 000000389 BRB AP 000000362 000000256 0 0 0 0 0 0 DP 0 0 0 0 0 0 0 000000024 DA 000000176 000000206 00000039 00000033 000000292 000000296 000000255 0000002 CEF AP 000000016 000000013 000000012 00000001 000000002 0 000000007 000000001 0 DP 0 0 0 0 0 0 0 0 0 DA 0 000000001 0 0 000000008 00000001 0 000000008 000000009 Fonte elaborada a partir dos dados da pesquisa 2019 Exceto pelos bancos BANESE BANRISUL e BNB no ano de 2010 todas as ponderações calculadas e apresentadas na Tabela 3 eliminaram pelo menos uma das variáveis Ou seja foi atribuído peso 0 zero a essas variáveis para melhorar o resultado de eficiência do banco Freaza Guedes e Gomes 2006 chegaram a constatações análogas quando no seus estudos tanto os bancos eficientes quanto os bancos ineficientes desprezaram pelo menos uma variável na apuração de eficiência Verificase que a variável DA é o item que menos vezes é zerado pela ferramenta Ou seja na maior parte das vezes é uma variável que contribui para o resultado em termos de eficiência Entretanto todos os bancos apresentam um crescimento de suas despesas ao longo do período com exceção do BB que apresenta queda Já a variável DP é a segunda menos zerada E a melhoria desse item DP exige assim como o verificado por Venturini Souza e Bianchi 2018 um esforço adicional pela característica mais estável do empregado público dificultando a redução do quadro de pessoal 13 A última análise efetuada diz respeito ao distanciamento que as bancos ineficientes encontramse da fronteira de eficiência Os percentuais que cada banco ineficiente precisa diminuir em suas entradas são apresentados na Tabela 4 Tabela 4 Ineficiências Banco Inputs 2010 2011 2012 2013 2014 2015 2016 2017 2018 BASA AP 3346 6832 5059 6516 5472 6250 DP 2953 6603 3846 6522 3974 5098 DA 1903 6980 2554 5474 297 2369 BANRISUL AP 3269 273 DP 2553 273 DA 2553 1417 BNB AP 6516 5493 DP 8841 8598 DA 8269 8015 BANESTES AP 6021 6151 5883 5204 6248 5736 5106 5049 DP 4545 4793 4195 2990 2875 1869 1551 1829 DA 6319 5588 4442 1563 2875 1869 1551 1115 BRB AP 2503 6658 6671 7403 7908 8570 7059 5658 DP 3316 7114 6176 7438 7137 7699 5566 3624 DA 2503 6658 3837 5498 5576 6945 5351 3624 CEF AP 4442 1410 9032 715 DP 8056 7321 9182 6691 DA 7125 5603 9007 5560 Fonte elaborada a partir dos dados da pesquisa 2019 A apuração das unidades ineficientes na Tabela 4 apresenta necessidade de melhoria nas três variáveis de entrada A DA é a variável que menos precisa sofrer ajustes para atingir os mesmos resultados dos bancos eficientes Entretanto o BANRISUL foge desse padrão no ano de 2018 quando a DA apresentou necessidade elevada de melhoria em relação às demais variáveis O ganho de eficiência mediante a redução do AP por sua vez foi o que caracterizou os resultados do BASA BANESTES e BRB Por fim podese verificar que a necessidade de melhora na variável DP foi impactante no BNB e na CEF em especial no ano de 2015 para a CEF A necessidade de redução de AP e DP em bancos ineficientes do ponto de vista da DEA também é observada nos estudos de Henriques et al 2018 O menor impacto da variável DA não surpreende pois coaduna com o observado na Tabela 3 Já a ineficiência por conta da variável AP pode ser explicada em parte pela estratégia dos bancos em disponibilizar unidades em regiões com menor apelo comercial dada o seu caráter público ARAÚJO CINTRA 2011 Em relação à variável DP conforme observado por Venturini Souza e Bianchi 2018 há dificuldades para melhoria de eficiência pela estabilidade dos empregados públicos 43 COMPARAÇÃO DOS RESULTADOS NAS ANÁLISES PELO IEO E BCC A confrontação dos resultados obtidos em cada uma das ferramentas de análise permite ampliar as avaliações efetuadas Isso ocorre em parte pelo fato de se utilizar diferentes variáveis em cada abordagem Essa forma de estudo utilizando mais de uma 14 técnica para comparar ou complementar as análises de eficiência encontra paralelo na literatura nos trabalhos de Neves Júnior et al 2007 Martin et al 2011 e Yin Yu e Huang 2018 A Tabela 5 exibe os bancos eficientes E e ineficientes I por modeloano Tabela 5 Comparação de resultados Ano 2010 2011 2012 2013 2014 2015 2016 2017 2018 Modelo DEA IEO DEA IEO DEA IEO DEA IEO DEA IEO DEA IEO DEA IEO DEA IEO DEA IEO BASA I I I I I I I I I I E I E I E I I I BB E E E E E E E E E E E E E E E E E E BANESE E I E I E I E I E I E I E I E I E I BANPARÁ E I E I E I E I E I E I E I E I E I BANRISUL E E E E E E E E I E E E E E E E I E BNB E I E I I I E I E I I I E I E I E I BANESTES E E I E I E I E I E I E I E I E I E BRB I I I I I I I I I I I I I I I I CEF I E E E E E I E E E I E I E E E E E Fonte elaborada a partir dos dados da pesquisa 2019 Os resultados nas análises IEO e BCC apresentaram correspondências mas também apresentaram situações antagônicas conforme verificado na Tabela 5 Como similaridade podese destacar nos bancos eficientes os resultados do BB e BANRISUL O BB ficou no bloco de bancos eficientes em todo o período avaliado pelo IEO tendo o mesmo comportamento na avaliação DEA Já o BANRISUL apesar de ter apresentado leve piora no IEO nos últimos três anos permaneceu a maior parte do período como banco eficiente em ambas metodologias Nos bancos ineficientes também há bancos que têm ocorrências uniformes nas duas perspectivas Os bancos BRB e BASA compõem esse grupo pois mostram fraco desempenho no IEO e coincidiram com resultados de ineficiência na análise DEA no intervalo de estudo Assim como Martin et al 2011 o estudo identificou falta de convergência em algumas situações As divergências foram observadas por exemplo nos bancos BANESTES e BANPARÁ No IEO o BANESTES atingiu excelentes resultados chegando ao melhor índice entre todos os bancos no ano de 2016 Por outro lado quando a avaliação ocorreu pela DEA o banco computou sucessivos resultados de ineficiência A variável AP utilizada apenas na DEA não responde isoladamente o motivo desse distanciamento nos resultados mas é inegável que impactou negativamente tendo em vista que as ineficiências apontadas na Tabela 4 são mais representativas nessa variável O oposto ocorreu com o banco BANPARÁ cujo resultado no IEO foi fraco e ao mesmo tempo apresentou eficiência em todos os anos na análise DEA A variável que merece destaque nesse caso é a DA Essa variável é considerada nas duas medidas porém no cálculo da DEA ela é zerada em sete dos nove anos da avaliação para esse banco influenciando menos os resultados nessa abordagem Outra forma de avaliar a influência das variáveis na eficiência é através de cálculos estatísticos como fizeram Neves Júnior et al 2007 5 CONSIDERAÇÕES FINAIS O propósito do artigo foi analisar a eficiência operacional dos principais bancos públicos do Brasil BASA BB BANESE BANPARÁ BANRISUL BNB BANESTES BRB e CEF no período de 2010 a 2018 O alcance desse objetivo foi conduzido mediante 15 avaliações da performance dos bancos através de DEA e do IEO bem como relacionando os resultados dos dois instrumentos Após o cálculo do IEO foi possível verificar que os bancos estudados dividiamse em dois grupos O grupo composto por CEF BANRISUL BB e BANESTES apresentaram melhores resultados ou seja um IEO baixo chegando a valores inferiores a 19 O outro grupo BASA BENESE BANPARÁ BNB e BRB exibiu um nível inferior de eficiência ficando o destaque negativo com o banco BASA que atingiu 4783 em 2013 Ambos os grupos sinalizaram evolução no IEO em que pese a leve piora da eficiência no último biênio A análise pela DEA apontou os bancos BB BANESE e BANPARÁ como eficientes em todos os anos Entretanto mais do que fazer um ranqueamento entre os bancos a avaliação por DEA permitiu identificar as variáveis que tiveram maior influência nos resultados de eficiência apurados Nessa avaliação percebeuse que a DA é uma variável que no geral contribuiu positivamente para o resultado dos bancos Evidenciouse ainda o quanto cada banco ineficiente precisa melhorar em cada variável para chegar no nível dos bancos eficientes A DA em relação às demais variáveis ficou menos distante dos resultados dos bancos eficientes A relação entre os resultados das duas avaliações é outro aspecto interessante que foi avaliado Constatouse que nem sempre um banco considerado eficiente na avaliação DEA possui o mesmo comportamento no IEO Essas divergências foram ocasionadas tanto pela diferença de variáveis utilizadas em cada ferramenta como pela influência que variáveis específicas exerceram na apuração dos bancos individualmente Do ponto de vista da análise IEO verificase que há grande margem para melhoria dos bancos ineficientes Ao mesmo tempo para todos os bancos expõe a trajetória dos resultados obtidos possibilitando correções e melhorias Na ótica da DEA os bancos ineficientes têm o direcionamento de onde atuar variáveis para alcançar o resultado do benchmarking Quando há o confrontamento das análises surgem diferenças que também apontam ineficiências que subsidiam a gestão dos bancos envolvidos Esse estudo contribui para o incremento da eficiência operacional nos bancos públicos pois explora e evidencia as lacunas existentes nessas instituições Embora forneça informações concretas e úteis para a análise de eficiência não se pode ser definitivo em relação aos resultados produzidos Sabese que a escolha de outras variáveis pode modificar as posições encontradas na análise DEA e mesmo o IEO pode ser apurado de forma distinta influenciando os resultados Por isso a fim de expandir o conhecimento recomendase a aplicação das análises com outras variáveis ou incluir bancos privados em estudos futuros REFERÊNCIAS ALTUNBAS Y EVANS L MOLYNEUX P Bank ownership and efficiency Journal of Money Credit and Banking Columbus v 33 n 4 p 926954 2001 Disponível em httpswwwjstororgstable2673929 Acesso em 19 nov 2018 ARAÚJO V L CINTRA M A M O papel dos bancos públicos federais na economia brasileira Brasília IPEA Instituto de Pesquisa Econômica Aplicada abr 2011 Disponível em httpwwwipeagovbrportalimagesstoriesPDFsTDstd1604pdf Acesso em 27 out 2018 ARAÚJO M V M Investimento em tecnologia nas instituições financeiras e a influência das fintechs 2018 Dissertação Mestrado em Economia Escola de Economia Fundação Getúlio Vargas São Paulo 2018 Disponível em httpbibliotecadigitalfgvbrdspacehandle1043824740 Acesso em 20 ago 2018 16 ASSAF NETO A Estrutura e análise de balanços um enfoque econômicofinanceiro 11 ed São Paulo Atlas 2015 BANCO CENTRAL DO BRASIL Informações para análise econômicofinanceiro Brasília 2018 Disponível em httpswww3bcbgovbrifdata Acesso em 13 out 2018 BANCO CENTRAL DO BRASIL Relatório de estabilidade financeira Brasília 2016 Disponível em httpswwwbcbgovbrhtmsestabilidade201609refPubpdf Acesso em 15 out 2018 BANCO CENTRAL DO BRASIL Relatórios anuais Brasília 2017 Disponível em httpswwwbcbgovbridREVSFNano2017 Acesso em 21 nov 2018 BANCO CENTRAL DO BRASIL Sistema financeiro nacional Brasília 2019 Disponível em httpswwwbcbgovbrestabilidadefinanceirasfn Acesso em 14 abr 2019 BANKER A CHARNES A COOPER W Some models for estimating technical and scale inefficiencies in data envelopment analysis Management Science Columbus v 30 n 9 p 10781092 1984 Disponível em httpswwwjstororgstable2631725 Acesso em 22 mar 2019 BANCO DO ESTADO DO RIO GRANDE DO SUL Demonstrações financeiras Porto Alegre 2019 Disponível em httpribanrisulcombrbanrisulwebconteudoptaspidioma0tipo8223conta28id 257233 Acesso em 19 maio 2019 BATEN M KASIM M RAHMAN M Cost and profit efficiency of online banks do national commercial banks perform better than private banks The Journal of Internet Banking and Commerce Candler v 20 n 3 p 119 2015 Disponível em httpwwwicommercecentralcomopenaccesscostandprofitefficiencyofonlinebanks donationalcommercialbanksperformbetterthanprivatebanksphpaid64359 Acesso em19 abr 2019 BERGER A MESTER L Inside the black box what explains differences in the efficiencies of financial institutions Journal of Banking Finance Sl v 1 n 21 p 895947 1997 Disponível em httpswwwsciencedirectcomsciencearticlepiiS0378426697000101 Acesso em 15 mar 2019 BRANCO A et al Efficiency of the brazilian banking system an assessment using DEA under three approaches Journal of Applied Finance Banking London v 6 n 4 p 27 42 2016 Disponível em httpswwwscienpresscomjournalfocusaspmainid56SubidIVIssue1835 Acesso em 10 nov 2018 BRASIL Presidência da República Lei Complementar nº 4595 de 31 de dezembro de 1964 Dispõe sobre a política e as instituições monetárias bancárias e creditícias cria o conselho monetário nacional e dá outras providências Brasília 1964 Disponível em httpwwwplanaltogovbrccivil03LEISL4595htm Acesso em 14 abr 2019 17 BRASIL Presidência da República Lei Complementar nº 6385 de 7 de dezembro de 1976 Dispõe sobre o mercado de valores mobiliários e cria a Comissão de Valores Mobiliários Brasília 1976 Disponível em httpwwwplanaltogovbrccivil03LEISL6385htm Acesso em 14 abr 2019 BRASIL Presidência da República Lei Complementar nº 6404 de 15 de dezembro de 1976 Brasília 1976 Dispõe sobre as sociedades por ações Disponível em httpwwwplanaltogovbrccivil03leisl6404compiladahtm Acesso em 14 abr 2019 BRASIL Presidência da República Lei Complementar nº 10303 de 31 de outubro de 2001 Altera e acrescenta dispositivos na Lei nº 6404 de 15 de dezembro de 1976 que dispõe sobre as Sociedades por Ações e na Lei nº 6385 de 7 de dezembro de 1976 que dispõe sobre o mercado de valores mobiliários e cria a Comissão de Valores Mobiliários Brasília 2001 Disponível em httpwwwplanaltogovbrCCIVIL03LEISLEIS2001L10303htm Acesso em 14 abr 2019 CAIXA ECONÔMINA FEDERAL Análise de desempenho 2018 Brasília 2018 Disponível em httpwwwcaixagovbrDownloadscaixademonstrativo financeiroRelatorioAnaliseDesempenho2018pdf Acesso em 19 maio 2019 CAMARGO JÚNIOR A MATIAS A Eficiência operacional de bancos no Brasil América Latina e EUA In CONGRESSO INTERNACIONAL DE CUSTOS 9 2005 Florianópolis Anais Florianópolis CIC 2005 Disponível em httpsanaiscbcemnuvenscombranaisissueview9 Acesso em 21 mar 2019 CARAFFINI J SOUZA R BEHR A Transformação digital e desempenho no setor bancário In CONGRESSO TRANSFORMAÇÃO DIGITAL 1 2018 São Paulo Anais São Paulo CTD 2018 Disponível em httpbibliotecadigitalfgvbrocsindexphpctdctd2018paperview6965 Acesso em 20 maio 2019 CASADO F Análise envoltória de dados conceitos metodologia e estudo da arte na educação superior Revista Sociais e Humanas Santa Maria v 20 n 1 p 5971 2007 Disponível em httpsperiodicosufsmbrsociaisehumanasarticleview907 Acesso em 23 maio 2019 CHARNES A COOPER W RHODES E Measuring the efficiency of decision making units European Journal of Operational Research Sl v 2 p 429444 1978 Disponível em httpswwwsciencedirectcomsciencearticlepii0377221778901388 Acesso em 28 abr 2019 CLEMENTE A KÜHL M Intermediação financeira no Brasil influência da taxa de captação sobre a taxa de aplicação In CONGRESSO USP CONTROLADORIA E CONTABILIDADE 6 2006 São Paulo Anais São Paulo CUCC 2006 Disponível em httpscongressouspfipecafiorganaisartigos62006anresumoaspcodtrabalho236 Acesso em 25 maio 2019 DEPARTAMENTO INTERSINDICAL DE ESTATÍSTICAS E ESTUDOS SOCIOECONÔMICOS Desempenho dos bancos 2017 São Paulo 2017 Disponível em 18 httpswwwdieeseorgbrdesempenhodosbancos2018desemprenhoDosBancos2017html Acesso em 26 out 2018 FORTUNA E Mercado financeiro produtos e serviços 18 ed Rio de Janeiro Qualitymark 2011 FREAZA F GUEDES L GOMES L Análise de eficiência do mercado bancário brasileiro utilizando a metodologia da análise envoltória de dados In SIMPÓSIO BRASILEIRO DE PESQUISA OPERACIONAL 38 2006 Goiânia Anais Goiânia SBPO 2006 Disponível em httpwwwdinuembrsbposbpo2006trabalhoidx00htm Acesso em 07 nov 2018 GIL A Como elaborar projetos de pesquisa 4 ed São Paulo Atlas SA 2008 GOMES M OLIVEIRA S MATIAS A Eficiência do setor bancário brasileiro no período de 2006 a 2013 bancos domésticos x bancos estrangeiros Revista Nova Economia Belo Horizonte v 27 n 3 p 641670 2017 Disponível em httpsrevistasfaceufmgbrindexphpnovaeconomiaarticleview3057 Acesso em 11 dez 2018 HENRIQUES I et al E Efficiency in the Brazilian banking system using data envelopment analysis Future Business Journal Sl v 4 n 1 p 157178 2018 Disponível em httpswwwsciencedirectcomsciencearticlepiiS2314721017300208 Acesso em 30 out 2018 NEVES JÚNIOR I et al M Eficiência operacional uma análise exploratória dos 50 maiores bancos brasileiros pelo ranking BACEN In CONGRESSO USP DE CONTROLADORIA E CONTABILIDADE 7 2007 São Paulo Anais São Paulo CUCC 2007 Disponível em httpscongressouspfipecafiorganaisartigos72007anresumoasppagina9con1 Acesso em 25 mar 2019 KWON H LEE J Twostage production modeling of large US banks A DEAneural network approach Expert Systems with Applications Greenburgh v 1 n 42 p 67586766 2015 Disponível em httpsdlacmorgcitationcfmid2799079 Acesso em 03 maio 2019 LAPLANTE A PARADI J Evaluation of bank branch growth potencial using data envelopment analysis The International Journal of Management Science Omega Sl v 52 n 1 p 3341 2015 Disponível em httpswwwsciencedirectcomsciencearticlepiiS0305048314001303 Acesso em 18 maio 2019 MAÇADA A Impacto dos investimentos em tecnologia da informação nas variáveis estratégicas e na eficiência dos bancos brasileiros 2001 Tese Doutorado em Administração Escola de Administração Universidade Federal do Rio Grande do Sul Porto Alegre 2001 Disponível em httpslumeufrgsbrhandle101831845 Acesso em 05 jun 2019 MACEDO A SANTOS R SILVA F Desempenho organizacional no setor bancário brasileiro uma aplicação da análise envoltória de dados Revista de Administração 19 Mackenzie São Paulo v 7 n 1 p 1144 janmar 2006 Disponível em httpeditorarevistasmackenziebrindexphpRAMarticleview89 Acesso em 10 dez 2018 MARTIN D et al Determinantes da lucratividade de grandes bancos brasileiros uma abordagem em painel com DEA e MPI In ENCONTRO DA ANPAD 35 2011 Rio de Janeiro Anais Rio de Janeiro ENANPAD 2011 Disponível em httpwwwanpadorgbradminpdfFIN930pdf Acesso em 01 jun 2019 MATTOS E TERRA R Avaliação da qualidade do gasto público e mensuração da eficiência Brasília Secretaria do Tesouro Nacional p 211233 2015 Disponível em httpwwwtesourofazendagovbrtesouronacionallancalivrosobrequalidadedogasto Acesso em 22 ago 2018 MELLO J et al Curso de análise envoltória de dados In SIMPÓSIO BRASILEIRO DE PESQUISA OPERACIONAL 37 2005 Gramado Anais Gramado SBPO 2005 Disponível em httpwwwdinuembrsbposbpo2005trabalhoidx00htm Acesso em 17 set 2018 MEZA L BIONDI NETO L RIBEIRO P SIAD V20 Sistema integrado de apoio à decisão uma implementação de modelos de análise envoltória de dados e um método multicritério In SIMPÓSIO BRASILEIRO DE PESQUISA OPERACIONAL 37 2005 Gramado Anais Gramado SBPO 2005 Disponível em httpwwwdinuembrsbposbpo2005trabalhoidx00htm Acesso em18 set 2018 PEDROSA P A E Eficiência bancária no Brasil uma análise por envoltória de dados 2014 Dissertação Mestrado em Economia Faculdade de Economia Universidade de Brasília Brasília 2014 Disponível em httprepositoriounbbrhandle1048217349 Acesso em 02 dez 2018 PÉRICO A REBELATTO D SANTANA N Eficiência bancária os maiores bancos são os mais eficientes Uma análise por envoltória de dados Revista Gestão e Produção São Carlos v 159 n 2 p 421431 maiago 2008 Disponível em httpwwwscielobrscielophppidS0104 530X2008000200016scriptsciabstracttlngpt Acesso em 15 out 2018 PRIMO U et al L Determinantes da rentabilidade bancária no Brasil Revista de Administração e Contabilidade da Unisinos São Leopoldo v10 n 4 p 308323 outdez 2013 Disponível em httprevistasunisinosbrindexphpbaseissueview387showToc Acesso em 05 maio 2019 RICHARDSON R Pesquisa social métodos e técnicas 3 ed São Paulo Atlas 1999 ROSA M A O depósito compulsório e o mercado financeiro 2014 Dissertação Mestrado em Economia Faculdade de Ciências Econômicas Universidade Federal do Rio Grande do Sul Porto Alegre 2014 Disponível em httpslumeufrgsbrbitstreamhandle10183103964000921620pdfsequence1isAllowed y Acesso em 02 dez 2018 ROSSETTI J Introdução à economia 20 ed São Paulo Atlas 2003 20 SENRA L et al Estudo sobre métodos de seleção de variáveis em DEA Pesquisa Operacional Rio de Janeiro v 27 n 2 p 191207 2007 Disponível em httpwwwscielobrscielophpscriptsciarttextpidS010174382007000200001 Acesso em 19 mar 2019 SPONG K SULLIVAN R DEYOUNG R What makes a bank efficient a look at financial characteristics and bank management and ownership structure FRB of Kansas City Review Kansas p 119 Dec 1995 Disponível em httpswwwkansascityfedorgpublicatfipPRS951SpongDeYoungPDF Acesso em 10 mar 2019 STAUB R SOUZA G TABAK B Evolution of bank efficiency in Brazil a DEA approach European Journal of Operational Research Sl v 202 n 1 p 204213 2010 Disponível em httpswwwsciencedirectcomsciencearticlepiiS0377221709002951 Acesso em 10 nov 2018 VENNET R Cost and profit efficiency of financial conglomerates and universal banks in Europe Journal of Money Credit and Banking Columbus v 34 n 1 p 254282 2002 Disponível em httpswwwjstororgstable3270685seq1metadatainfotabcontents Acesso em 20 abr 2019 VENTURINI L SOUZA A BIANCHI M Eficiência na alocação de recursos públicos do poder judiciário um estudo na 4ª região federal In CONGRESSO BRASILEIRO DE CUSTOS 25 2018 Vitória Anais Vitória CBC 2018 Disponível em httpsanaiscbcemnuvenscombranaisissueview50 Acesso em 10 maio 2019 YIN Z YU Y HUANG J Evaluation and evolution of bank efficiency considering heterogeneity technology An empirical study from China PLoS ONE San Francisco v 14 n 4 2018 Disponível em httpsdoiorg101371journalpone0204559 Acesso em 10 mar 2019 1 IMPACTO DOS GASTOS COM TECNOLOGIA NA EFICIÊNCIA DOS PRINCIPAIS BANCOS BRASILEIROS NO PERÍODO DE 2012 A 2022 UMA AVALIAÇÃO USANDO FRONTEIRA ESTOCÁSTICA Filipe de Oliveira Gomes v3 n5 85 2 IMPACTO DOS GASTOS COM TECNOLOGIA NA EFICIÊNCIA DOS PRINCIPAIS BANCOS BRASILEIROS NO PERÍODO DE 2012 A 2022 UMA AVALIAÇÃO USANDO FRONTEIRA ESTOCÁSTICA Filipe de Oliveira Gomes1 1 Filipe de Oliveira Gomes é Mestre em Economia pelo Instituto Brasileiro de Ensino Desenvolvimento e Pesquisa IDP Email filgomeshotmailcom httpsorcidorg0009000424309093 3 O IDP é um centro de excelência no ensino na pesquisa e na extensão nas áreas da Administração Pública Direito e Economia O Instituto tem como um de seus objetivos centrais a profusão e difusão do conhecimento de assuntos estratégicos nas áreas em que atua constituindose um think tank independente que visa contribuir para as transformações sociais políticas e econômicas do Brasil DIREÇÃO E COORDENAÇÃO Diretor Geral Francisco Schertel Coordenador do Mestrado em Economia José Luiz Rossi CONSELHO EDITORIAL Coordenação Thiago Caldeira Renan Holtermann Milton Mendonça Supervisão e Revisão Luiz Augusto Magalhães Mathias Tessmann Apoio Técnico Igor Silva Projeto Gráfico e Diagramação Juliana Vasconcelos wwwidpedubr Revista Técnica voltada à divulgação de resultados preliminares de estudos e pesquisas aplicados em desenvolvimento por professores pesquisadores e estudantes de pósgraduação com o objetivo de estimular a produção e a discussão de conhecimentos técnicos relevantes na área de Economia Convidamos a comunidade acadêmica e profissional a enviar comentários e críticas aos autores visando o aprimoramento dos trabalhos para futura publicação Por seu propósito se concentrar na recepção de comentários e críticas a Revista Debates em Economia Aplicada não possui ISSN e não fere o ineditismo dos trabalhos divulgados As publicações da Revista estão disponíveis para acesso e download gratuito no formato PDF Acesse wwwidpedubr As opiniões emitidas nesta publicação são de exclusiva e inteira responsabilidade dos autores não exprimindo necessariamente o ponto de vista do IDP Qualquer citação aos trabalhos da Série só é permitida mediante autorização expressa dos autores 4 1 Introdução 6 2 Referencial Teórico 11 3 Metodologia 16 31 Estimação da Eficiência 16 32 Base de Dados 17 4 Resultados 23 41 Testes de Especificação 23 42 Resultado da Estimação 25 5 Conclusão 30 6 Referências 32 5 RESUMO Este estudo avalia a relação entre os gastos com tecnologia da informação TI e a eficiência dos principais bancos brasileiros A evolução tecnológica tem impactado fortemente o setor financeiro levando à disponibilização de plataformas digitais e canais virtuais de atendimento aos clientes O maior nível de tecnologia adotada pelos bancos pode contribuir para reduzir custos operacionais e elevar o nível de eficiência das instituições podendo resultar na melhoria dos serviços prestados aos clientes e maior facilidade de acesso ao crédito pela população A análise da relação entre os gastos com tecnologia e a eficiência foi realizada por meio da estimação de eficiência de Fronteira Estocástica SFA com base nas demonstrações financeiras dos cinco maiores bancos brasileiros no período de 2012 a 2022 que representam 76 dos ativos totais do setor Os resultados evidenciaram uma relação positiva entre o fluxo de gastos com TI e o nível de eficiência dos bancos o que significa maior nível de lucratividade Paralelamente o saldo de investimentos de TI não apresentou relevância estatística o que pode ser entendido como um indicativo de mudança no perfil de gastos com tecnologia dos bancos com a migração do modelo proprietário de negócios e soluções de TI para um modelo mais focado em serviços Além disso observouse que uma maior quantidade de pontos de atendimento especializados e menores gastos administrativos favorecem a eficiência dos bancos demonstrando a necessidade de reduzir despesas e alocar recursos da forma mais eficiente possível PALAVRASCHAVE Instituições Financeiras Eficiência Investimentos em Tecnologia da Informação Fronteira estocástica ABSTRACT This study evaluates the relationship between information technology IT expenditures and the efficiency of the main Brazilian banks Technological evolution has strongly impacted the financial sector leading to the availability of digital platforms and virtual customer service channels The higher level of technology adopted by banks can contribute to reducing operating costs and raising the level of efficiency of institutions which may result in improved services provided to customers and easier access to credit by the population The analysis of the relationship between spending on technology and efficiency was performed using the Stochastic Frontier Approach SFA based on the financial statements of the five largest Brazilian banks in the period from 2012 to 2022 which represent 76 of sector totals assets The results showed a positive relationship between the flow of IT expenses and the level of efficiency of the banks which means a higher level of profitability At the same time the balance of IT investments was not statistically significant which can be understood as an indication of a change in the profile of technology spending by banks with the migration from the IT proprietary model of business and solutions to a model more focused on services In addition it was observed that a greater number of specialized service points and lower administrative expenses favor the efficiency of banks demonstrating the need to reduce expenses and allocate resources in the most efficient way possible 6 KEYWORDS Financial Institutions Efficiency Investments in Information Technology Stochastic frontier CLASSIFICAÇÃO JEL C13 G21 O33 7 1 INTRODUÇÃO As Instituições Financeiras são fundamentais para a dinâmica da economia e desenvolvimento socioeconômico do País tendo em vista que os Bancos disponibilizam serviços financeiros acesso ao crédito e soluções de pagamentos dentre outros serviços presentes nas relações econômicas entre indivíduos e empresas Os bancos aparecem como um dos segmentos que mais têm investido em Tecnologia da Informação TI no Brasil e mundo Febraban 2023 tendo grande parte de seus produtos e serviços dependentes dessa tecnologia o que tem resultado na elevação do uso de soluções e compartilhamento de informações seguras e rápidas redução do custo de transações e geração de maior comodidade e satisfação aos usuários É razoável que os gastos em TI sejam indutores da transformação digital e que contribuam para a melhoria de produtos e serviços bem como para a automação e otimização de processos operacionais Sistemas de gerenciamento de riscos e compliance por exemplo podem reduzir o tempo necessário para cumprir regulamentações e aumentar a precisão das avaliações de risco Além disso sistemas de processamento de dados em tempo real permitem que as instituições financeiras tomem decisões mais fundamentadas e ágeis contribuindo para a melhoria da eficiência O conceito de eficiência se refere à capacidade de utilizar os recursos disponíveis da maneira mais produtiva possível para maximizar a produção de bens e serviços Farrel 1957 A eficiência é um aspecto fundamental na teoria econômica e na gestão da produção pois busca alcançar o equilíbrio entre o uso adequado dos recursos e a obtenção dos melhores resultados A teoria da produtividade total dos fatores PTF mede a eficiência com que os fatores de produção incluindo o capital o trabalho e a tecnologia são combinados para gerar a produção Solow 1957 Kendrick 1961 Em linha com esse raciocínio a PTF poderia ser utilizada para relacionar gastos com tecnologia da informação e a melhoria da eficiência de uma instituição financeira 8 Assim se uma instituição financeira aumenta seus gastos com tecnologia isso pode contribuir para melhorar sua PTF e consequentemente sua eficiência Isso ocorre porque a tecnologia pode ajudar a reduzir custos operacionais melhorar o processamento de dados automatizar processos e melhorar a tomada de decisões Contudo o fator determinante para a elevação do nível de eficiência da empresa medido pela sua lucratividade não é o volume dos gastos com tecnologia mas o retorno desses mesmos investimentos e seus impactos na geração de receitas na redução de custos e na mitigação de riscos A partir desse entendimento este estudo empírico pretende analisar se os gastos com tecnologia da informação nos principais bancos brasileiros possuem relação com seus índices de eficiência E se sim pontuar se de forma positiva ou negativa Para atingimento deste objetivo geral serão avaliados outros aspectos relacionados ao tema como i estimar a eficiência dos principais bancos brasileiros ii estabelecer um ranking dos bancos quanto à eficiência iii analisar como os investimentos em tecnologia impactam nos índices de eficiência e iv analisar se a eficiência dos bancos brasileiros é impactada pelos ciclos econômicos representados pela taxa real de juros da economia Na análise da relação entre os gastos com tecnologia da informação e os índices de eficiência dos bancos brasileiros será utilizado o método paramétrico de estimação de eficiência de Fronteira Estocástica SFA Stochastic Frontier Approach ampliando a literatura sobre o tema por considerar dados em painel no período de 2012 a 2022 e por analisar não somente a relação com o saldo estoque dos investimentos em TI mas também quanto aos fluxos correntes de despesas com tecnologia Conhecer os gastos em TI dos principais bancos brasileiros e o seu impacto na eficiência dessas empresas é uma questão essencial para o setor bancário que atua em um ambiente extremamente competitivo e cada vez com menos barreiras a entrantes sendo informação relevante para direcionar a atuação dos gestores o alinhamento dos negócios e o esforço organizacional 9 A elevação da eficiência dos Bancos se mostra relevante na medida em que pode auxiliar na melhoria dos serviços prestados e facilitar o acesso ao crédito à população Além disso a atuação eficiente dos grandes conglomerados financeiros é fator relevante para a estabilidade da economia elevação do grau de confiança do investidor e promoção do desenvolvimento econômico do país A Figura 1 nos mostra a evolução anual do lucro líquido dos cinco maiores bancos brasileiros nos últimos 11 anos 2012 a 2022 onde observamos elevação significativa em valores absolutos que passou do patamar de R 854 bilhões em 2012 para R 1062 bilhões em 2022 correspondendo a um aumento de lucratividade de 24 Figura 1 Lucro líquido anual dos principais bancos brasileiros Fonte Elaborado pelo autor Dados do BCB valores deflacionados trazidos à data base 122022 No mesmo período 2012 a 2022 observamos elevados gastos com tecnologia da informação TI conforme constante na Figura 2 cujo montante anual chegou a atingir em 2016 o equivalente a 38 sobre o lucro líquido anual das instituições o que corresponde ao valor de R 271 bilhões 854 883 890 996 713 902 971 1249 861 1112 1062 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 Lucro R bilhão Variação 10 Figura 2 Gastos com TI dos principais bancos brasileiros Fonte Elaborado pelo autor com base nas demonstrações financeiras dos bancos valores deflacionados trazidos à data base 122022 Cabe ressaltar que uma das principais fontes de receita dos bancos é a margem financeira proveniente da taxa de juros cobrada de seus clientes pelos empréstimos tomados estando esse patamar de juros relacionado à taxa de juros da economia Os juros são uma forma de remunerar o risco que os bancos assumem ao conceder empréstimos aos seus clientes e quanto maior o risco maior deve ser a taxa de juros cobrada Quando a taxa de juros da economia está alta os bancos tendem a aumentar as suas taxas de empréstimo e consequentemente aumentar a sua margem de lucro Por outro lado quando a taxa de juros da economia está baixa os bancos podem ter uma redução na sua margem de lucro já que as taxas de empréstimo também tendem a cair No entanto a baixa taxa de juros pode estimular a demanda por crédito e aumentar o volume de empréstimos o que pode compensar a redução na margem de lucro Além disso uma taxa de juros mais baixa também pode incentivar o consumo e investimento na economia o que pode gerar novas oportunidades de negócios e lucro para os bancos 289 286 292 272 271 274 275 268 256 238 239 34 32 33 27 38 30 28 21 30 21 22 0 10 20 30 40 50 60 70 0 5 10 15 20 25 30 35 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 Gastos TI R Bi Gastos TI Lucro 11 Para considerar os efeitos da inflação calculamos a taxa real de juros da economia ajustando a taxa SELIC ao índice de inflação tendo sido considerado o Índice Nacional de Preços ao Consumidor Amplo IPCAIBGE onde observamos no mesmo período uma variação na taxa real de juros entre o percentual negativo de 378 aa e o patamar máximo de juros positivos de 752 aa no fechamento de 2022 conforme Figura 3 Figura 3 Comparação entre a Taxa SELIC e a Taxa Real de Juros aa 2012 a 2022 Fonte Elaborado pelo autor Dados do BCB e IBGE 2 REFERENCIAL TEÓRICO A Produtividade Total dos Fatores PTF é uma importante forma de análise econômica que busca explicar a relação entre os insumos utilizados na produção e a produção final Essa teoria foi desenvolvida ao longo do tempo por diversos autores como Solow 1956 Jorgenson e Griliches 1967 e Mankiw Romer e Weil 1992 A PTF defende que a produção final de uma economia depende não somente dos insumos utilizados na produção mas também da maneira como esses insumos são combinados e utilizados Assim é possível que uma economia que utiliza uma quantidade 378 752 6 4 2 0 2 4 6 8 10 12 14 16 1T12 2T12 3T12 4T12 1T13 2T13 3T13 4T13 1T14 2T14 3T14 4T14 1T15 2T15 3T15 4T15 1T16 2T16 3T16 4T16 1T17 2T17 3T17 4T17 1T18 2T18 3T18 4T18 1T19 2T19 3T19 4T19 1T20 2T20 3T20 4T20 1T21 2T21 3T21 4T21 1T22 2T22 3T22 4T22 SELIC Taxa Real de Juros 12 menor de insumos possa produzir mais do que outra que utiliza uma quantidade maior de insumos simplesmente porque os insumos são utilizados de forma mais eficiente Além disso a elevação do nível de tecnologia tem importância fundamental na produtividade dos fatores pois permite a utilização mais eficiente dos insumos reduzindo o desperdício de recursos e aumentando a produção A introdução de novas tecnologias pode aumentar a produtividade dos trabalhadores reduzir o tempo de produção e aumentar a qualidade do produto final A importância da tecnologia na produtividade foi destacada por diversos autores a exemplo de Solow 1957 que mostrou que a maior parte do crescimento econômico nos Estados Unidos desde a Segunda Guerra Mundial não podia ser explicada pelo aumento da quantidade de capital e trabalho mas sim pela elevação do nível de tecnologia A PTF consiste portanto em uma importante forma de entender a relação entre os insumos utilizados na produção e a produção final sendo a elevação do nível de tecnologia um fator essencial para aumentar a eficiência na utilização dos insumos e consequentemente a produtividade dos fatores A transformação digital tem sido um dos principais fatores de mudança nos últimos anos no âmbito do setor financeiro não restando dúvidas de que os bancos estão se adaptando rapidamente a essa nova realidade Os bancos estão investindo em tecnologia para melhorar sua eficiência operacional oferecer serviços personalizados e elevar a satisfação dos clientes Por exemplo a implementação de chatbots e assistentes virtuais está melhorando a experiência dos clientes ao lidar com questões simples e rotineiras Além disso a tecnologia de análise de dados tem permitido aos bancos oferecer serviços mais personalizados aos seus clientes como ofertas de produtos financeiros específicos a cada perfil A transformação digital dos bancos é condição necessária para o setor acompanhar o rápido avanço de novas tecnologias do mundo corporativo cuja disseminação tem alterado as condições de mercado e facilitado a entrada de novos competidores sejam eles empresas 13 de tecnologia voltadas ao segmento financeiro as chamadas fintechs ou mesmo grandes empresas de tecnologia que começaram a disponibilizar serviços financeiros as denominadas big techs Essa transformação digital vai além de apenas disponibilizar produtos ou serviços que cada vez mais se tornam commodities financeiras mas sim altera a forma de atender e de se relacionar com os consumidores o que exige uma alteração de posicionamento e de estratégia por parte dos bancos sendo fundamental oferecer aos seus clientes uma melhor experiência com atendimento conveniente ágil e integral interligados em todos os canais físicos e digitais Os gastos e investimentos em TI se mostram relevantes nesse cenário contribuindo para a sustentabilidade dos negócios que cada vez mais se tornam fortemente dependentes de tecnologia Assim esperase observar uma relação positiva entre os gastos com tecnologia da informação e o nível de eficiência dos bancos demonstrando que tais gastos de fato contribuem para a elevação da eficiência O estudo de Feng e Wu 2018 analisou a relação entre o aumento dos investimentos em TI e a performance e valor de mercado de bancos americanos listados em bolsa no período de 2000 a 2017 utilizando modelo de efeitos fixos com análise de regressão linear Os resultados indicaram que investimentos em TI melhoram a eficiência dos bancos principalmente nos grandes bancos Por outro lado o estudo de Gupta et al 2018 utilizando a análise de fronteira estocástica SFA não confirmou a presença de uma relação positiva entre os gastos com TI e a eficiência do lucro no setor bancário indiano tendo como base os dados coletados dos relatórios anuais de dez bancos nacionalizados e três bancos do setor privado no período de 2006 a 2013 Zuo et al 2021 utilizandose da técnica nãoparamétrica de análise envoltória de dados DEA analisou o impacto da transformação digital em 20 bancos comerciais chineses de 2011 a 2019 tendo verificado que os investimentos em tecnologia contribuem para uma 14 melhoria substancial na eficiência produtiva dos bancos embora tenha constatado heterogeneidade entre os bancos Outro estudo de Farouk e Dandago 2015 investigou o impacto do investimento em TI no desempenho financeiro dentre 24 bancos nigerianos no período de 2006 a 2010 por meio do modelo de regressão de dados em painel tendo verificado que deter elevado nível de investimento em TI não aumenta a lucratividade dos bancos Os estudos em âmbito internacional indicam que a relação entre investimentos em TI e a eficiência dos bancos depende de vários fatores tais como o tamanho do banco composição do seu capital social e o nível de concorrência no mercado No âmbito dos estudos já realizados no setor bancário brasileiro Becker Lunardi e Maçada 2003 utilizando DEA para medir a eficiência técnica de 74 bancos brasileiros no ano de 2000 concluíram que os bancos que mais investem em TI apresentam melhores índices de eficiência Ruiz et al 2008 analisando a eficiência técnica das instituições financeiras no Brasil entre 1995 e 2005 por meio de fronteira estocástica de custo nos mostraram que o aumento da taxa de juros da economia representado pela taxa média do CBond implica no aumento de custos para as instituições financeiras e resulta em uma perda de eficiência Bombonatti Filho et al 2016 com base na análise qualitativa de dados dos cinco principais bancos brasileiros do período de 2009 a 2014 verificaram que os bancos que mais investem em tecnologia são os que mais crescem em termos de receitas e lucro líquido apesar das turbulências econômicas no Brasil e no mundo Mendonça et al 2018 utilizaram SFA para avaliar a eficiência de 47 Instituições financeiras no Brasil no período de 2008 a 2015 adotando a forma funcional translog para função de custo tendo identificado a existência de relação positiva entre a eficiência na gestão de custos e a rentabilidade das instituições financeiras Freitas e Kirch 2019 a partir de um modelo de regressão linear verificaram que o investimento em tecnologia da informação pelos bancos possui uma relação 15 significativamente positiva com as variáveis de rentabilidade tendo por base 19 bancos brasileiros listados em bolsa no período 2011 a 2017 Barreto et al 2021 por meio de análise de correlação e regressão de dados em painel identificaram uma relação positiva entre os investimentos em TI e o Retorno sobre os Ativos ROA nos cinco maiores bancos brasileiros no período de 2009 a 2018 Este estudo pretende contribuir com a avaliação da relação entre gastos com TI e a lucratividade no setor bancário brasileiro adicionando uma nova dimensão à literatura existente e fornecendo subsídios para futuras pesquisas nesta área como se destaca mais especificamente abaixo Este trabalho difere do realizado por Becker Lunardi e Maçada 2003 por adotar a metodologia SFA e considerar dados em painel para capturar a evolução de eficiência ao longo do período de 2012 a 2022 Diferente da abordagem de Bombonatti Filho 2016 esta pesquisa irá analisar a relação entre os gastos de TI e o índice de eficiência das instituições financeiras utilizando a abordagem de fronteira estocástica Adicionalmente à análise de Freitas e Kirch 2019 e de Barreto 2021 este trabalho irá verificar a relação do fluxo de gastos com tecnologia da informação sobre a lucratividade dos bancos além de também analisar o impacto do saldo estoque dos investimentos em TI Em complemento aos achados de Ruiz 2008 este trabalho irá verificar a relação entre a taxa real de juros da economia representada pela diferença entre a taxa de juros da economia taxa SELIC e o índice de inflação do período INPC e a eficiência das instituições financeiras 16 3 METODOLOGIA 31 Estimação da Eficiência As técnicas paramétricas e não paramétricas de medição da eficiência são amplamente utilizadas para avaliar a eficiência de unidades de decisão em diversas áreas como finanças saúde e educação A Análise Envoltória de Dados DEA e a Análise de Fronteira Estocástica SFA são duas técnicas populares de medição da eficiência que utilizam abordagens diferentes para avaliar a eficiência de unidades de decisão A DEA foi proposta por Charnes et al 1978 e é uma técnica não paramétrica que avalia a eficiência relativa de unidades de decisão que possuem múltiplas entradas e saídas A SFA foi proposta por Aigner et al 1977 e é uma técnica paramétrica que utiliza a especificação de uma função de produção para avaliar a eficiência de unidades de decisão Tanto a DEA quanto a SFA têm vantagens e limitações e a escolha da técnica mais adequada depende das características dos dados e dos objetivos da análise Schmidt e Sickles 1984 propuseram a SFA como uma alternativa à DEA para lidar com dados de produção e discutiram as principais diferenças entre as duas técnicas Coelli et al 2005 discutiram as principais vantagens e limitações da DEA e SFA e apresentaram estudos comparativos entre elas A DEA se mostra mais adequada para dados em que não há conhecimento prévio da distribuição e pode lidar com várias entradas e saídas A SFA por outro lado requer a especificação de uma função de produção e pressupõe uma distribuição probabilística específica para os erros mas pode ser mais eficiente na identificação de unidades ineficientes Como forma de avaliar a eficiência dos bancos brasileiros ao longo do tempo será utilizado neste trabalho o método paramétrico de estimação de eficiência de Fronteira Estocástica SFA tendo em vista este decompor a ineficiência em duas partes uma decorrente de efeitos aleatórios eou erros de mensuração e a outra representando a ineficiência resultante de fatores que estão sob influência da gestão 17 Modelo de Fronteira Estocástica SFA y f x vit uit 1 Relação de produção entre insumos x vetor e um produto y com vit efeitos aleatórios iid N 0 σv2 uit ineficiência produtiva iid NT mit σu2 Será estimada uma fronteira de produção que permite apenas um output baseado na forma funcional Translog e na distribuição de probabilidades normal truncada sendo considerados os efeitos variáveis no tempo de forma a capturar a evolução tecnológica do setor bancário no período em análise Tais definições são fundamentadas pelos testes de especificação realizados para definição do modelo mais adequado aos dados conforme se verá adiante na seção de resultados Também serão incluídas variáveis explicativas da ineficiência com objetivo de evidenciar a relação entre as variáveis de interesse e a eficiência das instituições 31 Estimação da Eficiência Na estimação da função de produção a especificação do output dos inputs e das variáveis explicativas será Função de produção Output Lucro Líquido LL Input 1 Ativo Total AT Input 2 Intermediação Financeira Despesa sobre Receita IF Input 3 1 Despesas Totais DT Variáveis explicativas da ineficiência Gastos com Tecnologia da Informação GTI 18 Saldo dos Investimentos em Tecnologia da Informação SITI Taxa Real de Juros da Economia JUROS Despesas Administrativas DA Despesas de Pessoal DP Número de Agências AG Número de Pontos de Atendimento PA Assim o modelo de estimação na forma funcional translog pode ser expresso por ln 𝐿𝐿𝑖𝑡 𝛽0 𝛽t t 𝛽AT ln 𝐴𝑇𝑖𝑡 𝛽IF ln 𝐼𝐹𝑖𝑡 𝛽DT ln 𝐷𝑇𝑖𝑡 1 2 𝛽tt 𝑡2 1 2 𝛽AT AT 𝑙𝑛 𝐴𝑇𝑖𝑡2 1 2 𝛽IF IF 𝑙𝑛 𝐼𝐹𝑖𝑡2 1 2 𝛽DT DT 𝑙𝑛 𝐷𝑇𝑖𝑡2 1 2 𝛽AT IF 𝑙𝑛 𝐴𝑇𝑖𝑡𝑙𝑛 𝐼𝐹𝑖𝑡 1 2 𝛽AT DT 𝑙𝑛 𝐴𝑇𝑖𝑡𝑙𝑛 𝐷𝑇𝑖𝑡 1 2 𝛽IF DT 𝑙𝑛 𝐼𝐹𝑖𝑡𝑙𝑛 𝐷𝑇𝑖𝑡 𝛽AT t 𝑙𝑛 𝐴𝑇𝑖𝑡 𝑡 𝛽IF t 𝑙𝑛 𝐼𝐹𝑖𝑡 𝑡 𝛽DT t 𝑙𝑛 𝐷𝑇𝑖𝑡 𝑡 𝑣it 𝑢it 2 Onde uit δ0 δ1 GTIit δ2 SITIit δ3 JUROSit δ4 DAit δ5 DPit δ6 AGit δ7 PAit 3 O principal indicador de resultado dos bancos considerado pelo mercado é o lucro líquido sendo este definido como output da função de produção Cabe destacar que os bancos se caracterizam como uma indústria financeira de serviços não produzindo unidades físicas e sim se baseando em unidades monetárias Os bancos fundamentam sua atividade principal pela intermediação financeira cujo valor agregado está representado pelo Ativo Total o que abrange dentre outros o saldo das operações de crédito títulos e valores mobiliários e aplicações interfinanceiras A eficiência se fundamenta na busca por melhorar a relação entre resultados e custos buscando maximizar resultados e minimizar custos Nesse sentido analisar a relação entre as receitas e despesas com a intermediação financeira é importante para avaliar a capacidade dos bancos em gerar resultado As despesas totais foram calculadas pela soma das despesas administrativas e as despesas de pessoal sendo definidas como inputs do modelo por estarem fortemente 19 relacionadas aos gastos necessários para suporte às operações dos bancos e ao funcionamento da infraestrutura de atendimento Os Gastos com Tecnologia da Informação GTI foram considerados como sendo a soma das despesas com Processamento de Dados despesas com Comunicações de dados despesas com a amortização de softwares adquiridos e desenvolvidos internamente ativo intangível e despesas com a depreciação de sistemas de processamento de dados ativo tangível Importante destacar que essas variáveis representam fluxos correntes de despesas com TI Por outro lado o saldo dos Investimentos em Tecnologia da Informação SITI será representado pelo saldo estoque do ativo intangível correspondente à aquisição ou desenvolvimento de softwares e pelo saldo estoque do ativo imobilizado de uso referente aos sistemas de processamentos de dados Cabe destacar que não existe uma nomenclatura padrão na divulgação dos investimentos em TI pelos bancos brasileiros Alves et al 2018 sendo que as definições acima permitem abranger de forma mais assertiva os gastos e investimentos com tecnologia da informação nas empresas sejam eles investimentos diretos CAPEX despesas de custeio OPEX ou mesmo desenvolvimentos internos ou contratados de softwares Este critério também contempla a recente mudança do perfil de gastos em tecnologia dos bancos com a migração das tecnologias que sustentam os negócios para serviços XaaS Everything as a Service a exemplo de infraestrutura software plataformas digitais etc Grande parte do custo da indústria financeira que é baseada na prestação de serviços reside na estrutura física de atendimento aos clientes representados pelo número de agências bancárias e de postos de atendimento sendo também selecionadas como variáveis explicativas A taxa real de juros será a taxa nominal de juros ajustada pela inflação calculada pela fórmula taxa real de juros 1 taxa nominal de juros 1 taxa de inflação 1 A taxa nominal de juros considerada será a SELIC que representa a taxa básica de juros da economia 20 e o índice de inflação a ser considerado será o Índice Nacional de Preços ao Consumidor Amplo IPCAIBGE Todas as variáveis monetárias são mensuradas em moeda nacional R obtidas com valores nominais e ajustadas para considerar os efeitos da inflação pelo IPCA trazendoas a valor presente na data base de dezembro2022 último trimestre em avaliação O escopo do estudo será limitado à análise dos cinco maiores bancos do Brasil representados pelo Banco do Brasil BB Bradesco Caixa Econômica Federal Caixa Itaú e Santander que representam 76 dos Ativos Totais das instituições classificadas na categoria B1 do Banco Central do Brasil BCB atualmente composta por 99 empresas do tipo Banco Comercial Banco Múltiplo com Carteira Comercial ou Caixas Econômicas A restrição aos cinco principais bancos se justifica na falta de padronização das informações dos gastos e investimentos em TI nos bancos brasileiros cuja estruturação de uma base de dados com alto nível de precisão e completude exige a análise detalhada das demonstrações financeiras das instituições e extração das informações contidas em suas Notas Explicativas A alternativa à análise das demonstrações financeiras seria acessar de forma padronizada o saldo das contas contábeis dos bancos junto ao Banco Central do Brasil BCB no Portal do Cosif Plano Contábil das Instituições do Sistema Financeiro Nacional Porém os saldos dessas contas não correspondem exclusivamente aos investimentos e gastos com TI mas sim aos valores por categoria do ativo ou do passivo sendo necessário portanto assumir valores por meio do estabelecimento de proxy o que poderia prejudicar as estimações do modelo Assim serão utilizados dados trimestrais dos cinco maiores bancos no período de 44 trimestres ou 11 anos de março2012 a dezembro2022 obtidos por meio do IFdata Base de dados selecionados de instituições financeiras brasileiras mantida pelo Banco Central do Brasil BCB Os demais dados financeiros relacionados aos gastos e investimentos de TI serão extraídos das demonstrações contábeis trimestrais dos bancos disponíveis nos respectivos 21 sites de Relações com Investidores RI o que implica na análise de 220 demonstrações financeiras O período selecionado está alinhado ao tempo máximo de um ciclo econômico definido por Juglar que possui duração de 7 a 11 anos Korotayev e Tsirel 2010 abrangendo os movimentos de expansão e contração da economia embora os ciclos econômicos possam ser afetados por fatores como mudanças nas políticas econômicas eventos geopolíticos e inovações tecnológicas conforme Burns 1934 Além disso o período de 11 anos corresponde a duas vezes o tempo de evolução das tecnologias digitais estabelecido por Brynjolfsson e Mcafee 2011 que ponderam que o prazo de 5 a 6 anos é suficiente para que uma nova tecnologia seja desenvolvida testada aprimorada e adotada em larga escala pela sociedade A tabela 1 a seguir apresenta a análise descritiva das variáveis estabelecidas no modelo de estimação de eficiência de Fronteira Estocástica SFA abrangendo o output e os inputs da função de produção bem como as variáveis explicativas da ineficiência cujos valores foram apurados com base trimestral e no período de 2012 a 2022 Tabela 1 Análise descritiva das variáveis Variáveis Média Mediana Desvio padrão Máximo Mínimo Lucro Líquido LL a 4769 4760 2258 13069 1839 Ativo Total AT a 1566337 1632403 388689 2359739 796640 Intermediação Financeira Despesa sobre Receita IF c 0791 0747 0222 2173 0410 Despesas Totais DT a 10402 10929 2512 15468 4415 Gastos com TI GTI a 1346 1178 510 2360 370 Saldo dos Investimentos em TI SITI a 5875 4727 2778 12742 1495 Taxa Real de Juros JUROS d 2686 3131 2827 7524 3784 Despesas Administrativas DA a 5080 5438 1550 8333 2012 Despesas de Pessoal DP a 5321 5654 1766 9198 1736 Número de Agências AG b 3746 3408 913 5556 2372 Número de Pontos de Atendimento PA b 4891 4519 1042 7048 2976 22 Dados com base em 220 observações a Em R Milhão em valores deflacionados trazidos à data base 122022 b Em unidades de pontos c Razão entre as Despesas de Intermediação Financeira e as Receitas de Intermediação Financeira d Em aa A visão da evolução temporal das variáveis explicativas da ineficiência a serem consideradas GTI Gastos com TI e SITI Saldo dos Investimentos em TI permite observar a variação em valores deflacionados do período em análise trazidos à data base 122022 conforme demonstrado nas Figuras 4 e 5 a seguir cuja relação com a eficiência das empresas se pretende verificar neste estudo Figura 4 Comparativo dos gastos com TI entre os bancos 20122022 Fonte Elaborado pelo autor com base nas demonstrações financeiras dos bancos Consideradas as despesas com Processamento de Dados Comunicações de Dados Amortização de Softwares e Depreciação de Sistemas de Processamento de Dados 2228 6551 3525 7292 4258 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 R MILHÃO BB Bradesco Caixa Itaú Santander 23 Figura 5 Comparativo do saldo de investimentos em TI entre os bancos 20122022 Fonte Elaborado pelo autor com base nas demonstrações financeiras dos bancos Considerados os saldos com Softwares Adquiridos e Desenvolvidos Internamente Ativo Intangível e com Sistema de Processamento de Dados Ativo Tangível 4 RESULTADOS 41 Testes de Especificação Com o objetivo de determinar a especificação ideal do modelo a ser analisado foi realizado teste de razão de verossimilhança unicaudal Teste LR para determinar se o modelo ajustado com base na forma funcional Translog supera o modelo CobbDouglas O teste LR rejeitou a hipótese nula referente à versão restrita do modelo Cobb Douglas em apoio à forma funcional Translog Os valores resultantes dos testes foram maiores do que os valores listados na Tabela de Kodde e Palm2 para os respectivos graus de 2 Os valores críticos para o Teste de Razão de Verossimilhança encontramse na Tabela 1 de David A Kodde e Franz C Palm 1986 sendo justificado o uso da tabela uma vez que a estatística apresenta uma distribuição qui quadrada mista 5493 10473 2193 12742 5055 0 2000 4000 6000 8000 10000 12000 14000 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 R MILHÃO BB Bradesco Caixa Itaú Santander 24 liberdade indicando que o modelo Translog onde todos os parâmetros foram considerados superou o modelo restrito conforme dados dispostos na tabela 2 Tabela 2 Teste LR de Especificação para Fronteira Estocástica Teste Unrestricted log likelihood function ULLF Restricted log likelihood function RLLF λ a X2 gl b Valores Críticos c Resultado Translog 57187 62044 9713 5 8574 Aceitase Translog a λ 2ULLF RLLF diferença entre o modelo completo de Translog e o modelo restrito Ho b Graus de Liberdade df c Valores Críticos constantes da Tabela de Kodde e Palm para 10 de significância Também foi realizado o teste de Hausman 1978 para comparar os efeitos fixos e efeitos aleatórios do modelo de fronteira estocástica e avaliar qual modelo é mais apropriado para os dados O teste de Hausman apresentou valorp de 00463 inferior ao nível de significância de 5 rejeitando a hipótese nula onde o modelo de efeitos fixos é mais apropriado e sugerindo que os efeitos aleatórios são consistentes e que não há correlação entre as variáveis explicativas e os efeitos não observados Quanto à definição do modelo de distribuição de probabilidades foram realizadas as estimações considerando os modelos Semi Normal e Normal Truncada A distribuição Semi Normal assume que os erros da fronteira estocástica são simétricos e permite que tenha uma maior variabilidade em relação à média Por outro lado a distribuição Normal Truncada assume que os erros são normalmente distribuídos mas com truncamento em uma determinada região Essa distribuição é mais restritiva pois impõe a suposição de normalidade nos erros o que pode ser apropriado em algumas situações em que a variabilidade dos erros é relativamente pequena e não há evidências de assimetria ou outliers 25 As estimações realizadas nos modelos de distribuição não apresentaram variações nos valores de loglikelihoods e resultaram nos mesmos valores no Critério de Informação de Akaike AIC e no Critério Bayesiano de Schwarz BIC indicando que ambos os modelos têm um ajuste similar em relação à complexidade Assim optouse pela definição da distribuição Normal Truncada por não haver evidências de assimetria ou de outliers na análise dos resíduos assumido que os erros na estimação da fronteira estocástica seguem uma distribuição normal e não apresentam variabilidade excessiva O modelo proposto possui 3 inputs na função de produção e 7 variáveis explicativas para uma base com 220 observações número amplo em relação ao número de preditores o que contribui para a sua robustez O valorp associado ao teste F resultou no valor 42e13 bem abaixo do nível de significância prédeterminado de 005 sugerindo que modelo de regressão como um todo é estatisticamente significativo e tem poder preditivo e explicativo sobre a variável dependente 42 Resultado da Estimação A estimação dos escores de eficiência dos principais bancos brasileiros resultou em valores de eficiência média calculada pela média aritmética das eficiências dispostas na Figura 6 onde verificamos leve tendência de elevação ao longo do período avaliado em que pese a ocorrência de oscilações em períodos específicos como o observado no último trimestre2022 que se explica pelo aumento na provisão para créditos de difícil liquidação eou efetivação da contabilização de perdas como o ocorrido no caso das Lojas Americanas 26 Figura 6 Eficiência média dos bancos 20122022 A evolução da eficiência de cada banco está representada na Figura 7 sendo que os índices devem assumir valores entre 0 e 1 e quanto mais próximo de 1 mais eficiente será o banco em sua capacidade de gerar lucro Figura 7 Evolução dos índices de eficiência dos bancos 515 25 35 45 55 65 75 85 95 1T12 2T12 3T12 4T12 1T13 2T13 3T13 4T13 1T14 2T14 3T14 4T14 1T15 2T15 3T15 4T15 1T16 2T16 3T16 4T16 1T17 2T17 3T17 4T17 1T18 2T18 3T18 4T18 1T19 2T19 3T19 4T19 1T20 2T20 3T20 4T20 1T21 2T21 3T21 4T21 1T22 2T22 3T22 4T22 ESCORES DE EFICIÊNCIA ÚLTIMOS 44 TRIMESTRES EFFMedia Linear EFFMedia 00 01 02 03 04 05 06 07 08 09 10 1T12 2T12 3T12 4T12 1T13 2T13 3T13 4T13 1T14 2T14 3T14 4T14 1T15 2T15 3T15 4T15 1T16 2T16 3T16 4T16 1T17 2T17 3T17 4T17 1T18 2T18 3T18 4T18 1T19 2T19 3T19 4T19 1T20 2T20 3T20 4T20 1T21 2T21 3T21 4T21 1T22 2T22 3T22 4T22 ESCORE DE EFICIÊNCIA BB Bradesco Caixa Itaú Santander 27 A tabela 3 apresenta o resultado da estimação com os coeficientes das variáveis de entrada do modelo inputs e das variáveis explicativas da ineficiência dispostas na primeira coluna Tabela 3 Resultado da estimação Coeficiente Estimativa Desvio Padrão valor z Prz Intercepto 𝛽0 32 e05 84 e04 38 00002 logAtivo Total 𝛽AT 58 e04 12 e04 50 00000 logIntermediação Financeira 𝛽IF 90 e03 56 e03 16 01068 log1Despesa Total 𝛽DT 20 e04 60 e03 33 00008 logAtivo Total2 𝛽AT AT 48 e03 99 e02 49 00000 logAtivo Total x logIntermediação Financeira 𝛽AT IF 10 e03 53 e02 20 00492 logAtivo Total x log1Despesa Total 𝛽AT DT 12 e03 74 e02 17 00975 logIntermediação Financeira2 𝛽IF IF 55 e02 37 e02 15 01349 logIntermediação Financeira x log1Despesa Total 𝛽IF DT 60 e02 38 e02 16 01102 log1Despesa Total2 𝛽DT DT 29 e02 97 e02 03 07643 ZGastos com TI GTI δ1 13 e00 45 e01 28 00046 ZSaldo dos Investimentos em TI SITI δ2 92 e02 57 e02 16 01077 ZTaxa Real de Juros JUROS δ3 96 e01 29 e01 33 00010 ZDespesas Administrativas DA δ4 12 e01 13 e01 09 03613 ZDespesas de Pessoal DP δ5 45 e01 84 e02 53 00000 ZNúmero de Agências AG δ6 26 e00 59 e01 44 00000 ZNúmero de Pontos de Atendimento PA δ7 17 e00 41 e01 42 00000 sigmaSq σ2 29 e02 62 e01 47 00000 gamma ƴ 91 e02 33 e01 278 2e16 não significativo significativo a 10 significativo a 5 significativo a 1 significativo a 01 Observamos que no resultado da função de produção se destacou a contribuição do Ativo Total sobre o índice de eficiência resultando no valor positivo para o coeficiente 𝛽AT e com alta significância estatística 01 o que demonstra que a capacidade de geração de resultados nos bancos está positivamente relacionada ao volume de ativos dessas 28 instituições cujo montante desses cinco maiores bancos brasileiros totalizou em dezembro2022 o valor de R 848 trilhões Verificamos que controlar as despesas totais eleva o lucro dos bancos de forma estatisticamente significativa corroborando os achados de Mendonça et al 2018 que pondera que quanto mais eficiente é a gestão de custos da empresa maior será sua rentabilidade A estimativa com relação à razão do volume das despesas sobre as receitas com intermediação financeira não apresentou significância estatística sugerindo que os bancos ponderam a realização de suas operações às condições de vantajosidade do negócio e de exposição ao risco refletindo as condições econômicas do momento cuja estratégia de atuação pode implicar na variação do índice de relação entre as despesas e receitas com intermediação financeira decorrente por exemplo de ações de reforço de provisões para crédito de difícil liquidação restrição à concessão de crédito ponderada pelo risco ou da manutenção do patamar de captação de recursos Em consonância ao trabalho de Becker Lunardi e Maçada 2003 verificamos que os gastos com Tecnologia da Informação contribuem de forma relevante para a eficiência dos bancos brasileiros demonstrando a necessidade de alocação de recursos em TI para a elevação do nível de tecnologia e aumento de seu patamar de eficiência Cabe ressaltar que os gastos com Tecnologia da Informação definidos no modelo compreendem tanto as despesas administrativas dispendidas com TI quanto também as despesas com a depreciação e amortização dos investimentos realizados com TI abrangendo dessa forma todas as iniciativas relacionadas aos gastos com tecnologia da informação Nesse sentido observase que o saldo de investimentos de TI não apresentou relevância estatística em relação ao nível de eficiência dos bancos Isso pode ser explicado em razão de que o saldo dos investimentos em TI representa um estoque dos valores dispendidos em Tecnologia da Informação relacionados especificamente à aquisição de softwares desenvolvimento interno de softwares ou compra de sistemas de processamento 29 de dados cujo período prazo de vida útil pode ter duração de até 120 meses dependendo da natureza do investimento Dada a rápida evolução tecnológica experimentada atualmente é possível que este estoque não mais represente adequadamente os impactos da tecnologia mais atual nos lucros correntes o que não acontece com a variável GTI fluxo Tal observação reforça a tendência de mudança do perfil de gastos em tecnologia dos bancos onde verificamos uma migração do modelo proprietário de negócios de TI para o modelo de serviços XaaS Everything as a Service onde as empresas deixam de adquirir os produtos e passam assinar os serviços Observamos que a elevação na Taxa Real de Juros da economia contribui para a ineficiência dos Bancos analisados em consonância aos achados de Ruiz et al 2008 que considerou como indicador macroprudencial a taxa média do CBond Verificamos ainda que reduzir despesas de pessoal contribuem para a elevação da eficiência dos bancos sugerindo que a evolução dos serviços bancários está vinculada à evolução tecnológica do setor que possui como característica ser menos dependente e até mesmo poupadora de mão de obra implicando que funcionários antes alocados em atividades que passaram a ser abrangidas por novas tecnologias são realocados para lidar com outras atividades complexas de negócios e de atendimento das necessidades dos clientes No entanto observamos não haver significância estatística na variação das despesas administrativas excetuadas as despesas de pessoal o que pode ser explicado pela necessidade da empresa em custear não apenas os gastos com TI mas todas as despesas não operacionais necessárias para dar suporte aos negócios dos bancos e ao funcionamento da infraestrutura de atendimento Também constatamos que ter uma maior quantidade de Pontos de Atendimento implica na perda de lucratividade para as instituições o que pode ser explicado pelos custos associados ao funcionamento da infraestrutura física de atendimento desses locais sendo considerado no total desses pontos a soma da quantidade de Agências bancárias e dos Postos 30 de Atendimento estes últimos de menor tamanho e de atendimento restrito seja em público ou em serviços especializados Por outro lado observamos que uma maior quantidade de agências bancárias caracterizadas como pontos de atendimento especializados favorece o lucro das instituições demonstrando a necessidade de se manter uma adequada e ampla rede de atendimento para o atendimento qualificado e segmentado aos clientes 5 CONCLUSÃO O objetivo deste trabalho foi analisar o impacto dos gastos com tecnologia da informação nos índices de eficiência do setor bancário brasileiro por meio da estimação de eficiência usando a metodologia de fronteira estocástica SFA Os resultados permitiram observar correlação positiva entre os fluxos dos gastos com tecnologia da informação e a melhoria dos índices de eficiência dos principais bancos brasileiros no período de 2012 a 2022 tendo por base os dados das demonstrações financeiras trimestrais das instituições Por outro lado o saldo de investimentos de TI estoque não apresentou relevância estatística em relação ao nível de eficiência dos bancos Isso pode ser interpretado como um indicativo de mudança do perfil de gastos em tecnologia dos bancos onde há migração do modelo proprietário de negócios de TI para o modelo de serviços onde as empresas deixam de adquirir os produtos e passam a contratar os serviços Nesse caso a rápida evolução tecnológica experimentada faz com que o estoque acumulado possivelmente defasado não mais represente adequadamente os impactos da tecnologia mais atual nos níveis correntes de lucratividade Não obstante os gastos com TI comporem as despesas administrativas das Instituições Financeiras verificouse que quanto mais controlados são os gastos com pessoal maior será o índice de eficiência média dos bancos demonstrando a necessidade permanente da administração em reduzir despesas ou alocar recursos da forma mais eficiente possível 31 Apesar da transformação digital dos Bancos nos últimos anos e a migração do atendimento para canais digitais a exemplo do Internet Banking e aplicativos em dispositivos móveis APPs observamos que ter uma maior quantidade de pontos de atendimento especializados agências bancárias para atendimento aos clientes e comercialização de produtos e serviços financeiros favorece o lucro dos bancos em que pese uma maior quantidade total de pontos de atendimento agências bancárias somadas aos postos de atendimento ser um fator que contribui para a ineficiência das instituições Constatamos ainda que a elevação na taxa real de juros da economia contribui para a ineficiência dos bancos possivelmente decorrente do aumento de custos para as instituições refletindo questões relacionadas à concorrência restrições e alterações das condições impostas tanto na ponta da captação quanto na da aplicação de recursos financeiros Ainda que este trabalho tenha sido baseado nos cinco maiores bancos brasileiros que representam 76 dos Ativos Totais das instituições classificadas na categoria B1 do Banco Central do Brasil a análise da relação entre a eficiência dos bancos e os seus gastos com TI poderia ser ampliada para todas as 99 empresas da referida categoria de forma a avaliar o impacto nos diversos tipos de segmentos foco de atuação e tamanho das instituições financeiras Também poderiam ser analisadas possíveis alterações no custo da tecnologia ao longo desse período e sua relação comparativa quanto ao volume de investimentos em TI realizados A relação entre o spread bancário e a elevação da eficiência dos bancos fundamentada pelos investimentos em TI é outro ponto possível de ser explorado Esperase por fim que este trabalho traga subsídios para a atuação dos gestores de bancos na tomada de decisão sobre gastos e investimentos em TI promovendo a melhoria de produtos e serviços bancários o aumento da eficiência das instituições financeiras e favorecendo o desenvolvimento econômico do país 32 6 REFERÊNCIAS AIGNER D LOVELL C K SCHMIDT P Formulation and estimation of stochastic frontier production function models Journal of econometrics v 6 n 1 p 2137 1977 ALVES A T A R B A GALEGALE N V SANTOS F A Inversiones en ti los bancos brasileños revelan estas inversiones Revista La Junta v 1 n 1 2018 BARRETO L S PEREIRA V S PENEDO A S T Impacto dos investimentos em tecnologia sobre a rentabilidade do setor bancário brasileiro Future Studies Research Journal Trends and Strategies v 13 n 1 p 94111 2021 BECKER J LUNARDI G MAÇADA A Análise de eficiência dos bancos brasileiros um enfoque nos investimentos realizados em Tecnologia de Informação TI Revista Produção 13 2 p 7081 2003 BOMBONATTI FILHO O GASPAR M A CARDOSO M V COSTA I Investimentos em TI e o acesso da população brasileira aos serviços e facilidades bancárias Revista Espacios v 37 n 26 p 23 2016 BRYNJOLFSSON E MCAFEE A Race against the machine How the digital revolution is accelerating innovation driving productivity and irreversibly transforming employment and the economy 2011 BURNS A F Production trends in the United States since 1870 National Bureau of Economic Research 1934 CHARNES A COOPER W W RHODES E Measuring the efficiency of decision making units European journal of operational research v 2 n 6 p 429444 1978 COELLI T J RAO D S P ODONNELL C J BATTESE G E An introduction to efficiency and productivity analysis Springer science business media 2005 33 FAROUK B K U DANDAGO K I Impact of investment in information technology on financial performance of Nigerian banks Is there a productivity paradox Journal of Internet Banking and Commerce v 20 n 1 p 122 2015 FARREL M J A measurement of productive efficiency Journal of The Royal Statistical Society v120 p254290 1957 Febraban Federação Brasileira de Bancos Pesquisa FEBRABAN de Tecnologia Bancária 2023 2023 FENG Z WU Z Technology investment firm performance and market value evidence from Banks 2018 FREITAS O D KIRCH G Performance dos bancos brasileiros no contexto de digitalização Revista Brasileira de Finanças v 17 n 2 p 3855 2019 GUPTA S D RAYCHAUDHURI A HALDAR S K Information technology and profitability evidence from Indian banking sector International Journal of Emerging Markets 2018 HAUSMAN J A Specification tests in econometrics Econometrica Journal of the econometric society p 12511271 1978 JORGENSON D W GRILICHES Z The explanation of productivity change The review of economic studies v 34 n 3 p 249283 1967 KENDRICK J W Productivity trends in the United States Productivity trends in the United States 1961 KODDE D PALM F Table 1 Upper and lower bounds for the critical value for jointly testing equality and inequality restrictions 1986 34 KOROTAYEV A V TSIREL S V A spectral analysis of world GDP dynamics Kondratieff waves Kuznets swings Juglar and Kitchin cycles in global economic development and the 20082009 economic crisis Structure and Dynamics v 4 n 1 2010 MANKIW N G ROMER D WEIL D N A contribution to the empirics of economic growth The quarterly journal of economics v 107 n 2 p 407437 1992 MENDONÇA D J SOUZA J A E CARVALHO F D M DE BENEDICTO G C Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil Revista Capital Científico v 16 n 3 2018 RUIZ C TABAK B M CAJUEIRO D O Mensuração da eficiência bancária no Brasil a inclusão de indicadores macroprudenciais Revista Brasileira de Finanças v 6 n 3 art 136 p 411436 2008 SCHMIDT P SICKLES R C Production frontiers and panel data Journal of Business Economic Statistics v 2 n 4 p 367374 1984 SOLOW R M A contribution to the theory of economic growth The quarterly journal of economics v 70 n 1 p 6594 1956 SOLOW R M Technical Change and the Aggregate Production Function Review of Economics and Statistics 39 3 August 1957 pp 312320 Real Business Cycles p 543551 2013 ZUO L STRAUSS J ZUO L The digitalization transformation of commercial banks and its impact on sustainable efficiency improvements through investment in science and technology Sustainability v 13 n 19 p 11028 2021 35 FUNDAÇÃO GETULIO VARGAS ESCOLA DE ECONOMIA DE SÃO PAULO LUIS FELIPE GAZQUEZ RUA IMPACTO DE FATORES MACROECONÔMICOS E DA SOLVÊNCIA SOBRE A EFICIÊNCIA BANCÁRIA DOS PRINCIPAIS BANCOS BRASILEIROS SÃO PAULO 2020 LUIS FELIPE GAZQUEZ RUA IMPACTO DE FATORES MACROECONÔMICOS E DA SOLVÊNCIA SOBRE A EFICIÊNCIA BANCÁRIA DOS PRINCIPAIS BANCOS BRASILEIROS Dissertação apresentada à Escola de Economia da Fundação Getulio Vargas como requisito para obtenção do título de Mestre em Economia Campo de Conhecimento Economia bancária Orientador Prof Dr Jéfferson A Colombo SÃO PAULO 2020 Rua Luis Felipe Gazquez Impacto de fatores macroeconômicos e da solvência sobre a eficiência bancária dos principais bancos brasileiros Luis Felipe Gazquez Rua 2020 112 f Orientador Jéfferson Augusto Colombo Dissertação mestrado profissional MPFE Fundação Getulio Vargas Escola de Economia de São Paulo 1 Bancos Brasil 2 Eficiência organizacional 3 Análise estocástica 4 Análise envoltória de dados I Colombo Jéfferson Augusto II Dissertação mestrado profissional MPFE Escola de Economia de São Paulo III Fundação Getulio Vargas IV Título CDU 3367181 Ficha Catalográfica elaborada por Raphael Figueiredo Xavier CRB SP009987O Biblioteca Karl A Boedecker da Fundação Getulio Vargas SP LUIS FELIPE GAZQUEZ RUA IMPACTO DE FATORES MACROECONÔMICOS E DA SOLVÊNCIA SOBRE A EFICIÊNCIA BANCÁRIA DOS PRINCIPAIS BANCOS BRASILEIROS Dissertação apresentada à Escola de Economia da Fundação Getulio Vargas como requisito para obtenção do título de Mestre em Economia Campo de Conhecimento Economia bancária Data de Aprovação 27102020 Banca Examinadora Prof Dr Jéfferson A Colombo orientador FGVEESP Prof Dr Emerson Fernandes Marçal FGVEESP Prof Dr Peter Fernandes Wanke UFRJCOPPEAD Dedico esse trabalho aos meus alicerces Aos meus pais pelo que sou A minha esposa que me faz melhor a cada dia Ao meu filho que nasceu durante o mestrado e fez essa jornada valer ainda mais AGRADECIMENTOS Quero agradecer a minha esposa Jéssica pelo apoio constante e incondicional as palavras de incentivo e a perseverança em acreditar em mim Eu não teria conseguido sem você Obrigado por ter segurado a barra sozinha enquanto eu focava nesse trabalho Ao meu filho que nasceu durante essa titulação Me perdoe pelas ocasiões em que não consegui estar mais próximo Aos meus pais e familiares que fizeram o possível e impossível para que eu tivesse a melhor educação e as melhores condições Tenho muito orgulho de vocês Obrigado pelo que me proporcionaram Obrigado pela família maravilhosa que são Aos amigos pessoais pelas ausências e aos amigos e líderes do trabalho que se dispuseram ao máximo em me ajudar Ao orientador Prof Dr Jéfferson A Colombo pela confiança suporte e prestatividade que conduziu todas as interações A Deus que me proporcionou participar desse desafio e conhecer todas essas pessoas O mundo não será salvo pelos caridosos mas pelos eficientes Roberto Campos RESUMO A literatura acerca da eficiência bancária cresceu muito nos últimos anos No Brasil por conta das transformações no cenário macro e microeconômico que tornaram o ambiente mais competitivo e dinâmico isso não foi diferente Nesse cenário esse estudo tem como objetivo estimar o impacto parcial de choques em variáveis macroeconômicas que representassem expectativas futuras medida de riscopaís atividade econômica e inflação e microeconômicas bancoespecífica que representasse a solvência na eficiência bancária das principais instituições financeiras do Brasil do primeiro trimestre de 2009 até o quarto trimestre de 2019 Os escores de eficiência são estimados através da análise de fronteira estocástica SFA e a análise de sua interrelação com as demais variáveis se dá por meio da aplicação de Modelos Autorregressivos Vetoriais VAR seguidos de suas análises estruturais Em linhas gerais os resultados contribuem para a literatura destacando que cada instituição financeira é própria e que não há uma convergência entra as amostras analisadas para o período destacado Palavraschave Eficiência bancária brasileira Análise de fronteira estocástica SFA Modelo Autorregressivo Vetorial VAR Análise Estrutural VAR ABSTRACT The literature on banking efficiency has grown a lot in recent years In Brazil due to the changes in the macro and microeconomic scenario which made the environment more competitive and dynamic this was no different In this scenario this study aims to estimate the partial impact of shocks on macroeconomic variables representing future expectations country risk measure economic activity and inflation and microeconomic variables bankspecific representing solvency on the banking efficiency of the main financial institutions in Brazil from the first quarter of 2009 to the fourth quarter of 2019 Efficiency scores are estimated through the stochastic frontier analysis SFA and the analysis of their interrelation with the other variables is done through the application of Vector Autoregression Models VAR followed by their structural analysis In general the results contribute to the literature highlighting that each financial institution is its own and that there is no convergence between the samples analyzed for the highlighted period Keywords Brazilian banking efficiency Stochastic frontier analysis SFA Vector autoregression model VAR Structural Analysis VAR LISTA DE FIGURAS Figura 1 Principais etapas da análise VAR 42 Figura 2 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Itaú 61 Figura 3 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Banco do Brasil 62 Figura 4 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Bradesco 63 Figura 5 Função Impulsoresposta das diversas variáveis ao escore de eficiência da Caixa 64 Figura 6 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Santander 65 Figura 7 Função Impulsoresposta das diversas variáveis ao escore de eficiência da média ponderada dos Privados 66 Figura 8 Função Impulsoresposta das diversas variáveis ao escore de eficiência da média ponderada dos Públicos 67 Figura 9 Função Impulsoresposta das diversas variáveis ao escore de eficiência da média ponderada dos bancos 68 Figura 10 Escores de eficiência de custos e demais variáveis no tempo 102 Figura 11 Correlograma cruzado do escore de eficiência de custos do Itaú para cada uma das variáveis do Modelo VAR 106 Figura 12 Correlograma cruzado do escore de eficiência de custos do Banco do Brasil para cada uma das variáveis do Modelo VAR 106 Figura 13 Correlograma cruzado do escore de eficiência de custos do Bradesco para cada uma das variáveis do Modelo VAR 107 Figura 14 Correlograma cruzado do escore de eficiência de custos do Caixa para cada uma das variáveis do Modelo VAR 107 Figura 15 Correlograma cruzado do escore de eficiência de custos do Santander para cada uma das variáveis do Modelo VAR 108 Figura 16 Correlograma cruzado do escore de eficiência de custos da média ponderada dos Privados para cada uma das variáveis do Modelo VAR 108 Figura 17 Correlograma cruzado do escore de eficiência de custos da média ponderada dos Públicos para cada uma das variáveis do Modelo VAR 109 Figura 18 Correlograma cruzado do escore de eficiência de custos da média ponderada Geral para cada uma das variáveis do Modelo VAR 109 Figura 19 Gráficos da Estabilidade dos Modelos VAR para todas as amostras 110 LISTA DE QUADROS Quadro 1 Síntese dos estudos considerados para a identificação das hipóteses 35 Quadro 2 Resultado dos testes DieboldMariano para os critérios MSE MAE e MAPE 76 Quadro 3 Síntese comparativa entre os sinais esperados e os resultados apurados 84 LISTA DE GRÁFICOS Gráfico 1 Abordagem de estimativas nos 100 principais artigos 30 LISTA DE TABELAS Tabela 1 Estatística descritiva das variáveis 56 Tabela 2 Critérios de seleção da ordem de defasagens 57 Tabela 3 Probabilidades do teste de causalidade de Granger para Eficiência de Custos 60 Tabela 4 Resumo das Funções Impulsoresposta das diversas variáveis ao escore de eficiência de todas as amostras 69 Tabela 5 Estimativas pontuais das decomposições de variância do erro de previsão do Itaú 70 Tabela 6 Estimativas pontuais das decomposições de variância do erro de previsão do Banco do Brasil 70 Tabela 7 Estimativas pontuais das decomposições de variância do erro de previsão do Bradesco 71 Tabela 8 Estimativas pontuais das decomposições de variância do erro de previsão da Caixa 72 Tabela 9 Estimativas pontuais das decomposições de variância do erro de previsão do Santander 73 Tabela 10 Estimativas pontuais das decomposições de variância do erro de previsão da média ponderada dos privados 73 Tabela 11 Estimativas pontuais das decomposições de variância do erro de previsão da média ponderada dos públicos 74 Tabela 12 Estimativas pontuais das decomposições de variância do erro de previsão da média ponderada dos bancos 75 Tabela 13 Efeito das Funções Impulsoresposta dos choques do INEC à eficiência de custos de todas as amostras 78 Tabela 14 Efeito das Funções Impulsoresposta dos choques do Risco Brasil à eficiência de custos de todas as amostras 79 Tabela 15 Efeito das Funções Impulsoresposta dos choques do PIB à eficiência de custos de todas as amostras 81 Tabela 16 Efeito das Funções Impulsoresposta dos choques do IGPM à eficiência de custos de todas as amostras 82 Tabela 17 Efeito das Funções Impulsoresposta dos choques de Basileia à eficiência de custos de todas as amostras 83 Tabela 18 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável INEC 86 Tabela 19 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável Risco Brasil 86 Tabela 20 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável PIB 86 Tabela 21 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável IGPM 87 Tabela 22 Estatística descritiva dos escores de eficiência de receita e lucro 88 Tabela 23 Probabilidades do teste de causalidade de Granger para Eficiência de Receita Lucro e Custos 89 Tabela 24 Matriz de correlação das variáveis dos modelos VAR 103 Tabela 25 Teste ADF para principais variáveis do estudo 104 Tabela 26 Teste LM de cada amostra 105 Tabela 27 Probabilidades do teste de causalidade de Granger para as demais variáveis 111 Tabela 28 Matriz de correlação dos escores de eficiência de custo lucro e receita 112 SUMÁRIO 1 INTRODUÇÃO 17 2 REVISÃO DA LITERATURA 21 21 Eficiência bancária definição 21 22 Técnicas de estimação de eficiência 26 23 Eficiência bancária e suas relações com outras variáveis 30 24 As hipóteses de pesquisa 34 3 METODOLOGIA 40 31 Análise de fronteira estocástica 40 32 Modelo Autorregressivo Vetorial 42 321 Especificação e estimação do modelo VAR 43 322 Checagem do Modelo 48 323 Análise Estrutural 48 3231 Causalidade de Granger 49 3232 Função ImpulsoResposta 50 3233 Análise da Decomposição da Variância dos Erros de Previsão 52 324 Previsão 52 33 Dados e amostra 54 4 RESULTADOS E DISCUSSÃO 57 41 Resultados Modelo VAR 57 42 Análises Estruturais 58 421 Análises de causalidade de Granger 58 422 Análises das Funções ImpulsoResposta 60 4221 Análise individual dos cinco grandes bancos 61 4222 Análise das médias ponderadas dos públicos privados e geral 65 423 Análises das Decomposições das Variâncias dos Erros de Previsão 69 43 Previsão 75 44 Discussão dos resultados à luz das hipóteses 77 45 Teste de Robustez 84 451 Utilização de variáveis alternativas 84 452 Variação na ordem das variáveis 85 453 Eficiência de receita e lucro 87 16 5 CONCLUSÃO 91 6 REFERÊNCIAS 93 APÊNDICE 101 17 1 INTRODUÇÃO O tema eficiência bancária cresceu muito nos últimos anos fato evidenciado pelo aumento exponencial de publicações sobre esse conteúdo Esse fator se deu por diversas razões dentre as quais destacamos os desafios de medir e analisar uma das indústrias de serviço com a estrutura de receitas mais complexa BERGER HUMPHREY 1992 a divergência metodológica acerca do melhor modo de mensuração da eficiência AIELLO BONANNO 2015 os interesses pelas consequências e impactos das crises financeiras AHMAD et al 2019 a desregulamentação e mudanças tecnológicas FIORDELISI MARQUES IBANEZ MOLYNEUX 2011 e até mesmo o alto grau de dispersão dos estudos sobre o tema evidenciando que apesar de tradicional a eficiência bancária ainda não é uma área de conhecimento bem estabelecida ABREU KIMURA SOBREIRO 2019 No Brasil um dos maiores e mais complexos sistemas bancários do mundo essa evolução não foi diferente e seu efeito é oriundo das diversas transformações como aumento da concorrência onda de concentração com fusões e aquisições surgimento de novos produtos financeiros e a revolução na tecnologia da informação NAKANE WEINTRAUB 2005 Tivemos também a recente entrada de bancos estrangeiros logo após a estabilização financeira trazida pelo Plano Real que contribuiu para esse ambiente mais competitiva e dinâmico GOMES OLIVEIRA MATIAS 2017 e uma nova realidade do cenário macroeconômico com um ambiente de inflação mais baixa que forçou os bancos a se adaptarem NAKANE WEINTRAUB 2005 Como consequência dessa imensa quantidade de estudos várias foram as análises das relações entre a eficiência bancária e outras variáveis Artigos focaram em variáveis socioculturais como a cultura do país ver por exemplo Barros et al 2018 ou macroeconômicas como inflação Athanasoglou Brissimis e Delis 2008 Primo et al 2013 AlGasaymeh 2016 Tan e Anchor 2017 e produção nacional Sufian 2009 Chen Higgins e Mason 2005 Chortareas Girardone e Ventouri 2012 Primo et al 2013 AlGasaymeh 2016 Tan e Anchor 2017 Chen et al 2018 Kallel Hamad e Triki 2019 Houveram 18 também estudos relacionandoa com variáveis do setorindústria como concentração de mercado Berger e Hannan 1998 Athanasoglou Brissimis e Delis 2008 concorrência Casu e Girardone 2009 e liberalização do sistema financeiro Aiello e Bonanno 2018 ou até mesmo variáveis bancoespecíficas como empréstimos vencidos Berger e DeYoung 1997 o tamanho e ramificação da instituição Elyasiani e Mehdian 1990 Athanasoglou Brissimis e Delis 2008 Henriques et al 2018 os riscos bancários Athanasoglou Brissimis e Delis 2008 Fiordelisi MarquesIbanez e Molyneux 2011 Tan e Anchor 2017 o nível de capital Fiordelisi MarquesIbanez e Molyneux 2011 Kallel Hamad e Triki 2019 o custo do crédito Shamshur e Weill 2019 e até a qualidade dos serviços Kallel Hamad e Triki 2019 O propósito deste estudo é estimar o impacto parcial de choques em variáveis macro e microeconômicas sobre flutuações futuras no grau de eficiência bancária nas principais instituições financeiras do Brasil Para isso são utilizadas variáveis macroeconômicas como proxy para expectativas futuras medida de riscopaís atividade econômica inflação e uma variável bancoespecífica relacionada a solvência além da própria variável de eficiência A estimação ocorre por meio de modelos de Vetores Autorregressivos VAR com posterior análise das funções de impulso resposta IRFs causalidade de Granger e decomposição da variância dos erros de previsão FEVD Embora haja uma vasta literatura sobre o tema nosso artigo traz algumas contribuições para a literatura sobre a relação entre a eficiência bancária e variáveis macroeconômicas e específicas Em primeiro lugar nossa amostra considera dados bancários trimestrais recentes do 1º trimestre de 2009 até o 4º trimestre de 2019 das cinco maiores instituições financeiras brasileiras e outras três séries de dados oriundas de uma média ponderada desses cinco bancos Em segundo lugar utilizamos a técnica paramétrica mais difundida para estimação da eficiência a análise de fronteira estocástica pois ela nos permitiria analisar tanto a eficiência de custos quanto outras eficiências como lucro e receita BERGER MESTER 1997 que poderiam evidenciar outras qualidades e características gerenciais FIORDELISI MARQUESIBANEZ MOLYNEUX 2011 Logo utilizamos a eficiência de custos como referência principal e as demais mensurações como teste de robustez Em terceiro lugar optamos por 19 um modelo de séries temporais o Modelo Autorregressivo Vetorial Vector Autoregressive VAR para analisar as interrelações entre a eficiência e as demais variáveis Essa escolha nos permitiu realizar uma análise estrutural para cada um dos bancos ou seja o kit de ferramentas padrão na análise VAR STOCK WATSON 2001 análise de causalidade de Granger as funções impulso resposta e a decomposição da variância dos erros de previsão e testar a acurácia preditiva dos modelos através do teste de DieboldMariano Surpreendentemente não encontramos outros estudos que adotaram essa metodologia com esse fim para esse tema Os resultados obtidos mostram que a expectativa futura a atividade econômica e a inflação Grangercausam a eficiência de custos para a maioria das amostras o que não se mostrou verdadeiro para o riscopaís e o indicador de solvência das instituições Além disso a eficiência de custos Grangercausa a expectativa futura a atividade econômica e o indicador de solvência na maioria das amostras As demais análises das funções impulsoreposta e das decomposições de variância do erro de previsão nos mostraram que os resultados de cada instituição financeira é próprio e que não há uma convergência entra as amostras analisadas para o período destacado evidenciando que as habilidades e características próprias de gerenciamento são fato determinante para o comportamento da eficiência de custos Como medida de impacto ou seja quantificação as funções impulsoreposta também se mostraram heterogêneas entre os bancos brasileiros Após oito trimestres um choque positivo de um desvio padrão nas expectativas futuras ceteris paribus afetou a eficiência de custos entre 200 Santander e 47 pontosbase ou bps Bradesco Para o riscopaís os efeitos variaram de 27 bps Banco do Brasil até 156 bps Santander para a atividade econômica entre 96 bps Bradesco e 155 bps Santander para a inflação entre 37 bps Itaú e 56 bps Santander e entre 150 bps Santander e 90 bps Caixa para o indicador de solvência Este estudo está organizado da seguinte maneira A seção 2 fornece uma breve revisão da literatura para a eficiência bancária e suas técnicas de apuração além das suas relações com outras variáveis e apresenta as principais hipóteses da pesquisa Na seção 3 são apresentadas as metodologias para definição dos 20 escores de eficiência e para estimação do modelo autorregressivo vetorial e uma breve explicação das variáveis e dados Os resultados empíricos as análises e discussões acerca dos modelos e os diversos testes de robustez são detalhados na seção 4 A seção 5 conclui o estudo com as suas contribuições para a literatura suas limitações e sugestões de estudos futuros 21 2 REVISÃO DA LITERATURA Nesta seção apresentaremos um resumo da literatura sobre a eficiência bancária A primeira subseção conceitua o assunto citando desde os primeiros estudos sobre eficiência na década de 50 até os estudos mais atuais já específicos do universo bancário A segunda subseção mergulha nas diferentes técnicas de estimação de eficiência e seus respectivos usos na literatura A terceira subseção apresenta alguns destaques da literatura nos estudos que analisaram as relações entre variáveis diversas e a eficiência bancária A quarta e última subseção descreve as hipóteses de pesquisa desse estudo e algumas das suas respectivas referências 21 Eficiência bancária definição Farrel 1957 foi um dos primeiros autores a estudar a eficiência e a produtividade Seus estudos deram origem tanto a definição de eficiência e produtividade quanto às suas estimações Kumbhakar e Lovell 2010 destacaram esse pioneirismo do autor na medição da eficiência técnica e econômica que apresentou o método de medição da eficiência técnica através dos insumos a eficiência alocativa desses insumos e a eficiência de custos Uma premissa fundamental do estudo de Farrel 1957 foi permitir alocações imperfeitas de inputs e outputs definindoas como a distância de uma empresa a uma função de fronteira de produção referência Ou seja se a produção de uma empresa estiver na fronteira essa é perfeitamente eficiente mas se estiver abaixo será ineficiente Adicionalmente a razão entre a produção realizada ou auferida e a produção potencial ou fronteira de produção é o nível de eficiência de cada empresa ou unidade tomadora de decisão decision making units DMU Ele ainda propõe uma decomposição da eficiência em eficiência técnica ET e eficiência alocativa EA sendo a sua soma a representação da medida de eficiência total A eficiência técnica é a capacidade de uma empresa ou DMU de minimizar os inputs e maximizar os outputs já a eficiência alocativa 22 pressupõe a capacidade de usar os insumos nas melhores proporções possíveis levando em conta questões como preço e tecnologia de produção Mais tarde Havrylchyk 2006 utilizando os estudos de Farrel 1957 como referência o complementou descrevendo a eficiência técnica como a capacidade de produzir o máximo de outputs em um determinado nível de inputs ou a capacidade de usar o nível mínimo de inputs em um determinado nível de outputs v 30 p 1982 tradução nossa e a eficiência alocativa como a capacidade de selecionar a combinação ideal de insumos à luz de determinados preços a fim de produzir um determinado nível de produtos v 30 p 1982 tradução nossa Leibenstein 1966 outro precursor do tema o primeiro a utilizar a expressão X ineficiência para descrever a ineficiência de custo empregou o termo como a ineficiência oriunda de razões organizacionais e não por razões tecnológicas como por exemplo quando os interesses dos líderes e dos liderados não estão perfeitamente alinhados com os dos acionistas proprietários Elyasiani e Mehdian 1990 diferindo da literatura até então que na sua maioria adotava uma estrutura de função de custo mas seguindo na linha dos estudos de Farrel 1957 adotaram uma abordagem de fronteira de produção que segundo os autores era mais apropriado para a mensuração de eficiência enquanto as funções de custo eram mais úteis para a mensuração de economias de escala e escopo Embora o tema seja muito relevante sua análise é muito complexa e desafiadora para o setor bancário Berger e Humphrey 1992 afirmaram que a indústria bancária é uma das indústrias de serviço mais difíceis de medir a eficiência por conta do desafio de escolher quais funções constituem os produtos importantes dado que muitas receitas são implícitas e combinadas Apesar dessas dificuldades sua análise é essencial Allen e Rai 1996 afirmaram que à medida que políticas regulatórias e as realidades do mercado aumentam a concorrência a sobrevivência dos economicamente mais aptos deve prevalecer Com a harmonização dos ambientes regulatórios os bancos mais eficientes terão uma vantagem competitiva ALLEN RAI 1996 v 20 p 656 tradução nossa 23 Carvallo e Kasman 2005 afirmaram em seu estudo com 481 bancos de 16 países da América Latina que a pressão competitiva do setor é particularmente importante para os países emergentes pois neles as instituições financeiras são os principais intermediários para viabilizar economias poupança e investimentos Fiorentino Karmann e Koetter 2006 contextualizaram a importância do tema em seu estudo sobre o mercado bancário alemão Eles pontuaram que em todo o mundo o mercado financeiro e consequentemente as instituições financeiras vivenciaram mudanças substancias tanto no ambiente externo quanto no interno Questões como o aumento da concorrência que foi fruto não só de outros bancos mas de novos entrantes não bancários como a desregulamentação ou assimetria regulatória fomentada pelos órgãos responsáveis custos de informação reduzidos e o progresso tecnológico Até por esses fatores eles ponderam seja pelos reguladores ou pelos profissionais do ramo a dependência cada vez maior da teoria econômica para mensuração da eficiência e sua análise Eles também afirmaram que os índices de eficiência conforme descrição acima de Farrel 1957 assumem um valor entre zero e um sendo que se o índice for igual a um a empresa ou DMU é totalmente eficiente Eles mencionam um exemplo de um escore de eficiência em uma fronteira de custos de 90 que significaria que a DMU conseguiria reduzir os custos em 10 sem alterar os outputs Além das necessidades de mercado a análise da eficiência bancária é relevante tanto pelo aspecto microeconômico quanto pelo aspecto macroeconômico No aspecto microeconômico a busca pela eficiência do sistema bancário é uma realidade para aumento da concorrência e para a melhoria da regulação e supervisão No aspecto macroeconômico a eficiência bancária auxilia a estabilidade de todo o sistema financeiro e impacta o custo de intermediação financeira Além disso conforme afirmou Andries e Cocris 2010 uma maior eficiência dos bancos poderá acarretar numa melhor alocação dos recursos financeiros e consequentemente em um aumento dos investimentos que favoreceriam o crescimento econômico Matthews 2010 complementa em seu estudo de eficiência bancária em mercados emergentes que vários estudos sobre o tema utilizaram 24 equivocadamente a expressão de Leibenstein Xineficiência como eficiência técnica apesar do conceito semelhante Ele afirma que o a eficiência técnica é oriunda da teoria da firma neoclássica e compreende a maximização do lucro Um banco poderia ser tecnicamente ineficiente por razões técnicas como falta de capacitação para seus recursos humanos ou a defasagem tecnológica Como o surgimento de novas tecnologias não é constante e nem todas são perenes e penetram de fato no mercado algum banco poderia não acompanhar um movimento necessário e com isso ficar para trás na comparação com seus concorrentes A Xineficiência não é fruto da variabilidade de habilidades ou da variação temporal da difusão de novas tecnologias mas pela forma como cada banco usa explora e se organiza com tais habilidades e tecnologias O estudo de Aiello e Bonanno 2015 que analisou mais de 120 trabalhos publicados entre os anos de 2000 e 2014 destacou que a crescente atratividade dos estudos de eficiência passa pela divergência metodológica mencionando o questionamento sobre qual seria a melhor fronteira bancária e como estimála adequadamente que segundo eles ainda não obteve uma resposta conclusiva Eles afirmam que isso acarretou numa proliferação de pesquisas empíricas e uma heterogeneidade relevante nos resultados v 30 p 113 tradução nossa Gomes Oliveira e Matias 2017 no seu estudo que comparou a eficiência dos bancos domésticos e estrangeiros no Brasil entendem que para o mercado financeiro brasileiro questões relacionadas à eficiência ganharam maior relevância após a entrada de bancos estrangeiros principalmente a partir da década de 1990 logo após o Plano Real que iniciou a estabilização financeira fazendo com que o setor fosse cada vez mais impactado pelas forças de mercado onde o ambiente é mais competitivo e dinâmico O estudo que difere bastante dos demais é o de Chen et al 2018 que apresenta um caminho contrário tentando propor uma combinação de fatores viáveis e seus ajustes necessários para alcançar a eficiência total que eles comprovam ser atingível Eles utilizaram uma amostra com os principais bancos de Taiwan durante os anos de 1999 e 2011 e concluíram que é possível que esses bancos economizem 103 dos custos totais no curto prazo e 88 no longo prazo 25 Um fato curioso destacado em outro recente estudo de Huljak Martin e Moccero 2019 sobre eficiência de custo e produtividade dos bancos na zona do euro é a crítica aos indicadores de eficiência Average Cost AC razão entre os custos totais e o total de ativos e o Cost to Income Ratio CIR razão entre os custos administrativos e a receita operacional que são os indicadores mais comumente utilizados analistas de mercado para medir eficiência Apesar de fáceis de apurar eles alegam que não capturam a eficiência do setor Sendo que o AC é dependente do modelo de negócio da instituição financeiro e do seu tamanho Por exemplo bancos de atacado tendem a investir menos em estrutura física presencial em agências do que bancos de varejo que necessitam de capilaridade Eles também alegam que o AC depende de fatores específicos do país e as vezes até da região como o custo de mãodeobra e as exigências regulatórias Logo o AC seria um índice dos preços médios dos insumos para a produção de alguma unidade bancária ativos serviços entre outros e não um indicador de eficiência Já o CIR além de também serem impactados por fatores regionais e locais específicos reúne aspectos diferentes do desempenho além da eficiência como produtividade por exemplo Outra desvantagem do CIR é que ele é impactado pelo risco de crédito o que distorce a estimativa de eficiência Em contrapartida os autores afirmam que uma análise de fronteira estocástica traria vantagens pois controla o fato de os bancos produzirem diferentes produtos e consumirem ou pagarem por diferentes insumos permitindo então a comparação entre bancos de diferentes modelos de negócio especialização tamanho e estrutura Curiosamente apesar de vários estudos mencionarem a quantidade considerável de papers publicados sobre o tema eficiência Abreu Kimura e Sobreiro 2019 ao analisarem os 87 artigos publicados mais bem avaliados segundo a Academic Journal Guide of Association of Business Schools entre janeiro de 2011 e julho de 2017 encontraram alto grau de dispersão dos estudos e um número relativamente limitado de autores dedicados ao tema Segundo os autores esse seria um indício de que apesar da tradição do tema a eficiência bancária ainda não é uma área de conhecimento bem definida 26 22 Técnicas de estimação de eficiência Há muitas técnicas métodos para estimar eficiência e consequentemente muita literatura a respeito Ferrier e Lovell 1990 por exemplo elaboraram um artigo comparando duas técnicas para estimar a eficiência bancária uma técnica de estimativa econométrica de fronteira de custos e uma de série de programas lineares de fronteira de produção Elas apresentaram resultados diferentes em relação à eficiência de custos Eles atribuem as correlações entre as duas técnicas não serem estatisticamente significativas a quatro fatores 1 A abordagem de programação não é estocástica e portanto interpreta o ruído como ineficiência 2 A abordagem econométrica impõe estrutura paramétrica à tecnologia e à distribuição da ineficiência e assim mistura erro de especificação com ineficiência 3 A eficiência alocativa está disponível apenas como um valor médio sobre a amostra para a abordagem econométrica de modo que os valores reais da eficiência alocativa e portanto técnica sejam sempre superestimados ou subestimados 4 Para evitar uma categorização excessiva dos bancos e consequentemente um número excessivo de bancos ostensivamente eficientes dentro de cada categoria a abordagem de programação implementada exclui as variáveis categóricas do tipo de instituição Duas dessas categorias têm um impacto estatisticamente significativo no custo na abordagem econométrica esse impacto é atribuído à ineficiência na abordagem de programação FERRIER LOVELL 1990 p 243 tradução nossa Berger e Mester 1997 constataram que as técnicas mais comuns de estimativa de eficiência são como técnicas não paramétricas a análise envoltória de dados data envelopment analysis DEA e a livre disposição de envoltória free disposable hull analysis FDH e como técnicas paramétricas análise de fronteira estocástica stochastic frontier analysis SFA a abordagem de fronteira espessa thick frontier approach TFA e a abordagem da distribuição livre distributionfree approach DFA Eles afirmaram que os métodos não paramétricos geralmente ignoram os preços e portanto podem ser responsáveis apenas pela ineficiência técnica no uso de muitos insumos ou na produção de poucos produtos v 21 p 905 tradução nossa Eles 27 complementam que eles não explicam a ineficiência alocativa na resposta as escolhas de produtos e insumos ou na comparação de empresas que se especializam em diferentes produtos e insumos dado que isso não é possível sem o benefício do preço Logo as técnicas não paramétricas geralmente se concentram na otimização tecnológica e não na otimização econômica e não correspondem aos conceitos de eficiência de custo e lucro Outra desvantagem das técnicas não paramétricas é que elas geralmente não permitem erros aleatórios nos dados assumindo o erro de medição e a sorte como fatores que afetam os resultados embora algum progresso esteja sendo feito nesse sentido De fato eles separam as diferenças de eficiência do erro aleatório assumindo que o erro aleatório é zero BERGER MESTER 1997 v 21 p 905 tradução nossa Berger e Humphrey 1997 complementam ao analisaram 130 estudos de análise de eficiência bancária em 21 países relatando uma distribuição muito similar entre os métodos paramétricos e não paramétricos Fiorentino Karmann e Koetter 2006 também compararam as duas técnicas afirmando encontrarem uma consistência muito baixa entre ambas Eles afirmaram que os métodos não paramétricos são mais sensíveis à heterogeneidade da amostra e que a correlação entre a ordem de classificação de eficiência entre as duas técnicas é positiva mas baixa Segundo os autores esse modelo dos artigos de 1977 é indicado como ln𝑦𝑗 ln 𝑥𝑗𝛽 𝑣𝑗 𝑢𝑗 1 onde 𝑥𝑗 corresponde a um input para a empresa j 𝑣𝑗 corresponde a um erro aleatório adicionado ao termo de ineficiência não negativo o 𝑢𝑗 Esse erro aleatório e efeitos combinados de variáveis de entrada não especificadas na função de produção que explicam o erro de medição e outros eventuais fatores aleatórios que podem afetar o valor do output Como o limite superior é determinado pela variável estocástica exp𝑥𝑗𝛽 𝑣𝑗 o modelo é estocástico sendo o erro aleatório podendo ser negativo ou positivo logo os resultados dessa fronteira variam em relação à parte determinística do modelo nesse caso 28 exp𝑥𝑗𝛽 Para a utilização da técnica de análise de fronteira estocástica é necessário assumir uma forma funcional para estimar o modelo A complexidade do setor bancário conforme descrito anteriormente como um setor de produção múltipla inviabiliza a especificação de uma função de produção logo se faz necessária a especificação de uma fronteira de custos A fronteira de custos é apresentada pelos autores como a forma geral de log ln 𝐶𝑗 𝑓 ln 𝑦𝑟𝑗 ln 𝑐𝑖𝑗 𝜀𝑗 onde 𝐶𝑗 é o custo total da empresa j 𝑦𝑟𝑗 mede o résimo output da empresa j o 𝑐𝑖𝑗 é o preço do iésimo input da empresa j e o termo do erro 𝜀𝑗 é o erro aleatório 𝑣𝑗 e o termo de ineficiência 𝑢𝑗 descrito na equação 1 acima O termo de erro aleatório 𝑣𝑗 é classificado como independente e identicamente distribuída iid como 𝑣𝑗 𝑁 0 𝜎𝑣 2 e independente das variáveis explicativas O termo de ineficiência 𝑢𝑗 também é iid com 𝑢𝑗 𝑁 0 𝜎𝑢 2 independente do termo de erro e é extraído de uma distribuição não negativa truncada em zero Sobre o surgimento da análise de fronteira estocástica Kumbhakar e Lovell 2010 detalharam no seu capítulo introdutório as suas origens contando que surgiu em dois trabalhos publicados quase que simultaneamente por duas equipes distintas Meeusen e van den Broek em junho de 1977 e Aigner Lovell e Schmidt em julho de 1977 sendo esse último uma versão que reunia e mesclava um artigo do Aigner e outro do Lovell e Schmidt Apesar das publicações terem ocorrido em continentes distintos com somente um mês de diferença entre elas os documentos eram muito semelhantes sendo que estavam em preparação há três anos Logo após um terceiro artigo também muito conhecido foi lançado por Battese e Corra também em 1977 sendo que esses autores já tinham servido de referência para o artigo do Aigner Lovell e Schmidt Um estudo mais recente de Aiello e Bonanno 2015 realizou uma Análise de MetaRegressão MetaRegressionAnalysis MRA examinando 1661 escores de eficiência tendo encontrado como um dos seus principais resultados que os métodos paramétricos produzem níveis mais baixos de eficiência que os métodos não paramétricos Eles destacam também algumas diferenças relevantes por conta da qualidade dos estudos e o número de observações e variáveis Em estudo posterior dos mesmos autores 2018 que utilizou a 29 mesma base para análise e teve como objetivo mensurar o impacto da seleção do método e fatores específicos do país ou região na variabilidade do escore de eficiência destacou que não há consenso sobre a superioridade de um método de estimativa sobre outro v 32 n 1 p 195 tradução nossa O estudo de Silva et al 2017 que comparou as eficiências dos bancos locais chineses contrapõe a afirmação sobre heterogeneidade de Fiorentino Karmann e Koetter 2006 acima ao fornecer testes de robustez com resultados semelhantes através da remoção da heterogeneidade da amostra ao estimar níveis de eficiência no DEA e utilização de efeitos fixos para absorver as heterogeneidades dos bancos no SFA Eles atribuem a diferença entre as duas técnicas à natureza diferente de fronteiras eficientes que cada método estima Outra contribuição importante desse estudo foi que apesar dos resultados não fornecerem classificações de eficiência semelhantes entre os dois métodos na análise individual eles forneceram resultados semelhantes em relação às pontuações médias para todo o mercado ao longo dos anos analisados Caminhando na mesma linha o recente estudo de Ahmad et al 2019 afirmou que entre os 100 principais mais citados artigos sobre desempenho do setor bancário conforme Gráfico 1 abaixo 26 utilizaram índices financeiros exemplos retorno sobre o patrimônio líquido ou return on equity ROE retorno sobre os ativos ou return on assets ROA margem de juros líquida ou net interest margin NIM crédito não produtivo ou nonperforming loan NPL e os outros 74 utilizaram análise de fronteira na qual destacamos que 34 utilizaram métodos paramétricos predominantemente SFA e análise fatorial explanatória ou exploratory factor analysis EFA além de outros já mencionados acima e 40 não paramétricos sendo majoritariamente a DEA destaque também para a DEA em redes ou network DEA 30 Gráfico 1 Abordagem de estimativas nos 100 principais artigos Fonte adaptado de AHMAD ET AL 2019 p 16 23 Eficiência bancária e suas relações com outras variáveis Dentre os estudos que analisam comparam e associam a eficiência com outras variáveis destacamos o estudo de Elyasiani e Mehdian 1990 que investigou o impacto do tamanho e a ramificação quantidade de agências ou distribuição da rede Eles concluíram que o tamanho nesse caso medido pelo total de ativos ou receita total tem um efeito positivo na eficiência ou seja bancos maiores são mais eficientes Os autores vincularam a maior parte desse aumento na eficiência à eficiência de escala afirmando que isso ocorre pois os bancos conseguem se aproximar do ponto mínimo da curva de custo médio de longo prazo Outro ponto interessante e até controverso com outros estudos futuros é que eles afirmam que uma parte pequena dessa maior eficiência dos bancos grandes é associada a diferenças na eficiência técnica Não paramétrica 40 Paramétrica 34 Indicadores Financeiros 26 DEA 33 DEA em redes 3 Outros 4 SFA 31 EFA 1 DFA 2 TFA 1 ROA 8 ROE 7 Outros 8 NIM 2 NPL 1 31 Berger e DeYoung 1997 abordaram a relação entre os empréstimos problemáticos vencidos e a eficiência bancária empregando causalidade de Granger para testar hipóteses sobre a relação entre a qualidade do empréstimo e a eficiência e concluíram que os empréstimos problemáticos precedem ineficiência e que a eficiência precede reduções nos empréstimos problemáticos Além disso eles analisaram se os estudos devem ou não considerar os empréstimos problemáticos nas estimativas de eficiência mas não foram conclusivos apresentando resultados ambíguos Vale ressaltar que segundo o estudo de Ahmad et al 2019 esse artigo do Berger e DeYoung 1997 foi o mais citado enaltecendo sua relevância Outro estudo de Berger e Hannan 1998 tentou relacionar a concentração de mercado com a eficiência operacional testando a hipótese de que pressões mais brandas para minimizar os custos por conta da concorrência podem resultar em menor eficiência operacional para empresas em mercados concentrados Eles encontraram fortes evidências de que essa hipótese é verdadeira Fiorentino Karmann e Koetter 2006 concluíram que as medidas de eficiência tanto a técnica paramétrica quanto a não paramétrica apresentam baixa correlação com os indicadores de desempenho mais comumente utilizados como custobenefício e taxas de retorno O estudo de Athanasoglou Brissimis e Delis 2008 investigou os efeitos de determinantes específicos do próprio banco eficiência operacional risco financeiro e tamanho específicos da indústria a concentração da indústria e a participação de bancos privados no setor e macroeconômicos comportamento cíclico ou relação com o ciclo de negócios e inflação esperada na lucratividade dos bancos gregos entre 1985 e 2001 Eles encontraram resultados empíricos que sugerem que os determinantes do próprio banco com exceção do tamanho afetam significativamente a lucratividade já os da indústria não são relevantes pelo menos no sistema bancário grego que segundo os autores evoluiu muito ao longo do período analisado Por último o ciclo de negócio afeta significativamente o lucro dos bancos assim como a inflação esperada que apresentou relação positiva 32 Casu e Girardone 2009 testaram através da causalidade Granger a relação entre concorrência e a eficiência bancária para cinco países da União Europeia E diferentemente do que pressupõe a teoria do monopólio elas não encontraram evidências de que o poder de mercado dos bancos se traduz em uma diminuição na eficiência de custos porém elas não puderam concluir que a rejeição da hipótese de relaxamento na gestão oriunda de um monopólio suporte o paradigma da estrutura eficiente Elas testaram também a causalidade reversa que seria da eficiência à concorrência mas não encontraram resultados que sugerissem que o aumento da eficiência promoveria um aumento no poder de mercado Já Fiordelisi MarquesIbanez e Molyneux 2011 publicaram um estudo onde examinaram duas questões do setor bancário europeu a relação entre 1 eficiência bancária e risco bancário e 2 o impacto do capital na troca entre risco e eficiência Na primeira relação analisaram se baixa eficiência ou ineficiência poderia acarretar numa tentativa de aumento dos retornos e consequentemente prejudicar os padrões operacionais como gestão do crédito risco e na outra direção se um aumento dos riscos poderia ser um predecessor de uma redução na eficiência Na segunda relação analisaram por exemplo se os baixos índices de capitalização são fruto da eficiência ou da ineficiência Para tal utilizaram medidas prospectivas e retrospectivas de risco capital e três medidas de eficiência bancária receita custo e lucro Esse olhar para outras medidas de eficiência é fundamental pois refletem habilidades gerenciais diferentes ou seja as habilidades para maximizar as receitas e minimizar os custos v 35 p 1316 tradução nossa por exemplo Seus resultados mostraram que uma menor eficiência tanto para a medida de custo quanto receita Grangercausa maior risco bancário confirmando o que eles chamaram de hipótese de má gestão que aumentos no capital precedem ganhos de eficiência de custos que a eficiência de custos Grangercausa aumentos de capital e níveis mais elevados de capital tendem a ter um efeito positivo nos níveis de eficiência Outro estudo que também analisou a relação da eficiência bancária com risco foi o de Tan e Anchor 2017 que testou o impacto do risco em nível micro e da concorrência na eficiência do setor bancário chinês no período de 2003 a 2013 Eles consideraram formas abrangentes de comportamento de risco incluindo 33 risco de crédito risco de liquidez e risco de insolvência Eles destacaram o risco de liquidez que afeta significativamente e negativamente a eficiência e que uma maior concorrência precederia um declínio na eficiência Eles mencionam também algumas relações com variáveis macroeconômicas onde os resultados sugerem que um ambiente de alta inflação ou de boom econômico taxa de crescimento do PIB a eficiência é maior No estudo de Aiello e Bonanno 2018 que teve como um dos seus objetivos mensurar o impacto de fatores específicos do país na variabilidade do escore de eficiência e para isso analisou 1661 escores de eficiência destacou que a liberalização do sistema financeiro também desempenha um papel positivo no escore mencionando que países e regiões com maior liberalização no setor como os Estados Unidos União Europeia Europa Oriental e Oceania possuíam pontuações mais altas Por outro lado países de regiões com regulamentação mais rígida como na América Latina e na África números mais baixos Outra análise muito interessante foi feita por Barros et al 2018 onde realizaram uma avaliação da eficiência bancária em um cenário cultural comum selecionando países em desenvolvimento de mesmo idioma e passado colonial Angola Brasil e Moçambique Para os autores dessa forma seria possível afirmar que os resultados estão puramente relacionados ao tamanho da escala e às questões políticas e econômicas de cada país Como os resultados encontrados revelaram que o Brasil possui o mais alto nível de eficiência e retorno de escala puderam afirmar que os países com mais alto grau de desenvolvimento são mais eficientes Assim como diversos outros estudos que relacionaram a eficiência com o tamanho da organização Henriques et al 2018 que avaliou a eficiência bancária dos 37 maiores bancos brasileiros entre 2012 e 2016 concluíram que os cinco maiores bancos brasileiros demonstraram uma ineficiência diretamente relacionada ao nível de escala Eles complementam que uma das possíveis causas para isso seria a necessidade que eles apresentam em atenderem um grande número de clientes em todo o país acarretando uma estrutura de ativos e despesas muito elevada em relação aos bancos menores 34 Kallel Hamad e Triki 2019 também contribuem com a análise das relações da eficiência bancária com outras variáveis Em sua comparação entre as eficiências dos bancos tunisianos e marroquinos entre 2005 e 2014 eles puderam concluir como já era a expectativa deles que bancos mais capitalizados têm um nível de custos mais elevado que segundo eles é necessário para cobrir os riscos assumidos Já ao contrário do que eles esperavam o que eles chamaram de a qualidade do serviço bancário controlado por eles pelo índice de outras receitas operacionais sobre ativos totais apresentou um efeito negativo no custo total ou seja os bancos com maior receita de serviços tendem a ser mais eficientes no gerenciamento de custos Eles interpretaram esse resultado com fruto do alto nível de inadimplência nos sistemas bancários analisados Outra interessante contribuição foi do estudo de Shamshur e Weill 2019 que investigou o impacto da eficiência de custo dos bancos no custo do crédito para as empresas testando a hipótese de que um banco mais eficiente poderia fornecer empréstimos mais baratos a seus clientes Eles encontraram evidências de que essa hipótese é verdadeira porém com algumas ressalvas A eficiência reduz o custo do crédito para as pequenas e médias empresas mas não exerceu influência significativa para as grandes ou microempresas 24 As hipóteses de pesquisa Antes de introduzirmos a seção sobre a metodologia definimos as principais hipóteses de pesquisa Para tal consideramos nosso propósito de identificar variáveis através de suas séries temporais que contenham informações sobre o desenvolvimento futuro da eficiência bancária das principais instituições financeiras do Brasil Formulamos portanto cinco hipóteses de pesquisa para os testes empíricos tendo por base a razoabilidade econômica e referências discutidas ao longo da revisão de literatura O Quadro 1 apresenta os estudos sintetizados desenvolvidos em diferentes ambientes econômicos 35 Quadro 1 Síntese dos estudos considerados para a identificação das hipóteses Fonte elaborado pelo próprio autor H1 Expectativas futuras Eficiência bancária A primeira hipótese H1 levantada é a de que indicadores prospectivos relacionados à previsão da atividade econômica e às expectativas futuras possuem uma relação positiva com a eficiência de custos das maiores instituições financeiras do Brasil Em Beccalli Casu e Girardone 2006 essa associação foi analisada através dos preços das ações bancárias e os resultados H Tema Descrição Referências Objeto amostra do estudo Beccalli Casu e Girardone 2006 concordam França Alemanha Itália Espanha e Reino Unido Fiordelisi MarquesIbanez e Molyneux 2011 concordam União Euroéia Chen Higgins e Mason 2005 concordam Estados Unidos Chen Higgins e Mason 2005 concordam Estados Unidos Tamadonnejad et al 2013 concordam Leste da Ásia Borges 2012 concorda União Euroéia AlGasaymeh 2016 concorda Conselho de Cooperação do Golfo Chen et al 2018 não encontrou resultados significativos Taiwan Chortareas Girardone e Ventouri 2012 discordam União Euroéia Sufian 2009 discorda Malásia Kallel Hamad e Triki 2019 discordam Tunísia e Marrocos Tan e Anchor 2017 concordam China AlGasaymeh 2016 concorda Conselho de Cooperação do Golfo Primo et al 2013 concordam Brasil Chen Higgins e Mason 2005 concordam Estados Unidos Tan e Anchor 2017 discordam China Primo et al 2013 concordam Brasil AlGasaymeh 2016 concordam Conselho de Cooperação do Golfo Chortareas Girardone e Ventouri 2012 concordam União Euroéia Sufian 2009 concorda Malásia Zhang et al 2013 concordam Brasil Rússia Índia e China BRIC Fiordelisi MarquesIbanez e Molyneux 2011 discordam União Euroéia Athanasoglou Brissimis e Delis 2008 discordam Grécia Tecles e Tabak 2010 discordam Brasil Mendonça 2018 discordam Brasil H3 Atividade econômica A eficiência de custos das maiores instituições financeiras do país é positivamente relacionada com a atividade econômica H4 Inflação A eficiência de custos das maiores instituições financeiras do país é negativamente relacionada com a inflação H2 Medida de riscopaís A eficiência de custos das maiores instituições financeiras do país é negativamente relacionada com a medida de riscopaís H5 Solvência A eficiência de custos das maiores instituições financeiras do país é negativamente relacionada com a solvência H1 Expectativas futuras A eficiência de custos das maiores instituições financeiras do país é positivamente relacionada com a expectativa futura de mercado 36 sugeriram que a relação é positiva Em Fiordelisi MarquesIbanez e Molyneux 2011 com o objetivo de analisar um indicador prospectivo de risco bancário que conseguisse capturar a probabilidade de inadimplência dos bancos ou seja que avalie diversos riscos bancários não só o risco de crédito na visão do mercado financeiro recorreram ao Expected Default Frequency EDF apurado pela Moodys KMV um ano à frente Os resultados encontrados mostraram que melhorias na eficiência de custo bancário Grangercausam uma menor probabilidade de inadimplência Também corroborando com a nossa hipótese Chen Higgins e Mason 2005 encontraram uma associação positiva entre um indicador de previsão da atividade econômica futura1 está associado a expectativa de aumento do crescimento econômico e a eficiência de custos Nesse mesmo estudo também como um indicador preditivo do desempenho econômico analisaram a relação entre a eficiência de custos e o prêmio dos títulos federais fed funds premium FFP apurado como a diferença entre a taxa de fundos federais e a taxa de três meses de letras do Tesouro Eles encontraram que um spread alto que indica baixo crescimento econômico está negativamente associada à eficiência de custos H2 Risco país Eficiência bancária A segunda hipótese H2 levantada é a de que a eficiência de custos das maiores instituições financeiras do país é negativamente relacionada com a medida de riscopaís Corroborando com essa afirmação Tamadonnejad et al 2013 apresentaram resultados que mostraram que a relação entre a eficiência bancária e o riscopaís são negativamente relacionadas Os autores argumentaram que elevações no riscopaís acarretaram em falta de confiança da população e corridas repentinas aos bancos que consequentemente geraram um aumento nos empréstimos mal executados e o risco de inadimplência Na mesma linha Borges 20122 afirmou que os bancos oriundos de países com classificações piores ratings que expressam uma opinião desfavorável em relação ao país e consequentemente não atrairiam investidores provavelmente sofrerão reduções nos escores de eficiência uma vez que o coeficiente é 1 Os autores utilizaram um indicador de Ciclo de Negócio apurado pelo The Conference Board e disponível no seu site 2 Nesse estudo os ratings não foram considerados estatisticamente significativos a um nível de significância de 5 ou 10 somente num nível de 15 37 negativo Também concordando com a afirmação AlGasaymeh 2016 encontrou evidências de que as variáveis de riscopaís desempenham um papel importante na influência da eficiência dos bancos nos países analisados Segundo o autor os riscos mais baixos criam um ambiente de trabalho que favorece a eficiência pois os bancos têm um desempenho melhor nesse cenário estável e equilibrado H3 Atividade econômica Eficiência bancária A terceira hipótese H3 levantada é a de que a eficiência de custos das maiores instituições financeiras do país é positivamente relacionada com a atividade econômica Concordando com essa afirmação Tan e Anchor 2017 encontraram resultados que mostram que a taxa de crescimento do PIB está significativa e positivamente relacionada à eficiência dos bancos analisados indicando que são mais eficientes durante os períodos de boom econômico Eles interpretam isso como resultado de uma demanda maior por serviços bancários uma melhora da qualidade dos tomadores de empréstimos que acarreta uma redução no custo do monitoramento de riscos e uma redução no custo de economias de escala e escopo que precedem melhorias na eficiência Na mesma linha AlGasaymeh 2016 encontrou resultados de que quanto maior o PIB per capita menor o custo dos bancos e consequentemente maior a eficiência Outro estudo que também encontrou evidências que corroboram com essa hipótese foi o de Primo et al 2013 que evidenciaram a relevância estatística do PIB dado que nos momentos de maior desenvolvimento econômico a rentabilidade dos bancos aumentou Eles explicam o fato pelo aumento na demanda de crédito a redução do nível de inadimplência e um aumento nas receitas de tarifas oriundo da inclusão financeira Da mesma forma Chen Higgins e Mason 2005 apresentaram resultados de que a variação percentual do PIB está positivamente associada à eficiência de custos Já Chen et al 2018 não encontraram um impacto significativo do PIB sobre a eficiência Do lado contrário à nossa hipótese Chortareas Girardone e Ventouri 2012 encontraram resultados de que o PIB per capita é significativo na eficiência bancária mas de forma negativa o que indicaria que bancos em expansão poderiam ser menos eficientes no controle de custos Da mesma forma Kallel Hamad e Triki 2019 verificaram que o coeficiente associado ao crescimento do 38 PIB é significativamente positivo com relação ao aumento de custos ou seja redução da eficiência Também contrário à nossa hipótese Sufian 2009 encontrou que as taxas de crescimento da renda nacional medido pelo PIB exibiram uma relação negativa e estatisticamente significativa com a eficiência bancária nos bancos analisados que segundo ele seriam fruto da sua relativa volatilidade no período analisado H4 Inflação Eficiência bancária A quarta hipótese H4 levantada é a de que a eficiência de custos das maiores instituições financeiras do país é negativamente relacionada com a inflação Tan e Anchor 2017 apresentaram resultados que contrariam essa afirmação evidenciando que um aumento da inflação geraria um aumento na eficiência Para eles no cenário analisado como medida de combate à inflação o governo e as autoridades regulatórias adotam medidas como aumento da taxa de juros eou aumento dos compulsórios que reduzem os volumes de empréstimos concedidos e por isso reduzem o custo de monitoramento e gerenciamento de riscos que aprimora a eficiência Em outro estudo mais alinhado com a concordância da hipótese Primo et al 2013 utilizaram o retorno sobre o patrimônio líquido return on equity ROE e o retorno sobre os ativos return on assets ROA para a eficiência em dois modelos distintos para cada variável sendo que encontrou resultados mistos entre eles e a inflação Todos foram significativos mas em três com sinal negativo e em um com sinal positivo Os autores afirmaram que a predominância da relação negativa indicaria que os efeitos relacionados à perda de valor sobre os ativos seriam superiores aos ganhos da redução de valor real dos passivos Já de forma menos ambígua Al Gasaymeh 2016 encontrou uma relação significativa e positiva que segundo ele indicaria que os bancos em países com inflação mais elevada teriam custos potenciais mais elevados H5 Solvência Eficiência bancária A quinta hipótese H5 levantada é a de que a eficiência de custos das maiores instituições financeiras do país é negativamente relacionada com a solvência Em linha com essa afirmação Chortareas Girardone e Ventouri 2012 concluíram que as exigências de capital que apresentaram um coeficiente 39 negativo e estatisticamente significativo em todos os modelos testados aumentam a eficiência do banco Do mesmo modo Sufian 2009 pontuou que sua variável relacionada ao capital do banco Patrimônio Líquido sobre o Total de Ativos exibiu uma relação negativa e estatisticamente significativa com a eficiência bancária Segundo o autor esses resultados sugeririam que os bancos mais eficientes operariam mais alavancados ou seja com menos patrimônio e que os bancos menos eficientes poderiam estar envolvidos em operações com maior risco que exigiriam maior colchão de capital Também alinhado com essa hipótese Zhang et al 2013 encontrou evidências de que um excesso de capital também medido como o Patrimônio Líquido sobre o Total de Ativos poderia induzir uma ineficiência restringindo o crescimento dos ativos mais rentáveis Já na contramão da nossa afirmação Fiordelisi MarquesIbanez e Molyneux 2011 encontraram que a razão de capital definida por eles da mesma forma que os autores anteriores Grangercausa positivamente a eficiência de custo estatisticamente significativa Ou seja um aumento na soma dos coeficientes do índice de capital defasado precede temporalmente os aumentos de eficiência de custos p 1319 tradução nossa Também contrariando nossa hipótese Athanasoglou Brissimis e Delis 2008 constataram que a variável que representa o capital também utilizaram o mesmo indicador é estatisticamente significativo e positivo em relação à eficiência Segundo os autores isso significaria que um banco com uma estrutura de capital sólida teria condições de perseguir e conquistar negócios mais eficientes e até mesmo mais tempo e flexibilidade para enfrentar desafios com perdas inesperadas Outro estudo que contraria essa hipótese é o de Tecles e Tabak 2010 que também encontrou um sinal negativo no coeficiente de patrimônio dos bancos na sua relação com a eficiência bancária Eles afirmaram que isso corroboraria com a teoria do risco moral dado que quando uma proporção maior de capital acionário está em risco a alta liderança da instituição financeira tem mais incentivos para monitorar a eficiência Na mesma linha Mendonça 2018 também encontrou significância estatística na variável de capital e coeficiente positivo que indicaria que quanto maior for a capitalização maior será a rentabilidade Ele alegou que isso é decorrente da sinalização ao mercado de uma menor probabilidade de falência que lhe permitiria reduzir seus custos de empréstimo e captação 40 3 METODOLOGIA 31 Análise de fronteira estocástica Para esse estudo assim como Williams 2004 Fiorentino Karmann e Koetter 2006 Fiordelisi MarquesIbanez e Molyneux 2011 e tantos outros a técnica paramétrica selecionada foi a análise de fronteira estocástica Além de bastante difundida essa técnica conforme afirmou Berger e Mester 1997 corresponde bem tanto aos conceitos de eficiência de custo quanto de lucro Conforme Fiordelisi MarquesIbanez e Molyneux 2011 ou Williams 2004 a eficiência de custos foi mensurada através da análise de fronteira estocástica Para esse estudo foi utilizado o modelo de fronteira estocástica de Battese e Coelli 1995 ln 𝑇𝐶𝑖𝑡 𝑥𝑖𝑡 𝛽 𝑉𝑖𝑡 𝑈𝑖𝑡 2 onde 𝑡 representa a dimensão de tempo 𝑇𝐶𝑖 é o logaritmo do custo de produção do iésimo banco 𝑥𝑖 é um vetor 1 𝑥 𝑘 de inputs e outputs de preços padronizados do iésimo banco 𝛽 é um 𝑘 𝑥 1 vetor de parâmetros desconhecidos 𝑉𝑖 é uma variável aleatória assumida iid distribuída como uma 𝑁0 𝜎𝑣 2 e independente de 𝑈𝑖 𝑈𝑖 é uma variável aleatória não negativa que representa a ineficiência de custo na produção assumida como iid de modo que 𝑈𝑖 é obtida por truncamento em zero da distribuição 𝑁𝜎 𝜎𝑈 2 sendo 𝜎 um parâmetro a ser estimado Para esse estudo foi utilizada a seguinte função de custo translog para estimar uma fronteira de referência comum ln 𝑇𝐶 𝛼0 𝛼𝑖 ln 𝑦𝑖 3 𝑖1 𝛽𝑖 ln 𝑤𝑖 3 𝑖1 𝜏1 ln 𝐸 𝑡1𝑇 1 2 𝛿𝑖𝑗 ln 𝑦𝑖 ln 𝑦𝑗 3 𝑗1 3 𝑖1 𝛾𝑖𝑗 ln 𝑤𝑖 ln 𝑤𝑖 ln 𝑤𝑗 3 𝑗1 3 𝑖1 𝜙1 ln 𝐸 ln 𝐸 𝑡1 𝑇2 𝜌𝑖𝑗 ln 𝑦𝑖 ln 𝑤𝑗 3 𝑗1 3 𝑖1 𝜓𝑖 ln 𝑦𝑖 ln 𝐸 3 𝑗1 𝜑𝑖𝑇 ln 𝑦𝑖 3 𝑗1 𝜃𝑖 ln 𝑤𝑖 ln 𝐸 3 𝑖1 𝜐𝑖𝑇 ln 𝑤𝑖 3 𝑖1 ln 𝑢𝑐 ln 𝜀𝑐 3 41 onde ln 𝑇𝐶 é o logaritmo do custo de produção 𝑦𝑖𝑖 123 são outputs de quantidade 𝑤𝑗𝑗 123 são inputs de preço ln 𝐸 é o logaritmo natural do capital total 𝑇 é uma tendência temporal 𝑢𝑐 são os componentes da ineficiência de custo E para garantir a homogeneidade linear nos preços dos fatores 𝛽𝑗 3 𝑗1 1 𝛾𝑖𝑗 3 𝑖1 0 𝜌𝑖𝑗 3 𝑗1 0 4 ainda é suficiente e necessário aplicar as seguintes restrições a simetria padrão que de acordo com esta restrição assumese que 𝛿𝑖𝑗 𝛿𝑗𝑖 e 𝛾𝑖𝑗 𝛾𝑗𝑖 Como restrição linear da função custo os inputs e outputs desse trabalho foram definidos também seguindo como referência o trabalho de Fiordelisi Marques Ibanez e Molyneux 2011 de acordo com a abordagem de valor agregado Consideramos as variáveis trabalho despesa de pessoal sobre ativo total manutenção despesas administrativas sobre ativo total e despesas de captação despesas de captação sobre o volume de captações como insumos ou seja inputs e saldo de captações saldo da carteira de crédito e saldo de títulos e valores mobiliários TVM e derivativos como produtos ou outputs Também estimamos utilizando o mesmo modelo funcional adotado para eficiência de custos a eficiência da receita e do lucro usando como variável dependente a receita total e o lucro total respectivamente A definição de inputs e outputs assim como no trabalho referência é a mesma adotada para estimar a eficiência de custos Conforme Belotti et al 2013 as medidas ou escores de eficiência foram estimadas via 𝐸𝑒𝑥𝑝𝑢𝜀 como o estimador criado por Battese e Coelli em seu estudo publicado em 1988 e assumem valores entre 0 e 1 sendo 1 a indicação de um banco totalmente eficiente Esse valor indica a porcentagem de custos observados que seriam suficientes para produzir o resultado observado se o banco fosse totalmente eficiente Por exemplo numa medida de eficiência de 080 poderíamos afirmar que com 80 dos custos totais seria possível produzir os mesmos resultados 42 32 Modelo Autorregressivo Vetorial Conforme a descrição de Lütkepohl 2005 de que se fosse possível obter uma série temporal de dados para uma variável de interesse e esses dados contiverem informações sobre o desenvolvimento futuro de uma variável seria plausível usar como previsão alguma função desses dados coletados optamos por um modelo de séries temporais o Modelo Autorregressivo Vetorial Vector Autoregressive VAR Cavalcanti 2010 destacou que o uso desse método se difundiu rapidamente e que hoje estão entre os instrumentos mais usados em investigações empíricas na área de macroeconomia v 14 n 2 p 251 Ele credita esse feito à percepção de que com ele é possível analisar as interrelações entre diversas variáveis através de um conjunto mínimo de restrições para identificar o componente exógeno de cada variável portanto é possível estimar o efeito de um choque nessa variável sobre as demais Christiano 2012 lembrou que o VAR sugerido por Sims 1980 teria como um dos seus propósitos e o que teve maior impacto projetar e avaliar modelos econômicos Segundo ele os economistas estão acostumados a pensar em modelos econômicos em termos de impulsos e mecanismos de propagação e os VARs são um dispositivo para organizar os dados precisamente nessas categorias v 114 p 1086 tradução nossa Isso reforça nossa escolha Lütkepohl 2005 apresentou as principais etapas de uma análise VAR Figura 1 abaixo que foi a referência seguida Figura 1 Principais etapas da análise VAR Fonte adaptado de LÜTKEPOHL 2005 p 6 43 Christiano 2012 apresenta o VAR como um vetor 𝑦𝑡N x 1 que denota o conjunto de variáveis selecionadas na análise A suposição de que 𝑦𝑡 segue um VAR de ordem 𝑝 significa que ele pode ser expresso como 𝑦𝑡 𝐵𝑜 𝐵1𝑦𝑡1 𝐵𝑝𝑦𝑡𝑝 𝑢𝑡 𝐸𝑢𝑡𝑢 𝑡 𝑉 5 Onde 𝑢𝑡 não está correlacionado com 𝑦𝑡1 𝑦𝑡𝑝 Supõese que 𝑝 tenha um valor suficientemente grande para que 𝑢𝑡 não seja correlacionado automaticamente ao longo do tempo Os distúrbios do VAR 𝑢𝑡 são considerados uma transformação linear dos choques economicamente fundamentais 𝜀𝑡 𝑢𝑡 𝐶𝜀𝑡 𝐶𝐶 𝑉 321 Especificação e estimação do modelo VAR Conforme descrito em Cavalcanti 2013 o primeiro passo é a estimação de um modelo VAR na forma reduzida incluindo como variáveis endógenas além da medida de eficiência outras variáveis domésticas e específicas das instituições financeiras que de acordo com a teoria econômica seriam possíveis candidatas a explicar a evolução da eficiência bancária Como a inclusão simultânea de todas as possíveis variáveis relevantes não é viável por conta dos graus de liberdade na estimação foram testados modelos com diferentes subconjuntos de variáveis Para esse exercício na especificação do modelo utilizamos dimensões econômicas complementares entre si tanto em nível macroeconômico quanto no nível das instituições financeiras Especificamente além da variável que mede o grau de eficiência bancária utilizamos as seguintes variáveis endógenas no VAR a Índice Nacional de Expectativa do Consumidor INEC um indicador criado pela Confederação Nacional da Indústria que sintetiza a opinião dos brasileiros capturada em pesquisa de opinião pública sobre alguns 44 aspectos capazes de afetar as suas decisões de consumo CNI CONFEDERAÇÃO NACIONAL DA INDÚSTRIA 2010 v 2 p 3 Seus resultados variam de 0 a 100 e quanto maior mais confiante é a opinião da população b Risco Brasil ou EMBI Brasil é um índice que avalia a diferença dos retornos financeiros obtidos a cada dia por uma carteira selecionada de títulos brasileiros e a taxa de retorno oferecida por títulos emitidos pelo Tesouro americano Essa diferença é chamada de spread ou spread soberano Quanto maior esse índice maior é a percepção de risco dos investidores O índice é apurado pelo JPMorgan IPEAData 2020 c Produto Interno Bruto PIB oscilação entre os trimestres dos dados do Boletim da Seção de Atividade Econômica estimado mensalmente pelo Banco Central IPEAData 2020 d Índice Geral de Preços do Mercado IGPM registra o ritmo evolutivo de preços como medida síntese da inflação nacional É composto pela média ponderada do Índice de Preços por Atacado IPA 60 Índice de Preços ao Consumidor IPCFGV 30 e Índice Nacional de Preços da Construção Civil INCC 10 e compreende o período entre os dias 20 do mês de referência e 21 do mês anterior É apurado pela área de Conjuntura Econômica da Fundação Getulio Vargas FGV Os dados foram acumulados para representar o trimestre IPEAData 2020 e Índice de Basileia IB índice que representa a relação entre o Patrimônio de Referência3 e os Ativos ponderados pelo risco4 Riskweighted asset RWA Esses dados são divulgados trimestralmente pelo Banco Central do Brasil BANCO CENTRAL DO BRASIL 2020 Além das variáveis domésticas acima descritas outras especificações alternativas foram testadas com variáveis como Concessão de Crédito Inadimplência População Residente Salário Mínimo Taxa de Câmbio Nominal 3 Capital regulatório formado pela soma das parcelas de Capital Nível I e Capital Nível II O Capital Nível I compreende a soma da parcela do capital de melhor qualidade e imediatamente disponível para absorver perdas também denominado Capital Principal e instrumentos de capital e dívida perpétuos elegíveis como patrimônio regulatório ou Capital Complementar O Capital Nível II compreende a parcela do capital composta por instrumentos subordinados elegíveis como patrimônio regulatório aptos a absorver perdas durante o funcionamento da instituição BANCO CENTRAL DO BRASIL 2020 4 Soma da exposição aos riscos de Crédito Mercado e Operacional BANCO CENTRAL DO BRASIL 2020 45 e Real e Taxa de Juros SELIC Foram também testadas variáveis específicas das instituições financeiras como Quantidade de Agências e Postos de Atendimento PA Índice de Imobilização e Retorno sobre o Capital Como próximo passo da identificação do modelo conforme descreveu Cavalcanti 2010 é necessário definir um método de identificação do fator exógeno de cada uma das variáveis sendo assim possível estimar o efeito de um choque em uma das variáveis sobre as demais Ele afirma que o método mais utilizado é a ortogonalização dos resíduos do VAR através da chamada decomposição de Cholesky que impõe uma estrutura recursiva à matriz de relações contemporâneas entre as variáveis do modelo de modo que a primeira variável não seja afetada contemporaneamente por nenhuma das demais a segunda seja afetada apenas pela primeira a terceira seja afetada pelas primeiras duas e assim por diante Ao pesquisador cabe apenas selecionar a ordenação causal adequada das variáveis sob análise CAVALCANTI 2010 v 14 n 2 p 252 Lütkepohl 2005 descreveu esse processo como ordenação causal instantânea das variáveis que por ter sida defendida por Herman Wold é também chamada de Causalidade de Wold Para ele estabelecer essa ordem pode ser um exercício bastante difícil na prática p 61 tradução nossa Cavalcanti 2013 endossa essa afirmação classificando essa etapa como uma das mais difíceis e controversas Na sua opinião a principal dificuldade desse método de identificação reside na apresentação de argumentos convincentes em favor de uma estrutura recursiva com certa ordenação causal específica n 18 p 80 De acordo com os argumentos a seguir a ordenação causal adotada para esse estudo de acordo com a razoabilidade econômica foi a INEC é a variável que ajuda a antecipar variações no ritmo da atividade econômica portanto tendem a ser um indicador antecedente das variações no consumo das famílias e por conseguinte do nível de atividade da economia CNI 2019 v20 Por essas características pode 46 ser classificada como uma variável de resposta rápida ou seja deve preceder as demais variáveis b Risco Brasil como os desempenhos registrados pelo índice ajudam os investidores a formar opinião e a planejar suas decisões futuras IPEAData 2020 ele também contribui para o nível futuro de atividade da economia Como também é resultado das percepções dos agentes assim como o INEC também pode ser classificado como uma variável de resposta rápida Contudo como ele mede apenas a oscilação de preços dos títulos praticados de um dia para o outro IPEAData 2020 como uma variável real deve vir após o INEC c PIB como é razoável supor que as variáveis INEC e Risco Brasil afetem as decisões de produção e que haja uma defasagem entre esse momento e o momento em que a produção efetivamente ocorra que é o que o PIB mede portanto ele deve suceder essas duas variáveis d IGPM como a inflação é uma variável que pode oscilar rapidamente em função de variações no ambiente econômico e é razoável supor que responda contemporaneamente aos choques nas demais variáveis logo deve ser a última variável econômica e Índice de Basileia Considerando a representação dos bancos na economia nacional apesar de um dos setores mais relevantes variáveis nacionais devem preceder as variáveis específicas pois não são por estas afetadas Conforme descrito anteriormente outras especificações alternativas foram testadas nesse caso invertendose a ordenação causal com o objetivo de verificar a robustez dos resultados obtidos Além das etapas descritas acima é fundamental testarmos a estacionariedade da série pois segundo Bueno 2011 esse é o principal conceito que se deve ter ao estimar uma série temporal Gujarati e Porter 2011 reforçam esse ponto ao afirmar que o trabalho empírico baseado nos dados de séries temporais supõe que a série temporal subjacente seja estacionária p 731 grifo do autor Utilizaremos para tal o teste de raiz unitária que segundo os dois últimos autores se tornou amplamente utilizado nos últimos anos Dentre diversas opções optamos por aquele que foi definido por Elder e Kennedy 2001 como 47 o mais popular de muitos testes concorrentes na literatura v 32 n 2 p 138 tradução nossa o teste de DickeyFuller aumentado augmented DickeyFuller ADF Conforme descrito por Gujarati e Porter 2011 nele é possível adotarmos três variedades 1 um passeio aleatório puro 2 um passeio aleatório com deslocamento e 3 um passeio aleatório com deslocamento e tendência p 753 Eles também afirmam que se a série temporal original for não estacionária muito frequentemente suas primeiras diferenças tornamse estacionárias p 445 o que nos foi útil para algumas variáveis A próxima etapa da estimação do modelo foi a seleção da ordem de defasagens das variáveis endógenas do VAR Lütkepohl 2005 afirma que escolher uma ordem de defasagem desnecessariamente grande reduzirá a precisão do modelo e consequentemente da estimativa das respostas ao impulso Para tal foram realizados os critérios mais utilizados nos trabalhos aplicados e selecionada a ordem que mais vezes resultou ou seja que teve maioria Conforme descrito em Ivanov e Kilian 2005 e Lütkepohl 2005 os critérios mais utilizados são o erro de predição final final prediction error FPE o critério de informação de Akaike Akaikes information criterion AIC o critério de informação de Hannan e Quinn Hannan and Quinn information criterion HQIC e o critério de informação bayesiano de Schwarz Schwarzs Bayesian information criterion SBIC Bueno 2011 menciona os três últimos AIC HQIC e SBIC como os principais critérios de informação e acrescenta que seu uso seria uma forma de encontrar o modelo com a quantidade ideal de parâmetros Ele afirma que como os critérios associam uma penalidade ao aumento no número de regressores que por sua vez diminuem a soma dos resíduos eles são a maneira para avaliar se o regressor adicional traz mais benefícios do que custos 48 322 Checagem do Modelo Para nos utilizarmos do que Lütkepohl 2005 descreveu como ferramentas estatísticas para verificar as propriedades de autocorrelação dos resíduos de um determinado modelo VAR p 157 tradução nossa realizaremos o teste de multiplicador de Lagrange Lagrange Multiplier LM ou também chamado teste de BreuschGodfrey para testar as hipóteses nulas de ausência de autocorrelação residual para as ordens de defasagem até a ordem selecionada nos testes mencionados acima Outra checagem relevante é a estabilidade do modelo VAR Segundo afirmou Baum 2013 só é possível analisar as funções impulso resposta e a decomposição da variância se o modelo estimado for estável ou seja não possuir comportamento explosivo Para tal analisamos o comportamento das raízes inversas do polinômio característico do sistema VAR Conforme descreveu Biage Correa e Neder 2008 se todas as raízes encontraremse dentro do círculo ou seja terem módulo menor do que 1 o sistema é estável se se alguma for maior do que 1 o sistema é instável ou seja comportamento explosivo e se for igual a 1 o sistema é nãoestacionário Enders 2014 descreveu essa análise de forma simplificada afirmando que uma maneira sucinta de caracterizar as condições de estabilidade é afirmar que as raízes características devem estar dentro do círculo unitário 4th ed p 29 tradução nossa 323 Análise Estrutural Conforme descrito por Lütkepohl 2005 os modelos VAR são muito utilizados para análise dos relacionamentos entre as variáveis de interesse pois representam as correlações entre um conjunto de variáveis Stock e Watson 2001 afirmaram que o kit de ferramentas padrão na análise VAR é a análise de causalidade de Granger as funções impulso resposta e a decomposição da 49 variância dos erros de previsão Outra abordagem interessante dada pelos autores é que por conta da complexa dinâmica dos modelos VAR esses itens de sua análise são mais informativos que os coeficientes de regressão estimados ou as estatísticas R² 3231 Causalidade de Granger Assim como vários autores como Berger e DeYoung 19975 que tiveram sua abordagem replicada por Williams 2004 e foram citados por Fiordelisi MarquesIbanez e Molyneux 2011 e tantos outros que abordaram o tema eficiência mencionados na seção 2 nós também utilizaremos as técnicas de causalidade de Granger Porém diferentemente desses nós utilizaremos a análise de Granger como uma das ferramentas analíticas no escopo do modelo VAR Conforme mencionado por Berger e DeYoung 1997 e Fiordelisi Marques Ibanez e Molyneux 2011 como essa técnica produz associações estatísticas grosseiras não geraremos prova de causalidade econômica mas como as hipóteses testadas são relações únicas ordenadas por tempo e firmadas entre pares de variáveis ela poderá indicar quais hipóteses são consistentes ou inconsistentes O segundo estudo ainda destaca que apesar desse método ter várias limitações tem sido amplamente utilizado na literatura econômica e nos estudos bancários mencionando os trabalhos acima citados Outros estudos mais recentes já citados aqui também adotaram a mesma técnica como Tan e Anchor 2017 Casu e Girardone 2009 por exemplo mencionaram que a extensão da metodologia original de Granger aos dados do painel permite o uso de séries temporais para testar qualquer relação causal entre duas variáveis v 105 p 5Os autores mencionaram que a aplicação da causalidade de Granger na microeconomia bancária era incomum até então porém não inédita mencionando um trabalho de BERGER Allen N The Relationship between Capital and Earnings in Banking Journal of Money Credit and Banking v 27 n 2 p 432456 1995 50 135 tradução nossa que é o que faremos Lütkepohl 2005 afirmou que em condições adequadas a combinação dessa técnica com modelos VAR é bastante fácil de lidar Berger e DeYoung 1997 descrevem a causalidade de Granger como uma técnica que tenta determinar se a variável 𝑌𝑖 Grangercausa a variável 𝑌𝑗 testando se o histórico passado de 𝑌𝑖 melhora a previsão de 𝑌𝑗 depois de ter considerado o histórico passado de todos os outros 𝑌𝑠 Eles especificam o modelo da seguinte forma 𝑌𝑖𝑡 𝑓1𝑌𝑖𝑙𝑎𝑔 𝑋𝑖𝑙𝑎𝑔 𝑍𝑖𝑙𝑎𝑔 𝑊𝑖𝑙𝑎𝑔 𝐾𝑖𝑙𝑎𝑔 𝑉𝑖𝑙𝑎𝑔 𝑢𝑖𝑡 6 Onde 𝑌𝑖𝑡 é a variável dependente no nosso caso o escore de eficiência sendo 𝑖 a instituição financeira e 𝑡 o trimestre As variáveis do lado direito da equação consideram valores defasados das variáveis dependentes 𝑌 𝑋 𝑍 𝑊 𝐾 e 𝑉 no nosso caso escore de eficiência INEC Risco Brasil PIB IGPM e Índice de Basileia como é o procedimento para análises de causalidade de Granger Lütkepohl 2005 descreve de forma muito similar afirmando que se 𝑌𝑗 pode ser previsto com mais eficiência se as informações no processo 𝑌𝑖 forem levadas em consideração além de todas as outras informações no universo então 𝑌𝑖 Grangercausa para 𝑌𝑗 Ele afirmou que como a causalidade de Granger pode não explicar as interações entre as variáveis de um sistema e que em qualquer estudo pode ser necessário conhecer a resposta de uma variável a um impulso em outra variável em um sistema que envolve várias variáveis podemos recorrer a Análise de Impulso Resposta 3232 Função ImpulsoResposta Segundo Lütkepohl 2005 pode ser útil conhecer o comportamento de uma variável a um impulso em outra variável em um sistema com várias variáveis Em outras palavras como uma variável se comporta a um impulso em outra variável Ele complementa que se houver uma reação a esse impulso podemos 51 supor que há um tipo de causalidade como um efeito de um choque ou inovação exógena em uma algumas ou até mesmo em todas as variáveis do modelo Para tal análise utilizaremos a Função ImpulsoResposta Impulse Response Function IRF que segundo Cavalcanti 2013 é a descrição dos efeitos de variações inesperadas nas variáveis do sistema sobre a trajetória de cada variável ao longo do tempo p 81 Baum 2013 a define de forma muito similar afirmando que elas medem os efeitos de um choque e esclarece que se o modelo VAR for estável subseção 322 podemos reescrever o modelo na forma de média móvel 𝑌𝑡 𝜇 𝐷𝑖𝑋𝑡𝑖 𝑖0 𝜙𝑖𝑈𝑡𝑖 𝑖0 7 Para ele essa equação acima é a representação da média móvel vetorial vector moving average VMA e demonstra que todos os valores passados de 𝑌𝑡 foram substituídos Ele denomina as matrizes 𝐷𝑖 como as funções multiplicadores dinâmicas ou funções de transferência6 e a sequência de coeficientes de média móvel 𝜙𝑖 são as funções de impulso resposta no horizonte 𝑖 Conforme descrito na subseção 321 deste estudo como os resíduos do modelo proposto foram ortogonalizados através da decomposição de Cholesky e de acordo com a proposição de Sims 1980 as IRFs submetidas a essa escolha são conhecidas como IRFs ortogonalizadas orthogonalized IRFs OIRFs Baum 2013 Como nosso intuito será analisalas de forma cumulativa faremos uso do conceito de OIRF cumulativas cumulative OIRF COIRF Vale destacar conforme o estudo de Kim 2019 no qual nos baseamos que há uma abordagem alternativa e que não estaria sujeita ao problema da ordenação das variáveis a função generalizada de resposta ao impulso Generalized Impulse Response Function GIRF Contudo apesar de ter sido empregada por muitas pesquisas o próprio autor conclui que lhe parece mais razoável a OIRF 6 Bueno 2011 denomina esses elementos da matriz como multiplicadores de impacto de um choque sobre as variáveis endógenas p 217 52 3233 Análise da Decomposição da Variância dos Erros de Previsão Outra importante análise é a decomposição de variância dos erros de previsão forecast error variance decomposition FEVD descrita por Cavalcanti 2013 como uma opção que permite quantificar a relevância que em média os choques em cada variável do sistema possuem na explicação das flutuações dessas variáveis em torno de seus valores previstos p81 Para Baum 2013 ela mensura o quanto percentualmente da variação do erro de previsão de uma variável no caso endógena pode ser atribuída através de choques ortogonais para outra variável endógena ou para ela mesma Adicionalmente conforme descrito no item acima sobre a Função ImpulsoResposta a FEVD também é sensível ao problema da ordenação Baum 2013 Enders 2014 afirmou que a análise de impulso e a decomposição de variância denominadas conjuntamente de contabilidade da inovação podem ser ferramentas úteis para examinar as relações entre variáveis econômicas p 302 tradução nossa Por isso optase por incluir a análise do FEVD neste estudo de maneira complementar às funções de impulsoresposta na medida em que quantificam a importância relativa de cada variável do sistema na explicação das flutuações futuras da variável de eficiência bancária 324 Previsão Conforme Figura 1 acima um modelo VAR tem como principais usos a análise estrutural subseção 323 e a previsão conforme descrito por Lütkepohl 2005 Bueno 2011 afirmou que quando se conhece o processo gerador de dados no nosso caso o modelo VAR estimado a previsão ℎ passos à frente é descrita como 𝐸𝑋𝑡ℎ𝐼𝑡 𝑋𝑡ℎ𝑡 𝜙1𝑋𝑡ℎ1𝑡 𝜙2𝑋𝑡ℎ2𝑡 𝜙𝑝𝑋𝑡ℎ𝑝𝑡 8 53 em que 𝑋𝑡𝑗𝑡 𝑋𝑡𝑗 para 𝑗 0 E como temos que 𝑋𝑡 é estacionário podemos transformálo em um modelo de médias móveis infinito 𝑋𝑡ℎ 𝐼 𝜙𝑗𝐿𝑗 𝑝 𝑗1 1𝑒𝑡ℎ 𝑒𝑡ℎ Ψ1𝑒𝑡ℎ1 Ψ𝑝𝑒𝑡ℎ2 9 Logo podemos reescrever a previsão correspondente como 𝑋𝑡ℎ𝑡 Ψ𝑗𝑒𝑡ℎ𝑗 𝑗ℎ 10 De acordo com Stock e Watson 2001 analisamos o desempenho do modelo VAR com pseudoprevisões fora da amostra ou seja estimamos o modelo até um período anterior aos últimos dados utilizados para que a previsão possa ser comparada com dados reais Adicionalmente assim como eles também rodamos pseudoprevisões para uma autorregressão AR com quatro defasagens ou seja uma regressão da variável de eficiência em defasagem de seus quatro últimos valores passados Conforme sugestão de Enders 2014 para mensurar a precisão da previsão e comparar os métodos VAR e AR utilizamos o teste de DieboldMariano7 Esse teste calcula uma medida de precisão preditiva proposta pelos autores de mesmo nome 1995 De forma sucinta conforme descreveu Baum 2003 dada uma série de dados real e duas previsões aplicase um critério de perda nós utilizaremos três conforme descrição abaixo e calculase medidas de precisão preditiva que permitem que a hipótese nula de igual precisão seja testada Isso se dá através do teste de que a diferença média entre os critérios de perda para as duas previsões é zero Os três critérios de perda utilizados foram erro médio absoluto mean absolute error MAE erro quadrático médio mean square error MSE e erro percentual médio absoluto mean absolute percentage error MAPE Como nenhum deles é perfeito e apresenta vantagens e desvantagens utilizaremos os três como referência de análise Kennedy 2008 os descreve da seguinte maneira 7 Diebold e Mariano 1995 complementam que seu teste de comparação da previsão seria um dos muitos diagnósticos que devem ser examinados ao comparar modelos p 262 tradução nossa 54 Erro médio absoluto também conhecido como desvio médio absoluto é a média dos valores absolutos dos erros de previsão e é uma medida sensível ao dimensionamento Erro quadrático médio é a raiz quadrada da média dos valores quadrados dos erros de previsão também é sensível ao dimensionamento e além disso é sensível à outliers É segundo o autor8 a medida mais usada Erro percentual médio absoluto é a média dos valores absolutos dos erros percentuais e diferentemente das demais não é sensível ao dimensionamento porém apresenta outras desvantagens como a base para medir a porcentagem pode ser zero ou muito grande tornandoa indefinida ou distorcida e penaliza mais os erros positivos do que os negativos 33 Dados e amostra Nosso exercício para levantamento dos dados pode ser dividido em duas frentes 1 os dados utilizados para a análise de fronteira estocástica específicos de cada instituição financeira que serviram como referência para apuração dos escores de eficiência e 2 os dados utilizados para investigação das inter relações entre eficiência e outras variáveis ambos disponíveis no Apêndice A Os dados utilizados na primeira frente os específicos dos bancos foram extraídos do IFData portal do Banco Central do Brasil que disponibiliza trimestralmente informações das instituições que autoriza a funcionar e que estejam em operação normal Para a segunda frente além dos dados já levantados para a primeira e os dados de eficiência resultantes desse exercício as variáveis macroeconômicas foram capturadas do Ipeadata a base de dados macroeconômicos financeiros e regionais do Brasil mantida pelo Instituto de Pesquisa Econômica Aplicada IPEA 8 Lütkepohl 2005 reforça essa afirmação dizendo que os preditores que minimizam os erros quadráticos médios de previsão são os mais utilizados p 33 tradução nossa 55 Utilizamos dados trimestrais recentes dos cinco maiores bancos brasileiros9 Itaú Banco do Brasil Caixa Bradesco e Santander para o período de 2009 e 2019 Além destes foram criadas três séries de dados uma média ponderada entre esses cinco bancos aqui denominada Geral outra entre os três bancos privados Itaú Unibanco Bradesco e Santander aqui denominada Privados e outra entre os dois bancos públicos Banco do Brasil e Caixa Econômica Federal aqui denominada Públicos O intuito foi possibilitar a realização de mais comparações além dos bancos entre si Estas séries foram elaboradas conforme critério utilizado pelo Banco Central do Brasil 2017 ao estabelecer a sua segmentação das instituições financeiras ou seja o conceito de Exposição total10 Tentando não perpetuar o problema abordado por DeYoung 1997 de que os estudos existentes incluem poucas justificativas estatísticas para suas escolhas de muitas ou poucas observações de séries temporais nossa decisão por partir de 2009 não foi arbitrária Nós não objetivávamos avaliar nesse estudo a crise de 2008 a qual imaginávamos que afetaria nossa análise Os dados da Matriz de correlação das variáveis do modelo VAR disponível no Apêndice B se olharmos somente as correlações de cada uma das oito amostras ou seja sem analisar a correlação entre as variáveis de diferentes instituições financeiras a correlação mais alta foi de 056 e a mais baixa de 040 indicando que as variáveis em questão não apresentaram alta correlação sugerindo que não há problema de colinearidade De acordo com Kennedy 2008 a multicolinearidade é um problema quando a correlação está acima de 080 9 De acordo com a segmentação do Banco Central BANCO CENTRAL DO BRASIL 2017 os bancos classificados como S1 são aqueles que i tenham porte ExposiçãoProduto Interno Bruto superior a 10 ou ii exerçam atividade internacional relevante ativos no exterior superiores a US 10 bilhões Para efeito desse estudo como buscamos analisar e comparar a eficiência bancária Banco BTG Pactual foi a única instituição desconsiderada dentre os classificados como S1 data base Dez19 Além de não atender o item i da classificação ele não possui uma rede de agências ampla o que inviabilizaria as comparações 10 Disponíveis no IFData sem ponderação de risco conforme definido na Circular 3748 de 27 de fevereiro de 2015 56 As estimativas da primeira e da segunda frente a modelagem e todos os testes foram feitos através do software Stata 142 e a estatística descritiva das variáveis selecionadas está na Tabela 1 abaixo Tabela 1 Estatística descritiva das variáveis Fonte Elaborado pelo próprio autor Nota O prefixo dif diz respeito ao operador de primeira diferença Por conta dos resultados dos testes onde algumas variáveis apresentaram não estacionaridade conforme Bueno 2011 descreveu bastou a primeira diferença para estacionarizálas Isso ocorreu para os Índices de Basileia e o Risco Brasil Os resultados desses testes estão disponíveis no Apêndice C Tipo Indicação Média Desvio Padrão Mínimo Máximo Itaú ITAUcosteffic 0715 0016 0678 0744 BB BBcosteffic 0735 0022 0685 0781 Bradesco BRADESCOcosteffic 0724 0020 0682 0770 Caixa CAIXAcosteffic 0725 0052 0548 0785 Santander SANTANDERcosteffic 0724 0022 0669 0767 Privados PRIVcosteffic 0720 0016 0684 0751 Públicos PUBcosteffic 0729 0036 0611 0783 Geral GERALcosteffic 0724 0022 0654 0763 Itaú difITAUbasileia 0000 0009 0022 0018 BB difBBbasileia 0001 0007 0022 0019 Bradesco difBRADESCObasileia 0000 0010 0031 0023 Caixa difCAIXAbasileia 0000 0011 0027 0024 Santander difSANTANDERbasileia 0002 0014 0030 0051 Privados difPRIVbasileia 0001 0008 0015 0015 Públicos difPUBbasileia 0000 0008 0025 0020 Geral difGERALbasileia 0000 0006 0012 0012 INEC 0610 0056 0504 0723 difRiscoBrasil 0000 0006 0015 0013 PIB 0020 0037 0073 0079 IGPM 0014 0015 0027 0039 Específicas Domésticas Variável Escore de Eficiência de custo Índice de Basileia Índice Nacional de Expectativa do Consumidor Risco Brasil Produto Interno Bruto Índice Geral de Preços do Mercado 57 4 RESULTADOS E DISCUSSÃO 41 Resultados Modelo VAR Começamos essa seção apresentando os resultados do Modelo VAR para as oito séries de dados cinco bancos e as três séries criadas Conforme descrito na seção anterior como etapa inicial nesse processo foram testadas a estacionaridade das séries através do teste de ADF Os resultados nas três variedades disponíveis no Apêndice C apresentaram resultado de que as séries das variáveis selecionadas são estacionárias com deslocamento Seguindo as etapas descritas na Metodologia o próximo passo foi a seleção da ordem de defasagens das variáveis endógenas do VAR Para todos os modelos na maioria dos critérios quando não na totalidade a quarta ordem de defasagem foi a que apresentou os melhores resultados o que seria intuitivo e razoável para dados trimestrais O estudo de Ivanov e Kilian 2005 reforça esse resultado pois a maioria dos estudos empíricos com levantamento de dados trimestrais a ordem de defasagens encontrada foi de quatro períodos Esses resultados podem ser observados na Tabela 2 abaixo Tabela 2 Critérios de seleção da ordem de defasagens Fonte Elaborado pelo próprio autor Nota Os testes consideraram o nível de significância de 5 Após a especificação e estimação dos modelos foi realizada sua checagem através do teste LM Lagrange Multiplier para verificar a ausência de Estatística Lag Estatística Lag Estatística Lag Estatística Lag Itaú 36E24 4 385202 4 362203 4 327502 0 Banco do Brasil 14E24 4 394435 4 394435 4 336314 1 Bradesco 22E23 4 367112 4 344113 4 318718 1 Caixa 14E23 4 372002 4 349003 4 316686 1 Santander 16E23 4 370092 4 347093 4 319436 0 Geral 69E25 4 400266 4 377310 4 336556 1 Privados 42E25 4 405372 4 382415 4 341388 4 Públicos 24E24 4 387812 4 364856 4 333312 1 Série FPE AIC HQIC SBIC 58 autocorrelação residual Embora tenha rejeitado a hipótese nula em alguns lags o modelo não apresenta fortes evidências de presença de autocorrelação serial Os resultados estão disponíveis no Apêndice D Os gráficos de correlograma cruzado Apêndice E corroboram com essa afirmação Outra checagem realizada foi a estabilidade dos modelos VAR checando se o modelo não possuía comportamento explosivo ou seja se as raízes características estão dentro do círculo unitário Foi encontrada estabilidade para os oito modelos e os resultados estão disponíveis no Apêndice F 42 Análises Estruturais 421 Análises de causalidade de Granger Demos sequência aos modelos VAR através das Análises Estruturais mais especificamente na análise de causalidade de Granger nas funções impulso resposta e na decomposição da variância dos erros de previsão Vale lembrarmos que os testes de causalidade de Granger testarão a hipótese nula de que os coeficientes estimados dos valores defasados das variáveis do modelo para os dados analisados são conjuntamente zero Ou seja se não rejeitarmos a hipótese nula não rejeitaremos a hipótese de que a variável excluída representada na tabela como b não Grangercausa a variável da equação representada na tabela como a De modo a facilitar a interpretação e visualização dos resultados as probabilidades da Tabela 3 abaixo destacadas em verde expressam os casos em que rejeitamos a hipótese nula ou seja há evidências estatísticas de que as variáveis da coluna b Grangercausa as variáveis da coluna a De forma análoga as destacadas em vermelho não Grangercausam Os resultados revelaram que o INEC Grangercausa a Eficiência de Custo para a maioria dos bancos sendo o Banco do Brasil e a média ponderada dos bancos públicos as exceções O efeito descrito por Gujarati e Porter 2011 como 59 causalidade bilateral ou feedback também é percebido nessa relação ou seja a Eficiência de Custos também Grangercausa o INEC para a maioria dos bancos sendo o Bradesco e novamente os bancos públicos as exceções Os dados também revelaram que o Risco Brasil Grangercausa a Eficiência de Custos para o Itaú Bradesco Caixa e a média ponderada dos bancos privados mas não Grangercausa para Banco do Brasil Santander a média ponderada dos públicos e a média ponderada geral E o efeito feedback foi descartado nas cinco amostras Foi possível também evidenciar que o PIB Grangercausa a Eficiência de Custos e que a Eficiência de Custos Grangercausa o PIB com exceção do Banco do Brasil e da média ponderada geral dos bancos O IGPM apresentou evidências de que Grangercausa a Eficiência de Custos para Itaú Bradesco Santander e as três médias ponderadas da nossa análise mas não Granger causa para os dois bancos públicos o Banco do Brasil e a Caixa O efeito feedback nesse caso também evidenciou para a maioria dos bancos que a Eficiência de Custos não Grangercausa o IGPM com exceção da Caixa e da média ponderada dos bancos públicos O Índice de Basileia não Grangercausa a Eficiência de Custos para Itaú Banco do Brasil Bradesco e as médias ponderadas dos privados e a geral mas Grangercausa para Caixa Santander e a média ponderada dos públicos já o efeito feedback nesse caso se mostrou verdadeiro para as oito amostras ou seja a Eficiência de Custos Grangercausa o Índice de Basileia Os demais resultados dos testes de causalidade de Granger como não foram foco desse estudo estão disponíveis no Apêndice G 60 Tabela 3 Probabilidades do teste de causalidade de Granger para Eficiência de Custos Fonte Elaborado pelo próprio autor Notas Os números expressos na tabela representam a probabilidade de aceitar a hipótese nula sendo que a fonte verde representa os casos em que rejeitamos e em vermelho os casos em que aceitamos Consideramos o nível de significância de 5 422 Análises das Funções ImpulsoResposta Seguindo a análise dos modelos VAR estimamos as funções impulsoresposta a partir dos coeficientes estimados Como descrito acima nosso foco está na resposta do escore de eficiência de cada Instituição Financeira aos impulsos também chamados de choques nas demais variáveis do modelo INEC Risco Brasil PIB IGPM e Basileia isoladamente mantendo todas as demais constantes As Figuras 2 3 4 5 e 6 respectivamente Itaú Banco do Brasil Bradesco Caixa Santander na subseção Análise Individual dos cinco grandes bancos e 7 8 e 9 respectivamente Privados Públicos e Geral na subseção Análise das médias ponderadas mostram essas funções ortogonalizadas cumulativas ao longo do período considerando oito trimestres após o choque11 11 Foi considerada significância de 68 indicada na legenda do gráfico como Intervalo de Confiança IC para todas as demonstrações das Funções ImpulsoResposta Equação a Excluído b Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral Eficiência de Custo INEC 0005 0487 0000 0000 0000 0000 0104 0000 INEC Eficiência de Custo 0003 0032 0285 0017 0000 0000 0146 0003 Eficiência de Custo Risco Brasil 0014 0876 0000 0041 0500 0001 0561 0053 Risco Brasil Eficiência de Custo 0671 0105 0078 0288 0060 0118 0538 0782 Eficiência de Custo PIB 0000 0000 0000 0000 0000 0000 0000 0000 PIB Eficiência de Custo 0000 0539 0000 0004 0006 0002 0007 0128 Eficiência de Custo IGPM 0013 0916 0009 0091 0000 0001 0023 0000 IGPM Eficiência de Custo 0570 0353 0905 0000 0976 0561 0000 0108 Eficiência de Custo Basileia 0522 0660 0287 0000 0001 0193 0006 0714 Basileia Eficiência de Custo 0000 0000 0002 0046 0005 0000 0002 0011 Eficiência de Custo Todas 0000 0000 0000 0000 0000 0000 0000 0000 61 De forma sucinta o eixo das abcissas mostra a quantidade de trimestres após o choque e o eixo das ordenadas a resposta cumulativa em pontos percentuais do choque de um desviopadrão 4221 Análise individual dos cinco grandes bancos Figura 2 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Itaú Fonte elaborado pelo próprio autor Dos resultados do Itaú inferese pela Figura 2 acima que o INEC e o próprio escore de eficiência afetam positivamente os movimentos futuros do escore de eficiência mantendo tudo o mais constante Já as respostas do escore de eficiência aos choques sobre o IGPM PIB e Basileia também foram estatisticamente diferentes de zero mas negativas nos três casos exceção feita ao efeito no último trimestre no caso 8º para o PIB e no 1º trimestre para Basileia onde os efeitos não foram estatisticamente diferentes de zero Para o 62 Risco Brasil com exceção do primeiro trimestre onde a resposta do escore de eficiência foi negativa para os demais trimestres não apresentou relação estatisticamente diferente de zero Figura 3 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Banco do Brasil Fonte elaborado pelo próprio autor Para os resultados do Banco do Brasil podemos inferir pela Figura 3 acima que o Risco Brasil o PIB o IGPM e Basileia não apresentaram relação estatisticamente diferente de zero com o escore de eficiência A reposta do escore de eficiência ao choque no INEC só foi estatisticamente diferente de zero a partir do 7º trimestre sendo negativa nesse caso Diferentemente das demais variáveis choque do próprio escore de eficiência afeta positivamente os seus movimentos futuros mantendo tudo o mais constante da mesma forma que o resultado encontrado nessa relação para o Itaú 63 Figura 4 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Bradesco Fonte elaborado pelo próprio autor No caso do Bradesco na Figura 4 acima os resultados mostram que o Risco Brasil e Basileia assim como no Banco do Brasil não apresentaram relação estatisticamente diferente de zero com o escore de eficiência O mesmo vale para o IGPM com exceção do 1º trimestre que apresentou relação negativa Já o choque no PIB com exceção do 1º trimestre onde o resultado não foi estatisticamente diferente de zero afetou negativamente os movimentos futuros do escore de eficiência Para o INEC e o próprio escore de eficiência os choques afetaram positivamente sendo que no caso do INEC esse efeito só foi estatisticamente diferente de zero até o 4º trimestre 64 Figura 5 Função Impulsoresposta das diversas variáveis ao escore de eficiência da Caixa Fonte elaborado pelo próprio autor Já os resultados da Caixa conforme Figura 5 acima mostram que o PIB assim como no Banco do Brasil e o IGPM não apresentaram relação estatisticamente diferente de zero com o escore de eficiência com exceção do 1º trimestre para o IGPM onde o efeito foi positivo No caso do Risco Brasil o choque nos movimentos futuros do escore de eficiência só foi estatisticamente diferente de zero nos 2º e 3º trimestres tendo efeito positivo Para o próprio escore de eficiência e para Basileia os efeitos dos choques no escore de eficiência foram positivos em todos os trimestres No efeito do INEC nos movimentos futuros do escore de eficiência houve uma oscilação nos três primeiros trimestres os efeitos foram positivos e no 8º foi negativo Nos demais trimestres os efeitos não foram estatisticamente diferentes de zero 65 Figura 6 Função Impulsoresposta das diversas variáveis ao escore de eficiência do Santander Fonte elaborado pelo próprio autor Como último banco dessa etapa da análise conforme Figura 6 acima os resultados do Santander mostram que o os choques no INEC e na Basileia afetaram negativamente em todos os trimestres os movimentos futuros do escore de eficiência Já os choques no Risco Brasil e no PIB afetaram positivamente o escore de eficiência sendo que os resultados não foram estatisticamente significativos somente para o 1º trimestre no caso do Risco Brasil e somente no 3º trimestre no caso do PIB Para o IGPM com exceção do 1º trimestre que apresentou resultado negativo no escore de eficiência os demais trimestres não foram estatisticamente diferentes de zero Sobre o efeito do choque do próprio escore de eficiência sobre ele mesmo assim como todos os bancos foi positivo 4222 Análise das médias ponderadas dos públicos privados e geral 66 Figura 7 Função Impulsoresposta das diversas variáveis ao escore de eficiência da média ponderada dos Privados Fonte elaborado pelo próprio autor Ao compararmos os resultados dessa média ponderada dos bancos privados com os resultados individuais das instituições financeiras que o compõe podemos destacar que um choque no INEC afetou os movimentos futuros da média ponderada do escore de eficiência positivamente nos primeiros trimestres como no Itaú e Bradesco e negativamente no último como no Santander No caso do Risco Brasil o efeito foi negativo no início como no Itaú e positivo no 8º trimestre como no Santander Para o PIB o início foi positivo como no Santander e negativo no 3º e 7º trimestres como no Itaú e Bradesco No caso do IGPM o choque foi estatisticamente diferente de zero somente para o 2º e 5º trimestres com efeito negativo assim como o Itaú que contudo teve efeito em todos O choque em Basileia afetou o escore de eficiência de forma negativa similar ao Itaú e Santander 67 Figura 8 Função Impulsoresposta das diversas variáveis ao escore de eficiência da média ponderada dos Públicos Fonte elaborado pelo próprio autor Já na comparação dos resultados da média ponderada dos bancos públicos com os resultados do Banco do Brasil e da Caixa notamos que um choque no INEC afetou os movimentos futuros da média ponderada do escore de eficiência positivamente nos primeiros trimestres como na Caixa No Risco Brasil o efeito foi positivo no 1º trimestre e estatisticamente não diferente de zero nos demais parecido com o movimento da Caixa Para o PIB diferentemente do Banco do Brasil e da Caixa onde ambos apresentaram resultados estatisticamente não diferentes de zero o efeito foi positivo a partir do 6º trimestre De similar o IGPM só apresentou o resultado do 1º trimestre positivo como na Caixa E por último o choque em Basileia afetou o escore de eficiência de forma positiva como na Caixa em todos os trimestres Em linhas gerais não houve muita semelhança entre os resultados da média ponderada dos bancos públicos e os resultados do Banco do Brasil 68 Figura 9 Função Impulsoresposta das diversas variáveis ao escore de eficiência da média ponderada dos bancos Fonte elaborado pelo próprio autor Na última análise dos gráficos das Funções ImpulsoResposta Ortogonalizadas Cumulativas ao analisarmos a média ponderada das cinco instituições financeiras notamos uma semelhança dos resultados do efeito do choque no INEC nos movimentos futuros do escore de eficiência ao resultado da Caixa e dos Bancos Privados com um efeito positivo nos primeiros trimestres e negativo nos últimos No Risco Brasil o efeito foi positivo somente no 8º trimestre que foi o que observamos somente na média ponderada dos privados No PIB o efeito foi estatisticamente não diferente de zero para todos os trimestres o que ocorreu na Caixa no Banco do Brasil e em alguns poucos trimestres do Itaú Bradesco e Santander Para o IGPM o efeito foi negativo no 4º e 5º trimestres e positivo no 8º Esse efeito positivo no final só foi encontrado nessa situação Basileia apresentou um resultado estatisticamente não diferente de zero como encontrado no Banco do Brasil e no Bradesco Por último como todos os gráficos apresentados nessa etapa da análise em todas as instituições e nas três médias ponderadas o efeito do choque do próprio escore de eficiência sobre os movimentos futuros dele mesmo foram positivos 69 Além das análises e interpretações dos gráficos de função impulsoresposta realizadas acima para cada uma das oito amostras cinco instituições financeiras e as três médias a Tabela 4 abaixo representa a consolidação das funções impulsoresposta das diversas variáveis ao escore de eficiência de todas as amostras Tabela 4 Resumo das Funções Impulsoresposta das diversas variáveis ao escore de eficiência de todas as amostras Fonte elaborado pelo próprio autor Nota Os casos estatisticamente indistinguíveis de zero foram identificados como indist e os números entre parênteses indicam os trimestres após o choque 423 Análises das Decomposições das Variâncias dos Erros de Previsão A nossa última Análise Estrutural dos modelos VAR é a decomposição da variância dos erros de previsão conforme descrito na seção 3 Ela nos permitirá avaliar conforme descrito por Stock e Watson 2001 o quanto em percentual da variância do erro cometido na previsão do escore de eficiência é devido a um choque em uma das variáveis do modelo no caso o INEC o Risco Brasil o PIB o IGPM o Basileia e o próprio escore de eficiência As tabelas 5 6 7 8 9 10 11 e 12 abaixo apresentarão a fração da variância do erro de previsão do escore de eficiência que pode ser atribuída a cada variável do modelo em cada um dos oito trimestres após o choque Variável Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral INEC Positivo Indist 16 e negativo 78 Positivo 14 e indist 58 Positivo 13 indist 47 e negativo 8 Negativo Positivo 12 indist 37 e negativo 8 Positivo 13 e indist 48 Positivo 12 indist 34 e negativo 58 Risco Brasil Negativo 1 e indist 28 Indist Indist Positivo 23 e indist 14 8 Indist 1 e positivo 28 Negativo 1 indist 27 e positivo 8 Positivo 1 e indist28 Indist 17 e positivo 8 PIB Negativo e indist 8 Indist Negativo 28 e indist 1 Indist Positivo e indist 3 Positivo 1 indist 28 e negativo 37 Indist 15 e positivo 68 Indist IGPM Negativo Indist Negativo 1 e indist 28 Positivo 1 e indist 28 Negativo 1 e indist 28 Indist 168 e negativo 2 5 Positivo 1 indist 2 68 e negativo 3 5 Indist 13 6 7 negativo 4 5 e positivo 8 Basileia Negativo e indist 1 Indist Indist Positivo Negativo Negativo Positivo Indist Eficiência de Custo Positivo Positivo Positivo Positivo Positivo Positivo Positivo Positivo 70 Tabela 5 Estimativas pontuais das decomposições de variância do erro de previsão do Itaú Fonte elaborado pelo próprio autor A tabela 5 acima mostra as estimativas da decomposição de variância do erro de previsão para o escore de eficiência do Itaú Podemos notar que os choques no escore de eficiência são responsáveis por cerca de 55 das oscilações futuras do próprio índice revelando uma persistência das estruturas que o compõe Por outro lado outros fatores representam os 45 restantes com destaque para o Risco Brasil cerca de 15 Podemos também observar o aumento da importância do Risco Brasil na explicação dos movimentos do escore de eficiência do Itaú começando a partir do 4º trimestre e saltando novamente depois do 7º atingindo cerca de 15 conforme mencionado As demais variáveis oscilaram entre cerca de 8 e 6 tendo destaque o salto de relevância a partir do 3º trimestre Tabela 6 Estimativas pontuais das decomposições de variância do erro de previsão do Banco do Brasil Fonte elaborado pelo próprio autor Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 894 21 35 21 27 02 3 789 36 42 41 62 31 4 742 39 71 50 59 40 5 706 37 73 70 76 38 6 652 40 74 76 78 79 7 568 63 157 65 70 78 8 555 75 154 64 68 84 Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 973 03 07 10 00 07 3 955 03 15 20 02 06 4 887 55 24 26 02 08 5 883 64 29 18 01 05 6 884 72 23 15 01 04 7 873 80 22 19 02 05 8 847 106 22 18 02 05 71 No caso do Banco do Brasil de acordo com a tabela 6 acima notamos que os choques no escore de eficiência são responsáveis por cerca de 85 das oscilações futuras do próprio índice revelando uma persistência das estruturas que o compõe muito superior à do Itaú Já os outros fatores que representam os 15 restantes somente o INEC merece destaque com cerca de 11 Cabe destacar o aumento da importância dele na explicação dos movimentos do escore de eficiência do Banco do Brasil a partir do 4º trimestre e novamente a partir do 8º Vale lembrarmos que um choque positivo no INEC leva a uma resposta negativa no escore de eficiência do Banco do Brasil As demais variáveis oscilaram entre cerca de 0 e 2 Tabela 7 Estimativas pontuais das decomposições de variância do erro de previsão do Bradesco Fonte elaborado pelo próprio autor Na análise do Bradesco de acordo com a tabela 7 acima podemos observar que os choques no escore de eficiência são responsáveis por cerca de 61 das oscilações futuras do próprio índice revelando uma persistência das estruturas que o compõe mais próxima as do Itaú e menos relevantes que a do Banco do Brasil Merece menção o PIB com cerca de 17 que saltou de cerca de 0 para cerca de 8 a partir do 3º trimestre que conforme vimos na análise das funções de impulsoresposta afeta negativamente o escore de eficiência deste banco Outra variável muito relevante para explicação das oscilações futuras é o INEC assim como foi com o Banco do Brasil nesse caso com cerca de 13 muito similar ao valor observado logo no 2º trimestre Para essa variável nesse banco Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 827 128 14 00 26 06 3 743 129 12 81 23 12 4 700 117 40 111 21 12 5 703 112 39 108 26 11 6 694 106 43 122 24 11 7 643 96 51 172 24 14 8 609 134 51 169 24 14 72 choques positivos levaram a uma resposta também positivo no escore de eficiência Tabela 8 Estimativas pontuais das decomposições de variância do erro de previsão da Caixa Fonte elaborado pelo próprio autor Para a Caixa de acordo com a tabela 8 acima os choques no escore de eficiência são responsáveis por cerca de 43 das oscilações futuras do próprio índice revelando que outras variáveis contribuem significativamente para os movimentos futuros do seu escore de eficiência O principal destaque diferentemente dos demais bancos analisados é o Basileia que representou cerca de 33 tendo saltando do 2º para o 3º trimestre de cerca de 5 para cerca de 41 lembrando que choques positivos em Basileia afetavam positivamente o escore de eficiência da Caixa Outra variável muito relevante é o INEC assim como foi para o Banco do Brasil e Bradesco que nesse caso representou cerca de 13 das oscilações futuras Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 910 25 00 05 09 51 3 458 47 61 22 06 407 4 445 45 61 25 09 415 5 455 50 58 36 14 387 6 474 114 56 30 11 315 7 466 118 60 31 14 311 8 430 129 63 37 15 328 73 Tabela 9 Estimativas pontuais das decomposições de variância do erro de previsão do Santander Fonte elaborado pelo próprio autor Já para o Santander de acordo com a tabela 9 acima podemos observar que os choques no escore de eficiência são responsáveis por cerca de 33 das oscilações futuras do próprio índice revelando que as demais variáveis somadas explicam melhor as oscilações futuras do Índice do que ele próprio Com exceção do IGPM com cerca de 7 as demais variáveis representam mais de 10 das oscilações futuras do Índice do Santander A mais relevante é o INEC como foi para o Bradesco e o Banco do Brasil com cerca de 27 tendo atingido essa relevância logo no 4º trimestre da análise As demais variáveis na ordem decrescente de relevância foram Basileia com cerca de 13 PIB com cerca de 11 e Risco Brasil com cerca de 10 Tabela 10 Estimativas pontuais das decomposições de variância do erro de previsão da média ponderada dos privados Fonte elaborado pelo próprio autor Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 798 38 11 34 17 102 3 586 114 73 32 39 156 4 468 264 54 25 36 153 5 468 257 47 50 34 145 6 391 220 97 88 68 136 7 362 246 90 97 69 136 8 326 268 101 113 67 126 Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 861 60 34 22 04 20 3 804 81 32 46 18 19 4 754 104 39 64 17 22 5 780 75 46 46 37 17 6 764 75 46 54 45 17 7 679 65 112 48 80 16 8 630 114 114 44 82 15 74 A tabela 10 acima mostra as estimativas da decomposição de variância do erro de previsão para o escore de eficiência da média ponderada dos três bancos privados da nossa amostra Podemos notar que os choques no escore de eficiência são responsáveis por cerca de 63 das oscilações futuras do próprio índice Dentre os demais fatores destacamos o INEC e o Risco Brasil ambos com cerca de 11 sendo que o segundo atingiu essa relevância somente a partir do 7º trimestre Outra variável que merece destaque é o IGPM que representou cerca de 8 da oscilação futura Tabela 11 Estimativas pontuais das decomposições de variância do erro de previsão da média ponderada dos públicos Fonte elaborado pelo próprio autor A tabela 11 acima mostra as estimativas da decomposição de variância do erro de previsão para o escore de eficiência da média ponderada dos dois bancos públicos da nossa amostra onde podemos notar que os choques no escore de eficiência são responsáveis por cerca de 72 das oscilações futuras do próprio índice evidenciando uma persistência maior que a média dos bancos privados Para as demais variáveis merecem destaque o INEC e o PIB com cerca de 10 e 7 respectivamente lembrando que ambos apresentaram resultados que evidenciaram efeito positivo no escore de eficiência após choques positivos em seus resultados O INEC logo no 2º trimestre apresentou cerca de 5 seguido por cerca de 9 no 3º Já o PIB saltou de cerca de 2 para 5 entre o 4º e 5º trimestres As demais variáveis oscilaram entre cerca de 3 e 4 tendo atingido em alguns momentos uma relevância próxima aos 6 Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 897 49 11 00 13 30 3 820 86 13 13 34 34 4 757 75 31 15 65 58 5 760 63 38 45 51 44 6 756 94 34 37 42 38 7 732 91 33 63 45 36 8 718 96 33 71 44 37 75 Tabela 12 Estimativas pontuais das decomposições de variância do erro de previsão da média ponderada dos bancos Fonte elaborado pelo próprio autor Por último nessa etapa da análise a tabela 12 acima mostra as estimativas da decomposição de variância do erro de previsão para o escore de eficiência da média ponderada dos cinco bancos desse estudo É possível notar uma persistência do próprio escore de eficiência na explicação das suas oscilações futuras sendo responsável por cerca de 57 delas Dentre as demais variáveis merece destacarmos o INEC com cerca de 23 com saltos expressivos no 4º 7º e 8º trimestre Lembramos que choques positivos nessa variável afetaram positivamente o escore de eficiência nos primeiros trimestres mas negativamente nos últimos As outras variáveis oscilaram entre cerca de 3 e 6 43 Previsão Conforme descrito acima na seção 3 após as Análises Estruturais utilizamos o modelo VAR especificado rodando somente até o 4º trimestre de 2018 para prever os dados do último ano os quatro trimestres de 2019 ou seja uma pseudoprevisão fora da amostra e comparamos com uma autorregressão AR com quatro defasagens para os mesmos períodos Os testes foram realizados para as oito amostras nos três critérios especificados e conforme Quadro 2 abaixo Trimestre Eficiência de Custo INEC Risco Brasil PIB IGPM Basileia 1 1000 00 00 00 00 00 2 920 72 00 01 05 01 3 872 71 20 02 13 22 4 772 122 24 41 22 19 5 765 133 17 35 34 16 6 735 147 28 31 31 28 7 649 186 46 28 47 44 8 565 234 62 31 54 54 76 Quadro 2 Resultado dos testes DieboldMariano para os critérios MSE MAE e MAPE Fonte elaborado pelo próprio autor Nota os resultados em verde identificam a melhor previsão Os resultados acima mostram que os modelos AR apresentam menor erro de previsão para as amostras do Itaú Banco do Brasil Bradesco Caixa e Santander Para a média ponderada dos bancos privados somente no critério MSE o VAR apresentou melhor resultado Já nas médias ponderadas dos bancos públicos e na média ponderada geral o VAR apresentou melhores resultados nos três critérios Apesar de esperarmos que os resultados do modelo VAR previssem melhor os escores de eficiência como o resultado das previsões são extrapolados para um único ano ou seja quatro trimestres que teve suas diversas idiossincrasias com os parâmetros do modelo mudando constantemente regulação ciclo econômico choques políticos etc é natural termos resultados diferentes das expectativas Portanto conforme Clements e Hendry 2002 afirmaram os principais problemas que afligem as previsões econômicas surgem das coisas que não sabemos que não sabemos e dessas as mudanças em termos determinísticos podem ser as mais perniciosas p 16 Critério Modelo Itaú Banco do Brasil Bradesco Caixa VAR 000026 000172 000212 000050 AR 000011 000021 000041 000033 Diferença 000015 000151 000171 000017 VAR 001271 003275 003862 001918 AR 000940 001087 001811 001447 Diferença 000331 002188 002051 000471 VAR 001793 004613 005157 002576 AR 001312 001530 002415 001950 Diferença 000481 003084 002742 000626 Critério Modelo Santander Privados Públicos Geral VAR 000140 000010 000009 000004 AR 000022 000013 000020 000005 Diferença 000118 000003 000011 000000 VAR 002966 000815 000617 000462 AR 001365 000814 000930 000612 Diferença 001601 000000 000313 000150 VAR 004099 001119 000831 000639 AR 001885 001103 001282 000840 Diferença 002214 000016 000451 000201 MSE MAE MAPE MSE MAE MAPE 77 tradução nossa ou seja previsões foradajanela devem ser analisados com cautela para avaliar se um modelo é melhor ou pior 44 Discussão dos resultados à luz das hipóteses Para uma discussão mais abrangente e profunda na sequência compararemos cada hipótese da pesquisa com os resultados encontrados nas Análises Estruturais causalidade de Granger nas funções impulsoresposta e na decomposição da variância dos erros de previsão dos modelos VAR das oito amostras abordadas H1 Expectativas futuras eficiência bancária A hipótese H1 pressupunha que indicadores prospectivos que pudessem de alguma forma indicar expectativas futuras representado nesse estudo pelo INEC possuem uma relação positiva com a eficiência de custos das instituições financeiras Pela análise da causalidade de Granger pudemos constatar que na maioria das amostras o INEC Grangercausa a Eficiência de Custos ou seja os valores passados do INEC ajudam a prever os valores da Eficiência de Custos Outra evidência encontrada para a maioria dos casos foi o efeito feedback ou seja a Eficiência de Custos também Grangercausa o INEC Esses resultados corroboram com os estudos de Beccalli Casu e Girardone 2006 Fiordelisi MarquesIbanez e Molyneux 2011 e Chen Higgins e Mason 2005 Já pela análise das funções impulsoresposta foi possível observar que o INEC afeta os movimentos futuros dos escores de eficiência porém observamos que esse efeito é diferente entre as instituições financeiras De forma sucinta o efeito é positivo para Itaú com efeito subindo gradativamente após o choque até seu ápice no 4º trimestre de 40 bps até 30 bps no 8º e para o Bradesco com maior efeito no 2º e 3º trimestres na casa de 50 bps depois decrescendo Esses efeitos ratificam os três estudos citados Contudo encontramos resultados negativos para Banco do Brasil com efeito decrescente até atingir o 8º trimestre com 110 bps Santander também com aumento gradativo do efeito 78 negativo até o último trimestre com 200 bps e a média ponderada geral que até começou com efeito positivo mas foi decrescendo ao longo dos trimestres até atingir 50 bps no 8º trimestre Para Caixa e as médias ponderadas dos privados e públicos encontramos resultados positivos nos primeiros trimestres e negativos nos demais os quais não podemos discordar das referências pois eles utilizaram metodologias diferentes que não nos permitem essa comparação dos efeitos de um choque ao longo de um período subsequente Vale destacar a Caixa que apresentou um efeito que é decrescente chegando até 50 bps no último trimestre A tabela 13 abaixo mostra esses efeitos trimestre a trimestre em bps por amostra uma releitura dos gráficos de função impulsoresposta apresentados acima mas que facilitarão a visualização dos efeitos Vale destacar como efeito de conceito e interpretação dos dados que um efeito de 100 bps significa que um choque positivo de 1 desvio padrão na variável mencionada melhora o escore de eficiência de custos em 1 ponto percentual pp ou seja seria possível produzir os mesmos resultados com 1 menos custos totais Tabela 13 Efeito das Funções Impulsoresposta dos choques do INEC à eficiência de custos de todas as amostras Fonte elaborado pelo próprio autor Nota os números apresentados estão em bps De forma complementar a análise da decomposição da variância dos erros de previsão mostrou que choques no INEC são os principais responsáveis pela variância do erro de previsão dos escores de eficiência de cada banco com exceção é claro do próprio escore de eficiência reforçando a relevância do mesmo Destaque para o Santander e a média ponderada geral dos bancos que tiveram cerca de 25 de representatividade na variância Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 8 5 28 9 15 9 13 8 2 16 4 46 25 42 16 26 9 3 20 17 47 24 88 8 30 1 4 21 38 44 15 117 7 21 8 5 26 58 41 12 136 5 2 16 6 38 73 46 21 167 5 0 27 7 29 99 22 35 200 8 9 41 8 33 114 2 52 200 21 11 47 79 H2 Riscopaís eficiência bancária A hipótese H2 era de que a medida de riscopaís representada nesse estudo pelo Risco Brasil possuem uma relação negativa com a eficiência de custos das instituições financeiras analisadas A análise da causalidade de Granger nos mostrou que não há um consenso entre as diferentes amostras analisadas sendo que para Itaú Bradesco Caixa e a média ponderada dos privados o Risco Brasil Grangercausa a eficiência de custos mas para as outras quatro amostras essa afirmação é falsa Contudo de forma unânime a Eficiência de Custos não Grangercausa o Risco Brasil Na interpretação das análises das funções impulsoresposta também não há um resultado similar nas oito amostras De modo geral com exceção de um ou outro trimestre e do Santander os resultados não foram estatisticamente diferentes de zero O caso do Santander cujo efeito nos choques do Risco Brasil causaram um efeito positivo na Eficiência de Custos com efeito crescente chegando até 156 bps no último trimestre contraria as afirmações de Tamadonnejad et al 2013 Borges 2012 e AlGasaymeh 2016 A tabela 14 abaixo facilita a interpretação dos dados e ressalta a discrepância dos efeitos no Santander em relação às demais amostras Tabela 14 Efeito das Funções Impulsoresposta dos choques do Risco Brasil à eficiência de custos de todas as amostras Fonte elaborado pelo próprio autor Nota os números apresentados estão em bps A análise da decomposição da variância dos erros de previsão mostrou que choques no Risco Brasil são outros importantes responsáveis pela variância do erro de previsão dos escores de eficiência de cada banco Destacamos o Itaú o Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 10 7 9 1 8 7 6 0 2 3 1 5 20 32 6 10 4 3 7 11 12 25 34 2 0 7 4 12 25 14 19 44 5 11 5 5 7 27 22 10 78 3 15 0 6 15 25 10 2 83 11 13 7 7 12 18 2 11 105 5 10 1 8 6 15 19 24 156 25 8 19 80 Santander e a média ponderada dos bancos privados todos com mais de 10 de representatividade na variância H3 Atividade econômica eficiência bancária A hipótese H3 era de que a atividade econômica representada no nosso estudo pelo PIB afeta positivamente a eficiência de custos das amostras analisadas Para essa hipótese diferentemente das H1 e H2 não havia consenso entre as referências bibliográficas citadas A análise da causalidade de Granger como era esperado nos mostrou que para todas as amostras o PIB Grangercausa a eficiência de custos e que o efeito feedback é verdadeiro na maioria dos casos exceção ao Banco do Brasil e à média ponderada geral o que independe do sinal desse efeito corrobora com a opinião da maioria dos estudos mencionados nessa hipótese e contraria o estudo de Chen et al 2018 que não encontrou um impacto significativo do PIB sobre a eficiência Já a interpretação das análises das funções impulsoresposta nos mostrou uma divisão clara entre os resultados Para o Itaú o Bradesco e para a média ponderada dos privados em linhas gerais os choques no PIB causaram um efeito negativo na eficiência de custos como posto nos estudos de Chortareas Girardone e Ventouri 2012 Sufian 2009 e Kallel Hamad e Triki 2019 Os resultados do Itaú apesar de estatisticamente diferentes de zero e negativos representaram um efeito absoluto no escore de eficiência pouco significativo Já para o Bradesco nos últimos trimestres o efeito se aproximou de 100 bps No caso dos dois bancos públicos e sua média ponderada e para o Santander apesar de alguns trimestres onde os resultados foram indistinguíveis de zero notamos indicações de um efeito positivo na eficiência de custos Esses resultados ratificam as considerações de Tan e Anchor 2017 AlGasaymeh 2016 e Primo et al 2013 que afirmaram que os bancos são mais eficientes durante os períodos de crescimento econômico O efeito no escore de eficiência do Santander merece ser destacado pois chegou a 155 bps no 8º trimestre E como era esperado depois desse resultado divergente a média ponderada geral foi indistinguível de zero em todos os trimestres A tabela 15 abaixo transcreve visualmente esses efeitos Apesar de termos encontrado evidências que 81 ratificam todos os estudos abordados efeito positivo negativo e sem resultado significativo boa parte das razões desses efeitos descritas nesses estudos não nos parecem adequadas pois seriam efeitos gerais que afetariam todo mercado independentemente de ser um banco público ou privado como melhora na qualidade dos tomadores e aumento na demanda de crédito Com certeza existem fatores e ações específicas itens de gestão e governança adotadas por cada uma dessas Organizações que causaram esses resultados Tabela 15 Efeito das Funções Impulsoresposta dos choques do PIB à eficiência de custos de todas as amostras Fonte elaborado pelo próprio autor Nota os números apresentados estão em bps Como último elemento de análise dessa hipótese a decomposição da variância dos erros de previsão mostrou que choques no PIB foram responsáveis em seis amostras por menos de 10 da variância do erro de previsão dos escores de eficiência sendo o Bradesco e o Santander com cerca de 17 e 11 respectivamente as exceções H4 Inflação eficiência bancária A penúltima hipótese a H4 era de que a inflação representada pelo IGPM afeta negativamente a eficiência de custos das instituições e médias analisadas Da mesma forma que a anterior para essa hipótese também não havia consenso entre as referências A primeira análise da causalidade de Granger mostrou que para os três bancos privados e para as três médias ponderadas o IGPM Grangercausa a eficiência de custos mas a eficiência de custos não Grangercausa o IGPM exceção feita à Caixa e à média ponderada dos públicos Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 8 8 0 4 14 6 1 1 2 17 1 27 8 23 1 7 0 3 23 9 46 3 19 7 2 6 4 14 3 41 14 41 8 18 2 5 20 1 56 10 71 13 20 4 6 19 8 83 15 90 15 36 2 7 20 9 96 24 115 15 45 6 8 5 29 80 44 155 11 64 3 82 A interpretação das análises das funções impulsoresposta nos mostrou que com exceção do Itaú os choques no IGPM na maioria dos trimestres causaram efeitos indistinguíveis de zero No caso do Itaú os choques causam um efeito negativo na eficiência de custos como era nossa hipótese inicial alinhado com o constatado por Primo et al 2013 e AlGasaymeh 2016 porém com pouco efeito absoluto no escore chegando no ápice no 5º trimestre com 37 bps conforme tabela 16 abaixo Tabela 16 Efeito das Funções Impulsoresposta dos choques do IGPM à eficiência de custos de todas as amostras Fonte elaborado pelo próprio autor Nota os números apresentados estão em bps Na decomposição da variância dos erros de previsão a variável IGPM foi a que mostrou menos relevância na variância do erro de previsão da eficiência de custos sendo menor que 9 em todas as amostras H5 Solvência eficiência bancária A quinta e última hipótese a H5 pressupunha que a solvência que no nosso estudo foi abordada como o índice de Basileia era relacionada negativamente com a eficiência de custos e assim como as duas últimas hipóteses não havia um consenso na literatura considerada com estudos e indicações confirmando ou negando essa afirmação As análises da causalidade de Granger mostraram que somente para Caixa Santander e média ponderada dos públicos Basileia Grangercausa eficiência de custos e nas demais amostras a maioria não Grangercausa Já no efeito contrário de forma unânime a eficiência de custos Grangercausa Basileia Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 9 2 13 6 10 2 7 2 2 20 2 7 2 5 7 3 1 3 23 2 6 3 4 9 16 4 4 31 5 14 10 14 17 22 10 5 37 8 15 10 14 12 20 8 6 33 5 10 4 27 1 14 1 7 34 1 4 0 37 5 11 5 8 31 2 8 4 56 12 14 10 83 Esses resultados contrariam as evidências apresentadas nos estudos relacionados a essa hipótese Nas análises das funções impulsoresposta pudemos observar que os choques no índice de basileia apresentaram um efeito negativo na eficiência de custos para o Itaú Santander e média ponderada dos privados que foi o que pressupunha a nossa hipótese e que afirmou os estudos de Chortareas Girardone e Ventouri 2012 Sufian 2009 e Zhang et al 2013 No caso do Santander os efeitos absolutos relativos foram bem representativos com 150 bps no último trimestre Já para a Caixa e a média ponderada dos bancos públicos o efeito foi positivo alinhado às conclusões de Fiordelisi Marques Ibanez e Molyneux 2011 Athanasoglou Brissimis e Delis 2008 Tecles e Tabak 2010 e Mendonça 2018 No caso da Caixa os últimos trimestres também tiveram efeitos de destaque chegando a 90 bps nos últimos 2 trimestres Para as outras três amostras os resultados foram indistinguíveis de zero Tabela 17 Efeito das Funções Impulsoresposta dos choques de Basileia à eficiência de custos de todas as amostras Fonte elaborado pelo próprio autor Nota os números apresentados estão em bps Na decomposição da variância dos erros de previsão a variável Basileia foi extremamente relevante para a Caixa com cerca de 32 da variância do erro de previsão da eficiência de custos mas para as demais amostras teve valores bem mais inferiores com destaque para o Santander com cerca de 13 e o Itaú com cerca de 8 Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 3 7 6 13 25 5 10 1 2 13 8 2 65 51 6 16 5 3 19 4 2 78 73 9 28 5 4 19 4 1 65 93 10 33 2 5 33 7 1 64 114 9 37 3 6 40 11 6 68 130 11 37 4 7 33 12 3 90 139 11 41 10 8 26 14 3 88 150 10 53 7 84 Depois de apresentado e os resultados as análises e as interpretações das hipóteses de pesquisa apresentamos um resumo sintético comparando a relação esperada e os resultados obtidos para cada hipótese no Quadro 2 abaixo Quadro 3 Síntese comparativa entre os sinais esperados e os resultados apurados Fonte elaborado pelo próprio autor Nota os campos indicados com os dois sinais significam que ambos os efeitos ocorreram ao longo dos trimestres de acordo com a respectiva ordem Os campos identificados como indist significam que os resultados foram predominantemente indistinguíveis de zero 45 Teste de Robustez 451 Utilização de variáveis alternativas Uma série de verificações de robustez para confirmar a validade dos modelos e resultados empíricos relatados acima foram feitas Itaú Banco do Brasil Bradesco Caixa H1 Expectativas futuras INEC e H2 Medida de riscopaís Risco Brasil indist indist H3 Atividade econômica PIB indist indist H4 Inflação IGPM indist indist indist H5 Solvência Basileia indist indist Santander Privados Públicos Geral H1 Expectativas futuras INEC e e H2 Medida de riscopaís Risco Brasil e indist indist H3 Atividade econômica PIB indist H4 Inflação IGPM indist e H5 Solvência Basileia indist Hipótese Hipótese Tema Variável Sinal Esperado Sinal Apurado nos Testes Empíricos Sinal Apurado nos Testes Empíricos Sinal Esperado Variável Tema 85 Primeiramente outras variáveis foram testadas tanto no aspecto de mercado quanto no específico das instituições financeiras Seguindo a metodologia descrito na seção 3 testamos a estacionaridade de todas as variáveis levantadas e não foi possível estacionarizar nem por primeira diferença por aplicação de logaritmo ou por ambos somente a quantidade de agências e postos de atendimento de todas as amostras levantadas logo essa variável foi descartada As demais variáveis testadas e não utilizadas foram descartadas na checagem do modelo com destaque para a estabilidade do modelo VAR que nos fez descartar por conta de resultados desfavoráveis em pelo menos uma das amostras modelos com a variável de primeira diferença do logaritmo da quantidade de agências12 primeira diferença do Retorno sobre o Capital ambos casos de variáveis específicas ou Taxa de Juros variável doméstica 452 Variação na ordem das variáveis Numa segunda etapa também descrita na mesma seção testamos especificações alternativas com inversão ou trocas na ordenação causal dado que essa é uma etapa muito relevante e crítica para nossa estrutura do estudo conforme descrito na seção 3 Como quanto mais correlacionado forem os resíduos mais sensível poderão ser as respostas a diferentes ordenações das variáveis do sistema COLOMBO LONCAN e CALDEIRA 2018 e até mesmo como forma de testar a sensibilidade dos resultados a essas alterações decidimos supor que ao invés das variáveis que descrevemos como preditoras de expectativas futuras e indicadores de riscopaís afetaram as decisões de produção as expectativas futuras e o riscopaís seriam afetados pelos resultados do produto interno Testamos portanto a ordenação IGPM PIB Risco Brasil e INEC mantendo obviamente a eficiência de custos como variável dependente e Basileia variável específica de cada instituição como última Comparamos os resultados das análises das funções impulsoresposta deste exercício de robustez com os resultados do modelo final escolhido no mesmo 12 A variável quantidade de agências e postos de atendimento mencionada acima corresponde a soma dessas duas informações disponíveis na fonte de dados mencionada na seção específica 86 formato apresentado acima conforme tabelas 18 19 20 e 21 abaixo para cada variável que teve sua posição alterada Obviamente as variáveis eficiência de custos e Basileia por não sofrerem alteração na posição não tiveram diferença nos seus resultados Tabela 18 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável INEC Fonte elaborado pelo próprio autor Nota os números apresentados estão em pp Tabela 19 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável Risco Brasil Fonte elaborado pelo próprio autor Nota os números apresentados estão em pp Tabela 20 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável PIB Fonte elaborado pelo próprio autor Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 01 01 00 00 00 00 01 00 2 01 00 01 01 00 00 01 00 3 01 01 01 01 01 00 00 00 4 01 02 01 01 01 00 01 00 5 02 02 01 00 00 00 01 00 6 02 02 01 01 01 00 01 01 7 02 02 01 00 01 01 01 00 8 01 02 01 00 01 01 01 00 Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 01 01 01 00 01 00 00 00 2 01 00 02 00 04 01 00 00 3 01 01 03 00 05 01 01 00 4 01 01 03 00 07 01 01 00 5 01 01 03 01 12 01 01 00 6 01 00 05 01 15 01 00 00 7 01 01 05 02 18 01 00 01 8 00 01 03 04 23 01 01 02 Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 00 01 00 00 01 01 00 00 2 00 00 00 00 01 01 00 00 3 01 01 01 00 03 00 00 00 4 01 02 01 00 03 01 01 01 5 01 02 02 01 03 01 01 01 6 00 02 01 00 03 00 00 01 7 01 02 01 00 03 00 00 01 8 00 01 02 01 02 02 00 01 87 Nota os números apresentados estão em pp Tabela 21 Diferença do efeito das Funções Impulsoresposta do Teste de Robustez com o Modelo escolhido para a variável IGPM Fonte elaborado pelo próprio autor Nota os números apresentados estão em pp A respeito das Tabelas acima podemos concluir que não há grandes oscilações que mudem o sentido da análise ou seja que indicavam que choques geravam efeito positivo e passaram a gerar negativo ou viceversa A única exceção a essa afirmação foi a interpretação do choque da variável Risco Brasil na amostra do Santander que no modelo selecionada apresentava indícios de que o efeito era positivo mas nesse teste de robustez apresentou evidências de que o efeito seria negativo Contudo interpretamos que esse foi um caso isolado de uma das amostras portanto além da razoabilidade econômica ser mais factível no modelo escolhido esses resultados corroboram para a escolha do modelo e evidenciam que para os dados analisados em linhas gerais nossos resultados são qualitativa semelhantes independentemente de como as variáveis são ordenadas na decomposição de Cholesky da matriz de covariância dos resíduos 453 Eficiência de receita e lucro Na terceira e última etapa reestimamos os mesmos modelos aprovados porém com foco nos escores de eficiência de receita e lucro muito similar ao exercício feito por Fiordelisi MarquesIbanez e Molyneux 2011 A estatística descritiva dos escores de eficiência de receita e lucro estão na Tabela 22 abaixo Trimestre Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral 1 00 01 02 00 01 00 01 00 2 02 00 02 02 01 00 02 01 3 02 02 02 02 01 00 02 00 4 04 03 02 02 00 01 02 00 5 04 04 02 00 01 00 01 01 6 03 04 01 00 00 01 01 02 7 03 05 01 00 01 01 01 02 8 04 05 02 00 04 00 02 02 88 Tabela 22 Estatística descritiva dos escores de eficiência de receita e lucro Fonte elaborado pelo próprio autor Como primeiro desafio por conta da não estacionaridade disponível no Apêndice C ao invés de usar diretamente o escore de receita e lucro extraímos sua primeira diferença conforme mencionado na subseção 33 Os dados da Matriz de correlação das variáveis de escore de eficiência de custo lucro e receita para cada uma das oito amostras disponível no Apêndice H apresentaram uma média de 017 sendo que a correlação mais alta foi de 070 e a mais baixa de 012 ou seja sugerindo que não há problema de colinearidade abaixo de 08 conforme descreveu Kennedy 2008 Desse modo foi possível rodar os Modelos VAR para as oito amostras considerando um cenário de escore de receita e outro cenário de escore de lucro Para testar a estabilidade desses Modelos VAR para o escore de eficiência de lucro no caso das amostras da Caixa média ponderada do públicos e média ponderada geral o modelo se mostrou instável O mesmo ocorreu também para a Caixa mas nesse caso para o escore de receita Essa instabilidade do modelo conforme mencionado anteriormente nos inviabiliza de analisar os resultados das funções impulsoresposta Em todo caso como exercício de reflexão e Média Desvio Padrão Mínimo Máximo Itaú 0997224 0000044 0997118 0997326 BB 0997299 0000040 0997182 0997367 Bradesco 0997282 0000049 0997167 0997378 Caixa 0997268 0000045 0997123 0997336 Santander 0997283 0000076 0997149 0997534 Privados 0997257 0000042 0997161 0997359 Públicos 0997282 0000038 0997181 0997351 Geral 0997267 0000024 0997209 0997313 Itaú 0995559 0000029 0995506 0995607 BB 0995592 0000038 0995521 0995676 Bradesco 0995585 0000041 0995496 0995693 Caixa 0995589 0000053 0995459 0995736 Santander 0995597 0000062 0995489 0995709 Privados 0995576 0000026 0995527 0995628 Públicos 0995591 0000034 0995525 0995686 Geral 0995582 0000023 0995544 0995635 Variável Escore de Eficiência de receita Escore de Eficiência de lucro 89 alinhamento quanto aos resultados do escore de eficiência de custos comparamos na Tabela 23 os resultados dos testes de causalidade de Granger Tabela 23 Probabilidades do teste de causalidade de Granger para Eficiência de Receita Lucro e Custos continua Equação a Excluído b Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral Eficiência de Receita INEC 0020 0307 0732 0019 0051 0694 0206 0110 Eficiência de Lucro INEC 0263 0107 0006 0014 0263 0835 0003 0019 Eficiência de Custo INEC 0005 0487 0000 0000 0000 0000 0104 0000 INEC Eficiência de Receita 0119 0046 0765 0050 0002 0393 0038 0210 INEC Eficiência de Lucro 0000 0001 0002 0052 0497 0050 0022 0013 INEC Eficiência de Custo 0003 0032 0285 0017 0000 0000 0146 0003 Eficiência de Receita Risco Brasil 0000 0575 0033 0798 0000 0235 0898 0008 Eficiência de Lucro Risco Brasil 0597 0030 0012 0019 0434 0004 0029 0060 Eficiência de Custo Risco Brasil 0014 0876 0000 0041 0500 0001 0561 0053 Risco Brasil Eficiência de Receita 0000 0000 0161 0000 0000 0000 0000 0000 Risco Brasil Eficiência de Lucro 0272 0000 0281 0374 0000 0001 0074 0001 Risco Brasil Eficiência de Custo 0671 0105 0078 0288 0060 0118 0538 0782 Eficiência de Receita PIB 0000 0025 0010 0014 0024 0270 0263 0000 Eficiência de Lucro PIB 0020 0007 0000 0546 0144 0020 0449 0149 Eficiência de Custo PIB 0000 0000 0000 0000 0000 0000 0000 0000 PIB Eficiência de Receita 0992 0020 0109 0000 0055 0241 0037 0277 PIB Eficiência de Lucro 0650 0000 0000 0372 0000 0025 0000 0006 PIB Eficiência de Custo 0000 0539 0000 0004 0006 0002 0007 0128 Eficiência de Receita IGPM 0001 0274 0754 0922 0051 0502 0743 0558 Eficiência de Lucro IGPM 0572 0296 0000 0000 0274 0640 0014 0010 Eficiência de Custo IGPM 0013 0916 0009 0091 0000 0001 0023 0000 IGPM Eficiência de Receita 0000 0000 0487 0115 0001 0007 0000 0124 IGPM Eficiência de Lucro 0057 0000 0815 0000 0000 0850 0003 0343 IGPM Eficiência de Custo 0570 0353 0905 0000 0976 0561 0000 0108 90 Tabela 23 Probabilidades do teste de causalidade de Granger para Eficiência de Receita Lucro e Custos conclusão Fonte Elaborado pelo próprio autor Nota Os números expressos na tabela representam a probabilidade de aceitar a hipótese nula sendo que a fonte verde representa os casos em que rejeitamos e em vermelho os casos em que aceitamos Consideramos o nível de significância de 5 Em linhas gerais apesar de resultados diferentes para cada uma das oito amostras e para os três tipos de escore de eficiência O que temos de grande divergência entre os resultados dos três escores são o PIB Grangercausa a eficiência de custos que foi unânime para todas as amostras mas não se mostrou verdadeiro para todas no caso dos escores de lucro e receita e para a Eficiência de Custos que Grangercausava Basileia mas que não foi verdade para eficiência de receita e de lucro Equação a Excluído b Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral Eficiência de Receita Basileia 0000 0000 0276 0001 0000 0091 0299 0001 Eficiência de Lucro Basileia 0022 0056 0000 0397 0559 0000 0195 0088 Eficiência de Custo Basileia 0522 0660 0287 0000 0001 0193 0006 0714 Basileia Eficiência de Receita 0005 0985 0000 0000 0000 0152 0013 0067 Basileia Eficiência de Lucro 0001 0001 0249 0531 0002 0009 0501 0025 Basileia Eficiência de Custo 0000 0000 0002 0046 0005 0000 0002 0011 Eficiência de Receita Todas 0000 0000 0000 0000 0000 0000 0025 0000 Eficiência de Lucro Todas 0002 0002 0000 0000 0085 0000 0000 0000 Eficiência de Custo Todas 0000 0000 0000 0000 0000 0000 0000 0000 91 5 CONCLUSÃO Este estudo teve por objetivo quantificar através de suas séries temporais as variáveis que continham informações sobre as flutuações futuras da eficiência bancária das principais instituições financeiras do Brasil Primeiramente foi definida uma metodologia para apuração dos escores de eficiência a análise de fronteira estocástica Para representar esses escores de eficiência bancária optamos por utilizar a eficiência de custos sendo que realizamos testes de robustez com os demais escores lucro e receita e não obtivemos conclusões gerais distintas Na próxima etapa da metodologia optamos pelo Modelo VAR para analisar as interrelações entre diversas variáveis e com isso estimar o efeito de um choque nessa variável sobre a eficiência bancária Foram realizados alguns testes e concluise que o modelo apresentado era consistente e as hipóteses que se buscavam responder seriam endereçadas Para esse exercício foram selecionadas algumas variáveis específicas de cada instituição que serviriam para a definição dos escores de eficiência e outras que serviriam como variáveis endógenas e que juntamente com algumas variáveis macroeconômicas seriam possíveis candidatas a explicar a evolução da eficiência bancária Os dados capturados são trimestrais entre 2009 e 2019 e além dos cinco maiores bancos atuando no Brasil Itaú Banco do Brasil Bradesco Caixa e Santander foram definidas três médias ponderadas com base nessas cinco instituições bancos privados bancos públicos e geral totalizando oito amostras que permearam todo estudo Os resultados apurados através na análise estrutural do Modelo VAR evidenciaram que as expectativas futuras o PIB e a inflação Grangercausam a eficiência de custos para a maioria das amostras como afirma a maioria da literatura e nossas hipóteses de pesquisa Porém a análises não se mostrou verdadeira para solvência e para o riscopaís contrariando nossas hipóteses e as referências bibliográficas citadas Já o efeito contrário da eficiência de custos Grangercausar as variáveis se mostrou verdadeiro na maioria das amostras para as expectativas futuras o PIB e a solvência 92 De forma complementar a análise das funções impulsoreposta e as decomposições de variância do erro de previsão nos mostraram que os resultados entre as instituições financeiras são divergentes sendo que não há um padrão entre os resultados de cada variável para cada amostra ou seja o escore de eficiência de custos de cada instituição financeira se comporta de maneira distinta a cada choque nas variáveis analisadas As constatações das análises estruturais do Modelo VAR descritas acima complementam a literatura abordando a reflexão de que não há um padrão único entre as variáveis analisadas e seus efeitos na eficiência bancária dada as características específicas de cada uma das instituições financeiras estudadas Adicionalmente o modelo também foi testado com pseudoprevisões fora da amostra que não evidenciaram melhores critérios de previsão do que uma autorregressão simples Sobre as limitações do estudo podemos citar o período relativamente curto para a metodologia selecionada 44 trimestres e o fato de que o estudo só abrangeu as cinco maiores instituições financeiras do Brasil o que inviabilizaria a ampliação das evidências empíricas às instituições medianas pequenas ou de nichos de mercado Como sugestões para estudos futuros podemos imaginar além dos pontos mencionados nas limitações desse estudo a utilização de amostras de bancos de outros países que poderiam trazer uma diversidade relevante às conclusões e a eventual utilização de outras metodologias para definição dos escores de eficiência bancário como forma de comparação dos resultados das análises de interrelação das variáveis do modelo 93 6 REFERÊNCIAS ABREU Emmanuel Sousa de KIMURA Herbert SOBREIRO Vinicius Amorim What Is Going On with Studies on Banking Efficiency Research in International Business and Finance v 47 p 195219 2019 AHMAD Nisar et al Banking Sector Performance Profitability and Efficiency a citationbased systematic literature review Journal of Economic Surveys v 0 n 0 p 134 2019 AIELLO Francesco BONANNO Graziella Efficiency in banking a meta regression analysis International Review of Applied Economics v 30 p 112 149 2015 AIELLO Francesco BONANNO Graziella On the sources of heterogeneity in banking efficiency literature Journal of Economic Surveys v 31 n 1 p 194 225 2018 ALGASAYMEH Anwar Bank efficiency determinant Evidence from the gulf cooperation council countries Research in International Business and Finance v 38 p 214223 2016 ALLEN Linda RAI Anoop Operational efficiency in banking An international comparison Journal of Banking Finance v 20 p 655672 1996 ANDRIES Alin Marius COCRIS Vasile A Comparative Analysis of the Efficiency of Romanian Banks Romanian Journal of Economic Forecasting v 13 p 5475 2010 ATHANASOGLOU Panayiotis P BRISSIMIS Sophocles N DELIS Matthaios D Bankspecific industryspecific and macroeconomic determinants of bank profitability Journal of International Financial Markets Institutions Money v 18 p 121136 2008 BANCO CENTRAL DO BRASIL Resolução nº 4553 de 30 de janeiro de 2017 Estabelece a segmentação do conjunto das instituições financeiras e demais instituições autorizadas a funcionar pelo Banco Central do Brasil para fins de aplicação proporcional da regulação prudencial Diário Oficial da União Brasília 31 jan 2017 Seção 1 p 25 94 BANCO CENTRAL DO BRASIL IFdata Disponível em httpswww3bcbgovbrifdata Acesso em 04 abr 2020 BARROS C et al Efficiency in banking of developing countries with the same cultural background A novel distance function frontier model Journal of Economic Studies v 45 n 3 p 638659 2018 BATTESE G E COELLI T J A Model for Technical Inefficiency Effects in a Stochastic Frontier Production Function for Panel Data Empirical Economics v 20 p 325332 1995 BAUM Christopher F VAR SVAR and VECM models Applied Econometrics Notes Boston College 2013 BAUM Christopher F Statistical Software Components S433001 DMARIANO Stata module to calculate DieboldMariano comparison of forecast accuracy Boston College Department of Economics 2003 revisado em 25 abr 2011 Disponível em httpfmwwwbcedurepecbocodeddmarianohlp Acesso em 19 set 2020 BECCALLI Elena CASU Barbara GIRARDONE Claudia Efficiency and Stock Performance in European Banking Journal of Business Finance Accounting v 33 n 1 p 245262 2006 BELOTTI Federico et al Stochastic frontier analysis using Stata The Stata Journal v 13 n 4 p 719758 2013 BERGER Allen N DEYOUNG Robert Problem Loans and Cost Efficiency in Commercial Banks Journal of Banking Finance v 21 p 849870 1997 BERGER Allen N HANNAN Timothy H The Efficiency Cost of Market Power in the Banking Industry A Test of the Quiet Life and Related Hypotheses The Review of Economics and Statistics v 80 n 3 p 454465 1998 BERGER Allen N HUMPHREY David B Measurement and Efficiency Issues in Commercial Banking In GRILICHES Zvi Ed Output Measurement in the Service Sectors Chicago University of Chicago Press National Bureau of Economic Research 1992 cap 7 p 245300 95 BERGER Allen N HUMPHREY David B Efficiency of financial institutions International survey and directions for future research European Journal of Operational Research v 98 p 175212 1997 BERGER Allen N MESTER Loretta J Inside the black box What explains differences in the efficiencies of financial institutions Journal of Banking Finance v 21 p 895947 1997 BIAGE Milton CORREA Vanessa Petrelli NEDER Henrique Dandas Risco país fluxos de capitais e determinação da taxa de juros no Brasil Uma análise de impactos por meio da metodologia VEC Revista Economia v 9 n1 p 63 113 2008 BORGES Pedro Raposo Bank performance during the European Debt Crisis a DEA approach 2012 52 f Dissertação Mestrado em Finanças Instituto Superior de Economia e Gestão da Universidade Técnica de Lisboa Lisboa 2012 BUENO R L S Econometria de séries temporais 2 ed São Paulo Cengage Learning 2011 325 p CARVALLO Oscar KASMAN Adnan Cost efficiency in the Latin American and Caribbean banking systems Journal of International Financial Markets Institutions Money v 15 p 5572 2005 CASU Barbara GIRARDONE Claudia Testing the relationship between competition and efficiency in banking A panel data analysis Economics Letters v 105 p 134137 2009 CAVALCANTI Marco A F H Identificação de modelos VAR e causalidade Granger uma nota de advertência Economia Aplicada v 14 n 2 p 251260 2010 CAVALCANTI Marco A F H Uma análise econométrica da evolução da indústria de transformação brasileira no período 20022012 IPEA 2013 Nota Técnica Carta de Conjuntura IPEA N18 CHEN KuHsieh et al Is full efficiency achievable An empirical study of the banking industry in Taiwan Contemporary Economic Policy v 37 n 1 p 195 209 2018 96 CHEN YiKai HIGGINS Eric J MASON Joseph R Is Bank Efficiency Cyclical The Relationship Between Economic and Financial Market Conditions and Bank Performance Drexel University working paper 2005 CHORTAREAS Georgios E GIRARDONE Claudia VENTOURI Alexia Bank supervision regulation and efficiency Evidence from the European Union Journal of Financial Stability v 8 p 292302 2012 CHRISTIANO Lawrence J Christopher A Sims and Vector Autoregressions The Scandinavian Journal of Economics v 114 p 10821104 2012 CLEMENTS Michael P HENDRY David F A Companion to Economic Forecasting Malden MA Blackwell Publishing Ltd 2002 597 p CNI CONFEDERAÇÃO NACIONAL DA INDÚSTRIA Metodologia do Índice Nacional de Expectativa do Consumidor INEC Versão 20 Brasília CNI 2019 COELLI Tim RAO D S Prasada BATTESE George E An Introduction to Efficiency and Productivity Analysis 1st ed New York NY Kluwer Academic Publishers 1998 275 p COLOMBO J A LONCAN TR CALDEIRA JF Do foreign portfolio capital flows affect domestic investment Evidence from Brazil International Journal of Finance Economics v 24 p 129 2018 DEYOUNG Robert A diagnostic test for the distributionfree efficiency estimator An example using US commercial bank data European Journal of Operational Research v 98 p 243249 1997 DIEBOLD Francis X MARIANO Roberto S Comparing Predictive Accuracy Journal of Business Economic Statistics v 13 n 3 p 253263 ELDER John KENNEDY Peter E Testing for unit roots what should students be taught Journal of Economic Education v 32 n 2 p 137146 2001 ELYASIANI Elyas MEHDIAN Seyed Efficiency in the commercial banking industry a production frontier approach Applied Economics v 22 p 539551 1990 97 ENDERS Walter Applied Econometrics Time Series 4th ed New York John Wiley and Sons Press 2014 485 p FARREL M J The Measurement of Productive Efficiency Journal of the Royal Statistical Society v 120 n 3 p 253290 1957 FERRIER Garry D LOVELL C A Knox Measuring cost efficiency in banking Econometric and Linear Programming Evidence Journal of Econometrics v 46 p 229245 1990 FIORDELISI Franco MARQUESIBANEZ David MOLYNEUX Efficiency and risk in European banking Journal of Banking Finance v 35 p 13151326 2011 FIORENTINO Elisabetta KARMANN Alexander KOETTER Michael The cost efficiency of German banks a comparison of SFA and DEA Deutsche Bundesbank Series 2 Banking and Financial Studies v 10 2006 Discussion Paper GOMES Matheus da Costa OLIVEIRA Sonia Valle Walter Borges de MATIAS Alberto Borges Eficiência do setor bancário brasileiro no período de 2006 a 2013 bancos domésticos x bancos estrangeiros Nova Economia v 27 n 3 p 641670 2017 GUJARATI D N PORTER D C Econometria básica 5 ed Porto Alegre AMGH 2011 924 p HAVRYLCHYK O Efficiency of the Polish banking industry foreign versus domestic banks Journal of Banking and Finance v 30 n 7 p 19751996 2006 HENRIQUES Iago Cotrim et al Efficiency in the Brazilian banking system using data envelopment analysis Future Business Journal v 4 p 157178 2018 HULJAK Ivan MARTIN Reiner MOCCERO Diego The CostEfficiency and Productivity Growth of Euro Area Banks European Central Bank Working Paper No 2305 2019 98 IPEAData EMBI Risco Basil Disponível em httpwwwipeadatagovbrExibeSerieaspxserid40940moduleM Acesso em 13 jun 2020 IVANOV Ventzislav KILIAN Lutz A Practitioners Guide to Lag Order Selection For VAR Impulse Response Analysis Studies in Nonlinear Dynamics Econometrics v9 n 1 p 136 2005 KALLEL Hella HAMAD Salah Ben TRIKI Mohamed Modeling the efficiency of Tunisian and Moroccan banks using the SFA approach International Journal of Productivity and Performance Management v 68 n 5 p 879908 2019 KENNEDY Peter A Guide to Econometrics 6th ed Malden Blackwell Publishing 2008 585 p KIM Hyeongwoo Generalized Impulse Response Analysis General or Extreme EconoQuantum v 10 n2 p135141 2013 KUMBHAKAR S C LOVELL C A K Stochastic Frontier Analysis Cambridge Cambridge University Press 2010 LEIBENSTEIN H Allocative Efficiency vs XEfficiency The American Economic Review v 56 n 3 p 392415 1966 LÜTKEPOHL Helmut New Introduction to Multiple Time Series Analysis Berlim SpringerVerlag 2005 MATTHEWS Kent Banking efficiency in emerging market economies Cardiff Economics Working Papers No E201012 2010 MENDONÇA Douglas José et al Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil Revista Capital Científico Eletrônica RCCe v 16 n 3 p 521 2018 NAKANE Márcio I WEINTRAUB Daniela B Bank privatization and productivity Evidence for Brazil Journal of Banking Finance v 29 p 2259 2259 2005 99 PRIMO Uverlan Rodrigues et al Determinantes da Rentabilidade Bancária no Brasil BASE Revista de Administração e Contabilidade da Unisinos v 10 n 4 p 308323 2013 SHAMSHUR Anastasiya WEILL Laurent Does bank efficiency influence the cost of credit Journal of Banking and Finance v 105 p 6273 2019 SILVA Thiago Christiano et al A comparison of DEA and SFA using micro and macrolevel perspectives Efficiency of Chinese local banks Physica A Statistical Mechanics and its Applications v 469 p 216223 2017 SIMS C A Macroeconomics and Reality Econometrica v 48 n 1 p 148 1980 STOCK James H WATSON Mark W Vector Autoregressions Journal of Economic Perspectives v 15 n 4 p 102115 2001 SUFIAN Fadzlan Determinants of bank efficiency during unstable macroeconomic environment Empirical evidence from Malaysia Research in International Business and Finance v 23 p 5477 2009 TAMADONNEJAD A et al Political Stability Country Risk and Bank Efficiency in East Asian Countries Prosiding Perkem VIII v 2 p 616628 2013 TAN Yong ANCHOR John The impacts of risktaking behaviour and competition on technical efficiency Evidence from the Chinese banking industry Research in International Business and Finance v 41 p 90104 2017 TECLES Patricia Langsch TABAK Benjamin M Determinants of bank efficiency The case of Brazil European Journal of Operational Research v 207 p 15871598 2010 WILLIAMS Jonathan Determining management behaviour in European banking Journal of Banking Finance v 28 p 24272460 2004 WOOLDRIDGE Jeffrey M Introdução à Econometria uma abordagem moderna Tradução Priscilla Rodrigues da Silva Lopes e Livia Marina Koeppl Revisão técnica Heloisa Pinna Bernardo 6a ed São Paulo SP Cengage Learning 2016 848 p 100 ZHANG Jianhua et al Market concentration risktaking and bank performance Evidence from emerging economies International Review of Financial Analysis v 30 p 149157 2013 101 APÊNDICE 102 APÊNDICE A ESCORES DE EFICIÊNCIA DE CUSTO E DEMAIS VARIÁVEIS Figura 10 Escores de eficiência de custos e demais variáveis no tempo Fonte Elaborado pelo próprio autor 103 APÊNDICE B MATRIZ DE CORRELAÇÃO Tabela 24 Matriz de correlação das variáveis dos modelos VAR Fonte Elaborado pelo próprio autor Itaú BB Bradesco Caixa Santander Privados Públicos Geral Itaú 1000 BB 0671 1000 Bradesco 0708 0600 1000 Caixa 0465 0612 0432 1000 Santander 0616 0418 0432 0230 1000 Privados 0921 0673 0865 0455 0761 1000 Públicos 0582 0806 0532 0962 0317 0574 1000 Geral 0824 0841 0761 0830 0578 0857 0914 1000 0028 0358 0043 0419 0404 0097 0438 0228 1000 0188 0366 0254 0364 0109 0223 0400 0362 0309 1000 0557 0334 0426 0043 0392 0542 0084 0322 0192 0050 0063 0021 0036 0054 0095 0012 0048 0036 0168 0342 Itaú 0059 0128 0106 0050 0143 0116 0007 0052 0024 0133 BB 0063 0167 0069 0073 0110 0038 0112 0053 0045 0146 Bradesco 0136 0021 0029 0019 0110 0104 0007 0056 0160 0139 Caixa 0036 0170 0112 0240 0000 0063 0239 0120 0112 0132 Santander 0185 0211 0065 0304 0212 0173 0301 0276 0104 0442 Privados 0149 0123 0063 0020 0089 0118 0058 0095 0068 0139 Públicos 0153 0144 0090 0228 0148 0072 0221 0176 0097 0185 Geral 0190 0165 0001 0133 0143 0125 0157 0162 0100 0196 Itaú BB Bradesco Caixa Santander Privados Públicos Geral 1000 0108 1000 Itaú 0172 0115 1000 BB 0037 0069 0383 1000 Bradesco 0151 0144 0348 0016 1000 Caixa 0125 0129 0183 0474 0078 1000 Santander 0032 0186 0295 0110 0153 0144 1000 Privados 0270 0160 0235 0192 0103 0325 0077 1000 Públicos 0057 0193 0154 0064 0060 0038 0179 0269 1000 Geral 0165 0216 0089 0172 0104 0218 0041 0861 0721 1000 PIB IGPM Índice de Basileia PIB IGPM Índice de Basileia Índice de Basileia Variáveis Variáveis IGPM PIB difRiscoBrasil INEC Escore de Eficiência Escore de Eficiência difRisco Brasil INEC 104 APÊNDICE C TESTES ADF Tabela 25 Teste ADF para principais variáveis do estudo Fonte Elaborado pelo próprio autor Notas dif é o indicativo das variáveis em que foi tirada a primeira diferença e ln é o indicativo de que foi tirado o logaritmo natural Os testes consideraram o nível de significância de 10 Absoluto Drift Trend Absoluto Drift Trend 0361 0038 0841 Não Estacionária Estacionária Não Estacionária 0449 0053 0686 Não Estacionária Estacionária Não Estacionária 0012 0001 0077 Estacionária Estacionária Estacionária 0276 0026 0895 Não Estacionária Estacionária Não Estacionária 0002 0000 0010 Estacionária Estacionária Estacionária Itaú 0009 0001 0039 Estacionária Estacionária Estacionária BB 0582 0085 0906 Não Estacionária Estacionária Não Estacionária Bradesco 0393 0043 0715 Não Estacionária Estacionária Não Estacionária Caixa 0001 0000 0018 Estacionária Estacionária Estacionária Santander 0298 0028 0529 Não Estacionária Estacionária Não Estacionária Privados 0345 0035 0666 Não Estacionária Estacionária Não Estacionária Públicos 0089 0007 0311 Estacionária Estacionária Não Estacionária Geral 0092 0007 0299 Estacionária Estacionária Não Estacionária Itaú 0080 0006 0088 Estacionária Estacionária Estacionária BB 0022 0344 0243 Estacionária Não Estacionária Não Estacionária Bradesco 0292 0028 0088 Não Estacionária Estacionária Estacionária Caixa 0979 0631 0745 Não Estacionária Não Estacionária Não Estacionária Santander 0452 0053 0568 Não Estacionária Estacionária Não Estacionária Privados 0057 0004 0267 Estacionária Estacionária Não Estacionária Públicos 0856 0256 0622 Não Estacionária Não Estacionária Não Estacionária Geral 0136 0011 0407 Não Estacionária Estacionária Não Estacionária Itaú 0042 0003 0051 Estacionária Estacionária Estacionária BB 0029 0243 0303 Estacionária Não Estacionária Não Estacionária Bradesco 0194 0016 0069 Não Estacionária Estacionária Estacionária Caixa 0371 0039 0645 Não Estacionária Estacionária Não Estacionária Santander 0891 0308 0810 Não Estacionária Não Estacionária Não Estacionária Privados 0006 0001 0050 Estacionária Estacionária Estacionária Públicos 0147 0012 0545 Não Estacionária Estacionária Não Estacionária Geral 0079 0006 0464 Estacionária Estacionária Não Estacionária Itaú 0208 0018 0392 Não Estacionária Estacionária Não Estacionária BB 0912 0350 0547 Não Estacionária Não Estacionária Não Estacionária Bradesco 0367 0038 0163 Não Estacionária Estacionária Não Estacionária Caixa 0344 0035 0438 Não Estacionária Estacionária Não Estacionária Santander 0013 0001 0902 Estacionária Estacionária Não Estacionária Privados 0872 0278 0557 Não Estacionária Não Estacionária Não Estacionária Públicos 0064 0005 0229 Estacionária Estacionária Não Estacionária Geral 0138 0011 0340 Não Estacionária Estacionária Não Estacionária Itaú 0155 0013 0390 Não Estacionária Estacionária Não Estacionária BB 0001 0000 0005 Estacionária Estacionária Estacionária Bradesco 0013 0001 0007 Estacionária Estacionária Estacionária Caixa 0352 0036 0532 Não Estacionária Estacionária Não Estacionária Santander 0064 0005 0000 Estacionária Estacionária Estacionária Privados 0101 0008 0303 Não Estacionária Estacionária Não Estacionária Públicos 0015 0001 0055 Estacionária Estacionária Estacionária Geral 0077 0006 0159 Estacionária Estacionária Não Estacionária Itaú 0009 0001 0053 Estacionária Estacionária Estacionária BB 0772 0175 0342 Não Estacionária Não Estacionária Não Estacionária Bradesco 0056 0004 0058 Estacionária Estacionária Estacionária Caixa 0609 0095 0381 Não Estacionária Estacionária Não Estacionária Santander 0639 0105 0905 Não Estacionária Não Estacionária Não Estacionária Privados 0874 0281 0573 Não Estacionária Não Estacionária Não Estacionária Públicos 0008 0001 0008 Estacionária Estacionária Estacionária Geral 0314 0031 0064 Não Estacionária Estacionária Estacionária dif ln Agências PAs Resultado Escore de Eficiência de Receita Escore de Eficiência de Lucro Risco Brasil Pvalor Índice de Basileia difRiscoBrasil PIB IGPM dif Índice de Basileia Variáveis Escore de Eficiência de Custos INEC 105 APÊNDICE D TESTE LM Tabela 26 Teste LM de cada amostra Fonte Elaborado pelo próprio autor Amostra Lag chi2 df Prob chi2 Amostra Lag chi2 df Prob chi2 1 3329 36 0598 1 5781 36 0012 2 3964 36 0311 2 2310 36 0953 3 2808 36 0824 3 4026 36 0287 4 4050 36 0278 4 4392 36 0171 1 5305 36 0033 1 4036 36 0284 2 3964 36 0311 2 3589 36 0474 3 4645 36 0114 3 3059 36 0723 4 4664 36 0110 4 2365 36 0944 1 4049 36 0279 1 3469 36 0531 2 4869 36 0077 2 4485 36 0148 3 3119 36 0697 3 4867 36 0077 4 3367 36 0580 4 3981 36 0304 1 5159 36 0045 1 4066 36 0273 2 2932 36 0777 2 5830 36 0011 3 7117 36 0000 3 5950 36 0008 4 3656 36 0443 4 5022 36 0058 Geral Públicos Privados Santander Caixa Bradesco BB Itaú 106 APÊNDICE E CORRELOGRAMA CRUZADO Figura 11 Correlograma cruzado do escore de eficiência de custos do Itaú para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor Figura 12 Correlograma cruzado do escore de eficiência de custos do Banco do Brasil para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of ITAUcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of ITAUcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of ITAUcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of ITAUcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of ITAUcosteffic and difITAUbasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BBcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BBcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BBcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BBcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BBcosteffic and difBBbasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 107 Figura 13 Correlograma cruzado do escore de eficiência de custos do Bradesco para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor Figura 14 Correlograma cruzado do escore de eficiência de custos do Caixa para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BRADESCOcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BRADESCOcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BRADESCOcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BRADESCOcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of BRADESCOcosteffic and difBRADESCObasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of CAIXAcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of CAIXAcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of CAIXAcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of CAIXAcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of CAIXAcosteffic and difCAIXAbasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 108 Figura 15 Correlograma cruzado do escore de eficiência de custos do Santander para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor Figura 16 Correlograma cruzado do escore de eficiência de custos da média ponderada dos Privados para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of SANTANDERcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of SANTANDERcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of SANTANDERcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of SANTANDERcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of SANTANDERcosteffic and difSANTANDERbasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PRIVcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PRIVcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PRIVcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PRIVcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PRIVcosteffic and difPRIVbasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 109 Figura 17 Correlograma cruzado do escore de eficiência de custos da média ponderada dos Públicos para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor Figura 18 Correlograma cruzado do escore de eficiência de custos da média ponderada Geral para cada uma das variáveis do Modelo VAR Fonte Elaborado pelo próprio autor 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PUBcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PUBcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PUBcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PUBcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of PUBcosteffic and difPUBbasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of GERALcosteffic and INEC 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of GERALcosteffic and difRiscoBrasil 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of GERALcosteffic and PIB 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of GERALcosteffic and IGPM 10 5 0 5 10 Lag Crosscorrelogram 100 050 000 050 100 100 050 000 050 100 Crosscorrelations of GERALcosteffic and difGERALbasileia 10 5 0 5 10 Lag Crosscorrelogram INEC Risco Brasil PIB IGPM Basileia 110 APÊNDICE F ESTABILIDADE DO MODELO VAR Figura 19 Gráficos da Estabilidade dos Modelos VAR para todas as amostras Fonte Elaborado pelo próprio autor 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix Santander Itaú BB Privados 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix Bradesco Caixa Públicos Geral 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix Santander Privados 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix 1 5 0 5 1 Imaginary 1 5 0 5 1 Real Roots of the companion matrix Públicos Geral 111 APÊNDICE G TESTES DE CAUSALIDADE DE GRANGER Tabela 27 Probabilidades do teste de causalidade de Granger para as demais variáveis Fonte Elaborado pelo próprio autor Notas Os números expressos na tabela representam a probabilidade de aceitar a hipótese nula sendo que a fonte verde representa os casos em que rejeitamos e em vermelho os casos em que aceitamos Consideramos o nível de significância de 5 Itaú Banco do Brasil Bradesco Caixa Santander Privados Públicos Geral Risco Brasil 0001 0009 0332 0001 0000 0000 0001 0000 PIB 0102 0057 0080 0321 0000 0002 0135 0001 IGPM 0000 0000 0004 0000 0000 0000 0000 0000 Basileia 0000 0136 0908 0042 0004 0000 0077 0002 Todas 0000 0000 0000 0000 0000 0000 0000 0000 INEC 0004 0010 0078 0001 0003 0101 0040 0326 PIB 0294 0016 0056 0047 0066 0280 0166 0615 IGPM 0261 0039 0405 0044 0221 0093 0193 0205 Basileia 0847 0069 0602 0006 0020 0001 0502 0039 Todas 0017 0000 0000 0000 0000 0000 0025 0001 INEC 0047 0221 0005 0001 0228 0083 0130 0323 Risco Brasil 0002 0830 0115 0158 0271 0042 0417 0167 IGPM 0003 0100 0814 0003 0231 0020 0029 0053 Basileia 0179 0118 0042 0429 0479 0465 0012 0234 Todas 0000 0035 0000 0000 0002 0000 0000 0001 INEC 0862 0873 0060 0000 0802 0834 0068 0571 Risco Brasil 0052 0005 0014 0000 0310 0068 0000 0001 PIB 0629 0732 0543 0000 0711 0502 0061 0474 Basileia 0004 0098 0046 0000 0039 0005 0000 0001 Todas 0010 0080 0055 0000 0007 0021 0000 0002 INEC 0032 0009 0000 0021 0154 0180 0000 0000 Risco Brasil 0001 0000 0317 0149 0266 0428 0005 0000 PIB 0004 0000 0000 0166 0000 0078 0006 0091 IGPM 0707 0000 0091 0007 0339 0652 0003 0002 Todas 0000 0000 0000 0001 0002 0000 0000 0000 Risco Brasil PIB IGPM Basileia Equação Excluído Probabilidade INEC 112 APÊNDICE H MATRIZ DE CORRELAÇÃO DOS ESCORES DE EFICIÊNCIA Tabela 28 Matriz de correlação dos escores de eficiência de custo lucro e receita Fonte Elaborado pelo próprio autor Custo Lucro Receita Custo 1000 Lucro 0152 1000 Receita 0023 0239 1000 Custo 1000 Lucro 0073 1000 Receita 0003 0259 1000 Custo 1000 Lucro 0128 1000 Receita 0060 0703 1000 Custo 1000 Lucro 0091 1000 Receita 0124 0504 1000 Custo 1000 Lucro 0105 1000 Receita 0114 0198 1000 Custo 1000 Lucro 0179 1000 Receita 0063 0381 1000 Custo 1000 Lucro 0079 1000 Receita 0035 0553 1000 Custo 1000 Lucro 0083 1000 Receita 0076 0465 1000 Privados Públicos Geral Bradesco Caixa Santander Escore de Eficiência Amostra Escore de Eficiência Itaú Banco do Brasil Research in International Business and Finance 23 2009 5477 Contents lists available at ScienceDirect Research in International Business and Finance journal homepage wwwelseviercomlocateribaf Determinants of bank efficiency during unstable macroeconomic environment Empirical evidence from Malaysia Fadzlan Sufian ab a Planning and Research Department CIMB Bank Berhad Malaysia b Department of Economics Faculty of Economics and Management Universiti Putra Malaysia UPM Malaysia a r t i c l e i n f o Article history Received 5 November 2007 Received in revised form 7 July 2008 Accepted 9 July 2008 Available online 17 July 2008 JEL classification G21 G28 Keywords Financial crisis Bank efficiency Malaysia a b s t r a c t The present study investigates for the first time the efficiency of Malaysian banking sector around the Asian financial crisis 1997 The efficiency estimates of individual banks are evaluated by using the Data Envelopment Analysis DEA approach To examine the robustness of the estimated efficiency scores under various alter natives and to differentiate how efficiency scores vary with changes in inputs and outputs the present study focuses on three major approaches viz intermediation approach value added approach and operating approach The analysis further links the variation in calculated efficiencies to a set of explanatory variables ie bank size profitability and ownership The empirical findings clearly bring forth the high degree of inefficiency in the Malaysian bank ing sector particularly a year after the East Asian crisis The results suggest that the decline in technical efficiency is more abrupt under the intermediation approach relative to the value added approach and operating approach The regression results focusing on bank efficiency and other bank specific traits suggest that efficiency is negatively related to expense preference behavior and economic conditions while bank efficiency is positively related to loans inten sity 2008 Elsevier BV All rights reserved Correspondence address 21st Floor 6 Jalan Tun Perak 50050 Kuala Lumpur Malaysia Tel 60 3 2693 1722x6455 fax 60 3 2691 4415 Email addresses fadzlansufiancimbcom fadzlan14gmailcom 02755319 see front matter 2008 Elsevier BV All rights reserved doi101016jribaf200807002 F Sufian Research in International Business and Finance 23 2009 5477 55 1 Introduction The world was stunned when East Asia the highest growth region during the 1990s was hit by a banking crisis in 1997 The crisis which started in Thailand and spread rapidly throughout the region drew widespread attention from economists and financial analysts It was characterized by skyrock eting interest rates 34 in Korea 13 in Indonesia a dramatic drop in stock price indexes 55 in Thailand 52 in Malaysia real exchange rate depreciation 97 in Korea 87 in Thailand a decline in net capital flow to the region 20 billion and a drop in the regions gross domestic product 481 billion In order to understand the nature of the East Asian crisis as well as similar crises around the world economists have used a variety of methodologies while focusing primarily on macrolevel vari ables Papers in the macrotradition include Frankel and Rose 1996 Sachs et al 1996 Kaminsky et al 1998 Corsetti et al 1998 Eichengreen and Rose 1998 Chinn 1998 Kaminsky and Reinhart 1998 Krugman 1999 Berg and Pattillo 1999 Alba et al 1999 and Tanner 2000 among oth ers In addition to these studies a number of papers have adopted a microperspective deriving from banks balance sheet and income statements Generally the variables examined in these papers fol low the wellknown CAMEL framework1 These include Sinkey 1975 Barth et al 1985 Lane et al 1986 Thomson 1991 and GonzalezHermosillo et al 1997 There were also papers which com bined macro and microvariables including those of Goldstein and Turner 1996 Honohan 1997 and GonzalezHermosillo 1999 The characteristics of banking systems that are more prone to experience banking and financial crisis and its impact has also drawn academic interests in recent years Barth et al 2000 suggests that greater regulatory restrictions on bank activities are associated with higher probability of suffering a major banking crisis Beck et al 2006 find that the likelihood of financial crises is lower in more concentrated banking systems yet higher in less competitive and countries with less developed legal systems Daniel and Jones 2007 suggests that even if a banking system is well designed countries will enjoy an initial period of rapid low risk growth before entering a period with an elevated risk of banking crisis By using aggregate and bank level data for 35 developed and developing countries Dermiguc Kunt et al 2006 find that depositors leave weaker banks for stronger ones following a banking crisis Kroszner et al 2007 find that sectors that are highly dependent on external finance tend to experience greater contraction during a banking crisis in countries with deeper financial systems than in countries with shallower financial systems The literature examining the efficiency of financial institutions with parametric andor non parametric frontier techniques has expanded rapidly in recent times While a large body of literature spanning a halfcentury exists on banking efficiency in the United States see surveys in Berger et al 1993 Berger and Humphrey 1997 Berger 2007 and references therein more recent studies examine several other countries such as India Ataullah and Le 2006 Hong Kong Drake et al 2006 Greece Pasiouras 2008b Singapore Sufian 2007 and Ukraine Kyj and Isik 2008 Apart from focusing on various countries these studies also examine several other issues of bank efficiency ie the impact of risk on bank efficiency eg Drake and Hall 2003 the impact of offbalance sheet activities on bank efficiency eg LozanoVivas and Pasiouras 2008 the relationship between bank efficiency and share prices eg Pasiouras et al 2008 the impact of mergers on bank efficiency eg AlSharkas et al 2008 The comparison of efficiency between foreign and domestic banks has also been studied extensively eg Bhattacharya et al 1997 Isik and Hassan 2002 Ataullah and Le 2006 Despite its severity and deep influence on both the real and financial sectors the impact of the East Asian crisis of 1997 on the efficiency of the financial industry has not been critically examined yet Fukuyama 1995 Humphrey and Pulley 1997 Leightner and Lovell 1998 and Isik and Hassan 2003a have suggested that frontier techniques can be used to assess the impact of major economic events such as economic crisis or financial liberalization on the performance of banking firms How 1 The CAMEL framework is used by financial regulators to rate the health of financial institutions The framework criteria are associated with Capital Adequacy C Asset Quality A Management M Earnings E and Liquidity L 56 F Sufian Research in International Business and Finance 23 2009 5477 Table 1 Shortterm claims on the East Asian economies change YoY Year Hong Kong Indonesia Korea Malaysia Philippines Singapore Taiwan Thailand 1990 744 6032 3101 1103 1210 1774 2602 1706 1991 538 1228 1725 4496 028 270 8487 3923 1992 389 1387 1149 3690 230 959 2469 346 1993 2239 938 1017 8149 2487 911 2413 1871 1994 2009 1327 3679 1102 3442 1783 4921 2011 1995 133 2953 3520 2000 2826 2546 4053 316 1996 1811 2419 2438 4158 9024 1406 2537 397 1997 169 250 1290 2899 5381 1134 475 1228 1998 4419 3248 4949 3543 2718 4013 3960 2224 1999 2149 1969 1811 1677 1199 1814 3085 710 Source Bank of International Settlements Units are in millions of US Dollars ever to the best of our knowledge Isik and Hassans 2003a study is the only empirical research performed to directly examine the impact of financial disruptions on bank efficiency by using the frontier technique By employing a unique data set of all banks operating in Malaysia during the period of 19951999 this study contributes to the existing literature by providing new empirical evidence on the impact of the East Asian crisis on the Malaysian banking sectors efficiency Given that banks are the dominant financial institution in Malaysia their health is very critical to the health of the general economy at large as demonstrated during the East Asian crisis which left many financial institutions in distress However at the present time this type of analysis is completely missing in the literature To do so the nonparametric frontierbased Data Envelopment Analysis DEA method is used to estimate the technical pure technical and scale efficiency To examine the robustness of the estimated efficiency scores under various alternatives the present study focuses on three major approaches namely intermediation approach value added approach and operating approach Finally we analyze how ownership structure and different bank characteristics such as capitalization problem loans ratio and size influences the efficiency estimates The following section identifies those factors that characterize the Asian financial crisis Section 3 describes the data sources and model specifications which are employed in the study Section 4 presents the results of the analysis of Malaysian banking sector using DEA and the panel regression techniques Finally we conclude in Section 5 2 The Asian financial crisis in perspective Over the last quarter of the 20th century both developed and developing countries have experi enced severe banking crisis Chile Argentina and Mexico 1980s Sweden 1990s Thailand Malaysia Korea Philippines and Indonesia 1997 Paraguay 19951998 Russia 1998 Turkey 1994 2000 and 2001 Argentina 2001 The main causes of these crises are poor banking practices and lack of revenue diversification inadequate capital shortcomings in the assessment of credit risk lending to connected enterprises excessive maturity or currency mismatches and rapid rise of nonperforming loans Due to less than fully developed stock markets in the East Asian region equity financing was not an important source of financing Rather banks borrowing from both offshore and onshore became an important focus of financing in East Asia Suetorsak 2006 A huge accumulation of foreign debt financing relative to equity increasingly raised banks leverage risk Interest rate risk and maturity risk were also a concern of the East Asian banks Interest rate risk reflects a maturity mismatch between banks assets and liabilities For example East Asian banks often borrowed abroad with a shortterm maturity and lent domestically with a longterm maturity for domestic projects The total shortterm maturity claims of the East Asian region are shown in Table 1 Banks in East Asia also assumed exceptional liquidity risk From Table 2 it is clear that East Asian banks had a high proportion of shortterm debts relative to longterm debts They exposed themselves to liquidity risk due to problems with shortterm debt renewal It is observed from Table 2 that prior to F Sufian Research in International Business and Finance 23 2009 5477 57 Table 2 Percentage of shortterm maturity claims to longterm maturity claims on banks East Asian economies Shortterm maturity claims on banks Longterm maturity claims on banks Percentage of shortterm claims to longterm maturity claims on banks 1995 Indonesia 5516 3384 163 Korea 35016 14984 2337 Malaysia 2115 2385 887 Philippines 1078 1122 961 Thailand 8073 17727 455 Hong Kong 7845 19819 396 Singapore 1251 7171 174 1996 Indonesia 7282 4518 1612 Korea 44531 21369 2084 Malaysia 3273 3227 1014 Philippines 3048 2152 1416 Thailand 6989 19011 368 Hong Kong 14186 23917 593 Singapore 1978 7879 251 1997 Indonesia 7428 4972 1494 Korea 38268 29032 1318 Malaysia 5257 5243 1003 Philippines 4609 891 5173 Thailand 7968 18132 439 Hong Kong 10305 26031 396 Singapore 2765 9747 284 Source Bank of International Settlements Units are in millions of US Dollars the crisis in 1997 among the crisis affected countries Korea had the highest percentage of shortterm debts relative to longterm debts followed by Indonesia while Thailand and Malaysia the least On the other hand Singapore and Hong Kong which were relatively unscathed by the East Asian crisis had relatively low percentage of shortterm debts relative to longterm debts Another risk resulting from excessive external debt accumulations was exchange rate risk After increasing their offshore borrowing in foreign currencies East Asian banks lent out money domestically with little restraint in local currencies This lending boom led to excessive domestic credit growth Frankel and Rose 1996 Nevertheless banks were also exposed to foreign exchange losses risk in the event of domestic currency depreciation Chang and Velasco 1999 This was problematic since the domestic credit expansion was inconsistent with the East Asian quasifixed exchange rate regimes causing a depletion of foreign exchange reserves and eventually causing a collapse of the fixed exchange rate regime Krugman 1999 This suggests that the internal imbalance of the banking industry triggered an unsustainable external balance In other words microdecisions of banks caused macroimbalances in these countries Regardless of the cause of the East Asian crisis all economists have agreed that the crisis has been severe Although initially only financial in nature the crisis has led to significant real economic losses in the once fast growing developing economies The Malaysian situation possesses some important characteristics Basic macroeconomic fun damentals were not signaling imminent danger Low inflation and falling unemployment rate characterized the country over the decade preceding the East Asian crisis Despite the sound macroeco nomic fundamentals a few other major indicators pointed towards serious problems by the mid1990s further compounded by the financial systems weakness The financial institutions were weakened by largescale exposure to the property sector high nonperforming loans and shortterm loans that were unhedged against currency movements Furthermore the credit expansion especially to the private sector was pronounced during 19951997 particularly towards the real estate and property sectors Prior to the crisis the banking system had relatively high exposures to the property and equity sectors which amounted to approximately 426 of total loans compared to only 21 exposure towards the manufacturing sector Once devaluation and drops in real estate and stock markets got underway the crisis took on a life of its own The drop in property and equity prices caused large losses and outright defaults by many property companies as well as those who participate in the equity market by margin accounts resulting in a dramatic increase in the amount of nonperforming loans NPLs In sum the domestic banks were hit simultaneously on many fronts namely the withdrawal of foreign funds foreign exchange losses a sharp rise in NPLs and losses on equity holdings The erosion of the banking systems capital base due to all these factors had severely constrained their ability to lend to even solvent companies in the midst of the crisis due to the need to comply with international capital adequacy rules 3 Methodology and data 31 Data Envelopment Analysis The present study employs the nonparametric frontier DEA approach with variable returns to scale VRS assumption to measure inputoriented technical efficiency of Malaysian banks DEA involves constructing a nonparametric production frontier based on the actual inputoutput observations in the sample relative to which efficiency of each firm in the sample is measured Coelli 1996 Let us give a short description of the DEA 2 Assume that there is data on K inputs and M outputs for each N bank For the ith bank these are represented by the vectors xi and yi respectively Let us call the K N input matrix X and the M N output matrix Y To measure the efficiency for each bank we calculate a ratio of all inputs such as uyivxi where u is an M 1 vector of output weights and v is a K 1 vector of input weights To select optimal weights we specify the following mathematical programming problem min uy uyiv xi uyivxi 1 j 1 2 n u v 0 1 The above formulation has a problem of infinite solutions and therefore we impose the constraint vxi 1 which leads to min μ φ μy1φxi 1 μyi φkj 0 j 1 2 N μ φ 0 2 where we change notation from u and v to μ and φ respectively in order to reflect transformations Using the duality in linear programming an equivalent envelopment form of this problem can be derived min θ λ yi Yλ 0 θxi Xλ 0 λ 0 3 where θ is a scalar representing the value of the efficiency score for the ith bank which will range between 0 and 1 λ is a vector of N 1 constants The linear programming has to be solved N times once for each bank in the sample In order to calculate efficiency under the assumption of VRS the convexity constraint N1λ 1 will be added to ensure that an inefficient bank is only compared against banks of similar size and therefore provides the basis for measuring economies of scale within the DEA concept The convexity constraint determines how closely the production frontier envelops the observed inputoutput combinations and is not imposed in the constant returns to scale CRS case 32 Multivariate Tobit regression analysis Coelli et al 1998 suggested several ways in which environmental variables can be accommodated in a DEA analysis The term environmental variables is usually used to describe factors which could influence the efficiency of a firm In this case such factors are not traditional inputs and are assumed to be outside the control of the manager As defined in Eq 3 the DEA score falls between the interval 0 2 Good reference books on efficiency measures are Coelli et al 1998 Thanassoulis 2001 and Cooper et al 2000 and 1 0 θ 1 making the dependent variable a limited dependent variable A commonly held view in previous studies is that the use of the Tobit model can handle the characteristics of the distribution of in efficiency measures and thus provide results that can provide important policy guidelines to improve performance Accordingly DEA efficiency scores obtained in the first stage is used as a dependent variable in the second stage and are regressed against a set of bank characteristics and other environmental variables The standard Tobit model can be defined as follows for observation bank i yi β x ε y y if y 0 and y 0 otherwise 4 where εi N0 σ2 xi and β are vectors of explanatory variables and unknown parameters respectively while y is a latent variable and yi is the DEA efficiency score 33 Specification of bank inputs outputs and data It is commonly acknowledged that the choice of variables in efficiency studies significantly affects the results The problem is compounded by the fact that variable selection is often constrained by the paucity of data on relevant variables The cost and output measurements in banking are especially difficult because many of the financial services are jointly produced and prices are typically assigned to a bundle of financial services The role of commercial banks is generally defined as collecting the savings of households and other agents to finance the investment needs of firms and consumption needs of individuals Three approaches dominate the literature the production approach the intermediation approach and more recently the revenue or value added approach The operating approach of defining inputs and outputs of banks has also been popular Jemric and Vujcic 2002 The first two approaches apply the traditional microeconomic theory of the firm to banking and differ only in the specification of banking activities The third and fourth approach goes a step further and incorporates some specific activities of banking into the classical theory and thereby modifies it Under the production approach pioneered by Benston 1965 a financial institution is defined as a producer of services for account holders that is they perform transactions on deposit accounts and process documents such as loans Hence according to this approach the number of accounts or its related transactions is the best measure for output while the number of employees and physical capital are considered as inputs This approach has primarily been employed in studying the efficiency of bank branches Berger and Humphrey 1992 The intermediation approach on the other hand assumes that financial firms act as an intermediary between savers and borrowers and posits total loans and securities as outputs whereas deposits along with labour and physical capital are defined as inputs The operating approach or income based approach views banks as business units with the final objective of generating revenue from the total cost incurred for running the business Leightner and Lovell 1998 Accordingly it defines banks output as total revenue interest and noninterest income and inputs as the total expenses interest and noninterest expenses More recently Drake et al 2006 proposed the revenue approach or value added approach in DEA The value added approach identifies those balance sheet categories as outputs that contribute to the bank value added In general under this approach deposits and loans are viewed as outputs because they are responsible for the significant proportion of value added The appropriateness of each approach varies according to the circumstances It is apparent that banks undertake simultaneous functions However based on practical considerations and to examine the robustness of the estimated efficiency scores under various alternatives the present study focuses on three major approaches intermediation approach value added approach and operating approach Under the intermediation approach we assume deposits x2 labour x3 and capital x1 as inputs 3 likelihood function L is maximized to solve β and σ based on 171 observations banks of yi and xi is L yi 0 1 F yi 0 12Πσ2 12 e12σ2yβxi2 where Fi βxiσ 12Π12et22 dt The first product is over the observations for which the banks are 100 efficient y 0 and the second product is over the observations for which banks are inefficient y 0 Fi is the distribution function of the standard normal evaluated at βxiσ 60 F Sufian Research in International Business and Finance 23 2009 5477 Table 3 Descriptive statistics for inputs and outputs 1995 RMm 1996 RMm 1997 RMm 1998 RMm 1999 RMm Outputs Total loans Mean 528906408 679483803 835088961 969727785 972539809 SD 872785964 1011235357 1192196075 1437264431 1443451230 Investments Mean 161352108 205769856 253581888 289988042 293979955 SD 333493592 336822895 422983641 465112056 523261154 Interest income Mean 57731719 78759181 104331055 147455503 105609500 SD 95501644 122835793 156100693 222718785 172386525 Noninterest income Mean 8887419 10628472 12633661 14038148 13928582 SD 15051184 17125609 19600668 19773007 21481235 Inputs Capital Mean 9910225 11936536 13206267 16308094 17344830 SD 18322180 20036486 20568534 22365871 24843507 Total deposits Mean 718373928 873139222 1114000858 1215176988 1289761248 SD 1210740871 1354051762 1664963238 1759391510 1882915776 Labour Mean 7598469 8612244 10318052 11536691 10278524 SD 12388690 13770891 16303227 16699356 13970262 Interest expense Mean 34141642 47306778 64503800 98911603 66229870 SD 55924949 74647921 94199438 147673993 110340995 Noninterest expense Mean 6758075 8061311 9271452 11818221 10339167 SD 11035623 12738836 13915652 16604321 13341085 Source Individual Banks Annual Reports The table presents summary statistics of the variables used to construct the efficiency frontier for the DEA intermediation operating and value added approaches over the period 19951999 for producing loans y1 and investments y2 Under the value added approach labour x3 capital x1 and interest expenses x4 are used as inputs producing outputs like deposits x2 loans y1 and investments y2 Under the operating approach three types of inputs are considered namely interest expenses x4 labour x3 and other operating expenses excluding employee expenses x5 The relevant outputs are interest income y3 and noninterest income y4 emanating mostly from commission exchange brokerage etc We use annual bank level and macroeconomic data of all Malaysian commercial banks over the period 19951999 The variables are obtained from published balance sheet information in annual reports of each individual institution while the macroeconomic variable is sourced from various issues of Bank Negara Malaysias annual reports The dataset covers the whole gamut of the industrys total assets The final sample consists of 36 banks in 1995 and 1996 and 33 banks in 1997 1998 and 1999 yielding a total of 171 bankyear observations Table 3 presents summary statistics of the output and input variables used to construct the DEA model Several bank and industry specific attributes may influence a particular banks efficiency level Some of these factors may be neither inputs nor outputs in the production process but rather circumstances faced by a particular bank We use an array of bank specific variables to control banks production technologies the input and product market share they are facing and other factors that might confound the empirical relationship between bank characteristics and efficiency The independent variables used to explain Malaysian banks efficiency are grouped under two main characteristics The first represent bankspecific attributes while the second encompass economic conditions during the period examined The bankspecific variables included in the regressions are LNDEPO ROA LOANSTA LNTA LLPTL NIITA NIETA and EQASS To measure the relationship between economic conditions and bank efficiency LNGDP is used Agency issues associated with different types of firm ownership are an area of concern in many banking systems To capture the effects of organizational forms and governance on Malaysian bank efficiency binary dummy variables DUMFORB DUMPUBL and DUMGOVT are included in the regression models Finally to explore the home field and global advantage hypotheses suggested by Berger et al 2000 we have further classified foreign banks operating in Malaysia into three major groups according to their origins namely DUMAMER DUMEURO and DUMASIA During the period of investigation there were four banks headquartered in the North America and Canada four banks headquartered in the European countries and six foreign banks from other Asian countries operating in Malaysia The independent variables their hypothesized relationship with efficiency and notations are detailed in Table 4 To test the relationship between Malaysian bank efficiency and the bank specific and macroeconomic determinants described above the following regression model is estimated θjt β0β1 Characteristics β2 Ecom β3 Ownership β4 Origins εjt 5 where θjt is the technical efficiency of the jth bank in period t obtained from DEA intermediation operating and value added approaches Characteristics is a set of bank characteristics Econ is a vector of economic conditions Ownership is an array of bank ownership characteristics and finally Origins is a set of bank origins variables 4 Results and discussion In this section we will discuss the technical efficiency TE change of the Malaysian banking sector measured by the DEA method and its decomposition into pure technical efficiency PTE and scale efficiency SE components The efficiency of Malaysian commercial banks is first examined by employing the DEA method for each year under investigation The results are classified into two broad heads First we describe the estimates of technical efficiency during the East Asian crisis under three alternative approaches Second to substantiate the results under the DEA approach a multivariate regression framework is employed to relate bank efficiency level to a set of bank specific traits and other macroeconomic determinants 41 Efficiency of the Malaysian banking sector The summary results of technical pure technical and scale efficiency estimates under the three approaches are presented in Tables 57 respectively The average technical efficiency estimate M represents the average of all optimal values obtained from CRS model for each commercial bank Table 5 4 The empirical results suggest a large asymmetry between banks regarding their technical efficiency scores In particular the different approaches of classifying inputs and outputs of banks produced divergent sets of efficiency estimates The estimates of technical efficiency were observed to be consistently higher under operating approach visàvis the intermediation and value added approaches On the other hand under the intermediation approach banks are characterized by relatively low level 4 The CRS assumption is only justifiable when all decisionmaking units DMUs are operating at an optimal scale However firms or DMUs in practice might face either economies or diseconomies of scale Thus if one makes the CRS assumption when not all DMUs are operating at the optimal scale the computed measures of technical inefficiency will be contaminated with scale inefficiency Banker et al 1984 extended the CCR model by relaxing the CRS assumption The resulting BCC model was used to assess the efficiency of DMUs characterized by VRS The VRS assumption provides the measurement of PTE which is the measurement of technical efficiency devoid of the SE effects If there appears to be a difference between the TE and PTE scores of a particular DMU then it indicates the existence of scale inefficiency 62 F Sufian Research in International Business and Finance 23 2009 5477 Table 4 Descriptive of the variables used in the regression models Variable Description Hypothesized relationship with efficiency Remarks Bank characteristics ROA Return on assets Is used as a proxy of bank profitability A positive relationship with bank efficiency is expected LNDEPO Natural logarithm of total deposits Is used as a proxy of market share We do not have a priori expectation on the variable sign LOANSTA Total loans over total assets Is used as an indicator for bank liquidity position Liquidity is an indication of banks ability to meet its customers daytoday cash needs and respond to sudden cash withdrawals The variable is expected to enter the regression model positively LNTA Natural logarithm of total assets Is used as a proxy of bank size to capture the possible cost advantages associated with size economies of scale The variable is expected to take a positive sign LLPTL Loan loss provisions over total loans Is used as a proxy for asset quality It is expected that higher levels of nonperforming assets will result in lower efficiency levels NIITA Noninterest income over total assets Is used as a proxy for banks diversification strategy into nontraditional activities We do not have a priori expectation on the variable sign NIETA Noninterest expense over total assets Is used as a proxy for management quality An inverse relationship is expected between this variable and bank efficiency level EQASS Total book value of shareholders equity over total assets Is used as a proxy of capital adequacy A positive sign is expected Regulators view a higher level of equity as a cushion of future losses Economic conditions LNGDP Natural logarithm of gross domestic products Is used as a proxy for economic conditions We do not have a priori expectation on the variable sign Ownership DUMFORB Dummy variable that takes a value of 1 for foreign banks 0 otherwise Is used to examine the relationship between foreign bank ownership with efficiency The variable is expected to take a positive sign DUMPUBL Dummy variable that takes a value of 1 for publicly listed banks 0 otherwise Is used to examine the relationship between publicly listed banks with efficiency We do not have a priori expectation on the variable sign DUMGOVT Dummy variable that takes a value of 1 for government links banks 0 otherwise Is used to examine the relationship between government ownership and efficiency The variable is expected to take a negative sign Origins DUMAMER Dummy variable that takes a value of 1 for American banks 0 otherwise Is used to test the home field advantage versus global advantage and liability of unfamiliarness hypotheses DUMEURO Dummy variable that takes a value of 1 for European banks 0 otherwise Is used to test the home field advantage versus global advantage and liability of unfamiliarness hypotheses DUMASIA Dummy variable that takes a value of 1 for Asian banks 0 otherwise Is used to test the home field advantage versus global advantage and liability of unfamiliarness hypotheses Table 5 Average technical efficiency of Malaysian banks Year of banks of efficient banks Average efficiency M Average inefficiency 1 MM Standard deviation σ Minimum Interval I M σ I M σ of banks in I Intermediation approach 1995 36 3 0519 0927 0220 0135 0299 0739 750 1996 36 3 0515 0942 0200 0307 0315 0715 806 1997 33 3 0570 0754 0179 0371 0391 0749 727 1998 33 2 0555 0802 0195 0339 0360 0750 758 1999 33 1 0327 2058 0179 0107 0148 0506 818 Value added approach 1995 36 3 0551 0815 0200 0294 0351 0751 806 1996 36 6 0817 0224 0138 0404 0679 0955 667 1997 33 6 0852 0174 0132 0518 0720 0984 697 1998 33 11 0898 0114 0114 0562 0784 1012 848 1999 33 5 0702 0425 0173 0323 0529 0875 727 Operating approach 1995 36 7 0889 0125 0095 0584 0794 0984 667 1996 36 5 0865 0156 0079 0750 0786 0944 639 1997 33 5 0872 0147 0093 0684 0779 0965 606 1998 33 4 0773 0294 0115 0594 0658 0888 727 1999 33 4 0752 0330 0129 0540 0623 0881 727 Table 6 Average pure technical efficiency of Malaysian banks Year of banks of efficient banks Average efficiency M Average inefficiency 1 MM Standard deviation σ Minimum Interval I M σ I M σ of banks in I Intermediation approach 1995 36 7 0784 0276 0210 0141 0574 0994 667 1996 36 7 0840 0190 0140 0502 0700 0980 583 1997 33 10 0884 0131 0114 0643 0770 0998 515 1998 33 10 0838 0193 0163 0421 0675 1001 879 1999 33 8 0759 0318 0229 0245 0530 0988 576 Value added approach 1995 36 11 0834 0199 0163 0428 0671 0997 556 1996 36 16 0912 0096 0124 0463 0788 1036 833 1997 33 18 0930 0075 0106 0633 0824 1036 758 1998 33 19 0944 0059 0091 0691 0853 1035 879 1999 33 10 0790 0266 0188 0411 0602 0978 545 Operating approach 1995 36 16 0930 0075 0080 0719 0850 1010 806 1996 36 15 0926 0080 0080 0770 0846 1006 778 1997 33 12 0936 0068 0084 0731 0852 1020 848 1998 33 10 0880 0136 0109 0654 0771 0989 485 1999 33 13 0878 0139 0140 0600 0738 1018 788 Table 7 Average scale efficiency of Malaysian banks Year of banks of efficient banks Average efficiency M Average inefficiency 1 MM Standard deviation σ Minimum Interval I M σ I M σ of banks in I Intermediation approach 1995 36 3 0665 0504 0187 0395 0478 0852 639 1996 36 3 0616 0623 0207 0388 0409 0823 722 1997 33 3 0644 0553 0170 0402 0474 0814 636 1998 33 2 0663 0508 0172 0432 0491 0835 667 1999 33 1 0445 1247 0196 0141 0249 0641 758 Value added approach 1995 36 3 0662 0511 0187 0404 0475 0849 611 1996 36 6 0896 0116 0090 0637 0806 0986 722 1997 33 6 0917 0091 0099 0559 0818 1016 909 1998 33 11 0950 0053 0063 0803 0887 1013 818 1999 33 5 0902 0109 0140 0504 0762 1042 879 Operating approach 1995 36 7 0957 0045 0075 0584 0882 1032 972 1996 36 5 0936 0068 0065 0760 0871 1001 833 1997 33 5 0934 0071 0070 0767 0864 1004 818 1998 33 4 0882 0134 0098 0666 0784 0980 576 1999 33 4 0863 0159 0103 0626 0760 0966 667 66 F Sufian Research in International Business and Finance 23 2009 5477 of technical efficiency Illustratively in year 1999 ie a year after the East Asian crisis only one bank 3 was found to be efficient and the average technical efficiency for all banks stood at 327 under the intermediation approach The number of efficient banks during the sample period ranged from three banks in 19951997 to only one bank in 1999 under the intermediation approach and seven banks in 1995 to four banks in 1998 and 1999 under the operating approach It is interesting to note that the number of efficient banks was the highest under the value added approach ranging from 3 banks in 1995 to 11 banks in 1998 In sum in and around the East Asian crisis period there was no perceptible change in number of efficient banks under the intermediation and operating approaches although a noticeable increase was observed under the value added approach Under the operating approach the dispersion of technical efficiency scores as measured by its standard deviation depicts an increasing trend during the years 19971999 On the other hand the percentage of banks wherein technical efficiency lies within the interval of 1 SD around the mean hovered around 61 in 1997 to 73 in 1998 and 1999 under the operating approach and 67 in 1996 to 85 in 1998 under the value added approach These numbers were higher under the intermedi ation approach As the technical efficiency estimates itself is time varying these proportions do not necessarily corroborate the degree of in efficiency of the banking system For example under the intermediation approach around 75 in 1995 and around 82 in 1999 of banks recorded technical efficiency within the interval of 1 SD around the mean Yet banks were much more efficient in 1995 than in 1999 As against the changing benchmark of comparison these proportions quantify the number of banks that are close to the average over time and thus merely capture the kurtosis of the efficiency distribu tion depending on the approach For instance under the intermediation approach the efficiency scores displays a leptokurtic distribution ie the efficiency scores has a high peak with a small variance sug gesting that a lot of scores fall in the center of the distribution On the other hand under the operating approach the efficiency of the Malaysian banking sector seem to follow a mesokurtic distribution ie the efficiency scores displays a moderate peak with gradual curves suggesting a normal number of scores in the middle range of the distribution Overall the empirical findings presented in Table 5 clearly bring forth the high degree of inefficiency of Malaysian banking sector around the East Asian crisis period particularly a year after the East Asian crisis Most of the inefficiency stemmed from the under utilization of resources inputs Finally considering the evolution of efficiency over time a clear temporal pattern does not emerge from these different approaches However under the intermediation approach inefficiency exists in the production of banking services and appears to be an important determinant of banks costs It is also observed from Table 5 that although in general technical efficiency level seems to deteriorate abruptly a year after the East Asian crisis under all approaches the deterioration seems to be more pronounced under the intermediation approach model The set of inputoutput is run under both CRS and VRS assumptions If the efficiency score of each bank produced by these models varies then the bank is said to experience VRS Avkiran 1999 Once pure technical efficiency is estimated using VRS scale efficiency is derived by dividing the technical efficiency by pure technical efficiency Table 6 presents the PTE estimates while SE estimates under all three approaches are presented in Table 7 It is observed that over the sample period both pure technical and scale efficiency measures especially under intermediation approach display significant variation and the Malaysian banking sector did not achieve sustained efficiency gains Estimates of pure technical efficiency under the intermediation approach vary from a low of 76 in 1999 to a high of 88 in 1997 In most of the years about 20 of banks were found to be pure technically efficient under the intermediation approach Interestingly the percentage of banks whose pure technical efficiency falls within the interval of 1 SD around the mean displayed a large asymmetry particularly during the period 19971999 under all approaches It is observed from Table 6 that under the intermediation approach the percentage stood at around 52 in 1997 to 88 in 1998 while under value added approach and operating approach the figures stood at around 55 in 1999 to 88 in 1998 and 49 in 1998 to 85 in 1997 respectively It is interesting to note that the number of efficient banks under CRS technical efficiency assump tion and VRS pure technical efficiency assumption differs markedly irrespective of the choice of various inputs and outputs This clearly demonstrates the existence of sizable scale inefficiency among Table 8 Multivariate Tobit regression analysis Explanatory variables Model 1 Intermediation approach operating approach Value added approach CONSTANT 7500 2103 4462 1149 3383 1132 Bank characteristics LNDEPO 0231 0541 0026 0028 0043 0028 LOANSTA 0541 0107 0194 0058 0159 0058 LNTA 0208 0056 0015 0031 0029 0030 LLPTL 0056 0291 0292 0159 0162 0157 NIITA 0448 2008 3208 1097 3308 1080 NIETA 16007 2269 10413 1239 8475 1221 EQASS 0542 0449 0282 0245 0552 0242 ROA 0005 0009 0033 0005 0033 0005 Economic conditions LNGDP 0626 0194 0322 0106 0213 0105 No of observations 171 171 171 Log likelihood 81395 184796 187381 R2 0495 0500 0460 Adj R2 0464 0469 0426 θjt α β1 LNDEPO β2 LOANSTA β3 LNTA β4 LLPTL β5 NIETA β6 EQUITYTA β7 ROA β8 LOGGDP εj The dependent variable is banks efficiency scores derived from DEA intermediation operating and value added approaches LNDEPO is a measure of banks market share calculated as a natural logarithm of total bank deposits LOANSTA is a measure of banks loans intensity calculated as the ratio of total loans to bank total assets LNTA is the size of the banks total asset measured as the natural logarithm of total bank assets LLPTL is a measure of banks risk calculated as the ratio of total loan loss provisions divided by total loans NIETA is a measure of bank management quality calculated as total noninterest expenses divided by total assets NIITA is a measure of banks diversification towards non interest income calculated as total noninterest income divided by total assets EQUITYTA is a measure of banks leverage intensity measured by banks total shareholders equity divided by total assets ROA is a proxy measure for bank profitability calculated as bank profit after tax divided by total assets and LNGDP is natural logarithm of gross domestic product Values in parentheses are standard errors and indicate significance at 1 5 and 10 levels respectively 68 F Sufian Research in International Business and Finance 23 2009 5477 others Hauner 2005 and Pasiouras 2008ab QML HuberWhite standard errors and covariates are calculated Several general comments regarding the test results are warranted The model performs reasonably well in at least two respects For one results for most variables remain stable across the various regressions tested The R2s are also reasonably high ranging from 43 to 47 The findings sug gest that all explanatory variables have the expected signs and in most cases are statistically different from zero In models 27 regressions when we add the other group of variables to the baseline specification that include the bank specific attribute variables the coefficients of the baseline variables stay mostly the same they keep the same sign the same order of magnitude they remain significant as they were so in model 1 regressions albeit sometimes at different levels and with few exceptions do not become significant if they were not in model 1 regressions Therefore for models 27 regressions we will only discuss the results of the new variables added to the baseline specification From Table 8 it is observed that LNDEPO reveals a negative relationship and is statistically significant in intermediation approach regression model suggesting that the more efficient banks are associated to the banks with lower market share thus diminishing market leadership argument The results imply that banks with small market share like the foreign banks can be at least as efficient as market dominant banks in their intermediation function because maintaining or expanding market share might involve extra costs and inputs and thus exacerbating inefficiency LOANSTA reveals positive relationship and is statistically significant in all regression models The findings imply that banks with higher loanstoasset ratios tend to have higher efficiency scores Thus bank loans seem to be more highly valued than alternative bank outputs ie investments and securities The positive relationship found between technical efficiency and LOANSTA may be supporting the efficient market hypothesis Market power in loan markets may be the result of efficient operations Due to their ability to manage operations more productively relatively efficient banks might have lower production costs which enable them to offer more reasonable loan terms and ultimately gaining larger market shares over inefficient banks Likewise LNTA shows positive coefficients and is statistically significant in the intermediation approach regression model suggesting that the larger the bank the more efficient the bank will be purely because of the economies of scale arguments Hauner 2005 offers two potential explanations for which size could have a positive impact on bank efficiency First if it relates to market power large banks should pay less for their inputs Second there may be increasing returns to scale through the allocation of fixed costs eg research or risk management over a higher volume of services or from efficiency gains from a specialized workforce Thus assuming that the average cost curve for Malaysian banks is Ushaped the recent growth policies of the small and medium Malaysian banks seem to be consistent with cost minimization As expected LLPTL shows negative relationship with banks technical efficiency and is statisti cally significant in the operating approach regression model The finding is consistent with earlier findings by among others Kwan and Eisenbeis 1995 Resti 1997 and Barr et al 2002 Further more most research conducted on explaining the causes of bank or thrift industry failures have found that failing institutions carried a large proportion of nonperforming loans in their books prior to failure DermigucKunt 1989 Whalen 1991 Barr and Siems 1994 Berger and Humphrey 1992 Barr and Siems 1994 and Wheelock and Wilson 1995 have also found that banks approaching failure tend to have low cost efficiency and experience high ratios of problem loans and that failing banks tend to be located far from the best practice frontiers The results imply that Malaysian banks should focus more on credit risk management which has been proven to be problematic in the recent past Serious banking problems have arisen from the failure of banks to recognize impaired assets and create reserves for writing off these assets An immense help towards smoothing these anoma lies would be provided by improving the transparency of the financial systems which in turn will assist banks to evaluate credit risk more effectively and avoid problems associated with hazardous exposure The empirical findings seem to suggest that NIITA consistently possesses positive relationship with banks technical efficiency and are statistically significant in all regression models The findings also suggest that the elasticity of technical efficiency with respect to NIITA is quite high particularly in the operating approach and value added approach regression models The results imply that banks tend F Sufian Research in International Business and Finance 23 2009 5477 69 to become more managerially efficient as they increase their income emanating from noninterest sources The empirical finding is consistent with earlier findings by among others Jeon and Miller 2005 The findings seem to suggest that NIETA consistently exhibit negative relationship with bank effi ciency and are statistically significant in the operating approach and value added approach regression models The finding is in consonance with the bad management hypothesis of Berger and DeYoung 1997 Low measure of technical efficiency is a signal of poor senior management practices which apply to inputusage daytoday operations and managing the loan portfolio Sub par managers do not sufficiently monitor and control their operating expenses Managers in these banks might not practice adequate loan underwriting monitoring and control Clearly efficient cost management is a prerequisite for the improved efficiency of the Malaysian banking system ie the high elasticity of efficiency to this variable denotes that banks have much to gain if they improve their managerial prac tices Furthermore the Malaysian banking system has not reached the maturity level required to link quality effects pending from increased spending to higher bank efficiency From Table 8 it is clear that EQASS exhibits negative relationship with bank efficiency and is sta tistically significant in the value added approach regression model which is in line with the findings of Akhigbe and McNulty 2005 The findings imply that the more efficient banks ceteris paribus use more leverage less equity compared to their peers The results seems to suggest that the less efficient banks could have been involved in riskier operations and in the process tend to hold more equity vol untarily or involuntarily ie the reason might be banks deliberate efforts to increase safety cushions and in turn decrease the cost of funds or perhaps regulatory pressures that mandate riskier banks to carry more equity It is observed that ROA exhibits positive relationship and is statistically significant in the operat ing and value added approaches regression models These findings indicate that the more profitable banks tend to exhibit lower inefficiency which corroborates similar findings of some previous studies Isik and Hassan 2002 Hasan and Marton 2003 Miller and Noulas 1996 Banks reporting higher profitability ratios are usually preferred by clients and therefore attract the biggest share of deposits as well as the best potential creditworthy borrowers Such conditions create a favourable environment for the profitable banks to be more efficient from the point of view of intermediation activities Another factor which could explain Malaysian banks inefficiency is the relatively volatile rates of national income growth during the period of analysis measured by LNGDP which exhibit negative and statistically significant relationship in all regression models Demand for financial services tends to grow as economies expand and societies become wealthier However during the period of study the Malaysian economy had experienced a volatile economic growth ie from a robust 995 average growth in 1995 and 1996 declining to record 730 growth in 1997 falling into a recession in 1998 before recovering to register 615 growth in 1999 The volatile economic growth could have resulted in banks to suffer from lower demand for their financial services increased loan defaults and thus lower output 43 Bank ownership and efficiency Banks of different ownership forms may react differently to the same efficiency determinants Thus in the preceding analysis we repeat the regression models above to examine the factors that influence the efficiency of the foreign banks publicly listed banks and government linked banks The regression results focusing on the relationship between bank ownership and efficiency are presented in Table 9 As expected DUMFORB entered the regression models positively and is statistically significant in the value added approach regression models The findings imply that banks with controlling share of foreign ownership are likely to be more efficient compared to their domestically owned counter parts This should come as no surprise because of the ability of foreign owned banks to capitalize on their access to better risk management and operational techniques which is usually made available through their parent banks abroad In addition since foreign ownership is likely to be concentrated foreign owned banks are less prone to typical corporate governance conflicts dispersed owners and the management The evidence seems to suggest that foreign owned banks are more likely to cherry pick the best borrowers available in the market especially those from their own countries of origin 70 F Sufian Research in International Business and Finance 23 2009 5477 Table 9 Multivariate Tobit regression analysis Explanatory variables Model 2 Model 3 Intermediation approach Operating approach Value added approach Intermediation approach Operating approach Value added approach CONSTANT 7511 2102 4484 1142 3427 1104 7362 3943 4555 3683 8068 3414 Bank characteristics LNDEPO 0228 0052 0020 0028 0032 0027 0237 5169 0022 0946 0043 0900 LOANSTA 0533 0109 0179 0059 0130 0057 0537 5029 0197 2820 0483 4035 LNTA 0206 0056 0012 0031 0022 0029 0220 4155 0007 0287 0022 0397 LLPTL 0060 0291 0284 0158 0146 0153 0086 0593 0313 0936 0369 1270 NIITA 0020 2267 2396 1232 1676 1191 0039 0016 3485 2852 0579 0243 NIETA 15800 2324 10022 1263 7688 1220 14930 4829 11142 8674 8475 2681 EQASS 0537 0449 0273 0244 0534 0236 0467 0887 0333 1037 1650 2735 ROA 0005 0009 0033 0005 0032 0005 0005 0648 0034 4631 0008 0669 Economic conditions LNGDP 0629 0194 0327 0106 0023 0102 0624 3536 0324 2826 0763 3413 Bank ownership DUMFORB 0012 0030 0023 0016 0047 0016 DUMPUBL 0039 1512 0027 1371 0061 1449 DUMGOVT No of observations 171 171 171 171 171 171 Log likelihood 81478 185791 191615 81970 185678 63024 R2 0496 0506 0486 0498 0505 0287 Adj R2 0461 0472 0451 0464 0471 0237 Explanatory variables Model 4 Intermediation approach Operating approach Value added approach CONSTANT 7584 4057 4504 3628 7667 3256 Bank characteristics LNDEPO 0232 5108 0026 1096 0050 1057 LOANSTA 0539 5087 0193 2718 0485 4083 LNTA 0212 4114 0017 0665 0032 0604 LLPTL 0039 0270 0300 0897 0455 1715 NIITA 0293 0122 3130 2553 0291 0125 NIETA 16143 5576 10481 8440 10450 3769 EQASS 0535 1029 0278 0867 1548 2675 ROA 0005 0583 0033 4448 0009 0739 Explanatory variables Model 4 Intermediation approach Operating approach Value added approach Economic conditions LNGDP 0638 3600 0328 2852 0734 3307 Bank ownership DUMFORB DUMPUBL DUMGOVT 0023 0950 0012 0620 0052 1537 No of observations 171 171 171 Log likelihood 81661 185017 62979 R2 0497 0501 0286 Adj R2 0462 0467 0267 θjt α β1 LNDEPO β2 LOANSTA β3 LNTA β4 LLPTL β5 NIETA β6 EQUITYTA β7 ROA β8 LOGGDP β9 DUFORB β10 DUPUBL β11 DUMGOVT εj The dependent variable is banks efficiency scores derived from DEA intermediation operating and value added approaches LNDEPO is a measure of banks market share calculated as a natural logarithm of total bank deposits LOANSTA is a measure of banks loans intensity calculated as the ratio of total loans to bank total assets LNTA is the size of the banks total asset measured as the natural logarithm of total bank assets LLPTL is a measure of banks risk calculated as the ratio of total loan loss provisions divided by total loans NIETA is a measure of bank management quality calculated as total noninterest expenses divided by total assets NIITA is a measure of banks diversification towards non interest income calculated as total noninterest income divided by total assets EQUITYTA is a measure of banks leverage intensity measured by banks total shareholders equity divided by total assets ROA is a proxy measure for bank profitability calculated as bank profit after tax divided by total assets LNGDP is natural logarithm of gross domestic product DUMFORB DUMPUBL and DUMGOVT are dummy variables that take a value of 1 for foreign banks publicly listed banks and government related banks respectively 0 otherwise Values in parentheses are standard errors and indicate significance at 1 5 and 10 levels respectively 72 F Sufian Research in International Business and Finance 23 2009 5477 thereby improving the quality of their loan portfolios The empirical observation that foreign banks perform better than domestic banks in developing countries also implies the technical savvy of banks from developed countries generally overcomes the home field advantage in developing countries especially when the domestic economy has relatively unsophisticated financial markets and institu tions Jeon and Miller 2005 The results are in accordance with earlier findings by Isik and Hassan 2003b on Turkish banks Hasan and Marton 2003 on Hungarian banks and Sathye 2003 on Indian banks To investigate the relationship between publicly traded and Malaysian bank efficiency a binary dummy variable DUMPUBL is introduced as an explanatory variable in model 2 regressions It is observed from columns 46 of Table 9 that the variable entered the regression models with a negative sign but is never significant in all the regression models Although the market discipline hypothesis suggests that banks whose shares are publicly traded should exhibit higher efficiency the empirical findings seem to suggest that the Malaysian capital market exerts no discipline over bank management The results are in accord with Chu and Lim 1998 who suggests that stock markets respond more strongly to profit rather than cost efficiency And finally to examine the relationship between government linked banks and efficiency DUM GOVT is included in model 3 regressions The results are presented in columns 79 of Table 9 The estimated coefficients entered the regression models with a negative sign but are never significant in any of the regression models 44 Bank origins and efficiency As pointed out by Berger et al 2000 the methodologies used in the previous studies may not be able to distinguish properly the home field and global advantage hypotheses Following the procedures set by Berger et al 2000 we address this drawback by distinguishing among nations of origin of foreign banks to test for the limited form of the global advantage hypothesis Accordingly we repeat Eq 5 by further classifying foreign banks operating in Malaysia into three major groups according to their parents continents In each set these regressions are performed by considering each bank origin at a time The regression results are presented in Table 10 Under the intermediation approach the empirical findings seem to suggest that foreign banks from the Americas and Europe are relatively more efficient compared to their domestic bank counterparts On the other hand efficiency is negatively related to foreign banks from Asian countries suggesting that foreign banks from other Asian countries were the least efficient banking group during the period of study Unlike the intermediation approach it is observed from Table 10 that foreign banks from the Asian countries were the most efficient banking group under the operating approach and is statistically significant at the 1 level of significance This is followed by banks from the Americas although is not statistically significant at any conventional levels It is clear from column 5 of Table 10 that efficiency is negatively related to banks from other European countries statistically significant at the 1 level implying that banks from the European countries were the least efficient banking group under the operating approach Columns 79 of Table 10 presents the regression results for the value added approach Similar to the operating approach the empirical findings seem to suggest that banks from other Asian countries to be the most efficient banking group statistically significant at the 1 level followed by banks from the Americas Again it is clear from column 8 of Table 10 that efficiency is negatively related to banks from other European countries under the value added approach It is worth highlighting several important points from the findings Firstly the results highlight the importance to differentiate foreign banks according to their origins The empirical findings clearly sug gest that generalizations made in regard to the higher efficiency levels of the foreign banks without tak ing into account their nations or regions of origins could lead to bias conclusions And secondly differ ent inputs and outputs may result in different conclusions The empirical findings from this study sug gest that foreign banks from European countries were found to be relatively more efficient than their domestic bank counterparts under the intermediation approach but were relatively inefficient under the value added and operating approaches Likewise although foreign banks from Asian countries were found to be relatively efficient compared to their domestic bank peers under the value added and operating approaches they were found to be relatively inefficient under the intermediation approach F Sufian Research in International Business and Finance 23 2009 5477 73 Table 10 Multivariate Tobit regression analysis Explanatory variables Model 5 Model 6 Intermediation approach Operating approach 1 2 3 1 2 3 CONSTANT 7562 2105 7520 2105 7504 2107 4464 1151 4376 1139 4199 1108 Bank characteristics LNDEPO 0207 0070 0232 0051 0231 0056 0025 0038 0025 0028 0067 0029 LOANSTA 0551 0108 0541 0107 0542 0113 0194 0059 0194 0058 0121 0060 LNTA 0184 0073 0208 0056 0207 0061 0014 0040 0014 0303 0062 0032 LLPTL 0050 0291 0050 0292 0054 0295 0292 0159 0266 0158 0198 0155 NIITA 0342 2016 0314 2103 0470 2105 3204 1102 3793 1139 1979 1107 NIETA 16133 2280 16113 2321 16051 2630 10417 1246 9954 1258 7836 1382 EQASS 0540 0449 0523 0458 0539 0455 0282 0245 0365 0248 0429 0239 ROA 0005 0009 0005 0009 0005 0009 0033 0005 0033 0005 0032 0005 Economic conditions LNGDP 0632 0194 0628 0194 0626 0194 0322 0106 0314 0105 0306 0102 Bank nationality DUMAMER 0031 0059 0001 0038 DUMEURO 0009 0040 0038 0022 DUMASIA 0001 0040 0078 0021 No of observations 171 171 171 171 171 171 Log likelihood 81532 81418 81396 184796 186272 191337 R2 0496 0495 0495 0500 0509 0537 Adj R2 0461 0460 0460 0465 0475 0505 Explanatory variables Model 7 Value added approach 1 2 3 CONSTANT 3446 1130 3363 1132 3125 1091 Bank characteristics LNDEPO 0018 0021 0043 0028 0083 0029 LOANSTA 0170 0058 0160 0057 0088 0059 LNTA 0004 0023 0028 0030 0075 0032 LLPTL 0167 0156 0156 0157 0069 0153 NIITA 3202 1083 3446 1131 2098 1090 Table 10 Continued Explanatory variables Model 7 Value added approach 1 2 3 NIETA 8603 1224 8367 1248 5938 1362 EQASS 0551 0181 0572 0246 0697 0236 ROA 0032 0005 0032 0005 0031 0005 Economic conditions LNGDP 0219 0104 0211 0105 0197 0101 Bank nationality DUMAMER 0031 0032 DUMEURO 0009 0022 DUMASIA 0077 0021 No of observations 171 171 171 Log likelihood 187862 187465 193914 R2 0463 0461 0500 Adj R2 0426 0423 0465 θjt α β1 LNDEPO β2 LOANSTA β3 LNTA β4 LLPTL β5 NIETA β6 EQUITYTA β7 ROA β8 LOGGDP β9 DUMAMER β10 DUMEURO β11 DUMASIA εj The dependent variable is banks efficiency scores derived from DEA intermediation operating and value added approaches LNDEPO is a measure of banks market share calculated as a natural logarithm of total bank deposits LOANSTA is a measure of banks loans intensity calculated as the ratio of total loans to bank total assets LNTA is the size of the bank total asset measured as the natural logarithm of total bank assets LLPTL is a measure of banks risk calculated as the ratio of total loan loss provisions divided by total loans NIETA is a measure of bank management quality calculated as total noninterest expenses divided by total assets EQUITYTA is a measure of banks leverage intensity measured by banks total shareholders equity divided by total assets ROA is a proxy measure for bank profitability calculated as bank profit after tax divided by total assets LNGDP is natural logarithm of gross domestic product DUMAMER DUMEURO and DUMASIA are dummy variables that take a value of 1 for foreign banks originated from the Americas European countries and Asia respectively 0 otherwise Values in parentheses are standard errors and indicate significance at 1 5 and 10 levels respectively F Sufian Research in International Business and Finance 23 2009 5477 75 In essence the findings which suggest that the American banks were relatively more efficient compared to other foreign banks and their domestic bank peers do not lend support to the limited form of global advantage hypothesis Berger et al 2000 argued that under the limited form of global advantage hypothesis only the efficient institutions in one or a limited number of nations with specific favourable market or regulatory conditions in their home countries can operate more efficiently than domestic banks in other nations 5 Concluding remarks The present study investigates the efficiency of the Malaysian banking sector around the Asian financial crisis 1997 The efficiency estimates of individual banks are evaluated by using the DEA approach Three different approaches viz intermediation approach value added approach and oper ating approach have been employed to differentiate how efficiency scores vary with changes in inputs and outputs A multivariate Tobit model is employed to examine the relationship between efficiency scores derived from the DEA to a set of explanatory variables ie bank size profitability and ownership The estimates of technical efficiency were observed to be consistently higher under operating approach visàvis the intermediation and value added approaches On the other hand under the intermediation approach banks are characterized by relatively low level of technical efficiency The empirical findings clearly bring forth the high degree of inefficiency of the Malaysian banking sector around the East Asian crisis period particularly a year after the East Asian crisis The results suggest that although in general technical efficiency level seems to deteriorate abruptly a year after the East Asian crisis under all approaches the deterioration seems to be more pronounced under the intermediation approach model The empirical findings suggest that the number of efficient banks under CRS technical efficiency technology and VRS technology pure technical efficiency differs markedly irrespective of the choice of various inputs and outputs suggesting the existence of sizable scale inefficiency among Malaysian commercial banks during the East Asian crisis period During the preand postcrisis periods scale inefficiency seems to outweigh pure technical inefficiency in determining the total technical ineffi ciency of the Malaysian banking sector under the intermediation approach On the other hand under the value added approach while scale inefficiency outweighs pure technical inefficiency in determin ing the total technical inefficiency of the Malaysian banking sector during the precrisis period the empirical findings seem to suggest that pure technical inefficiency outweigh scale inefficiency during the postcrisis period Finally under the operating approach although pure technical efficiency is gen erally lower during the precrisis period the trend is less clear during the postcrisis period Thus with respect to their scale of operations Malaysian banks are likely to lose sizeable output particularly in the case of the intermediation approach although at a lesser degree under the value added approach The results from the multivariate regression analysis suggest that technical efficiency is positively and significantly associated with loans intensity suggesting that banks with higher loanstoasset ratios exhibits higher efficiency scores On the other hand it appears that the expense preference behaviour not to be holding in the Malaysian banking sector thus supporting Berger and DeYoungs 1997 bad management hypothesis The findings clearly suggest that efficient cost management to be a prerequisite for the improved efficiency of the Malaysian banking system and that banks have much to gain if they improve their managerial practices Furthermore the Malaysian banking system has not reached the maturity level required to link quality effects pending from increased spending to higher bank efficiency Similarly LNGDP is also negatively related to Malaysian banks efficiency lev els This could be explained by the volatile economic growth which could have resulted in banks to suffer from lower demand for their financial services increases loan defaults and thus lower output We find that the foreign banks have succeeded in capitalizing on their advantages and exhibit a higher level of efficiency compared to their domestic bank peers However we do not find evidence of higher efficiency levels of the publicly listed banks Similarly efficiency is not significantly related to the government link banks The empirical findings do not support for the limited form of global advantage hypothesis and rejects the home field advantage hypothesis as the results suggest that the American banks are relatively more efficient compared to their domestic bank counterparts 76 F Sufian Research in International Business and Finance 23 2009 5477 Due to its limitations the paper could be extended in a variety of ways Firstly the scope of this study can be extended to examine the efficiency of the conventional banks compared to the Islamic banks Secondly the present study can also be extended to estimate the determinants of pure technical and scale efficiency along with the technical efficiency estimates Finally investigation of changes in productivity over time as a result of technical change or technological progress or regress by employing the Malmquist Productivity Index MPI could also be another extension to the present paper Acknowledgements We would like to thank an anonymous reviewer for valuable comments that helped improve an earlier version of the manuscript Any remaining errors are of course our own References Akhigbe A McNulty JE 2005 Profit efficiency sources and differences among small and large US commercial banks J Econ Finan 29 3 289299 Alba P Hernandez L Klingebiel D 1999 Financial liberalization and the capital account Thailand 198897 The World Bank Working Paper AlSharkas AA Hassan MK Lawrence S 2008 The impact of mergers and acquisitions on the efficiency of the US banking industry further evidence J Business Finan Acc 35 5070 Ataullah A Le H 2006 Economic reforms and bank efficiency in developing countries the case of the Indian banking industry Appl Finan Econ 16 653663 Avkiran NK 1999 The evidence on efficiency gains the role of mergers and the benefits to the public J Bank Finan 23 9911013 Banker RD Charnes A Cooper WW 1984 Some models for estimating technical and scale inefficiencies in data envelopment analysis Manage Sci 30 9 10781092 Barr R Siems T 1994 Predicting bank failure using DEA to quantify management quality Working Paper Federal Reserve Bank of Dallas Barr RS Killgo KA Siems TF Zimmel SA 2002 Evaluating the productive efficiency and performance of US commercial banks Manage Finan 28 8 325 Barth JR Brumbaugh DR Suaerhaft D Wang G 1985 Thrift Institution Failures Causes and Policy Issues Bank Structure and Competition Federal Reserve Bank of Chicago Barth JR Caprio J Levine R 2000 Banking Systems around the globe do regulation and ownership affect performance and stability In Mishkin F Ed Prudential Regulation and Supervision What Works and What Doesnt National Bureau of Economic Research Washington DC Beck TA DermigucKunt A Levine R 2006 Bank concentration competition and crises first results J Bank Finan 30 15811603 Benston GJ 1965 Branch banking and economies of scale J Finan 20 2 312331 Berg A Pattillo C 1999 Predicting currency crises the indicators and an alternative J Int Money Finan 18 561586 Berger AN DeYoung 1997 Problem loans and cost efficiency in commercial banks J Bank Finan 21 6 849870 Berger AN Humphrey DB 1992 Measurement and efficiency issues in commercial banking In Griliches Z Ed Measure ment Issues in the Service Sectors National Bureau of Economic Research University of Chicago Press pp 245279 Berger AN Humphrey DB 1997 Efficiency of financial institutions international survey and directions for future research Eur J Oper Res 98 175212 Berger AN DeYoung R Genay H Udell GF 2000 Globalization of financial institutions evidence from cross border banking performance BrookingsWharton Paper Finan Serv 3 23158 Berger AN Hunter WC Timme SG 1993 The efficiency of financial institutions a review and preview of research past present and future J Bank Finan 17 221249 Berger AN 2007 International comparisons of banking efficiency Finan Mark Instit Instrum 16 119144 Bhattacharya A Lovell CAK Sahay P 1997 The impact of liberalization on the productive efficiency of Indian commercial banks Eur J Oper Res 98 332345 Chang R Velasco A 1999 Liquidity crises in emerging markets theory and policy NBER Working Paper Chinn MD 1998 Before the fall were East Asian currencies overvalued NBER Working Paper Chu SF Lim GH 1998 Share performance and profit efficiency of banks in an oligopolistic market evidence from Singapore J Multinatl Finan Manage 8 155168 Coelli T 1996 A Guide to DEAP Version 21 A Data Envelopment Analysis Computer Program Working Paper CEPA University of New England Armidale Coelli T PrasadaRao DS Battese GE 1998 An Introduction to Efficiency and Productivity Analysis Kluwer Academic Publishers Boston Cooper WW Seiford LM Tone K 2000 Data Envelopment Analysis Kluwer Academic Publishers Boston Corsetti G Pesenti P Roubini N 1998 What caused the Asian currency and financial crisis NBER Working Paper Daniel BC Jones JB 2007 Financial liberalization and banking crises in emerging economies J Int Econ 72 1 202221 DermigucKunt A Detragiache E Gupta A 2006 Inside the crisis an empirical analysis of banking systems in distress J Int Money Finan 25 5 702718 DermigucKunt A 1989 Deposit institutions failure a review of the empirical literature Federal Reserve Bank Cleveland Econ Rev 25 4 218 F Sufian Research in International Business and Finance 23 2009 5477 77 Drake L Hall MJB 2003 Efficiency in Japanese banking an empirical analysis J Bank Finan 27 891917 Drake L Hall MJB Simper R 2006 The impact of macroeconomic and regulatory factors on bank efficiency a nonparametric analysis of Hong Kongs banking system J Bank Finan 30 5 14431466 Eichengreen B Rose AK 1998 Staying afloat when the wind shifts external factors and emerging market banking crises NBER Working Paper Frankel JA Rose AK 1996 Currency crashes in emerging markets empirical indicators NBER Working Paper Fukuyama H 1995 Measuring efficiency and productivity growth in Japanese banking a nonparametric approach Appl Finan Econ 5 2 95107 Goldstein M Turner P 1996 Banking Crises in Emerging Economies Origins and Policy Options BIS Economic Papers GonzalezHermosillo B Pazarbasioglu C Billings R 1997 Determinants of banking system fragility a case study of Mexico Int Monet Fund 44 295314 GonzalezHermosillo B 1999 Determinants of exante banking system distress a macro micro empirical exploration of some research episodes IMF Working Paper Hasan I Marton K 2003 Development and efficiency of the banking sector in a transitional economy a Hungarian experience J Bank Finan 27 12 22492271 Hauner D 2005 Explaining efficiency differences among large German and Austrian banks Appl Econ 37 9 969980 Honohan P 1997 Banking system failures in developing and transition countries diagnosis and prediction BIS Working Paper Humphrey DB Pulley LB 1997 Banks responses to deregulation profits technology and efficiency J Money Credit Bank 29 7393 Isik I Hassan MK 2003a Financial disruption and bank productivity the 1994 experience of Turkish banks Q Rev Econ Finan 43 291320 Isik I Hassan MK 2003b Efficiency ownership and market structure corporate control and governance in the Turkish banking industry J Business Finan Acc 30 910 13631421 Isik I Hassan MK 2002 Technical scale and allocative efficiencies of Turkish banking industry J Bank Finan 26 4 719766 Jemric I Vujcic B 2002 Efficiency of banks in Croatia a DEA approach Comp Econ Stud 44 169193 Jeon Y Miller SM 2005 Performance of domestic and foreign banks the case of Korea and the Asian financial crisis Global Econ Rev 34 2 145165 Kaminsky GL Reinhart CM 1998 Financial crises in Asia and Latin America then and now Am Econ Rev 88 2 444448 Kaminsky GL Lizondo S Reinhart CM 1998 Leading indicators of currency crises IMF Staff Papers 45 1 148 Kroszner RS Laeven L Klingebiel D 2007 Banking crises financial dependence and growth J Finan Econ 84 1 187228 Krugman P 1999 Currency crises NBER Conference Report Series Kwan SH Eisenbeis R 1995 An analysis of inefficiencies in banking J Bank Finan 19 34 733734 Kyj L Isik I 2008 Bank Xefficiency in Ukraine an analysis of service characteristics and ownership J Econ Business 60 369393 Lane WR Looney SW Wansley JW 1986 An application of the Cox proportional hazards and international overborrowing J Bank Finan 10 4 511531 Leightner JE Lovell CAK 1998 The impact of financial liberalization on the performance of Thai banks J Econ Business 50 115131 LozanoVivas A Pasiouras F 2008 The impact of nontraditional activities on the estimation of bank efficiency international evidence Working Paper University of Bath Miller SM Noulas AG 1996 The technical efficiency of large bank production J Bank Finan 20 3 495509 Pasiouras F 2008a International evidence on the impact of regulations and supervision on banks efficiency an application of twostage data envelopment analysis Rev Quant Finan Account 30 2 187223 Pasiouras F 2008b Estimating the technical and scale efficiency of Greek commercial banks the impact of credit risk off balance sheet activities and international operations Res Int Business Finan 22 3 301318 Pasiouras F Liadaki A Zopounidis C 2008 Bank efficiency and share performance evidence from Greece Appl Finan Econ 18 14 11211130 Resti A 1997 Evaluating the cost efficiency of the Italian banking system what can be learned from the joint application of parametric and nonparametric techniques J Bank Finan 21 2 221250 Sachs JA Tornell A Velaso A 1996 Financial crises in emerging markets the lessons from 1995 NBER Working Paper Sathye M 2003 Efficiency of banks in a developing economy the case of India Eur J Oper Res 148 3 662671 Saxonhouse GR 1976 Estimated parameters as dependent variables Am Econ Rev 66 1 178183 Sinkey JF 1975 A multivariate statistical analysis of the characteristics of problem banks J Finan 30 1 2136 Suetorsak R 2006 Banking crisis in East Asia a micromacro perspective Rev Quant Finan Account 26 219248 Sufian F 2007 Trends in the efficiency of Singapores commercial banking groups a nonstochastic frontier DEA window analysis approach Int J Product Perform Manage 56 99136 Tanner E 2000 Exchange market pressure and monetary policy Asia and Latin America in the 1990s IMF Staff Papers 47 3 311333 Thanassoulis E 2001 Introduction to the Theory and Application of Data Envelopment Analysis A Foundation Text with Integrated Software Kluwer Academic Publishers Boston Thomson JB 1991 Predicting Bank Failures in the 1980s Federal Reserve Bank Cleveland Econ Rev 27 1 920 Whalen G 1991 A proportional hazards model of bank failure an examination of its usefulness as an early warning tool Federal Reserve Bank Cleveland Econ Rev 27 1 2131 Wheelock DC Wilson PW 1995 Explaining bank failures deposit insurance regulation and efficiency Rev Econ Stat 77 4 689700 Revista Capital Científico Eletrônica RCCe ISSN 21774153 Disponível em revistasunicentrobrindexphpcapitalcientificoindex www3unicentrobrppgadm www3unicentrobr Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil Relationship between efficiency in cost management measured by the method of Stochastic Frontier Analy sis SFA and the profitability of financial institutions in Brazil Douglas José Mendonça1 Júlia Alves e Souza2 Francisval De Melo Carvalho3 e Gideon Carvalho de Benedicto4 Recebido em 17122017 Revisado em 10022018 Aprovado em 28032018 Disponível em 01072018 1 Universidade Federal de Lavras Brasil Doutorando em Administração email mendoncadouglasyahoocombr 2 Universidade Federal de Lavras Brasil Doutoranda em Administração email juliajasouzagmailcom 3 Universidade Federal de Lavras Brasil Doutor em Administração email francarvdaeuflabr 4 Universidade Federal de Lavras Brasil Doutor em Contabilidade e Controladoria email gideonbenedictogmailcom Resumo Este estudo tem por objetivo investigar a relação entre a eficiência na gestão de custos e a rentabilidade das instituições financeiras bancárias no Brasil Utilizouse a abordagem de intermediação para a estimação da eficiência na gestão de custos das instituições financeiras aplicando o método da Análise da Fronteira Estocástica SFA modelado com indicadores econômicofinanceiros Para investigar a relação entre eficiência e a rentabilidade foi ajustado um modelo de regressão linear múltipla com dados em painel A amostra da pesquisa é composta por 47 instituições financeiras classificadas como banco múltiplo comercial e Caixa Econômica e o período analisado compreende os anos de 2008 a 2015 Diante da aplicação do modelo econométrico foi identificada a existência da relação direta e estatisticamente significativa entre a eficiência na gestão de custos e a rentabilidade das instituições financeiras indicando que maiores níveis de eficiência na gestão de custos estão relacionados a maiores níveis de rentabilidade Os sinais obtidos para os coeficientes da variável de investigação e das variáveis de controle corroboraram as expectativas baseadas na literatura científica Assim os achados deste estudo indicam que a eficiência na gestão de custos é um fator determinante para melhorar a rentabilidade do sistema bancário brasileiro Palavraschave Eficiência Rentabilidade Instituições Financeiras Gestão de Custos Análise da Fronteira Estocástica Abstract This study aims to investigate the relationship between efficiency in cost management and the profitability of banking financial institutions in Brazil The intermediation approach was used to estimate the efficiency in cost management of financial institutions applying the Stochastic Frontier Analysis SFA method modeled with economicfinancial indicators To investigate the relationship between efficiency and profitability a multiple linear regression Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 6 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 model with panel data was fitted The research sample is made up of 47 financial institutions classified as multiple commercial and Caixa Econômica and the period analyzed covers the years from 2008 to 2015 In view of the application of the econometric model the existence of a direct and statistically significant relationship was identified between efficiency in the management of costs and the profitability of financial institutions indicating that higher levels of efficiency in the management of costs are related to higher levels of profitability The signals obtained for the coefficients of the research variable and the control variables corroborated the expectations based on the scientific literature Thus the findings indicate that efficiency in cost management is a determining factor to improve the profitability of the Brazilian banking system Keywords Efficiency Profitability Financial Institutions Cost Management Stochastic Frontier Analysis 1 INTRODUÇÃO As instituições financeiras bancárias são importantes para o desenvolvimento econômico de um país Assaf Neto 2012 afirma que os bancos são protagonistas na prestação de serviços de intermediação financeira e por esse motivo auxiliam no financiamento da economia A intermediação financeira é o processo que utiliza os recursos financeiros acumulados dos agentes poupadores e os coloca novamente na economia em forma de empréstimos para os agentes tomadores Como fruto dessa operação as instituições bancárias recebem o spread já que captam recursos a uma determinada taxa de juros e os emprestam a uma outra taxa que é superior à de captação PHILIPPON 2015 Devido ao processo de evolução do mercado financeiro a intermediação financeira deixou de ser a única fonte de receita de tais instituições SCHOLTENS WENSVEEN 2000 Nesse sentido Brighi e Venturelli 2015 afirmam que as evoluções nesse mercado resultaram em uma expansão do escopo das atividades bancárias que passaram a ir além da fonte de renda tradicional ganhos de spread para incorporar também a utilização de fontes nãotradicionais de renda tais como seguros e previdência privada As implicações de tais mudanças nas fontes de receitas dos bancos têm sido amplamente abordadas na literatura embora ainda não exista um consenso sobre os efeitos gerados na rentabilidade das instituições financeiras BRIGHI VENTURELLI 2015 Nota se que essa rentabilidade é alvo de muitas discussões no mercado financeiro e na literatura científica Conforme Doumpos e Cohen 2014 a literatura científica na área de gestão tem explorado a relação entre eficiência e rentabilidade bancária pois a eficiência é um potencial fator determinante do desempenho financeiro das instituições Segundo Dietrich e Wanzenried 2011 a rentabilidade é um critério importante para analisar o resultado das instituições financeiras pois permite identificar o sucesso das estratégias competitivas das instituições que operam em ambientes similares Já a eficiência é um conceito que tem relação com o modo de se fazer as coisas representando a relação entre a quantidade produzida e os recursos consumidos dessa forma a eficiência compara o que foi produzido dado os recursos disponíveis com o que poderia ter sido produzido com os mesmos recursos MELLO et al 2005 Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 7 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 A eficiência na gestão de custos é mensurada por meio da estimação de uma fronteira ótima de custo Uma vez que essa fronteira é estimada ela se torna o padrão em relação ao qual é medida a eficiência das empresas A fronteira de custo indica o custo mínimo necessário para produzir um conjunto de produtos ou serviços dados os preços dos insumos o nível de produção e a tecnologia existente NGUYEN et al 2016 Assim as empresas consideradas mais eficientes na gestão de custos são aquelas que incorrem nos menores custos para produzirem uma determinada quantidade de produtos ou serviços Nesse sentido Philippon 2015 afirma que a eficiência na gestão dos custos oriundos da operação de intermediação financeira impacta diretamente o desenvolvimento da economia no longo prazo O custo de intermediação financeira afeta o custo do utilizador do financiamento externo DIAMOND 1984 Uma gestão eficiente dos custos das instituições financeiras pode representar taxas menores de repasse afetando o fluxo monetário e contribuindo para a redução do custo de capital dos agentes tomadores de empréstimos O método da Análise da Fronteira Estocástica SFA é utilizado na literatura para mensurar a eficiência na gestão de custos consistindo em uma metodologia paramétrica e estocástica projetada para avaliar eficiência AIGNER LOVELL SCHMIDT 1977 O que a SFA faz é estimar a função da fronteira ótima de custo utilizandose de regressões com múltiplas variáveis Para isso tal método trata os insumos como variáveis independentes e o produto ou indicador que sintetize vários produtos como variável dependente e faz uma comparação da eficiência das empresas analisadas em relação aos padrões técnicos ou comportamentais das próprias empresas SILVA 2014 Diante desse contexto o presente artigo tem como objetivo investigar a relação entre a eficiência na gestão de custos mensurada pela SFA e a rentabilidade das instituições financeiras bancárias no Brasil Com base na literatura e nas evidências de estudos anteriores KIRKWOOD NAHM 2006 DOLIVEIRA 2014 MENDONÇA et al 2017 é formulada a hipótese de que instituições mais eficientes na gestão de custos tendem a ter maior rentabilidade Nessa perspectiva pretendese responder à seguinte questão qual a relação existente entre a eficiência na gestão de custos mensurada pela SFA e a rentabilidade das instituições financeiras bancárias no Brasil Devido à importância das instituições financeiras bancárias para a economia brasileira tornase relevante compreender se a eficiência está relacionada à rentabilidade Muitos países que implementaram novas políticas de regulamentação do setor estão vivenciando um cenário com instituições mais competitivas Este também é o caso do Brasil no período entre 2001 e 2010 o setor bancário brasileiro apresentou evolução no lucro líquido e uma expansão no volume de operações de crédito demonstrando um forte crescimento MODRO SANTOS 2015 A importância do setor bancário brasileiro está relacionada com a significativa participação do setor no Produto Interno Bruto PIB do país e os altos lucros que os bancos têm apresentado na casa dos bilhões de reais dado que as instituições financeiras bancárias estão entre as maiores e mais lucrativas empresas no Brasil MANTOVANI SANTOS 2015 Conforme Nguyen et al 2016 um setor bancário eficiente pode refletir uma economia eficiente uma vez que as instituições financeiras bancárias são as principais financiadoras da economia Os resultados desse estudo poderão contribuir com os gestores investidores e analistas financeiros gerando subsídios para tomada de decisão uma vez que oferece informações a respeito da estrutura do setor financeiro e sobre a importância da eficiência Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 8 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 para a rentabilidade O estudo também tem o potencial de contribuir para a ampliação das discussões acadêmicas a respeito dessa temática A estrutura do presente trabalho é composta por cinco seções sendo esta introdução a primeira delas A plataforma teórica que norteia o estudo é apresentada na segunda seção e na terceira são descritos os procedimentos metodológicos adotados Na quarta seção são apresentados os resultados obtidos pela pesquisa Já as considerações finais constam na quinta e última seção 2 REFERENCIAL TEÓRICO 21 Teoria da intermediação financeira A intermediação financeira é uma atividade produtiva em que as instituições financeiras captam recursos com o propósito de repassálos por meio de empréstimos Nesse sentido os intermediários financeiros unem os ofertantes com os tomadores de recursos financeiros facilitando assim a realocação de tais recursos na economia DIAMOND 1984 O estudo das instituições financeiras pode ser feito por meio da teoria da intermediação financeira Conforme Allen e Santomero 1998 essa teoria trata de modelos de alocação de recursos e dos custos decorrentes da operação de intermediação de recursos financeiros Os intermediários financeiros são agentes aos quais é delegada a autoridade para investir em ativos financeiros Tais intermediários podem ser divididos em dois grupos os intermediários financeiros bancários e os intermediários financeiros não bancários DIAMOND 1984 Os intermediários bancários efetuam diretamente a intermediação de dinheiro em espécie no mercado financeiro por exemplo os bancos múltiplos e bancos comerciais entre outros Já os intermediários não bancários efetuam a intermediação de títulos de valores mobiliários com o objetivo de gerar liquidez a estes títulos dessa forma podem atuar também no mercado de capitais por exemplo as sociedades de leasing as companhias de seguros e as corretoras PHILIPPON 2015 Scholtens e Wensveen 2000 afirmam que no processo de evolução do mercado financeiro os bancos comerciais se tornaram empresas globais desenvolvendo suas atividades em várias partes do mundo e passando a obter outras fontes de receitas para além daquelas geradas pela atividade tradicional de intermediação financeira Essas mudanças acarretaram em uma evolução na teoria da intermediação para captar as mudanças ocorridas no setor financeiro mundial Visando captar tais mudanças a abordagem do gerenciamento de risco surge como uma crítica à teoria tradicional da intermediação financeira Essa abordagem parte do pressuposto de que os intermediários financeiros não têm como finalidade apenas a correção das imperfeições de mercado mas também absorvem o risco que seria do poupador no processo de transferência dos recursos financeiros para o tomador PHILIPPON 2015 Os intermediários financeiros são compensados pelos serviços de intermediação prestados O ganho pela intermediação financeira é conhecido como spread que é a diferença entre a taxa de captação e a de repasse ou seja as instituições intermediadoras captam recursos a uma determinada taxa de juros e os emprestam a uma outra taxa que é superior à de captação Os resultados da intermediação são mensurados por meio do total da receita de intermediação menos os custos agregados da operação de intermediação financeira custo Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 9 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 de captação no mercado custo da operação provisão para perdas e despesas de pessoal dentre outros O resultado da intermediação é a soma de todos os spreads oriundos das operações de crédito ZAERNJUK et al 2014 Uma ilustração simplificada do processo de intermediação financeira bancária com a representação do spread e do custo é demonstrada na Figura 01 Nessa figura o valor intermediado é de R 10000 O banco capta o recurso do poupador e paga uma taxa de R 500 ou 5 O valor é repassado para o tomador a uma taxa de R 1200 ou 12 Figura 01 Ilustração do processo de intermediação financeira bancária Fonte Adaptada de Philippon 2015 O spread dessa operação é de R 700 ou 7 que é a diferença entre a taxa de captação e a de repasse Como o custo agregado da operação foi de R 300 valor dado no exemplo o resultado da intermediação financeira é de R 400 R 700 R 300 Vale destacar que o custo de intermediação financeira afeta diretamente o custo dos utilizadores do financiamento externo ou seja das empresas ou pessoas que utilizam os serviços dos intermediadores DIAMOND 1984 Philippon 2015 afirma que a eficiência na gestão dos custos oriundos da operação de intermediação financeira impacta diretamente o desenvolvimento da economia no longo prazo 22 A reestruturação do negócio bancário Os bancos desempenham um papel vital no desenvolvimento econômico de um país desenvolvendo três funções principais facilitação dos pagamentos mobilização de poupança e alocação de fundos de empréstimo KHALIL MEHMOOD AHMAD 2015 O negócio bancário está mudando rapidamente e está se tornando mais competitivo Ao redor do mundo as empresas desse setor passaram por mudanças drásticas em termos de globalização financeira e competição tecnológica Ao mesmo tempo os bancos também experimentaram o crescimento dos custos como consequência de fatores como o aumento do rigor nos requisitos regulatórios as inovações nos serviços financeiros o desenvolvimento tecnológico e os desafios das crises financeiras NGUYEN et al 2016 De acordo com Brighi e Venturelli 2015 as mudanças no ambiente regulatório a abertura dos mercados e os desenvolvimentos tecnológicos vêm criando um novo cenário competitivo no setor bancário A reestruturação no setor alterou as fontes de receitas e o modelo de gestão As instituições que tinham alta dependência dos ganhos de spread perderam muito desse ganho com a regulação do setor e a padronização das taxas Diante disso para compensar a perda de receitas com o spread as chamadas receitas de serviço abertura e manutenção de contas tarifas em geral administração de fundos dentre outros passaram a ter maior representatividade no total das receitas bancárias PHILIPPON 2015 Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 10 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Visando o aumento das receitas os bancos se utilizaram dos avanços tecnológicos e passaram a investir em meios alternativos de serviços com destaque para o internet banking pessoa física e o office banking pessoa jurídica além destes podese citar o aumento na rede de terminais de autoatendimento caixas eletrônicos e o investimento em correspondentes bancários que são empresas terceirizadas e remuneradas pelos bancos para prestar alguns serviços que antes eram prestados dentro das agências por exemplo o pagamento de títulos SOUZA MACEDO 2009 As reestruturações do setor no Brasil começaram a ocorrer após a implantação do Plano Real e seguiram a tendência internacional Conforme Nguyen et al 2016 as instituições financeiras passaram a ter duas fontes principais de receitas as oriundas da intermediação financeira e as da prestação de serviços Esse processo forçou uma alteraçção estrutural dos bancos que resultou não apenas em uma mudança do foco sobre as atividades rentáveis mas também em um novo desenho do mercado BARROS WANKE 2014 Giambiagi e Villela 2005 dividem o período de reestruturação no sistema bancário brasileiro em três etapas A primeira etapa compreende o período entre os anos de 1991 a 1994 marcada pela hiperinflação e pelo binômio privatização do setor ou abertura para a participação estrangeira essa etapa forçou o setor a se modernizar para encarar o novo ambiente A segunda etapa compreende o período entre os anos de 1995 a 1998 quando o Plano Real estabilizou a economia e marcou uma revolução comportamental no setor privado Já a terceira etapa compreende o período entre os anos de 1999 a 2002 e foi marcada por uma tríplice mudança de regime cambial monetário e fiscal Diante das mudanças ocorridas Philippon 2015 explica que as fontes de receita bancária e a eficiência na gestão dos custos oriundos dessas operações passaram a ter uma importância ainda maior visto que os bancos possuem produtos e tecnologias muito similares Assim a manutenção da rentabilidade das empresas do setor está relacionada com a capacidade de gerenciar os custos do negócio bancário 23 Evidências empíricas sobre a eficiência e a rentabilidade de instituições bancárias Conforme Gitman 2010 a eficiência na gestão empresarial cria valor e melhora o desempenho organizacional minimizando o uso dos recursos Observase que o resultado financeiro é fruto de decisões tomadas nos processos empresariais Diante disso o autor argumenta que a utilização eficiente dos recursos leva à redução dos custos e ao aumento no nível de serviço agregando valor e podendo representar um aumento da rentabilidade das empresas Nesse sentido podese identificar alguns trabalhos empíricos que buscaram obter evidências sobre a relação entre a eficiência e a rentabilidade das instituições financeiras Por exemplo o estudo de Kirkwood e Nahm 2006 utilizou a Análise Envoltória de Dados DEA para mensurar a eficiência e analisou a relação entre eficiência e a rentabilidade dos dez maiores bancos da Austrália entre os anos de 1995 a 2002 Os resultados evidenciaram que as grandes instituições financeiras bancárias apresentaram maior eficiência e revelaram a existência de uma relação significativa entre eficiência e rentabilidade em que os bancos maiores e mais eficientes apresentaram maior rentabilidade Já o estudo de Mendonça et al 2017 investigou a relação entre a eficiência econômico financeira mensurada pela DEA e a lucratividade no setor bancário brasileiro no período Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 11 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 de 2011 a 2015 As evidências obtidas indicaram que quanto maior a eficiência econômico financeira da instituição bancária maior sua lucratividade Por fim notase que o estudo de DOliveira 2014 teve uma abordagem diferenciada utilizando o método de Análise da Fronteira Estocástica SFA para mensurar a eficiência na gestão de custos O trabalho examinou quais variáveis específicas dos bancos quais específicas da indústria e quais variáreis macroeconômicas determinaram a rentabilidade dos bancos brasileiros no período de 1995 a 2013 Dentre outros aspectos os resultados demonstraram uma relação significativa e positiva entre a eficiência no custo e a rentabilidade dos bancos evidenciando que a eficiência é um fator que influencia o resultado das empresas do setor Dessa forma a literatura fornece indícios de que a eficiência pode ser considerada um fator que influencia a rentabilidade bancária uma vez que a gestão mais eficiente pode contribuir para o melhor desempenho econômicofinanceiro Entretanto não está claro qual a magnitude do impacto que essa eficiência tem sobre a rentabilidade o que representa uma oportunidade de pesquisa que pode contribuir para o avanço do conhecimento na área 3 METODOLOGIA 31 Tipo de pesquisa amostra e coleta de dados Este trabalho é caracterizado como um estudo quantitativo e descritivo Uma pesquisa quantitativa segundo Gil 2010 é aquela que utiliza uma análise estatística e matemática considerando aquilo que pode ser quantificável A pesquisa descritiva para Martins 2002 tem como objetivo descrever as características de uma determinada população ou fenômeno estabelecendo relações entre variáveis e fatos A pesquisa utilizou dados secundários obtidos por meio do relatório anual do Banco Central do Brasil BCB intitulado 50 Maiores Bancos e o Consolidado do Sistema Financeiro Nacional disponível no website do BCB 2016 Esse relatório contém os dados consolidados referentes a todas as instituições que compõem o Sistema Financeiro brasileiro Ressaltase que o modelo de demonstrações adotado segue o padrão do Plano Contábil das Instituições do Sistema Financeiro Nacional COSIF conforme a Circular n 1273 emitida pelo BCB BCB 1987 Foram coletados os dados contidos nas demonstrações contábeis anuais das instituições englobando o período de 2008 a 2015 A amostra do estudo foi formada por todas as instituições bancárias cujos dados estavam disponíveis para todos os anos em análise Dessa forma a amostra final da pesquisa é composta por 47 instituições financeiras bancárias as quais são listadas no Quadro 01 Observase que a amostra é diversificada e que contém não apenas instituições financeiras nacionais mas também filiais internacionais que atuam no Brasil Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 12 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Quadro 01 Instituições financeiras que compõem a amostra INSTITUIÇÃO FINANCEIRA TIPO DE BANCO INSTITUIÇÃO FINANCEIRA TIPO DE BANCO Banco ABC Brasil SA Múltiplo BPN Brasil Banco Múltiplo SA Múltiplo Banco Alfa SA Comercial Banco Bradesco SA Múltiplo Banco Cooperativo do Brasil SA Múltiplo BRB Banco De Brasília SA Múltiplo Banestes SA Banco do Est do ES Múltiplo Banco BTG Pactual SA Múltiplo Banco Indo Funchal Brasil SA Múltiplo Caixa Econômica Federal Caixa Econômica Banco do Est do RS SA Múltiplo Citibank NA Com Estrangeiro Banco do Brasil SA Múltiplo Banco Credit Suisse Brasil SA Múltiplo Banco BBM SA Múltiplo Deutsche Bank SA Múltiplo Banco AJ Renner SA Múltiplo Banco Fator SA Múltiplo Banco Capital SA Múltiplo HSBC Bank Brasil SA Múltiplo Banco Cargill SA Múltiplo Banco CNH Ind Capital SA Múltiplo Banco Cédula SA Múltiplo Banco Indusval SA Múltiplo Banco Fibra SA Múltiplo ING Bank NV Com Estrangeiro Banco Guanabara SA Múltiplo Itaú Unibanco SA Múltiplo Banco Keb Hana do Brasil SA Comercial Banco John Deere SA Múltiplo Banco de la Provde Buenos Aires Com Estrangeiro Banco Mercantil do Brasil SA Múltiplo Banco Luso Brasileiro SA Múltiplo Banco Pine SA Múltiplo Banco Pottencial SA Comercial Banco Rendimento SA Comercial Banco Rabobank Inte Brasil SA Múltiplo Banco J Safra SA Múltiplo Banco Ribeirão Preto SA Múltiplo Banco Santander SA Múltiplo Banco Triangulo SA Múltiplo Banco Paulista SA Múltiplo Banco BMG SA Múltiplo Banco Sofisa SA Múltiplo Banco BNP Paribas Brasil SA Múltiplo Banco Votorantim SA Múltiplo Bank of Ame Merrill Lynch SA Múltiplo Fonte elaborado pelos autores com base em BCB 2016 A amostra possui instituições financeiras classificadas como bancos múltiplos bancos comerciais e Caixa Econômica Assaf Neto 2003 explica os aspectos referentes a cada um desses tipos de instituição Conforme o autor os bancos comerciais são constituídos obrigatoriamente na forma de sociedades anônimas Eles têm a finalidade de executar operações de crédito de curto e de médio prazo atendendo tanto às necessidades de recursos para capital de giro de pessoas jurídicas como também às pessoas físicas e terceiros em geral Já os bancos múltiplos são organizados sob forma de sociedade anônima e em sua denominação social deve constar a expressão banco Podem ser privados ou públicos e realizam operações ativas passivas e acessórias normalmente realizadas por distintos tipos de instituições financeiras E a Caixa Econômica é o principal agente das políticas públicas do Governo Federal e executa atividades características dos bancos comerciais e múltiplos Ela é responsável pela operacionalização das políticas públicas do governo federal para a habitação popular e saneamento básico utilizando os recursos de cadernetas de poupança e depósitos judiciais e pela gestão dos recursos do Fundo de Garantia por Tempo de Serviço FGTS ASSAF NETO 2003 32 Cálculo da eficiência utilizando o método SFA De acordo com Jiang Yao e Feng 2013 a abordagem de intermediação é considerada a mais adequada para mensurar a eficiência dos bancos e é amplamente utilizada na literatura internacional GAGANIS PASIOURAS 2013 DUYGUN SENA SHABAN 2013 KHALIL MEHMOOD AHMAD 2015 GHROUBI ABAOUB 2016 Considerando esse aspecto o presente estudo emprega a abordagem de intermediação com o intuito de mensurar a eficiência na gestão de custos das instituições financeiras bancárias brasileiras Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 13 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Sob a ótica dessa abordagem as despesas incorridas na captação de recursos são consideradas como inputs assim como as despesas operacionais e de pessoal Já os recursos emprestados e gerados por meio de investimentos são tratados como outputs divididos em dois resultados distintos o resultado da própria intermediação e o resultado operacional que engloba os lucros das demais atividades bancárias GHROUBI ABAOUB 2016 Na Figura 02 são apresentados os inputs e outputs utilizados para estimar a eficiência na gestão de custos das instituições financeiras Figura 02 Indicadores de input e output utilizados para mensurar a eficiência de custo Fonte Adaptada de Khalil Mehmood e Ahmad 2015 Conforme representado na Figura 02 o termo eficiência na gestão de custos pode ser orientado de duas maneiras para a maximização do resultado outputs ou para a minimização do custo inputs Especificamente na mensuração da eficiência na gestão de custos por meio do método da SFA é necessária a estimação de uma função custo Essa função estima a fronteira do custo mínimo necessário para produzir um dado produto a partir dos insumos inputs disponíveis Assim para ser eficiente uma instituição deve continuar produzindo o mesmo volume de produtos outputs utilizando um volume menor de inputs Assim o método SFA é orientado para minimizar o nível de inputs e estimar os níveis de eficiência na gestão de custos das instituições financeiras KHALIL MEHMOOD AHMAD 2015 Os escores estimados pelo modelo para cada empresa variam entre 0 zero e 1 um Quanto maior o escore maior a eficiência na gestão de custos o escore igual a 1 um indicaria uma instituição totalmente eficiente enquanto que o escore igual a 0 zero indicaria uma instituição totalmente ineficiente Os inputs e outputs do modelo foram operacionalizados utilizando indicadores econômicofinanceiros que representam as características em análise A fórmula para o cálculo e as respectivas definições dos indicadores são apresentadas no Quadro 02 Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 14 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Quadro 02 Indicadores de input e output utilizados no modelo SFA INPUTS NOME SIGLA FÓRMULA DEFINIÇÃO Indicador de Despesa de Pessoal IDP Identifica o custo com pessoal necessário para o banco exercer a atividade bancária seja de intermediação ou prestação de serviço Indicador de Despesa Operacional IDO Mede o custo operacional da instituição que seria a despesa que a instituição possui para prestar os demais serviços que não são de intermediação financeira Indicador de Despesa com Intermediação Financeira IDIF Mensura o custo da instituição bancária para prestar o serviço de intermediação financeira OUTPUTS Indicador de Resultado da Intermediação Financeira IRIF Indica o quanto a instituição financeira conseguiu apurar de resultado das operações de intermediação financeira Indicador de Resultado Operacional IRO Representa o quanto a instituição financeira conseguiu apurar de resultado operacional ou seja de receitas não oriundas da intermediação financeira Fonte Elaborado pelos autores Para analisar a eficiência da relação entre inputs e outputs a SFA requer uma especificação explícita da forma desta relação ou seja há a necessidade da estimação de uma função de custos Nesse estudo foi utilizada a função translog transcendental logarítmica que consiste em uma aproximação de segunda ordem de qualquer forma funcional BATTESE COELLI 1995 A função utilizada é representada pela Equação 01 Ln CTit β0 β1 LnIRIFit β2 LnIROit β3 LnIDPit β4 LnIDOit β5 LnIDIFit β6 LnIRIFit2 β7 LnIROit2 β8 LnIDPit2 β9 LnIDOit2 β10 LnIDIFit2 β11 LnIRIFit LnIROit β12 LnIRIFit LnIDPit β13 LnIRIFit LnIDOit β14 LnIRIFit LnIDIFit β15 LnIROit LnIDPit β16 LnIROit LnIDOit β17 LnIROit LnIDIFit υit νit Onde i indica a observação da iésima instituição financeira na amostra t indica o tésimo período Ln CTit é o logaritmo natural do custo total IRIFit é o Indicador de Resultado da Intermediação Financeira IROit é o Indicador de Resultado Operacional IDPit é o Indicador de Despesa de Pessoal IDOit é o Indicador de Despesa Operacional IDIFit é o Indicador de Despesa com Intermediação Financeira e o υit e νit são os termos de erro Já o Custo Total CTit utilizado na estimação da função de custo é o somatório das despesas de intermediação financeira despesas de pessoal despesas administrativas despesas tributárias e outras despesas operacionais dividido pelo ativo total ou seja calculado como percentual do ativo Para a realização dos cálculos da Fronteira Estocástica foi utilizado o software Frontier 41 Dessa forma foram mensurados os índices de eficiência para cada uma das instituições financeiras bancárias em cada ano e esses índices compõem a variável eficiência na gestão de custos EFI 33 Aplicação do modelo econométrico com dados em painel Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 15 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Na primeira etapa da pesquisa foram calculados os escores de eficiência utilizando se do método da Análise da Fronteira Estocástica SFA Já na segunda fase o estudo investigou por meio de um modelo de regressão com dados em painel se a eficiência foi fator determinante para a rentabilidade das instituições financeiras bancárias Para investigar a relação entre a eficiência mensurada pela SFA e a rentabilidade bancária foi utilizado um modelo econométrico de regressão linear múltipla com dados em painel A variável dependente é o Retorno sobre Ativos ROA que representa a rentabilidade das instituições bancárias e é calculado pela divisão do lucro líquido pelo ativo total O ROA mede a capacidade de gestão dos ativos ou seja indica o lucro líquido obtido para cada R 100 em ativos A variável de investigação desse estudo é a EFI eficiência na gestão de custos composta pelos indicadores de eficiência calculados para cada banco conforme descrito no tópico 32 Adicionalmente foram incluídas variáveis de controle no modelo As variáveis de controle são aquelas que poderiam afetar a variável dependente ROA conforme indicado pela literatura e por evidências empíricas anteriores e são utilizadas com o intuito de isolar o seu efeito sobre a variável dependente Para este estudo foram utilizadas as variáveis tamanho dos bancos capitalização grau de intensidade de empréstimo indicador de despesas operacionais em relação ao ativo total e grau de diversificação Tais variáveis juntamente às suas definições e aos sinais esperados são apresentadas no Quadro 03 Quadro 03 Variáveis utilizadas no modelo de regressão com dados em painel e sinais esperados NOME SIGLA DEFINIÇÃO E CÁLCULO SINAL ESPERADO Eficiência EFI Mensurada pela SFA esse indicador pondera as receitas de intermediação e operacional com as despesas incorridas dessas operações visando refletir a dinâmica da atividade bancária Tamanho dos Bancos TAM Calculado pelo o logaritmo dos ativos totais É utilizado para medir o ganho em escala ao maximizar a utilização de seus fatores de produção Capitalização CAP É calculada pelo logaritmo da razão entre patrimônio líquido e total de ativos Mede o grau de capitalização ou seja é a capacidade que uma instituição possui de captar depósito à vista Grau de Intensidade de Empréstimo GIE É calculado pelo logaritmo da razão entre empréstimos e total de ativos Referese indiretamente ao nível de liquidez Assim volume de empréstimo maior implica em baixa liquidez Indicador de Despesas Operacionais em Relação ao Ativo DOA É calculado pelo logaritmo da razão entre despesas operacionais e total de ativos Referese aos gastos operacionais relativamente ao porte do banco Grau de Diversificação GDI É calculado pelo logaritmo da razão entre receitas operacionais e total de ativos É referente às fontes alternativas de receita como por exemplo receitas com serviços e taxas Fonte Elaborado pelos autores Assim o modelo é composto de cinco variáveis de controle além da variável de investigação EFI e da dependente ROA A estrutura utilizada no modelo operacionalizado nesta pesquisa é representada pela Equação 02 ROAit β0 β1 EFIit β2 TAMit β3 CAPit β4 GIEit β5 DOAit β6 GDIit ε Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 16 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Onde ROAii é a rentabilidade Retorno sobre Ativos da instituição financeira bancária i no período t β0 é o intercepto da reta β1 a β6 são os coeficientes angulares EFIit TAMit CAPit GIEit DOAit são as variáveis independentes para a instituição financeira bancária i no período t e ε é o termo de erro da regressão O coeficiente β1 indicará a relação entre a eficiência na gestão de custos e a rentabilidade ROA das instituições financeiras Esperase que esse coeficiente seja positivo e estatisticamente significativo o que indicaria uma relação direta entre tais variáveis 4 RESULTADOS E DISCUSSÕES Inicialmente são apresentados os resultados obtidos pela mensuração da eficiência pelo método SFA Na Tabela 01 constam as estatísticas descritivas dos escores de eficiência na gestão de custos mensurados pela SFA para cada ano Tabela 01 Estatísticas descritivas da eficiência mensurada no período Ano Média Desviopadrão Mínimo Máximo 2008 08233 00894 05943 09365 2009 09221 00073 09035 09482 2010 08662 00595 06981 09503 2011 08590 00999 05666 09735 2012 07625 01730 03580 09859 2013 08613 00800 05683 09671 2014 08927 00551 07243 09636 2015 08789 00562 07206 09577 Fonte Elaborada pelos autores A eficiência média na gestão de custos variou entre 7625 e 9221 durante o período analisado O ano com a menor média foi o de 2012 já o de maior média foi 2009 Com relação aos valores mínimos o ano de 2012 apresentou o menor valor 03580 e o valor mínimo mais elevado foi apresentado no ano de 2009 09035 Já sobre os valores máximos do período o ano de 2012 apresentou o maior valor 09859 enquanto que o valor máximo menos elevado 09365 foi obtido no ano de 2008 Podese observar que o desviopadrão apresentou valores entre 00551 a 01730 no período O ano de 2012 teve a maior variabilidade nos escores de eficiência na gestão de custos das instituições financeiras analisadas desviopadrão de 01730 apresentando tanto o menor valor mínimo do período quanto o maior valor máximo Com base nos valores máximos estimados observase que nenhuma instituição foi totalmente eficiente na gestão de custos nenhuma delas atingiu o escore igual a 1 ou 100 Já com base nos valores mínimos é possível notar também que nenhuma instituição foi totalmente ineficiente na gestão de custos não houve instituições com escore igual a 0 Após a análise das estatísticas descritivas o método de regressão linear múltipla com dados em painel foi utilizado para investigar a relação entre a eficiência na gestão de custos e a rentabilidade das instituições do setor bancário brasileiro Para a aplicação do modelo de regressão inicialmente foram verificadas as correlações entre as variáveis do estudo A estatística VIF fatores de inflação de variância confirmou a ausência de multicolinearidade sendo que o indicador variou entre 1105 e 2200 Para que a regressão Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 17 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 seja aceitável é necessário que o VIF seja menor que cinco conforme Gujarati 2006 Dessa forma não houve problemas com relação à multicolinearidade das variáveis Com relação à normalidade dos resíduos o estudo pautouse no proposto no Teorema do Limite Central o qual afirma que para amostras com um número superior a 30 observações presumese que a distribuição das médias é aproximadamente normal WOOLDRIDGE 2012 Considerando que foram utilizadas 376 observações podese assumir a distribuição normal para as variáveis e os resíduos da regressão Após aplicar o modelo por meio do método de Mínimos Quadrados Ordinários MQO foram efetuados os testes para a identificação da abordagem de dados em painel que melhor se ajustaria aos dados O resultado do teste de BreuschPagan foi igual a 64373 com pvalor de 0000 validando que a abordagem de efeitos aleatórios é preferível em relação à POLS Procedendo ao teste de Hausman obtevese o resultado igual a 15806 com pvalor de 00148 indicando que o modelo efeitos aleatórios é preferível em relação ao modelo de efeitos fixos Assim concluiuse que a abordagem de painel que melhor se ajusta aos dados é a de efeitos aleatórios Para verificar a homocedasticidade dos resíduos utilizouse o teste de White Foi detectada a heterocedasticidade e esta foi corrigida por meio da matriz de covariância de White errospadrão robustos Os resultados finais obtidos a partir da aplicação do modelo são apresentados na Tabela 02 Tabela 02 Resultados da regressão pelo modelo de efeitos aleatórios com errospadrão robustos Variável dependente ROA Retorno sobre Ativos Rentabilidade Variável Coeficiente Erropadrão robusto Estatística t EFI 00122 00059 2070 TAM 00011 00006 1812 CAP 00086 00019 4468 GIE 00001 00009 0122 DOA 00035 00014 2513 GDI 00017 00011 1580 Const 00104 00123 0842 Wald chi25 3658 Prob chi2 00099 Nº de observações 376 R2 total 01359 Fonte Elaborado pelos autores Observase que o modelo é significativo a 1 uma vez que o pvalor do teste qui quadrado é inferior a 001 e que a estimativa por efeitos aleatórios é consistente Como resultado mais relevante destacase que a variável de investigação eficiência EFI mensurada pela SFA foi significativa a 5 95 de confiança apresentando relação positiva O resultado para essa relação corrobora o encontrado por DOliveira 2014 que também indica que maiores níveis de eficiência na gestão de custos estão relacionados a maiores níveis de rentabilidade das instituições financeiras Observase que o coeficiente da variável EFI apresentou um valor de 00122 o que indica que o aumento de uma unidade na eficiência geria um aumento de 00122 unidade na rentabilidade Diante disso é possível inferir que o fato de a instituição financeira obter máxima eficiência escore de eficiência igual a 1 estaria associado a um ROA cerca de 122 superior ao de uma instituição totalmente ineficiente escore de eficiência igual a 0 Também foram obtidos resultados estatisticamente significativos para as variáveis de controle TAM CAP e DOA A variável TAM foi significativa a 10 e esse achado evidencia que o tamanho da instituição financeira pode influenciar na rentabilidade DOliveira 2014 Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 18 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 e Mendonça et al 2017 também encontraram resultados semelhantes para essa variável A relação encontrada foi positiva o que significa que quanto maior for o tamanho da instituição maior será a sua rentabilidade uma vez que as instituições maiores reduzem seus riscos ao diversificarem as carteiras e com isso diminuem o custo de captação A variável CAP foi significativa a 1 Seu coeficiente positivo indica que quanto maior for a capitalização maior será a rentabilidade pois bancos bem capitalizados reduzem seus custos de financiamento e captação ao sinalizarem uma menor possibilidade de falência Os estudos de Kirkwood e Nahm 2006 DOliveira 2014 e Mendonça et al 2017 encontraram resultados análogos a este para essa variável Já a variável DOA foi significativa a 5 Os estudos de Kirkwood e Nahm 2006 e Mendonça et al 2017 encontraram resultados semelhantes A relação encontrada foi indireta sinal negativo o que significa que quanto maiores forem os gastos operacionais relativamente ao porte do banco menor será a rentabilidade Diante disso evidenciase a relevância de que as instituições financeiras controlem os gastos operacionais para terem uma maior rentabilidade Por fim vale destacar que os sinais esperados dos coeficientes que seriam obtidos para a variável de investigação e as variáveis de controle embasados na literatura científica foram todos confirmados para os coeficientes encontrados no presente estudo 5 CONSIDERAÇÕES FINAIS Esta pesquisa objetivou investigar a relação entre eficiência na gestão de custos e a rentabilidade das instituições financeira bancárias no Brasil Diante da aplicação do modelo econométrico foi identificada a existência da relação estatisticamente significativa entre essas variáveis Dessa forma a hipótese teórica testada foi confirmada Considerando o aspecto gerencial os resultados obtidos evidenciam que a eficiência na gestão de custos se relaciona de forma direta com a rentabilidade demonstrando que a eficiência exerce um impacto significativo no retorno sobre o total de ativos aplicados ROA Assim os achados deste estudo indicam que a eficiência na gestão de custos é um fator determinante para melhorar a rentabilidade do sistema bancário brasileiro Destacase que foi possível identificar a magnitude dessa relação o fato de a instituição financeira obter máxima eficiência escore de eficiência igual a 1 estaria associado a um ROA cerca de 122 superior ao de uma instituição totalmente ineficiente escore de eficiência igual a 0 Do ponto de vista econômico é importante ressaltar que a eficiência na operação de intermediação financeira não afeta apenas a rentabilidade destas instituições mas pode impactar o desenvolvimento da economia no longo prazo gerando taxas menores de repasse e influenciando o fluxo monetário Assim estudar os aspectos referentes à eficiência das instituições bancárias assume crescente relevância no atual cenário econômico brasileiro de instabilidade e de altas taxas de juros praticadas pelos bancos Vale mencionar também que as evidências encontradas e discutidas nesta pesquisa devem ser consideradas levando em conta os critérios de seleção da amostra as 47 instituições financeiras no período de 2008 a 2015 e as limitações da metodologia adotada Existe ainda a limitação das variáveis utilizadas nos modelos econométricos uma vez que existem outras variáveis que podem influenciar a rentabilidade e que não foram consideradas nesse estudo Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 19 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 Sugerese para futuros estudos investigar quais variáveis são determinantes da eficiência bancária bem como identificar outros fatores que possam influenciar a rentabilidade dessas instituições Por fim esperase que as evidências apresentadas neste trabalho contribuam para aumentar o conhecimento sobre a importância da eficiência na gestão de custos para a rentabilidade das instituições financeiras bancárias brasileiras fomentando a discussão sobre esta temática REFERÊNCIAS AIGNER D LOVELL C A K SCHMIDT P Formulation and estimation of stochastic frontier production function models Journal of Econometrics v 6 n 1 p 2137 1977 ALLEN F SANTOMERO A M The theory of financial intermediation Journal of Banking Finance v 21 n 11 p 14611485 1998 ALLEN F SANTOMERO A M What do financial intermediaries do Journal of Banking Finance v 25 n 2 p 271294 2001 ASSAF NETO A Estrutura e análise de balanços 10 ed São Paulo Atlas 2012 ASSAF NETO A Mercado Financeiro 5 ed São Paulo Atlas 2003 BARROS C P WANKE P Banking efficiency in Brazil Journal of International Financial Markets Institutions and Money v 28 p 5465 2014 BATTESE G E COELLI T J A model for technical inefficiency effects in a stochastic frontier production function for panel data Empirical Economics v 20 n 2 p 325332 1995 BCB BANCO CENTRAL DO BRASIL 50 maiores bancos e o consolidado do Sistema Financeiro Nacional 2016 Disponível em httpwww4bcbgovbrtop50porttop50 asp Acesso em 24 mai 2017 BCB BANCO CENTRAL DO BRASIL Plano Contábil das Instituições do Sistema Financeiro Nacional COSIF 1987 Disponível em httpwwwbcbgovbrCOSIF Acesso em 23 mai 2017 BRIGHI P VENTURELLI V How functional and geographic diversification affect bank profitability during the crisis Finance Research Letters v16 p 110 2015 DOLIVEIRA E H Determinantes da lucratividade bancária no Brasil 2014 51 f Dissertação Mestrado em Economia Programa de PósGraduação em Economia Universidade de Brasília Brasília 2014 DIAMOND D W Financial intermediation and delegated monitoring The Review of Economic Studies v 51 n 3 p 393414 1984 Relação entre a eficiência na gestão de custos mensurada pelo método de Análise da Fronteira Estocástica SFA e a rentabilidade das instituições financeiras no Brasil 20 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 DIETRICH A WANZENRIED G Determinants of bank profitability before and during the crisis Journal of International Financial Markets Institutions Money v 21 p 307327 2011 DOUMPOS M COHEN S Applying data envelopment analysis on accounting data to assess and optimize the efficiency of Greek local governments Omega International Journal of Management Science v 46 p 7485 2014 DUYGUN M SENA V SHABAN M Schumpeterian competition and efficiency among commercial banks Journal of Banking Finance v 37 n 12 p 51765185 2013 GAGANIS C PASIOURAS F Financial supervision regimes and bank efficiency International evidence Journal of Banking Finance v 37 n 12 p 54635475 2013 GHROUBI M ABAOUB E A metafrontier function for the estimation of Islamic and conventional banks cost and revenue efficiency the case of Malaysia from 2006 to 2012 International Journal of Business and Management v 11 n 5 p 254 2016 GIAMBIAGI F VILLELA A A Economia brasileira contemporânea Rio de Janeiro Elsevier 2005 GIL A C Como elaborar projetos de pesquisa 5 ed São Paulo Atlas 2010 GITMAN L J Princípios de administração financeira 12 ed São Paulo Harbra 2010 GUJARATI D N Econometria básica 4 ed Rio de Janeiro Elsevier 2006 JIANG C YAO S FENG G Bank ownership privatization and performance Evidence from a transition country Journal of Banking Finance v 37 n 9 p 33643372 2013 KHALIL S MEHMOOD B AHMAD N Cost efficiency of Pakistani banking sector a Stochastic Frontier Analysis The Journal of Commerce v 7 n 3 p110126 2015 KIRKWOOD J NAHM D Australian banking efficiency and its relation to stock returns The Economic Record v 82 n 258 p 253267 2006 MANTOVANI M H C SANTOS J O dos Análise da relação entre alavancagem e rentabilidade dos bancos brasileiros listados da Bolsa de Valores de São Paulo do período de 2001 a 2010 REGE Revista de Gestão v 22 n 4 p 509524 2015 MARTINS G de A Manual para elaboração de monografias e dissertações São Paulo Atlas 2002 MELLO J C C B S MEZA L A GOMES E G NETO L B Curso de análise de envoltória de dados In Simpósio Brasileiro de Pesquisa Operacional 37 2005 Gramado Anais Rio de Janeiro SOBRAPO 2005 Douglas José Mendonça Júlia Alves e Souza Francisval De Melo Carvalho Gideon Carvalho de Benedicto 21 Revista Capital Científico Volume 16 n3 JulSet 2018 DOI 1059352177415320180017 MENDONÇA D J SOUZA J A e BENEDICTO G C de SILVA S N A da Relação entre eficiência econômicofinanceira e lucratividade em instituições bancárias brasileiras Revista de Finanças e Contabilidade da Unimep v 4 n 1 p 2037 2017 MODRO W M SANTOS J O dos A relação entre o retorno das ações ordinárias métricas de desempenho e fatores econômicos Revista Administração em Diálogo v 17 n 3 2015 NGUYEN T P T NGHIEM S H ROCA E SHARMA P Bank reforms and efficiency in Vietnamese banks Applied Economics v 48 n 30 p 28222835 2016 PHILIPPON T Has the US finance industry become less efficient On the theory and measurement of financial intermediation The American Economic Review v 105 n 4 p 14081438 2015 SCHOLTENS B WENSVEEN D V A critique on the theory of financial intermediation Journal of Banking Finance v 24 n 8 p 12431251 2000 SILVA M R M Medindo a eficiência dos municípios brasileiros na provisão de políticas trabalhistas 2014 69 f Dissertação Mestrado em Economia Programa de PósGraduação em Economia Universidade de Brasília Brasília 2014 SOUZA M F A S MACEDO M A S Análise de desempenho contábilfinanceiro no setor bancário brasileiro por meio da aplicação da análise envoltória de dados DEA Revista BASE v 6 n 2 p 81100 2009 WOOLDRIDGE J M Introductory econometrics a modern approach 5 ed Mason Cengage Learning 2012 ZAERNJUK V M KRYUKOVA E M BOKAREVA E V CHERNIKOVA L I A study of the theoretical approaches to the banking financial intermediation and its development trends World Applied Sciences Journal v 30 n 12 p 17231725 2014 See discussions stats and author profiles for this publication at httpswwwresearchgatenetpublication256033901 Stochastic Frontier Analysis Using Stata Article in The Stata Journal Promoting communications on statistics and Stata September 2012 DOI 102139ssrn2145803 CITATIONS 240 READS 49275 4 authors Federico Belotti University of Rome Tor Vergata 60 PUBLICATIONS 2355 CITATIONS SEE PROFILE Silvio Daidone Food and Agriculture Organization of the United Nations 77 PUBLICATIONS 1777 CITATIONS SEE PROFILE Giuseppe Ilardi Banca dItalia 17 PUBLICATIONS 745 CITATIONS SEE PROFILE Vincenzo Atella University of Rome Tor Vergata 137 PUBLICATIONS 2370 CITATIONS SEE PROFILE All content following this page was uploaded by Federico Belotti on 07 October 2017 The user has requested enhancement of the downloaded file The Stata Journal Editors H Joseph Newton Department of Statistics Texas AM University College Station Texas editorsstatajournalcom Nicholas J Cox Department of Geography Durham University Durham UK editorsstatajournalcom Associate Editors Christopher F Baum Boston College Nathaniel Beck New York University Rino Bellocco Karolinska Institutet Sweden and University of MilanoBicocca Italy Maarten L Buis WZB Germany A Colin Cameron University of CaliforniaDavis Mario A Cleves University of Arkansas for Medical Sciences William D Dupont Vanderbilt University Philip Ender University of CaliforniaLos Angeles David Epstein Columbia University Allan Gregory Queens University James Hardin University of South Carolina Ben Jann University of Bern Switzerland Stephen Jenkins London School of Economics and Political Science Ulrich Kohler University of Potsdam Germany Frauke Kreuter Univ of MarylandCollege Park Peter A Lachenbruch Oregon State University Jens Lauritsen Odense University Hospital Stanley Lemeshow Ohio State University J Scott Long Indiana University Roger Newson Imperial College London Austin Nichols Urban Institute Washington DC Marcello Pagano Harvard School of Public Health Sophia RabeHesketh Univ of CaliforniaBerkeley J Patrick Royston MRC Clinical Trials Unit London Philip Ryan University of Adelaide Mark E Schaffer HeriotWatt Univ Edinburgh Jeroen Weesie Utrecht University Ian White MRC Biostatistics Unit Cambridge Nicholas J G Winter University of Virginia Jeffrey Wooldridge Michigan State University Stata Press Editorial Manager Lisa Gilmore Stata Press Copy Editors David Culwell and Deirdre Skaggs The Stata Journal publishes reviewed papers together with shorter notes or comments regular columns book reviews and other material of interest to Stata users Examples of the types of papers include 1 expository papers that link the use of Stata commands or programs to associated principles such as those that will serve as tutorials for users first encountering a new field of statistics or a major new technique 2 papers that go beyond the Stata manual in explaining key features or uses of Stata that are of interest to intermediate or advanced users of Stata 3 papers that discuss new commands or Stata programs of interest either to a wide spectrum of users eg in data management or graphics or to some large segment of Stata users eg in survey statistics survival analysis panel analysis or limited dependent variable modeling 4 papers analyzing the statistical properties of new or existing estimators and tests in Stata 5 papers that could be of interest or usefulness to researchers especially in fields that are of practical importance but are not often included in texts or other journals such as the use of Stata in managing datasets especially large datasets with advice from hardwon experience and 6 papers of interest to those who teach including Stata with topics such as extended examples of techniques and interpretation of results simulations of statistical concepts and overviews of subject areas The Stata Journal is indexed and abstracted by CompuMath Citation Index Current ContentsSocial and Behav ioral Sciences RePEc Research Papers in Economics Science Citation Index Expanded also known as SciSearch Scopus and Social Sciences Citation Index For more information on the Stata Journal including information for authors see the webpage httpwwwstatajournalcom Subscriptions are available from StataCorp 4905 Lakeway Drive College Station Texas 77845 telephone 9796964600 or 800STATAPC fax 9796964601 or online at httpwwwstatacombookstoresjhtml Subscription rates listed below include both a printed and an electronic copy unless otherwise mentioned US and Canada Elsewhere Printed electronic Printed electronic 1year subscription 98 1year subscription 138 2year subscription 165 2year subscription 245 3year subscription 225 3year subscription 345 1year student subscription 75 1year student subscription 99 1year institutional subscription 245 1year institutional subscription 285 2year institutional subscription 445 2year institutional subscription 525 3year institutional subscription 645 3year institutional subscription 765 Electronic only Electronic only 1year subscription 75 1year subscription 75 2year subscription 125 2year subscription 125 3year subscription 165 3year subscription 165 1year student subscription 45 1year student subscription 45 Back issues of the Stata Journal may be ordered online at httpwwwstatacombookstoresjjhtml Individual articles three or more years old may be accessed online without charge More recent articles may be ordered online httpwwwstatajournalcomarchiveshtml The Stata Journal is published quarterly by the Stata Press College Station Texas USA Address changes should be sent to the Stata Journal StataCorp 4905 Lakeway Drive College Station TX 77845 USA or emailed to sjstatacom Copyright c 2013 by StataCorp LP Copyright Statement The Stata Journal and the contents of the supporting files programs datasets and help files are copyright c by StataCorp LP The contents of the supporting files programs datasets and help files may be copied or reproduced by any means whatsoever in whole or in part as long as any copy or reproduction includes attribution to both 1 the author and 2 the Stata Journal The articles appearing in the Stata Journal may be copied or reproduced as printed copies in whole or in part as long as any copy or reproduction includes attribution to both 1 the author and 2 the Stata Journal Written permission must be obtained from StataCorp if you wish to make electronic copies of the insertions This precludes placing electronic copies of the Stata Journal in whole or in part on publicly accessible websites fileservers or other locations where the copy may be accessed by anyone other than the subscriber Users of any of the software ideas data or other materials published in the Stata Journal or the supporting files understand that such use is made without warranty of any kind by either the Stata Journal the author or StataCorp In particular there is no warranty of fitness of purpose or merchantability nor for special incidental or consequential damages such as loss of profits The purpose of the Stata Journal is to promote free communication among Stata users The Stata Journal ISSN 1536867X is a publication of Stata Press Stata Stata Press Mata and NetCourse are registered trademarks of StataCorp LP The Stata Journal 2013 13 Number 4 pp 719758 Stochastic frontier analysis using Stata Federico Belotti Centre for Economic and International Studies University of Rome Tor Vergata Rome Italy federicobelottiuniroma2it Silvio Daidone Centre for Health Economics University of York York UK silviodaidoneyorkacuk Giuseppe Ilardi Economic and Financial Statistics Department Bank of Italy Rome Italy giuseppeilardibancaditaliait Vincenzo Atella Centre for Economic and International Studies University of Rome Tor Vergata Rome Italy atellauniroma2it Abstract This article describes sfcross and sfpanel two new Stata commands for the estimation of crosssectional and paneldata stochastic frontier models sfcross extends the capabilities of the frontier command by including addi tional models Greene 2003 Journal of Productivity Analysis 19 179190 Wang 2002 Journal of Productivity Analysis 18 241253 and command functional ities such as the possibility of managing complex survey data characteristics Similarly sfpanel allows one to fit a much wider range of timevarying ineffi ciency models compared with the xtfrontier command including the model of Cornwell Schmidt and Sickles 1990 Journal of Econometrics 46 185200 the model of Lee and Schmidt 1993 in The Measurement of Productive Efficiency Techniques and Applications a production frontier model with flexible temporal variation in technical efficiency the flexible model of Kumbhakar 1990 Journal of Econometrics 46 201211 the inefficiency effects model of Battese and Coelli 1995 Empirical Economics 20 325332 and the true fixed and randomeffects models of Greene 2005a Journal of Econometrics 126 269303 A brief overview of the stochastic frontier literature a description of the two commands and their options and examples using simulated and real data are provided Keywords st0315 sfcross sfpanel stochastic frontier analysis production frontier cost frontier crosssectional panel data 1 Introduction This article describes sfcross and sfpanel two new Stata commands for the estima tion of parametric stochastic frontier SF models using crosssectional and panel data c 2013 StataCorp LP st0315 720 Stochastic frontier using Stata Since the publication of the seminal articles by Meeusen and van den Broeck 1977 and Aigner Lovell and Schmidt 1977 this class of models has become a popular tool for efficiency analysis a stream of research has produced many reformulations and exten sions of the original statistical models generating a flourishing industry of empirical studies An extended review of these models can be found in the recent survey by Greene 2012 The SF model is motivated by the theoretical idea that no economic agent can exceed the ideal frontier and deviations from this extreme represent individual inefficiencies From the statistical point of view this idea has been implemented by specifying a regression model characterized by a composite error term in which the classical idiosyn cratic disturbance aiming at capturing measurement error and any other classical noise is included with a onesided disturbance that represents inefficiency1 Whether cross sectional or panel data production or cost frontier time invariant or varying inefficiency parametric SF models are usually estimated by likelihoodbased methods and the main interest is on making inference about both frontier parameters and inefficiency The estimation of SF models is already possible using official Stata routines How ever the available commands cover a restricted range of models especially in the panel data case The sfcross command mirrors the frontier commands functionality but adds new features such as i the estimation of normalgamma models via simulated maximum likelihood SML Greene 2003 ii the estimation of the normaltruncated normal model proposed by Wang 2002 in which both the location and the scale parameters of the inefficiency distribution can be expressed as a function of exogenous covariates and iii the opportunity to manage complex survey data characteristics via the svyset command As far as paneldata analysis is concerned the Stata xtfrontier command allows the estimation of a normaltruncated normal model with timeinvariant inefficiency Battese and Coelli 1988 and a timevarying version named the time decay model Battese and Coelli 1992 Our sfpanel command allows one to fit a wider range of timevarying inefficiency models including the model of Cornwell Schmidt and Sick les 1990 that of Lee and Schmidt 1993 the flexible model of Kumbhakar 1990 the time decay and the inefficiency effects models of Battese and Coelli 1992 1995 and the true fixed TFE and randomeffects TRE models of Greene 2005a For the last two models the command allows different distributional assumptions provid ing the modeling of both inefficiency location and scale parameters Furthermore the command allows the estimation of the randomeffects timeinvariant inefficiency mod els of Pitt and Lee 1981 and Battese and Coelli 1988 as well as the fixedeffects version of the model of Schmidt and Sickles 1984 which is characterized by no distri butional assumptions on the inefficiency term In addition because the main objective 1 The literature distinguishes between production and cost frontiers The former represents the maximum amount of output that can be obtained from a given level of inputs while the latter characterizes the minimum expenditure required to produce a bundle of outputs given the prices of the inputs used in its production F Belotti S Daidone G Ilardi and V Atella 721 of SF analysis is the estimation of inefficiency we provide postestimation routines to compute both inefficiency and efficiency scores as well as their confidence intervals Jondrow et al 1982 Battese and Coelli 1988 Horrace and Schmidt 1996 Finally sfcross and sfpanel also allow the simultaneous modeling of heteroskedasticity in the idiosyncratic error term In the development of these new commands we make extensive use of Mata to speed up the estimation process We allow for the use of Stata factor variables weighted esti mation constrained estimation resamplingbased variance estimation and clustering Moreover by using Mata structures and libraries we provide a very readable code that can be easily developed further by other Stata users All of these features make the commands simple to use extremely flexible and fast they also ensure the opportunity to fit stateoftheart SF models Finally we would like to emphasize that sfpanel offers the possibility of performing a constrained fixedeffects estimation which is not yet available with xtreg Moreover the models of Cornwell Schmidt and Sickles 1990 and Lee and Schmidt 1993 al though proposed in the SF literature are linear paneldata models with timevarying fixed effects and thus potentially very useful in other contexts The article is organized as follows In section 2 we present a brief review of the SF approach evolution focusing on the models that can be estimated using the proposed commands Sections 3 and 4 describe the syntax of sfcross and sfpanel focusing on the main options Sections 5 and 6 illustrate the two commands using simulated data and two empirical applications from the SF literature Finally section 7 offers some conclusions 2 A review of stochastic frontier models We begin our discussion with a general formulation of the SF crosssectional model and then review extensions and improvements that have been proposed in the literature focusing on those models that can be estimated using sfcross and sfpanel Given the large number of estimators allowed by the two commands we deliberately do not discuss the derivation of the corresponding criterion functions We refer the reader to the cited works for details on the estimation of each model A summary of all estimable models and their features is reported in table 1 21 Crosssectional models Consider the following SF model yi α x iβ εi i 1 N 1 εi vi ui 2 vi N0 σ2 v 3 ui F 4 where yi represents the logarithm of the output or cost of the ith productive unit xi is a vector of inputs input prices and quantities in the case of a cost frontier and β is the vector of technology parameters The composed error term εi is the sum or the difference of a normally distributed disturbance vi representing measurement and specification error and a onesided disturbance ui representing inefficiency 2 Moreover ui and vi are assumed to be independent of each other and independent and identically distributed across observations The last assumption about the distribution F of the inefficiency term is needed to make the model estimable Aigner Lovell and Schmidt 1977 assumed a halfnormal distribution that is ui N0σu2 while Meeusen and van den Broeck 1977 opted for an exponential one ui E σu Other commonly adopted distributions are the truncated normal Stevenson 1980 and the gamma distributions Greene 1980ab 2003 The distributional assumption required for the identification of the inefficiency term implies that this model is usually fit by maximum likelihood ML even if modified ordinary leastsquares or generalized method of moments estimators are possible often inefficient alternatives3 In general SF analysis is based on two sequential steps in the first estimates of the model parameters θ are obtained by maximizing the loglikelihood function ℓθ where θ αβσu2σv2 4 In the second step point estimates of inefficiency can be obtained through the mean or the mode of the conditional distribution fui εi where εi yi α xi β The derivation of the likelihood function is based on the independence assumption between ui and vi Because the composite model error εi is defined as εi vi ui its probability density function is the convolution of the two component densities as fεεi 0 to fuui fv εi ui dui 5 Hence the loglikelihood function for a sample of n productive units is ℓθ i1 to n log fεεi θ The marginalization of ui in 5 leads to a convenient closedform expression only for the normalhalf normal normalexponential and normaltruncated normal models In all other cases for example the normalgamma model numerical or simulationbased techniques are necessary to approximate the integral in 5 The second estimation step is necessary because the estimates of the model parameters allow for the computation of residuals ε but not for the inefficiency estimates 2 In this section we consider only production functions However the sign of the ui term in 2 is positive or negative depending on whether the frontier describes a cost or a production function respectively 3 Notice that when a distributional assumption on u is made sfcross and sfpanel estimate model parameters by likelihoodbased techniques 4 Different model parameterizations are used in the SF literature as for example θ αβσ2λ where σ2 σu2 σv2 and λ σu σv Because the main objective of SF analysis is the estimation of technical or cost efficiency a strategy for disentangling this unobserved component from the compounded error is required As mentioned before the most wellknown solutions to this problem proposed by Jondrow et al 1982 and Battese and Coelli 1988 exploit the conditional distribution of u given ε Thus a point estimate of the inefficiencies can be obtained using the mean Eu ε or the mode Mu ε of this conditional distribution Once point estimates of u are obtained estimates of the technical cost efficiency can be derived as Eff exp u where u is either Eu ε or Mu ε5 22 Paneldata models The availability of a richer set of information in panel data allows one to relax some of the assumptions previously imposed and to consider a more realistic characterization of the inefficiencies Pitt and Lee 1981 were the first to extend the model 14 to longitudinal data They proposed the ML estimation of the following normalhalf normal SF model yit α xit β εit i 1N t 1Ti 6 εit vit ui vit N0σv2 ui N0σu2 The generalization of this model to the normaltruncated normal case has been proposed by Battese and Coelli 19886 As pointed out by Schmidt and Sickles 1984 the estimation of an SF model with timeinvariant inefficiency can also be performed by adapting conventional fixedeffects estimation techniques thereby allowing inefficiency to be correlated with the frontier regressors and avoiding distributional assumptions about ui However the timeinvariant nature of the inefficiency term has been questioned especially in the presence of empirical applications based on longpanel datasets To relax this restriction Cornwell Schmidt and Sickles 1990 have approached the problem by proposing the following SF model with individualspecific slope parameters yit α xit β vit uit i 1N t 1Ti u it ωi ωi1 t ωi2 t2 7 in which the model parameters are estimated by extending the conventional fixed and randomeffects paneldata estimators This quadratic specification allows for a unitspecific temporal pattern of inefficiency but requires the estimation of a large number of parameters N 3 5 A general presentation of the postestimation procedures implemented in the sfcross and sfpanel routines is given by Kumbhakar and Lovell 2000 and Greene 2012 to whom we refer the reader for further details 6 The normalexponential model is another straightforward extension allowed by sfpanel Following a slightly different estimation strategy Lee and Schmidt 1993 proposed an alternative specification in which the uit is specified as uit gt ui where gt is represented by a set of time dummy variables This specification is more parsimonious than 7 and it does not impose any parametric form but it is less flexible because it restricts the temporal pattern of uit to be the same for all productive units7 Kumbhakar 1990 was the first to propose the ML estimation of a timevarying SF model in which gt is specified as gt 1 expγt δt21 This model contains only two additional parameters to be estimated γ and δ and the hypothesis of timeinvariant technical efficiency can be easily tested by setting γ δ 0 A similar model called time decay has been proposed by Battese and Coelli 1992 in which gt expη t Ti The common feature of all of these timevarying SF models is that the intercept α is the same across productive units thus generating a misspecification bias in the presence of timeinvariant unobservable factors which are unrelated with the production process but affect the output Therefore the effect of these factors may be captured by the inefficiency term producing biased results Greene 2005a approached this issue through a timevarying SF normalhalf normal model with unitspecific intercepts obtained by replacing 6 by the following specification yit αi xit β εit 8 Compared with previous models this specification allows one to disentangle timevarying inefficiency from unitspecific timeinvariant unobserved heterogeneity For this reason Greene 2005a termed these models TFE or TRE according to the assumptions on the unobserved unitspecific heterogeneity While the estimation of the TRE specification can be easily performed using simulationbased techniques the ML estimation of the TFE variant requires the solution of two major issues related to the estimation of nonlinear paneldata models The first is purely computational because of the large dimension of the parameter space Nevertheless Greene 2005ab showed that a maximumlikelihood dummy variable MLDV approach is computationally feasible also in the presence of a large number of nuisance parameters αi N 1000 The second the socalled incidental parameters problem is an inferential issue that arises 7 Han Orea and Schmidt 2005 and Ahn Hoon Lee and Schmidt 2001 propose estimating the models of Cornwell Schmidt and Sickles 1990 and Lee and Schmidt 1993 respectively through a generalized method of moments GMM approach They show that GMM is preferable because it is asymptotically efficient Currently sfpanel allows for the estimation of the models of Cornwell Schmidt and Sickles 1990 and Lee and Schmidt 1993 through modified leastsquares dummy variables and iterative leastsquares approaches respectively We leave for future updates the implementation of the GMM estimator F Belotti S Daidone G Ilardi and V Atella 725 when the number of units is relatively large compared with the length of the panel In these cases the unitspecific intercepts are inconsistently estimated as N with fixed T because only Ti observations are used to estimate each unitspecific parame ter Neyman and Scott 1948 Lancaster 2002 As shown in Belotti and Ilardi 2012 because this inconsistency contaminates the variance parameters which represent the key ingredients in the postestimation of inefficiencies the MLDV approach appears to be appropriate only when the length of the panel is large enough T 108 Although model 8 may appear to be the most flexible and parsimonious choice among the several existing timevarying specifications one can argue that a portion of the timeinvariant unobserved heterogeneity does belong to inefficiency or these two components should not be disentangled at all The sfpanel command provides options for the estimation of these two extremes for the models of Schmidt and Sickles 1984 Pitt and Lee 1981 and Battese and Coelli 1988 in which all timeinvariant unob served heterogeneity is considered as inefficiency and for the two true specifications in which all timeinvariant unobserved heterogeneity is ruled out from the inefficiency component As pointed out by Greene 2005b neither formulation is completely satis factory a priori and the choice should be driven by the features of the data at hand9 Despite the usefulness of SF models in many contexts a practical disclaimer is in order in both crosssectional and paneldata models the identification through dis tributional assumptions of the two components u and v heavily depends on how the shape of their distributions is involved in defining the shape of the ε distribution Iden tification problems may arise when either the shapes are very similar as pointed out by Ritter and Simar 1997 in the case of small samples for the normalgamma cross sectional model or one of the two components is responsible for most of the shape of the ε distribution The latter is the case where the ratio between the inefficiency and measurement error variability the socalled signaltonoise ratio σuσv is very small or very large In these cases the profile of the log likelihood becomes quite flat producing nontrivial numerical maximization problems 23 Exogenous inefficiency determinants and heteroskedasticity A very important issue in SF analysis is the inclusion in the model of exogenous vari ables that are supposed to affect the distribution of inefficiency These variables which usually are neither the inputs nor the outputs of the production process but nonetheless 8 A common approach to solve this problem is based on the elimination of the αi through a data transformation The consistent estimation of the fixedeffects variant of Greenes 2005a model is still an open research issue in SF literature Promising solutions have been pro posed by Chen Schmidt and Wang 2011 for a homoskedastic normalhalf normal model and Belotti and Ilardi 2012 for a more flexible heteroskedastic specification in normalhalf normal and normalexponential models We are currently working to update the sfpanel command along these directions 9 A way to disentangle unobserved heterogeneity from inefficiency is to include explanatory vari ables that are correlated with inefficiency but not with the remaining heterogeneity The use of untestable exclusion restrictions is a quite standard econometric technique to deal with identifi cation issues affect the productive unit performance have been incorporated in a variety of ways i they may shift the frontier function and the inefficiency distribution ii they may scale the frontier function and the inefficiency distribution and iii they may shift and scale the frontier function and the inefficiency distribution Moreover Kumbhakar and Lovell 2000 stress that in contrast with the linear regression model in which the misspecification of the second moment of the errors distribution determines only efficiency losses the presence of uncontrolled observable heterogeneity in uᵢ and vᵢ may affect the inference in SF models Indeed while neglected heteroskedasticity in vᵢ does not produce any bias for the frontiers parameter estimates it leads to biased inefficiency estimates as we show in section 53 In this section we present the approaches that introduce heterogeneity in the location parameter of the inefficiency distribution and heteroskedasticity of the inefficiency as well as in the parameter of the idiosyncratic error term for the models implemented in the sfcross and sfpanel commands Because these approaches can be easily extended to the paneldata context we deliberately confine the review to the crosssectional framework As pointed out by Greene 2012 researchers have often incorporated exogenous effects using a twostep approach In the first step estimates of inefficiency are obtained without controlling for these factors while in the second the estimated inefficiency scores are regressed or otherwise associated with them Wang and Schmidt 2002 show that this approach leads to severely biased results thus we shall focus only on model extensions based on simultaneous estimation A natural starting point for introducing exogenous variables in the inefficiency model is in the location of the distribution The most wellknown approaches are those suggested by Kumbhakar Ghosh and McGuckin 1991 and Huang and Liu 1994 They proposed to parameterize the mean of the pretruncated inefficiency distribution Basically model 13 can be completed with uᵢ N μᵢ σᵤ² μᵢ zᵢψ where uᵢ is a realization from a truncated normal random variable zᵢ is a vector of exogenous variables including a constant term and ψ is the vector of unknown parameters to be estimated the socalled inefficiency effects One interesting feature of this approach is that the vector zᵢ may include interactions with input variables thus allowing one to test the hypothesis that inefficiency is neutral with respect to its impact on input usage¹⁰ ¹⁰ Battese and Coelli 1995 proposed a similar specification for panel data An alternative approach to analyzing the effect of exogenous determinants on inefficiency is obtained by scaling its distribution Then a model that allows heteroskedasticity in uᵢ and vᵢ becomes a straightforward extension For example Caudill and Ford 1993 Caudill Ford and Gropper 1995 and Hadri 1999 proposed to parameterize the variance of the pretruncated inefficiency distribution in the following way uᵢ N 0 σᵤᵢ² 10 σᵤᵢ² exp zᵢψ 11 Hadri 1999 extends this last specification by allowing the variance of the idiosyncratic term to be heteroskedastic so that 3 can be rewritten as vᵢ N 0 σᵥᵢ² σᵥᵢ² exp hᵢφ 12 where the variables in hᵢ do not necessarily appear in zᵢ As in Wang 2002 both sfcross and sfpanel allow one to combine 9 and 10 for the normaltruncated normal model In postestimation it is possible to compute nonmonotonic effects of the exogenous factors zᵢ on uᵢ A different specification has been suggested by Wang and Schmidt 2002 in which both the location and variance parameters are scaled by the same positive monotonic function hzᵢ ψ Their model uᵢ hzᵢ ψuᵢ with uᵢ N μ σ² is equivalent to the assumption that uᵢ Nμhzᵢ ψ σ²hzᵢ ψ² in which the zᵢ vector does not include a constant term¹¹ ¹¹ We are currently working to extend the sfcross command to normaltruncated normal models with scaling property Wang and Schmidt 2002 728 Stochastic frontier using Stata Table 1 A summary of sfcross and sfpanel estimation and postestimation capabilities Reference Arguments of F 2 Est Loc eff Heter Heter JLMS4 BC5 CI6 d and m1 method3 in u in u in v Crosssectional models Aigner Lovell and Schmidt 1977 hnormal HN ML x x x x x Meeusen and van den Broeck 1977 exponential E ML x x x x x x Stevenson 1980 tnormal TN ML x x x x x x Greene 2003 gamma G SML x x x x Paneldata models Schmidt and Sickles 1984 fe W x Schmidt and Sickles 1984 regls GLS x Pitt and Lee 1981 pl81 HN ML x x x Battese and Coelli 1988 bc88 TN ML x x x Cornwell Schmidt and Sickles 1990 fecss MW x Lee and Schmidt 1993 fels ILS x Kumbhakar 1990 kumb90 HN ML x x x Battese and Coelli 1992 bc92 TN ML x x x Battese and Coelli 1995 bc95 TN ML x x x x x x Greene 2005a tfe E MLDV x x x x x x Greene 2005a tfe HN MLDV x x x x x Greene 2005a tfe TN MLDV x x x x x x Greene 2005a tre E SML x x x x x x Greene 2005a tre HN SML x x x x x Greene 2005a tre TN SML x x x x x x 1 The model option is available in sfpanel The distribution option is available in sfcross 2 Distribution F of u HN half normal E exponential TN truncated normal G gamma 3 Estimation method ML maximum likelihood SML simulated maximum likelihood GLS generalized least squares W within group MW modified within group ILS iterative least squares MLDV maximumlikelihood dummy variable 4 Inefficiency and efficiency estimation via the approach of Jondrow et al 1982 5 Efficiency estimation via the approach of Battese and Coelli 1988 6 Confidence interval for inefficiencies via the approach of Horrace and Schmidt 1996 3 The sfcross command The new Stata command sfcross provides parametric ML estimators of SF models where the default is represented by production The general syntax of this command is as follows sfcross depvar indepvars if in weight distributiondistname emeanvarlistm noconstant usigmavarlistu noconstant vsigmavarlistv noconstant svfrontierlistname svemeanlistname svusigmalistname svvsigmalistname cost simtypesimtype nsimulations base postscore posthessian This command and its panel analog sfpanel are written using the moptimize suite of functions the optimization engine used by ml and share the same features of all Stata estimation commands including access to the estimation results and options for the maximization process see help maximize Stata 112 is the earliest version that can be used to run the commands aweight fweight iweight and pweight are allowed see help weight sfcross supports the svy prefix see help survey The default is the normalexponential model Most options are similar to those of other Stata estimation commands A full description of all available options is provided in the sfcross help file 31 Main options for sfcross distributiondistname specifies the distribution for the inefficiency term as half normal hnormal truncated normal tnormal exponential exponential or gamma gamma The default is distributionexponential emeanvarlistm noconstant may be used only with distributiontnormal With this option sfcross specifies the mean of the truncated normal distribution in terms of a linear function of the covariates defined in varlistm Specifying noconstant suppresses the constant in this function usigmavarlistu noconstant specifies that the technical inefficiency component is heteroskedastic with the variance expressed as a function of the covariates defined in varlistu Specifying noconstant suppresses the constant in this function vsigmavarlistv noconstant specifies that the idiosyncratic error component is heteroskedastic with the variance expressed as a function of the covariates defined in varlistv Specifying noconstant suppresses the constant in this function svfrontierlistname specifies a 1 k vector of initial values for the coefficients of the frontier The vector must have the same length of the parameter vector to be estimated svemeanlistname specifies a 1 km vector of initial values for the coefficients of the conditional mean model This can be specified only with distributiontnormal svusigmalistname specifies a 1 ku vector of initial values for the coefficients of the technical inefficiency variance function svvsigmalistname specifies a 1 kv vector of initial values for the coefficients of the idiosyncratic error variance function cost specifies that sfcross fit a cost frontier model simtypesimtype specifies the method for random draws when distributiongamma is specified runiform generates uniformly distributed random variates halton and genhalton create respectively Halton sequences and generalized Halton sequences where the base is expressed by the prime number in base The default is simtyperuniform See help mata halton for more details on generating Halton sequences nsimulations specifies the number of draws used when distributiongamma is specified The default is nsimulations250 base specifies the number preferably a prime used as a base for the generation of Halton sequences and generalized Halton sequences when distributiongamma is specified The default is base7 Note that Halton sequences based on large primes 10 can be highly correlated and their coverage may be worse than that of the pseudorandom uniform sequences postscore saves an observationbyobservation matrix of scores in the estimation results list Scores are defined as the derivative of the objective function with respect to the parameters This option is not allowed when the size of the scores matrix is greater than the Stata matrix limit see R limits posthessian saves the Hessian matrix corresponding to the full set of coefficients in the estimation results list 32 Postestimation command after sfcross After the estimation with sfcross the predict command can be used to compute linear predictions inefficiency and score variables Moreover the sfcross postestimation command allows one to compute the inefficiency confidence interval through the option ci as well as nonmonotonic marginal effects in the manner of Wang 2002 using when appropriate the option marginal The syntax of the command is the following predict type newvar if in statistic predict type stubnewvarxb newvarv newvaru if in scores where statistic includes xb stdp u m jlms bc ci and marginal xb the default calculates the linear prediction stdp calculates the standard error of the linear prediction u produces estimates of technical or cost inefficiency via Euε using the estimator of Jondrow et al 1982 m produces estimates of technical or cost inefficiency via Muε the mode of the conditional distribution of uε This option is not allowed when the estimation is performed with the distributiongamma option jlms produces estimates of technical or cost efficiency via expEuε bc produces estimates of technical or cost efficiency via Eexpuε the estimator of Battese and Coelli 1988 This option is not allowed when the estimation is performed with the distributiongamma option ci computes the confidence interval using the approach proposed by Horrace and Schmidt 1996 It can be used only when u or bc is specified The default is level95 or a 95 confidence interval If the option level is used in the previous estimation command the confidence interval will be computed using the level This option creates two additional variables newvarLBcilevel and newvarUBcilevel the lower and the upper bound respectively This option is not allowed when the estimation is performed with the distributiongamma option marginal calculates the marginal effects of the exogenous determinants on Eu and Varu The marginal effects are observation specific and are saved in the new variables varnamemM and varnameuV the marginal effects on the mean and the variance of the inefficiency respectively varnamem and varnameu are the names of each exogenous determinant specified in options emeanvarlistm noconstant and usigmavarlistu noconstant marginal can be used only if the estimation is performed with the distributiontnormal option When they are both specified varlistm and varlistu must contain the same variables in the same order This option can be specified in two ways i together with u m jlms or bc and ii alone without specifying newvar scores calculates score variables When the argument of the option distribution is hnormal tnormal or exponential score variables are generated as the derivative of the objective function with respect to the parameters When the argument of the option distribution is gamma they are generated as the derivative of the objective function with respect to the coefficients This difference is due to the different moptmize evaluator type used to implement the estimators see help mata moptimize 4 The sfpanel command sfpanel allows for the estimation of SF paneldata models through ML and leastsquares techniques The general sfpanel syntax is the following sfpanel depvar indepvars if in weight modelmodeltype options As for its crosssectional counterpart version 112 is the earliest version of Stata that can be used to run sfpanel Similarly all types of weights are allowed but the declared weight variable must be constant within each unit of the panel Moreover the command does not support the svy prefix The default model is the timedecay model of Battese and Coelli 1992 A description of the main commandspecific estimation and postestimation options is provided below A full description of all available options is provided in the sfpanel help file 41 Main options for sfpanel True fixed and randomeffects models Greene 2005ab distributiondistname specifies the distribution for the inefficiency term as half normal hnormal truncated normal tnormal or exponential exponential The default is distributionexponential emeanvarlistm noconstant may be used only with distributiontnormal With this option sfpanel specifies the mean of the truncated normal distribution in terms of a linear function of the covariates defined in varlistm Specifying noconstant suppresses the constant in this function usigmavarlistu noconstant specifies that the technical inefficiency component is heteroskedastic with the variance expressed as a function of the covariates defined in varlistu Specifying noconstant suppresses the constant in this function vsigmavarlistv noconstant specifies that the idiosyncratic error component is heteroskedastic with the variance expressed as a function of the covariates defined in varlistv Specifying noconstant suppresses the constant in this function feshow allows the user to display estimates of individual fixed effects along with structural parameters Only for modeltfe simtypesimtype specifies the method to generate random draws for the unitspecific random effects runiform generates uniformly distributed random variates halton and genhalton create respectively Halton sequences and generalized Halton sequences where the base is expressed by the prime number in base The default is simtyperuniform See help mata halton for more details on generating Halton sequences Only for modeltre nsimulations specifies the number of draws used in the simulation The default is nsimulations250 Only for modeltre base specifies the number preferably a prime used as a base for the generation of Halton sequences and generalized Halton sequences The default is base7 Note that Halton sequences based on large primes 10 can be highly correlated and their coverage may be worse than that of the pseudorandom uniform sequences Only for modeltre ML randomeffects timevarying inefficiency effects model Battese and Coelli 1995 emeanvarlistm noconstant fits the Battese and Coelli 1995 conditional mean model in which the mean of the truncated normal distribution is expressed as a linear function of the covariates specified in varlistm Specifying noconstant suppresses the constant in this function usigmavarlistu noconstant specifies that the technical inefficiency component is heteroskedastic with the variance expressed as a function of the covariates defined in varlistu Specifying noconstant suppresses the constant in this function vsigmavarlistv noconstant specifies that the idiosyncratic error component is heteroskedastic with the variance expressed as a function of the covariates defined in varlistv Specifying noconstant suppresses the constant in this function ML randomeffects flexible timevarying efficiency model Kumbhakar 1990 btvarlistbt noconstant fits a model that allows a flexible specification of technical inefficiency handling different types of time behavior using the formulation uit ui1 expvarlistbt1 Typically explanatory variables in varlistbt are represented by a polynomial in time Specifying noconstant suppresses the constant in the function The default includes a linear and a quadratic term in time without constant as in Kumbhakar 1990 42 Postestimation command after sfpanel After the estimation with sfpanel the predict command can be used to compute linear predictions inefficiency and score variables Moreover the sfpanel postestimation command allows one to compute the inefficiency confidence interval through the option ci as well as nonmonotonic marginal effects in the manner of Wang 2002 using when appropriate the option marginal The syntax of the command is the following predict type newvar if in statistic predict type stubnewvarxb newvarv newvaru if in scores where statistic includes xb stdp u u0 m jlms bc ci marginal and trunctlevel xb the default calculates the linear prediction stdp calculates the standard error of the linear prediction u produces estimates of technical or cost inefficiency via Euε using the estimator of Jondrow et al 1982 u0 produces estimates of technical or cost inefficiency via Euε using the estimator of Jondrow et al 1982 when the random effect is zero This statistic can only be specified when the estimation is performed with the modeltre option m produces estimates of technical or cost inefficiency via Muε the mode of the conditional distribution of uε This statistic is not allowed when the estimation is performed with the option modelfecss modelfels modelfe or modelregls jlms produces estimates of technical or cost efficiency via expEuε bc produces estimates of technical or cost efficiency via Eexpuε estimator of the Battese and Coelli 1988 This statistic is not allowed when the estimation is performed with the option modelfecss modelfels modelfe or modelregls ci computes the confidence interval using the approach of Horrace and Schmidt 1996 This option can only be used with the u jlms and bc statistics but not when the estimation is performed with the option modelfels modelbc92 modelkumb90 modelfecss modelfe or modelregls The default is level95 or a 95 confidence interval If the option level is used in the previous estimation command the confidence interval will be computed using the level This option creates two additional variables newvarLBci level and newvarUBci level the lower and the upper bound respectively marginal calculates the marginal effects of the exogenous determinants on Eu and Varu The marginal effects are observation specific and are saved in the new variables varnamemM and varnameuV the marginal effects on the unconditional mean and the variance of inefficiency respectively varnamem and varnameu are the names of each exogenous determinant specified in options emeanvarlistm noconstant and usigmavarlistu noconstant marginal can only be used if estimation is performed with the modelbc95 option or if the inefficiency in modeltfe or modeltre is distributiontnormal When they are both specified varlistm and varlistu must contain the same variables in the same order This option can be specified in two ways i together with u m jlms or bc and ii alone without specifying newvar trunctlevel excludes from the inefficiency estimation the units whose effects are at least at one time period in the upper and bottom tlevel range trunc can only be used if the estimation is performed with modelfe modelregls modelfecss and modelfels scores calculates score variables This option is not allowed when the estimation is performed with the option modelfecss modelfels modelfe or modelregls When the argument of the option model is tfe or bc95 score variables are generated as the derivative of the objective function with respect to the parameters F Belotti S Daidone G Ilardi and V Atella 735 When the argument of the option model is tre bc88 bc92 kumb90 or pl81 they are generated as the derivative of the objective function with respect to the coefficients This difference is due to the different moptimize evaluator type used to implement the estimators see help mata moptimize 5 Examples with simulated data In this section we use simulated data to illustrate sfcross and sfpanel estimation capabilities focusing on some of the models that cannot be estimated using official Stata routines12 51 The normalgamma SF production model There is a large debate in the SF literature about the nonidentifiability of the normal gamma crosssectional model Ritter and Simar 1997 pointed out that this model is difficult to distinguish from the normalexponential one and that the estimation of the shape parameter of the gamma distribution may require large sample sizes up to several thousand observations On the other hand Greene 2003 argued that their result was a matter of degree not a definitive result and that the nonidentifiability of the true value of the shape parameter remains an empirical question In this section we illustrate the sfcross command by fitting a normalgamma SF production model We consider the following datagenerating process DGP yi 1 03x1i 07x2i vi ui i 1 N vi N0 1 ui Γ2 2 where the inefficiency is gamma distributed with shape and scale parameters equal to 2 the idiosyncratic error is N0 1 and the two regressors x1i and x2i are normally distributed with 0 means and variances equal to 1 and 4 respectively The sample size is set to 1000 observations a large size as noted by Ritter and Simar 1997 but in general not so large given the current availability of microdata Let us begin by fitting the normalexponential model using the following syntax 12 We report the Mata code used for the DGP and the models estimation syntax for each example in the sj examples simdatado ancillary file 736 Stochastic frontier using Stata sfcross y x1 x2 distributionexp nolog Stoc frontier normalexponential model Number of obs 1000 Wald chi22 41988 Prob chi2 00000 Log likelihood 24230869 y Coef Std Err z Pz 95 Conf Interval Frontier x1 3709605 068792 539 0000 2361306 5057904 x2 6810641 0339945 2003 0000 6144361 747692 cons 1474677 1131198 130 0192 3691784 0742431 Usigma cons 2173649 0957468 2270 0000 1985989 2361309 Vsigma cons 3827463 1498911 255 0011 0889652 6765274 sigmau 2964844 1419372 2089 0000 2699305 3256505 sigmav 1210911 0907524 1334 0000 1045487 140251 lambda 2448441 2058941 1189 0000 2044895 2851986 estimates store exp predict uhatexp u Note that the normalexponential model is the sfcross default so we might omit the option distributionexponential13 As can be seen although there is only one equation to be estimated in the model the command fits three of Matas moptimize equations see M5 moptimize Indeed given that sfcross allows both the inef ficiency and the idiosyncratic error to be heteroskedastic see table 1 the output also reports variance parameters estimated in a transformed metric according to 11 and 12 respectively In this example the inefficiency is assumed to be homoskedastic so sfcross estimates the coefficient of the constant term in 11 rather than directly esti mating σu To make the output easily interpretable sfcross also displays the variance parameters in their natural metric As expected the normalexponential model produces biased results especially for the frontiers constant term and the inefficiency scale parameter σu We also run the predict command using the u option In this way inefficiency estimates are obtained through the approach of Jondrow et al 1982 Because the inefficiencies are drawn from a gamma distribution a better fit can be obtained using the following command 13 The option nolog allows one to omit the display of the criterion function iteration log sfcross and sfpanel allow one to use all maximize options available for ml estimation commands see help maximize and the additional options postscore and posthessian which report the score and the Hessian as an e vector and matrix respectively F Belotti S Daidone G Ilardi and V Atella 737 sfcross y x1 x2 distributiongamma nsim50 simtypegenha base7 nolog Stoc frontier normalgamma model Number of obs 1000 Wald chi22 43802 Prob chi2 00000 Log simulatedlikelihood 24190008 Number of Randomized Halton Sequences 50 Base for Randomized Halton Sequences 7 y Coef Std Err z Pz 95 Conf Interval Frontier x1 3809769 0670487 568 0000 2495639 5123899 x2 6877634 0336088 2046 0000 6218914 7536354 cons 9361791 4121864 227 0023 1283087 174405 Usigma cons 153519 226486 678 0000 1091286 1979094 Vsigma cons 2734356 333033 082 0412 9261682 379297 sigmau 2154578 2439909 883 0000 1725717 2690016 sigmav 8722163 1452384 601 0000 6293397 1208825 lambda 2470234 1969744 1254 0000 2084171 2856297 gshape 1879186 3845502 489 0000 1125482 2632891 estimates store gamma predict uhatgamma u In the normalgamma crosssectional model the parameters are estimated using sim ulated maximum likelihood SML A better approximation of the loglikelihood function requires the right choice about the number of draws and the way they are created In this example we use generalized Halton sequences simtypegenhalton with base equal to 7 base7 and only 50 draws nsim50 Indeed a Halton sequence gen erally has a more uniform coverage than a sequence generated from pseudouniform random numbers Moreover as noted by Greene 2003 the computational efficiency when compared with that of pseudouniform random draws appears to be at least 10 to 1 Thus in our example the same results can be approximately obtained using 500 pseudouniform draws see help mata halton14 14 For all models fit using SML the default option of sfcross and sfpanel is simtypeuniform with nsim250 In our opinion small values for example 50 for Halton sequences and 250 for pseu douniform random draws are sufficient for exploratory work On the other hand larger values in the order of several hundreds are advisable for more precise results We suggest using Halton se quences rather than pseudorandom random draws However as pointed out by Drukker and Gates 2006 Halton sequences based on large primes d 10 can be highly correlated and their coverage can be worse than that of the pseudorandom uniform sequences 738 Stochastic frontier using Stata As expected in this example the parameters of the normalgamma model are prop erly estimated Furthermore this model is preferable to the normalexponential one as corroborated by the following likelihoodratio test15 lrtest exp gamma Likelihoodratio test LR chi21 817 Assumption exp nested in gamma Prob chi2 00043 Similar conclusions may be drawn by comparing the estimated mean inefficiencies with the true simulated one even if the Spearman rank correlation with the latter is high and very similar for both uhat gamma and uhat exp16 summarize u uhatgamma uhatexp Variable Obs Mean Std Dev Min Max u 1000 4097398 291035 0259262 1990251 uhatgamma 1000 4048885 2839368 4752663 2027557 uhatexp 1000 2964844 264064 363516 1895619 spearman u uhatgamma uhatexp obs1000 u uhatga uhatexp u 10000 uhatgamma 09141 10000 uhatexp 09145 09998 10000 52 Paneldata timevarying inefficiency models Cornwell Schmidt and Sickles 1990 and Lee and Schmidt 1993 provide a fixed effects treatment of models like those proposed by Kumbhakar 1990 and Battese and Coelli 1992 Currently sfpanel allows for the estimation of the models of Corn well Schmidt and Sickles 1990 and Lee and Schmidt 1993 by means of modified leastsquares dummy variables and iterative least squares ILS respectively An in teresting aspect of these models is that although they have been proposed in the SF literature they are actually linear paneldata models with timevarying fixed effects and thus potentially very useful in other contexts However their consistency requires white noise errors and they are less efficient than the GMM estimator proposed by Ahn Hoon Lee and Schmidt 2001 and Han Orea and Schmidt 2005 15 Notice that exp and gamma are the names of the exponential and gamma models estimation results saved with the estimates store command 16 In line with Ritter and Simar 1997 our simulation results indicate that in the normalgamma model a relatively large sample is needed to achieve a reasonable degree of precision in the estimates of inefficiency distribution parameters F Belotti S Daidone G Ilardi and V Atella 739 In this section we report the main syntax to fit such models We start by specifying the following stochastic production frontier translog model yit uit 02x1it 06x2it 06x3it 02x2 1it 01x2 2it 02x2 3it 015x1itx2it 03x1itx3it 03x2itx3it vit vit N0 025 i 1 n t 1 T As already mentioned the main feature of these models is the absence of any distri butional assumption about inefficiency In this example the DGP follows the Lee and Schmidt 1993 model where ui δiξ For each unit the parameter δi is drawn from a uniform distribution in 0 12τ 1 1 with τ 08 The elements of the vector ξ ξ1 ξT are equally spaced between 2 and 2 This setup implies a standard deviation of the inefficiency term σu 183 Once the sample is declared to be a panel see help xtset the models of Lee and Schmidt 1993 and Cornwell Schmidt and Sickles 1990 can be estimated using the following syntaxes sfpanel y x1 x2 x3 x1sq x2sq x3sq x1x2 x1x3 x2x3 modelfels output omitted estimates store fels predict uhatfels u sfpanel y x1 x2 x3 x1sq x2sq x3sq x1x2 x1x3 x2x3 modelfecss output omitted estimates store fecss predict uhatfecss u Notice that we use the predict command with the u option to postestimate inef ficiency As an additional source of comparison we use the same simulated data to assess the behavior of the Schmidt and Sickles 1984 timeinvariant inefficiency model The fixedeffects version of this model can be fit using sfpanel and the official xtreg command However when the estimation is performed using sfpanel the predict command with the option u can be used to obtain inefficiency estimates17 sfpanel y x1 x2 x3 x1sq x2sq x3sq x1x2 x1x3 x2x3 modelfe output omitted estimates store fesssf predict uhatfess u xtreg y x1 x2 x3 x1sq x2sq x3sq x1x2 x1x3 x2x3 fe output omitted estimates store fessxt Table 2 reports the estimation results from the three models Unsurprisingly both the frontier and variance parameters are well estimated in the ls93 and css90 models This result shows that when the DGP follows the model by Lee and Schmidt 1993 the 17 Both xtreg and sfpanel also allow for the estimation of the randomeffects version of this model through the feasible generalized leastsquares approach estimator by Cornwell Schmidt and Sickles 1990 provides reliable results On the other hand being the data generated from a timevarying model variance estimates from the ss84 model show a substantial bias Table 2 Schmidt and Sickles ss84 Cornwell Schmidt and Sickles css90 and Lee and Schmidt ls93 estimation results ss84 css90 ls93 x1 0254 0185 0171 00695 00237 00230 x2 0626 0619 0611 00354 00121 00117 x3 0602 0591 0596 00220 00073 00075 x1sq 0193 0204 0209 00234 00078 00076 x2sq 0099 0103 0101 00080 00027 00026 x3sq 0198 0201 0201 00036 00012 00012 x1x2 0149 0142 0145 00198 00066 00064 x1x3 0293 0295 0295 00130 00043 00043 x2x3 0306 0300 0301 00076 00026 00025 cons 0050 00866 σu 0223 1859 1832 σv 2096 0499 0497 We do not expect large differences with regard to inefficiency scores given the similarities in terms of variance estimates between css90 and ls93 Note that for these models including ss84 inefficiency scores are retrieved in postestimation with the assumption that the best decisionmaking unit is fully efficient18 As seen in the following summarize command both css90 and ls93 average inefficiencies are close to the true values while the Spearman rank correlations are almost equal to 1 As expected the ss84 estimated inefficiencies are highly biased and the corresponding units ranking is completely unreliable 18 This assumption involves calculating ûi α αi with α maxi1nαi in the case of timeinvariant inefficiency models and uit αt αit with αt maxi1nαit in the case of timevarying inefficiency models F Belotti S Daidone G Ilardi and V Atella 741 summarize u uhatfels uhatfecss uhatfess Variable Obs Mean Std Dev Min Max u 2500 1345894 1265155 0 4486315 uhatfels 2500 1669869 1419268 0 5550043 uhatfecss 2500 2214224 1314245 0 6500714 uhatfess 2500 645184 2232496 0 127254 spearman u uhatfels uhatfecss uhatfess obs2500 u uhatls uhatcss uhatess u 10000 uhatfels 09794 10000 uhatfecss 08974 09129 10000 uhatfess 00005 00092 01991 10000 Finally we show additional features of sfpanel i the possibility of computing elasticities via the official lincom command and ii the possibility of performing a constrained fixedeffects estimation which is not yet available with xtreg With respect to the former point it is well known that parameters in a translog pro duction frontier do not represent output elasticities In particular a linear combination of frontier parameters is needed for computing such elasticities Moreover to calculate output elasticities at means we first need to compute and store the mean for each input variable using the following syntax quietly summarize x1 scalar x1m rmean quietly summarize x2 scalar x2m rmean quietly summarize x3 scalar x3m rmean Then the lincom command can be used to combine estimated frontier parameters using the following standard syntax lincom x1 x1sq x1m x1x2x2m x1x3x3m 1 x1 1108946x1sq 1074533x1x2 105167x1x3 0 y Coef Std Err t Pt 95 Conf Interval 1 3203578 05348 599 0000 2154752 4252405 lincom x2 x2sq x2m x1x2x1m x2x3x3m 1 x2 1074533x2sq 1108946x1x2 105167x2x3 0 y Coef Std Err t Pt 95 Conf Interval 1 5751999 0254143 2263 0000 5253585 6250413 742 Stochastic frontier using Stata lincom x3 x3sq x3m x1x3x1m x2x3x2m 1 x3 105167x3sq 1108946x1x3 1074533x2x3 0 y Coef Std Err t Pt 95 Conf Interval 1 156379 0158945 984 0000 1252075 1875505 Finally the constant return to scale CRS hypothesis can be trivially tested by using the following syntax lincom x1 x1sq x1m x1x2x2m x1x3x3m x2 x2sq x2m x1x2x1m x2x3x3m x3 x3sq x3m x1x3x1m x2x3x2m 1 1 x1 x2 x3 1108946x1sq 1074533x2sq 105167x3sq 218348x1x2 2160617x1x3 2126204x2x3 1 y Coef Std Err t Pt 95 Conf Interval 1 0519367 0609852 085 0395 0676648 1715383 In this example the CRS hypothesis cannot be rejected To run a constrained fixed effects estimation we can define the required set of constraints to impose CRS through the official Stata command constraint using the following syntax Constraints definition constraint define 1 x1 x2 x3 1 constraint define 2 x1sq x1x2 x1x3 0 constraint define 3 x2sq x1x2 x2x3 0 constraint define 4 x3sq x1x3 x2x3 0 Then the constrained model can be estimated using sfpanel with the modelfe and constraints1 2 3 4 options F Belotti S Daidone G Ilardi and V Atella 743 sfpanel y x1 x2 x3 x1sq x2sq x3sq x1x2 x1x3 x2x3 modelfe constraints1 2 3 4 Timeinvariant fixedeffects model LSDV Number of obs 2500 Group variable id Number of groups 500 Time variable time Obs per group min 5 avg 50 max 5 1 x1 x2 x3 1 2 x1sq x1x2 x1x3 0 3 x2sq x1x2 x2x3 0 4 x3sq x1x3 x2x3 0 y Coef Std Err z Pz 95 Conf Interval x1 3530365 0851901 414 0000 1860671 520006 x2 5092917 0434568 1172 0000 4241179 5944655 x3 1376718 0270375 509 0000 0846792 1906644 x1sq 0343576 0287476 120 0232 0907019 0219868 x2sq 1282553 0098209 1306 0000 1090067 1475039 x3sq 21594 004442 4861 0000 2072339 2246461 x1x2 0610211 0242651 251 0012 0134624 1085799 x1x3 0266635 0159577 167 0095 0579401 0046131 x2x3 1892764 0092834 2039 0000 2074716 1710813 cons 2326412 1062126 219 0029 0244682 4408141 sigmau 7140381 sigmav 25700643 The constrained frontier estimates are more biased than the unconstrained ones but are still not too far from the true values This is an artifact of our DGP because the scale elasticity has been simulated without imposing CRS 53 True fixed and randomeffects models As already discussed in section 22 the true fixed and randomeffects models allow one to disentangle timeinvariant heterogeneity from timevarying inefficiency In this section we present the main syntax and some of the options to fit such models We start by specifying the following normalexponential stochastic production frontier model yit 1 αi 03x1it 07x2it vit uit vit N0 1 13 uit E 2 i 1 n t 1 T 14 where the nuisance parameters αi i 1 n are drawn from a N0 θ2 with θ 15 In the fixedeffects design TFEDGP the two regressors x1it and x2it are distributed for each unit according to a normal distribution centered in the corresponding uniteffect αi with variances equal to 1 and 4 respectively This design ensures correlation between regressors and individual effects a typical scenario in which the fixedeffects specification represents the consistent choice19 19 Notice that higher values of θ correspond to higher correlations between the regressors and the unitspecific effects 744 Stochastic frontier using Stata As far as the randomeffects design is concerned TREDGP x1it and x2it are not correlated with the unitspecific effects and are distributed according to a normal dis tribution with 0 mean and variances equal to 1 and 4 respectively The generated sample consists of a balanced panel of 1000 units observed for 10 periods for a total of 10000 observations Once the sample is declared as a panel we fit the following models i a normalexponential TFE model on TFEDGP data tfe120 sfpanel yf x1c x2c modeltfe distributionexp rescale output omitted estimates store tfec predict utfec u ii a normalexponential TRE model on TFEDGP data tre1 sfpanel yf x1c x2c modeltre distributionexp nsim50 simtypegenhalton base7 rescale output omitted estimates store trec predict utrec u iii a normalexponential TRE model on TREDGP data tre2 sfpanel yr x1nc x2nc modeltre distributionexp nsim100 simtypegenhalton base7 rescale output omitted estimates store trenc predict utrenc u predict u0trenc u0 As shown in the first column of table 3 when the model is correctly specified the frontier parameters are properly estimated However in this example the MLDV estimator of σv is slightly biased by the incidental parameter problem even if the length of the panel is quite large21 This problem does not seem to affect variance estimates in the tre1 model In this case the parameters are estimated using the SML technique assuming that the unobserved heterogeneity is distributed as N0 θ2 where θ represents the standard deviation of the unobserved heterogeneity and that Eαix1it x2it 0 Thus because the estimates are obtained using the TFEDGP data the frontier and θ parameter estimates are biased 20 Note that yf x1 c and x2 c are the variables from the TFEDGP while yr x1 nc and x2 nc are from the TREDGP 21 See section 22 for a discussion of the MLDV estimator problems in the TFE model F Belotti S Daidone G Ilardi and V Atella 745 Table 3 TFE and TRE estimation results tfe1 tre1 tre2 x1 c 0304 0776 00164 00198 x2 c 0700 0811 00081 00094 x1 nc 0295 00176 x2 nc 0706 00089 cons 1062 1090 00342 00540 σu 2075 2035 2023 σv 0770 1095 0973 θ 0602 1542 On the contrary by fitting a correctly specified TRE model on TREDGP data column tre2 in table 3 all parameters including the frontier ones are accurately estimated After each estimation we use the predict command to obtain inefficiency esti mates As already mentioned option u instructs the postestimation routine to com pute inefficiencies through the estimator of Jondrow et al 1982 see help sfpanel postestimation In the case of the TRE model the predict command also allows for the option u0 to estimate inefficiencies assuming the random effects are zero At this point we can summarize the estimated inefficiencies to compare them with the actual values summarize u utfec utrec utrenc u0trenc Variable Obs Mean Std Dev Min Max u 10000 2004997 200852 0003777 2083139 utfec 10000 2075017 1948148 2008319 2042197 utrec 10000 2034946 1818154 2430926 1876244 utrenc 10000 2025002 1831147 2656734 1998998 u0trenc 10000 2200728 2086419 1338385 1947738 spearman u utfec utrec utrenc u0trenc obs10000 u utfec utrec utrenc u0trec u 10000 utfec 07654 10000 utrec 07541 09291 10000 utrenc 07700 09925 09464 10000 u0trenc 06297 07313 08168 07965 10000 746 Stochastic frontier using Stata All the estimates of Jondrow et al 1982 are very close to the true simulated ones u Actually the estimated average inefficiency after a correctly specified TRE model shows a lower bias than the estimated average inefficiency after a correctly specified TFE model This is due to the incidental parameters problem Also note the good performances of the TRE model when it is fit on the TFEDGP data u tre c Introducing heteroskedasticity Finally we deal with the problem of heteroskedasticity a very important issue for applied research For both TFE and TRE models we compare the estimates obtained from a model that neglects heteroskedasticity with those obtained from a heteroskedastic one To introduce heteroskedasticity we replace equations 1314 with the following vit N0 σvit uit E σuit σvit exp 051 5 zvit σuit exp 052 1 zuit where both inefficiency and idiosyncraticerror scale parameters are now a function of a constant term and of an exogenous covariate zuit and zvit drawn from a standard normal random variable Note that because of the introduction of heteroskedasticity we will deal with average σu and σv which in our simulated sample are approximately 31 and 17 respectively In this case each observation has a different signaltonoise ratio which implies an average of about 19 We estimate four different models i a homoskedastic TFE model on heteroskedastic TFEDGP data tfe1 sfpanel yf x1c x2c modeltfe distributionexp rescale output omitted estimates store tfehom predict utfehom u ii a heteroskedastic TFE model on heteroskedastic TFEDGP data tfe2 sfpanel yf x1c x2c modeltfe distributionexp usigmazu vsigmazv output omitted estimates store tfehet predict utfehet u iii a homoskedastic TRE model on heteroskedastic TREDGP data tre1 sfpanel yr x1nc x2nc modeltre distributionexp nsim50 simtypegenhalton base7 rescale output omitted estimates store trehom predict utrehom u F Belotti S Daidone G Ilardi and V Atella 747 iv a heteroskedastic TRE model on heteroskedastic TREDGP data tre2 sfpanel yr x1nc x2nc modeltre distributionexp usigmazu vsigmazv nsim50 simtypegenhalton base7 rescale output omitted estimates store trehet predict utrehet u predict u0trehet u0 Estimation results are reported in table 4 As expected tfe1 variance estimates are biased by both the incidental parameters problem and the neglected heteroskedasticity in u and v These estimates can be significantly improved by considering both sources of heteroskedasticity using the options usigmavarlist u and vsigmavarlist v tfe2 Exactly the same argument applies in the TRE case tre1 versus tre2 but without the incidental parameters problem Table 4 TFE and TRE estimation results homoskedasticity versus heteroskedasticity tfe1 tfe2 tre1 tre2 x1 c 0324 0295 00271 00245 x2 c 0723 0732 00134 00121 x1 nc 0316 0310 00290 00264 x2 nc 0681 0689 00147 00135 cons 1576 1113 00637 00652 σu 3717 3264 3642 3168 σv 1185 1402 1526 1693 θ 1579 1565 As we mentioned in section 23 neglecting heteroskedasticity in u and v leads to biased inefficiency estimates This conclusion is confirmed by the summarize command summarize u utfehom utfehet utrehom utrehet u0trehet Variable Obs Mean Std Dev Min Max u 10000 3091925 3915396 000169 5220689 utfehom 10000 3717061 3941147 3442658 5154804 utfehet 10000 3271297 3828366 2642199 5206564 utrehom 10000 3641955 3788298 3739219 5176109 utrehet 10000 3173224 3709123 3241621 5183721 u0trehet 10000 32855 3844297 1828969 542632 748 Stochastic frontier using Stata The average inefficiency is upward biased by about 15 for both TFE and TRE models in which heteroskedasticity has been neglected A slightly better result is also obtained in terms of Spearmans rank correlation spearman u utfehom utfehet utrehom utrehet u0trehet obs10000 u utfem utfet utrem utret u0tret u 10000 utfehom 07287 10000 utfehet 07536 09589 10000 utrehom 07380 09830 09531 10000 utrehet 07623 09461 09835 09642 10000 u0trehet 07039 08173 08455 08944 09121 10000 6 Empirical applications In this section we illustrate sfcross and sfpanel capabilities through two empirical applications from the SF literature The first analyzes the cost inefficiency of Swiss railways using data from the Swiss Federal Office of Statistics on public transport companies the second focuses on the technical inefficiency of Spanish dairy farms using data from a voluntary recordkeeping program22 61 Swiss railways This application is based on an unbalanced panel of 50 railway companies over the period 19851997 which resulted in 605 observations We think that this application is interesting for at least two reasons i cost frontiers are much less diffuse in the literature compared with production frontiers given the lack of reliable cost and price data and ii the length of the panel makes this database quite unusual in the SF literature A detailed description of the Swiss railway transport system and complete information on the variables used are available in Farsi Filippini and Greene 2005 To estimate a CobbDouglas cost frontier we impose linear homogeneity by normalizing total costs and prices through the price of energy Therefore the model can be written as lnTCit PEit β0 βY ln Yit βQ ln Qit βN ln Nit βpk ln Pkit Peit βPl ln Plit Peit t19861997 βt dyeart uit vit 15 where i and t are the subscripts denoting the railway company and year respectively As is common uit is interpreted as a measure of cost inefficiency Two output measures are included in the cost function passenger output and freight output Length of network 22 Both datasets are freely available from the webpage of professor William Greene httppeoplesternnyueduwgreene F Belotti S Daidone G Ilardi and V Atella 749 is included as an output characteristic Further we have price data for three inputs capital labor and energy All monetary values including total costs are in 1997 Swiss Francs CHF We have also included a set of time dummies dyeart to control for unobserved timedependent variation in costs We consider three timevarying inefficiency specificationsthe Kumbhakar 1990 model kumb90 the Battese and Coelli 1992 model bc92 and the Greene 2005a randomeffects model treand three timeinvariant models With respect to the latter group we estimate the fixedeffects version of the Schmidt and Sickles 1984 model ss84 the Pitt and Lee 1981 pl81 model and the Battese and Coelli 1988 bc88 model All models are fit assuming that the inefficiency is halfnormally distributed that is all except bc88 and bc92 in which u Nμσu2 and ss84 in which no distributional assumption is made The choice of also including Greenes specification is driven by the multioutput technology that characterizes a railway company for which unmeasured quality captured by the random effects may play an important role in the production process Finally as a benchmark we fit a pooled crosssectional model pcs Table 5 shows the results Coefficient estimates of input prices and outputs are all significant across the seven models and have the expected signs positive marginal costs and positive ownprice elasticities Looking at table 6 we further observe that the three timeinvariant specifications provide inefficiency estimates that are highly correlated Perhaps the most interesting result is that inefficiency scores obtained from kumb90 and bc92 models are also highly correlated with those coming from timeinvariant models table 6 and figure 1 This is not surprising because the two timeinvariance hypotheses H0 t t2 0 in the kumb90 model and H0 η 0 in the bc92 specification cannot be rejected at the 5 level Hence we may conclude that there is evidence of timeinvariant technical inefficiency in the Swiss railway transport system at least for the study period Consistently with this result we also find that the tre model provides inefficiency estimates that have no link with those obtained from any of the other models Moreover because of a very low estimate of the inefficiency variance the estimated signaltonoise ratio λ is the lowest one In our opinion these results are driven from the peculiar timevarying inefficiency specification of this model Indeed when the inefficiency term is constant over time the tre specification does not allow one to disentangle timeinvariant unobserved heterogeneity from inefficiency This interpretation is supported by the fact that the estimated standard deviation of the random effects θ dominates the inefficiency one σu 750 Stochastic frontier using Stata Table 5 Swiss railways estimation results 50 firms for a total of 605 observations pcs ss pl81 bc88 kumb90 bc92 tre bse bse bse bse bse bse bse lnY 0492 0114 0200 0199 0193 0199 0201 0015 0032 0034 0033 0033 0033 0026 lnQ 0030 0014 0021 0021 0020 0020 0028 0006 0006 0006 0006 0006 0006 0005 lnN 0393 0448 0485 0503 0477 0499 0583 0027 0051 0045 0047 0044 0047 0034 lnpk 0171 0318 0310 0311 0311 0313 0311 0032 0017 0017 0017 0017 0017 0017 lnpl 0592 0546 0548 0546 0538 0543 0560 0074 0037 0037 0037 0037 0037 0037 dyear1986 0009 0010 0009 0009 0015 0008 0015 0056 0015 0015 0015 0015 0015 0015 dyear1987 0003 0020 0012 0012 0023 0009 0018 0056 0015 0015 0015 0017 0015 0015 dyear1988 0010 0039 0028 0027 0043 0023 0034 0057 0015 0015 0015 0019 0016 0016 dyear1989 0036 0065 0052 0052 0070 0046 0058 0057 0016 0016 0016 0021 0016 0016 dyear1990 0024 0084 0068 0068 0086 0060 0074 0058 0016 0016 0016 0022 0017 0016 dyear1991 0030 0098 0078 0078 0096 0069 0086 0058 0017 0018 0017 0024 0019 0018 dyear1992 0046 0111 0094 0094 0109 0083 0101 0058 0017 0017 0017 0023 0019 0017 dyear1993 0015 0100 0081 0081 0092 0069 0089 0057 0017 0017 0017 0023 0020 0017 dyear1994 0001 0082 0063 0063 0069 0049 0070 0056 0017 0017 0017 0022 0020 0017 dyear1995 0019 0059 0048 0047 0045 0031 0050 0057 0016 0016 0016 0022 0021 0016 dyear1996 0027 0037 0028 0027 0018 0010 0017 0057 0017 0016 0016 0022 0022 0017 dyear1997 0019 0038 0030 0029 0009 0009 0029 0060 0018 0017 0017 0023 0024 0017 Constant 8310 2682 4895 4929 4626 4871 4894 0976 0652 0643 0634 0637 0637 0531 Continued on next page F Belotti S Daidone G Ilardi and V Atella 751 pcs ss pl81 bc88 kumb90 bc92 tre bse bse bse bse bse bse bse t 0023 0015 t2 0002 0001 η 0002 0002 λ 2882 7803 11366 7716 23930 7887 1634 σ 0464 0560 0807 0551 1682 0562 0098 σu 0438 0555 0804 0546 1681 0557 0083 σv 0152 0071 0071 0071 0070 0071 0051 θ 0347 Estimated cost inefficiencies ûit Mean 0350 0807 0663 0679 0687 0682 0091 SD 0233 0550 0429 0425 0445 0428 0076 Min 0060 0000 0015 0020 0015 0019 0018 Max 1134 2507 2006 1991 2124 2031 0629 Log likelihood 116572 595159 596523 597649 597285 595516 Notes Standard errors for ancillary parameters are not reported Table 6 Swiss railways correlation of inefficiency estimates Variables pcs ss84 pl81 bc88 kumb90 bc92 tre pcs 1000 ss84 0439 1000 pl81 0595 0975 1000 bc88 0608 0971 0991 1000 kumb90 0573 0984 0990 0998 1000 bc92 0603 0974 0991 1000 0998 1000 tre 0140 0378 0405 0407 0400 0406 1000 Figure 1 Swiss railways inefficiency scatterplots and output volumes heterogeneity of output and less than ideal proxies for inputs The output variable is given by the liters of milk produced per year This measure explains only partially the final output of this industry milk can also be considered an intermediate input to produce dairy products Furthermore variables such as slaughtered animals should also be considered part of the final output The functional form employed in the empirical analysis is the following translog production function with time dummy variables to control for neutral technical change ln yit β0 sumj14 βj ln xjit 12 sumj14 sumk14 βjk ln xjit ln xkit sumt19931998 βt dyeart uit vit 16 where j and t are the subscripts denoting farm and year respectively Four inputs have been employed in the production frontier number of milking cows x1 number of manequivalent units x2 hectares of land devoted to pasture and crops x3 and kilograms of feedstuffs fed to the dairy cows x4 More details on these variables are available in Cuesta 2000 and Alvarez and Arias 2004 We have fit three models with timevarying inefficiency the normalhalf normal Kumbhakar 1990 model kumb90 a randomeffects model fit through the feasible generalized leastsquares method the Cornwell Schmidt and Sickles 1990 model css90 fit through the modified leastsquares dummy variable technique and finally the Lee and Schmidt 1993 model ls93 fit through ILS Note that the last two models are fit using approaches that do not allow intercept β0 and time dummies dyeart F Belotti S Daidone G Ilardi and V Atella 753 to be simultaneously included in the frontier equation Finally we also considered two models with timeinvariant inefficiency that is the uit term reduced to ui in 16 the first was proposed by Schmidt and Sickles 1984 and estimated without any distri butional assumption through the leastsquares dummy variable approach ss84 the second was proposed by Pitt and Lee 1981 and estimated through ML assuming a halfnormal inefficiency pl81 Table 7 reports the results of our exercise There is a certain degree of similarity between the different models because both parameter significance and magnitudes are comparable For the ss84 css90 and ls93 models the most efficient firm in the sample for each period is considered as fully efficient thus the smallest value of inefficiency is 0 On average and as expected the css90 model shows a higher level of inefficiency and its distribution also has more variability while the other models seem to behave very similarly in this application Finally as we can see in table 8 linear correlations between inefficiencies are very high This does not come as a surprise given the similarity of the estimated frontier parameters and it looks like an indication that in medium short panels and in certain economic sectors and contexts a timeinvariant inefficiency specification is a valid solution 754 Stochastic frontier using Stata Table 7 Spanish dairy farms estimation results 247 firms for a total of 1482 obs ss84 css90 ls93 kumb90 pl81 bse bse bse bse bse x1 0642 0527 0641 0661 0660 0036 0065 0036 0028 0028 x2 0037 0043 0037 0038 0041 0017 0027 0017 0015 0015 x3 0011 0079 0010 0050 0049 0025 0063 0025 0018 0018 x4 0308 0226 0307 0351 0356 0020 0035 0020 0018 0017 x11 0135 0187 0133 0308 0314 0157 0192 0155 0171 0178 x22 0002 0060 0001 0111 0112 0069 0111 0068 0064 0067 x33 0242 0168 0243 0129 0131 0188 0317 0187 0119 0115 x44 0105 0125 0105 0112 0118 0050 0084 0050 0048 0049 x12 0010 0059 0009 0060 0064 0073 0100 0072 0077 0081 x13 0084 0114 0085 0088 0091 0102 0158 0101 0090 0090 x14 0075 0142 0074 0140 0146 0083 0132 0082 0084 0088 x23 0001 0067 0002 0020 0011 0050 0107 0050 0049 0050 x24 0011 0062 0011 0025 0025 0041 0060 0041 0039 0040 x34 0012 0110 0013 0015 0017 0046 0085 0046 0041 0041 dyear1994 0035 0042 0027 0007 0010 0007 dyear1995 0062 0072 0048 0009 0014 0008 dyear1996 0072 0078 0052 0010 0016 0009 dyear1997 0075 0074 0051 0010 0017 0009 dyear1998 0092 0077 0064 0012 0018 0010 Constant 11512 11695 11711 0016 0019 0016 Continued on next page ss84 css90 ls93 kumb90 p181 bse bse bse bse bse t 0347 0212 t2 0045 0028 λ 1948 3711 2010 4485 2775 σ 0168 0237 0171 0356 0230 σu 0149 0229 0153 0348 0216 σv 0077 0062 0076 0077 0078 Estimated technical inefficiencies uit Mean 0315 0531 0316 0182 0179 SD 0149 0227 0150 0119 0117 Min 0000 0000 0000 0008 0009 Max 0873 1412 0879 0667 0623 Log likelihood 1355248 1351826 Notes Clusterrobust standard errors are in parentheses Standard errors for ancillary parameters are not reported Table 8 Spanish dairy farms correlation of inefficiency estimates Variables ss84 css90 ls93 kumb90 pl81 ss84 1000 css90 0868 1000 ls93 1000 0871 1000 kumb90 0938 0726 0936 1000 pl81 0931 0709 0929 0995 1000 7 Concluding remarks In this article we introduced the new Stata commands sfcross and sfpanel which implement an extensive array of SF models for crosssectional and panel data With respect to the available official Stata commands frontier and xtfrontier we add multiple features for estimating frontier parameters and for postestimating unit inefficiency and efficiency In the development of the commands we widely exploit Mata potentiality By using Mata structures we provide a very readable code that can be easily developed further by other Stata users We illustrated the commands estimation capabilities through simulated data focusing on some of the models that cannot be estimated using official Stata commands 756 Stochastic frontier using Stata Finally we illustrated the proposed routines using real datasets under different possible empirical scenarios short versus long panels cost versus production frontiers homoge neous versus heterogeneous outputs 8 Acknowledgments We are grateful to David Drukker and all participants at the 2009 Italian Stata Users Group meeting for useful comments We thank William Greene for valuable advice and discussions and for maintaining an excellent webpage and making available several databases two of which we have extracted and used in the empirical applications 9 References Ahn S C Y Hoon Lee and P Schmidt 2001 GMM estimation of linear panel data models with timevarying individual effects Journal of Econometrics 101 219255 Aigner D J C A K Lovell and P Schmidt 1977 Formulation and estimation of stochastic frontier production function models Journal of Econometrics 6 2137 Alvarez A and C Arias 2004 Technical efficiency and farm size A conditional analysis Agricultural Economics 30 241250 Battese G E and T J Coelli 1988 Prediction of firmlevel technical efficiencies with a generalized frontier production function and panel data Journal of Econometrics 38 387399 1992 Frontier production functions technical efficiency and panel data With application to paddy farmers in India Journal of Productivity Analysis 3 153169 1995 A model for technical inefficiency effects in a stochastic frontier production function for panel data Empirical Economics 20 325332 Belotti F and G Ilardi 2012 Consistent estimation of the true fixedeffects stochas tic frontier model CEIS Tor Vergata Research Paper Series Caudill S B and J M Ford 1993 Biases in frontier estimation due to heteroscedas ticity Economics Letters 41 1720 Caudill S B J M Ford and D M Gropper 1995 Frontier estimation and firm specific inefficiency measures in the presence of heteroscedasticity Journal of Business and Economic Statistics 13 105111 Chen YY P Schmidt and HJ Wang 2011 Consistent estimation of the fixed effects stochastic frontier model httpwwweconomicsuwocanewslettermisc2011schmidt nov11pdf Cornwell C P Schmidt and R C Sickles 1990 Production frontiers with cross sectional and timeseries variation in efficiency levels Journal of Econometrics 46 185200 F Belotti S Daidone G Ilardi and V Atella 757 Cuesta R A 2000 A production model with firmspecific temporal variation in tech nical inefficiency With application to Spanish dairy farms Journal of Productivity Analysis 13 139158 Drukker D M and R Gates 2006 Generating Halton sequences using Mata Stata Journal 6 214228 Farsi M M Filippini and W Greene 2005 Efficiency measurement in network indus tries Application to the Swiss railway companies Journal of Regulatory Economics 28 6990 Greene W 2005a Reconsidering heterogeneity in panel data estimators of the stochas tic frontier model Journal of Econometrics 126 269303 2005b Fixed and random effects in stochastic frontier models Journal of Productivity Analysis 23 732 Greene W H 1980a Maximum likelihood estimation of econometric frontier functions Journal of Econometrics 13 2756 1980b On the estimation of a flexible frontier production model Journal of Econometrics 13 101115 2003 Simulated likelihood estimation of the normalgamma stochastic frontier function Journal of Productivity Analysis 19 179190 2012 Econometric Analysis 7th ed Upper Saddle River NJ Prentice Hall Hadri K 1999 Estimation of a doubly heteroscedastic stochastic frontier cost function Journal of Business and Economic Statistics 17 359363 Han C L Orea and P Schmidt 2005 Estimation of a panel data model with para metric temporal variation in individual effects Journal of Econometrics 126 241267 Horrace W C and P Schmidt 1996 Confidence statements for efficiency estimates from stochastic frontier models Journal of Productivity Analysis 7 257282 Huang C J and JT Liu 1994 Estimation of a nonneutral stochastic frontier production function Journal of Productivity Analysis 5 171180 Jondrow J C A K Lovell I S Materov and P Schmidt 1982 On the estimation of technical inefficiency in the stochastic frontier production function model Journal of Econometrics 19 233238 Kumbhakar S C 1990 Production frontiers panel data and timevarying technical inefficiency Journal of Econometrics 46 201211 Kumbhakar S C S Ghosh and J T McGuckin 1991 A generalized production fron tier approach for estimating determinants of inefficiency in US dairy farms Journal of Business and Economic Statistics 9 279286 758 Stochastic frontier using Stata Kumbhakar S C and C A K Lovell 2000 Stochastic Frontier Analysis Cambridge Cambridge University Press Lancaster T 2002 The incidental parameters problem since 1948 Journal of Econo metrics 95 391414 Lee Y H and P Schmidt 1993 A production frontier model with flexible tempo ral variation in technical efficiency In The Measurement of Productive Efficiency Techniques and Applications ed H O Fried C A Knox Lovell and S S Schmidt 237255 New York Oxford University Press Meeusen W and J van den Broeck 1977 Efficiency estimation from CobbDouglas production functions with composed error International Economic Review 18 435 444 Neyman J and E Scott 1948 Consistent estimates based on partially consistent observations Econometrica 16 132 Pitt M M and LF Lee 1981 The measurement and sources of technical inefficiency in the Indonesian weaving industry Journal of Development Economics 9 4364 Ritter C and L Simar 1997 Pitfalls of normalgamma stochastic frontier models Journal of Productivity Analysis 8 167182 Schmidt P and R C Sickles 1984 Production frontiers and panel data Journal of Business and Economic Statistics 2 367374 Stevenson R E 1980 Likelihood functions for generalized stochastic frontier estima tion Journal of Econometrics 13 5766 Wang HJ 2002 Heteroscedasticity and nonmonotonic efficiency effects of a stochastic frontier model Journal of Productivity Analysis 18 241253 Wang HJ and P Schmidt 2002 Onestep and twostep estimation of the effects of exogenous variables on technical efficiency levels Journal of Productivity Analysis 18 129144 About the authors Federico Belotti is a researcher at the Centre for Economics and International Studies CEIS of the University of Rome Tor Vergata Silvio Daidone is a research associate at the Centre for Health Economics CHE of the Uni versity of York and an economist at the Agricultural Development Economics Division of the Food and Agriculture Organization of the United Nations Giuseppe Ilardi is a researcher at the Economic and Financial Statistics Department of the Bank of Italy Vincenzo Atella is an associate professor at the University of Rome Tor Vergata View publication stats CHAPTER 3 Theory of Bank Efficiency and Bank Competition 31 INTRODUCTION Efficiency is a commonly used measure in the production process There are inputs and outputs in production Higher efficiency mainly means any unitcompany has the ability to use the same amount of input to generate a higher volume of output whereas on the other hand higher efficiency can also be interpreted as any unitcompany has the ability to minimize the input to produce the same volume of output This interpretation of effi ciency purely considers the volumes of input and output whereas in the production process not only is the volume of output and input being considered by the managers and regulators but more attention has also been given to minimize cost as well as maximize profit and revenue Thus it is important to analyse efficiency in technical terms as well as evaluate efficiency from a cost profit and revenue perspective More specifically managers and regulators are greatly concerned about how to use a certain amount of input to minimize the cost and maximize the profit and revenue The chapter mainly explains the theories of bank efficiency as well as bank competition Specifically this chapter is organized as follows Section 32 discusses the framework of efficiency and is divided into four subsections which explain cost efficiency technical efficiency profit efficiency and revenue efficiency on a onebyone basis With regards to bank competition four indicators to measure bank competition will be explained which include Concentration ratio PanzareRosse Hstatistic Lerner index and Boone indicator 32 THE FRAMEWORK OF EFFICIENCY 321 Technical Efficiency A simple measure of firm efficiency is defined by Farrell 1957 whose work is derived from Debreu 1951 and Koopmans 1951 He argued that technical efficiency reflects a firms ability to obtain maximal output from a Efficiency and Competition in Chinese Banking ISBN 9780081000748 httpdxdoiorg101016B9780081000748000030 2016 Elsevier Ltd All rights reserved 45 given set of inputs Farrell explained his idea by making the assumption that firms use two inputs X1 and X2 to produce one output y and the production is under the assumption of Constant Return to Scale CRS see Fig 31 In other words an increase decrease in the inputs leads to the same proportional increase decrease of the output The unit isoquant SS0 describes the technological set to produce the certain amount of output using the combination of the inputs X1 and X2 In other words SS0 shows the minimum number of inputs needed to produce one unit of output All the production along this curve SS0 is supposed to be perfectly efficient whereas any other points above or located at the right of the curve such as the point P are regarded as inefficient production due to the fact that the number of inputs used in the production to produce one unit of output is more than the efficient production The distance QP represents the technical inefficiency of the firm It also represents the number of inputs which can be reduced without any in fluence on the output production In other words it represents the number of inputs which can be reduced without any decrease in the output The percentage of the input reductiontechnical inefficiency level for the point P can be represented by the ratio QPOP whereas the technical efficiency of a firm can be measured by the ratio OQOP The value of technical efficiency ranges from 0 to 1 The value of 1 means that the firm is fully technically efficient The condition for the assumption of constant return to scale CRS is that all the firms operate at the optimal scale which is impossible sometimes because of imperfect competition The variable return to scale VRS Q R S P A O X1Y X2Y A S Q Figure 31 The measurement of technical efficiency Coelli T 1996 A Guide to DEAP Version 21 A Data Envelopment Analysis Computer Program CEPA working paper 199608 46 Efficiency and Competition in Chinese Banking further decomposes the technical efficiency into pure technical efficiency PTE and scale efficiency SE In Fig 32 we make another assumption which is that firms use one input X to product one output Y The constant return to scale frontier is represented by CRS firms production can be laying on the curve or located to the right side the technical inefficiency for the firm P is PPc over AP whereas the technical efficiency can be repre sented by the ratio APcAP The SE considers the possibility that firms do not operate at the optimal size The CRS assumption is replaced by the VRS assumption to measure the SE The variable return to scale frontier is represented by VRS under which the technical inefficiency for firm P is PPv over AP the PTE for the firm P can be represented by the ratio APvAP whereas the SE can be represented by the ratio APcAPv If the firm operates under CRS the value of scale efficiency equals 1 whereas the value of SE will be less than 1 in a VRS situation In other words firms have scale inefficiency 322 Cost Efficiency Rather than focussing on manipulating the inputs and outputs in banking operations which is reflected by technical efficiency more practically bank managers and banking regulatory authorities emphasize minimizing the cost of banking operations one of the indicator used to measure this perfor mance is cost efficiency It is also called economic efficiency It can be Y CRS VRS NIRS A P Q Pv Pc O X Figure 32 The measurement of pure technical efficiency and scale efficiency CRS Constant Return to Scale NIRS nonincreasing returns to scale Coelli T 1996 A Guide to DEAP Version 21 A Data Envelopment Analysis Computer Program CEPA working paper 199608 Theory of Bank Efficiency and Bank Competition 47 represented by the ratio of minimal cost achieved by the bestpractice bank to the actual cost incurred by a specific bank The banks with higher cost efficiency normally have higher ability to choose their inputs and mixes according to their prices the purpose of which is to minimize the cost whereas there are two sources of cost inefficiency One is attributed to the technology in other words one part of cost inefficiency is derived from technical inefficiency whereas the other part of cost inefficiency is due to the suboptimal allocation of resources that is part of cost inefficiency is related to allocative inefficiency The overall cost efficiency can be calcu lated by multiplying the technical efficiency by allocative efficiency 323 Revenue Efficiency Berger and Mester 1997 argue that revenue efficiency measures the change in a banks revenue adjusted for a random error relative to the estimated revenue obtained from producing an output bundle as efficiently as the best practice bank Revenue efficiency emphasizes the fact that the banking operation aims to maximize revenue and revenue efficiency is measured by the ratio of actual revenue to maximum revenue There are two sources of revenue inefficiency More specifically revenue inefficiency can be attrib uted to outputoriented technical efficiency or outputoriented allocative inefficiency The technical inefficiency can be interpreted as producing too few outputs given the input quantities whereas the allocative inefficiency means nonoptimal combination of outputs are produced given their prices The following figure Fig 33 explains revenue efficiency in the frontier methodological framework We assume that the bank uses input X to produce two outputs y1 and y2 TT0 represents the production possibility curve Under the outputoriented assumption a bank can be regarded as technically efficient if the operation point is located on TT0 Thus we can say banks B C D and E are technically efficient under the outputoriented assumption Due to the fact that Bank A locates below the frontier it is regarded as inefficient In other words compared to revenueefficient banks the bank which is located at point A has the potential to increase the production levels of both outputs to point A0 located on the production possibility frontier Therefore for Bank A the distance AA0 represents the outputoriented technical inefficiency The outputoriented technical ef ficiency TE is measured by the ratio of actual output to maximum output of the bank Using the above diagram the outputoriented TE can be represented by the ratio of OAOA0 It is assumed that the isorevenue line is represented by PP0 Bank D is regarded as revenue efficient Because the 48 Efficiency and Competition in Chinese Banking Figure 33 Revenue efficiency Kumar S Gulati R 2013 Regulation and Efficiency of Indian Banks Springer India cost at point D equals the revenue at point F the outputoriented allocative efficiency can be measured by the ratio of OAOF If it is assumed that the production is located at the point A the increase in revenue can be represented by the distance AF Finally the revenue efficiency of the bank can be represented by the ratio of OAOF Due to the fact that revenue efficiency includes both outputoriented TE and allocative efficiency it is considered as a composite efficiency measure The revenue inefficiency for Bank A can be represented by the distance AF The relationship between revenue efficiency TE and allocative efficiency can be expressed as Revenue efficiency technical efficiency allocative efficiency 324 Profit Efficiency Profit efficiency is a more comprehensive type of efficiency concept which considers both cost and revenue performance The profit efficiency assumes that both revenue and cost are controlled by managers In empirical research the analysis of profit efficiency is preferred to cost efficiency due to the fact the operating revenues and loan losses are ignored by cost efficiency The profit efficiency can be defined as how close a bank is able to produce the maximum possible profit given the levels of input and output process and other exogenous conditions Under the given conditions profit efficiency can be improved if the banks profits move closer to the profit of a benchmark bank bestpractice bank with highest profit The profit efficiency can be measured by the ratio of actual profit over the maximum profit The score of profit efficiency is bounded above and has a maximum value of 1 In other words the value of profit efficiency ranges from minus infinity to 1 The value of profit efficiency for the bestpractice bank is 1 Negative values of profit efficiency are possible due to the fact that banks can throw away more than 100 of the potential profits The following figure is used to better illustrate the concept of profit efficiency In the following figure Fig 34 the production frontier is represented by the curve OQ The actual inputoutput combination of bank A which is represented by point A is xAyA Thus Bank A profit can be calculated as qAyA pAxA All the production points located on the line CD yield the same amount of profit as the point A The objective of bank A is to reach highest isoprofit line parallel to CD that can be attained at any point on or below the curve OQ Point B will be the highest point which is also the tangency of EF and OQ If it is assumed that the inputoutput combination for point B is x and y line OE represents the maximum profit achieved and Bank A is able to achieve the maximum profit as long as it reaches one point on Line EF Say point A The calculation of maximum profit can be expressed as qAyA pAxA this also equals qByB pBxB The profit efficiency of bank A can be calculated by the ratio of actual profit divided by the maximum profit Figure 34 Profit efficiency Kumar S Gulati R 2013 Regulation and Efficiency of Indian Banks Springer India 33 THE MEASUREMENT OF EFFICIENCY The methods used to measure efficiency can be mainly divided into two streams one is called the parametric approach and the other one is called the nonparametric approach The former can be more specifically divided into three analyses which include stochastic frontier analysis SFA distri bution free approach DFA and think frontier approach TFA The non parametric approach can be further divided into two analyses which are data envelopment analysis DEA and free disposal hull FDH analysis With regards to the parametric approach a prespecified functional form for the bestpractice frontier is required The efficiency of a firm is compared to the benchmark firm firm with best practice A firm is inefficient if the cost is higher than the best practice from or the profit is lower than the benchmark firm The nonparametric approach is a mathematical programming technique to estimate efficiency and Stephards distance function methodology is implemented The main argument for the DEA over a parametric technique such as SFA lies to the fact that it works particularly well with small samples Furthermore it is able to handle multiple inputs and outputs stated in different measurement units and does not necessitate knowledge of any functional form of the frontier see Charnes et al 1995 Most empirical papers show that using DEA to estimate the efficiency frontier can yield robust results see Seiford and Thrall 1990 DEA has a number of draw backs first the efficiency scores derived from DEA are sensitive to the selection of inputs and outputs Secondly the number of efficiency firms on the frontier tends to increase with the number of input and output vari ables Thirdly DEA assumes that it has no statistical noise and it is sensitive to extreme observations and measurement errors This section will be arranged as follows Section 331 will discuss the parametric approach As outlined previously three specific analyses are focussed upon which include SFA DFA as well as TFA In Section 332 two nonparametric approaches will be explained which are DEA and FDH 331 Parametric Approach 3311 Stochastic Frontier Approach One of the most widely used parametric technique to measure bank effi ciency is SFA which was proposed by Aigner et al 1997 Battese and Corra 1997 and Meeusen and Van Den Broeck 1997 The assumption of this method is that the decisionmaking unit DMU does not totally Theory of Bank Efficiency and Bank Competition 51 control the deviation from the frontier Random errors which are linked together with the choice of the functional form are allowed by SFA The SFA model is also called the composed error model in which two terms are included and considered the statistical noise term follows a symmetric distribution whereas another term which represents inefficiency follows a onesided distribution To make clearer with regards to the formation of the SFA the following stochastic cost function model is used ln Ci ¼ ln Cðyi wiÞ þ εi ¼ ln Cðyi wiÞ þ vi þ mi In the above function C represents the total cost i represents the specific bank y stands for a vector of output and w is the input price In the banking operation normally three input prices are considered which are price of funds price of labour and price of capital n represents the statistical noise and u stands for the nonnegative disturbance term mi measures the distance of a specific individual firm from the efficient cost frontier and this term represents the technical and allocative efficiency of the firm The deterministic cost frontier is represented by Cyiwi Aigner et al 1977 and Mester 1993 argue that vi follows the normal distribution and mi follows the halfnormal or exponential distribution Berger 1993 argues that the halfnormal or exponential distribution term on inefficiency assumes that most firms nearly reach the full efficiency and are clustered together The possibility of higher inefficiency is relatively low However this is not always the case due to the fact that some specific DMU would experience a higher degree of inefficiency Rather than using the halfnormal distribution for Ui Stevenson 1980 and Berger and DeYoung 1997 suggested that alternative distribution for inefficiency such as a truncated normal distribution is more appropriate Besides the truncated normal distribution the normalgamma distribution is considered by Greene 1990 One of the obvious disadvantages of the truncated normal distribution as well as normalgamma distribution lies in the fact that the inefficiency is difficult to separate from random error Berger and Humphrey 1997 argue that random error of these two distributions is close to a symmetric normal distribution There are mainly two methods being used to estimate the parameters of the frontier model and the error terms which are the maximum likelihood estimation or the corrected ordinary least square OLS see Battese and Coelli 1995 Coelli 1995 Fried et al 1993 In comparison the maximum 52 Efficiency and Competition in Chinese Banking likelihood estimation is preferred to the corrected OLS estimator due to the fact the maximum likelihood estimation produces more accurate results in large samples Coelli 1995 Olesen et al 1980 Jondrow et al 1982 argue that the distribution of the inefficiency term conditional on the estimate of the two error terms will be used to obtain estimates of inefficiency 3312 Distribution Free Approach DFA another parametric method was proposed by Schmidt and Sickles 1984 and Berger 1993 It is a panel estimation method The distributional assumptions on the error component were not imposed by this parametric method The similarity between DFA and SFA lies in the fact that the former specifies a functional form for the efficiency frontier The main difference between these two parametric methods is that the inefficiencies from the residual are separated in a different way DFA assumes that inefficiencies are relatively stable and should persist over time which disentangles them from random errors The random errors in DFA have the following characteristics 1 they are momentary and 2 the random errors tend to cancel one another through averaging DFA estimates the cost or profit function for each period of a panel data set There are two parts included in the residual which are inefficiency terms and random error term Due to the assumption that the error term tends to average out over time the inefficiency of the bank can be estimated by the average of a banks residuals Due to the fact that the distribution of inefficiency or error terms do not have any restrictive assumptions so compared to the SFA the DFA is easier to implement The reason lies in the fact that the maximum likelihood methods are not required by DFA to estimate the cost or profit function There are mainly two methods used to estimate the cost or profit function which is generalized least square GLS reflected by the work of Schmidt and Sickles 1984 The second method is OLS which is represented by the work of Berger 1993 The difference between average residual and average residual of the firm on the frontier gives the estimate of inefficiency There are some disadvantages to DFA Misleading results can be obtained if the period under examination is too long the results will also be biased if the inefficiency component of the error term is not constant The accuracy of the results will also be affected if the random error term is averaged out by an insufficient number of available data years The appropriate number of years enough to address all the above issues is six as argued by DeYoung 1997 Theory of Bank Efficiency and Bank Competition 53 3313 Thick Frontier Approach The last parametric method used to measure efficiency is the TFA which was developed by Berger and Humphrey 1991 1992 Similar to other parametric frontier approaches a functional form is specified for the frontier cost function in TFA The main difference between this method and other parametric methods lies to the fact that a thick frontier is estimated by this approach whereas with regards to other parametric techniques a frontier edge is used to measure the efficiency Furthermore the distributional assumption for crosssectional data is also avoided by this method This method is implemented by estimating the cost function Both the lowest averagecost quartile and the Highest average cost quartile of the firm are estimated Two groups of efficiencies are represented by firms in these two different quartiles More specifically the assumption is that firms in the lowest averagecost quartile have efficiency greater than the average level and they form the thick frontier On the other hand firms in the highest averagecost quartile have efficiency lower than the average level The difference between the two groups of firms with higher than average ef ficiency and lower than average efficiency represent the inefficiency as well as the exogenous difference in the regression Random error and luck are assumed to be represented by the error term within each of the frontiers The difference in the parameters between the upper cost frontier and lower cost frontier is used by TFA to measure cost efficiency whereas with regards to the exogenous factors they are estimated by calculating the difference between the lowest cost unction and highest cost function One disadvantage of TFA lies in the fact that the overall level of efficiency rather than the efficiency of the individual DMU is estimated TFA may not generate a robust estimate of the overall level of efficiency because the assumption made by this approach does not usually hold 332 NonParametric Approach 3321 Data Envelopment Analysis The idea of the nonparametric efficiency approach was first originated by Farrell 1957 in his seminal paper One of the disadvantages of his work lies to the fact that singleoutput case is considered whereas it cannot applied to large dataset with multiple inputs and outputs His work has been signifi cantly extended by Charnes et al 1978 in which a mathematical pro gramming algorithm was proposed namely DEA This method was used to measure the efficiency of a set of DMUs from which a significant 54 Efficiency and Competition in Chinese Banking improvement of their work compared to Farrell 1957 lies in the fact that this method can be applied to multiple inputs and outputs and until now many research studies use this method to measure efficiency in particular in the banking sector Two different DEA models are available with regards to the input or output orientation Inputoriented DEA aims to minimize the amount of input to produce a certain level of output On the other hand output oriented DEA focusses on keeping the same amount of input to maxi mize the output production With regards to the choice of orientation it depends over which aspect the managers have more control More spe cifically Coelli 1998 argued that inputoriented is more appropriate if producers can freely adjust the input usage and are required to meet the market demand whereas if only a certain amount of resources is available for the firm to maximize the output production the outputoriented model would be more appropriate Coelli and Perelman 1999 suggest that there is very minor impact of the choice of input or outputoriented model on the estimation of the efficiency score 3322 Free Disposal Hull One nonparametric technique used to measure efficiency is called FDH analysis The FDH model was firstly proposed by Deprins 1984 and then the model was extended by Tulkens 1993 The FDH analysis relaxes the convexity assumption of basic DEA models Both DEA and FDH are consistent estimators if the true production set is convex However due to the fact that FDH requires fewer assumptions thus a lower rate of conver gency is shown by FDH compared to DEA On the other hand if the true production set is not convex FDH has the ability to estimate the efficiency in a more precise way in comparison to DEA Because Both DEA and FDH are nonparametric methods compared to the parametric method such as SFA no particular functional form for the boundary is assumed and the measurement error is ignored by the nonparametric method The boundary of a reconstructed production possibility subset is regarded as the bestpractice technology which can be represented by enveloping a set of observations Berger and Humphrey 1997 argue that the set of production possibilities comprises only the DEA vertices and the FDH points interior to these vertices One of the assumptions made by the DEA is that linear substitution between observed input combinations on an isoquant is possible while the observations in Theory of Bank Efficiency and Bank Competition 55 piecewise linear form generate the isoquant In comparison one of the assumptions made by the FDH is that linear substitution between observed input combinations is impossible Therefore the isoquant is more like a step function formed by the intersection of lines drawn from observed Leontieftype input combinations Thus compared to DEA frontier the data are enveloped more tightly in the FDH which has a staircase shape The relationship between DEA and FDH frontiers can be described as two possible situations first the FDH frontier could be congruent with the DEA frontier The second situation is that the FDH frontier could be interior to the DEA frontier Tulkens 1993 argues that larger estimates of average efficiency are typically generated by FDH compared to DEA 34 THEORY OF BANK COMPETITION Similar to other industries the types of competition in the banking sector can be mainly classified into the following categories perfect competition monopolistic competition oligopoly and monopoly The perfect competitive environment in the banking sector has the following characteristics see Mankiw and Taylor 2006 1 there are a large number of banks in the market 2 banks offer a homogenous product with regards to the cost and attribute of the product 3 the cost for new banks to enter the market is very low In a perfectly competitive envi ronment banks are pricetakers rather than pricemakers The price of a product offered by banks will be determined by the industry supply or demand whereas they have no ability to influence the volumes of demand and supply in the market Another type of competition is called monop olistic competition Under this competitive environment lots of banks are in the market but unlike perfectly competitive markets these banks offer differentiated products to customers The cost of entry to and exit from the banking market is low Banks have a degree of control over the price of the product offered In other words they are price makers rather than price takers to some extent The third type of competition is called oligopoly Under this competitive environment there are a small number of banks in the markets whereas all of them provide either homogenous or hetero geneous products in the market the entry or exit to the market is quite expensive The banks operating in the oligopoly market have power to set the price in the market The degree of price control by banks in oligopoly is higher than in monopolistic competition One special characteristic of 56 Efficiency and Competition in Chinese Banking oligopoly over monopolistic competition and perfect competition is the interdependence among banking firms The market in the condition of oligopoly is made up of a few large banks because the size of the banks is very large their actions will affect the market conditions due to the fact that in the oligopoly market few banks are operating and each bank is large enough that its actions will affect market conditions Thus other banks will be aware of one banks action and respond appropriately to keep their competitive position in the market Finally the market in the con dition of monopoly has only one bank because this bank is the only firm that provides financial services to the market there is no competition and this bank has absolute power in setting the price in the market Further more other potential banks are unable to enter the market They also have the ability to charge a different price to different markets ie they may charge a lower price in a very elastic market to increase the quantity sold whereas a higher price would be charged to the consumers in a relatively inelastic market to maximize the profit 341 StructureeConductePerformance SCP Theory The SCP paradigm states that market structure would determine firm conduct which would determine performance Market structure can be measured by a number of factors such as the number of competitors in an industry the heterogeneity of product and the cost of entry and exit Conduct refers to a number of specific actions taken by a firm which include price taking product differentiation tacit collusion and exploitation of market power The performance of the firm can be measured from a number of indicators such as productive efficiency allocative efficiency and profitability The range of options and constraints facing a firm is defined by the attributes of the industry within which a firm operates In some industries with higher competition very few options are available to the firms and the firms have lots of constraints Firms in these industries generate maximized social welfare and in the long run the returns earned by the firms can only cover the cost of capital In summary the industry structure determines the firms conduct and longrun firm performance On the other hand the firms operating in a lower competitive industry environment have a greater range of conduct options and the number of constraints faced by the firms is limited Firms can make use of the available options to obtain the competitive advantage For instance the firms in these industries can use market power to set prices that generate significant economic value Theory of Bank Efficiency and Bank Competition 57 Structure Number of competitors Heterogeneity of product Cost of entry and exit Conduct Price taking Product differentiation Tacit collusion Exploitation of market power Performance Productive efficiency Allocative efficiency Profitability Figure 35 The structureconductperformance SCP model Own illustration However the sustainability of their advantages is determined by one of the attributes of industrystructure barriers to entry If there are no barriers to entry the competitive advantages of the firms in the industry will disappear when new competitors enter the market Therefore industry structure has an important effect on firm conduct and performance even though firms in these industries can sometimes have competitive advantages Barney and Clark 2007 The SCP model is summarized in Fig 35 342 Lerner Index The Lerner index was developed in 1934 by the American economist Abba Lerner The Lerner index is defined as the difference between price and marginal cost divided by price and it can be specified as LIit Pit MCitPit in which P is the price of banking outputs MC represents the margin cost whereas i and t represent the specific bank at specific year The value of Lerner index ranges from a minimum of zero to a maximum of one When P MC the Lerner index is zero which indicates that the firm has no pricing power As the value of Lerner index increases the difference between price and marginal cost becomes bigger which indicates that banks have higher market power Ariss 2010 In other words LI 0 indicates that there is perfect competition whereas LI 1 means the market is in a condition of monopoly Casu and Girardone 2009 argue that the Lerner index of monopoly power measures the degree of market power very well and is a good and widely used indicator in measuring competition in banking literature It represents the extent to which banks have the market power to set their price above the marginal cost Berger et al 2009 A similar opinion is held by DemirgucKunt and Peria 2010 who suggest that computing direct measures of market power is an alternative way to examine the competitive condition in the banking industry and the Lerner index defined as the difference between price and marginal cost relative to price is frequently used in the banking sector Lerner 1934 argues that market power is determined by demand elasticity and the Lerner index provides a number which links with the demandeprice elasticity inverse relationship Rojas 2011 suggests that the Lerner index is popular due to the fact that it shows the firms market power location between perfect competition and mo nopoly it also reflects the role that demand elasticity plays in determining a firms markup Compared to the traditional concentration ratio the Lerner index provides a more accurate measurement of bank market power The advantage of the Lerner index over the PanzareRosse Hstatistic which will be discussed later lies in the fact that the former is not a longrun equilibrium measure of competition and can be calculated at each point in time DemirgucKunt and Peria 2010 There are also some arguments relating to the disadvantages of the Lerner index Fernandez de Guevara et al 2005 argue that there are several problems with regards to the estimation of the Lerner index First the value of the Lerner index changes according to different revenues used by the study Frequently only interest revenue and costs are considered whereas other noninterest revenues and expenses are omitted The consideration of traditional loanedeposit services as the revenue ignores the banking activities of providing other services which have grown substan tially during recent years This will lead to an inaccurate result regarding the competitive condition in the banking sector Second the cost of risk which is very important in the profit and loss account of a banking system is not considered in general practice The ignorance of the cost of risk can be attributed to reasons such as insufficient data and calculation difficulties If the cost of risk in not included in the estimation of cost function it will lead to a wrong interpretation of the Lerner index because the margin is over estimated Third Bikker et al 2007 argue that the weakness of the Lerner index is attributed to the fact that the prices and costs required to calculate the index are not clearly identified by the available bank balancesheet data so that prices and costs can be proxied by many debatable choices 343 Panzar and Rosse 1987 HStatistic An empirical test is developed by John C Panzar and James N Rosse to discriminate between monopoly oligopolistic monopolistically competi tive and perfectly competitive markets They have provided a concise in dicator the socalled Hstatistic which is based on the static properties of reducedform revenue equation The Hstatistic can be interpreted as a continuous and increasing measure of the overall level of competition in a Theory of Bank Efficiency and Bank Competition 59 specific market This method makes the assumption that different pricing strategies will be employed in reaction to changes in input costs In other words market power is measured by the extent to which changes in factor prices unit price of funds capital and labour are reflected in revenue The Hstatistic ranges from minus infinity to unity see Table 31 A negative H arises when the competitive structure is a monopoly a perfect colluding oligopoly or a conjectural variation of shortrun oligopoly In all the cases an increase in input prices will translate into higher marginal costs a reduction of equilibrium output and subsequently a fall in total revenue Vesala 1995 If H lies between zero and unity the market structure is characterized by monopolistic competition Under monopolistic competition potential entry leads to contestable market equilibrium and revenue increases less than proportionally to the changes in input prices as the demand facing individual banks for banking products is inelastic Triole 1987 Under perfect competition the Hstatistic equals unity In this particular situation a proportional increase in factor input prices raises both marginal and average costs and induces an equiproportional change in revenues without distorting the optimal output of any firm The Hstatistic would also equal to one if a contestable market exists Baumol et al 1982 who put forward the contestability theory argue that under very restrictive circumstances such as free entry to and exit from the market and higher price elastic demand for industrys output the competition in a highly concentrated market still exists Due to the existence of these features larger firms in the market need to take competitive measures to price their outputs One assumption of PR Hstatistic is that the test can only be applied to firms which produce a single output Therefore banks are regarded as providers of traditional loandeposit services as well as nontraditional Table 31 Discriminatory power of Hstatistic Estimated value of H Competitive environment Market equilibrium H 0 Monopolistic market behaviour Conjectural variation shortrun oligopoly H 0 Equilibrium H 0 Disequilibrium 0 H 1 Monopolistic competition Natural monopoly in a perfectly contestable market Perfect competition H 1 activities using factor inputs such as labour funds and capital In other words all the activities should be considered to estimate the competitive condition of the specific market Nevertheless supposing the above assumption is omitted the competitive conditions of separate segments cannot be analysed due to the fact that there are no detailed data available for the estimation of the reducedform revenue equations The advantages of using PR Hstatistic to measure the competition are 1 It works well with firmspecific data on revenues and factor prices 2 it does not require information about equilibrium output prices and quantities for the firm andor industry 3 this method can generate concise competitive condition if the sample size is small Matthews et al 2007 Due to the fact that Hstatistic was developed on the basis of a static model there are no predictions on the Hvalue which is one of the weaknesses of this test In addition the overall market equilibrium required by the test cannot be fulfilled because of market entry and exit which leads to further limits on the interpretation of such analysis Due to the fact that the Hstatistic is a static approach one characteristic of the implementation is that the test needs to be undertaken on obser vations that are in longrun equilibrium In other words in the equilibrium test the dependent variable will be replaced by the profitability measure such as return on assets ROA or return on equity ROE rather than the revenues In equilibrium the resulting Hstatistic will be significantly equal to zero However the significant and negative Hstatistic indicates that the test is not in longrun equilibrium The purpose of the test is to justify on the ground that competitive markets will equalize riskadjusted return across firms such that in equilibrium rate of return should not be correlated with factor input prices 344 Boone Indicator Nowadays a growing number of studies in the literature use the Boone 2008 indicator to measure bankingsector competition This indicator is based on the efficiency hypothesis which was proposed by Demsetz 1973 The efficiency hypothesis argues that the growth of efficient firms leads to improvement in industry performance In other words the hypothesis emphasizes the relationship between efficiency and performance This indicator is based on the assumption that efficiency leads to improvement in performance This hypothesis mainly argues that banks with higher effi ciency have higher levels of performance in the form of higher profit and then have higher ability to attract a larger number of market shares The Theory of Bank Efficiency and Bank Competition 61 theoryassumption of the Boone indicator is in line with the studies by Stiorh 2000 and Stiroh and Straham 2003 who suggest that a large proportion of assets will be transformed from banks with lower profit to the ones with higher profit due to the increase in competition The basic idea under the Boone indicator argues that stronger competition leads to the reallocation of assets A number of reasons explain the increase in competition such as increase in the number of banks providing banking services or a higher degree of aggressive interaction between banks under the condition that individual banks have a certain level of efficiency The industry with a higher degree of competition will make the profit of the higherefficiency bank increase at the expense of banks with lower levels of efficiency Schaeck and Cihak 2010 argue that there are a number of advantages of the Boone indicator compared to other competition indicators such as Panzar and Rosse 1987 HStatistic and the Lerner index More specifically compared to the PR Hstatistic which requires one to identify a situation from the statistic the Boone indicator is monotonically related to competition Xu et al 2013 argue that the Boone indicator will provide robust results under binding interestrate regulations Finally the Boone indicator has an additional advantage for studies on developing countries This indicator only requires information on profit and cost and the computation of the Boone indicator does not require information on prices Leon 2014 A number of studies use the Boone indicator to measure competition in banking markets van Leuvensteijn et al 2007 Maslovych 2009 Schaeck and Cihak 2010 among others However Leon 2015 argues that the Boone indicator is still a young tool and therefore has not been thoroughly scrutinized by the literature 35 SUMMARY AND CONCLUSION Efficiency as one important measure of performance has received greater attention from government companies as well as academic researchers Both the government and company management try to use different ways to improve efficiency and further promote performance Efficiency can be measured from different perspectives More specifically technical efficiency mainly focusses on volume of production investment and output In other words technical efficiency mainly concentrates on using a certain amount of input to maximize output production or producing a certain volume of 62 Efficiency and Competition in Chinese Banking output using fewer numbers of inputs In firm operation in daily life not only does the volume of investment and output concern the managers but more importantly is how efficiently the firm controls the cost In other words how to use a certain amount of inputs to minimize cost is the goal which needs to be achieved within the firm It is well known that con trolling cost as efficiently as possible is not the final step for the company the company must aim to maximize the revenue and more importantly the profit Thus how efficiently a company uses input resources to generate profit attracts the greatest attention from the public This chapter explains the theories from technical cost revenue and profit perspectives thus systematically discussing the theories of technical cost revenueand profit efficiencies Not only the theories of efficiency being discussed this chapter also explains two mainly streams of methods used to measure efficiency which are parametric method as well as nonparametric method To be more specific three parametric approaches are discussed in the chapter which include SFA DFA as well as TFA on the other hand there are mainly two nonparametric approaches being discussed in the chapter which are DEA and FDH Not only are specific methods explained in this chapter but also the different methods are compared After the discussion about the theory and measures of efficiency this chapter systematically explains the theories of bank competition Starting from the traditional structureeconducteperformance hypothesis which mainly uses the concentration ratio or HerfindaleHirschman index as a competition indicator After the discussion of the concentration ratio the chapter further discusses the Lerner index competition indicator The ad vantages of this indicator lie in the fact that it not only measures competitive conditions but more importantly it reflects the market power of banks Then the chapter discusses the PanzarRosse H statistic Finally the chapter discusses the Boone competition indicator One characteristic of Boone compared to the other competition indicators lies in the fact that this in dicator can measure the competition of different banking markets ie loan market deposit market and noninterest income market This indicator is the newest competition indicator used in academic research today Through the systematic explanation and discussion of the theories of efficiency and competition not only can we have a clear and better understanding of efficiency and competition but provide a basis for our analysis and discussion in future chapters Theory of Bank Efficiency and Bank Competition 63 REFERENCES Aigner D Lovell CAK Schmidt P 1977 Formation and estimation of stochastic frontier production function models Journal of Econometrics 6 1 21e37 Ariss RT 2010 On the implications of market power in banking evidence from developing countries Journal of Banking and Finance 34 765e775 Barney JB Clark DN 2007 ResourceBased Theory Creating and Sustaining Competitive Advantage Oxford University Press Oxford Battese GE Coelli TJ 1995 A model for technical inefficiency effects in a stochastic frontiers production function for panel data Empirical Economics 20 2 325e332 Battese GE Corra GS 1977 Estimation of a production frontier model with appli cations to the pastoral zone of eastern Australia Australian Journal of Agricultural Economics 21 3 169e179 Baumol WJ Panzar JC Willig RD 1982 Contestable Markets and the Theory of Industry Structure Harcourt Brace Jovanovich San Diego CA Berger AN DeYoung R 1977 Problem loans and cost efficiency in commercial banks Journal of Banking and Finance 21 6 849e870 Berger AN Humphrey DB 1991 The dominance of inefficiencies over scale and product mix economies in banking Journal of Monetary Economics 28 1 117e148 Berger AN Humphrey DB 1992 Measurement and efficiency issues in commercial banking In Griliches Z Ed Output Measurement in the Service Sectors University of Chicago Press Chicago and London pp 245e279 Berger AN Humphrey DB 1997 Efficiency of financial institutions international survey and directions for future research European Journal of Operational Research 98 175e212 Berger AN Klapper LF TurkAriss R 2009 Bank competition and financial stability Journal of Financial Services Research 35 99e118 Berger AN Mester LJ 1997 Inside the black box what explains differences in the efficiencies of financial institutions Journal of Banking and Finance 21 895e947 Berger AN 1993 Distribution free estimates of efficiency of the US banking industry and tests of the standard distributional assumptions Journal of Productivity Analysis 4 3 261e292 Bikker JA Spierdijk L Finnie P 2007 The Impact of Market Structure Contestability and Institutional Environment on Bank Competition DNB working paper No 156 Boone J 2008 A new way to measure competition Economic Journal 118 1245e1261 Casu B Girardone C 2009 Testing the relationship between competition and efficiency in banking a panel data analysis Economics Letters 105 134e137 Charnes A Cooper WW Rhodes E 1978 Measuring the efficiency of decision making units European Journal of Operational Research 2 6 429e444 Charnes A Cooper WW Lewin AY Seiford LM 1995 Data Envelopment Anal ysis Theory Methodology and Applications Kluwer Boston Coelli T Perelman S 1999 A comparison of parametric and nonparametric distance functions with application to European Railways European Journal of Operational Research 117 2 326e339 Coelli TJ 1995 Estimators and hypothesis tests for a stochastic frontier function a Monte Carlo analysis Journal of Productivity Analysis 6 3 247e268 Coelli T 1996 A Guide to DEAP Version 21 A Data Envelopment Analysis Computer Program CEPA working paper 199608 Coelli T 1998 A multistage methodology for the solution of oriented DEA models Operations Research Letters 23 3e5 143e149 Debreu G 1951 The coefficient of resource utilization Econometrica 19 273e292 64 Efficiency and Competition in Chinese Banking DemirgucKunt A Peria MSM 2010 A Framework for Analysing Competition in the Banking Sector An Application to the Case of Jordan Policy Research working paper No 5499 1e24 Demsetz H 1973 Industry structure market rivalry and public policy Journal of Law and Economics 16 1e9 Deprins D Simar L Tulkens H 1984 Measuring labour efficiency in post offices In Marchand M Pestieau P Tulkens H Eds The Performance of Public Enterprises Concepts and Measurement NorthHolland Amsterdam DeYoung R 1997 A diagnostic test for the distribution fee efficiency estimator an example using US commercial bank data European Journal of Operational Research 98 2 243e249 Farrell MJ 1957 The measurement of productive efficiency Journal of the Royal Statistical Society 120 253e290 Fernandez de Guevara J Maudos J Perez F 2005 Market power in European banking Journal of Financial Services Research 27 109e137 Fried HO Lovell CAK Schmidt SS 1993 The Measurement of Productive Efficiency Techniques and Application Oxford University Press New York Greene WH 1990 A gammadistributional stochastic frontier model Journal of Econometrics 46 1e2 141e163 Jondrow J Knox Lovell CA Materor IS Schmidt P 1982 On the estimation of technical inefficiency in the stochastic frontier production function model Journal of Econometrics 19 2e3 233e238 Koopmans TC 1951 An analysis of production as an efficient combination of activities In Koopmans TC Ed Activity Analysis of Production and Allocation Wiley New York Kumar S Gulati R 2013 Regulation and Efficiency of Indian Banks Springer India Leon F 2014 Measuring competition in banking a critical review of methods Series etudes et documents du CERDI 12 1e14 Leon F 2015 Measuring Competition in Banking A Critical Review of Methods Working paper Halshs No 01015794 Lerner AP 1934 The concept of monopoly and the measurement of monopoly power The Review of Economic Studies 1 157e175 Mankiw NG Taylor M 2006 Economics Cengage Learning EMEA London Maslovych M 2009 The Boone Indicator as a Measure of Competition in the Banking Sector The Case of Ukraine Kier School of Economics MA thesis Matthews K Murinde V Zhao T 2007 Competitive conditions among the major British Banks Journal of Banking and Finance 31 2025e2042 Meeusen W Van Den Broeck J 1977 Efficiency estimation from CobbDouglas pro duction functions with composed error International Economic Review 18 2 435e444 Mester LJ 1993 Efficiency in the savings and loan industry Journal of Banking and Finance 17 2e3 267e286 Olesen JA Schmidt P Waldman DM 1980 A monte carlo study of estimators of stochastic frontier production function Journal of Econometrics 13 1 67e82 Panzar JC Rosse JN 1987 Testing for monopoly equilibrium Journal of Industrial Economics 35 443e456 Rojas C 2011 Market power and the lerner index a classroom experiment Journal of Industrial Organization Education 5 1e19 Schaeck K Cihak M 2010 Competition Efficiency and Soundness in Banking An Industrial Organization Perspective European Banking Center discussion paper No 2010e20 Schimidt P Sickles RC 1984 Production frontiers and panel data Journal of Business and Economic Statistics 2 4 367e374 Theory of Bank Efficiency and Bank Competition 65 Seiford LM Thrall RM 1990 Recent development in DEA the mathematical programming approach to frontier analysis Journal of Econometrics 46 7e38 Stevenson RE 1980 Likelihood functions for generalised stochastic frontier estimation Journal of Econometrics 13 1 57e66 Stiroh KJ Strahan PE 2003 Competitive dynamics of deregulation evidence from US banking Journal of Money Credit and Banking 35 801e828 Stiroh KJ 2000 Compositional Dynamics and the Performance of the US Banking Industry Federal Reserve Bank of New York staff report No 98 Triole J 1987 The Theory of Industrial Organization MIT press Cambridge Tulkens H 1993 On FDH efficiency analsysis some methodological issues and applica tions to retail banking courts and urban transit Journal of Productivity Analysis 4 183e210 Van Leuvensteijn M Bikker JA van Rixtel AARJM Kok Sorensen C 2007 A New Approach to Measuring Competition in the Loan Markets of the Euro Area European Central Bank working paper No 768 Vesala J 1995 Testing for Competition in Banking Behavioural Evidence From Finland Bank of Finland Studies p E1 Xu B van Rixtel A van Leuvensteijn M 2013 Measuring Bank Competition in China A Comparison of New Versus Conventional Approaches Applied to Loan Markets BIS working paper 422 pp 1e68 66 Efficiency and Competition in Chinese Banking Universidade do Estado do Rio de Janeiro Centro de Ciências Sociais Faculdade de Ciências Econômicas Giovanni Barboza Xavier Avaliação de desempenho no setor bancário uma comparação entre bancos digitais e tradicionais via Análise de Fronteira de Eficiência e Índice de Malmquist Rio de Janeiro 2023 Giovanni Barboza Xavier Avaliação de desempenho no setor bancário uma comparação entre bancos digitais e tradicionais via Análise de Fronteira de Eficiência e Índice de Malmquist Dissertação apresentada como requisito parcial para obtenção do título de Mestre ao Programa de PósGraduação em Ciências Econômicas da Universidade do Estado do Rio de Janeiro Área de concentração II Economia Aplicada Orientador Prof Dr Alexandre Marinho Coorientador Prof Dr Luiz Fernando Rodrigues de Paula Rio de Janeiro 2023 CATALOGAÇÃO NA FONTE UERJREDE SIRIUSBIBLIOTECA CCSB Bibliotecário Fabiano Salgueiro CRB76974 Autorizo apenas para fins acadêmicos e científicos a reprodução total ou parcial desta dissertação desde que citada a fonte Assinatura Data X3 Xavier Giovanni Barboza Avaliação de desempenho no setor bancário uma comparação entre bancos digitais e tradicionais via Análise de Fronteira de Eficiência e Índice de Malmquist Giovanni Barboza Xavier 2023 73 f Orientador Prof Dr Alexandre Marinho Coorientador Prof Dr Luiz Fernando Rodrigues de Paula Dissertação mestrado Universidade do Estado do Rio de Janeiro Faculdade de Ciências Econômicas Bibliografia f 6873 1 Análise de envoltório de dados Teses 2 Bancos Brasil Teses I Marinho Alexandre II Paula Luiz Fernando Rodrigues de III Universidade do Estado do Rio de Janeiro Faculdade de Ciências Econômicas IV Título CDU 33681005 Giovanni Barboza Xavier Avaliação de desempenho no setor bancário uma comparação entre bancos digitais e tradicionais via Análise de Fronteira de Eficiência e Índice de Malmquist Dissertação apresentada como requisito parcial para obtenção do título de Mestre ao Programa de PósGraduação em Ciências Econômicas da Universidade do Estado do Rio de Janeiro Área de concentração II Economia Aplicada Aprovada em 10 de março de 2023 Banca Examinadora Prof Dr Alexandre Marinho Orientador FCEUERJ Prof Dr Luiz Fernando Rodrigues de Paula Coorientador IEUFRJ e FCEUERJ Prof Dr Pedro James Frias Hemsley IEUFRJ Rio de Janeiro 2023 AGRADECIMENTOS Aos meus pais Janice e Carlos por sempre estarem ao meu lado me apoiando nos momentos difíceis Ao meu orientador Alexandre Marinho e ao meu coorientador Luiz Fernando de Paula pelos ensinamentos apoio e colaboração que me permitiram estar concluindo este trabalho Ao meu amigo Renan pela parceria durante todo o curso e aos meus amigos Natalia Henry e Catharine por todo companheirismo ao longo do período em que me dediquei a este trabalho Aos meus professores e colegas do Programa de PósGraduação em Ciências Econômicas PPGCEUERJ com quem convivi ao longo desses anos de curso pelo aprendizado e pelas experiências que certamente enriqueceram a minha formação acadêmica Agradeço à Coordenação de Aperfeiçoamento de Pessoal de Nível Superior Brasil CAPES Código de Financiamento 001 pelo apoio financeiro para o presente trabalho RESUMO XAVIER Giovanni Barboza Avaliação de desempenho no setor bancário uma comparação entre bancos digitais e tradicionais via Análise de Fronteira de Eficiência e Índice de Malmquist 2023 73 f Dissertação Mestrado em Ciências Econômicas Faculdade de Ciências Econômicas Universidade do Estado do Rio de Janeiro Rio de Janeiro 2023 O setor bancário tem fundamental importância para o desenvolvimento econômico de um país Através da intermediação financeira função primária dos bancos os recursos chegam aos agentes deficitários permitindo que os investimentos desejados sejam realizados Com os avanços tecnológicos e a sua incorporação na indústria financeira uma nova modalidade de bancos surgiu os bancos digitais Este trabalho tem como objetivo realizar uma análise comparativa da eficiência dos bancos digitais e dos bancos de varejo tradicionais no mercado de crédito brasileiro Para tanto foi utilizada a técnica nãoparamétrica DEA corrigida pelo método de Bootstrap com orientação para inputs para mensurar a eficiência técnica dos bancos a partir da abordagem da intermediação financeira A amostra ficou constituída de 73 bancos comerciais em que por meio de testes estatísticos adequados a combinação de inputs e outputs definida se mostrou convexa e com retornos variáveis de escala o que justificou a aplicação do modelo DEABCC Os resultados sugerem de modo geral que os bancos comerciais brasileiros apresentaram nível elevado de ineficiência nas atividades de intermediação financeira Foi observado que o grupo dos bancos digitais apresenta desempenho inferior quando comparado com o grupo de todos os bancos de varejo tradicionais Entretanto foi visto que quando comparados apenas com os bancos de varejo de mesmo porte a performance observada foi semelhante Por fim com a aplicação do Índice de Malmquist para o triênio de 20192021 foi percebida um aumento de produtividade dos bancos comerciais brasileiros atribuída a uma melhora de gestão dos insumos pelos bancos digitais e tradicionais Por outro lado não foi possível observar indícios de progresso tecnológico durante o período Palavraschave Análise Envoltória de Dados Bancos Digitais Intermediação Financeira ABSTRACT XAVIER Giovanni Barboza Banking benchmarking a comparison between digital and traditional banks with Efficiency Frontier Analysis and Malmquist Index 2023 73 f Dissertação Mestrado em Ciências Econômicas Faculdade de Ciências Econômicas Universidade do Estado do Rio de Janeiro Rio de Janeiro 2023 The banking sector is of fundamental importance for the economic development of a country Through financial intermediation the primary function of banks resources reach the deficit agents allowing the desired investments to be carried out With technological advances and their incorporation into the financial industry a new type of banking has emerged digital banks This work aims to carry out a comparative analysis of the efficiency of digital banks and traditional retail banks in the Brazilian credit market For this purpose the DEA nonparametric technique corrected by the Bootstrap method with input orientation was used to measure the technical efficiency of banks based on the financial intermediation approach The sample consisted of 73 commercial banks in which through adequate statistical tests the defined combination of inputs and outputs proved to be convex and with variable returns to scale which justified the application of the DEABCC model The results suggest in general that Brazilian commercial banks showed a high level of inefficiency in financial intermediation activities It was observed that the group of digital banks performs poorly when compared to the group of all traditional retail banks However it was seen that when compared only with retail banks of the same size the performance observed was similar Finally with the application of the Malmquist Index for the triennium 20192021 an increase in the productivity of Brazilian commercial banks was noticed attributed to an improvement in the management of inputs by digital and traditional banks On the other hand it was not possible to observe signs of technological progress during the period Keywords Data Envelopment Analysis Digital Banks Financial Intermediation LISTA DE ILUSTRAÇÕES Figura 1 Produtividade e eficiência 22 Figura 2 Modelo de Farrel para eficiência técnica e alocativa 24 Figura 3 Comparação das fronteiras de eficiência dos modelos CCR e BCC 30 Quadro 1 Principais características da DEA e da SFA 32 LISTA DE TABELAS Tabela 1 Abordagens para avaliar a eficiência bancária presentes na literatura 12 Tabela 2 Abordagens utilizadas na literatura nacional 19 Tabela 3 Inputs e output do modelo proposto 42 Tabela 4 Estatísticas descritivas dos inputs e output 20192021 44 Tabela 5 Testes de convexidade da amostra 45 Tabela 6 Testes de retornos de escala da amostra 46 Tabela 7 Escores de eficiência dos bancos no 3º trimestre de 2021 47 Tabela 8 Sumário estatístico das eficiências no 3º trimestre de 2021 49 Tabela 9 Eficiência dos bancos digitais no 3º trimestre de 2021 50 Tabela 10 Segmentação das instituições financeiras conforme o BCB 51 Tabela 11 Eficiência dos bancos de varejo no 3º trimestre de 2021 51 Tabela 12 Sumário da eficiência dos bancos no triênio 20192021 52 Tabela 13 Eficiência dos bancos digitais no triênio 20192021 53 Tabela 14 Eficiência dos bancos de varejo no triênio 20192021 53 Tabela 15 Teste 1 de equivalência de médias de Kneip et al 2016 54 Tabela 16 Teste 2 de equivalência de médias de Kneip et al 2016 55 Tabela 17 Eficiência de escala dos bancos digitais no 3º trimestre de 2021 57 Tabela 18 Eficiência de escala dos bancos de varejo no 3º trimestre de 2021 57 Tabela 19 Índice de produtividade de Malmquist para os bancos digitais 59 Tabela 20 Índice de alterações de eficiência para os bancos digitais 60 Tabela 21 Índice de alterações de tecnologia para os bancos digitais 61 Tabela 22 Índice de produtividade de Malmquist para os bancos de varejo 61 Tabela 23 Índice de alteração de eficiência para os bancos de varejo 62 Tabela 24 Índice de alterações de tecnologia para os bancos de varejo 63 LISTA DE ABREVIATURAS E SIGLAS AA AE AT BB BRB BCB BCC CCR CEF Cosif CRS DEA DMU FDH FGV IA IGP IM PA SFA TFA VAA VRS Asset Approach Alteração de Eficiência Alteração Tecnológica Banco do Brasil Banco Regional de Brasília Banco Central do Brasil Banker Charnes e Cooper Charnes Cooper e Rhodes Caixa Econômica Federal Padrão Contábil das Instituições Reguladas pelo Banco Central do Brasil Constant Returns to Scale Data Envelopment Analysis Decision Making Unit Free Disposal Hull Fundação Getúlio Vargas Intermediation Approach Índice Geral de Preços Índice de Malmquist Production Approach Stochastic Frontier Analysis Thick Frontier Analysis Value Added Approach Variable Returns to Scale SUMÁRIO INTRODUÇÃO 1 Objetivos 3 Estrutura da dissertação 4 1 REVISÃO DA LITERATURA SOBRE EFICIÊNCIA BANCÁRIA 6 11 Literatura internacional sobre eficiência no setor bancário 6 12 As principais abordagens de análise 11 13 A literatura nacional sobre eficiência bancária 14 2 METODOLOGIA 21 21 Produtividade e eficiência 21 22 Eficiência técnica e eficiência alocativa 23 23 A Análise Envoltória de Dados DEA 25 231 Modelo CCR 26 232 Modelo BCC 29 233 Comparação entre os modelos CCR e BCC 30 24 Características da técnica DEA e sua comparação com a SFA 31 25 O viés da DEA e a técnica de Bootstrap 33 26 A convexidade na DEA 35 27 Testes para os retornos de escala 37 28 Comparação de médias 38 29 Índice de Malmquist 39 3 RESULTADOS 41 31 Modelo proposto 41 32 A amostra 43 33 Convexidade e retornos de escala 45 34 Desempenho do setor bancário brasileiro em 2021 47 34 Evolução da eficiência dos bancos digitais e bancos de varejo entre 20192021 52 35 Comparação entre bancos digitais e bancos de varejo tradicionais 54 36 Análise da eficiência de escala 56 37 Análise temporal do setor bancário brasileiro 59 CONCLUSÕES 64 REFERÊNCIAS 68 1 INTRODUÇÃO O setor financeiro e principalmente o setor bancário tem fundamental importância para o desenvolvimento econômico de um país Este setor tem como função principal exercer a atividade de intermediação de recursos entre poupadores e tomadores fazendo com que valores poupados dos agentes superavitários cheguem aos agentes que demandam destes recursos para desempenhar suas atividades Instituições financeiras sólidas e eficientes na intermediação financeira permitem que os investimentos desejados sejam realizados o que é condição essencial para o desenvolvimento econômico de um país Cava et al 2016 explicam que o setor bancário brasileiro desempenha um papel importante na economia em que os mercados de capitais e de seguros ainda se encontram subdesenvolvidos o que destaca a participação dos bancos nos financiamentos de investimentos que contribuem para o desenvolvimento do país Sobre este aspecto Staub et al 2010 destacaram que o desenvolvimento do sistema bancário e aumento da sua eficiência estão relacionados com maiores crescimentos econômicos Conforme Périco et al 2016 ao longo do tempo praticamente em todos os lugares do mundo as mudanças vistas nas funções dos bancos são notáveis com o surgimento de novos produtos e com o aumento da disponibilidade dos serviços para os clientes entretanto o papel de intermediação financeira ainda prevalece Segundo os autores todos os outros recursos incorporados nas funções de um banco ainda são resultados de sua função primária a de intermediação financeira Boot et al 2020 explicam que a última década apresentou enormes desafios para as instituições financeiras incluindo crise financeira reforma regulatória baixa lucratividade e deterioração da confiança do público mas contra este cenário as mudanças tecnológicas na indústria financeira estão aumentando impulsionadas agora pela recente crise da COVID19 que tem aumentado a demanda por serviços digitais acelerando este processo Com os avanços tecnológicos e a sua incorporação na indústria financeira uma nova modalidade de banco emergiu os bancos digitais em que através da tecnologia determinadas funções são oferecidas ao público por meio de plataformas digitais sem o contato físico com o cliente De tal modo os serviços realizados em espaços específicos passam a ser realizados em ambientes virtuais onde tais bancos operam com poucas ou nenhumas agências e em sua maioria oferecem serviços atrativos para os clientes com isenção de taxas e tarifas bancárias 2 De acordo com Philippon 2019 Fintechs um termo para empresas que empregam tecnologia nos serviços financeiros abrangem inovações digitais e inovações nos modelos de negócios com tecnologia no setor financeiro e podem romper as estruturas recentes das empresas e a forma como estas criam e entregam produtos e também democratizar o acesso a serviços financeiros Segundo a OECD 2020 a disrupção digital tem potencial para prover uma maior eficiência no setor bancário por meio da inovação pois esta tendência tem como principal característica a desmaterialização dos processos trazendo benefícios como a redução de recursos e despesas o que pode promover a redução no valor de serviços para os clientes Ainda conforme a OECD 2020 este processo pode ser capaz de gerar um impacto na competitividade do setor em que os bancos de varejo tradicionais poderão ter que se mover para este modelo baseado em plataforma com atendimento centrado no cliente Dado a importância de um setor bancário eficiente para o desenvolvimento econômico pelo desempenho de sua função principal de intermediação financeira e a recente disrupção digital no setor com o surgimento dos bancos digitais surge o questionamento se estes bancos pelo seu modo de operação são mais ou menos eficientes que os bancos tradicionais Esta é a pergunta que este trabalho se propõe a responder É pretensão deste trabalho avaliar a eficiência do sistema bancário brasileiro no período recente de 2019 a 2021 período em que bancos digitais passaram a atuar mais intensamente no mercado bancário brasileiro Buscase com isso contribuir com a literatura nacional de eficiência bancária investigando se esta nova modalidade de instituição financeira os bancos digitais são mais ou menos eficientes que os bancos de varejo tradicionais Pretendese também examinar se a entrada destas instituições promoveu um aumento de eficiência no mercado de crédito brasileiro A Análise Envoltória de Dados DEA desenvolvida por Charnes Cooper e Rhodes 1978 é uma técnica utilizada para mensurar a eficiência de unidades produtivas que os autores denominam de Decision Making Units DMUs Essas unidades produtivas são caracterizadas por empregarem o uso dos mesmos insumos inputs para a produção de determinados produtos outputs A análise de eficiência na DEA tem como objetivo medir o desempenho da DMU em converter os insumos utilizados em produtos Esta técnica emprega a programação matemática para a construção da fronteira eficiente A DEA é uma técnica não paramétrica ou seja não necessita que seja especificada uma forma para a fronteira eficiente Existem outras técnicas para mensurar a eficiência de unidades produtivas como a Free Disposal Hull FDH e a Stochastic Frontier Analysis SFA Essas técnicas têm sido largamente utilizada pelos pesquisadores para investigar a eficiência 3 dos bancos Neste estudo é utilizada a técnica DEA que é mais aprofundada ao longo do capítulo 2 que é destinado a metodologia do trabalho Bancos podem ser avaliados por diferentes abordagens Neste trabalho será procurado avaliar a eficiência dos bancos brasileiros e realizar a comparação entre bancos digitais e bancos de varejo tradicionais no desempenho de sua principal função a intermediação financeira Pretendese fazer assim pois a concessão de crédito é a função dos bancos que mais tem impacto no desenvolvimento da economia de um país e permitirá investigar como os bancos digitais desempenham esta atividade Por fim é utilizado o Índice de Malmquist para o triênio de 2019 a 2021 com o objetivo de analisar a evolução do desempenho do setor bancário brasileiro ao longo deste período Este índice permite realizar uma análise completa da evolução da produtividade ao longo de um período determinado e sua decomposição permite obter informações sobre as variações de eficiências dos bancos e sobre as variações da tecnologia de produção Objetivos Fundamentandose na contextualização e justificativa do problema de pesquisa realizadas acima o presente estudo possui como objeto geral realizar a análise comparativa do desempenho dos bancos digitais com os bancos de varejo tradicionais atuantes no mercado de crédito brasileiro Ainda este trabalho pretende alcançar os seguintes objetivos específicos a Resumir as principais abordagens de avaliação dos bancos presentes na literatura sobre eficiência bancária b Definir o modelo com os inputs e outputs a serem utilizados de acordo com o que for mais indicado para avaliar os bancos na atividade de concessão de crédito c Avaliar a convexidade e os retornos de escala do conjunto de produção representado na amostra a ser selecionada d Mensurar a eficiência técnica dos bancos pela abordagem da intermediação financeira e Comparar os escores de eficiência médios dos bancos digitais com os bancos de varejo tradicionais 4 f Calcular e decompor o Índice de Produtividade de Malmquist para investigar as variações de eficiência e avanços tecnológicos no triênio de 2019 a 2021 g Contribuir com a literatura nacional sobre eficiência bancária investigando como os bancos digitais desempenham em comparação com os bancos de varejo e como impactam no mercado de crédito brasileiro Estrutura da dissertação Com a finalidade de alcançar o objetivo geral e os objetivos específicos propostos esta dissertação está dividida em 3 capítulos além desta introdução e das conclusões No capítulo 1 é feita uma revisão da literatura do setor bancário São apresentados os primeiros trabalhos que se propuseram a avaliar a eficiência de instituições financeiras discussão acerca das abordagens de análise que podem ser adotadas apresentada primeiramente por Favero e Papi 1995 é discorrida Ao final é feita uma contextualização das principais pesquisas aplicadas ao mercado de crédito brasileiro No capítulo 2 é apresentada toda a metodologia utilizada por este trabalho Inicialmente são explanadas as contribuições dos trabalhos da década de 50 considerados como o alicerce das técnicas utilizadas atualmente para se avaliar a eficiência Em seguida são apresentados modelos da Análise Envoltória de Dados DEA desenvolvidos por Charnes Cooper e Rhodes 1978 e Banker Charnes e Cooper 1984 Ainda são expostos os testes utilizados para se testar a convexidade do conjunto de produção e a presença de retornos constantes e variáveis de escala O teste de Kneip et al 2016 para se realizar a comparação da média das eficiências de grupos de produtores distintos utilizado para se alcançar o objetivo principal deste trabalho é explicado em sequência Por fim o Índice de Malmquist e a sua decomposição nos índices de alterações de eficiência e de alterações tecnológicas são expostos O capítulo 3 se inicia com a explicação do modelo proposto por este trabalho para analisar a eficiência bancária brasileira recente São discutidos os resultados obtidos descrevendo e analisando os desempenhos mensurados pelos bancos As comparações entre bancos digitais e de varejo tradicionais realizadas pelos testes estatísticos propostos são discorridas em seguir Por fim é realizada uma análise temporal do mercado de crédito 5 brasileiro por meio do Índice de Malmquist avaliando as alterações de eficiência técnica e as alterações da tecnologia ao longo do triênio de 2019 a 2021 Nas conclusões deste trabalho são apresentadas as principais conclusões obtidas no capítulo anterior e feita um diagnóstico geral do desempenho do mercado de crédito brasileiro Por fim são feitas sugestões para trabalhos futuros que abordem a mesma temática deste trabalho 6 1 REVISÃO DA LITERATURA SOBRE EFICIÊNCIA BANCÁRIA Este capítulo tem como objetivo fazer uma resenha da literatura sobre eficiência no setor bancário apresentando os mais relevantes trabalhos presentes na literatura internacional e nacional Na primeira seção serão apresentados os trabalhos considerados pioneiros que introduziram as técnicas de medição de eficiência para avaliação dos bancos e alguns dos principais trabalhos da literatura internacional Em seguida na segunda seção são discutidas as principais abordagens utilizadas para se avaliar as instituições financeiras apresentando o objetivo que cada uma delas pretende alcançar suas principais escolhas de inputs e outputs e os resultados alcançados Por fim são discutidas as pesquisas que buscaram avaliar a performance de bancos nacionais discorrendo sobre aspectos como a escolha dos insumos e produtos a formação da amostra e as conclusões alcançadas por estes estudos 11 Literatura internacional sobre eficiência no setor bancário Os primeiros autores a utilizar a técnica DEA para avaliar o desempenho do setor bancário foram Sherman e Gold 1985 O desempenho operacional de um branco era frequentemente medido usando índices contábeis como retorno sobre ativos e retorno sobre investimentos Sherman e Gold 1985 Segundo os autores a utilização destes índices apresentava limitações pois estes eram elaborados por meio de agregações extensas que não refletiam a eficiência operacional podendo levar a conclusões falsas Com o objetivo de aplicar uma técnica que permita superar as fraquezas que a utilização de índices contábeis apresenta Sherman e Gold 1985 aplicaram a DEA para identificar e desenvolver maneiras para melhorar a eficiência de agências bancárias O modelo CCR foi aplicado em 14 agências de um banco que desejava substituir os índices que utilizava por uma técnica que fosse superior e levando a melhores avaliações Os autores utilizaram como inputs uma medida de utilização de tempo de empregados o aluguel pago por cada agência como um representativo do capital utilizado e por fim o custo total dos suprimentos utilizados Como output foram utilizados o número de transações 7 realizadas pela agência como abertura de contas pedidos de empréstimos seguros de vida vendidos etc Os resultados apresentados pelos autores mostram que das agências analisadas oito apresentaram o escore de eficiência máximo enquanto seis agências apresentaram algum nível de ineficiência Nenhuma das cinco maiores agências foi identificada como ineficiente e duas das ineficientes eram as menores em termos de transações totais Os resultados foram validados pelos gerentes bancários que apontaram como possível explicação o fato de que as agências menores eram administradas por gerentes menos experientes Enquanto que Sherman e Gold 1985 estavam interessados em analisar as agências de um banco em particular Rangan et al 1988 foram os primeiros a avaliar o mercado bancário a partir de uma visão do setor como um todo Esses autores selecionaram uma amostra com 215 bancos em operação nos Estados Unidos e utilizaram da técnica da DEA para mensurar a eficiência técnica destas instituições Rangan et al 1988 tinham a intenção de analisar o desempenho das instituições na atividade de intermediação financeira no ano de 1986 Para isso utilizaram como inputs o número de trabalhadores de tempo integral ativos fixos e fundos arrecadados Como outputs foram utilizados os empréstimos imobiliários os empréstimos comerciais os empréstimos industriais os depósitos à vista e por fim os depósitos a prazo e de poupança A partir da eficiência média da amostra obtida através do modelo DEA Rangan et al 1988 concluíram que os bancos norteamericanos poderiam produzir o mesmo nível de outputs com 70 dos inputs que usavam Ferrier e Lovell 1990 compararam duas técnicas para avaliar o desempenho de uma amostra de 575 bancos americanos em atividade no ano de 1984 A primeira é uma abordagem econométrica de função translog e a segunda é a Análise Envoltória de Dados Os autores utilizaram como inputs o número total de empregados custos de aluguel e despesas com mobília e equipamentos Como outputs foram utilizados os números de contas de depósitos à vista depósitos a prazo empréstimos imobiliários e o número de empréstimos comerciais Os autores afirmaram que as técnicas estavam de substancial acordo em vários aspectos importantes Ferrier e Lovell 1990 concluíram que as duas técnicas estavam correlacionadas positivamente Berger e Humphrey 1991 propuseram uma nova técnica para mensurar e analisar as eficiências de todos os bancos dos Estados Unidos em 1984 o método paramétrico denominado Análise de Fronteira Espessa Thick Frontier Analysis TFA Os resultados do estudo sugerem 8 que as ineficiências dos bancos derivam dos inputs de capital físico e mãodeobra o que os autores chamaram de ineficiências operacionais Do ponto de vista político Berger e Humphrey 1991 indicam que a concorrência adicional promovida por uma desregulamentação do setor bancário pressionaria muitos dos bancos considerados ineficientes a cortar custos fundirse com instituições que apresentam eficiências operacionais ou deixar o mercado No estudo de Berg et al 1993 foram estudados os desempenhos do setor bancário de três países nórdicos A amostra foi composta com 503 bancos finlandeses 150 bancos noruegueses e 126 bancos suecos em atuação no ano de 1990 Os autores constataram que a amostra representava praticamente toda a indústria bancária destes países O objetivo era avaliar como bem preparados estariam os bancos de diferentes tamanhos destes países para encarar a competição mais intensa do Mercado Comum Europeu A técnica DEA foi utilizada por Berg et al 1993 considerando tanto o modelo com retornos constantes de escala quanto o modelo que permite retornos variáveis Os mercados bancários dos países foram analisados individualmente e em seguida o Índice de Malmquist foi aplicado para realizar uma comparação entre os países O modelo deste trabalho tem os empréstimos e os depósitos como outputs O número de agências é definido como output por representar a disponibilidade dos serviços bancários Os inputs considerados são o trabalho medido em horas de trabalho e capital medido pelos valores contábeis de máquinas e equipamentos Os resultados encontrados por Berg et al 1993 mostraram que o mercado de crédito sueco era o mais eficiente com os escores médios de 078 considerando retornos variáveis de escala e 069 considerando retornos constantes Com o objetivo de realizar uma comparação entre os países foi construído um Índice de Malmquist onde os países da Suécia foram considerados como base tecnológica por ser o país que se mostrou mais eficiente quando avaliado individualmente Os bancos suecos se mostraram 5263 mais produtivos que os bancos finlandeses com a maior parte da diferença tendo como origem o componente de eficiência A diferença tecnológica entre esses países era de aproximadamente 10 Quando comparados os bancos suecos com os bancos noruegueses a diferença de produtividade foi de 4046 com a maior parte da diferença se mantendo no componente de eficiência Os bancos da Noruega foram 10 mais produtivos que os bancos da Finlândia com o componente tecnológico apontando para esta direção nos modelos CCR e BCC Berg et al 1993 concluem que a maior parcela dos bancos que se encontram na fronteira nórdica de eficiência são suecos indicando que os bancos deste país estão na melhor posição para expansão em um futuro mercado bancário comum nórdico 9 A estimação da fronteira eficiente nos estudos de performance bancária é dividida entre os métodos paramétricos e não paramétricos Entre os métodos paramétricos a técnica mais utilizada é a Análise da Fronteira Estocástica Stochastic Frontier Analysis SFA que especifica uma forma funcional para a fronteira eficiente presente nos trabalhos de Ferrier e Lovell 1990 Berger e DeYoung 1997 Bonin et al 2005 e Berger et al 2009 Bonin et al 2005 usando dados do período de 1996 a 2000 avaliaram os efeitos do tipo de controle na eficiência bancária em onze países emergentes europeus Os autores utilizaram o método paramétrico SFA para mensurar os escores de eficiência destes bancos As variáveis de saída foram o total de depósitos o total de empréstimos e o total de ativos líquidos e investimentos Como entradas os autores selecionaram variáveis que representassem os preços do capital e os preços dos fundos arrecadados Os resultados mostraram que os bancos públicos não eram significativamente menos eficientes que os bancos privados sugerindo que a privatização não seria por si só suficiente para aumentar a eficiência do mercado de crédito destes países Bancos estrangeiros foram considerados mais eficientes em custos do que outros tipos de controle A técnica DEA foi amplamente utilizada para avaliar a eficiência de setores bancários de diversos países Barros et al 2012 aplicaram a DEA para analisar a eficiência técnica de bancos japoneses de 2000 a 2007 Azad et al 2019 avaliaram a eficiência do setor bancário de Bangladesh durante o ano de 2017 com a técnica DEA Ariff e Can 2008 avaliaram a eficiência de custos e de lucros de 28 bancos comerciais chineses com essa técnica não paramétrica Da mesma forma Kao e Liu 2013 examinaram a eficiência dos bancos taiwaneses A DEA é uma técnica não paramétrica de construção de fronteira eficiente enquanto que a SFA especifica uma forma funcional para a fronteira Vários estudos foram propostos na literatura para comparar essas técnicas Kuchler 2013 compararam a DEA e a SFA na estimação da eficiência dos bancos dinamarqueses durante a crise dos anos 2008 a 2010 O autor encontrou uma correlação considerável mas não perfeita entre os rankings de eficiência estimados pela DEA e SFA No mesmo sentido Tabak et al 2014 compararam as duas técnicas para estimar a eficiência de uma amostra de bancos chineses Os autores indicaram que a DEA e a SFA se mostraram consistentes para construir rankings de eficiência ao longo dos anos para o setor bancário como um todo porém indicaram que as correlações apontaram para divergências nos diagnósticos das performances individuais dos bancos Paradi et al 2011 construíram um modelo DEA de dois estágios com o objetivo de avaliar a performance de 816 agências de um dos cinco maiores bancos canadenses por três 10 abordagens diferentes produção rentabilidade e intermediação As eficiências obtidas nos três modelos são os outputs do segundo estágio onde uma variável dummy com o valor 1 é empregado como input para todas as agências Segundo os autores esta avaliação de múltiplas abordagens permitiria que os gestores através da aplicação DEA aprimorassem suas gestões em negócios reais O mercado bancário italiano foi estudado por Favero e Papi 1995 onde foram mensuradas as eficiências técnicas e de escala de uma amostra de 174 bancos coletada no ano de 1991 Os autores explicam que a indústria bancária estava sofrendo mudanças na legislação e na regulação devido ao processo de unificação do mercado europeu com aumento da competição e integração e portanto seria importante avaliar o impacto destas mudanças no mercado de crédito Favero e Papi 1995 avaliaram o mercado italiano sob duas abordagens diferentes a abordagem da intermediação financeira e a abordagem dos ativos Os autores aplicaram os modelos DEACCR e DEABCC e obtiveram como escores de eficiência média para a abordagem da intermediação os valores de 088 e 091 respectivamente Enquanto que na abordagem de ativos foram observados os valores médios de 079 e 084 Foi aplicado um teste de Wald para testar a hipótese nula de igualdade entre as eficiências medidas dos dois modelos que foi rejeitada para as duas abordagens de estudo Assim os autores afirmaram que parte da ineficiência do setor bancário italiano é derivada da existência de ineficiências de escala Favero e Papi 1995 aplicaram uma regressão para investigar os determinantes das ineficiências no mercado bancário italiano usando o modelo BCC devido ao resultado do teste aplicado Eles avaliaram o impacto do tamanho dos bancos da especialização produtiva do tipo de controle da estrutura de mercado e da localização Os autores indicaram que as ineficiências eram melhor explicadas pela especialização produtiva tamanho e localização dos bancos Uma importante contribuição de Favero e Papi 1995 foi a realização de uma discussão sobre as abordagens que podem ser adotadas para se avaliar a eficiência de um setor bancário Foram identificadas pelos autores cinco abordagens presentes na literatura bancária Foram discutidos os principais aspectos a atividade bancária que se pretende avaliar e como é feita a escolha de inputs e outputs em cada uma delas Na seção a seguir essa discussão será apresentada 11 12 As principais abordagens de análise Uma importante decisão que deve ser feita em relação a avaliação de desempenho do setor bancário é a escolha de inputs e outputs Ahmad et al 2020 afirmam que diferentes abordagens têm sido utilizadas pelos pesquisadores para realizar esta definição e que ainda não existe consenso sobre o assunto Fethi e Pasiouras 2010 no mesmo sentido afirmaram que existe uma discussão em curso na literatura sobre eficiência bancária sobre as definições de inputs e outputs mais apropriadas Esses autores destacaram a seguinte frase de Bergendahl 1998 Tem havido quase tantas suposições de entradas e saídas como tem havido aplicações de DEA Bhatia et al 2018 no mesmo sentido afirmam que devido a complexidade das transações financeiras existem diferentes escolas de pensamento para definir inputs e outputs para a eficiência bancária Favero e Papi 1995 discutiram as especificações de inputs e outputs dos trabalhos que buscaram avaliar o desempenho das instituições financeiras até aquele momento Os autores afirmaram que economistas estavam divididos sobre a definição correta de entradas e saídas e que uma definição da atividade bancária é essencial para que se possa avaliar a eficiência do setor bancário Foram identificadas por estes autores quatro abordagens para a especificação de inputs e outputs a abordagem da produção Production Approach PA a abordagem da intermediação Intermediation Approach IA a abordagem dos ativos Asset Approach AA e por fim a abordagem do valor agregado Value Added Approach VAA Alguns dos principais trabalhos sobre eficiência bancária foram classificados por Favero e Papi 1995 de acordo com a abordagem utilizada apresentando os inputs e outputs escolhidos Essa classificação é apresentada na tabela 1 12 Tabela 1 Abordagens para avaliar a eficiência bancária presentes na literatura Autores Inputs Outputs Abordagem Sherman e Gold 1985 Trabalho Capital Custo de suprimentos Número de transações PA Rangan et al 1988 Trabalho Capital Fundos arrecadados Empréstimos Depósitos IA Ferrier e Lovell 1990 Trabalho Gasto com materiais Aluguel e mobília Número de contas de depósitos Número de empréstimos PA Berger e Humphrey 1991 Trabalho Capital Fundos arrecadados Depósitos Empréstimos IA Berg et al 1993 Trabalho Capital Empréstimos Depósitos Número de agências VAA English et al 1993 Trabalho Fundos arrecadados Empréstimos Investimentos AA Fonte FAVERO PAPI 1995 p389 Segundo Favero e Papi 1995 na abordagem da intermediação bancos são vistos como intermediários em que são transformados e transferidos os recursos financeiros de unidades superavitárias para unidades deficitárias Ahmad et al 2020 afirmaram que esta é claramente a abordagem mais popular na amostra de estudos que estes autores selecionaram que consiste nos 100 artigos mais citados da literatura Bhatia 2018 afirmam que a abordagem da intermediação foi proposta pela primeira vez por Sealey e Lindley 1977 Favero e Papi 1995 definem a abordagem de ativos como uma variante da abordagem da intermediação em que são considerados como outputs outros ativos dos bancos como investimentos além dos empréstimos que a instituição fornece Enquanto que a abordagem da 13 intermediação foca na principal atividade desenvolvida pelos bancos a concessão de crédito a abordagem dos ativos tem uma visão mais ampla da instituição considerando outros ativos produzidos A principal deficiência destas duas abordagens é que elas levam em consideração a maioria dos serviços prestados pelos bancos Favero e Papi 1995 Bhatia et al 2018 explicam que outra variação da abordagem da intermediação é a abordagem do valor adicionado em que as atividades dos bancos capazes de gerar valor adicional são consideradas como outputs Nesta linha o estudo de Berg et al 1993 considera como produto o número de agências pois representa a disponibilidade dos serviços bancários De acordo com Bhatia et al 2018 a abordagem da intermediação é a mais popular entre os pesquisadores e eles concordam que os bancos são intermediários de serviços financeiros Apesar de depósitos serem considerados como outputs em alguns trabalhos que tinham como objetivo avaliar a eficiência do mercado bancário pela abordagem da intermediação Bhatia et al 2018 afirmam que é mais lógico consideralos como inputs pois bancos usam depósitos para praticar atividades capazes de gerar receitas como empréstimos e pagar juros aos depositantes Ainda sobre o papel dos depósitos no modelo de eficiência bancária Fethi e Pasiouras 2010 explicam que existe uma controvérsia na literatura sobre este assunto Estes autores afirmam que alguns estudos selecionam como outputs apenas aqueles de acordo com o proposto por Sealey e Lindley 1977 enquanto que outros utilizam os depósitos como um output adicional o que os deixa mais próximo da abordagem do valor adicionado do que da abordagem da intermediação A abordagem da produção considera o banco como um fornecedor de serviços bancários De acordo com Favero e Papi 1995 nesta abordagem o banco é considerado como produtor de contas de depósitos e serviços de empréstimos onde outputs são mensurados pelo número de contas ou transações processadas A abordagem da produção está presente no trabalho de Ferrier e Lovell 1990 Alguns trabalhos procuram avaliar os bancos a partir de uma abordagem de rentabilidade O estudo de Faria et al 2006 avaliou os bancos atuantes no mercado de crédito do Brasil pela sua rentabilidade usando com inputs despesas com juros e despesas não relacionadas a juros e como outputs receitas da intermediação financeira receitas de prestação de serviços e outras receitas operacionais Bhatia et al 2018 explicam que muitos pesquisadores usam mais de uma abordagem no mesmo estudo e outros modificam as especificações de inputs e outputs de forma que eles 14 não se enquadram em nenhuma abordagem mencionada para se adequar melhor ao objetivo pretendido pelos estudos Este presente trabalho propõe avaliar o setor bancário brasileiro de acordo com o modelo tradicional de intermediação financeira proposto por Sealey e Lindley 1977 em que os bancos são vistos como intermediários entre agentes poupadores e agentes tomadores de empréstimos A orientação do modelo aplicado será voltada para insumos que como Fethi e Pasiouras 2010 explicam a orientação para inputs é de longe a mais utilizada para estimar a eficiência dos bancos pois é assumido que os gestores tem maior controle sobre recursos empregados no processo do que sobre os produtos 13 A literatura nacional sobre eficiência bancária Vários estudos foram propostos para avaliar a eficiência no mercado de crédito brasileiro Ghilardi 2006 avaliaram o desempenho dos 50 maiores bancos que atuavam no Brasil por meio da técnica não paramétrica DEA Este trabalho avaliou o desempenho dos bancos considerando que o grau de eficiência pode ser medido pela capacidade de geração de receita utilizandose do patrimônio líquido e o número de funcionários os dois inputs do modelo Como outputs forma considerados a receita de serviços e outras receitas operacionais e o somatório das receitas de operações com arrendamento mercantil e as receitas de operações com títulos e valores mobiliários Ghilardi 2006 realizou uma comparação entre os modelos DEACCR e DEABCC com um índice de análise financeira tradicional o índice ROI Return On Investiment ROI Como resultado do estudo não foram observadas mudanças significativas entre os modelos DEA porém quando comparados com o índice ROI o autor afirma que parece que não se trata do mesmo assunto ou de que cada análise leva em conta princípios antagônicos dadas as contradições observadas na comparação Faria et al 2006 estavam interessados em avaliar o desempenho do setor bancário após um processo de 10 anos de mudanças profundas por uma onda de fusões e aquisições bancárias além da entrada de novas instituições estrangeiras no mercado brasileiro São avaliados o mercado bancário como um todo e individualmente os bancos envolvidos nestas operações por meio da técnica DEA A amostra foi composta de seis bancos varejistas que participaram 15 de fusões e aquisições no período analisado de 1995 à 2005 três bancos atacadistas e três bancos públicos Dois modelos foram construídos por Faria et al 2006 O primeiro modelo para a atividade de intermediação financeira tinha como inputs as seguintes rubricas contábeis captações totais despesas de pessoal e outras despesas administrativas e ativo permanente e imobilizado de arrendamento Como outputs foram utilizadas as operações de crédito e arrendamento mercantil e o somatório dos títulos e valores mobiliários e aplicações financeiras O modelo de resultados para avaliar a eficiência do banco em gerar renda tinha como inputs as despesas de intermediação financeira as despesas de pessoal e outras despesas As receitas de intermediação financeira e as receitas de prestações de serviços e outras receitas operacionais foram consideradas como as saídas deste modelo Os resultados apresentados por Faria et al 2006 indicam que o período de fusões e aquisições levou a um maior desempenho dos bancos nas atividades de intermediação financeira Os autores afirmaram que as instituições foram aumentando de tamanho e melhorando sua performance sendo possível atestar que o mercado de crédito brasileiro apresentou economias de escala na concessão de crédito Por outro lado os autores concluíram que no modelo de resultados o aumento do porte das instituições gerou impacto negativo na rentabilidade das mesmas Périco et al 2008 selecionaram os 12 maiores bancos comerciais para formarem a sua amostra Estes autores justificam a escolha pelo fato de não terem como objetivo explicar quais bancos são eficientes e quais não são mas sim de demonstrar de forma sucinta que nem sempre os maiores bancos são considerados os mais eficientes O modelo elaborado pelos autores teve como inputs o ativo total as operações de crédito os depósitos e o patrimônio líquido Receita financeira resultado financeiro resultado operacional e resultado líquido foram as quatro saídas do modelo O modelo escolhido foi o DEABCC com orientação para outputs Na amostra selecionada os cinco maiores bancos foram os únicos a apresentarem o escore de eficiência máximo Todos os outros bancos operavam com algum nível de ineficiência Périco et al 2008 destacaram que os 12 maiores bancos operavam com algum nível de ineficiência dado ao escore médio de 79 Os autores também afirmaram que a pergunta proposta inicialmente sobre a possibilidade de o maior banco ser o mais eficiente foi respondida positivamente sendo possível desde que ele otimize a utilização de seus recursos Outro ponto fundamental deste trabalho é a relação entre porte da instituição e eficiência que fica evidenciada no fato de que os cinco maiores bancos foram eficientes e os escores 16 diminuíam de acordo com o tamanho da instituição apontando para a presença de economias de escala no mercado de crédito brasileiro Paula e Faria 2008 avaliaram a evolução da eficiência técnica e de escala do setor bancário brasileiro no período de 2000 a 2006 Foram selecionadas para a amostra 38 instituições bancárias onde oito eram grandes bancos varejistas oito bancos de varejo regionais três bancos varejistas para alta renda doze bancos atacadistas e sete bancos especializados em crédito Os autores justificaram a segmentação pelo interesse em avaliar a eficiência bancária considerando uma dimensão ainda pouco estudada no mercado de crédito brasileiro o tipo e a natureza da instituição e seu nicho de mercado Para mensurar a eficiência técnica Paula e Faria 2008 optaram pelo modelo DEA BCC pois estavam interessados em avaliar a eficiência sem a influência da escala de operação do banco e este modelo permite retornos variáveis de escala A eficiência de escala foi calculada apenas para os bancos varejistas Os autores justificaram essa escolha pelo fato de que os bancos atacadistas possuem características semelhantes a um banco de investimento não tendo a concessão de crédito como sua atividade principal e por sua vez os bancos especializados em crédito possuem um foco de negócios bem específico para poderem ser comparados com bancos varejistas Nos resultados Paula e Faria 2008 destacam que a evolução da eficiência técnica no período de análise não apresentou melhorias acentuadas nem perdas Os autores apontaram que uma melhoria recente na eficiência técnica no modelo dos resultados coincidiu com o boom de crédito ocorrido no país Foram identificados níveis maiores de eficiência técnica nos grandes bancos varejistas e bancos de varejo para alta renda que os autores atribuíram a um possível aprimoramento operacional dos bancos e cortes nos custos administrativos Sobre a eficiência de escala Paula e Faria 2008 mostram que o grupo dos grandes bancos varejistas apresentaram o segundo melhor desempenho entre os segmentos selecionados enquanto que os bancos varejistas regionais apresentaram o pior desempenho da amostra Os autores indicam que isto parecia evidenciar que o tamanho da instituição afeta o seu desempenho Os resultados da eficiência de escala no modelo de intermediação mostram uma pequena perda de escala Os autores indicam que a opção de ser grande no mercado varejista parece ser vantajosa em função do maior potencial de geração de receitas e lucros decorrentes da ampliação no volume dos serviços ofertados e do potencial de vendas cruzadas que um banco pode oferecer Staub et al 2010 avaliaram o mercado de crédito brasileiro no período de 2000 a 2007 Foram estimadas as eficiências de custos técnica e alocativa aplicando a técnica DEA Os 17 bancos foram avaliados a partir da abordagem da intermediação em que os inputs representavam os usos de capital trabalho e fundos arrecadados Como outputs os autores utilizaram os empréstimos investimentos e depósitos A inclusão dos depósitos como outputs foi justificada pelo fato de que eles são proporcionais aos serviços de depósitos providos pelos bancos o que aproxima o modelo construído pelos autores da abordagem do valor adicionado variante do modelo da intermediação Os resultados de Staub et al 2010 indicam que bancos estrangeiros e bancos com participações estrangeiras são os menos eficientes comparados com outros tipos de controle sugerindo que a hipótese de vantagem global não estava prevalecendo no Brasil Os autores indicaram que o tamanho não era um importante fator determinante de eficiência Também não foram verificadas diferenças substanciais nos desempenhos dos bancos de diferentes áreas de atuação Por fim os autores afirmaram que os altos níveis de ineficiência técnica sugerem que ainda existe um largo caminho para melhor a eficiência bancária no Brasil A eficiência dos bancos brasileiros foi avaliada durante o longo período de 1998 a 2010 por Wanke e Barros 2014a com a utilização de um modelo paramétrico estocástico As variáveis foram escolhidas de acordo com a abordagem de intermediação financeira proposta por Sealey e Lindley 1977 onde o banco é visto como uma empresa financeira que emprega depósitos empregados e ativos totais para gerar empréstimos e investimentos em títulos Os autores indicaram que através da aplicação deste modelo foi verificado uma melhora no desempenho do setor bancário brasileiro ao longo do tempo Wanke e Barros 2014b aplicaram um modelo DEA de dois estágios para investigar a eficiência dos bancos brasileiros no ano de 2012 O número de agências e o número de funcionários foram os inputs do output intermediário de despesas administrativas e despesas de pessoal que posteriormente se tornaram inputs do segundo estágio que tinha como outputs capital próprio e ativos permanentes Os autores indicaram estarem interessados em olhar para os bancos como uma sequência de um processo produtivo que transforma recursos humanos e físicos em termos financeiros Os resultados apresentados por Wanke e Barros 2014b indicam que apenas um banco da amostra de 40 instituições se mostrou eficiente no primeiro estágio Os autores afirmaram que os bancos brasileiros não eram eficientes em usar os recursos humanos e físicos No segundo estágio nenhum banco se mostrou eficiente As escores médios de eficiência do primeiro e segundo estágio foram respectivamente 043 e 086 indicando que os bancos brasileiros são mais eficientes em converter despesas administrativas e de pessoal em ativos permanentes e capital próprio do que em gerir seus recursos humanos e físicos 18 Cava et al 2016 promoveram uma análise diferente do setor bancário brasileiro Estes autores se propuseram a avaliar a eficiência pela abordagem da produção No modelo elaborado pelos autores o banco é visto como um fornecedor de serviços depósitos representados pelo output de contas de depósitos e como um fornecedor de outros serviços representados pelas receitas não relacionadas à juros Como inputs foram utilizados o número de empregados as despesas operacionais líquidas de despesas da intermediação financeira e os ativos fixos Os bancos foram avaliados no ano de 2013 utilizando o modelo DEABCC pois os autores afirmaram que os bancos apresentam ganhos relativos ao tamanho dos serviços totais produzidos Os resultados obtidos por Cava et al 2016 indicam que os maiores bancos são mais eficientes visto que apresentam o maior escore médio de eficiência Os autores apontam que uma possível explicação seria as economias de escala alcançadas pelos maiores bancos Considerando o tipo de controle os bancos públicos foram considerados mais eficientes que os bancos privados e estrangeiros indicando que estes bancos são mais eficientes em produzir serviços Ainda bancos estrangeiros se mostraram mais eficientes que os bancos privados Bancos com risco classificado com AAA obtiveram os maiores escores de eficiência médio indicando que bancos com melhores serviços apresentam melhores resultados ao lado de melhor classificação de risco Carneiro et al 2016 pela abordagem da intermediação financeira avaliaram a eficiência de 99 instituições financeiras que atuavam no mercado brasileiro em 2013 Foram considerados como inputs o total de depósitos o número de trabalhadores e as despesas com juros enquanto que como outputs foram considerados os empréstimos as receitas de intermediação financeira e os investimentos e aplicações dos bancos O modelo escolhido foi o DEABCC pois os autores consideraram que as instituições financeiras atuam em ambientes onde há ganhos e perdas de escala No conjunto de instituições que apresentaram o nível máximo de eficiência no estudo de Carneiro et al 2016 destacamse as instituições financeiras privadas com participação estrangeira e as públicas com escores médios respectivos de 88 e 79 Os resultados vão de encontro com os obtidos por Staub et al 2010 onde as instituições financeiras públicas são mais eficientes que as privadas nacionais Por outro lado no trabalho de Staub et al 2010 as instituições privadas com participação estrangeira se mostraram economicamente ineficientes O tamanho foi identificado como determinante para o escore de eficiência com os bancos de grande porte se mostrando mais eficientes Também foi identificado uma forte relação entre risco de crédito com a eficiência 19 Mendonça 2017 investigou a relação entre a eficiência mensurada pela técnica DEA com a da SFA para uma amostra de 47 bancos em atuação no mercado brasileiro nos anos de 2008 a 2015 Os resultados deixam evidentes que cada método tem objetivos distintos que se refletem na forma como estimam a eficiência das instituições analisadas A DEA apresentou um maior número de instituições eficientes o que os autores atribuíram ao método de fazer um ranqueamento das instituições eficientes O método SFA apresentou uma menor dispersão nos escores de eficiência o que o autor explicaram pela atribuição de pesos iguais para as variáveis no momento de estimar a fronteira eficiente A tabela a seguir apresenta os inputs e outputs utilizados nos trabalhos apresentados nesta seção que avaliaram a eficiência do setor bancário brasileiro O resumo a seguir contém apenas os trabalhos que são passíveis de ser enquadrados na classificação das abordagens de análise apresentada por Favero e Papi 1995 Tabela 2 Abordagens utilizadas na literatura nacional continua Autores Inputs Outputs Abordagem Faria et al 2006 Captações Despesas de pessoal Ativo permanente e imobilizado de arrendamento Operações de crédito Arrendamento mercantil IA Paula e Faria 2008 Número de funcionários Depósitos totais Permanente e imobilizado de arrendamento Capital Operações de crédito e arrendamento mercantil Investimentos AA Staub et al 2010 Uso de capital Trabalho Fundos arrecadados Empréstimos Investimentos Depósitos AA 20 Tabela 2 Abordagens utilizadas na literatura nacional conclusão Wanke e Barros 2014a Cava et al 2016 Carneiro et al 2016 Depósitos Empregados Ativos totais Número de empregados Despesas operacionais líquidas Despesas da intermediação financeira Ativos fixos Total de depósitos Número de trabalhadores Despesas com juros Empréstimos Investimentos Contas de depósito Receitas não relacionadas à juros Empréstimos Receitas de intermediação financeira Investimentos e aplicações AA PA AA Fonte O autor 2022 21 2 METODOLOGIA Este capítulo apresenta a metodologia utilizada para a análise que se pretende Os conceitos de produtividade e eficiência são diferenciados na primeira seção Em seguida são apresentadas as contribuições dos trabalhos de Koopmans 1951 Debreu 1951 e Farrell 1957 que são considerados como o alicerce das técnicas utilizadas para se avaliar a eficiência atualmente Nas seções seguintes são apresentados a Análise Envoltória de Dados DEA seus principais conceitos e os modelos de Charnes Cooper e Rhodes 1978 conhecido como modelo CCR e Banker Charnes e Cooper 1984 conhecido como modelo BCC A explicação dos métodos utilizados neste trabalho continua com a apresentação das principais características da técnica nãoparamétrica DEA e sua comparação com a técnica paramétrica de Análise da Fronteira Estocástica SFA O viés característico da DEA é discutido e o método de Bootstrap introduzido no âmbito da DEA pelo trabalho de Simar e Wilson 1998 é exposto em sequência Em seguida são apresentados os testes utilizados para se testar a convexidade do conjunto de produção e a presença de retornos constantes de escala contra retornos variáveis O teste de Kneip et al 2016 utilizado para se alcançar o objetivo principal deste trabalho que permite realizar a comparação da média das eficiências de grupos de produtores distintos é explicado em sequência Por fim o Índice de Produtividade de Malmquist e a sua decomposição nos índices de alterações de eficiência e de alterações tecnológicas são expostos na última seção 21 Produtividade e eficiência Existem algumas expressões que são bastante utilizadas quando estamos nos referindo ao desempenho produtividade eficiência Apesar de tratarem de um assunto semelhante possuem conceitos diferentes que não se confundem A produtividade é definida como a razão entre o output produzido e o input que foi utilizado 22 𝒑𝒓𝒐𝒅𝒖𝒕𝒊𝒗𝒊𝒅𝒂𝒅𝒆 𝒐𝒖𝒕𝒑𝒖𝒕 𝒊𝒏𝒑𝒖𝒕 1 Coelli et al 1998 explicam que o cálculo da produtividade é simples quando o processo produtivo envolve um único input e um único output porém quando existirem mais de um algum método que realize a agregação desses inputs ou outputs precisa ser aplicado para se obter uma medida da produtividade Para ilustrar a diferença entre produtividade e eficiência Coelli et al 1998 apresentaram uma fronteira de produção que representa o máximo de um output y que pode ser obtido dado a quantidade do input x utilizado Figura 1 Produtividade e eficiência Fonte COELLI et al1998 p4 As unidades C e B encontramse posicionados na fronteira de produção F representando pontos eficientes A é ineficiente pois poderia aumentar o nível do output produzido dado a quantidade de input que utiliza Para ser eficiente A deveria produzir a mesma quantidade de output produzido pela unidade B Considerando um raio que passa pela origem e tem como inclinação yx uma medida de produtividade para as unidades consideradas caso A mudasse para o ponto de B teria sua produtividade aumentada pois a inclinação do raio seria maior No entanto caso A mudasse para o ponto C o raio da reta que passa pela origem é tangente à fronteira eficiente definindo o ponto de máxima produtividade possível 23 Coelli et al 1998 concluem que um ponto pode ser tecnicamente eficiente pois se encontra na fronteira de eficiência mas não ser a mais produtiva podendo aumentar a sua produtividade explorando economias de escala 22 Eficiência técnica e eficiência alocativa Os trabalhos de Koopmans 1951 Debreu 1951 e Farrell 1957 foram os alicerces para o desenvolvimento das técnicas que utilizadas atualmente para avaliar se um plano de produção é eficiente Casado e Souza 2015 explicam que os estudos dedicados a análise de eficiência da década de 1950 vieram a servir como base para o posterior desenvolvimento da técnica não paramétrica conhecida como Análise Envoltória de Dados Data Envelopment Analysis DEA por Charnes Cooper e Rhodes 1978 Koopmans 1951 definiu um conceito de eficiência conhecido como modelo de Pareto Koopmans Nesta definição a eficiência é alcançada se nenhum dos produtos pode ser aumentado sem que algum outro produto seja reduzido ou algum insumo seja aumentado e nenhum dos insumos possa ser reduzido sem que algum outro insumo seja aumentado ou algum produto seja reduzido Assim dentro da tecnologia que a unidade de produção tem disponível a unidade produtiva eficiente não consegue aumentar um produto mantendo constantes os níveis de insumos utilizados e o nível dos outros outputs produzidos para tanto precisa realocar algum destes recursos para alcançar o aumento no produto que deseja Debreu 1951 determinou em seus estudos o coeficiente de utilização de recursos onde a eficiência é alcançada pela redução dos insumos utilizados Casado e Souza 2015 explicam que Debreu 1951 estabeleceu o primeiro indicador de eficiência produtiva conhecido Este coeficiente é baseado na minimização dos insumos utilizados Consiste na redução radial equiproporcional de todos os recursos mantendo o nível atual de produção que tornaria a unidade produtiva eficiente Portanto segundo Debreu uma unidade é eficiente quando não é possível gerar a mesma produção com uma quantidade equiproporcionalmente menor dos insumos utilizados Ramos 2007 explica que Farrell 1957 desenvolveu seu conceito de eficiência técnica inspirado na análise da atividade de Koopmans 1951 e no conceito de coeficiente de utilização de recursos de Debreu 1951 Segundo Casado e Souza 2015 Farrel estendeu o 24 trabalho iniciado por Debreu para desenvolver um método para calcular o indicador de eficiência produtiva Conforme explicado por Ramos 2007 Farrell apresenta um modelo conforme a figura 2 a seguir com dois insumos X1 e X2 e um produto SS denota a isoquanta e AA a isocusto Figura 2 Modelo de Farrell para eficiência técnica e alocativa Fonte RAMOS 2007 p3 O ponto P representa uma firma que usa mais insumos que a firma Q na isoquanta considerada eficiente Farrell então considera que a razão OQOP poderia representar uma medida radial equiproporcional dos insumos X1 e X2 da eficiência técnica que como explica Ramos 2007 está presente em praticamente todos os modelos e aplicações de DEA usadas atualmente A eficiência técnica será igual a unidade se a firma no ponto P puder alcançar a isoquanta e se aproximará de zero quando a distância entre P e Q aumentar Podese perceber também que Q e Q são unidades de produção consideradas eficientes tecnicamente pois produzem o mesmo produto com combinações ótimas dos insumos X1 e X2 Uma firma que opera em um ponto da isoquanta QQ é tecnicamente eficiente mas isso não significa que estará operando com a combinação mais lucrativa dos fatores de produção dados os preços relativos Considerando a relação de preços dos insumos representada pela isocusto AA que tangencia a isoquanta SS a unidade Q é mais eficiente em termos de preço ou seja é mais eficiente alocativamente Assim Q e Q possuem a mesma eficiência técnica produtividade mas diferente eficiência alocativa custos Enquanto que para calcular a eficiência técnica precisamos apenas dos insumos e produtos para a eficiência alocativa é preciso conhecer o custo dos insumos seus preços Pela característica dos inputs utilizados neste trabalho que serão apresentados no próximo capítulo 25 por não apresentarem custos não é possível avaliar a eficiência dos bancos no emprego dos insumos em proporções adequadas dados os seus respectivos preços 23 A Análise Envoltória de Dados DEA A Análise Envoltória de Dados DEA foi apresentada primeiramente por Charnes Cooper e Rhodes 1978 A DEA é utilizada para a mensurar a eficiência de unidades produtivas homogêneas Estas unidades produtivas foram chamadas por estes autores de Decision Making Units DMUs e são caracterizadas por empregarem o uso dos mesmos inputs para a produção de determinados outputs A análise de eficiência tem como objetivo medir o desempenho da DMU em converter os insumos utilizados em produtos A DMU tecnicamente eficiente é aquela que opera na fronteira das possibilidades de produção Casado e Souza 2015 explicam que Charnes Cooper e Rhodes 1978 generalizaram os estudos de Farrel para trabalhar com múltiplos insumos e múltiplos outputs Essa generalização deu origem a DEA técnica que permite a construção de fronteiras de produção e indicadores de eficiência produtiva A DEA é uma técnica nãoparamétrica assim não necessita que seja especificada uma forma para a função que representa a fronteira de produção Esta técnica emprega programação matemática para a construção das fronteiras Os modelos básicos de DEA são o modelo originalmente proposto por Charnes Cooper e Rhodes 1978 conhecido como modelo CCR e o modelo proposto por Banker Charnes e Cooper 1984 conhecido como modelo BCC O primeiro modelo considera que a tecnologia de produção apresenta apenas retornos constantes de escala enquanto que o modelo BCC considera que as tecnologias podem apresentar retornos variáveis Os modelos DEA em sua aplicação podem ser divididos em modelos orientados para inputs e modelos orientados para outputs No primeiro buscase obter um menor uso de inputs dado o nível de outputs produzidos O pesquisador que utiliza a orientação para outputs busca obter o maior nível de outputs possível mantendo os inputs fixos A seguir serão apresentados os principais modelos DEA existentes CCR orientado para inputs CCR orientado para outputs BCC orientado para inputs e BCC orientado para outputs 26 231 Modelo CCR O modelo apresentado originalmente por Charnes Cooper e Rhodes 1978 trabalha com retornos constantes de escala Constant Returns to Scale CRS e foi inicialmente proposto como um modelo orientado para inputs Souza e Meza 2006 explicam que este modelo determina a eficiência pela otimização da divisão entre a soma ponderada dos outputs chamada de output virtual e a soma ponderada dos inputs chamada de input virtual A eficiência técnica de qualquer DMU no modelo CCR CRS original é obtida através de um Problema de Programação NãoLinear nas seguintes condições 𝒎á𝒙 𝒉𝟎 𝒖𝒋𝒚𝒋𝟎 𝑺 𝒋𝟏 𝒗𝒊𝒙𝒊𝟎 𝑹 𝒊𝟏 1 sa 𝒖𝒋𝒚𝒋𝒌 𝑺 𝒋𝟏 𝒗𝒊𝒙𝒊𝒌 𝑹 𝒊𝟏 𝟏 𝒌 𝟏 𝟐 𝒏 2 𝒖𝒋 𝒗𝒊 𝟎 𝒋 𝟏 𝑺 𝒊 𝟏 𝑹 3 Onde h0 Eficiência da DMU que está sendo avaliada yjk Quantidade do output yj para a DMU k xik Quantidade do input xi para a DMU k R Quantidade total de inputs S Quantidade total de outputs n Quantidade total de DMUs uj Peso referente ao output j vi Peso referente ao input i yj0 Quantidade do output yj para a DMU que está sendo avaliada xi0 Quantidade do input xi para a DMU que está sendo avaliada 27 A solução deste problema de maximização leva a determinação dos pesos u e v que maximizam a eficiência da DMU que está sendo avaliada A primeira condição do problema indica que as razões das demais DMUs deve ser menor ou igual à unidade enquanto que a segunda condição determina que os pesos devem ser maiores ou iguais a zero A maximização confere então para esta DMU em análise a ponderação mais favorável que as restrições permitem O modelo proposto por Charnes et al 1978 em uma forma fracionária pode ser formulado como um Problema de Programação Linear conhecido como modelo dos multiplicadores O modelo CCR input orientado na forma do modelo dos multiplicadores tem a seguinte formulação 𝒎á𝒙 𝒉𝟎 𝒖𝒋𝒚𝒋𝟎 𝑺 𝒋𝟏 4 sa 𝒗𝒊𝒙𝒊𝟎 𝑹 𝒊𝟏 𝟏 5 𝒖𝒋𝒚𝒋𝒌 𝑺 𝒋𝟏 𝒗𝒊𝒙𝒊𝒌 𝑹 𝒊𝟏 𝟎 𝒌 𝟏 𝒏 6 𝒖𝒋 𝒗𝒊 𝟎 𝒋 𝟏 𝑺 𝒊 𝟏 𝑹 7 O modelo dos multiplicadores pode ser derivado em um modelo dual que como explicam Souza e Meza 2006 apresentará uma menor quantidade de restrições e por ter solução computacional mais simples tem preferência sobre o modelo dos multiplicadores Esta forma dual conhecida como modelo da envoltória ou do envelope é apresentado a seguir 𝒎í𝒏 𝜽 8 sa 28 𝒚𝒋𝟎 𝒚𝒋𝒌𝝀𝒌 𝒏 𝒌𝟏 𝟎 𝒋 𝟏 𝑺 9 𝜽𝒙𝒊𝟎 𝒙𝒊𝒌𝝀𝒌 𝒏 𝒌𝟏 𝟎 𝒊 𝟏 𝑹 10 𝝀𝒌 𝟎 𝒌 𝟏 𝒏 11 Onde 𝜆𝑘 Peso da combinação convexa da DMU k No modelo da envoltória a DMU é eficiente quando 𝜃 1 caso for menor que a unidade reflete o valor que devemos multiplicar todos os inputs para produzir o nível de outputs atual de maneira eficiente Conforme Souza e Meza 2006 no modelo DEA é conveniente que o número de DMUs seja maior que o número de variáveis inputs e outputs visto que se n S R o modelo pode eventualmente não discriminar os escores das DMUs Logo quando o número de DMUs for maior que o número de variáveis ou sejam n S R o modelo do envelope possui menos restrições S R que o modelo dos multiplicadores n1 O modelo original CCR e os modelos dos multiplicadores e do envelope foram expostos para a orientação de inputs O modelo CCR output orientado na forma da envoltória tem a seguinte formulação 𝒎á𝒙 𝜽 12 sa 𝒙𝒊𝟎 𝒙𝒊𝒌𝝀𝒌 𝒏 𝒌𝟏 𝟎 𝒊 𝟏 𝑹 13 𝜽𝒚𝒋𝟎 𝒚𝒋𝒌𝝀𝒌 𝒏 𝒌𝟏 𝟎 𝒋 𝟏 𝑺 14 𝝀𝒌 𝟎 𝒌 𝟏 𝒏 15 29 No modelo CCR output orientado 𝜃 é o valor que devemos multiplicar todos os outputs para que a DMU em análise seja eficiente Desta forma 𝜃 é maior ou igual a 1 Logo devemos fazer 1 𝜃 se quisermos encontrar as eficiências no intervalo de 0 a 1 como no modelo CCR orientado para inputs 232 Modelo BCC Banker Charnes e Cooper 1984 desenvolveram um modelo que utiliza retornos variáveis de escala Variable Returns to Scale VRS assim de acordo com Casado e Souza 2015 evita problemas existentes em situações de competição imperfeita O modelo CCR é limitado no sentido de considerar que a tecnologia pode apresentar apenas retornos constantes de escala O Modelo BCC como é conhecido permite que a tecnologia apresente também retornos crescentes ou decrescentes de escala Na forma dos multiplicadores essa flexibilidade é acrescentada pela inclusão de um intercepto na função objetivo como pode ser visto nas equações a seguir que apresenta o modelo orientado para inputs 𝒎á𝒙 𝒉𝟎 𝒖𝒋𝒚𝒋𝟎 𝑺 𝒋𝟏 𝒖 16 sa 𝒗𝒊𝒙𝒊𝟎 𝑹 𝒊𝟏 𝟏 17 𝒖𝒋𝒚𝒋𝒌 𝑺 𝒋𝟏 𝒗𝒊𝒙𝒊𝒌 𝑹 𝒊𝟏 𝒖 𝟎 𝒌 𝟏 𝒏 18 𝒖𝒋 𝒗𝒊 𝟎 𝒋 𝟏 𝑺 𝒊 𝟏 𝑹 19 A fronteira é construída a partir dos pontos eficientes Como explicam Banker et al 1984 em um ponto eficiente retornos crescentes decrescentes ou constantes de escala são 30 verificados a depender do valor do intercepto Assim temos em um modelo orientado para inputs Para 𝑢 0 Retornos crescentes de escala Para 𝑢 0 Retornos constantes de escala Para 𝑢 0 Retornos decrescentes de escala 233 Comparação entre os modelos CCR e BCC Enquanto que no modelo CCR os pontos eficientes se encontram posicionados em uma reta que sai da origem no modelo BCC a fronteira é formada por segmentos de reta com inclinações diferentes devido ao intercepto de cada DMU eficiente que denota os retornos de escala presentes naquele ponto A Figura 3 a seguir ilustra essa diferença entre os modelos CCR e BCC Figura 3 Comparação das fronteiras eficientes dos modelos CCR e BCC Fonte BANKER et al 1984 p1089 31 Conforme explicam Tupy e Yamaguchi 1998 a eficiência produtiva tem dois componentes o puramente técnico físico e o alocativo preço o primeiro referese à habilidade de evitar perdas produzindo tantos produtos quanto os insumos utilizados permitem ou utilizando o mínimo de insumo possível no processo de produção o segundo componente referese à habilidade de combinar insumos e produtos em proporções ótimas dados os seus preços Para os autores uma firma além de ser eficiente na utilização dos insumos eficiência técnica e alocativa deverá também ser eficiente no produto eficiência de escala A eficiência de escala é obtida pela razão entre as eficiências dos modelos CCR e BCC ou seja entre as eficiências técnicas de longoprazo e de curtoprazo Uma firma eficiente no produto eficiência de escala opera onde uma mudança no nível dos outputs resultada em uma mudança proporcional nos inputs A ineficiência de escala é a ineficiência que advém de uma unidade produtiva estar operando fora do nível de produto com retornos constantes de escala A distância do modelo BCC para o modelo CCR representa a ineficiência de escala Na figura 3 podese observar que a DMU no ponto A é tecnicamente ineficiente enquanto que a DMU no ponto B representa uma firma que opera com o mesmo nível de produto mas com menos insumos sendo considerada tecnicamente eficiente A DMU do ponto E opera no ponto em que as curvas dos modelos CCR e BCC são iguais sendo além de eficiente o ponto mais produtivo O ponto E é eficiente em escala tendo produtividade máxima representando o tamanho ótimo de uma firma neste conjunto de produção 24 Características da técnica DEA e sua comparação com a SFA A principal característica da técnica de estimação DEA é ser baseada na programação linear não exigindo que se estabeleça especificações sobre a função de produção do processo que se está analisando Falcão e Correia 2012 destacam como uma importante característica desse método a sua flexibilidade que permite trabalhar com sistemas com diversos insumos e produtos De acordo com Falcão e Correa 2012 na DEA é analisada somente uma unidade por vez medindo o desempenho relativo dela o que gera um resultado para cada DMU que é utilizado para fazer a comparação e posteriormente ordenálos 32 Marinho 2001 elencou algumas das características mais interessantes da DEA que descrevem como essa técnica se configura a Caracteriza cada DMU como eficiente ou ineficiente através de uma única medida resumo de eficiência b Não faz julgamentos a priori sobre os valores das ponderações de inputs e outputs que levariam as DMUs ao melhor nível de eficiência possível c Pode prescindir mas não rejeita de sistemas de preços d Dispensa mas pode acatar préespecificações de funções de produção subjacentes e Pode considerar sistemas de preferências de avaliadores e de gestores f Baseiase em observações individuais e não em valores médios g Permite a incorporação na análise de insumos e de produtos avaliados em unidades de medidas diferentes h Possibilita a verificação de valores ótimos de produção e de consumo respeitando restrições de factibilidade i Permite a observação de unidades eficientes de referência para aquelas que forem assinaladas como ineficientes e j Produz resultados alocativos eficientes no sentido de Pareto Por sua vez a Análise da Fronteira Estocástica Stochastic Frontier Analysis SFA é um método paramétrico e estocástico Silva 2014 explica que a SFA estima a função fronteira ideal de produção ou de custo utilizandose regressão de múltiplas variáveis a partir de um conjunto de dados das DMUs na qual os insumos são colocados como variáveis independentes e o produto ou indicador que sintetiza vários produtos é colocado como variável independente Quadro 1 Principais características da DEA e da SFA DEA SFA Não paramétrica Não paramétrica Determinística Estocástica Não realiza suposições na distribuição do termo da ineficiência Realiza suposições na distribuição do termo da ineficiência Não inclui o erro como termo Inclui um termo composto do erro Não exige a especificação de uma função Exige a especificação de uma função Método Programação Linear Método Econométrico Fonte FALCÃO CORREIA 2012 p133 33 Coelli et al 2003 explicam que a seleção da melhor metodologia a ser utilizada em uma determinada aplicação pode ser influenciado por uma série de fatores como a disponibilidade de dados a importância do ruído dos dados o uso pretendido para os resultados A SFA tenta explicar o ruído enquanto a DEA assume que ele não existe Segundo Coelli et al 2003 se o pesquisador suspeitar que seus dados são de baixa qualidade poderá decidir se afastar da técnica DEA Segundo Coelli et al 2003 o método paramétrico da SFA reside na possibilidade de realizar inferências nas variáveis introduzidas tendo como objetivo estimar uma função para a fronteira eficiente O método DEA é uma técnica que avalia o posicionamento competitivo de organizações pois cria um ranking entre as empresas buscando identificar entre elas quais são eficientes e quais se mostram ineficientes Na DEA de acordo com Falcão e Correia 2012 quanto menor a amostra melhores serão as interpretações As duas técnicas apresentadas são metodologias distintas para construir fronteiras de eficiência uma considera técnicas de programação matemática e a outra usa técnicas econométricas A escolha entre a DEA e a SFA deverá ser feita pelo pesquisador considerando os dados disponíveis e os objetivos pretendidos por ele 25 O viés da DEA e a técnica de Bootstrap A DEA utiliza uma abordagem de Programação Linear para a estimação da fronteira Essa técnica não inclui o ruído ou erro aleatório pois não tem uma abordagem estatística Assaf e Matawie 2014 explicam que a DEA não é capaz de determinar a precisão das estimativas de eficiência ou fornecer uma base estatística para a fronteira estimada O principal problema dos estimadores DEA é que em tais estatísticas os estimadores de eficiência são avaliados em pontos de dados aleatórios retirados de uma população não observada em que não conhecemos o processo de geração de dados Conforme demonstrado por Wilson e Simar 1998 os teoremas usuais do limite central não são aplicáveis a menos que a dimensão do problema número de entradas e saídas seja pequeno Ou seja o problema vem do viés das estimativas da eficiência individual e o fato de que esse viés não desaparece exceto em casos envolvendo pequenas dimensões 34 A ferramenta de bootstrap foi introduzida na técnica DEA pelo trabalho de Simar e Wilson 1998 que forneceu uma metodologia geral de bootstrap em modelos de fronteiras não estocásticas Essa ferramenta foi introduzida por Efron 1979 e fornece base para a analisar a sensibilidade dos escores de eficiência à variação da amostragem Conforme a explicação de Bogetoft e Otto 2011 a ideia básica do bootstrap é por meio de reamostragem simular repetidamente o processo de geração de dados criando amostras do mesmo tamanho que a amostra original e a partir daí calcular as estatísticas necessárias chamadas de réplicas Esse processo gera uma amostra de réplicas que pode ser utilizada para tirar conclusões sobre a distribuição estatística em que se está interessado Neste caso estamos interessados na distribuição estatística dos estimadores de eficiência DEA O propósito desta abordagem segundo Assaf e Matawie 2014 é estimar a distribuição populacional dos escores de eficiência DEA fazendo ser possível realizar testes de hipóteses nas pontuações de eficiência Tziogkidis 2012 explicam que o estimador DEA possui um viés de amostragem que está associado ao fato de que a amostra observada é extraída de uma população subjacente não observada e que as pontuações de eficiência das DMUs na amostra dependem das DMUs que definem a fronteira Assim as pontuações de eficiência da DEA são superestimadas em comparação com a fronteira verdadeira da população com a exceção de que as DMUs que definem a fronteira populacional estejam todas incluídas na amostra Com base na demonstração realizada por Bogetoft e Otto 2011 se não houver erros de medição então todas as observações na amostra são da tecnologia 𝑇 𝑇 Temos que a estimativa DEA é tendenciosa para cima a eficiência estimada é maior ou igual a eficiência real pois estamos otimizando numa amostra menor que a população Para eliminar o viés ele precisa ser primeiramente estimado para logo obtermos uma estimativa corrigida pelo viés O viés pode ser estimado como demostrado a seguir 𝒗𝒊é𝒔𝒌 𝑬𝜽𝒌 𝜽𝒌 20 Onde 𝜃𝑘 A eficiência real baseada no conjunto de tecnologia não conhecido T 𝜃𝑘 A eficiência DEA estimada baseada na amostra 𝑇 35 Como não conhecemos a distribuição de 𝜃𝑘 não é possível calcular o valor esperado de 𝜃𝑘 E como dito por Bogetoft e Otto 2011 é aqui que o bootstrap entra Considerando que 𝜃𝑘𝑏 é uma estimativa réplica bootstrap de 𝜃𝑘 a estimativa bootstrap do viés é dada por 𝒗𝒊é𝒔𝒌 𝟏 𝑩 𝜽𝒌𝒃 𝑩 𝒃𝟏 𝜽𝒌 𝜽𝒌 𝜽𝒌 21 Logo o estimador corrigido do viés para 𝜃𝑘 é expresso na equação a seguir 𝜽𝒌 𝜽𝒌 𝒗𝒊é𝒔𝒌 𝜽𝒌 𝜽𝒌 𝜽𝒌 𝟐𝜽𝒌 𝜽𝒌 22 A precisão das estimativas pode ser determinada com base na variância da estimativa de bootstrap 𝝈𝟐 𝟏 𝑩 𝜽𝒌𝒃 𝑩 𝒃𝟏 𝜽𝒌² 23 26 A convexidade na DEA A DEA é caracterizada por gerar um ranking de eficiência para todas as unidades produtivas tornando possível comparar uma DMU com todas as outras que compõem a amostra A DEA também permite que uma DMU eficiente localizada na fronteira sirva como benchmark para uma outra unidade ineficiente que produza a mesma quantidade de produtos com a utilização de uma quantidade maior de insumos em comparação com a unidade eficiente ou ainda que produza uma quantidade de produtos inferior com a utilização da mesma quantidade de insumos da DMU eficiente A convexidade do conjunto de produção é fundamental para que a DEA possa cumprir todas as comparações e projeções que um ranking deve permitir A fronteira eficiente é formada por segmentos de reta que conectam todas as DMUs eficientes assim as combinações 36 convexas dos inputs e outputs de uma DMU eficiente será capaz de gerar uma outro ponto igualmente eficiente localizado na fronteira Desta forma a convexidade é necessária para que um ponto ineficiente tenha um ponto virtual na fronteira eficiente que ela seja capaz de alcançar promovendo a redução de seus inputs mantendo os outputs constantes Logo cada DMU tem uma referência de projeção que pode ser uma outra DMU eficiente ou um ponto virtual constituído por uma combinação convexa de outro ponto eficiente Dois testes desenvolvidos na segunda metade da última década apresentamse como opções para se testar a hipótese de convexidade do conjunto de produção Kneip et al 2015 mostram que os teoremas do limite central tradicionais não se aplicam para a realização de inferência de valores médios das técnicas não paramétricas DEA e FDH Free Disposal Hull FDH devido ao fato de que o viés dos escores individuais é de ordem superior à variância ou covariância Os autores desenvolveram novos teoremas do limite central para permitir a realização de inferências na DEA e na FDH O primeiro teste proposto por Kneip et al 2016 baseado nos resultados obtidos em Kneip et al 2015 testa a hipótese nula de convexidade contra a não convexidade do conjunto de produção O teste divide a amostra em duas subamostras aleatórias onde na primeira a convexidade é imposta e os retornos variáveis de escala são permitidos e na segunda subamostra a não convexidade da fronteira é permitida São comparadas a média da eficiência DEA da primeira subamostra com a média FDH da segunda Segundo Kneip et al 2016 na maioria dos trabalhos aplicados a escolha entre técnicas DEA e FDH é feita de forma arbitrária O teste permite que o investigador possa escolher entre a primeira que impõe convexidade no conjunto de produção e a segunda que não faz esta imposição O segundo teste para avaliar a convexidade do conjunto de produção é o teste de Simar e Wilson 2020 que utiliza o método de bootstrap para realizar múltiplas divisões de amostras e eliminar a sensibilidade que ocorre quando é realizada apenas uma divisão A maioria dos trabalhos empíricos que avaliam a eficiência do setor bancário brasileiro utilizam a técnica não paramétrica DEA Entretanto poucos realizam uma análise acerca da convexidade ou não do conjunto de produção Kneip et al 2016 afirmam que a habilidade de testar se o conjunto de produção é convexo ou não é crucial para realizar aplicações A flexibilidade da FDH de não impor qualquer restrição sobre o conjunto de produção gera regiões onde muitas firmas são comparáveis apenas com elas mesmas o que gera unidades que não são comparáveis com nenhuma outra mas mesmo assim são consideradas eficientes A técnica DEA é um estimador mais eficiente que o FDH mas somente se a tecnologia for convexa 37 Logo a importância de se realizar os testes está em avaliar o conjunto de produção a disposição para saber qual técnica poderá ser aplicada para avaliar a eficiência dos bancos brasileiros na atividade de intermediação financeira dado os inputs e outputs escolhidos neste trabalho 27 Testes para os retornos de escala O modelo CCR é limitado no sentido de permitir apenas retornos constantes de escala enquanto que o modelo BCC permite que sejam observados retornos crescentes e retornos decrescentes na tecnologia de produção Com o objetivo de utilizar o modelo que melhor represente o conjunto de produção os testes de Kneip et al 2016 e Simar e Wilson 2020 podem ser aplicados para testar a hipótese nula de retornos constantes de escala contra a hipótese de retornos variáveis de escala Os testes envolvem a divisão da amostra em duas subamostras onde na primeira retornos constantes de escala são impostos e na segunda retornos variáveis de escala são permitidos São comparadas as médias das eficiências DEA das duas subamostras que caso iguais apontam para a aplicação do modelo CCR e se diferentes apontam para a aplicação do modelo que permite retornos variáveis de escala o modelo BCC Os testes de retorno de escala têm o mesmo plano de fundo que os testes para a convexidade da tecnologia O teste de Kneip et al 2016 é baseado nos teoremas de limite central desenvolvidos em Kneip et al 2015 E o teste de Simar e Wilson 2020 utiliza o método de bootstrap para realizar múltiplas divisões da amostra É importante testar se os bancos na atividade de intermediação financeira possuem retornos variáveis ou constantes de escala Na década de 90 e no início do século XXI ocorreu uma forte tendência mundial no mercado financeiro de concentração das operações em um pequeno número de conglomerados financeiros em decorrência de operações de aquisições e fusões nestes mercados Faria et al 2006 Com isso é importante avaliar se estes movimentos e a concentração do setor proporciona economias de escala por essas instituições A aplicação dos testes neste trabalho permitirá averiguar se o conjunto de produção se apresenta como o esperado com retornos variáveis de escala validando estatisticamente a aplicação do modelo BCC ou se apresenta retornos constantes de escala diferentemente do esperável para a tecnologia do setor apontando para a aplicação do modelo CCR 38 28 Comparação de médias O objetivo principal deste trabalho envolve a comparação da média de grupos de bancos distintos Para alcançar este objetivo é necessário aplicar inferência estatística para realizar a comparação entre bancos tradicionais e bancos digitais Um teste de hipóteses pode nos dizer se a média entre os dois grupos de bancos são distintas estatisticamente permitindo averiguar se um grupo possui melhor desempenho que o outro Entretanto realizar inferência estatística sobre médias populacionais dos modelos DEA onde os estimadores são viesados não é uma tarefa simples É preciso lidar com os problemas derivados do viés descobertos por Kneip et al 2015 Os autores explicam que os estimadores de eficiência não paramétricos como a DEA têm viés que desaparece assintoticamente à medida que o tamanho da amostra aumenta mas apenas na taxa em que o estimador particular converge Conforme explicado por Kneip et al 2015 a menos que o número de entradas e saídas do modelo seja muito pequeno os resultados dos Teoremas do Limite Central padrões não valem para eficiência média estimados pelos estimadores não paramétricos Kneip et al 2015 para superar o problema gerado pelo viés dos estimadores individuais usaram uma média calculada sobre uma subamostra de dados O tamanho da subamostra é determinado para ajustar o viés e a variância com o objetivo de se obter uma distribuição estável e não degenerada Kneip et al 2015 Os autores desenvolveram uma Teorema do Limite Central priorizando o problema do viés da DEA que é a base dos testes amostrais desenvolvidos nos trabalhos de Kneip et al 2016 e Simar e Wilson 2020 Para realizar esta inferência o teste desenvolvido por Kneip et al 2016 será aplicado Este teste compara a eficiência média de dois grupos de DMUs distintos para testar a hipótese nula de equivalência de médias dos grupos 1 e 2 Este teste assim como os demais testes desenvolvidos em Kneip et al 2016 usa o teorema do limite central derivado de Kneip et al 2015 Existem hoje no mercado bancário brasileiro bancos de varejo de diferentes tamanhos Enquanto que ainda não existem bancos digitais de grande porte conforme a segmentação definida pelo BCB explicada no capítulo a seguir Este trabalho pretende comparar estes bancos digitais com dois grupos distintos de bancos de varejo O primeiro grupo inclui a 39 totalidade dos grupos de varejo enquanto que o segundo somente inclui as instituições com o mesmo porte dos bancos digitais Com isso pretendemos avaliar como estes novos bancos operam na atividade de intermediação financeira em comparação com a totalidade do mercado de varejo e verificar como o seu desempenho se posiciona em relação aos tradicionais de porte semelhante 29 Índice de Malmquist O Índice de Malmquist proposto originalmente por Malmquist 1953 e Caves et al 1982 permite a análise comparativa dos desempenhos de uma amostra de DMUs durante um período de tempo a partir da elaboração de um número índice Caves et al 1982 explicam que o Índice de Malmquist tem a capacidade de medir mudanças temporais de produtividade total dos fatores de produção considerando as alterações de eficiência e as mudanças tecnológicas relativas à amostra considerada Considerando que temos um período inicial 0 e um período t E0 t é a projeção dos insumos e produtos correspondentes ao período 0 na fronteira de eficiência construída no período t A alteração de eficiência AE medida a partir da modificação de desempenho no horizonte temporal considerado é dado pela equação a seguir 𝑨𝑬 𝑬𝒕 𝒕 𝑬𝟎 𝟎 24 A alteração tecnológica AT é obtida tal como expresso 𝑨𝑻 𝑬𝒕 𝟎𝑬𝟎 𝟎 𝑬𝒕 𝒕𝑬𝟎 𝒕 25 O índice Malmquist que representa a alteração da produtividade total dos fatores de produção em um determinado período de tempo 0 t é obtido pela multiplicação de AE e AT como expresso a seguir 40 𝑰𝑴 𝑨𝑻 𝑨𝑬 𝑬𝒕 𝟎𝑬𝟎 𝟎 𝑬𝒕 𝒕𝑬𝟎 𝒕 𝑿 𝑬𝒕 𝒕 𝑬𝟎 𝟎 26 𝑰𝑴 𝑬𝒕 𝟎𝑬𝟎 𝟎 𝑬𝒕 𝒕𝑬𝟎 𝒕 𝑿 𝑬𝒕 𝒕 𝑬𝟎 𝟎 𝟐 𝑬𝒕 𝟎𝑬𝒕 𝒕 𝑬𝟎 𝟎𝑬𝟎 𝒕 27 Para calcular as projeções dos períodos 0 t 0 0 t 0 t t utilizadas para se obter os índices de alteração de eficiência alteração de tecnologia e o Índice de Malmquist é utilizado o seguinte problema de programação linear onde E é o inverso de 𝑑0 𝒅𝟎 𝒑𝒙𝒒 𝒚𝒒𝟏 𝒎á𝒙𝝓𝝀 𝝓 28 sa 𝝓𝒚𝒊𝒒 𝒀𝒑𝝀𝒊 𝟎 𝒙𝒊𝒒 𝑿𝒑𝝀𝒊 𝟎 𝝀𝟏 𝝀𝒏 𝟎 29 Em que p q 𝜖 t t t1 t1 t t1 t1 t Em um modelo orientado para inputs o Índice de Malmquist entre os períodos t e t1 pode ser M 1 A produtividade aumentou M 1 A produtividade ficou constante M 1 A produtividade diminuiu Em um modelo orientado para outputs ocorre o inverso Quando o Índice é menor que a unidade produtividade diminuiu e quando o Índice é maior que unidade significa que a produtividade aumentou durante o período analisado Esta pesquisa irá utilizar as fronteiras de eficiência construídas nos extremos dos períodos analisados para mensurar as alterações de produtividade total dos fatores de produção incorridas pelos bancos no decorrer do período estudado Também serão descritas as fontes dessas alterações sendo alterações de eficiência ou alterações tecnológicas 41 3 RESULTADOS 31 Modelo proposto A eficiência das instituições bancárias será avaliada com enfoque na principal atividade exercida por elas Berger e Humphrey 1997 explicaram que os bancos exercem o papel de instituição organizadora de fundos entre poupadores e investidores fazendo com que valores poupados dos agentes superavitários cheguem aos agentes que demandam destes recursos para desenvolver seus projetos A abordagem da intermediação financeira é a mais adotada para a avaliação de desempenho dessas instituições pois é a que melhor representa as atividades que os bancos realizam De acordo com Mccaing e Stengos 2005 esta abordagem capta as atividades financeiras que possuem impacto nas economias nacionais O modelo de eficiência que será implementado neste trabalho vai de acordo com o modelo desenvolvido por Sealey e Lindley 1977 em que a instituição financeira é uma Decision Making Unit DMU que utiliza dos fundos captados e dos usos de capital e trabalho para produzir ativos remunerados Estes autores concluem que o output apropriado do ponto de vista do processo de decisão da firma financeira é o serviço provido para os agentes deficitários A partir da abordagem da intermediação financeira o modelo deste trabalho foi construído utilizando como primeiro input as captações de recursos realizadas pelos bancos que são compostas de depósitos totais depósitos à vista depósitos de poupança depósitos interfinanceiros depósitos a prazo conta de pagamento prépaga e outros depósitos recursos de aceites e emissões de títulos e por fim obrigações por empréstimos e repasses e por operações compromissadas As captações foram obtidas da conta de passivo do balanço dos bancos Para representar o capital empregado neste processo foi utilizado como input o ativo permanente ajustado que é o ativo permanente dos bancos menos o ativo imobilizado destinado para operações de arrendamento mercantil As despesas de pessoal constituem o último input do modelo obtida da demonstração de resultados dos bancos e representam o emprego de trabalho que é utilizado neste processo de transformação das captações em crédito para os agentes deficitários 42 As operações de crédito constituem o output único do modelo representando o resultado do processo descrito por Sealey e Lindley 1977 sendo o serviço fornecido pelas instituições financeiras para os agentes que demandam recursos Foram utilizadas as operações líquidas de provisões retirando assim as estimativas de perdas prováveis que o banco terá com devedores inadimplentes Tabela 3 Inputs e outputs do modelo de proposto Inputs I Captações Depósitos totais recursos de aceites e emissões de títulos obrigações por empréstimos e repasses obrigações por operações compromissadas II Trabalho Despesas de pessoal III Capital Ativo permanente imobilizado de arrendamento Output I Crédito Operações de crédito líquidas de provisão Fonte O autor 2023 A definição dos inputs nos modelos de intermediação financeira presentes na literatura pouco sofre variação São escolhidas variáveis que representem os usos de capital trabalho e depósitos arrecadados em quase todos os trabalhos A escolha realizada neste trabalho de utilizar o modelo de intermediação financeira puro sem a inclusão de outros ativos dos bancos como aplicações interfinanceiras de liquidez e títulos e valores mobiliários por exemplo se deve ao fato de que queremos avaliar os bancos no desempenho da atividade de concessão de crédito exclusivamente Isso deve ao fato de que esta atividade como foi visto é a principal exercida pelos bancos e todas as demais derivam dela e é a atividade com a maior capacidade de ter impactos na economia 43 32 A amostra A coleta de dados deste trabalho baseouse nas informações divulgadas pelo Banco Central do Brasil BCB das instituições financeiras que autoriza a funcionar e que estejam em operação O BCB disponibiliza trimestralmente relatórios constituídos de informações contábeis selecionadas destas instituições de acordo com o Padrão Contábil das Instituições Reguladas pelo Banco Central do Brasil Cosif Os relatórios podem ser obtidos através do portal IFData Apesar de as Fintechs apresentaremse no mercado brasileiro sob diversas formas de constituições instituições de pagamento sociedades de crédito direto etc neste trabalho buscouse trabalhar apenas com bancos comerciais bancos múltiplos que contenham a carteira comercial e caixas econômicas Tal escolhe se deve ao fato de que apenas estas instituições autorizadas a receber depósitos à vista do público são capazes de criar moeda escritural e possuem uma dinâmica particular de difícil comparação com outras instituições financeiras Para a construção da amostra foram selecionadas as instituições de maneira consolidada onde as instituições pertencentes a um mesmo conglomerado financeiro são apresentadas como se representassem uma única entidade Foram excluídos da amostra os bancos sem dados disponíveis bancos que não possuíam operações de crédito e por fim os bancos que não estavam em operação durante o intervalo de 3 anos definidos para a análise que compreende os anos de 2019 2020 e 2021 A amostra ficou constituída de 73 instituições bancárias sendo 9 bancos de controle público BB BRB Banrisul Banestes Banpará Banese Banco da Amazônia Banco do Nordeste do Brasil CEF 34 de controle privado nacional e 30 instituições de controle privado estrangeiro Para atingir os objetivos propostos neste trabalho os bancos foram divididos conforme a sua área de atuação podendo ser identificados 14 bancos de varejo e 5 bancos digitais Agibank C6 Bank BS2 Inter e Original Foram elaboradas matrizes de inputs e output com dados trimestrais 1º e 3º trimestres para o período compreendido entre o 1º trimestre de 2019 e 3º trimestre de 2021 último relatório trimestral disponível a época de elaboração deste trabalho A fim de tornar os dados comparáveis ao longo do período os valores foram atualizados para setembro de 2021 corrigidos pelo índice IGPDI da Fundação Getúlio Vargas FGV 44 Tabela 4 Estatísticas descritivas dos inputs e output 20192021 2019 2020 2021 1º T 3º T 1º T 3º T 1º T 3º T Captações Mín 2255 5501 8356 25012 10262 12584 Média 112122161 110902324 116304954 117124465 103370298 101347113 Median a 7164467 7513996 7617505 8614391 8010199 7767950 DP 361819196 352708625 364496190 362904728 316415706 308689288 Máx 179021948 3 170035563 6 172512485 0 170000797 6 151986689 9 155274888 6 Despesas de Pessoal Mín 546 585 608 278 250 227 Média 526739 561758 506637 456285 403756 386009 Median a 38629 41321 39697 32564 33578 30942 DP 1680637 1788337 1594519 1428698 1274718 1174353 Máx 8468064 8299113 7732324 6829318 6507198 6000818 Permanente Mín 425 167 159 141 122 114 Média 5865851 5975737 5764752 5219187 4519680 4221997 Median a 122026 127217 123876 132887 171427 179970 DP 24211560 24785450 23458973 21109324 17844795 16169585 Máx 156314940 158669017 155346711 131604007 113043317 99067117 Operações de Crédito Mín 36 6095 8472 4513 536 104 Média 51163803 51758849 53243754 50817065 46787852 46821505 Median a 2329947 2360252 2724170 2923333 2764280 3234986 DP 173005129 172699425 174591301 165276376 151067118 149474464 Máx 955053080 935523344 914826941 882294254 806209307 795546705 Fonte O autor 2023 Em R mil 45 33 Convexidade e retornos de escala A fronteira eficiente na DEA é formada por segmentos de reta que conectam todas as DMUs eficientes assim as combinações convexas desses pontos também estarão na fronteira Logo combinações convexas dos inputs e outputs de uma DMU eficiente será capaz de gerar uma outra unidade que estará na fronteira sendo igualmente eficiente A convexidade do conjunto de produção é necessária para que uma DMU ineficiente que promove a redução de seus inputs mantendo os outputs constantes seja capaz de alcançar um ponto na fronteira eficiente Com o objetivo de testar a hipótese de convexidade do conjunto de produção foram utilizados os testes de Kneip et al 2016 e Simar e Wilson 2020 implementados através do Software R Os testes realizam uma divisão aleatória da amostra em duas subamostras independentes em que são comparadas a média das estimativas de eficiência DEA da primeira subamostra onde a convexidade é imposta e retornos variáveis de escala permitidos com a média das eficiências Free Disposal Hull FDH da segunda subamostra permitindo a não convexidade da fronteira O teste de Kneip et al 2016 trabalha com apenas uma divisão aleatória da amostra enquanto que o teste de Simar e Wilson 2020 aplica o método de bootstrap permitindo o uso de múltiplas divisões de amostra Os testes foram aplicados para o conjunto de produção da amostra em todos os seus recortes dentro do período de análise Em todos os trimestres analisados a hipótese nula não foi rejeitada ao nível de 5 de significância apontando para a convexidade da amostra Tabela 5 Testes de convexidade da amostra 2019 2020 2021 1º T 3º T 1º T 3º T 1º T 3º T Kneip et al 2016 Estatística 1001 0976 0382 0149 0395 0363 pvalor 0839 0882 0595 0235 0736 0163 Simar e Wilson 2020 Estatística 0330 0347 0188 0253 0190 0271 pvalor 0301 0196 0751 0452 0712 0398 Fonte O autor 2023 46 Um modelo DEA pode ser construído considerando retornos constantes ou retornos variáveis de escala O modelo CCR desenvolvido por Charnes Cooper e Rhodes 1978 é utilizado quando a tecnologia de produção apresenta retornos constantes de escala enquanto que o modelo BCC desenvolvido por Banker Charnes e Cooper 1984 é adequado para estimar eficiências em tecnologias que apresentam retornos variáveis de escala Os testes para avaliar os retornos de escala de uma tecnologia são similares aos testes de convexidade Para testar a hipótese nula de retornos constantes de escala a amostra é subdividida em duas subamostras independentes onde é feita a comparação entre a média das eficiências DEA da primeira subamostra onde retornos constantes são impostos contra a média das eficiências DEA da segunda subamostra onde são permitidos retornos variáveis de escala O teste de Kneip et al 2016 realiza apenas uma divisão da amostra enquanto que o teste de Simar e Wilson 2020 permite múltiplas divisões da amostra utilizando o método de Bootstrap Tabela 6 Testes de retornos de escala da amostra 2019 2020 2021 1º T 3º T 1º T 3º T 1º T 3º T Kneip et al 2016 Estatística 1448 1605 1752 4055 2143 1144 pvalor 0046 0036 0033 0006 0011 0039 Simar e Wilson 2020 Estatística 0865 0650 0683 1001 0761 0715 pvalor 0012 0037 0035 0005 0009 0017 Fonte O autor 2023 Ao nível de 5 de significância em todos os trimestres do período selecionado a hipótese nula de retornos constantes de escala foi rejeita o que indica que os bancos possuem em relação a atividade de intermediação financeira uma tecnologia que apresenta retornos variáveis de escala 47 34 Desempenho do setor bancário brasileiro em 2021 Para avaliar a eficiência dos bancos comerciais brasileiros e alcançar o objetivo proposto por este trabalho de comparar o desempenho dos bancos comerciais digitais com os bancos comerciais do segmento de varejo a técnica DEA foi aplicada para medir o desempenho das organizações As estimativas foram obtidas pelo programa R com uso do pacote Benchmarking mesmo programa utilizado para realizar os testes de convexidade e retornos de escala com o uso do pacote FEAR Foram calculadas as eficiências para todo o período de análise com a técnica DEA utilizandose do modelo BCC orientado para inputs que permite retornos variáveis de escala o que vai de acordo com os resultados obtidos nos testes realizados A técnica de Bootstrap foi aplicada para estimar as eficiências sem o viés existente no estimador de eficiência da DEA Os resultados para as instituições bancárias da amostra no 3º trimestre de 2021 são apresentadas na Tabela 7 a seguir Tabela 7 Escores de eficiência dos bancos no 3º trimestre de 2021 continua Bancos 5 Eficiência 95 ABCBRASIL 04705 05499 07178 AGIBANK 04558 05322 06511 ALFA 04252 04896 05732 ANDBANK 00798 00959 01354 BANCO ARBI 01158 01380 01757 BANCO CAPITAL 05223 06318 09248 BANCO CARGILL 05165 06253 09287 BANCO CEDULA 05360 06384 09474 BANCO COOPERATIVO SICREDI 05708 06581 09332 BANCO DA AMAZONIA 07022 08028 09404 BANCO DA CHINA BRASIL 02786 03234 04221 BANCO DO ESTADO DE SERGIPE 02193 02473 02931 BANCO DO ESTADO DO PARÁ 04310 05053 05967 BANCO DO NORDESTE DO BRASIL 05228 06186 07546 BANCO FIBRA 02149 02525 03140 BANCO INBURSA 03629 04164 05260 BANCO JOHN DEERE 05427 06436 09317 BANCO KEB HANA DO BRASIL 02798 03266 04230 BANCO LUSO BRASILEIRO 04863 05560 06343 BANCO MIZUHO DO BRASIL 00809 00946 01267 BANCO MUFG BRASIL 01814 02126 02497 BANCO RABOBANK 05194 06303 09166 48 Tabela 7 Escores de eficiência dos bancos no 3º trimestre de 2021 continuação BANCO RIBEIRAO PRETO 03026 03489 04239 BANCO RODOBENS 03996 04618 05601 BANCO SEMEAR 04549 05328 06472 BANCO SUMITOMO MITSUI 02345 02838 03625 BANCO TOPÁZIO 00681 00809 01039 BANCO TRIANGULO 01615 01876 02424 BANCOOB 01327 01598 02186 BANESTES 01015 01176 01474 BANRISUL 03852 04606 05576 BB 04529 05646 07724 BMG 04217 04912 05986 BNP PARIBAS 01912 02263 02747 BOCOM 03961 04610 05457 BOFA MERRILL LYNCH 00370 00441 00533 BRADESCO 04033 04920 06452 BRB 06750 07965 09447 BTG PACTUAL 04268 05090 06681 C6 BANK 03835 04458 05159 CEF 05150 06280 09209 CAIXA GERAL 01490 01819 02287 CCB 03148 03884 05284 CITIBANK 01161 01387 01857 COMMERZBANK BRASIL 04137 04796 05596 CREDIT AGRICOLE 01190 01409 01738 CREDIT SUISSE 03106 03566 04077 CREFISA 05422 06290 09534 DAYCOVAL 05017 05885 07392 DEUTSCHE BANK 01057 01250 01517 FATOR 00446 00547 00735 GRUPO BONSUCESSO BS2 00444 00509 00599 ICBC DO BRASIL 02950 03514 04420 INDUSTRIAL DO BRASIL 03477 04012 04683 ING 00484 00579 00741 INTER 03140 03613 04162 INTESA SANPAOLO BRASIL 03468 04132 05362 ITAU 05600 07043 09502 JP MORGAN CHASE 00285 00344 00449 MERCANTIL DO BRASIL 05075 05921 06803 MODAL 00654 00747 00942 NOVO BANCO CONTINENTAL 05829 06861 08223 OMNI 04768 05593 06573 ORIGINAL 02961 03442 03922 PARANÁ BANCO 04618 05393 06533 PINE 01561 01809 0236 RENDIMENTO 01423 01603 01920 49 Tabela 7 Escores de eficiência dos bancos no 3º trimestre de 2021 conclusão SAFRA 03903 04675 05915 SANTANDER 05706 06775 09527 SCOTIABANK 01729 02049 02440 SOCIETE GENERALE 01297 01527 01858 SOFISA 03934 04636 05673 VOTORANTIM 05680 06839 08668 Fonte O autor 2023 As eficiências foram calculadas com para inputs pois o objetivo é verificar a performance dos bancos no ajuste de captações despesas de pessoal e ativo permanente para obter maior eficiência na intermediação financeira Segundo Fethi e Pasiouras 2010 a orientação para inputs é de longe a mais utilizada para estimar a eficiência dos bancos na abordagem da intermediação financeira pois é assumido que os gestores tem maior controle sobre os recursos empregados no processo do que sobre os produtos Os dados estão expostos na medida de DebreuFarrel para inputs onde o valor estimado para a eficiência é o valor pelo qual podemos multiplicar todos os insumos para produzir os mesmos valores de output crédito de maneira eficiente O escore de eficiência corrigido pelo método bootstrap é contínuo assim a probabilidade de uma DMU ter escore de eficiência igual a unidade é nula Os três bancos que se mostraram mais eficientes foram o Banco da Amazônia o Banco de Brasília BRB e o Itaú com os respectivos escores de eficiência 08028 07965 e 07043 Tabela 8 Sumário estatístico das eficiências no 3º trimestre de 2021 Eficiência Mín 00344 Média 03907 Mediana 04164 DP 02145 Máx 08028 Fonte O autor 2023 O banco com a menor eficiência apurada no período o banco JP Morgan Chase com escore de eficiência de 00344 é um banco que atua nas áreas de investment bank corporate bank asset management private banking e treasury and securities services A instituição é 50 focada em oferecer serviços financeiros para corporações governos e instituições não tendo como a concessão de crédito para seus clientes Assim o JP Morgan Chase na intermediação financeira é o banco mais ineficiente atuando no mercado brasileiro atualmente Bancos tem diferentes áreas de atuação Analisar a eficiência de todos os bancos juntos pode ser um erro porque estariam sendo agrupados juntos bancos com tipos de operações diferentes Além disso o foco principal deste trabalho é realizar a análise comparativa dos bancos digitais com os bancos de varejo tendo em vista que ambos os tipos de bancos buscam oferecer serviços financeiros para pessoas físicas em geral Tabela 9 Eficiência dos bancos digitais no 3º trimestre de 2021 Porte 5 Eficiência 95 Agibank S4 04558 05322 07178 C6 Bank S4 03835 04458 05159 BS2 S3 00444 00509 00599 Inter S3 03140 03613 04162 Original S3 02961 03442 03922 Média 03469 DP 01816 Fonte O autor 2023 Os bancos digitais como um todo operaram de maneira ineficiente durante o terceiro trimestre de 2021 com destaque para o banco BS2 que para se tornar eficiente deveria produzir a mesma quantidade de crédito com aproximadamente 5 dos insumos que utiliza atualmente A média dos bancos digitais ficou em 03469 indicando a ineficiência elevada dos bancos digitais na intermediação financeira O BCB segmenta as instituições financeiras que autoriza a funcionar conforme o porte que cada uma apresenta Como medida do porte das instituições a autoridade monetária considera a razão entre o valor da exposição total da instituição e o valor do PIB do Brasil O Segmento 1 S1 é composto pelos bancos que tenham porte igual ou superior a 10 do Produto Interno Bruto PIB O Segmento 2 S2 compreende as instituições com porte inferior a 10 e igual ou superior a 1 do PIB As instituições com porte inferior a 1 do PIB e igual ou superior a 01 do PIB integram o Segmento 3 S3 Por fim o Segmento 4 S4 é composto pelas instituições de porte inferior a 01 51 Tabela 10 Segmentação das instituições financeiras conforme o BCB Segmento Porte S1 10 do PIB S2 1 e 10 do PIB S3 01 e 1 do PIB S4 01 do PIB Fonte O autor 2023 Os bancos que apresentaram os maiores escores de eficiência entre os bancos digitais foram os bancos Agibank e C6bank bancos atualmente enquadrados segundo seu porte no segmento S4 Os bancos digitais que possuem porte um segmento acima dos dois primeiros Inter BS2 Original apresentam escores de eficiência menores Tabela 11 Eficiência dos bancos de varejo no 3º trimestre de 2021 Bancos Segmento Controle 5 Eficiência 95 BB S1 Público 04529 05646 07724 Bradesco S1 Privado Nacional 04033 04920 06452 CEF S1 Público 05150 06280 09209 Itaú S1 Privado Nacional 05600 07043 09502 Santander S1 Privado Estrangeiro 05706 06775 09527 Banco do Nordeste S2 Público 05228 06186 07546 Banrisul S2 Público 03852 04606 05576 Safra S2 Privado Nacional 03903 04675 05915 Banco da Amazônia S3 Público 07022 08028 09404 Banpará S3 Público 04310 05053 05967 Banestes S3 Público 01015 01176 01474 BRB S3 Público 06750 07965 09447 Mercantil do Brasil S3 Privado Nacional 05075 05921 06803 Banese S4 Público 02193 02473 02931 Média 05482 DP 01844 Fonte O autor 2023 Os três bancos mais eficientes de toda a amostra são bancos de varejo Banco da Amazônia BRB e Itaú Os dois primeiros são bancos de varejo regionais de controle público enquanto o Itaú é um banco de varejo nacional com controle privado nacional Os bancos de 52 varejo apresentaram média de 05482 de eficiência o que indica que para se tornarem eficientes deveriam utilizar em média 5482 dos insumos que utilizam hoje para produzir o mesmo nível de crédito Quando se considera apenas os bancos dos segmentos S1 e S2 o valor da média dos escores dos bancos de varejo aumenta para 05766 enquanto que considerando apenas os segmentos S3 e S4 a média das eficiências é de 05102 puxada para baixo pelo Banese e pelo Banestes que apresentaram as duas menores eficiências dos bancos de varejo respectivamente 02473 e 01176 34 Evolução da eficiência dos bancos digitais e bancos de varejo entre 20192021 Esta seção tem por objetivo demonstrar o que aconteceu com as posições relativas dos bancos digitais e dos bancos de varejo ao longo do período selecionado Os escores individuais obtidos pelos modelos DEA de cada ano não são passíveis de serem comparados entre si Entretanto é interessante ver o que aconteceu com a posição relativa de cada instituição do setor ao longo do triênio A tabela a seguir apresenta um sumário da eficiência dos bancos entre os anos de 2019 e 2021 Tabela 12 Sumário da eficiência dos bancos no triênio 20192021 1T 2019 3T 2019 1T 2020 3T 2020 1T 2021 3T 2021 Mín 00213 00114 00151 00128 00241 00343 Média 03142 03181 03192 03415 03657 03901 Mediana 02721 02836 02847 03006 03768 04194 DP 01959 01984 01990 01984 02025 02140 Máx 07013 06958 06809 07581 07033 07991 CV 6234 6237 6234 5809 5537 5485 Fonte O autor 2023 A partir da Tabela 12 é possível observar que o coeficiente de variação CV dos escores de eficiência dos bancos diminuiu gradativamente ao longo do período analisado O coeficiente de variação do primeiro trimestre de 2019 foi de 6234 e ao fim intervalo no terceiro trimestre de 2021 o coeficiente de variação era de 5485 apresentando uma redução de aproximadamente 8 53 Tabela 13 Eficiência dos bancos digitais no triênio 20192021 1T 2019 3T 2019 1T 2020 3T 2020 1T 2021 3T 2021 Agibank 01979 02590 02401 03415 04027 05325 C6 Bank 00213 00114 00151 01134 04898 04426 BS2 00544 0078 00540 00526 00435 00506 Inter 03309 02596 03035 02828 03484 03588 Original 01973 02163 02847 02725 03294 03412 Média 01603 01649 01795 02126 03227 03451 CV 7788 6883 7520 5787 5206 5253 Fonte O autor 2023 A partir da tabela 13 observase que o coeficiente de variação reduziu ao longo dos anos mais de 20 demonstrando que houve uma aproximação do desempenho dos bancos digitais ao longo do período Um ponto relevante é que os escores de eficiência de todos os bancos calculados pelo modelo DEABCC com orientação para inputs corrigido por bootstrap apresentaram valores dentro do intervalo de confiança de 95 em todos os períodos da amostra Tabela 14 Eficiência dos bancos de varejo no triênio 20192021 1T 2019 3T 2019 1T 2020 3T 2020 1T 2021 3 T 2021 Santander 0679286 0663284 0652354 0658722 0670965 06775 BB 06157 0506604 0497904 0502561 0460595 05646 CEF 0607291 0594171 0594228 0604319 0623342 0628 Itaú 0557846 0485505 0581137 0585735 0565432 07043 BRB 0504778 0536556 0680954 075814 0703383 07965 Safra 0468496 0414131 0485886 0558196 0511322 04675 Banrisul 0465373 0462167 0513722 0503145 0471492 04606 Bradesco 0446083 0483121 0463643 0463077 0478732 0492 Mercantil do Brasil 0397718 0327467 0373468 0431105 0595035 05921 Banpará 0370831 0357785 046786 0424338 0494943 05053 Banco do Nordeste 0338431 0349436 0395456 0360905 0446095 06186 Banese 0193963 0184841 0208032 0206879 0226816 02473 Banco da Amazônia 0182649 0208964 0281441 0397964 0568823 08028 Banestes 0074688 0091347 0080148 0078686 0107175 01176 Média 0421652 040467 0448302 0466698 0494582 0548193 CV 4049077 3866078 3640876 3635215 3142862 3364399 Fonte O autor 2023 54 35 Comparação entre bancos digitais e bancos de varejo tradicionais Com o objetivo de comparar as eficiências médias dos grupos de bancos digitais e bancos de varejo será implementado o teste desenvolvido por Kneip et al 2016 Os autores explicam que o teste para a eficiência média de dois grupos foi sugerido no trabalho de Charnes et al 1981 onde foi considerado dois grupos de escolas um recebendo um tratamento específico enquanto o outro grupo não recebia Utilizandose dos resultados obtidos em Kneip et al 2015 onde foi desenvolvido um teorema do limite central específico para realizar inferências sobre eficiências médias Kneip et al 2016 apresentaram um teste para dois grupos distintos de DMUs onde a hipótese nula de médias iguais é avaliada Foram realizados os testes considerando apenas os grupos dos bancos digitais e dos bancos de varejos Os bancos digitais que se encontram em atividade no mercado brasileiro estão classificados nos segmentos S3 e S4 enquanto que existem bancos de varejo em todos os segmentos Devido a isso dois testes serão realizados O primeiro teste considerará todos os bancos de varejo em atuação comparando a média de eficiências da totalidade dos bancos digitais com a média de eficiências da totalidade dos bancos de varejo Um segundo teste será realizado em que os bancos digitais serão comparados apenas com os bancos de varejo em atuação no mesmo segmento S3 e S4 Portanto foram feitos dois testes comparando os bancos digitais com dois grupos diferentes dos bancos de varejo com e sem os maiores bancos do setor O primeiro teste de comparação dos escores médios de eficiência dos bancos digitais com os escores médios de todos os bancos de varejo foi realizado considerando retornos variáveis de escala e também que a fronteira é a mesma para os dois grupos Os resultados obtidos neste teste estão resumidos na tabela a seguir Tabela 15 Teste 1 de equivalência de médias de Kneip et al 2016 Grupo 1 bancos digitais Grupo 2 bancos de varejo Eficiência média 0797 0872 Variância 0037 0024 Viés 0617 0229 Estatística 2343 pvalor 0019 bilateral 0010 unilateral 1 0990 unilateral 2 Fonte O autor 2023 55 Notase que as eficiências médias calculadas no teste diferem das eficiências médias calculadas e apresentadas anteriormente para este mesmo período 3º T de 2021 Isso se deve ao fato de que o teste considera apenas os dois grupos que foram selecionados da amostra primária para a estimação das eficiências enquanto que o modelo deste trabalho emprega a técnica DEA para uma amostra de 73 bancos comerciais A eficiência dos bancos de varejo apresentada é superior ao do grupo dos bancos digitais enquanto a variância é um pouco menor São apresentados pvalores distintos para os 3 testes que são realizados O primeiro teste é bilateral que tem como hipótese nula a igualdade das médias dos dois grupos Essa hipótese foi rejeita ao nível de significância de 5 indicando a diferença entre as médias de eficiência O teste seguinte é unilateral e tem como hipótese nula que a média do primeiro grupo é superior à média do segundo Esta hipótese foi rejeitada O segundo teste unilateral avalia a hipótese nula de a média do grupo 2 ser maior que a média do grupo 1 Esta hipótese foi aceita apontando para um desempenho médio superior dos bancos de varejo sobre os bancos digitais na intermediação financeira no terceiro trimestre de 2021 Tabela 16 Teste 2 de equivalência de médias de Kneip et al 2016 Grupo 1 bancos digitais Grupo 2 bancos de varejo S3 e S4 Eficiência média 0870 0906 Variância 0021 0019 Viés 0390 0282 Estatística 0398 pvalor 0690 bilateral 0345 unilateral 1 0655 unilateral 2 Fonte O autor 2023 Considerando que os bancos digitais que atuam no Brasil atualmente possuem porte menor do que 1 do PIB enquanto existem cinco bancos de varejo com porte maior ou igual a 10 foi realizada uma comparação dos escores médios de eficiência dos grupos considerando apenas aquelas instituições que se encontram nos segmentos S3 e S4 definidos pelo BCB A eficiência média estimada para o grupo 1 foi de 0870 próxima a média grupo 2 de 0906 O teste bilateral não rejeita a hipótese de igualdade dos escores médios Os resultados contrários dos testes unilaterais apontam também para essa igualdade Logo bancos digitais e bancos de varejo de médio e pequeno porte não apresentaram diferenças estatisticamente relevantes no seu desempenho médio durante o 3º trimestre de 2021 56 A segmentação do BCB utilizada para definir quais bancos são grandes e quais são pequenos é exógena ao modelo Para fazer essa divisão deveríamos olhar para a eficiência de escala dos bancos observando em qual parte da fronteira eficiente eles se situam Ou seja se apresentam retornos crescentes de escala retornos constantes ou retornos não crescentes segmentando os bancos de acordo com os dados calculados Porém uma limitação verificada neste trabalho é de que não temos disponível um pacote para o programa R que possibilite realizar testes de hipótese para testar a eficiência de escala dos grupos de bancos Portanto a segmentação do BCB de acordo com o porte das instituições foi uma alternativa de fora do modelo para classificar os bancos pelo seu tamanho A partir dos resultados apresentados depreendese que os bancos digitais em média apresentam resultados inferiores em termos de eficiência quando comparados com os bancos de varejo considerando tanto os grandes bancos quanto os bancos regionais Porém quando são considerados apenas os bancos de varejo de porte similar aos bancos digitais hoje em atuação os resultados não apontam diferenças entre o desempenho destes grupos 36 Análise da eficiência de escala As tabelas 17 e 18 mostram os resultados da eficiência de escala para os bancos digitais e para os bancos varejistas tradicionais respectivamente A eficiência de escala é obtida pela razão entre as eficiências dos modelos CCR e BCC ou seja entre as eficiências técnicas de longoprazo e de curtoprazo A ineficiência de escala é a ineficiência que advém de uma unidade produtiva estar operando fora do nível de produto com retornos constantes de escala 57 Tabela 17 Eficiência de escala dos bancos digitais no 3º trimestre de 2021 Segmento Eficiência de escala Agibank S4 07056 C6 Bank S4 06581 BS2 S3 06800 Inter S3 05644 Original S3 05029 Média 06222 DP 00853 Fonte O autor 2023 Entre os bancos digitais o que apresentou o maior nível de eficiência de escala foi o banco Agibank sendo este o que opera mais próximo da região que apresenta retornos constantes de escala Em média os bancos digitais apresentam uma eficiência de escala de 06222 Tabela 18 Eficiência de escala dos bancos de varejo no 3º trimestre de 2021 Segmento Eficiência de escala BB S1 03403 Bradesco S1 03665 CEF S1 05509 Itaú S1 03098 Santander S1 03711 Banco do Nordeste S2 06379 Banrisul S2 03729 Safra S2 03846 Banco da Amazônia S3 06617 Banpará S3 07433 Banestes S3 08064 BRB S3 04849 Mercantil do Brasil S3 05269 Banese S4 08502 Média 05291 DP 01835 Fonte O autor 2023 58 Pela Tabela 18 é possível observar que os bancos de varejo dos segmentos S3 e S4 são os que mais estão próximos de retornos constantes de escala com destaque para os bancos Banese e Banestes com eficiências de escala de 08502 e 08064 Já entre os grandes bancos varejistas presentes nos segmentos S1 e S2 observase altos níveis de ineficiência de escala com destaque para os bancos BB Bradesco Itaú e Santander em que os escores de eficiência de escala não ultrapassam o valor de 04 Os bancos varejistas médios são os que em média 06768 apresentam eficiência de escala maiores estando mais próximos de retornos constantes de escala Os bancos digitais operam em um nível abaixo de retornos constantes com eficiência de escala em média de 06222 o que significa que estes bancos podem se tornar mais eficientes em escala e se aproximar dos bancos de varejo médios Por outro lado os grandes bancos do setor apresentam em média 04167 a maior distância para os retornos constantes de escala e por serem os pontos mais afastados em relação a origem possuem retornos não crescentes de escala Os grandes bancos em comparação com os bancos digitais os mais próximos da origem possuem inclinação inferior Em resumo bancos digitais são pequenos demais e precisariam crescer para se tornarem mais eficientes em escala aproximandoos dos bancos de varejo médios A DEA resolve o problema e calcula as eficiências considerando que as DMUs são livres para escolher o seu nível de produção sem competirem entre si ou seja não leva em consideração a organização do mercado É preciso saber se existem condições de mercado para que os bancos digitais cresçam A experiência simplificada e de baixo custo permitiu aos bancos digitais crescerem de forma acelerada nos primeiros anos de vida Bancos digitais operam com uma estrutura menor pois não contam com as estruturas físicas das agências operam com um número menor de funcionários o que permite atender os clientes a um custo mais baixo No entanto pode ser que ainda falte aos bancos digitais a escala necessária para diluir sua estrutura de custo tornando se assim mais eficientes A ascensão dos bancos digitais no mercado de crédito é viabilizada pela capacidade que estes têm de reduzir as barreiras de entrada através da inovação A prática dos grandes bancos de preços elevados e rentabilidade elevada também é um incentivo a participação dos bancos digitais no mercado de crédito Não é possível prever qual será o destino da competição entre bancos tradicionais e bancos digitais no Brasil Um cenário possível é que o mercado seja quebrado em nichos atendidos por bancos tradicionais e digitais Em outro caminho os bancos de tradicionais podem não sobreviver e serem substituídos pelos novos bancos Outro cenário 59 é de os bancos tradicionais digitalizarem suas operações mantendo sua base consumidora e a liderança do mercado 37 Análise temporal do setor bancário brasileiro Com o objetivo de analisar a evolução do desempenho dos bancos comerciais atuantes no mercado brasileiro ao longo do triênio de 2019 a 2021 foi calculado o Índice de Malmquist entre os extremos do período analisado o primeiro trimestre de 2019 e o terceiro trimestre de 2021 O índice permite realizar uma análise completa da evolução da produtividade de uma DMU ao longo de um período determinado Seus componentes permitem obter informações sobre as variações de eficiências e sobre as variações da tecnologia de produção O índice foi implementado através do pacote FEAR Frontier Efficiency Analysis with R de Paul W Wilson no programa R Os intervalos de confiança foram construídos com base nos intervalos percentuais de Hall com base nas diferenças O índice foi calculado com orientação para inputs portanto resultados menores que a unidade apresentam melhoras na produtividade enquanto que valores superiores indicam pioras O intervalo de confiança não pode incluir a unidade para que se possa afirmar estatisticamente que o movimento de alta ou baixa na produtividade ocorreu Ou seja o intervalo de confiança deve estar totalmente à esquerda da unidade para que haja uma melhora na produtividade ou totalmente à direita para que haja uma queda na produtividade durante o período selecionado Tabela 19 Índice de produtividade de Malmquist para os bancos digitais Bancos 25 Malm 975 Agibank 0383 0573 0673 C6 Bank 0024 0007 0006 BS2 1028 1206 1329 Inter 1963 1902 2340 Original 0539 0713 0750 Média conclusivos 0880 Fonte O autor 2023 60 Na tabela 19 estão expressos os índices de produtividade de Malmquist para o período do primeiro trimestre de 2019 para o terceiro trimestre de 2021 No intervalo considerado apenas os bancos BS2 e Inter tiveram redução na produtividade dos seus fatores de produção entre os bancos digitais O banco C6 bank teve o melhor desempenho do período apresentando um aumento da sua produtividade na ordem de mais de 99 segundo o índice de Malmquist Os bancos digitais como um todo conforme a média do índice apresentam um avanço entre 2019 e 2021 de aproximadamente 12 Tabela 20 Índice de alterações de eficiência para os bancos digitais Bancos 25 Eff 975 Agibank 0051 0442 0561 C6 Bank 0017 0002 0002 BS2 0285 0950 1204 Inter 0893 1425 1879 Original 0125 0553 0701 Média conclusivos 0332 Fonte O autor 2023 Os bancos digitais apresentaram uma melhora relevante na eficiência técnica obtida no intervalo de análise O banco C6 bank foi o que apresentou a melhora mais acentuada na gestão de seus inputs Em relação à média considerando todos os bancos digitais nada se pode afirmar pois no índice de alterações de eficiência alguns bancos apresentam intervalos de confiança inconclusivos pois incluem a unidade Para os bancos BS2 e Inter não se podem fazer afirmações quanto a alterações na eficiência observada pois a unidade encontrase dentro dos intervalos de confiança estimados A média foi calculada tendo como base apenas os bancos com intervalos de confiança conclusivos apontando em média para uma melhora na eficiência técnica de aproximadamente 66 61 Tabela 21 Índice de alterações de tecnologia para os bancos digitais Bancos 25 Tech 975 Agibank 0911 1297 1828 C6 Bank 1052 3067 4822 BS2 0853 1270 1760 Inter 0975 1335 1833 Original 0681 1289 1798 Média conclusivos 3067 Fonte O autor 2023 A mudança da tecnologia de produção não pode ser considerada determinante para a produtividade positiva observada para os bancos digitais pelo índice de Malmquist Para quatro dos cincos bancos o intervalo de confiança inclui a unidade Para o banco C6 Bank o índice aponta que não houve progresso tecnológico no período Tabela 22 Índice de produtividade de Malmquist para os bancos de varejo Bancos 25 Malm 975 Banco da Amazônia 0205 0311 0310 Banese 1097 1094 1233 Banpará 0634 0783 0800 Banco do Nordeste 0708 0733 0753 Banestes 0835 0824 0953 Banrisul 1001 1053 1139 BB 0979 1018 1102 Bradesco 0765 0832 0863 BRB 0621 0728 0737 CEF 0963 0980 1067 Itaú 0726 0797 0879 Mercantil do Brasil 0884 0883 0927 Safra 1004 1015 1038 Santander 0761 0811 0972 Média conclusivos 0822 Fonte O autor 2023 Conforme os valores expressos na Tabela 22 percebese que os bancos de varejo tiveram um aumento na sua produtividade de em média 17 considerando apenas os bancos com intervalos de confiança conclusivos Apenas os bancos Safra Banese e Banrisul 62 apresentaram redução no seu desempenho segundo o índice de Malmquist O Banco da Amazônia que possui o maior escore de eficiência no terceiro trimestre de 2021 apresentou o maior aumento de produtividade do período de aproximadamente 70 Para BB e CEF nada se pode afirmar pois apresentam intervalos de confiança inconclusivos Tabela 23 Índice de alterações de eficiência para os bancos de varejo Bancos 25 Eff 975 Banco da Amazônia 0004 0204 0238 Banese 0393 0798 1011 Banpará 0184 0576 0678 Banco do Nordeste 0300 0583 0723 Banestes 0277 0678 0898 Banrisul 0173 0722 0886 BB 0304 0830 1075 Bradesco 0238 0682 0874 BRB 0106 0494 0586 CEF 0394 0783 1003 Itaú 0207 0670 0876 Mercantil do Brasil 0183 0578 0726 Safra 0359 0842 1085 Santander 0265 0703 0938 Média conclusivos 0589 Fonte O autor 2023 63 Tabela 24 Índice de alterações de tecnologia para os bancos de varejo Bancos 25 Tech 975 Banco da Amazônia 1031 1526 2155 Banese 0983 1371 1886 Banpará 0966 1359 1841 Banco do Nordeste 0862 1258 1667 Banestes 0803 1215 1722 Banrisul 1066 1459 2097 BB 0796 1226 1709 Bradesco 0712 1220 1689 BRB 1039 1474 2062 CEF 0857 1252 1722 Itaú 0725 1190 1674 Mercantil do Brasil 1032 1527 2167 Safra 0728 1205 1656 Santander 0714 1155 1616 Média conclusivos 1412 Fonte O autor 2023 O desmembramento do índice de Malmquist para os bancos de varejo apresenta resultados parecidos com os observados para os bancos digitais Esses bancos obtiveram um aumento relevante na eficiência técnica de em média 41 Para algumas instituições os intervalos de confiança apontam um retrocesso na tecnologia de produção Podemos afirmar que não ouve um progresso tecnológico positivo no período analisado Os resultados apresentados pelo Índice de Malmquist indicam que entre o primeiro trimestre de 2019 e o terceiro trimestre de 2021 os bancos comerciais como um todo se tornaram mais produtivos Bancos digitais e bancos tradicionais de varejo apresentaram melhoras relevantes na gestão de seus inputs aproximandose assim da fronteira eficiente Por outro lado não foram observadas mudanças tecnológicas positivas não existindo indícios de avanços tecnológicos no período Logo o aumento de produtividade observado se deve integralmente ao comportamento dos bancos no emprego de seus insumos no processo de intermediação financeira 64 CONCLUSÕES Este trabalho realizou através da técnica DEA a comparação entre os desempenhos de cinco bancos digitais Agibank C6 Bank BS2 Inter e Original com os bancos de varejo tradicionais Ainda foi realizada uma análise temporal do setor bancário brasileiro através da construção do Índice de Malmquist para o triênio de 20192021 A abordagem escolhida para avaliar os bancos foi a da intermediação financeira onde foi possível mensurar o desempenho das instituições no gerenciamento de seus inputs para a concessão de crédito Como foi visto a literatura sobre eficiência bancária internacional e nacional é vasta Muitos trabalhos avaliaram bancos de diversos países com várias abordagens diferentes Neste trabalho o modelo proposto por Sealey e Lindley onde o banco é visto como uma DMU que utiliza os fundos captados e o uso de capital e trabalho para produzir ativos remunerados neste caso operações de crédito Esta abordagem como foi percebido é a mais presente nos estudos sobre eficiência bancária pois avalia os bancos na atividade que mais pode ter impacto nas economias nacionais A DEA se mostrou como a melhor opção para o trabalho devido ao fato de não exigir que se especifique uma forma funcional para a fronteira eficiente como acontece com a SFA A utilização da técnica não paramétrica frequentemente utilizada nos trabalhos sobre eficiência bancária se mostra como a melhor opção para representar a atividade bancária visto que a literatura diverge sobre como seria a melhor forma de especificar uma função para representar a intermediação financeira pelos bancos A convexidade do conjunto de produção averiguada pelos testes de hipótese realizados validou a aplicação da DEA em comparação com outras técnicas não paramétricas Foi visto que a orientação para inputs é a mais adequada para a estimar a eficiência dos bancos pois é assumido que os gestores tem maior controle sobre recursos empregados no processo produtivo do que sobre os produtos Assim as eficiências foram calculadas com orientação para inputs sendo eles as captações dos bancos as despesas de pessoal variável proxy para o trabalho empregado e os ativos permanentes líquidos de imobilizado de arrendamento mercantil Foi utilizado o output de operações de crédito líquidas de provisão representando o produto produzido pelos bancos na atividade de intermediação financeira As eficiências foram calculadas com a correção dos escores de eficiência pelo método de bootstrap Foi apresentado que as estimativas da DEA são tendenciosas para cima apresentando valores superiores ao da eficiência real Este método permite estimar as 65 eficiências DEA sem esse viés O método de bootstrap também é empregado nos testes realizados neste trabalho sendo eles os testes de convexidade os testes de retornos de escala e os testes de comparação de médias aplicados Foi visto que a convexidade do conjunto de produção é fundamental para que a DEA possa cumprir todas as comparações e projeções que se deve esperar de um ranking de eficiência Foram aplicados os testes de Kneip et al 2015 e Simar e Wilson 2020 para testar a convexidade da amostra selecionada Os testes foram aplicados para o conjunto de produção da amostra em todos os seus recortes dentro do período de análise sendo constatado que em todos eles a amostra se apresentou convexa permitindo que a técnica DEA fosse aplicada O modelo DEA pode ser construído considerando que o conjunto de produção apresenta retornos constantes de escala com o modelo CCR ou retornos variáveis de escala com o modelo BCC Com o objetivo de testar qual modelo seria mais apropriado foram aplicados testes nos primeiros e terceiros trimestres dos anos de 2019 a 2021 A rejeição da hipótese de retornos constantes de escala em todo o período de análise indicou fortemente que os bancos brasileiros em relação a atividade de intermediação financeira possuem uma tecnologia que apresenta retornos variáveis de escala Foi observado que o setor bancário brasileiro apresenta níveis significativos de ineficiência visto que o escore médio obtido foi de 039 aproximadamente O Banco da Amazônia banco de varejo regional apresentou a maior desempenho nesta análise atemporal para o terceiro trimestre de 2021 com o escore de eficiência próximo de 080 O Banco de Brasília outro banco de varejo regional apresentou desempenho muito próximo com escore de eficiência de 079 aproximadamente O Itaú banco varejista de âmbito nacional apresentou o terceiro melhor desempenho com o escore de 070 Foi visto que entre os bancos digitais os que apresentaram os maiores escores de eficiência foram os bancos Agibank e C6bank 053 e 045 bancos enquadrados pelo BCB segundo seu porte no segmento S4 um segmento abaixo dos demais bancos digitais que apresentam escores de eficiência menores O banco BS2 apresentou escore de eficiência no valor de 005 sendo mais ineficiente deste grupo O banco Inter e o banco Original apresentaram os escores de 036 e 034 respectivamente Com o objetivo de realizar a objetivo principal deste trabalho de comparar os bancos digitais com os bancos de varejo na atividade de intermediação financeira foi aplicado o teste estatístico de Kneip et al 2016 para comparar a eficiência média de grupos produtivos distintos A primeira comparação realizada foi a do grupo dos bancos digitais com o grupo da totalidade dos bancos de varejo em atuação no setor bancário brasileiro O teste realiza o cálculo 66 de eficiência dos grupos desconsiderando o restante da amostra para depois comparálos O grupo dos bancos digitais apresentou eficiência média de 0797 enquanto que o grupo dos bancos de varejo apresentou eficiência média de 0872 O resultado do teste apontou para a diferença estatística entre os valores de eficiência com os bancos de varejo sendo considerados mais eficientes que os bancos digitais Dado que no mercado brasileiro bancos digitais são recentes e não é possível constatar um banco de grande porte entre eles diferentemente do que acontece com os bancos de varejo onde é possível encontrar bancos nos maiores segmentos S1 e S2 foi proposto a realização de uma outra comparação agora entre os bancos digitais e os bancos de varejo que se encontram no mesmo segmento em que são observados bancos digitais os segmentos S3 e S4 Neste caso os bancos digitais apresentaram eficiência média de 087 enquanto que os bancos de varejo ficaram por volta de 09 Os pvalores do teste não rejeitaram a hipótese de igualdade dos escores médios destes dois grupos Logo bancos digitais e bancos de varejo com o mesmo porte não apresentaram diferenças estatisticamente relevantes no seu desempenho médio no 3º trimestre de 2021 A partir dos resultados apresentados depreendese que os bancos digitais em média apresentam resultados inferiores em termos de eficiência quando comparados com os bancos de varejo considerando tanto os grandes bancos quanto os bancos regionais Porém quando são considerados apenas os bancos de varejo de porte similar aos bancos digitais hoje em atuação os resultados não apontam diferenças entre o desempenho destes grupos na abordagem da intermediação financeira Foi calculado o Índice de Malmquist para o triênio de 2019 a 2021 com o objetivo de analisar a evolução do desempenho dos bancos comerciais do mercado brasileiro O índice foi calculado com orientação para inputs onde resultados menores que 1 apresentam melhoras na produtividade enquanto que valores superiores a 1 indicam pioras Foi percebido que os bancos digitais como um todo conforme a média obtida pelo índice apresentam um avanço entre 2019 e 2021 de aproximadamente 12 em sua produtividade Os bancos de varejo apresentaram um aumento de produtividade de em média 18 Em relação a fonte do aumento de produtividade dos bancos digitais no período por meio da decomposição do Índice de Malmquist foi verificado que não houve ganhos de produtividade pelo progresso tecnológico no período para estes bancos enquanto que esses bancos tiveram um aumento médio de 76 em sua eficiência técnica Resultados similares foram vistos para os bancos de varejo onde não foi possível constatar um progresso tecnológico 67 no período analisado mas por outro lado foi verificado um aumento de 42 na eficiência técnica deste grupo Foi visto que os bancos comerciais como um todo se tornaram mais produtivos pela análise temporal realizada com o Índice de Malmquist Bancos digitais e bancos de varejo tradicionais apresentaram melhoras na gestão de seus inputs aproximandose assim da fronteira eficiente Por outro lado não foi possível observar indícios de progresso tecnológico durante o período Para concluir é sugerido para trabalhos futuros buscar avaliar comparativamente os bancos digitais com outros grupos de bancos com outras abordagens diferentes da abordagem da intermediação financeira utilizada neste trabalho A abordagem da produção de serviços por exemplo parece ser capaz de apresentar resultados interessantes para futuras análises porém exigirá dos pesquisadores encontrar dados que permitam realizar este estudo visto que os dados que são disponibilizados publicamente pelo BCB se referem basicamente a rubricas contábeis que podem ser difíceis de representar os serviços bancários 68 REFERÊNCIAS AHMAD N NAVEED A AHMAD S BUTT I Banking sector performance profitability and efficiency a citationbased systematic literature review Journal of Economic Surveys v 34 n 1 p 185218 2020 ARIFF M CAN L Cost and profit efficiency of Chinese banks a nonparametric analysis China Economic Review v 19 p 260273 2008 ASSAF A MATAWIE K M Improving the accuracy of DEA efficiency analysis a bootstrap application to the health care food service industry Applied Economics v 42 p 35473558 2010 AZAD A K WANKER P RAIHAN M Z ANWAR R MUSTAFA R Bank efficiency in Bangladesh revisited a slackbased network DEA approach Journal of Economic Studies v 46 2019 BANKER R D CHARNES A COOPER W W Some models for estimating technical and scale inefficiencies in data envelopment analysis Management Science v 30 p 10781092 1984 BANKER R D CUMMINS J D Performance measurement in the financial services sector frontier efficiency methodologies and other innovative techniques Journal of banking and finance v 34 p 14131416 2010 BERGENDAHL G DEA and benchmarks An application to nordic banks Annals of Operations Research v 82 p 233249 1998 BERG S A FORSUND F R HJALMARSSON L SOUMINEN M Banking efficiency in the Nordic countries Journal of Banking and Finance v 17 p 37188 1993 BERGER A N HUMPHREY D B The dominance of inefficiencies over scale and product mix economies in banking Journal of Monetary Economics v 28 p 11748 1991 BERGER A N DEYOUNG R Problem loans and cost efficiency in commercial banks Journal of Banking Finance v 21 p 849870 1997 69 BERGER A M HASAN I ZHOU M Bank ownership and efficiency in China what will happen in the worlds largest nation Journal of Banking and Finance v 33 p 113130 2009 BHATIA V BASU S MITRA S K DASH P A review of bank efficiency and productivity OPSEARCH v 55 p 557600 2018 BOGETOFT P OTTO L Benchmarking with DEA SFA and R International Series in Operations Research and Management Science v 157 2011 BONIN J P HASAN I WACHTEL P Bank performance efficiency and ownership in transition countries Journal of banking finance v 29 p 3153 2005 BOOT A HOFFMANN P LAEVEN L RATNOVSKI L Fintech whats old whats new Journal of Financial Stability forthcoming 2020 CARNEIRO M SALGADO JUNIOR A P MACORIS L S Avaliação da eficiência bancária por meio da abordagem de intermediação uma análise comparativa de instituições financeiras brasileiras Revista Eletrônica de Administração Porto Alegre v 85 p 336359 2016 Disponível em httpswwwscielobrjreadafDVKjqJyjMCSsDyGrXKjpVklang pt Acesso em 10 jun 2022 CASADO F L SOUZA A M Análise Envoltória de Dados conceitos metodologia e estudo da arte na educação superior UFSM 2015 CAVA P B SALGADO JUNIOR A P BRANCO A M F Evaluation of bank efficiency in Brazil a DEA approach Revista de Administração Mackenzie v 17 n 4 p 6284 2016 CAVES D W CHRISTENSEN L R DIEWERT W E The economic theory of index numbers of the measurement of input output and productivity Econometrica v 50 n 6 p 13931414 1982 COELLI T RAO D S P BATTESE G E An introduction to efficiency and productivity analysis Kluwer Academic Publishers 1998 COELLI T ESTACHE A PERELMAN S TRUJILLO L A primer of efficiency measurement for utilities and transport regulators World Bank Institute Development Studies 26062 2003 70 CHARNES A COOPER W W RHODES E Measuring efficiency of decision making units European Journal of Operational Research v 2 n 6 p 429444 1978 DEBREU G The coeficiente of resource utilization Econometrica v 19 p 273292 1951 EFRON B Bootstrap methods another look at the jackknife Annals of Statistics v 9 p 1 26 1979 ENGLISH M GROSSKOPF S HAYES K YAISAWARNG S Output allocative and technical efficiency of banks Journal of Banking and Finance v 17 34966 1993 FALCÃO V A CORREIA A R Eficiência portuária análise das principais metodologias para o caso dos portos brasileiros Journal of Transport Literature v 6 p 133146 2012 FARE R GROSSKOPF S LOVELL C A K Production frontiers Cambridge University Press 1994 FARIA J A de Eficiência no setor bancário brasileiro a experiência recente das fusões e aquisições 2006 126 f Dissertação Mestrado em Ciências Econômicas Faculdade de Ciências Econômicas Universidade Estadual do Rio de Janeiro Rio de Janeiro 2006 FARIA J A de PAULA L F de MARINHO A Fusões e aquisições bancárias no Brasil uma avaliação da eficiência técnica e de escala Texto para Discussão IPEA v 1233 p 134 2006 FARREL M J The measurement of productive efficiency Journal of the Royal Statistical Society v 120 p 253281 1957 FAVERO C A PAPI L Technical efficiency and scale efficiency in Italian banking sector a nonparametric approach Appl Econ v 27 p 385395 1995 FERRIER G D LOVELL C A K Measuring cost efficiency in banking econometric and linear programming evidence Journal of Econometrics v 46 p 22945 1990 FETHI M PASIOURAS F Assessing bank efficiency and performance with operational research and artificial intelligence techniques a survey European Journal of Operational Research v 204 n 2 189198 2010 71 FUKUYAMA H MATOUSEK R Modelling bank performance a network DEA approach European Journal of Operational Research v 259 p 721732 2017 KAO C LIU S T Multiperiod efficiency measurement in data envelopment analysis the case of Taiwanese commercial banks Omega v 47 p 9098 2014 KNEIP A SIMAR L WILSON P W When bias kills the variance central limit theorems for DEA and FDH efficiency scores Econometric Theory v 31 p 394422 2015 KNEIP A SIMAR L WILSON P W Testing hypotheses in nonparametric models of production Journal of Business and Economic Statistics v 34 n 3 p 435456 2016 KOOPMANS T C Activity analysis of production and allocation New York Wiley 1951 MCCAIG B STENGOS T Financial intermediation and growth some robustness results Economics Letters v 88 n 3 p 306312 2005 MALMQUIST S Index numbers and indifference surfaces Trabajos de Estatística v 4 p 209242 1953 MARINHO A Estudo de eficiência em alguns hospitais públicos e privados com a geração de rankings Texto para Discussão IPEA n 794 2001 MENDONÇA D J Relação entre eficiência e rentabilidade em instituições financeiras bancárias no Brasil 2017 95 f Dissertação Mestrado em Administração Programa de PósGraduação em Administração Universidade Federal de Lavras Minas Gerais 2017 OECD Digital disruption in banking and its impact on competition OECD Competition Committee p 150 2020 Disponível em httpswwwoecdorgcompetitiondigital disruptioninbankinganditsimpactoncompetition2020pdf Acesso em 10 jun 2022 PARADI J C HAIYAN ZHU H A survey on bank branch efficiency and performance Research with data envelopment analysis Omega v 41 p 6179 2013 PARADI J C ROUATT S ZHU H Twostage evaluation of bank branch efficiency using data envelopment analysis Omega v 39 p 99109 2011 72 PAULA L F de FARIA J A de Eficiência do setor bancário brasileiro por segmento de mercado uma avaliação recente Mesa 3 ANPEC Macroeconomia e Finanças 2008 PÉRICO A E SANTANA N B REBELATTO D A N Eficiência bancária os maiores bancos são os mais eficientes Uma análise por envoltória de dados Gest Prod São Carlos v 15 n 2 p 421431 2008 PÉRICO A E SANTANA N B REBELATTO D A N Estimating the efficiency from Brazilian banks a bootstrapped Data Envelopment Analysis DEA Production v 26 n 3 p 551561 2016 PHILIPPON T On fintech and financial inclusion BIS Working Papers v 841 2020 RAMOS R E B Michael J Farrell e a medição da eficiência técnica In ENCONTRO NACIONAL DE ENGENHARIA DE PRODUÇÃO XXVII 2007 Foz do Iguaçu Anais Foz do Iguaçu 2007 p 110 RANGAN N GRABOWSKY R ALY H Y PASURKA C The technical efficiency of US banks Economics Letters v 28 p 169175 1988 SEALEY C LINDLEY J Inputs outputs and a theory of production and cost at depository financial institutions Journal of Finance v 32 p 12511266 1977 SOUZA P O MEZA L A Uma pesquisa sobre índices de eficiência não radiais na análise envoltória de dados In SIMPÓSIO DE PESQUISA OPERACIONAL E LOGÍSTICA DA MARINHA 2006 Rio de Janeiro Anais Rio de Janeiro 2006 p 2741 SHERMAN H D GOLD F Bank branch operating efficiency evaluation with data envelopment analysis J Bank Finance v 9 n 2 p 297316 1985 SIMAR L WILSON P W Sensitivity analysis of efficiency scores how to bootstrap in nonparametric Frontier models Management Science v 44 n 1 p 4961 2020 SIMAR L WILSON P W Hyphotesis testing in nonparametric models of production using multiple sample splits Journal of Productivity Analysis v 53 p 287303 2020 STAUB R B SOUZA G S TABAK B M Evolution of bank efficiency in Brazil a DEA approach European Journal of Operation Research v 202 p 204213 2010 73 TABAK B M CAJUEIRO D O DIAS M V B The efficiency of chinese local banks a comparison of DEA and SFA BCB working papers v 346 p 115 2014 TZIOGKIDIS P Bootstrap DEA and hypothesis testing Cardiff Economics Working Papers v 18 p 121 2012 WANKE P BARROS C P Banking efficiency in Brasil Journal of International Financial Markets Institutions and Money v 28 p 5465 2014a WANKE P BARROS C P Twostage DEA an application to major Brazilian banks Expert Systems with Applications v 41 p 23372344 2014b ARTICLE IN PRESS JID EOR m5G November 29 20161015 European Journal of Operational Research 0 0 0 2016 111 Contents lists available at ScienceDirect European Journal of Operational Research journal homepage wwwelseviercomlocateejor Innovative Applications of OR Comparison of variable selection techniques for data envelopment analysis in a retail bank Juha Eskelinen Aalto University School of Business Department of Information and Service Economy PO Box 21220 0 0 076 Aalto Helsinki Finland a r t i c l e i n f o Article history Received 8 January 2015 Accepted 2 November 2016 Available online xxx Keywords Data envelopment analysis OR in banking Variable selection Productive efficiency analysis a b s t r a c t If there are too few units compared to inputs and outputs the efficiency evaluation based upon the data envelopment analysis suffers from a lack of discrimination The literature has proposed various statistical techniques when the value judgments do not guide the selection of the inputsoutputs Two techniques the variable reduction procedure of Jenkins and Anderson 2003 and the approach based upon the ef ficiency contribution measure of Pastor Ruiz and Sirvent 2002 were compared in an empirical retail bank context The objective was to select a representative set of outputs from the services the bank provides As the techniques take different approaches to selecting influential variables the output sets proposed by the techniques diverged This created some significant differences in the efficiency evalua tions of the bank branches The bank management gave feedback on the techniques and the results from a practical perspective The techniques led to different managerial interpretations of the performance complementing each other Thus the techniques can be utilized to evaluate the units from multiple per spectives 2016 Elsevier BV All rights reserved 1 Introduction Data envelopment analysis DEA is a method used to evaluate the efficiencies of comparable units with multiple inputs andor outputs It takes an optimistic perspective by choosing for each decisionmaking unit DMU the most beneficial nonnegative ie those weights will maximize its efficiency score This virtue has a consequence Having a low number of units under assessment compared to inputs and outputs leads to finding a large portion of the units efficient or nearly efficient The identification of many units falsely efficient and too optimistic efficiency scores is called a lack of discrimination see eg Podinovski Thanassoulis 2007 Such an efficiency estimation may not serve the purpose of the evaluation Banks consume several resources inputs to produce several services or other outcomes outputs Even though there are large nationwide and international banks the majority of the branch networks are small with a relatively low number of branches 1 In Corresponding author Email addresses juhapeskelinenaaltofi juhaeskelinenmelkiorfi 1 In the United States the commercial banks had on average 132 branches in 2011 data source US Federal Deposit Insurance Corporation In the United King dom the monetary financial institutions have 287 in Germany 189 in France 318 and in Finland 38 branches data source European Central Bank 2011 UK data from 2010 The exceptionally low figure in Finland is explained by the high proportion of small local cooperative and savings banks addition in the large banks the branches are often clustered into smaller homogeneous subsets Thus the lack of discrimination can be an issue in bank branch efficiency evaluations see eg the dis cussion of Paradi Zhu 2013 This essay focuses on the variable side ie inputs and outputs of the discrimination problem The number of variables can be re duced by selecting the most important outputsinputs to depict the activity Numerous statistical techniques have been proposed for situations where judgmental knowledge is not available or not sufficient enough for selecting the variables The following review illustrates the variety of techniques 2 Some selective techniques omit variables prior to DEA Lewin Morey and Cook 1982 among others used correlation and re gression analysis to test redundant variables to be omitted Jenkins and Anderson 2003 proposed a variable reduction procedure based on partial covariance abbreviated as VR in this paper where a combination of variables is selected based upon the proportion of the total variance retained GonzalezBravo 2007 2 Besides selective techniques that are the focus of this paper there is another branch of statistical techniques that deals with the lack of discrimination by deter mining statistically the weights for summing up the data in a reduced set of vari ables Various techniques ending up in aggregation have been proposed by Sengupta 1990 SinuanyStern and Friedman 1998 Ueda and Hoshiai 1997 Adler and Golany 20 01 20 02 Bian 2012 and Amirteimoori Despotis and Kordrostami 2014 httpdxdoiorg101016jejor201611009 03772217 2016 Elsevier BV All rights reserved Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 2 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 ARTICLE IN PRESS JID EOR m5G November 29 20161015 proposed a procedure to help model specification by using an out putinput ratiobased analysis prior to DEA Another subbranch of selective techniques tests the impact of the retained variables on DEA results Banker 1996 proposed the use of statistical tests for different inefficiency distribution assumptions to indicate the significance of a variable in the production process ie efficiency estimates Simar and Wilson 2001 recommended bootstrapping to test whether the input or output variables are relevant Pastor Ruiz and Sirvent 2002 in troduced an efficiency contribution measure ECM to test for the significance of a variable in efficiency estimation Fanchon 2003 Ruggiero 2005 and Sharma and Yu 2015 regressed the effi ciency scores against the variables and tested the regression coeffi cient in dropping the insignificant variables Wagner and Shimshak 2007 proposed a stepwise procedure that evaluated the impact of the variable set on the average efficiency scores There are a few published studies comparing some of the sta tistical variable selection techniques see Nataraja Johnson 2011 Sirvent Ruiz Borras Pastor 2005 These comparisons are based upon Monte Carlo simulations and the efficiency estimates are compared to the known true frontier We take another approach in this paper The main interest is in the implications that the use of variable selection could have in the interpretation and acceptance of the DEA results for performance management We use data that have been gathered during the re search collaboration with a Finnish retail bank The variable selec tion has a specific contextual focus to specify a model for evaluat ing the operational efficiencies of the branches during 20072010 using DEA The techniques and their outcomes are reviewed by the branch network management We focus on two techniques VR and ECM Our prior assump tion is that both techniques could be helpful in our empirical case where the output variables represent the sales quantities of bank ing services that are highly positively correlated Both techniques arise from the critique against intuitive variable exclusion based upon high correlations between variables and present statistical approaches for variable selection see Jenkins Anderson 2003 Pastor et al 2002 3 Both techniques aim to minimize information loss when omit ting variables but they use very different definitions of informa tion loss While the objective of VR is to minimize the reduction in data variability ECM focuses on minimizing distortion in the dis tribution of efficiency scores This paper describes what kinds of consequences the different definitions can have in the case con text The techniques should be considered as tools to help the man agement when deciding on inputs and outputs for efficiency eval uation If the purpose is solely the correct classification of efficient and inefficient DMUs in a certain context and the purpose is not variable selection there may be other choices to reduce dimen sionality 4 Even though the industry is less relevant in variable selection for DEA it is worth noting that VR has been associated as an option to improve discrimination in bank and bank branch effi ciency evaluation Paradi Zhu 2013 Paradi Yang Zhu 2011 The reader may also be interested in the application of ECM in 3 Even though we focus on the outputs both techniques can be applied to inputs or both Jenkins and Anderson 2003 present an example where they apply their approach to the entire set when there are high correlations between some of the inputs and outputs 4 ECM performed quite well in most tested scenarios when compared with a number of selection and aggregationbased approaches Nataraja Johnson 2011 VR has been compared with a technique that combines principal component anal ysis PCA and DEA Adler Yazhemsky 2010 The latter outperformed VR in the simulations Even though the combination PCA and DEA reduces dimensionality it is an aggregation approach rather than a tool for guiding the selection of variables the evaluation of a Spanish bank branch network Pastor Lovell Tulkens 2006 The rest of this paper is organized as follows Section 2 presents the two selected techniques used in this study Section 3 intro duces the empirical case including the data used and the ini tial estimation without variable reduction Section 4 describes how the techniques were applied to construct alternative specifications for efficiency estimation Section 5 discusses the key findings Section 6 summarizes the management feedback to the compari son Section 7 makes concluding remarks 2 Variable selection techniques First we briefly introduce a DEA estimator that provides the baseline for this study and then we explain both techniques ap plied for variable selection in this comparison 21 DEA estimator The efficiency of a DMU can be evaluated using the generalized DEA estimator presented by Banker Charnes and Cooper 1984 We use the outputoriented multiplier form of DEA here Assume there are n DMUs each consuming m inputs and producing p out puts The inputs are denoted by x ℜ m and outputs by y ℜ p Let X ℜ m n and Y ℜ pn be the matrices of observed inputs and outputs for the DMUs DEA finds the most favorable input weights ν and output weights μ for the DMU under assessment denoted by superscript r Scalar u o is associated with the returns to scale RTS assumption If constant returns to scale CRS are assumed u o equals zero and the estimator is called CCR Charnes Cooper Rhodes 1978 If variable returns to scale VRS are assumed u o is free and the estimator is called BCC Banker et al 1984 The generalized DEA estimator is the following min νT x r u o 1 st μT y r 1 μT Y νT X 1 T u o 0 T μ ν 0 The objective function provides an efficiency score for the DMU the smaller the better because of the output orientation If the score is one and all variable weights ν and μ are strictly positive the DMU is efficient The first constraint normalizes the sum of weighted outputs to one This normalization has a convenient sideproduct because the weighted value of the output μi y r i depicts directly the importance of the particular variable in the efficiency evaluation of the DMU This weighted value is called the virtual output and it is an impor tant concept in this comparison The second constraint limits the weights in such a way that the performance of the unit is scaled to the best performance among all units 22 Variable reduction based on partial covariance As variables in the efficiency evaluation are often highly cor related an intuitive approach is to omit some of the highly cor related variables without a significant loss of information Jenkins and Anderson 2003 criticized this approach because interrela tions between variables are less obvious and cannot be determined directly from the correlation matrix They wanted to introduce a systematic statistical procedure to help make decisions for variable selection The authors have not given any specific name or abbre viation to their procedure We call it variable reduction based on partial covariance similarly to Adler and Yazhemsky 2010 VR is a multivariate statistical procedure used to find a reduced set of initial input or output variables to be included in DEA The Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 3 ARTICLE IN PRESS JID EOR m5G November 29 20161015 aim is to minimize the information loss of original data when re moving variables The indicator of the information contained by a variable is the variance around the mean The following summary of the procedure is based on Jenkins and Anderson 2003 The conditional variance of a remaining i variable when the ef fect of variable i is removed is denoted by σ ii If the variables are perfectly correlated then the variance remaining is σ ii 0 and i can be removed without any loss of information This conditioning can be extended to the sets of retained and omitted variables To illustrate the procedure we apply it to the output variables The data are normalized to zero mean and unit variance to treat all outputs equally Then the sum of variances of the normalized outputs is equal to the number of outputs p The outputs are divided into the set of variables omitted i 1 k and variables retained i k 1 p where the variables are in an arbitrary order The variance covariance matrix of outputs denoted by V can be partitioned as variancecovariance matrices of omitted variables denoted by V 11 and retained variables V 22 covariance of omitted and retained variables V 12 and vice versa V 21 The partial variancecovariance matrix of omitted variables given the retained variables is computed V 11 2 V 11 V 12 V 1 22 V 21 The trace of V 112 represents the size of the remaining variance of omitted variables after conditioning the retained variables The sum of trace measures the information loss The proportion of in formation retained can be obtained by comparing the information loss to the original sum of variance p The test of the partial covariance is repeated for all combina tions of retained and remaining variables The information losses of the combinations are monitored A reduced variable set with minimal loss of information is chosen and this variable set is used with the appropriate DEA estimator Jenkins and Anderson 2003 do not give exact criteria for choosing the set of initial variables retained The number of vari ables is externally defined to avoid the lack of discrimination Jenk ins and Anderson state that the ultimate decision to omit particu lar variables depends upon the importance of the variables to the management and that the VR procedure plays a supportive role in this decision 23 Variable selection based upon the efficiency contribution measure Even though the ECM approach is also a selective technique it differs from VR in the role of DEA VR makes the variable selection using a statistical test without any use of DEA The ECM approach uses the appropriate DEA estimator to test the marginal contribu tion of each variable to the efficiency estimation Following Pastor et al 2002 the test procedure is presented for an outputoriented radial efficiency estimation with efficiency scores 1 where 1 is efficient The ECM approach is based on two variable specifications that are called total model and reduced model The total model con tains a set of variables In the reduced model one of the input or output variables called the candidate variable denoted by z will be removed As the first step the data used in the total model is adjusted in such a way that all the DMUs become efficient Then the candidate variable is removed and DEA is executed with the re duced model DMU r s efficiency score denoted by ϕ r represents the change in the efficiency scores between the total model and the reduced model ϕ r is called the efficiency contribution measure of variable z to DMU r If ϕ r 1 then we can conclude that the candidate variable z has no influence on the efficiency evaluation of the DMU Analogously ϕ r 11 indicates that the removal of the candidate variable has a 10 impact on the efficiency score of the DMU These scores are calculated for all DMUs The relevance of the candidate variable z for the total model can be tested statistically using a binomial test The null hypothesis is that the efficiency score distributions of the original model and the reduced model do not differ significantly and the candidate variable could be eliminated The test has two parameters The first one is a parameter for tolerated change between the models de noted by ϕ Then T is the number of DMUs that exceed this toler ance The second parameter is the probability level denoted by p 0 for the maximum allowed proportion of DMUs exceeding the tol erance Sirvent et al 2005 suggest that ϕ 11 and p 0 015 are a suitable conservative choice for practical purposes These parame ters mean that the proportion of DMUs affected by more than 10 should not exceed 15 at the given significance level Otherwise the variable cannot be eliminated The use of the ECM is a stepwise process Pastor et al 2002 propose both a backward procedure variable elimination and a forward procedure variable incorporation As the backward procedure is relevant for variable reduction it will be our focus First all variables are tested The candidate variable with the low est value of T is eliminated Then a new original model with the remaining variables is built The process continues until no vari ables can be eliminated 3 The empirical case The case is an application to evaluate the operational efficien cies of the bank branches in a Finnish retail bank Helsinki OP Bank Plc 5 This is a retail bank operating in the Helsinki metropolitan area It belongs to OP Group which is a leading financial service provider in Finland The variable selection techniques are applied to a set of out puts services for operational efficiency estimation when assum ing that all outputs are initially equal Two alternative output spec ifications are defined one using the VR technique and one using the ECM technique The impacts of the use of these techniques are assessed by the following questions How discriminative are the alternative output specifications in relation to the distribution of efficiency scores and the identifi cation of efficient units What are the implications for the ranking of the units based upon the efficiency scores of the alternative output specifica tions What is the importance of each main group of services to the efficiency evaluation when alternative output specifications are used Management feedback on the techniques and the results will be also considered 31 Outputs and inputs The bank branches provide more than 50 banking products that are tracked in the banks sales system Many of them are only vari ants such as different types of credit cards The output set can be reduced to ten services by aggregating different product vari ants These services denoted by i are presented in Table 1 The unit of measurement related to each service is the number of new agreements such as new housing loans granted mutual fund in vestments sold or current accounts opened during the period The services comprise three main groups financing invest ment and daily banking In addition to housing loans and consumer loans financing services include payment protection in surance The investment services include savings in an account or investment funds either directly or as insurance savings accounts Daily banking services include services that the customers need for 5 As of April 1 2016 the banks name has been Helsinki Area Cooperative Bank Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 4 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Table 1 Initial outputs for the bank branch efficiency evaluation Notation i Service Service group 1 Housing loans Financing 2 Consumer loans Financing 3 Payment protection insurance Financing 4 Mutual funds Investment 5 Insurance savings Investment 6 Savings accounts Investment 7 Critical illness cover Investment 8 Current accounts Daily banking 9 Online banking agreements Daily banking 10 Cards Daily banking payments and other daily affairs Thus new agreements for daily services reflect well the new customer base Critical illness cover is actually insurance but sold by banking professionals It is a rel evant output of the bank branch activity but it is not significant enough to be a fourth group in this analysis The bank manage ment decided to locate it under investment services in the effi ciency evaluation The initial inputs represent the working time of employees fulltime equivalent FTE in five different job roles in the sales activity The bank had standard unit costs for different job roles These unit costs were used to aggregate the cost of the workforce used as single input The unit of measurement presented in this paper is working year FTE standardized with these unit costs 32 Data This study is based on annual data of the 25 bank branches from the years 20072010 As some of the branches were founded during the time frame there are 95 observations There are several temporal options for DEA when panel data is used see Tulkens Vanden Eeckaut 1995 We use an intertemporal approach in the variable selection meaning that all observations specify the model used in the efficiency estimation If we applied a contemporaneous or sequential approach each period would have a different specifi cation which would complicate the comparison of the techniques 6 In this study DMUs are the observations of branchyear combina tions As the variable selection techniques are based upon statistical analysis the distributions and correlations are relevant when ex plaining both the specifications and outcomes of the efficiency es timations The cost of the work force is denoted by c The out puts ie services are denoted by y i i 1 10 The descriptive statistics are presented in Table 2 The distributions of the variables are positively skewed There are more small and mediumsize branches than large branches in the network The skewness differs by service One reason for these differences is that in the panel data the distribution is affected not only by the branch size but also by the yearly fluctuation of the service The correlations between the variables are presented in Table 3 They are relatively high between the outputs and between the out puts and the input There are however some lower correlations even between services within the same group The lowest corre lation is between y 4 mutual funds and y 6 savings accounts The reason for this is in the customers behavior during the turbu lent times of 20072010 During the bull market before the finan cial crisis customers sought profits from mutual funds but as the 6 Even though we select a single set of variables for all periods changes over time can also be allowed Chen and Johnson 2010 demonstrate how a variable selection technique in their case ECM can be used to study the change of the set of relevant variables and bring new insights to the dynamics of the performance share prices dropped customers changed to less risky instruments such as fixedterm savings accounts 33 Initial sales efficiency estimation The baseline estimation utilized a sales efficiency model that consists of all ten outputs We do not utilize the logical group ing of services The branches intent was to maximize the sales of banking products with their resources rather than minimize the resources to a given demand Thus the outputoriented measure ment of efficiency will be used We apply a radial outputoriented DEA estimator 1 Additive DEA estimators based on the difference between weighted outputs and weighted inputs are sometimes used in bank branch efficiency analysis They allow negative inputoutput values which we do not have 7 The radial DEA is more appropriate for our purposes to rank the units The efficiencies will be estimated using both CRS and VRS assumptions and a test will be used to determine the appro priateness of the CRS assumption When applying the initial sales efficiency estimation with ten outputs 17 units 18 of all units are found to be efficient when CRS is assumed and 27 units 28 when VRS is assumed The av erage efficiency scores are 082 CRS and 087 VRS As this is a real application we cannot obviously say which of the units are actually correctly or incorrectly identified as efficient or inefficient As the purpose is to give help to managers in specifying the output set we look at the initial estimation from their perspec tive If they are satisfied with the outcome of the initial estimation they would accept a high proportion of efficient units compared to a reduced variable set When used in targetsetting less improve ment would be required from the branch network Some of the improvement potential may be lost and there would be a risk of picking poor benchmark DMUs from which the others have little to learn They would also accept that a unit can be determined to be efficient based upon high performance in one output even if the larger number of other outputs would have significant space for improvement If the discrimination is increased the benchmarks are fewer and they represent more likely good practices but on the other hand some relevant benchmarks may be ignored The targets based upon an efficient frontier tend to be more of a stretch If the lo cal management and employees consider the targets impossible to achieve these targets may be potentially demotivating In our case the management considered the variable selection to be justified 4 Applying the variable selection techniques This section presents how the reduced output specifications were constructed using the variable selection techniques 41 Applying VR In this study VR was applied to select one output from each service category Thus the number of retained outputs was three Appendix A presents the ranking of the 36 combinations meeting this criterion We computed the matrices of partial variancecovariance for each possible combination of retainedomitted variables and sorted the combinations according to the variance retained as proposed by Jenkins and Anderson 2003 7 See for example Portela and Thanassoulis 2007 where the outputs include the change in the number of clients and differences in the monetary values of var ious financial services These changes can obviously also be negative Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 5 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Table 2 Descriptive statistics of variables Mean Median Std Dev Skewness Kurtosis Min Max c Cost of sales force 12 0 103 6 3 0 73 0 61 42 274 y 1 Housing loans 392 5 353 227 4 0 82 0 14 92 1074 y 2 Consumer loans 286 0 244 179 7 0 95 0 25 52 802 y 3 Payment protection 157 4 153 95 9 0 73 0 01 33 432 y 4 Mutual funds 684 1 459 568 6 1 67 2 66 100 2781 y 5 Insurance savings 131 9 101 88 6 1 01 0 83 16 458 y 6 Savings accounts 918 4 836 670 7 1 26 1 26 153 3003 y 7 Critical illness cover 91 4 72 71 5 1 07 0 29 12 297 y 8 Current accounts 849 2 728 471 5 0 81 0 01 277 2291 y 19 Online banking agreements 622 7 585 339 0 0 67 0 35 172 1619 y 10 Cards 896 7 756 528 8 0 69 0 57 185 2206 n 95 For c the unit of measurement is the working year FTE Outputs y 1 y 10 represent quantities of transactions Table 3 Correlation matrix between variables C y 1 y 2 y 3 y 4 y 5 y 6 y 7 y 8 y 9 y 10 c 1 y 1 0941 1 y 2 0781 0836 1 y 3 0743 0799 0686 1 y 4 0842 0777 0720 0620 1 y 5 0833 0834 0834 0710 0771 1 y 6 0734 0742 0685 0588 0433 0711 1 y 7 0812 0761 0695 0812 0682 0707 0678 1 y 8 0846 0850 0725 0825 0593 0707 0799 0853 1 y 9 0790 0811 0780 0784 0528 0724 0852 0825 0958 1 y 10 0766 0786 0885 0733 0577 0758 0814 0788 0892 0954 1 All correlations are significant at the 0001 level The first ranked combination y 3 Payment protection insurance y 4 Mutual funds and y 9 Online banking agreements was selected for use in the comparison We call it the VR specification in this comparison The differences between the topranked combinations were rather small We will discuss the sensitivity of this selection later in 53 42 Applying the ECM We used the backward procedure of the ECM to delete superflu ous variables from the initial variable set The DEA efficiency scores used as the efficiency contribution measure are potentially affected by the RTS specification As we do not have a predefined RTS speci fication the procedure was executed both for CCR and BCC estima tors separately Thus it could be possible that the procedure ends up with a different reduced variable set As Sirvent et al 2005 suggest parameters ϕ 11 and p 0 015 were used when eliminating outputs in the stepwise process Binomial distribution B94 085 was used with a 5 level of sta tistical significance to determine if the output can be eliminated 8 However in some steps two or more variables were tied for the highest effect with the initial tolerance level In such cases the tol erance was decreased ϕ 105 ϕ 103 etc until the variable to be removed was identified The stepwise variable elimination process is presented in Appendix B for both CRS and VRS assumptions Even though there were some differences between the intermediate steps both as sumptions ended up with the same three variables after which no further variable eliminations were possible When following the ECM procedure the ideal combination for evaluating efficiency is variables y 1 Housing loans y 4 Mutual funds and y 8 Current ac counts The rest of the outputs can be considered superfluous We 8 Binomial distribution B94 085 was used because there are only n 1 DMUs that can be affected At least one of the DMUs remains efficient in the reduced model call the combination of these three variables the ECM specification in this comparison This combination of y 1 y 4 and y 8 also meets the criterion to select one variable from each service group similarly to our appli cation of the VR procedure It retains 8353 of the total variance of the initial output It is however only the 21th of the variable alternatives according to the VR procedure see Appendix A At this stage we can conclude that y 4 Mutual funds is a statis tically highly influential service from the performance perspective as it is included in the reduced variable set in both the VR and ECM variable sets 5 Findings The efficiencies of the DMUs were estimated by employing both CRS and VRS assumptions The descriptive statistics of the effi ciency scores and the number of efficient units are presented in Table 4 The scores are presented in a form where scores below one represent inefficiency High average median and minimum ef ficiency scores and low standard deviation are also indicators of possible overestimation of efficiencies The initial DEA estimation which contains all ten variables is by default the least discriminative specification If VRS was as sumed 28 of the DMUs would be considered efficient The dis crimination can clearly be improved by adopting either of the se lective techniques The scores of the VR specification are on aver age lower than the scores of the ECM specification On the other hand the ECM specification finds less DMUs efficient than the VR specification In order to examine which RTS assumption should be used in the model the nonparametric KolmogorovSmirnov KS test Banker 1993 Banker Natarajan 2011 is used The test com pares CCR and BCC efficiency score distributions The null hypoth esis is that the frontier exhibits CRS and the alternative hypoth esis is VRS The results of the KS test are presented in Table 5 The initial estimation retains the null hypothesis of CRS if the 5 Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 6 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Table 4 Descriptive statistics of efficiency scores and number of efficient DMUs under CRS and VRS assumptions Number of efficient DMUs Efficiency scores Mean Median StdDev Min Constant returns to scale CCR Initial estimation 17 082 082 013 057 VR specification 4 069 068 015 041 ECM specification 3 076 075 012 056 Variable returns to scale BCC Initial estimation 27 087 091 012 065 VR specification 16 077 076 016 042 ECM specification 13 082 082 012 057 n 95 Table 5 KolmogorovSmirnov test of the constant returns to scale specification Most extreme differences D pvalue Initial estimation 0179 0095 VR 0242 0008 ECM 0221 0019 Critical D 5 sign is 0197 Fig 1 Rankings of the DMUs by VR and ECM specifications statistical significance level is required Both the VR and ECM spec ifications reject CRS for VRS The statistical techniques do not only increase discrimination but they may increase the difference be tween the scores of RTS assumptions as less information is used Based upon the RTS test we assume VRS in the remaining part of this paper 51 Ranking differences Some significant differences occur in the rankings of individ ual DMUs between the output specifications selected by the tech niques Fig 1 plots the ranking of DMUs The figure illustrates that the evaluations based upon the two variable sets identified by dif ferent techniques are related The point in the lower left corner ranking 1 with both the ECM and VR specifications includes 11 observations that are identified as efficient using both techniques There are however many significant differences The biggest difference in rankings was related to branch B23 which according to the ECM specification is efficient but the VR Table 6 Rank correlations between DMUs VRS Spearmans rank correlation coefficient Initial estimation VR ECM Initial estimation 1 VR specification 0796 1 ECM specification 0830 0765 1 All correlations are significant at the 0001 level specification ranks it as 54th with an efficiency score 075 This is illustrated in Fig 1 The reason for the high ECMbased score was the high housing loan sales when the market peaked in 2008 However the branch was not able to utilize that opportunity in the sales of payment protection insurance to the housing loan cus tomers which would have provided high scores when the VR spec ification was used too There are several similar examples where the excellent performance of a branch becomes mediocre using the other technique or a mediocre performance becomes poor The rank correlations of the DMUs are presented in Table 6 The rank correlation also indicates that even though the rankings based upon the two techniques are significantly correlated they draw a different picture of the relative performance The differences between the ECM and VR emerge from the dif ferent aims in the variable selection VR focuses on the variance between the variables and thus emphasizes the differences in out puts between the units On the other hand the ECM tends to retain variables that cause significant changes in the efficiency scores of the units These variables that the ECM retains are not necessarily the ones with big differences between the DMUs For example y 1 Housing loans is a core service in which most branches perform relatively well thus a large number of DMUs operate close to the frontier and y 1 cannot be deleted without a significant change in the efficiency scores of a large proportion of DMUs As different techniques take different approaches to dealing with the dimensionality issue some variations in the results are natural 52 Importance of outputs in efficiency evaluations The analysis of virtual outputs proposed by Boussofiane Dyson and Thanassoulis 1991 can be insightful Virtual outputs depict the importance of a particular output for the unit under assess ment for the maximum efficiency rating Virtual outputs can be utilized to identify units with good practices that could be dissem inated Table 7 presents the descriptive statistics of virtual outputs in the initial estimation and the two alternative specifications Because a lot of attention is paid to the service groups in the case bank a question arises about how important each main ser vice group is to the efficiency evaluation We use virtual outputs to provide management with information on this matter We com pare the expected virtual outputs and the mean of the actual Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 7 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Table 7 Descriptive statistics of virtual outputs all DMUs μ1 y 1 μ2 y 2 μ3 y 3 μ4 y 4 μ5 y 4 μ6 y 6 μ7 y 6 μ8 y 8 μ9 y 9 μ10 y 10 Initial estimation Mean 0221 00533 0104 0158 00455 00811 01122 0128 00588 00400 StdDev 0284 0128 0225 0199 0091 0187 0193 0245 0156 0113 Min 0 0 0 0 0 0 0 0 0 0 Max 1 0762 0937 0922 04 4 4 0921 0839 0833 0834 0674 VR specification Mean 0197 0278 0525 StdDev 0258 0258 0287 Min 0 0 0 Max 1 1 1 ECM specification Mean 0370 0268 0362 StdDev 0379 0247 0371 Min 0 0 0 Max 1 1 1 The weights of the efficient DMUs are not unique in DEA and may vary by solver used Table 8 Mean virtual outputs by service group all DMUs Financing services Investment services Daily banking services t 2 test statistics pvalue Initial estimation 38 40 23 7 511 0028 VR specification 20 28 53 46 021 00 0 0 ECM specification 37 27 36 6 706 0041 The test is Hotellings tsquared statistics 8 The test is indicative only virtual outputs on the service group level in the set of all DMUs Even though the weights are free for an individual unit in a ser vice business like retail banking it may not be desirable for the overall efficiency evaluation to be too strongly ruled by one group of services Table 8 presents the mean virtual outputs of each service group by alternative There are clear differences between alternatives in how the service groups are weighted The virtual outputs of the ECM specification are most equally balanced between the service groups If balanced weighting of the service groups is wanted the initial estimation underweights daily banking services The VR specification underweights financing services and overweights daily services The importance of the variables among all DMUs depends on how far the observations generally are from the frontier The dis tributional anomalies affect the mean virtual outputs For exam ple positive skewness indicates that the majority of the variables observations are far away from the tails defining the frontier As the skewness differs by variable it is generally more favorable to weight variables whose observations tend to be closer to the fron tier The VR procedure is especially sensitive to the distributional differences as it is based upon the few retained variables selected prior to DEA 9 The low sum of the virtual output of financing ser vices indicates that a high performance in the representative vari able payment protection insurance is rare In addition the ECM procedure is potentially affected by extreme performances in sin gle service However the ECM procedure tends to delete outputs that have the vast majority of observations far away from the fron tier as it is favorable for the DMUs not to weight them 8 The null hypothesis is that the vector of mean virtual outputs of the groups does not deviate from the expected mean vector The expected mean is approxi mately 33 13 of each service group when VR and ECM specifications are used When an initial specification is used the expected mean for financing services and daily services is 30 310 and 40 410 for investment services These tests are only very indicative because the virtual outputs of service groups depend on each other 9 VR assumes implicitly that the variable distributions are normal when using partial covariance to select variables The issue of normality requirement for VR in empirical cases has been brought up previously by Adler and Yazhemsky 2010 An interesting comparison from the managerial perspective is how the virtual outputs reflect the cost of the sales force As the initial input data contained the work of the sales force by job role and the standard unit costs we can make an indicative split of the costs of the sales force to the service groups The cost of the team leaders is allocated to the service groups by the FTEs of the direct sales force An indicative split of the cost of the sales force is fi nancing services 41 investment services 30 and daily banking services 28 The VR specification violates this split clearly while the ECM specification weights are quite well aligned with the re source allocation of the branch network The results show that the apparent equality of the variables may lead in practice to unexpected differences in the importance of the variables and their logical groups in the efficiency evalua tion This may be against the implicit knowledge of management Thus it is worth studying the virtual outputs and inputs gen erated through the variable selection and discussing the findings with management 53 Sensitivity of the variable selection Jenkins and Anderson 2003 have pointed out that the re sults of omitting or including variables in DEA are difficult to pre dict even if the variables are highly correlated The examples that the authors present show that one should be cautious and study the differences between the candidate variable sets We checked the extent to which the top four ranked output combinations see Appendix A provide different results These combinations are compared in Table 9 Interestingly the 1st output combination that we have used as the VR specification in the comparison does not provide the high est discrimination It identifies more efficient DMUs than the other three selections The 4th output combination has both lower aver age scores and higher variance between the scores However the rank correlations indicate that the four options give fairly similar rankings to DMUs The rank correlations of the 1st 2nd and 4th output combinations with the ECM specification are very similar rank correlations 07090764 but the rankings of the 3rd com bination are closer to the ECM rankings rank correlation 0877 Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 8 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Table 9 Comparison of four top output combinations according to the VR procedure Output combination Number of efficient DMUs Average Median Std Dev Min Spearman s rank correlation with 1st selection 1st y 3 y 4 y 9 16 077 076 016 042 1 2nd y 2 y 4 y 9 12 075 075 015 042 0889 3rd y 2 y 4 y 8 13 079 077 013 046 0850 4th y 3 y 4 y 10 13 074 072 018 038 0926 All rank correlations are significant at the 0001 level Thus we can conclude that the 2nd and 4th output combinations take a fairly similar perspective of efficiency to the 1st output combination and the change of one variable does not have a sig nificant impact in our case As described earlier in 42 the results of the ECM technique could potentially be impacted by the RTS assumption However both RTS assumptions converged to the same final variable sets The forward selection procedure of the ECM was also tested The single variable y 4 Mutual funds was considered to be essential and other variables were tested and added to the model in a step wise process following the steps described by Pastor et al 2002 The forward procedure also converged to the same the variable set y 1 y 4 and y 8 Thus in our case the ECM technique ends up with a single variable set regardless of the RTS assumption or the form of procedure 6 Management feedback The management of the branch network reviewed the main results of the initial efficiency estimation and the alternative se lection techniques The managers were provided with a graphi cal presentation of the efficiency scores of branches Because each branch had several yearly observations from 20072010 averages of efficiency scores were used The branchlevel results were more convenient for the managers This however reduced the visibil ity of some extreme observed performances and yearly perfor mance fluctuation The management also reviewed the virtual out puts of the service groups and compared them to the costs of different job roles that were attached to these groups In addi tion the fundamentals of each technique were discussed with the management The management stated that even though it preferred the ap proach of evaluating efficiency using an aggregation of variables based upon the revenue value of services an operational efficiency evaluation could be a supplementary analysis 10 If a wide data set is gathered then it would attempt to use all that information However the management found it hard to accept that a branch could claim itself to be efficient only on the basis of any individ ual service that the bank provides as would be possible with the initial variable set Thus variable selection is justified The managers commented that the principle of using the out put set with the largest overall differences between the observa tions was reasonable ie variance retained In addition the set of variables selected by VR was justified as the management had put emphasis on these services during the years under evaluation However the disadvantage in this case was that the results were impacted by the strongly overweighted daily services The principle of eliminating the least significant outputs accord ing to their contribution to efficiency was considered attractive by the management The output specification obtained from the ECM approach contained the core services with which the branches are constantly engaged The virtual outputs of the services group were 10 Eskelinen and Kuosmanen 2013 and Eskelinen Halme and Kallio 2014 present applications using the revenue valuebased variable aggregation for DEA at the case bank also balanced and aligned with the costs Thus the management favored the ECM over VR if only one reduced set of variables is to be selected 7 Concluding remarks It has been well known that the DEA results are sensitive to variable selection Even in a case where the variables are highly correlated the choice of the variables can affect the results of the efficiency evaluation significantly Dyson et al 2001 This study has demonstrated that the implications may concern not only the efficiency scores and rankings of individual DMUs but also the im portance that the variables and their logical groups have in the ef ficiency evaluation Thus it may affect the conclusions drawn from the results regarding the entire branch network On the other hand this case study also suggests that some con tradictions are natural The techniques compared represented two strategies of variable selection VR emphasized the combination of services with the largest performance differences between the units while the ECM ended up with a set of core outputs in which a high number of units perform relatively well The management found that the techniques resulted in different interpretations of the efficiency measurement The fundamental differences between the selection approaches can benefit decisionmaking in companies and public organiza tions Cooper Seiford and Tone 2007 p 166 state that If we cannot identify the characteristics of the production frontiers by preliminary surveys it may be risky to rely on only one particu lar model When the results from different selection techniques are uniform they confirm the basis for decisions Contradictory re sults regarding an important issue indicate uncertainty and further study may be needed before the decision Both techniques have benefits and drawbacks VR has been criticized because variance retained has little value to managers Pastor et al 2006 In our case the management appreciated the principle of VR and the reduced variable set made sense to them Thus it might be insightful to consider which set of outputs con tains the biggest differences and to discuss whether these differ ences are relevant from the performance improvement point of view The unpredictability of VR has been highlighted by Jenkins and Anderson 2003 The results are sensitive to extreme per formances skewness and also noise which Adler and Yazhemsky 2010 have recognized in their experiments In our case the sen sitivity to anomalies caused an unbalanced weighting of services contradicting the prior knowledge of management The ECMs criterion for variable selection contribution to effi ciency measurement was straightforward to the management and thus it was easily acceptable in our case As the technique avoids elimination with a large overall impact on the efficiency scores it seems to emphasize existing strengths in the network rather than areas where the units have the biggest opportunities to stretch their capabilities This study complements the previous simulationbased compar ative studies of variable selection by employing an empirical case The author hopes that this article can also be helpful to the prac titioners of DEAbased performance management Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 9 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Table A1 Proportion of variance retained three output combinations Ranking of combination Combinations of three output variables Proportion of variance retained 1 y 3 y 4 y 9 87 10 2 y 2 y 4 y 9 86 98 3 y 2 y 4 y 8 86 92 4 y 3 y 4 y 10 86 35 5 y 3 y 5 y 9 85 83 6 y 1 y 4 y 9 85 78 7 y 1 y 4 y 10 85 14 8 y 2 y 7 y 9 84 92 9 y 1 y 5 y 9 84 88 10 y 1 y 7 y 9 84 83 11 y 3 y 5 y 10 84 70 12 y 2 y 5 y 8 84 61 13 y 2 y 6 y 8 84 53 14 y 1 y 7 y 10 84 46 15 y 2 y 5 y 9 84 31 16 y 3 y 5 y 8 84 24 17 y 2 y 7 y 8 83 90 18 y 1 y 6 y 9 83 87 19 y 1 y 6 y 10 83 77 20 y 3 y 4 y 8 83 70 21 y 1 y 4 y 8 83 53 22 y 1 y 5 y 10 83 52 23 y 1 y 5 y 8 83 43 24 y 2 y 7 y 10 83 43 25 y 2 y 4 y 10 83 38 26 y 1 y 6 y 8 83 16 27 y 2 y 6 y 9 82 54 28 y 1 y 7 y 8 82 39 29 y 3 y 6 y 10 81 85 30 y 2 y 5 y 10 81 49 31 y 3 y 7 y 10 80 96 32 y 3 y 7 y 9 80 20 33 y 3 y 6 y 9 80 04 34 y 2 y 6 y 10 79 68 35 y 3 y 6 y 8 79 40 36 y 3 y 7 y 8 77 29 Acknowledgments The author would like to thank Jussi Huttunen and Marko He lin from Helsinki Area Cooperative Bank for the cooperation that has made this paper possible He extends his thanks to Pekka Korhonen Aalto University C A Knox Lovell University of Queensland Maria C A S Portela Portuguese Catholic Univer sity Andy L Johnson Texas AM and two anonymous referees for their suggestions for improving this paper Appendix A Proportion of variance retained according to VR Table A1 presents the three output combinations between ser vice groups The combinations have been ranked by the proportion of variance retained according to the variable reduction procedure of Jenkins and Anderson 2003 Appendix B ECMbased backward elimination procedure by step Table B1 presents the steps of the ECMbased backward elim ination of variables for both the constant and variable returns to scale assumptions The critical value for binomial test B94085 at the 5 statistical significance level is 19 The null hypothesis states that the distributions of the efficiency scores in the original model ie the model from the previous step and the reduced model are statistically indistinguishable Thus if T 19 at tolerance level ϕ 11 the null hypothesis is rejected and the output cannot be elimi nated If there are several outputs that could be removed the out put with the lowest T is eliminated If necessary the tolerance level is reduced in order to identify the variable to be eliminated Table B1 ECMbased backward elimination procedure by step T Constant returns to scale T Variable returns to scale ϕ 11 ϕ 105 ϕ 11 ϕ 105 Step 1 All 10 outputs delete y 1 18 28 delete y 1 10 18 delete y 2 0 0 delete y 2 0 1 delete y 3 1 6 delete y 3 0 6 delete y 4 11 26 delete y 4 14 27 delete y 5 0 2 delete y 5 0 3 delete y 6 1 2 delete y 6 0 2 delete y 7 6 18 delete y 7 4 13 delete y 8 7 12 delete y 8 3 9 delete y 9 0 1 delete y 9 0 1 delete y 10 0 2 delete y 10 0 0 Reduce to 9 outputs eliminate y 2 Reduce to 9 outputs eliminate y 10 ϕ 11 ϕ 105 ϕ 11 ϕ 105 Step 2 9 outputs delete y 1 20 30 delete y 1 10 18 delete y 3 1 7 delete y 2 0 3 delete y 4 13 28 delete y 3 0 6 delete y 5 1 2 delete y 4 16 27 delete y 6 1 4 delete y 5 0 3 delete y 7 6 18 delete y 6 0 4 delete y 8 7 12 delete y 7 5 14 delete y 9 0 1 delete y 8 3 9 delete y 10 0 4 delete y 9 0 1 Reduce to 8 outputs eliminate y 9 Reduce to 8 outputs eliminate y 9 continued on next page Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 10 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Table B1 continued T Constant returns to scale T Variable returns to scale ϕ 11 ϕ 105 ϕ 11 ϕ 105 ϕ 103 ϕ 101 Step 3 8 outputs delete y 1 20 31 delete y 1 10 18 25 35 delete y 3 1 5 delete y 2 0 3 3 8 delete y 4 13 28 delete y 3 0 6 8 13 delete y 5 1 2 delete y 4 16 27 34 42 delete y 6 1 4 delete y 5 0 3 3 6 delete y 7 6 17 delete y 6 3 7 9 14 delete y 8 9 14 delete y 7 6 14 17 23 delete y 9 2 1 delete y 8 8 13 15 19 Reduce to 7 outputs eliminate y 5 Reduce to 7 outputs eliminate y 5 ϕ 11 ϕ 11 ϕ 105 Step 4 7 outputs delete y 1 19 delete y 1 11 20 delete y 3 3 delete y 2 1 4 delete y 4 24 delete y 3 1 6 delete y 6 4 delete y 4 17 30 delete y 7 6 delete y 6 6 9 delete y 8 10 delete y 7 7 16 delete y 9 1 delete y 8 9 13 Reduce to 6 outputs eliminate y 9 Reduce to 6 outputs eliminate y 2 ϕ 11 ϕ 11 Step 5 6 outputs delete y 1 22 delete y 1 15 delete y 3 2 delete y 3 2 delete y 4 32 delete y 4 24 delete y 6 6 delete y 6 6 delete y 7 6 delete y 7 6 delete y 8 15 delete y 8 10 Reduce to 5 outputs eliminate y 3 Reduce to 5 outputs eliminate y 3 ϕ 11 ϕ 11 Step 6 5 outputs delete y 1 21 delete y 1 22 delete y 4 36 delete y 4 24 delete y 6 6 delete y 6 5 delete y 7 5 delete y 7 7 delete y 8 18 delete y 8 12 Reduce to 4 outputs eliminate y 7 Reduce to 4 outputs eliminate y 6 ϕ 11 ϕ 11 Step 7 4 outputs delete y 1 27 delete y 1 21 delete y 4 31 delete y 4 23 delete y 6 9 delete y 7 10 delete y 8 26 delete y 8 16 Reduce to 3 outputs eliminate y 6 Reduce to 3 outputs eliminate y 7 ϕ 11 ϕ 11 Step 8 3 outputs delete y 1 25 delete y 1 25 delete y 4 45 delete y 4 38 delete y 8 36 delete y 8 26 End up with 3 outputs End up with 3 outputs References Adler N Golany B 2001 Evaluation of deregulated airline networks using data envelopment analysis combined with principal component analysis with an application to Western Europe European Journal of Operational Research 132 260273 Adler N Golany B 2002 Including principal component weights to improve discrimination in data envelopment analysis The Journal of the Operational Re search Society 53 985991 Adler N Yazhemsky E 2010 Improving discrimination in data envelopment analysis PCADEA or variable reduction European Journal of Operational Re search 202 273284 Amirteimoori A Despotis D K Kordrostami S 2014 Variable reduction in data envelopment analysis Optimization 63 735745 Banker R D 1993 Maximum likelihood consistency and data envelopment anal ysis A statistical foundation Management Science 39 12651273 Banker R D 1996 Hypothesis test using data envelopment analysis Journal of Productivity Analysis 7 139159 Banker R D Charnes A Cooper W W 1984 Some models for estimating tech nical and scale inefficiencies in data envelopment analysis Management Science 30 10781092 Banker R D Natarajan R 2011 Statistical tests based on DEA efficiency scores In W W Cooper L Seiford J Zhu Eds Handbook on data envelopment analy sis 2nd ed pp 273295 New York Dordrecht Heidelberg London Springer Bian Y 2012 A GramSchmidt process based approach for improving DEA dis crimination in the presence of large dimensionality of data set Expert Systems with Applications 39 685705 Boussofiane A Dyson R G Thanassoulis E 1991 Applied data envelopment analysis European Journal of Operational Research 52 115 Charnes A Cooper W W Rhodes E 1978 Measuring the efficiency of decision making units European Journal of Operational Research 2 4294 4 4 Chen WC Johnson A L 2010 The Dynamics of performance space of ma jor league baseball pitchers 18762006 Annals of Operational Research 181 287302 Cooper W W Seiford L M Tone K 2007 Data envelopment analysis A com prehensive text with models applications references and DEASolver Software 2nd ed New York Springer Dyson R G Allen R Camanho A S Podinovski V V Sarrico C S Shale E A 2001 Pitfalls and protocols in DEA European Journal of Operational Research 132 245259 Eskelinen J Halme M Kallio M 2014 Bank branch sales evaluation using ex tended value efficiency analysis European Journal of Operational Research 232 654663 Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 J Eskelinen European Journal of Operational Research 0 0 0 2016 111 11 ARTICLE IN PRESS JID EOR m5G November 29 20161015 Eskelinen J Kuosmanen T 2013 Intertemporal efficiency analysis of sales teams of a bank Stochastic seminonparametric approach Journal of Banking Finance 37 51635175 Fanchon P 2003 Variable selection for dynamic measures efficiency in the com puter industry International Advances in Economic Research 9 175188 GonzalezBravo M I 2007 PriorRatioAnalysis procedure to improve data envel opment analysis for performance measurement Journal of the Operational Re search Society 58 12141222 Jenkins L Anderson M 2003 A multivariate statistical approach to reducing the number of variables in data envelopment analysis European Journal of Op erational Research 147 5161 Lewin A Y Morey R C Cook T J 1982 Evaluating the administrative effi ciency of courts Omega 10 401411 Nataraja N R Johnson A L 2011 Guidelines for using variable selection tech niques in data envelopment analysis European Journal of Operational Research 215 662669 Paradi J C Yang Z Zhu H 2011 Assessing bank and bank branch performance Modeling considerations and approaches In W W Cooper L M Seiford J Zhu Eds Handbook of data envelopment analysis 2nd ed pp 315362 New York Dordrecht Heidelberg London Springer Paradi J C Zhu H 2013 A survey on bank branch efficiency and performance research with data envelopment analysis Omega 41 6179 Pastor J T Lovell C A K Tulkens H 2006 Evaluating the financial perfor mance of bank branches Annals of Operations Research 145 321337 Pastor J T Ruiz J L Sirvent I 2002 A statistical test for nested radial DEA models Operations Research 50 728735 Podinovski V V Thanassoulis E 2007 Improving discrimination in data envel opment analysis Some practical suggestions Journal of Productivity Analysis 28 117126 Portela M C A S Thanassoulis E 2007 Comparative efficiency analysis of Portuguese bank branches European Journal of Operational Research 177 12751288 Ruggiero J 2005 Impact assessment of input omission on DEA International Jour nal of Information Technology Decision Making 4 359368 Sengupta J K 1990 Test of Efficiency in data envelopment analysis Computers and Operations Research 17 2 121132 Sharma M J Yu S J 2015 Stepwise regression data envelopment analysis for variable reduction Applied Mathematics and Computation 253 126134 Simar L Wilson P W 2001 Testing restrictions in nonparametric efficiency models Communications in Statistics Simulation and Computation 30 159184 SinuanyStern Z Friedman L 1998 DEA and the discriminant analysis of ratios for ranking units European Journal of Operational Research 111 470478 Sirvent I Ruiz J L Borras F Pastor J T 2005 A Monte Carlo evaluation of several tests for the selection of variables in DEA models International Journal of Information Technology Decision Making 4 325343 Tulkens H Vanden Eeckaut P 1995 Nonparametric efficiency progress and regress measures for panel data Methodological aspects European Journal of Operational Research 80 474499 Ueda T Hoshiai Y 1997 Application of principal component analysis for parsi monious summarization of DEA inputs andor outputs Journal of the Operations Research Society of Japan 40 466478 Wagner J M Shimshak D G 2007 Stepwise selection of variables in data en velopment analysis Procedures and managerial perspectives European Journal of Operational Research 180 5767 Please cite this article as J Eskelinen Comparison of variable selection techniques for data envelopment analysis in a retail bank Euro pean Journal of Operational Research 2016 httpdxdoiorg101016jejor201611009 The efficiency of Chinese local banks a comparison of DEA and SFA Benjamin Miranda Tabak Daniel Oliveira Cajueiro e Marina V B Dias January 2014 346 ISSN 15183548 CGC 00038166000105 Working Paper Series Brasília n 346 January 2014 p 115 Working Paper Series Edited by Research Department Depep Email workingpaperbcbgovbr Editor Benjamin Miranda Tabak Email benjamintabakbcbgovbr Editorial Assistant Jane Sofia Moita Email janesofiabcbgovbr Head of Research Department Eduardo José Araújo Lima Email eduardolimabcbgovbr The Banco Central do Brasil Working Papers are all evaluated in double blind referee process Reproduction is permitted only if source is stated as follows Working Paper n 346 Authorized by Carlos Hamilton Vasconcelos Araújo Deputy Governor for Economic Policy General Control of Publications Banco Central do Brasil ComunDipivCoivi SBS Quadra 3 Bloco B EdifícioSede 14º andar Caixa Postal 8670 70074900 Brasília DF Brazil Phones 55 61 34143710 and 34143565 Fax 55 61 34141898 Email editorbcbgovbr The views expressed in this work are those of the authors and do not necessarily reflect those of the Banco Central or its members Although these Working Papers often represent preliminary work citation of source is required when used or reproduced As opiniões expressas neste trabalho são exclusivamente dos autores e não refletem necessariamente a visão do Banco Central do Brasil Ainda que este artigo represente trabalho preliminar é requerida a citação da fonte mesmo quando reproduzido parcialmente Citizen Service Division Banco Central do Brasil DeatiDiate SBS Quadra 3 Bloco B EdifícioSede 2º subsolo 70074900 Brasília DF Brazil Toll Free 0800 9792345 Fax 55 61 34142553 Internet httpwwwbcbgovbrCONTACTUS The efficiency of Chinese local banks a comparison of DEA and SFA Benjamin Miranda Tabak Daniel Oliveira Cajueiro Marina V B Dias The Working Papers should not be reported as representing the views of Banco Central do Brasil The views expressed in the papers are those of the authors and do not necessarily reect those of Banco Central do Brasil Abstract This study investigates to which extent results produced by a single frontier model are reliable based on the application of data envelopment analysis and stochastic frontier approach to a sample of Chinese local banks Our findings show they do produce a consistent trend on efficiency scores over the years However rank correlations indicate they diverge with respect to individual performance di agnosis This shows that these models provide steady information on the efficiency of the banking system as a whole but they become inconsistent at individual level Keywords Bank efficiency Stochastic frontier approach Data envelopment analysis JEL Classification Numbers G21 G28 Universidade Catlica de Brasilia Departamento de Estudos e Pesquisas Banco Cen tral do Brasil Departamento de Economia Universidade de Braslia Universidade de Braslia 3 1 Introduction This paper applies both data envelopment analysis DEA and stochastic frontier approach SFA to investigate the efficiency of Chinese local banks We analyze some features related to the efficiency scores of both approaches We determine whether the methodology applied is relevant to the determination of best and worst performers and to the conclusions drawn regarding the whole market These methods are not consistent in the determination of individual performance Nevertheless they do produce a similar trend on efficiency over the years Efficiency of financial institutions can be measured by parametric or nonparametric models Both approaches have their features but few studies apply both types of method ologies to the same sample to investigate the consistency between their results Some of these studies consider the conditions Bauer et al 1998 propose to investigate this issue Fiorentino et al 2006 There are mixed results regarding the similarity of these models Some authors find that the consistency between DEA and SFA in ranking banks accord ing to their performance is low or even statistically insignificant Fiorentino et al 2006 Ferrier and Lovell 1990 On the other hand comparison of their results at industry level can point out the similarities between them Resti 1997 There are two main points we make that differ from previous literature on the com parison of DEA and SFA First we do not rule the methodologies as either consistent or inconsistent We analyze at which level their results are more reliable produce similar information and to which extent they must be considered with care Second we argue that the comparison of the behavior of efficiency along the sample period is much more relevant to determine the consistency between both models than the average levels of efficiency Parametric and nonparametric methodologies are extensively applied to measure the efficiency of financial institutions In many cases the purpose of the existing studies is to assist policy makers with relevant information such as the impact of regulation measures over the performance of these firms Lee and Chih 2013 Barth et al 2013 Gaganis and Pasiouras 2013 However it is common to apply only one frontier model either parametric or nonparametric to measure efficiency Investigating if these models are consistent has the purpose of pointing out to which extent results and conclusions obtained from the application of only one frontier method are reliable especially to assist policy decisions Studies of bank efficiency in China are mostly related to the effects of deregulation re forms over the performance of financial institutions Berger et al 2009 Sun et al 2013 Some of them as Yin et al 2013 conclude that these reforms led to an improvement on bank efficiency Our study finds stable efficiency scores over the period 20012012 which is not in line with the idea that deregulation reforms still have a positive impact on the efficiency of Chinese financial institutions 4 2 Methodology and Data Bank efficiency can be measured through basically two different types of models parametric or nonparametric The most applied parametric model is the stochastic frontier approach and the most applied nonparametric model is data envelopment analysis In this paper we apply data envelopment analysis and stochastic frontier approach to measure the efficiency of Chinese commercial banks The advantages of using DEA is that it does not require prior knowledge of either the distributional form of the inefficiency term or the production technology used in the industry The main feature of the SFA is that it accounts for random shocks Berger and Humphrey 1997 Seiford and Thrall 1990 Mester 1996 21 Data Envelopment Analysis Data envelopment analysis was first proposed by Charnes et al 1978 and was developed based on the seminal work of Farrell 1957 It uses linear programming techniques to build a nonparametric efficiency frontier of the data sample Therefore the frontier is constructed by the practices combination of inputs and outputs of the most efficient firms in the sample The linear programming problem consists of the maximization of the firms weighted outputinput ratio In this study we apply DEA to measure inputoriented outputoriented economic and allocative efficiencies We estimate the variable returns to scale model proposed by Banker et al 1984 The inputoriented model consists of the following linear programming problem minϕz ϕ sa qi Qλj 0 ϕxi Xλj 0 i1n λi λi 0 1 in which ϕ is technical inefficiency qi are the outputs for firm i Q and X are matrices of outputs and inputs for all firms respectively xi is the vector of inputs for the ith firm and λi is a vector of weights Allocative efficiency measures the quality of the combination of inputs to produce certain outputs considering input prices Sengupta 1999 Economic efficiency analyzes both technical and allocative efficiencies Bauer et al 1998 They can be measured by a cost minimization problem that determines the optimal levels of inputs to be used given their prices wi minwiχ wiχi sa j λj yj yi 0 χi λj xj 0 λj 0 2 i λi 1 3 Economic efficiency EE will be determined by the following ratio EEi fracwi χiwi xi 4 Allocative efficiency will be measured by the ratio cost efficiency over inputoriented technical efficiency TEinputs AEi fracEEiTEinputsi 5 22 Stochastic Frontier Approach The stochastic frontier approach was simultaneously developed by Meeusen and Van den Broek 1977 and Aigner et al 1977 and it estimates a parametric frontier of the best possible practices given a standard cost or profit function Both cost and profit functions are comparable to DEA economic efficiency measures since they consider the same efficiency concept Bauer et al 1998 We estimate a cost function because it is more commonly applied in the literature Berger and Mester 1997 Since SFA builds a parametric frontier it is necessary to specify the production technology and the distribution of the inefficiency term We use a translog form of the cost function since it is a flexible functional form Berger et al 2009 LozanoVivas and Pasiouras 2010 Also the inefficiency term v has a halfnormal distribution while the random error ν is normally distributed The translog cost function is written as follows lnCTw2 δ0 j δ1 lnyjit frac12 j k δjk lnyjit lnykit β1 lnw1w2it frac12 β11 lnw1w2it lnw1w2it j θj lnyjit lnw1w2it ln vit ln νit 6 in which CT is the firms total costs i and t stand for bank and time respectively In the true fixed effects model the inefficiency term vit is composed by a set of dummy variables which determine its behavior across time This study considers three outputs and two inputs Thus w1 and w2 are the two inputs used to produce the four outputs yj The normalization by the price of the last input w2 guarantees price homogeneity In this study we apply two different specifications of the stochastic frontier model the true fixed effects model from Greene 2005 and the model proposed by Battese and Coelli 1995 The true fixed effects model fits the sample better so we refer to this model when considering the parametric efficiency results1 We obtain efficiency scores from the Jondrow et al 1982 estimator 23 Data Our sample comprises an unbalanced panel with 461 yearly observations from 65 Chinese local commercial banks for the period 20012012 The data source is Bankscope Our definition of local banks includes city commercial banks and rural commercial banks Berger et al 2009 To analyze the efficiency trend over the years we create subsamples for each year and apply the DEA model for all of them For the stochastic frontier model we use the specification of panel data This paper uses the intermediation approach to define inputs and outputs for the empirical application which states that banks capture borrowed funds and use capital and labor to turn them into loans and other assets Sealey and Lindley 1977 Therefore we specify the outputs as deposits loans and liquid assets The input quantities are total interest expenses and total noninterest expenses Input prices are the ratios total interest expenses over deposits and noninterest expenses over fixed assets Table 1 shows descriptive statistics on these variables Table 1 about here 3 Empirical Results Results regarding the individual performance of firms should be looked into with care while general results related to the performance of the whole market are consistent Both data envelopment analysis and stochastic frontier approach results show that efficiency scores are roughly stable over the years which we report on Figure 1 This finding does not confirm the upward trend Yin et al 2013 observe for bank efficiency in China after 2001 From the results we report on tables 2 and 3 our scores 1The loglikelihood for the true fixed effects model is 249051 while for the Battese and Coelli 1995 model it is 6179308 7 do seem to vary over the years at first sight DEA scores seem to show a drawback in the period 20012007 with posterior efficiency gains SFA scores seem to indicate a slight improvement on efficiency over the sample period However a closer look into mean efficiency scores and their volatility which we report on Figure 1 leads to the conclusion that the performance of Chinese local banks does not show a statistically significant change over the years A possible reason for divergent results is that our sample comprises only Chinese local banks which includes only city commercial banks and rural commercial banks while Yin et al 2013 use a sample with different types of Chinese commercial banks Table 2 about here Table 3 about here From Figure 1 we observe that efficiency levels are in some periods higher according to results from the parametric methodology Nevertheless we point out that this difference is not relevant if both models provide the same conclusion over the banking systems performance Efficiency levels themselves are not enough to rule for the inconsistency between different methodologies In this case both models indicate that Chinese local banks have not experienced efficiency improvements over the period 20012012 This shows that for this empirical study DEA and SFA do provide similar conclusions with respect to the performance of the industry of Chinese local banks as a whole Figure 1 about here To analyze the consistency between individual efficiency scores produced by the two methodologies we compute Spearmans rank correlation between DEA economic effi ciency and SFA cost efficiency since they use the same efficiency concept Bauer et al 1998 We find a rank correlation of 26 which is not statistically significant at 1 level and is consistent with the findings of Ferrier and Lovell 1990 who compute the correlation of 14 2 Fiorentino et al 2006 also find a positive but low correlation between data envelopment analysis and the stochastic frontier approach between 44 and 58 One possible explanation for the difference between both models is that DEA is more sensitive to sample heterogeneity than the SFA Fiorentino et al 2006 To check if this argument is plausible we run a robustness check in which we use the DEA results of efficiency to remove efficient banks from the original sample Sample A leaving us with a sample of banks that are originally economically inefficient Sample B and a sample of banks that are originally technically inefficient Sample C We find that efficiency scores 2We compute the spearman rank correlation between scores relative to the year 2009 which is the year which comprises the greater number of observations 8 do not differ among these samples and thus heterogeneity is not the reason why DEA and SFA do not produce similar results These results are available from the authors upon request We do not repeat the same to the SFA estimation because we use the results of the true fixed effects model which already accounts for heterogeneity Greene 2005 It should be expected that DEA and SFA might produce contradictory results in some empirical applications Berger and Humphrey 1997 point out the conflict be tween parametric and nonparametric methodologies based on the fact that they have different degrees of dispersion and rank banks differently This latter conflict is exactly the inconsistency we report here From a theoretical perspective DEA and SFA are very different Data envelopment analysis considers a deterministic frontier while the stochastic frontier model considers a parametric one which incorporates states of nature For this reason Bauer et al 1998 propose some consistency conditions these models should meet in order to be sure they are providing policy makers with reliable infor mation The fact that DEA does not account for random shocks may be the source of the inconsistency between both models Fiorentino et al 2006 The implication of this inconsistency in the framework analyzed here is that individual performance should be analyzed with caution The application of only one methodology to determine best and worst performers may lead to wrong conclusions especially when rank correlation among different models are not statistically significant 4 Conclusions We apply both data envelopment analysis and the stochastic frontier model to a sample of Chinese local banks We analyze the consistency between these measures from micro and macro perspectives The majority of studies on Chinese bank efficiency apply either a parametric or a nonparametric methodology to investigate the performance of the countrys financial institutions Our findings show that these models are not consistent in the individual analysis of efficiency and results obtained at a micro level should be dealt with care However they do provide similar results regarding the behavior of average efficiency scores for the whole market over the years We argue that this behavior is more important to determine the consistency between both models than the average efficiency levels themselves We also conclude that Chinese local banks do not show improvement on performance over the period 20012012 References Aigner D Lovell C and Schmidt P 1977 Formulation and estimation of stochastic frontier production function models Journal of Econometrics 62137 9 Banker R Charnes A and Cooper W 1984 Some models for estimating technical scale inefficiencies in data envelopment analysis Management Science 3010781092 Barth J R Lin C Ma Y Seade J and Song F M 2013 Do bank regulation supervision and monitoring enhance or impede bank efficiency Journal of Banking Finance 3782879 2892 Battese G and Coelli T 1995 A model for technical inefficiency effects in a stochastic frontier production for panel data Empirical Economics 20325332 Bauer P W Berger A N Ferrier G D and Humphrey D B 1998 Consistency conditions for regulatory analysis of financial institutions A comparison of frontier efficiency methods Journal of Economics and Business 5085114 Berger A Hasan I and Zhou M 2009 Bank ownership and efficiency in china What will happen in the worlds largest nation Journal of Banking Finance 33113130 Berger A and Mester L 1997 Inside the black box what explains differences in the efficiencies of financial institutions Journal of Banking Finance 21895947 Berger A N and Humphrey D B 1997 Efficiency of financial institutions Inter national survey and directions for future research European Journal of Operational Research 98175212 Charnes A Cooper W and Rhodes E 1978 Measuring the efficiency of decision making units European Journal of Operational Research 3429444 Farrell M 1957 The measurement of productive efficiency Journal of the Royal Statistical Society 120A253281 Ferrier G and Lovell C 1990 Measuring cost efficiency in banking Econometric and linear programming evidence Journal of Econometrics 46229245 Fiorentino E Karmann A and Koetter M 2006 The cost efficiency of german banks a comparison of sfa and dea Discussion Paper 102006 Deutsche Bundesbank Gaganis C and Pasiouras F 2013 Financial supervision regimes and bank efciency International evidence Journal of Banking Finance 3754635475 Greene W 2005 Reconsidering heterogeneity in panel data estimators of the stochastic frontier model Journal of Econometrics 26269303 Jondrow J CAK L IS M and Schmidt P 1982 On the estimation of technical inefficiency in the stochastic frontier production function model Journal of Economet rics 19233238 Lee TH and Chih SH 2013 Does financial regulation affect the profit efficiency and risk of banks evidence from chinas commercial banks The North American Journal of Economics and Finance 26705 724 10 LozanoVivas A and Pasiouras F 2010 The impact of nontraditional activities on the estimation of bank efficiency International evidence Journal of Banking Finance 3414361449 Meeusen W and Van den Broek K 1977 Efficiency estimation from cobbdouglas production functions with composed error International Economic Review 18435 444 Mester L 1996 A study of bank efficiency taking into account riskpreferences Journal of Banking Finance 2010251045 Resti A 1997 Evaluating the costefficiency of the italian banking system What can be learned from the joint application of parametric and nonparametric techniques Journal of Banking Finance 21221250 Sealey C and Lindley J 1977 Inputs outputs and a theory of production and cost of depository financial institutions Journal of Finance 3212511266 Seiford L and Thrall R 1990 Recent developments in dea The mathematical pro gramming approach to frontier analysis Journal of Econometrics 46738 Sengupta J K 1999 A dynamic efficiency model using data envelopment analysis International Journal of Production Economics 62209218 Sun J Harimaya K and Yamori N 2013 Regional economic development strate gic investors and efficiency of chinese city commercial banks Journal of Banking Finance 3716021611 Yin H Yang J and Mehran J 2013 An empirical study of bank efficiency in china after WTO accession Global Finance Journal 11 Figure 1 Comparison of efficiency levels from DEA and SFA over the years 12 Table 1 Variables used on efficiency estimation Year Total Expenses Loans Liquid Assets Deposits Total interest expenses Total noninterest expenses w1 w2 2001 Mean 612951 8602172 3543838 1253685 3599656 2529854 00309 00145 SD 2319506 2854423 4474053 4435583 196382 1201174 00149 00023 N 11 11 11 11 11 11 11 11 2002 Mean 6517423 1087913 5200032 1552868 3827805 2689618 00255 00125 SD 3359987 5313906 7924992 8472038 2365301 135607 00136 00043 N 18 18 18 18 18 18 18 18 2003 Mean 7869842 1330581 5680322 1954243 4762005 3107837 00252 00116 SD 3995986 7105158 7198977 1060500 2836826 1646898 00129 00024 N 19 19 19 19 19 19 19 19 2004 Mean 781695 1439702 5217968 2179900 4509097 3307853 00205 00121 SD 487802 7771709 6370221 1289392 3250524 2147764 00111 00032 N 22 22 22 22 22 22 22 22 2005 Mean 1185633 2178598 8771696 3418808 7011715 4844617 00235 00119 SD 101892 2021322 1167075 3289473 6299166 4699437 00147 00029 N 33 33 33 33 33 33 33 33 2006 Mean 1299445 2449558 8431636 3727860 7578404 5416049 00227 00119 SD 1224961 2367041 1045025 3798744 7248159 5711323 00122 00030 N 45 45 45 45 45 45 45 45 2007 Mean 1695035 3093186 1274486 4811405 1063485 6315506 00234 00114 SD 162724 2988969 1247269 4916175 1056608 6397875 00098 00039 N 51 51 51 51 51 51 51 51 2008 Mean 2698787 4219558 2038509 6917867 1765891 9328953 00265 00126 SD 2876577 4839686 2912078 8074015 2022744 9439254 00097 00047 N 60 60 60 60 60 60 60 60 2009 Mean 2703437 5809675 2693320 9468915 1564529 1138907 00169 00108 SD 2860058 6701814 3393745 111E07 1772829 1140117 00055 00028 N 61 61 61 61 61 61 61 61 2010 Mean 360655 7451999 5140042 128E07 2107509 1499042 00161 000981 SD 3914733 8544235 6012537 143E07 2423446 1555168 00054 00026 N 59 59 59 59 59 59 59 59 2011 Mean 6445085 9653769 745E06 162E07 4311683 2133402 00264 001068 SD 6906545 110E07 894E06 173E07 5023891 206458 00111 00027 N 57 57 57 57 57 57 57 57 2012 Mean 1399510 190E07 148E07 304E07 1008609 3909015 00326 00094 SD 1220445 169E07 128E07 255E07 9501927 2967428 00075 00018 N 25 25 25 25 25 25 25 25 Ratio total interest expenses over deposits Ratio total noninterest expenses over total assets 13 Table 2 DEA economic efficiency scores Year Mean Std Dev Min Max 2001 084 024 036 1 2002 076 025 033 1 2003 073 026 033 1 2004 070 024 036 1 2005 074 026 019 1 2006 073 024 028 1 2007 056 024 022 1 2008 071 017 035 1 2009 073 014 048 1 2010 078 014 048 1 2011 079 017 031 1 2012 087 014 060 1 14 Table 3 SFA efficiency scores Cost efficiency Year Mean Std Dev Min Max 2001 089 004 081 093 2002 088 004 075 093 2003 088 005 076 093 2004 091 004 082 096 2005 092 003 085 096 2006 092 003 083 095 2007 092 002 080 095 2008 093 002 079 096 2009 092 002 087 096 2010 092 002 087 096 2011 092 002 086 095 2012 093 002 085 095 15 Journal of Econometrics 6 1977 2137 0 NorthHolland Publishing Company FORMULATION AND ESTIMATION OF STOCHASTIC FRONTIER PRODUCTION FUNCTION MODELS Dennis AIGNER University of Wisconsin Madison WI 53706 USA University of Southern Califbrnia Los Angeles CA 90007 USA CA Knox LOVELL University of North Carolina Chapel Hill NC 27514 USA Peter SCHMIDT University of North Carolina Chapel Hill NC 27514 USA Received October 1975 final version received October 1976 Previous studies of the socalled frontier production function have not utilized an adequate characterization of the disturbance term for such a model In this paper we provide an ap propriate specification by defining the disturbance term as the sum of symmetric normal and negative halfnormal random variables Various aspects of maximumlikelihood estimation for the coefficients of a production function with an additive disturbance term of this sort are then considered 1 Introduction The theoretical definition of a production function expressing the maximum amount of output obtainable from given input bundles with fixed technology has been accepted for many decades And for almost as long econometricians have been estimating average production functions It has only been since the pioneering work of Farrell 1957 that serious consideration has been given to the possibility of estimating socalled frontier production functions in an effort to bridge the gap between theory and empirical work For a variety of reasons these efforts have not been completely successful In this paper we suggest a new approach to the estimation of frontier production functions This involves the specification of the error term as being made up of two components one normal and the other from a onesided distribution This approach enables us to overcome some of the major shortcomings of previous work in the area This work was supported in part by NSF Grant GS39995 DJA We are indebted to Takeshi Amemiya Forrest Nelson and Dale J Poirier for helpful advice at various stages of the research and to Chinbang Chung for research assistance None of them should be held responsible for any errors that remain 22 D Aigner et al Stochastic frontier production function models The outline of the paper is as follows In section 2 we review previous app roaches to the estimation of parametric frontier production functions and in section 3 we propose a new approach The statistical properties of our model are discussed in section 4 Some Monte Carlo experiments are reported in section 5 and empirical examples are given in section 6 Section 7 concludes 2 Previous parametric frontier models Previous work on the estimation of parametric frontier production functions as characterized by the work of Aigner and Chu 1968 Afriat 1972 and Richmond 1974 begins by assuming a function giving maximum possible output as a function of certain inputs For a given firm say the ith we write Yi Axi i PI 1 Here yi is the maximum output obtainable from xi a vector of nonstochastic inputs and 1 is an unknown parameter vector to be estimated Aigner and Chu 1968 suggest the estimation of I by mathematical program ming methods based on a crosssection of N firms within a given industry Specifically they suggest minimization of subject to yi 6 fxi p which is a linear programming problem if fxi p is linear in 1 Alternatively they suggest minimization of subject to the same constraint which is a quadratic programming problem if fXi fl is linear Obviously something magical has happened in moving from 1 to either of these estimation methods In order to characterize differences in output among firms with identical input vectors or to explain how a given firms output lies below the frontier fni fi a disturbance term has been implicitly assumed One problem with these approaches is extreme sensitivity to outliers This has led to the development of socalled probabilistic frontiers Timmer 1971 Dugger 1974 which are estimated by the same types of mathematical pro gramming techniques discussed above except that some specified proportion of the observations is allowed to lie above the frontier The selection of this proportion is essentially arbitrary lacking explicit economic or statistical justifi D Aigner et al Stochastic frontier production function models 23 cation Another problem involves reconciling the observations above the frontier with the concept of the frontier as maximum possible output Typically this is accomplished by appealing to measurement error in the extreme observations However it seems preferable to incorporate the possibility of measurement error and of other unobservable shocks in a less arbitrary fashion As they have been applied previously therefore the mathematical program ming techniques do not lead to estimates with known statistical properties In an attempt to give them a statistical basis Schmidt 1976 explicitly added a onesided disturbance to 1 above which yields the model Yi fXxi Bi i lN where si 5 0 Given a distribution assumption for the disturbance term the model can then be estimated by maximumlikelihood techniques In particular the assumption that si has an exponential distribution leads to the linear programming technique while the assumption that si has a halfnormal distri bution leads to the quadratic programming technique Therefore Aigner and Chus estimates can be viewed as maximumlikelihood estimates under particular error specifications Unfortunately the observation that the model can be estimated by maximum likelihood techniques and that under appropriate assumptions linear and quadratic programming are maximumlikelihood techniques is of little practical value This is so because the usual regularity conditions for the application of maximum likelihood are violated In particular since yi 5 fxi p the range of the random variable y depends on the parameters to be estimated Therefore the usual theorems cannot be invoked to determine the asymptotic distributions of parameter estimates Under these circumstances it is not clear just how much we know about the frontier after having estimated it In another recent paper Aigner Amemiya and Poirier 1976 construct a more reasonable error structure than a purely onesided one Specifically they assume Ei i VT1 a if E 0 i 1 N Tlze if ET 5 0 3 where the errors 7 are independent normally distributed random variables with zero means and variance c2 for 0 8 1 otherwise E has either the negative or positive truncated normal distribution when 0 1 or 0 0 respectively Their justification for this error specification is that firms are presumed to differ in their production of y for a given set of values for the inputs according to random variation in 1 their ability to utilize best practice technology 24 D Aigner et al Stochastic frontier production function mode a source of error that is onesided si 5 0 andor 2 an input quantity or measurement error in y a symmetric error The parameter f3 is interpreted as the measure of relative variability in these two error sources its values circum scribing the full frontier function 0 l the average function 0 and intermediate cases of some interest A primary contribution of this error structure to the literature is that it allows the placement of the fitted function to be estimated along with the usual para meters of interest through the parameter 8 Thus the criticism levied at the average function by proponents of the frontier eg Aigner and Chu 1968 and criticisms that accompany strict use of the frontier or envelope function as the appropriate industry production function cf Timmer 1971 are ameliorated by this more accommodating specification Nevertheless the interpretation of 6 as a measure of the relative variability of error sources is only implicit in the Aigner Amemiya Poirier formulation A more direct approach is to specifically model the error process implied by the behavioral considerations mentioned above 3 A stochastic frontier We now return to the model as given in eq 2 but under the error structure Ei UiUi i 1 N 4 The error component z represents the symmetric disturbance the pi are assumed to be independently and identically distributed as N0 0 The error component U is assumed to be distributed independently of vi and to satisfy ui 5 0 We will be particularly concerned with the case in which ui is derived from a N0 rr distribution truncated above at zero However other onesided distributions are tenable and we will also briefly consider the case in which ui has an exponential distribution This model collapses to a deterministic frontier model when 01 0 and it collapses to the Zellner Kmenta and Dreze 1966 stochastic production func tion model when 0 0 Note that yi 5 fXi p vi so that the frontier itself is now clearly stochastic The economic logic behind this specification is that the production process is subject to two economically distinguishable random disturbances with different characteristics We believe that there is ample precedent in the literature for IAs 0 f 1 the positive error component has a large variance hence small influence in the likelihood function and the negative error dominates This gives rise to the full frontier as the limiting case 0 1 A similar interpretation follows for the case of 0 P 0 athough a behavioral explanation for this situation is lacking When 0 3 the likelihood function has the form of a mixture of two halfnormals each with equal influence D Aigner et al Stochastic frontier production function models 25 such a view although our interpretation is clearly new2 And from a practical standpoint such a distinction greatly facilitates the estimation and interpreta tion of a frontier The nonpositive disturbance ui reflects the fact that each firms output must lie on or below its frontier fxi p vi Any such deviation is the result of factors under the firms control such as technical and economic inefficiency the will and effort of the producer and his employees and perhaps such factors as defective and damaged product But the frontier itself can vary randomly across firms or over time for the same firm On this interpretation the frontier is stochastic with random disturbance vi 0 being the result of favorable as well as unfavorable external events such as luck climate topo graphy and machine performance Errors of observation and measurement on y constitute another source of vi 5 0 One interesting byproduct of this approach is that we can estimate the variances of Vi and Ui so as to get evidence on their relative sizes Another implication of this approach is that productive efficiency should in principle be measured by the ratio YilLfCxi B 21i13 5 rather than by the ratio This simply distinguishes productive inefficiency from other sources of disturb ance that are beyond the firms control For example the farmer whose crop is decimated by drought or storm is unlucky on our measure 5 but inefficient by the usual measure 63 Our discussion of estimation will be simplified somewhat if we consider a linear model We therefore write in obvious matrix form y xgs 7 in place of 2 where now a vu Marschak and Andrews 1944 suggest that the sum vi ui reflects the technical efficiency and the will effort and luck of a producer Zellner Kmenta and Dreze 1966 suggest that it reflects factors such as weather unpredictable variations in machine or labor performance and so on and they were perhaps the first to propose a stochastic production function although they clearly did not have a frontier in mind Other characterizations of the error term exist Aigner and Chu 1968 explain it by reason of technical and economic inefficiencies as well as by pure random shocks in the production process that might be due to careless handling and defective or damaged output Timmer 1971 cites technical and economic inefficiency as well as definitional and measurement problems in the variables And agri cultural economists frequently cite variation across farms in such environmental conditions as climate topography and soil type as indicative of a random production function 3As defined 5 is a strange efficiency measure since it is stochastic and depends on an unobservable vi It is offered here only to support the argument 26 D Aigner et al Stochastic frontier production function modeis 4 Estimation of the stochastic frontier model The distribution function of the sum of a symmetric normal random variable and a truncated normal random variable was apparently first derived by MA Weinstein 1964 The derivation of the density function of E is straight forward so we shall not include it here The result is 2 fE f E lFdC 0 0 cosEfCo Cr where IS 0 og 1 cc and f and F are the standard normal density and distribution functions respectively This density is asymmetric around zero with its mean and variance given by 42 EE Eu JJ b VE Vu Vu n2 6 n D U as can be easily ascertained from elementary considerations and calculation of the moments of U The particular parameterization in 8 is convenient because A is thereby interpreted to be an indicator of the relative variability of the two sources of random error that distinguish firms from one another4 A2 0 implies 03 co andor 0 f 0 ie that the symmetric error dominates in the determination of E Eq 8 then becomes the density of a N0 02 random variable as can be seen by inspection Similarly when C 0 the onesided error becomes the domi nant source of random variation in the model and 5 takes on the form of a negative halfnormal 5 We prefer to use this interpretation of I even though aU2 is not the variance of UT 2n UUZ is Another useful parameterization is to use oz along with c ajo For oVz 0 and thus 1 03 5 becomes J2 f 4 no exp 1 E2 2a2 1 for E j 0 zz 0 otherwise D Aigner et al Stochastic frontier production function models 27 The estimation problem is posed by assuming we have available a random sample of N observations and then forming the relevant loglikelihood function 12 In 9 yjI A 02 N lndT N In CI 2 In lFl 10 il which is almost exactly the form of the Amemiya 1973 p 1015 eq 102 Taking derivatives 1 ikelihood function considered by a In 2 jyr a In 8 i iI YiBxi9 an alns 1 N i iFI YiPiri i xi afi il i 11 12 where a k x 1 vector consisting of elements in the ith row of X and f and FT are respectively the standard normal density and distribution func tions evaluated at yi xi lo Given 12 we have that N fi ill 1 FT YiBXi 0 at the optimum Inserting this result into ll the ML estimator for CJ is determined through N tf 2a4 YiBxi O 2a2 rl 14 which yields A2 0 iil YiBxi2 15 28 D Aigner et al Stochastic frontier production function models the basis for the usual ML estimator of residual variance in a regression model But the determination of fi is not independent of 8 from other equations In any event this result can be used as a basis for an iterative solution scheme I premultiplied into 13 gives Adding to this j2 times eq 12 and simplifying we get 16 17 which in conjunction with 13 gives a system of k 1 equations that cor responds very closely to the system of firstorder equations encountered in the socalled Tobit model6 Since our density function is continuous in the range of E it is not anticipated that the difficulties encountered in the Tobit model will occur here and therefore we claim all the usual maximumlikelihood properties for the values of p 1 and a2 which simultaneously equate ll IZ and 13 to zero Formal proof of this claim and an examination of the regularity conditions that support it follow the analysis in Amemiya 1973 Various solution algorithms are available for finding the optimizing values of R and u Most of these the FletcherPowell algorithm for exampe require analytical first or secondorder derivatives in addition to the likelihood function itself for their best performance at reasonable cost in computer time Since such algorithms are now readily availabIe7 we will not devote any space to a discussion of the ML computational problem except to note that this likelihood function seems to be wellbehaved based on our experience Second order derivatives are presented in the appendix to this paper for that use and as a basis for calculating asymptotic standard errors of the ML estimates We note in passing that if estimation of p alone is desired all but the coefficient in 3 corresponding to a column of ones in X is estimated unbiasedly and con sistently by least squares Moreover the components of C can be extracted ie consistent estimators for them can be found based on the least squares results by utilizing eq 9 for VE in terms of 0 and C and a similar relation ship for a higherorder moment of e since VE and higher order meancorrected See Amemiya 1973 p 1011 eqs 72 73 7For a good discussion of the available algorithms as of a few years ago see Goldfeld Quandt 1971 D Aigner et al Stochastic frontier production function models 29 moments of E are themselves consistently estimable from the computed least squares residuals8 Similar comments andderivations would applyunder alternative distributional assumptions for ui For example we could choose the simple oneparameter exponential distribution for u lf9 where 4 2 0 is the mean of ui The variance is 4 A little algebra reveals that the distribution of ei viui is given by the density IFczexpb 19 where again F represents the cumulative distribution function of the standard normal distribution The likelihood function for the model follows immediately 5 Monte Carlo results In order to discover some specific information about the small sample be havior of the ML estimators discussed in the previous section we constructed two limited Monte Carlo experiments which rest entirely on artificial data Study 1 The model considered here is yi ei i 1 N where Ei is generated by eq 8 with various values of 0 and 0 The results and further details about the range of parameter values considered are reported in table 1 So no regression is involved in this instance indeed not even a mean is esti mated and we assume that the conventionally unobservable Ei is observable Some easily distinguished patterns of bias and precision in estimation emerge Considering table la 1 is apparently biased upward ol a and a2 seems to be estimated very well in all cases both with regard to small bias and MSE Moving to table lb increasing sample size to N 100 for the one set of results reported shows little if any basis for alteration of the qualitative con clusions reached above but also gives to indication of the reduction in bias and increase in precision promised by the asymptotic properties of ML sFor instance the thirdorder moment of E is 203 Er13 22jZj 1 In the Monte Carlo results that follow no attempt is made to evaluate the properties of these estimators C 30 D Aigner et al Stochastic frontier production function models Table la Monte Carlo results for the model yi si number of replications 100 sample size 50 value of A 166 124 083 166 124 083 141 b 136 008 122 i2 122 006 101 CP 102 004 1 177 028 li 131 020 x 093 013 1036 101 007 074 82 075 007 041 BZ 047 005 038 8 036 002 048 BUZ 047 002 060 8u2 055 003 188 a2 181 014 163 tP 164 011 135 8 136 008 n 180 033 i 135 017 t 094 013 138 aU2 134 016 099 82 103 011 055 aU2 062 009 050 b2 047 003 064 6 061 004 080 82 074 004 281 Bz 281 033 245 bZ 246 030 203 6 205 019 z 176 035 1 134 022 z 084 008 206 6u2 206 042 149 a2 153 030 083 6 085 018 075 82 075 007 096 BUZ 093 010 120 B2 120 009 Values in parentheses to the right of estimates are MSEs Values in parentheses to the left of2 b 2 u and 8 are the true values for these parameters used in the simulations Table lb Monte Car10 results for the model yt cl number of replications 100 sample size 100 value of Iz 166 124 0838 166 124 083 188 a2 188 014 163 82 170 012 135 6 132 008 z 181 037 5 134 023 x 090 012 138 8u2 140 015 099 62 106 014 055 a2 057 007 050 802 048 003 064 2 064 004 080 8 075 005 Values in parentheses to the right of estimates are MSEs Values in parentheses to the left of I and B are the true values for these parameters used in the simulations Study 2 Here the experiments of table 1 are repeated using the model yi jLLi i 1 IV Since si already possesses a nonzero mean the question is what effect does the extraction of an explicit intercept term have on the previous case From table 2a we see that qualitatively the results for 1 are altered somewhat with some cases of negative bias occurring Moreover the additional parameter to be estimated has a deleterious effect on 8 and hence a2 Whereas in the previous cases these estimators provided sharp results now there is a tendency for them both to be biased downward I is also biased downward but without any apparent pattern of relationship with the values of other parameters Again increasing sample size to N 100 has no perceptible effect on our results and conclusions from the N 50 case D Aigner et al Stochastic frontier production function models 31 Table 2a Monte Carlo results for the model y p cl number of replications 100 sample size 50 fl 10 value of 1 166 124 O838 166 124 083 141 P 134 022 1226 111 025 101 r 082 013 n 172 122 a 104 120 1 023 066 088 011 103 2 095 034 075 021 074 2 055 025 058 044 041 euz 012 023 038 Boz 039 004 048 a2 053 005 060 6 070 004 188 82 175 039 163 a2 161 051 135 a2 130 030 z 193 299 145 273 070 091 084 018 138 au2 125 065 082 025 099 zz oo 083 071 023 055 050 052 050 BZ 050 007 064 8 061 009 080 a2 080 010 81 8 256 089 245 8 240 093 203 s2 202 048 z 184 239 146 132 2 081 079 083 023 206 zU2 180 135 085 029 149 155 148 078 025 083 zU2 086 105 075 8 076 014 096 8 086 018 120 82 116 022 BValues in to the of estimates in parentheses to the of are the true values for these parameters used in the simulations Table 2b Monte Carlo results for the model y t8 number of replications 100 sample size 100 p 10 value of 1 166 124 083 166 124 083 188 B2 190 047 163 a2 179 055 135 a2 117 022 i 203 248 z 157 155 a 043 061 088 017 138 2 140 079 093 019 099 zUz 122 087 062 024 055 z 029 039 050 a 050 008 064 B 057 008 080 802 087 007 Values in parentheses to the right of estimates are MSEs Values in parentheses to the left ofbZ B 2andbZ t are the true values for these parameters used in the simulations Contrasting these results to similar experiments conducted by the Aigner Amemiya and Poirier 1976 the overall performance of common estimators 2 and S2 is roughly comparable We note that their measure of the relative variability in error sources is generally overstated but has small MSE compared to 1 9Cf their table 5 on page 20 The disturbance variance VE is set at 05 in their results whereas it varies in our table 2a and is generally much larger than 05 32 D Aigner et al Stochastic frontier production function models 6 Empirical examples In this section we present some examples of estimates of the stochastic frontier model and compare these estimates to other earlier results The first example uses 195758 data on the US primary metals industry SIC 33 consisting of observations across 28 states This data set was previously analyzed by Hildebrand and Liu 1965 and by Aigner and Chu 1968 The production function to be estimated is of the form In Y per lnZP In RlnKvu where Y is value added per establishment 1 is a measure of labor input per establishment K is the gross book value of plant and equipment per establish ment and R is the ratio of net to gross book value of plant and equipment Various sets of parameter estimates are given in table 3 The first three sets of entries correspond to our results running OLS and using the maximum Table 3 Estimates of eq 20 by various methods Method I 81 82 OLS Stochastic frontier exponential Stochastic frontier halfnormal HildebrandLiu OLS AignerChu OLS AignerChu LP AignerChu QPl AignerChu QP2 09146 09168 004164 204 731 219 09601 09144 004125 220 771 229 09600 09105 004208 206 768 234 0988 004208 0908 00333 0873 00031 1071 00269 0822 00219 likelihood technique of section 4 with the onesided disturbance assumed to be halfnormal and exponential respectively The numbers in parentheses under the OLS estimates are t ratios The numbers in parentheses under the stochastic frontier maximumlikelihood estimates are asymptotic t ratios That is they are the ratio of the coefficient estimate to the square root of the appropriate diagonal element of the inverse of the information matrix These are asymp totically distributed as N0 1 under the null hypothesis that the associated coefficient is zero D Aigner et al Stochastic frontier production function models 33 The remaining entries are taken from Aigner and Chu 1968 p 836 They consist of the OLS results of Hildebrand and Liu and of Aigner and Chu the linear programming results of Aigner and Chu and the results of two variants of the quadratic programming technique of Aigner and Chu We did not succeed in duplicating exactly the OLS results of either Hildebrand and Liu or Aigner and Chu although the discrepancies are not large What is most interesting is the extremely close agreement between our OLS estimates and both sets of stochastic frontier maximumlikelihood estimates In particular the stochastic frontier estimates are much closer to the OLS estimates than they are to Aigner and Chus programming results The reason for this is clear if we look at the estimates of the parameters of the distributions of the disturbances lo In the OLS case the estimated variance of the disturbances is 0077640 In the halfnormal case we have A2 0 0000686 asymptotic t ratio 005 A2 C 00692 asymptotic t ratio 364 and in the exponential case we have f 00180 asymptotic t ratio 016 A2 0 00691 asymptotic t ratio 368 In both cases the symmetric component of the disturbance effectively swamps the onesided component In the exponential case the mean of the onesided component is 0018 and its variance is 0000325 which is only 0468 percent of the total disturbance variance of 000032500691 00694 Similarly in the halfnormal case the mean of the onesided component is 0021 and its variance is 0000251 which is only 0361 percent of the total disturbance vari ance Therefore the picture that emerges is one of substantial variation in the frontier across states but relatively little variation of observed output beneath the frontier As a final note the maximized value of the logarithm of the likelihood func tion is 2372 in the exponential case and 2368 in the halfnormal case indicating a marginally better fit by the halfnormal distribution Our second example uses US agricultural data for six years 196065 and the 48 contiguous states The data set is the one used by Timmer 1971 exclud ing the years 1966 and 1967 The function to be estimated is Y ppl Labor Capitalp LandP Fertilizer Is LivestockP Seed v u 21 In the empirical examples used here the likelihood function was explicitly parameterized in terms of the parameters of the two error component distributions oU2 and uV2 rather than in the equivalent way discussed in section 4 34 D Aigner et al Stochastic frontier production function models where Y is gross agricultural output and the remaining variables are more or less selfexplanatory see Timmer 1971 for details All variables are in loga rithms and are on a perfarm basis Various sets of parameter estimates are given in table 4 The first two sets of estimates are our OLS and stochastic frontier results based on an exponential distribution for u The other four sets are taken from Timmer 1971 p 785 They represent respectively his OLS results his linear programming results and his linear programming results using 98 percent and 97 percent of the observations Table 4 Estimates of eq 21 by various methods Method OLS 18072 01149 02976 006061 01411 02581 01956 367 885 481 1240 1672 644 Stochastic frontier 18143 01148 02776 006061 01411 02581 01956 exponential 429 371 896 487 1256 1693 652 Timmer OLS 17350 01919 03726 00458 01484 02510 01579 538 67 117 42 160 195 54 Timmer LPI 0 0 16693 06015 04887 01334 02347 01043 Timmer LPg 8 18578 03287 03689 00298 01428 02045 02243 Timmer LPg7 18828 02679 04842 00099 01693 01885 01712 Our OLS results are probably as close as can be expected to Timmers OLS results given that we are missing 96 of his observations Our stochastic frontier estimates approximate the OLS estimates as in the previous example They are also reasonably close to Timmers OLS LP and LP estimates It would appear that the linear programming LPe estimates are the only ones very unlike the others Our estimates of the parameters of the distributions of the disturbances are f 000710 asymptotic t ratio 030 a 001005 asymptotic t ratio 112 Once again the symmetric component of the disturbance completely swamps the exponential component the variance of the exponential component 00000504 is only about one half of one percent of the total variance 7 Conclusions We have described a linear model with an error specification that is con sidered appropriate for the estimation of an industry production function using D Aigner et al Stochastic frontier production function models 35 crosssection data The specification does not prejudge the placement of the function as if an average function or a frontier function were to be fitted to the data Indeed the interesting feature of the framework discussed herein is precisely that placement of the function is estimated along with other model parameters This is also the contribution of the model presented in Aigner Amemiya and Poirier 1976 but that model is not capable of direct interpreta tion in terms of the sources of random error that may cause firms with identical input vectors to differ Whether from an applications viewpoint one model dominates the other is as yet not clear since the empirical evidence from both studies is limited What we find in the Monte Carlo results of the present paper is not particularly encouraging In the presence of an intercept the placement parameter A and the intercept itself are apparently difficult to estimate by the ML technique even when sample size is as large as 100 Further smallsample studies would help in forming a more definite opinion about the relative merits of the alter native specifications for tie placement and estimation of the industry production studied by Aigner Amemiya and Poirier and in the present paper Additional research is also required to evaluate the performance of the moment estimators based on leastsquares residuals mentioned at the end of section 4 Tests of our model on two realworld data sets indicated relatively small onesided components of the disturbance This in turn suggests high levels of efficiency relative to a stochastic frontier Whether this finding based on state wide perestablishment aggregates would continue to hold for the individual establishments themselves is yet another interesting question to be answered Appendix Below are reproduced the secondorder derivatives of In fZ with respect to fi 2 and o2 for use in computational algorithms and as a basis for estimating asymptotic standard errors aln2 1 iv 32 7 il 132 ifixi2 c fl I ff lfTyjBXi 1 A 1 aln2 apap f i xixl 642 36 D Aigner et al Stochastic frontier production function models 1 FryiPXi331FyiBXi b43 d21n9 1 N anap d il lOf c f i lF lnyififXi 1 FYijXi2 I Xi a2kz Tg ii1 yiXiXi L t L 3 il lj72 i lFT tfYifiXi 1 FTyiBXi2 645 a21n9 1 5 fi 12l 57 iCl 1 Fy I lFTYijXJ afyiflXi 1 FTyiBXi3 1 646 References Afriat SN 1972 Efficiency estimation of production functions International Economic Review 13 Oct 568598 Aigner DJ and SF Chu 1968 On estimating the industry production function American Economic Review 58826839 Aigner DJ T Amemiya and DJ Poirier 1976 On the estimation of production frontiers International Economic Review 17 377396 Amemiya Takeshi 1973 Regression analysis when the dependent variable is truncated normal Econometrica 41 9971016 Cohen AC Jr 1950 Estimating the mean and variance of normal populations from singly and doubly truncated samples Annals of Mathematical Statistics 21 557569 Cohen AC Jr 1957 On the solution of estimating equations for truncated and censored samples from normal populations Biometrika 44 225236 Dugger R 1974 An application of bounded nonparametric estimating functions to the analysis of bank cost and production functions unpublished PhD dissertation University of North Carolina Chapel Hill NC D Aigner et al Stochastic frontier production function models 37 Farrell MJ 1957 The measurement of productive efficiency Journal of the Royal Statistical Society A general 120 pt 3 253281 Goldfeld Stephen and Richard Quandt 1971 Nonlinear methods in econometrics North Holland Amsterdam Halperin M 1952 Maximum likelihood estimation in truncated samples Annals of Mathe matical Statistics 23 226238 Halperin M 1952 Estimation in the truncated normal distribution Journal of the American Statistical Association 47 457465 Hartley HO 1958 Maximum likelihood estimation from incomplete data Biometrics 14 174194 Hildebrand G and TC Liu 1965 Manufacturing production functions in the United States 1957 Cornell University Press Ithaca NY Marschak J and WJ Andrews 1944 Random simultaneous equations and the theory of production Econometrica 12 JulyOct 143205 Richmond J 1974 Estimating the efficiency of production International Economic Review 15 June 515521 Schmidt P 1976 On the statistical estimation of parametric frontier production functions Review of Economics and Statistics 58 238239 Timmer C Peter 1971 Using a probabilistic frontier production function to measure technical efficiency Journal of Political Economy 79 776794 Weinstein MA 1964 The sum of values from a normal and a truncated normal distribution Technometrics 6 104105 with some additional material 469470 Zellner A Kmenta J and J De 1966 Specification and estimation of CobbDouglas production functions Econometrica 34 Oct 784795 Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p2243 JanAbril 2023 22 REVISTA EVIDENCIAÇÃO CONTÁBIL FINANÇAS João Pessoa v 11 n 1 p 2243 JanAbril 2023 ISSN 23181001 DOI 1022478ufpb231810012023v11n158619 Disponível em httpperiodicosufpbbrojs2indexphprecfin INTERMEDIAÇÃO FINANCEIRA E EFICIÊNCIA UM ESTUDO DOS BANCOS E COOPERATIVAS DE CRÉDITO DO BRASIL1 FINANCIAL INTERMEDIATION AND EFFICIENCY A STUDY OF BRAZILS CREDIT BANKS AND COOPERATIVES Wesley Paulo Santos Mestre em Ciências Contábeis UFPE Faculdade Santa Helena wesleypaulolivecom Elenildo Santos Bezerra Doutorando em Ciências Contábeis UFPE Universidade Federal de Sergipe UFS elenildoconsultoriagmailcom Mércia de Lima Pereira Doutoranda em Ciências Contábeis UFPB Universidade Federal da Paraíba UFPB profamercialimagmailcom RESUMO Objetivo Analisar quais fatores influenciam a eficiência das instituições financeiras e verificar quais índices exigidos por Basileia possuem relação com a eficiência delas Fundamento A forma que os gestores de instituições financeiras utilizam para o reconhecimento de despesas impacta nos resultados do período e consequentemente na eficiência demonstrada por aquelas instituições financeiras aos demais agentes e órgãos reguladores Macedo Kelly 2016 Santos et al 2019 Método Foi realizada uma fronteira de eficiência estocástica bayesiana para determinar a eficiên cia dos custos e isolar o fator de ineficiência seguindo o que foi realizado por Tecles e Tabak 2010 e Barros e Wanke 2014 Em seguida foi comparada a forma com que os índices obrigatórios por Basileia afetaram o termo de ineficiência A amostra foi de seiscentas e setenta e uma instituições financeiras entre 2015 e 2018 Resultado Os resultados demonstraram que os fatores que afetaram a eficiência dos bancos foram as obrigações por empréstimos as despesas operacionais o preço pelo trabalho e nível do capital próprio Quanto às cooperativas de crédito os fatores que afetaram a eficiência foram os depósitos totais o market share e o nível de capital próprio 1 Artigo recebido em 23032021 Revisado por pares em 09052023 Reformulado em 01102021 Recomendado para publicação 18052023 por Anna Paola Fernandes Freire Editora Adjunta Publicado em 16082023 Organização res ponsável pelo periódico UFPB Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 23 Contribuições Este estudo contribui para a literatura por evidenciar que em linhas gerais a boa gestão do índice de Alavancagem proposto por Basiléia aumenta a eficiência dos custos para as cooperativas porém o mesmo não ocorre para os bancos sendo assim tal ineficiência dáse devi do a outros fatores Palavraschave Fronteira de eficiência estocástica bayesiana Eficiência Bancos Cooperativas de crédito ABSTRACT Objective To analyze which factors influence the efficiency of financial institutions and to verify which mandatory indexes by Basel interfere in efficiency Background The way that managers of financial institutions use to recognize expenses influences the result for the period and consequently the efficiency shown by that financial institution to other agents and regulatory bodies Method A Bayesian stochastic efficiency frontier was used to determine efficiency and isolate the inefficiency factor following what was done by Tecles and Tabak 2010 and Barros and Wanke 2014 Then the way in which the Basel mandatory ratios affected the inefficiency term was com pared The sample consisted of 671 financial institutions between 2015 and 2018 Result The results showed that the factors that affected banks efficiency were loan obligations operating expenses price for labor and level of equity For credit unions the factors that affected efficiency were total deposits market share and equity level Contributions This study contributes to the literature by showing that the higher the Basel index the lower the banking inefficiency and the greater the degree of financial leverage the greater the level of inefficiency Keywords Bayesian stochastic efficiency frontier Efficiency Banks Credit unions 1 INTRODUÇÃO O setor bancário assim como qualquer outro setor tem como objetivo gerir da melhor for ma possível seus recursos em um processo eficiente de utilização de insumos para a geração de riqueza Tecles Tabak 2010 Nesse sentido eficiência bancária deve ser entendida como as me lhores práticas na gestão dos custos empresariais e a ineficiência por outro lado é assumida como um fator que aumenta o custo das organizações e reduz seu lucro No entanto diferente de outras entidades a eficiência bancária não pode ser medida ape nas em relação aos custos e lucros gerados uma vez que os bancos estão expostos a variações ma croeconômicas o que pode fazer com que o lucro obtido durante um período não se confirme no período seguinte mas sim que haja prejuízo Santos Rodrigues Anjos 2019 Assim é importante que os bancos gerenciem a qualidade do capital responsável por gerar seus lucros Esse fator im plica uma dicotomia pois por um lado os bancos precisam captar recursos e emprestálos a outros agentes entretanto por outro lado eles precisam analisar de forma cuidadosa a quem concedem esses valores porque afinal podem não ser mais reavidos Nesse contexto a partir da segunda metade da década de 1970 os estudos sobre eficiência passaram a construir uma fronteira de eficiência capaz de determinar quais instituições estariam utilizando melhor seus recursos criando assim um ranking Os estudos de Aigner Lovell e Sch midt 1977 e Meeusen e Broeck 1977 foram os primeiros a associar a ineficiência a uma distribui ção de probabilidade estabelecendo uma Fronteira de Eficiência Estocástica FEE Se aplicado ao setor bancário a FEE poderá identificar quais bancos são os mais eficientes dentro do Sistema Fi Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 24 nanceiro Nacional SFN bem como quais fatores interferem de modo a aumentar o seu grau de ineficiência A técnica utilizada para criar a FEE foi de abordagem bayesiana Esta se diferencia da a bordagem clássica por selecionar a distribuição de probabilidade para um evento específico de forma prévia sem contudo saber o número total de casos desse evento Tecles Tabak 2010 Meeusen e Broeck 1977 expõem que a abordagem bayesiana é uma importante alternativa diante da falta de informações precisas sobre determinado evento ou população A distribuição de proba bilidade considerada neste estudo foi a proposta por Griffin e Stell 2007 e validada para as insti tuições financeiras brasileiras por Tecles e Tabak 2010 e Barros e Wanke 2014 Verificar a eficiência das instituições financeiras de um país ou região é importante pois elas são os agentes responsáveis por fomentar a economia na forma de concessão de empréstimos e créditos para que agentes deficitários possam operar A relevância que as entidades que fazem parte do SFN têm é reconhecida mundialmente tanto que elas são supervisionadas por bancos centrais e obrigadas a atender legislação própria emitida pelos acordos de Basileia Apesar da supervisão dos bancos centrais e da regulação proposta por Basileia o sistema financeiro pode passar por crises Quase 20 anos após o primeiro acordo de Basileia de 1988 que definiu o capital mínimo que as entidades financeiras deveriam ter para suportar a sua exposição ao risco e cerca de 4 anos depois do segundo acordo de Basileia de 2004 que teve como foco os riscos de crédito de mercado e operacional das instituições financeiras o mundo testemunhou a crise financeira ocorrida no ano de 2008 A crise financeira de 2008 entre outros fatores deuse pela falência de um banco de inves timentos o Lehman Brothers o que abalou a economia de vários países provocando um efeito devastador Após a crise financeira houve um novo acordo de Basileia chamado de Basileia III que entrou em vigor no Brasil no ano de 2013 Esse último acordo introduziu a obrigatoriedade do índice de alavancagem liquidez e reformulou a estrutura de capital das organizações Com as alte rações trazidas por Basileia III três indicadores ganharam especial atenção do sistema financeiro o índice de risco de Basileia o índice de Imobilização além do índice de Alavancagem Apesar de toda regulamentação e supervisão o setor financeiro carece de constante apri moramento uma vez que a falência de uma entidade financeira pode desestabilizar diversos agen tes econômicos Cardoso Campos Dantas Medeiros 2019 Para Ferreira Zanini e Alves 2019 a segunda metade do século XX foi de diversificação das receitas bancárias o que levou alguns paí ses a flexibilizarem suas leis para atender esse novo padrão Porém tal medida favoreceu uma maior adoção de um mix de receitas bancárias que privilegiasse o curto prazo Bonfim Kim 2012 Nesse sentido os acordos de Basileia têm entre seus objetivos reforçar o quadro global de capitais e preservar a liquidez do sistema por meio de requisitos mínimos obrigatórios de capital para suas entidades Cardoso et al 2019 No contexto brasileiro estudos como os de Pinheiro Savóia e Securato 2015 Goes Sheng e Schiozer 2016 e Cardoso et al 2019 analisaram o impacto de Basileia III sobre as entidades que compõe o SFN De forma anterior à Basileia III os estudos de Tecles e Tabak 2010 e Barros e Wanke 2014 criaram uma fronteira de eficiência para o setor bancário brasileiro No entanto ne nhum desses estudos estabeleceu a relação entre eficiência bancária e os índices obrigatórios por Basileia Assim o objetivo deste estudo é analisar a forma com que os índices obrigatórios por Basi leia afetam a eficiência das instituições financeiras brasileiras no período entre 2015 e 2018 De forma específica este estudo demonstra a quantidade de bancos e cooperativas que são eficientes e ineficientes além de verificar quais fatores influenciam a eficiência dessas instituições como por exemplo despesas operacionais market share entre outros Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 25 Para alcançar o objetivo foi realizada uma fronteira de eficiência estocástica para os bancos e outra para as cooperativas de crédito com a finalidade identificar e isolar o fator de ineficiência de cada instituição financeira da amostra Em seguida a ineficiência dessas instituições financeiras foram analisadas levando em consideração os índices obrigatórios pelos acordos de Basileia Estu dar os bancos e as cooperativas de crédito sob a ótica da eficiência financeira se justifica pois em bora ambas possuam finalidades distintas uma vez que os bancos possuem o propósito de gerar lucro aos seus acionistas enquanto as cooperativas de crédito possuem como propósito atender aos seus cooperados ambas as entidades possuem como principal atividade a intermediação fi nanceira através da facilitação do acesso ao crédito Os resultados deste estudo são úteis de três formas apontam quais fatores operacionais ou de mercado influenciam nos custos das entidades demonstram se a maior parte das entidades que compõe o SFN atua de forma eficiente e por fim evidenciam se os índices de estrutura de capital alavancagem e imobilização propostos por Basileia reduzem a ineficiência bancária o que tende a aumentar a estabilidade financeira e atende à finalidade com a qual eles foram estabelecidos Des sa forma depreendese que este estudo é importante para reguladores normatizadores clientes e demais agentes do mercado financeiro por evidenciar que a maioria dos bancos e cooperativas se preocupam em atender apenas ao mínimo exigido por Basileia Um dos possíveis motivos para isso está na exposição do capital dessas instituições a maiores níveis de risco 2 FUNDAMENTO 21 Sistema financeiro brasileiro No Brasil o Sistema Financeiro Nacional SFN é composto por um conjunto de entidades e instituições que promovem a intermediação financeira entre credores e tomadores de recursos O SFN possui três órgãos normativos o Conselho Nacional de Seguros Privados o Conselho Nacio nal de Previdência Complementar e o Conselho Monetário Nacional CMN sendo este último o responsável por formular políticas de moeda e crédito e todos esses órgãos são subordinados ao Governo Federal do Brasil BCB 2019 O CMN além de estabelecer políticas macroeconômicas também controla os seguintes ór gãos supervisores a Comissão de Valores Mobiliários CVM órgão responsável por supervisionar a bolsa de valores mercadorias e futuro e o Banco Central do Brasil BCB órgão responsável por supervisionar bancos caixas econômicas e cooperativas de crédito BCB 2019a De acordo com o BCB 2019b a quantidade de conglomerados financeiros e instituições independentes em dezem bro de 2018 era de 1356 instituições Desse total 134 instituições eram bancos 962 eram cooperati vas de crédito e 260 instituições eram constituídas na forma de outras entidades como instituições não bancárias de crédito não bancárias de mercado de capitais e instituições de pagamento Em relação às características do SFN em dezembro de 2018 existiam 962 cooperativas de crédito que possuíam um ativo total superior a R 271 trilhões enquanto apenas 6 instituições sendo todas bancárias em todo SFN eram classificadas como S1 ou seja que possuíam participa ção superior a 10 do mercado Os bancos que fizeram parte desse grupo foram Itaú Banco do Brasil Caixa Econômica Federal Bradesco Santander e BTG Pactual sendo a soma dos ativos des ses 6 bancos estimada no valor de R 62 trilhões enquanto o total de ativos de todas as 134 institu ições bancárias era de R 9 trilhões BCB2019b Outra característica do SFN é a existência de insti tuições financeiras públicas que operam no mercado comum Em 2018 existiam 28 instituições fi nanceiras nessa situação sendo o Banco do Brasil e a Caixa Econômica Federal bancos públicos que figuravam dentro do grupo das 6 maiores instituições financeiras nacionais BCB 2019b Em relação aos bancos públicos e privados no Brasil o estudo de Sanches Silva Júnior e Srisuma 2018 teve como objetivo estudar a privatização dos bancos públicos no país Os resulta dos apontam que os bancos privados possuem o objetivo de maximizar seus lucros enquanto os Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 26 bancos públicos tendem a fornecer serviços a segmentos de mercado que não são lucrativos ou carecem de financiamento diferenciado dada a sua importância Os autores citam como exemplo o Banco do Brasil que financia o setor agrícola brasileiro e a Caixa Econômica Federal que é a res ponsável pela distribuição dos recursos de fundos governamentais para habitação saneamento assistência social entre outros Assim Sanches et al 2018 discorrem que o desenho institucional do mercado brasileiro faz com que bancos públicos operem oferecendo crédito para infraestrutura hipotecas e investimentos enquanto os bancos privados ofereçam em sua maioria crédito pessoal e de curto prazo para empresas Sobre a fonte de receita dos bancos e cooperativas o estudo de Ferreira Zanini e Alves 2019 demonstrou que apesar do crescente movimento por diversificação de receita dessas institu ições a receita de intermediação financeira interest é a principal fonte de retorno dos bancos e cooperativas do Brasil A receita noniterest que são comissões e receitas em geral exceto de inter mediação financeira foi responsável em média por 274 do faturamento dos bancos brasileiros em 2014 o que demonstra que 726 das receitas obtidas foram de intermediação financeira con figurando ainda sua principal fonte de recurso Ao analisar especificamente as receitas de interme diação financeira Ferreira et al 2019 sugerem que os bancos atuantes no Brasil deem preferência às atividades de crédito em vez de atividades com títulos e valores mobiliários pois seriam as mais rentáveis Ao que se refere as cooperativas de crédito elas possuem também uma função social uma vez que o interesse da instituição é fornecer crédito com taxas mais atrativas para os seus coopera dos do que um banco comum Maia Bressan Lamounier e Braga 2013 comentam que os mem bros da cooperativa são os responsáveis por prover o capital necessário para a entidade quanto a demanda a ser consumida Assim no aspecto conceitual as cooperativas de crédito não devem se valer de altas taxas de juros com intermediação financeira pois sua finalidade principal é atender aos seus cooperados No entanto Ventura Fontes Filho e Soares 2009 demonstram que ainda assim as cooperativas de crédito podem ser submetidas a pressões por performance financeira uma vez que as sobras líquidas podem ser distribuídas entre seus cooperados Sobre a estrutura de capital das cooperativas de crédito o estudo de Carvalho Diaz Bia loskorski Neto e Kalatzis 2015 enfatiza que a gestão dessas instituições pode tornarse difícil uma vez que elas podem ser geridas por pessoas ligadas diretamente aos cooperados porém sem a de vida capacidade técnica A falta de capacidade técnica por parte da gestão pode por sua vez pre judicar a estrutura de capital das cooperativas Sob outra perspectiva os autores expõem que as cooperativas de crédito podem ter ganhos de escala por meio de fusões de pequenas cooperativas e o auxílio mútuo através de cooperativas centrais o que elevaria o seu retorno financeiro Quanto às diferenças apresentadas entre os bancos e as cooperativas dois estudos busca ram se aprofundar no tema no contexto brasileiro Matias Quaglio Lima e Magnani 2014 anali sou qual tipo de entidade foi mais eficiente em relação a gestão de suas receitas se comparado com as despesas administrativas e de pessoal para os anos entre 2002 e 2012 A conclusão chegada por Matias et al 2014 é a de que as cooperativas de crédito eram menos eficientes que os bancos pú blicos e privados na gestão de suas despesas Outro estudo que abordou de forma conjunta os ban cos e as cooperativas brasileiras foi o de Bittencourt V G Bressan Goulart A A Bressan Costa e Lamounier 2016 em que foi analisada a rentabilidade dos bancos e das cooperativas em relação ao retorno sobre o patrimônio líquido e o retorno sobre o ativo entre os anos de 2009 a 2013 Já Bittencourt et al 2016 concluem que não há diferença significativa entre o retorno sobre o patri mônio líquido dos bancos e das cooperativas no entanto em relação ao retorno sobre o ativo os bancos apresentam retornos maiores que as cooperativas sendo essa diferença estatisticamente significativa Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 27 Além do estudo de Cardoso et al 2019 outros estudos levaram em consideração o impac to causado pelas variáveis macroeconômicas para o setor bancário brasileiro O estudo de Araújo Lustosa e Paulo 2018 que teve como objetivo analisar se as provisões para crédito de liquidação duvidosa em bancos eram prócíclicas anticíclicas ou acíclicas concluiu que elas eram prócíclicas e que o efeito das variações macroeconômicas repercutia no mesmo período contábil Brito Lopes e Coelho 2012 e Dantas Paulo Medeiros 2013 analisaram o conservadorismo nos lucros dos bancos brasileiros Ambos os estudos demonstraram que os bancos tendem a ser conservadores condicionais o que implica reconhecer de forma antecipada as perdas se comparado com os ga nhos e que bancos estatais são mais propensos a tal procedimento Corroborando com o exposto Santos Rodrigues Anjos e Tavares 2019 analisaram o cená rio econômico interno e externo na prática de income smoothing entre os anos de 2011 a 2017 e i dentificaram que os bancos brasileiros são mais conservadores em seus lucros no final de cada se mestre ao aumentarem suas despesas com crédito de liquidação duvidosa Sobre o gerenciamento de resultado ao final de cada semestre Sousa 2018 demonstrou que essa prática é comum nos bancos brasileiros principalmente nos bancos de pequeno porte Santos et al 2019 também iden tificaram que o cenário econômico externo influencia de forma significativa a prática de suavização de resultado Os resultados de Santos et al 2019 corroboram o que foi exposto anteriormente por Macedo e Kelly 2016 ao sugerir que bancos de capital estrangeiro praticam mais gerenciamento de resultado do que os bancos de capital nacional 22 Fronteira de eficiência bayesiana Fronteiras de eficiência são ferramentas muito utilizadas em economia e finanças para de terminar quantidades ótimas como exemplo há a fronteira de eficiência de uma carteira de ações há a fronteira de eficiência do consumidor ou do produtor entre outros Apesar de amplamente divulgada as fronteiras de eficiência em sua forma clássica necessitam do prévio conhecimento do total de casos de um determinado evento sendo algo fechado em si mesmo Ntzoufras 2009 Apenas a partir dos estudos de Aigner et al 1977 e Meeusen e Broeck 1977 é que as fronteiras de eficiência passaram a ser consideradas de forma estocástica ou seja associada a uma distribui ção de probabilidade a partir de um evento aleatório gerado por simulação Aigner et al 1977 formularam e estimaram uma fronteira de produção associando o ter mo de erro do modelo a duas distribuições de probabilidade Em seguida uma série de simulações foram estimadas via Simulação de Monte Carlo sendo a principal contribuição do estudo a possi bilidade de estimar o modelo junto com sua respectiva função de eficiência Aigner Lovell Sch midt 1977 De forma similar Meeusen e Broeck 1977 também estimaram uma função de produ ção ao escolherem uma distribuição exponencial para o termo de erro Por fim concluíram que a fragilidade desse método estava na escolha prévia da distribuição de probabilidade sendo este o principal fator Meeusen Broeck 1977 Zhag 2000 ao comparar o modelo bayesiano com o modelo de máxima verossimilhança identificou que o primeiro foi superior ao estimar a fronteira de eficiência Assaf Matousek e Tsionas 2013 ao compararem o modelo Bayesiano com o método dos momentos generalizados GMM identificaram que o primeiro é mais estável possuindo me nos variações do que o segundo A escolha da distribuição de probabilidade pode ser realizada com base na análise da quan tidade obtida dos eventos observados ou pode se fazer o inverso primeiro se escolhe a distribui ção de probabilidade para em seguida se observar os eventos Essa segunda abordagem é deno minada de bayesiana e deve ser utilizada quando o número de casos para o evento é desconheci do porém já existem algumas evidências sobre eles Kinas Andrade 2010 De acordo com Te cles e Tabak 2010 em um modelo bayesiano são atribuídos aos parâmetros distribuições prévias Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 28 que contenham as informações já conhecidas sobre o evento sem contudo saber de forma precisa o número de casos para cada evento Nesse sentido o modelo bayesiano pode ser uma boa escolha se já existem evidências sobre o objeto de estudo ao mesmo tempo que não é o mais indicado caso o parâmetro não possua ne nhuma base teórica anterior De acordo com Kumbhakar e Tsionas 2005 a inferência bayesiana é realizada utilizando a técnica de Monte Carlo por Cadeia de Markov MCMC para os sorteios aleatórios A filosofia desse método é que a função de densidade posterior é dada por onde é o vetor do parâmetro sendo o objetivo simular sorteios aleatórios da distribuição posterior por Ntzoufras 2009 argumenta que o sistema bayesiano necessita de alguns parâmetros O primeiro deles é a seleção das variáveis dependentes e independentes Em seguida a escolha da distribuição de probabilidade utilizada que deve conter as características conhecidas dos números de casos já observados de um determinado evento ou ser validada de acordo com a literatura pre viamente estabelecida de estudos anteriores Ntzoufras 2009 Esse segundo passo visa contornar a fragilidade na determinação da distribuição de probabilidade apontada por Meeusen e Broeck 1977 Por último o passo apontado por Ntzoufras 2009 é a análise dos parâmetros de saída que são a fronteira de eficiência e a significância das variáveis que compõe o modelo Entre os estudos que utilizaram a abordagem bayesiana para analisar o setor bancário des tacamse para o contexto pretendido os trabalhos de Kumbhakar e Tsionas 2005 Behr 2010 e Assaf et al 2013 no âmbito internacional e o de Tecles e Tabak 2010 e de Barros e Wanke 2014 para o contexto brasileiro 23 Acordos de Basileia e eficiência bancária O Acordo de Capital de 1988 também denominado de International convergence of capital measurement and capital standards ou Basileia I teve como objetivo a convergência do setor financei ro mundial a partir da segunda metade do século XX Pinheiro Savóia Securato 2015 Ele estabe leceu de forma global o conceito de capital mínimo regulatório que na ocasião era de 4 para o nível 1 e de 8 no total Goes Sheng Schiozer 2016 O Brasil adotou as regras estabelecidas por Basileia apenas em 1994 período posterior ao da hiperinflação brasileira e quando adotou optou por níveis de capitais mais conservadores sendo de 8 para o capital mínimo e 11 para o ativo ponderado pelo risco Goes Sheng Schiozer 2016 Em 2005 foi estabelecido o segundo acordo de Basileia Ele promoveu mudanças qualitati vas uma vez que alterou o conceito de ativo ponderado pelo risco Pinheiro Savóia Securato 2015 Uma outra mudança considerável foi a dedução do ativo intangível na composição do capi tal mínimo regulatório Pinheiro Savóia Securato 2015 O terceiro acordo de Basileia realizado em 2011 foi uma revisão do Acordo de Capital De acordo com Pinheiro Savóia e Securato 2015 as alterações trazidas por Basileia III fizeram com que os bancos fossem obrigados a levar em consideração três indicadores financeiros sendo eles 1 O índice de imobilização que informa o quanto do capital total da instituição financeira está alocado em ativos permanentes 2 O índice de risco de Basileia que mede o risco de insolvência de uma instituição financeira assim quanto menor for esse índice mais propenso o banco estará de falir 3 O índice de alavancagem que é a razão entre o capital seguro nível 1 e a exposição total da instituição Para Goes Sheng e Schiozer 2016 o terceiro acordo de Basileia teve como objetivo aumentar a capacidade dos bancos de suportar possíveis perdas transferindo o possível ônus de decisões equivocadas por parte das instituições para investidores e proprietários enquanto livra o governo e a sociedade Diante da importância que o setor financeiro tem para a economia nacional além do risco operacional ao qual essas entidades estão expostas alguns estudos no Brasil trataram de verificar Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 29 como se deu a adoção de Basileia II nas instituições financeiras do Brasil O estudo de Santos Ma cedo e Rodrigues 2014 analisou os fatores que determinaram o nível de divulgação dos 100 maio res bancos do país no ano de 2010 Os autores chegaram à conclusão de que apenas um maior por te do banco e um melhor índice de Basileia foram capazes de explicar uma maior divulgação de informações financeiras por parte dos bancos De forma anterior Beneditto e Silva 2008 expuse ram que os únicos fatores que poderiam aumentar o nível de divulgação dos bancos nacionais além da regulação seriam através das auditorias independentes e por meio da pressão exercida por seus stakeholders Ainda sobre os estudos realizados que tomaram como base os acordos de Basileia o estudo de Pinheiro et al 2015 ao analisar o impacto trazido pelo terceiro acordo de Basileia no setor fi nanceiro brasileiro concluiu que os bancos devem buscar fontes de receita que ofereçam maior rentabilidade alinhados a uma maior eficiência no que diz respeito às suas despesas operacionais Para eles o acordo de Basileia III ao deixar mais rígido a ponderação pelo risco impactou direta mente no nível de resultado operacional dos bancos o que deve fazer com que esses agentes reve jam suas estratégias operacionais Uma vez que essa estrutura trazida por Basileia III diminuiu o valor líquido dos ativos das instituições financeiras o estudo de Cardoso et al 2019 analisou os fatores macroeconômicos re lacionados à liquidez estrutural dessas instituições Os autores identificaram que quanto maior a taxa de juros o risco Brasil e a saída de recursos na balança comercial menor será a liquidez do setor bancário brasileiro Ainda sobre a estrutura de capital dos bancos o estudo de F Malaquias A Malaquias e Hwang 2018 apontou que é crescente a parcela da população brasileira que dei xou de procurar as agências pois resolve suas demandas por meio de aplicativo fazendo com que instituições financeiras reduzissem o número de agências ou optassem por agências menores Outros estudos utilizaram a abordagem bayesiana para analisar o setor bancário de diferen tes países Kumbhakar e Tsionas 2005 analisaram 500 bancos comerciais dos Estados Unidos para o período de 1996 e 2000 e identificaram que os parâmetros técnicos e alocativos influenciavam na eficiência bancária Behr 2010 analisou a eficiência dos bancos caixas econômicas e cooperativas de crédito alemãs o resultado encontrado foi que as instituições mais eficientes foram aquelas cu jos custos diferem consideravelmente de sua média Assaf et al 2013 estimou a eficiência dos bancos turcos para os períodos entre 2002 e 2010 identificando que a eficiência média dos bancos turcos foi negativa e que isso pode ter sido agravado pela crise financeira de 2008 No contexto brasileiro o estudo de Tecles e Tabak 2010 analisou 156 bancos no período de 2000 a 2007 Os fatores considerados foram participação no mercado empréstimo e capital próprio e seus resultados apontam que não houve uma grande variação da eficiência dos custos no merca do brasileiro para o período O estudo de Barros e Wanke 2014 foi outro a analisar os bancos bra sileiros considerando os 40 maiores bancos brasileiros entre o período de 1998 e 2010 Obtiveram como resultado que quanto menor o porte dos bancos menos eficientes eles foram Também veri ficaram que bancos estrangeiros e públicos possuíram maior ineficiência Além dos fatores de entrada e saída de recursos que são sempre utilizados para determinar a eficiência das instituições financeiras outras características dessas instituições foram levadas em consideração por outros estudos para determinar a fronteira de eficiência estocástica Behr 2010 levou em consideração a rentabilidade do patrimônio líquido e o percentual de ativo fixo Tecles e Tabak 2010 consideraram como determinantes para a eficiência se os bancos eram públicos pri vados ou estrangeiros Barros e Wanke 2014 por sua vez levaram em consideração o porte dos bancos para determinar a eficiência se os bancos haviam recebido ajuda do governo durante o ano de 2008 e se os bancos haviam passado por processo de fusão recente Porém nenhum desses es tudos analisou o nível de eficiência obtida de forma bayesiana com o índice de eficiência trazido e homologado pelos acordos de Basileia Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 30 Para as cooperativas de crédito os estudos de Garden e Ralston 1999 e McKillop Glass e Ferguson 2002 utilizaram a análise envoltória de dados ou DEA para traçar a fronteira de eficiên cia Garden e Ralston 1999 ao traçar uma fronteira de eficiência para as cooperativas de crédito australianas entre os anos de 1992 e 1997 teve como objetivo verificar se na ocasião de fusões es sas novas cooperativas teriam maior eficiência em relação às entidades que a antecederam ou em relação ao mercado Os resultados encontrados por Garden e Ralston 1999 foram que as coopera tivas que foram objetivos de fusão não apresentaram maior grau de eficiência em nenhuma das duas situações De forma semelhante McKillop Glass e Ferguson 2002 analisaram as cooperati vas de crédito do Reino Unido eles se concentraram especificamente no ano de 1996 ano de esta bilidade econômica local onde as entidades financeiras conseguiram reverter uma recessão que o setor atravessava Os resultados obtidos por McKillop Glass e Ferguson 2002 demonstram que apesar do contexto a maior parte das cooperativas de crédito eram ineficientes atuando próximo ao limite aceitável pela legislação Em um contexto bayesiano o trabalho de Feng e Wang 2021 analisou os bancos comunitá rios dos Estados Unidos entre os anos de 2001 e 2017 Os bancos comunitários são semelhantes às cooperativas uma vez que ambos visam atender aos seus membros no entanto o que os difere é o fato de as cooperativas poderem oferecer outros serviços de intermediação financeira além de a tender aos seus segurados ou exclusivamente a uma comunidade local Feng e Wang 2021 chega ram à conclusão de que os bancos comunitários apresentaram um baixo desempenho técnico cus tos financeiros mais altos e menor rentabilidade se comparado com os bancos comerciais Em rela ção ao Brasil o trabalho de V G Bressan Braga e A A Bressan 2010 utilizou uma abordagem estocástica para avaliar a eficiência das cooperativas de crédito de Minas Gerais para o período de 2001 a 2003 os resultados obtidos foram que apenas 15 das cooperativas eram eficientes em rela ção aos seus custos 3 MÉTODO 31 População e Amostra Inicialmente a população foi constituída de 671 instituições financeiras que compreendiam o total das instituições financeiras listadas pelo Banco Central do Brasil BCB em dezembro de 2018 Os dados foram obtidos no endereço eletrônico do BCB do relatório de Dados Selecionados de Entidades Supervisionadas IFdata BCB 2019b Desse modo os dados considerados são de natureza secundária uma vez que a primazia deles era de fornecer informações ao BCB sobre a situação econômica e financeira das instituições financeiras que compõem o Sistema Financeiro Nacional Da população foram excluídas instituições que no período 1 não possuíam os 16 trimes tres entre março de 2015 e dezembro de 2018 2 não possuíam carteira de crédito ativa 3 não tinham operações realizadas 4 o seu ativo era igual a zero 5 eram considerados como bancos de desenvolvimento ou não bancárias de crédito Após as exclusões a amostra considerada foi de 52 bancos comerciais banco múltiplo com carteira comercial ou caixa econômica categoria b1 18 bancos múltiplos sem carteira comercial ou banco de investimento ou banco de câmbio categoria b2 543 cooperativas de crédito singulares categoria b3s 16 instituições centrais e confederações de cooperativas de crédito carteira b3c As sim a amostra final compreende 70 instituições bancárias e 559 cooperativas O período escolhido se justifica pela alteração em relação a forma de divulgação do BCB Antes de 2015 o BCB apresentava os dados financeiros apenas das 50 maiores instituições financei ras do país o que inviabilizava uma análise macro do setor O ano de 2018 foi escolhido como úl timo período pois em 2019 os bancos classificados como S1 foram obrigados a adotar de forma definitiva a Resolução 4616 de 30112017 do Conselho Monetário Nacional que reestruturou o Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 31 indicador de liquidez de longo prazo dessas instituições Assim o período escolhido contém a maior parte das instituições financeiras do país e a trata de forma homogênea em relação a sua obrigatoriedade legal 32 Modelagem A modelagem desenvolvida se baseia na fronteira estocástica bayesiana apresentada por Batesse e Coelli 1992 Griffin e Stell 2007 e adaptada para as entidades financeiras brasileiras por Tecles e Tabak 2010 Assim a eficiência de forma genérica é dada pela regressão em um painel da variável y que é o logaritmo natural do custo de intermediação financeira de uma firma i obser vada no tempo t com um conjunto de fatores de saída e de produção Na equação anterior é a variável dependente são fatores de produção e saída de recursos é um fator de ineficiência e pode assumir qualquer valor entre zero e um sendo que quanto mais próximo de um maior será a ineficiência da entidade i o que pode aumentar o valor da variável dependente e é o erro padrão da regressão Uma vez que serão consideradas distri buições de probabilidade em uma ótica bayesiana algumas modificações devem ser realizadas Os dois primeiros passos de acordo com Ntzoufras 2009 é a escolha das variáveis depen dentes e independentes além da seleção das distribuições de probabilidade Assim a junção de que é a variável dependente com fatores de saída entrada e outras características que queiram ser observadas formam uma função de eficiência que por sua vez é condicionada a uma distribuição de probabilidade conforme a expressão que segue Sendo µ corresponde a Ao realizar o segundo passo estabelecido por Ntzoufras 2009 buscouse validar se a dis tribuição de probabilidade normal utilizada por Tecles e Tabak 2010 e por Barros e Kanke 2014 era a mais indicada para a série de dados Por meio do software estatístico Oracle Crystal Ball foi verificada qual distribuição de probabilidade era capaz de melhor retratar a variável dependente para os bancos e cooperativas Para os bancos o teste de KolmogorovSmirnov foi aplicado à vari ável dependente que apresentou coeficiente de 0421 e pvalue de 000 No caso das cooperativas o teste de KolmogorovSmirnov apresentou coeficiente de 04944 e pvalue de 000 Desse modo a distribuição de probabilidade normal foi validada estatisticamente além de já ter sido adotada em estudos anteriores por Tecles e Tabak 2010 e Barros e Kanke 2014 As variáveis selecionadas de a tiveram como base os estudo de Koop Osiewalski e Steel 1997 Tecles e Tabak 2010 e Barros e Wanke 2014 A primeira variável compreende aos depósitos totais e é dada pela divisão entre o valor total dos depósitos pelo somatório das o brigações com empréstimos e repasses mais depósitos totais A proporção de obrigações por em préstimos é a fração entre as obrigações por empréstimos e repasses pelo total de ativos A taxa dos fundos corresponde à divisão das despesas com intermediação financeira pelo total de obrigações por empréstimos e repasses A proporção de despesas operacionais é a divisão Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 32 entre o total das despesas operacionais despesas de pessoal despesas administrativas despesas tributárias e outras despesas operacionais pelo total de ativos O preço do trabalho é a despe sa de pessoal dividida pelo ativo total O Market Share se dá pela divisão entre a quantidade de clientes ativos para cada banco pelo total de clientes ativos no sistema financeiro nacional O nível de capital próprio é a fração entre o patrimônio líquido de cada instituição pelo seu total de ativos Por fim a variável nível de atraso é a provisão sobre operações de crédito para clientes dividido pelo valor total das operações de crédito para clientes Os coeficientes de α e β também são modelados por uma distribuição de probabilidade eles seguem uma distribuição normal com média 0 e variância 1 conforme utilizado anteriormente por Griffin e Stell 2007 A ineficiência considerada na variável modela a diferença entre a melhor prática e a prá tica real e presume que tal medida tenha um comportamento exponencial e que a eficiência da firma varia ao longo do tempo Meeusen Broeck 1977 Griffin Stell 2007 Sendo assim a ine ficiência pode ser descrito da forma a seguir Onde nas instituições financeiras brasileiras é igual a 085 e é o logaritmo negativo que compõe uma distribuição exponencial que compreende esse mesmo argumento foi utilizado por Tecles e Tabak 2010 por sua vez segue também uma distribuição exponencial que modela que é a ineficiência parcial Já é encontrado conforme segue A variável corresponde a uma distribuição normal de média 0 e variância 04 conforme o utilizado por Battese e Coelli 1992 e Griffin e Stell 2007 e poderá indicar um incremento de efi ciência ao longo do tempo o que poderá resultar em um maior O representa a eficiência ao longo do tempo e se positivo poderá reduzir a ineficiência da instituição financeira dada por em que o resultado disso será a ineficiência total da instituição i no tempo t que é expressa por Uma vez estabelecida a ineficiência para determinar o seu oposto a eficiência deverá seguir a ex pressão A eficiência nesses termos segue o utilizado por Griffin e Stell 2007 e Tecles e Tabak 2010 e é a exponencial do inverso da ineficiência O numerador é a função de uma empresa livre de ineficiência por outro lado a base ou denominador é a função de uma empresa levando em consideração a ineficiência da mesma A métrica adotada para a eficiência resultará na criação de coeficientes que possibilitará a aferição e criação de uma fronteira de eficiência estocástica A fronteira de eficiência estocástica permitirá avaliar a quantidade de instituições que eficientes dadas as variáveis e premissas esco lhidas Esse processo de aferição e análise dos coeficientes e da fronteira de eficiência é o terceiro passo estabelecido por Ntzoufras 2009 Sobre o ranking é importante destacar que tal medida não tem como objetivo denegrir ou exaltar qualquer instituição financeira em especifico mas o de evidenciar o comportamento das instituições financeiras ao longo do tempo com base na metodologia aplicada motivo pelo qual os nomes das instituições financeiras serão preservados Por fim serão estimados dois painéis sepa radamente um para as instituições bancárias e outro para as cooperativas de crédito o que resul Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 33 tará em uma fronteira de eficiência para cada grupo em que medida permitirá a comparação entre esses dois macrossegmentos do SFN 33 Simulações Uma vez estabelecido o modelo foi cotado para cada painel 10000 interações os 1000 pri meiros números gerados foram descartados e a cada 25 valores gerados uma amostra era conside rada sendo tal medida necessária para diminuir possíveis correlações existentes entre as variáveis O software utilizado na modelagem foi o WinBUGS14 software grátis de abordagem bayesiana capaz de gerar números aleatórios por meio da simulação de Monte Carlo por Cadeia de Markov 34 Regressão Após determinar um índice de ineficiência de cada instituição financeira foi realizada uma regressão em um modelo logístico ordinal com o objetivo de elencar quais fatores influenciam na perda de produtividade das instituições financeiras O modelo logístico foi o escolhido por ser um dos mais indicados para tratar de variáveis dependentes que possuem como característica índices ou variáveis que exijam ranqueamento Wooldrige 2010 Greene 2018 A regressão foi focada em um único período de tempo dezembro de 2018 ou seja cortetransversal Assim temse a regres são Na regressão a variável dependente é o índice de ineficiência financeira de cada banco ou cooperativa compreende ao índice de Basileia compreende ao índice de imobilização compreende ao índice de alavancagem financeira Assim foram realizadas duas regressões uma contendo apenas os bancos e outra contendo apenas as cooperativas Para aumentar o rigor da comparação foi utilizada como metodologia também uma varia ção do Propensity Score Matching conforme discutido de modo técnico por Angrist e Pischke 2009 e já utilizado na área de contabilidade e finanças por Armstrong Jagolinzer Lacker 2010 e Gue dhami Pittman Saffar 2014 O score utilizado foi a ineficiência presente nos bancos e nas coopera tivas Desse modo ao considerar o coeficiente de ineficiência de um banco buscouse nas coopera tivas uma instituição que possuísse um coeficiente de ineficiência igual ou semelhante ao apresen tado pelo banco Tal método é denominado de vizinho mais próximo e foi por ele que foi realiza da a correspondência Matching entre os bancos e cooperativas Angrist Pischke 2009 Por esse motivo tanto a regressão aplicada aos bancos quanto a regressão aplicada às cooperativas possu em cada uma 70 instituições 4 RESULTADOS 41 Fatores que influenciam a eficiência dos custos com intermediação financeira dos bancos O painel realizado para as instituições bancárias não possuía presença de autocorrelação e sua série temporal foi estacionária A presença ou ausência de autocorrelação foi avaliada por meio da função de autocorrelação ou FAC seguindo o proposto por Griffin e Stell 2007 e Tecles e Ta bak 2010 e Santos 2022 Significa dizer que há indícios de autocorrelação quando apesar dos números de lags os valores não tendem a zero o que não foi o caso aqui todos coeficientes já ten diam a zero antes de 20 lags Sobre a estacionariedade ela foi garantida por meio da Simulação de Monte Carlo por Cadeia de Markov de acordo com o que foi apresentado por Lunn Thomas Best e Spiegelhalter 2000 e Santos 2022 uma vez que ela assegura por meio das simulações que a série convirja para uma única média o que também aconteceu na regressão abaixo Assim o painel apresentou os seguintes resultados Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 34 Tabela 1 Fronteira de eficiência bancária Variáveis Coeficientes Desviopadrão MC erro 5 95 Significância Constante 70900 02360 00070 67040 74710 Depósitos 02141 02698 00054 02274 06580 Obrigações por em préstimos 15140 03025 00045 10100 20140 Taxa dos fundos 00004 00004 00000 00002 00010 Despesas operacio nais 32950 09832 00121 16860 49100 Preço do Trabalho 78750 36470 00425 19070 138100 Market Share 00359 00448 00015 00357 01107 Nível de capital próprio 25580 04312 00056 32720 18500 Nível de atraso 00079 00122 00001 00122 00280 Notas significante a 1 significante a 5 significante a 10 Fonte Elaborado pelos autores 2020 Os resultados da tabela 1 demonstram que os custos com intermediação financeira dos ban cos brasileiros são influenciados positivamente pelas obrigações por empréstimos com coeficiente de 15140 proporção de despesas operacionais com 3295 preço do trabalho com 7875 e influen ciado negativamente pelo nível do capital próprio com coeficiente de 2558 sendo todos estatisti camente significantes a um nível de 10 Para os bancos quando eles captam dinheiro com tercei ros são obrigados a pagar uma taxa desse modo quanto maior o nível de obrigação com terceiros maior será o seu custo de intermediação financeira Esse resultado é semelhante ao encontrado por Tecles e Tabak 2010 ao considerar os empréstimos como um fator de saída de recursos para os bancos pois também encontraram uma associação positiva e significativa entre ele e os custos A despesa com mãodeobra também foi estatisticamente significativa Isso se deve ao fato de os bancos precisarem de pontos físicos agências para vender seus principais produtos Esse resultado também é semelhante ao encontrado por Tecles e Tabak 2010 que obtiveram um resul tado positivo e significativo para esta variável Ainda que a maioria dos bancos esteja investido em suas plataformas digitais como apontado por F Malaquias et al 2018 e que tenham surgido bancos exclusivamente digitais tal fator não é capturado pelo modelo uma vez que os bancos to talmente digitais por exemplo foram excluídos da amostra por não satisfazerem ao período anali sado de 16 trimestres A despesa com trabalho foi significativa e positiva cujo resultado é semelhante ao encon trado por Kumbhakar e Tsionas 2005 e Tecles e Tabak 2010 Barros e Wanke 2018 considera ram as despesas com trabalho de três formas diferentes a primeira sendo apenas o valor do traba lho ponderado pelo capital da empresa o segundo foi o valor do trabalho ponderado pelo capital da empresa ao quadrado e o terceiro foi o valor do trabalho ponderado pelo capital multiplicado pelos empréstimos concedidos a clientes Das três considerações realizadas por Barros e Wanke 2018 para o valor do trabalho a primeira apresentou sinal diferente do que foi encontrado en quanto o segundo e terceiro teve adoção semelhante ao que foi achado O resultado negativo e significativo encontrado para a proporção do capital próprio sobre o total de ativo foi também encontrada por Koop et al 1997 e por Tecles e Tabak 2010 Esse resul tado aponta que quanto maior é o nível de capital próprio menor será a ineficiência Isso se deve Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 35 ao fato de que quanto maior for o volume de capital próprio aplicado pelos bancos mais seguras serão suas estruturas de capitais pois eles não dependerão tanto de uma fonte originada de tercei ros Após determinados os fatores que influenciam os custos dos bancos foi realizado um ran king de quais instituições estão acima da fronteira de eficiência estocástica A seguir é apresentada a fronteira de eficiência de bancos Figura 1 Fronteira de eficiência dos Bancos Fonte Elaborado pelos autores 2020 O índice de eficiência se dá em um espaço amostral que está entre 0 e 1 concebido por meio de uma distribuição exponencial sendo esse espaço entre 0 e 1 a área da distribuição Assim quan to mais próximo de 1 mais eficiente será a instituição Já a fronteira de eficiência é determinada pela posição média que cada instituição financeira ocupa nesse espaço bem como sua variância sendo esse valor individual atribuído a cada instituição por meio da modelagem bayesiana reali zada Desse modo a fronteira de eficiência é traçada ao levar em consideração a posição que to das as instituições ocupam para a amostra e representa uma aproximação das melhores práticas para se ter um menor nível de custos de modo a maximizar o conjunto de saída Griffin Stell 2007 Para os bancos essa fronteira de eficiência foi de 00478 nessa escala que varia entre 0 e 1 o que demonstra uma fronteira de eficiência mais próxima de zero no entanto ao posicionar cada banco nesse espaço amostral a maioria ficou abaixo da fronteira de eficiência O modelo considerou os 70 bancos da amostra desses apenas 16 foram eficientes o que corresponde a 2286 enquanto 54 ou 7714 ficaram abaixo da fronteira de eficiência Ao anali sar os que ficaram abaixo da fronteira de eficiência foi constatado que os maiores bancos do Brasil até 2018 estavam nessa região ou seja apesar do ganho de escala tido pelo market share eles são os menos eficientes em relação a gestão dos seus custos O motivo para que bancos de pequeno e mé dio porte sejam mais eficientes do que os 5 maiores bancos do país para o momento analisado têm como causa o fato de eles necessitarem operar com mais recursos de terceiros do que com os próprios para atender a quantidade de clientes Isso faz com que eles tenham uma maior despesa remunerando o capital de terceiros além de terem um maior volume financeiro com inadimplên cia de seus tomadores Esses resultados são semelhantes aos encontrados por Barros e Wanke 2014 quando constataram que os maiores bancos brasileiros possuíam menor eficiência em rela ção aos seus custos Fronteira de eficiência Bancos 100E5 100E4 0001 001 01 Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 36 42 Fatores que influenciam a eficiência dos custos com intermediação financeira das cooperati vas de crédito O painel realizado para as cooperativas de crédito não apresentou autocorrelação e sua série temporal foi estacionária A ausência de autocorrelação foi avaliada por meio da função de autocorrelação ou FAC também seguindo o proposto por Griffin e Stell 2007 e Tecles e Tabak 2010 e Santos 2022 Foi verificado que todos os coeficientes já tendiam a zero antes de 20 lags indicando ausência de autocorrelação A estacionariedade foi garantida por meio da Simulação de Monte Carlo por Cadeia de Markov também de acordo com o que foi apresentado por Lunn Thomas Best e Spiegelhalter 2000 e Santos 2022 uma vez que ela assegura por meio das simu lações que a série convirja para uma única média o que também aconteceu na regressão que em breve será apresentada Sobre a regressão os depósitos totais e o market share foram estatisticamente significativos para as cooperativas de crédito e de forma positiva os depósitos totais com coeficiente de 06025 e o market share com coeficiente de 3949 Com relação negativa apresentouse o nível de capital pró prio com 00136 Todos foram significativos a 10 A variável taxa dos fundos e o nível de atraso não foram estatisticamente significativas nem para os bancos nem para as cooperativas Os resul tados da tabela 2 demonstram que com exceção da variável nível de capital próprio que é estatisti camente significativa e que reduz o custo de intermediação financeira para os bancos e para as cooperativas os fatores que influenciam no custo de intermediação financeira para os bancos e para as cooperativas são diferentes O sinal negativo para o nível de capital próprio para os bancos e para as cooperativas demonstra que quanto maior o nível de capital próprio menor é o custo de intermediação financeira dessas instituições Os resultados do painel são apresentados a seguir Tabela 2 Fronteira de eficiência das cooperativas de crédito Variáveis Coeficientes Desviopadrão MC erro 5 95 Significância Constante 29810 00756 00012 28550 31040 Depósitos 06025 00828 00015 04669 07400 Obrigações por em préstimos 00322 00253 00003 00104 00737 Taxa dos fundos 00000 00000 00000 00000 00000 Despesas operacionais 00272 01177 00013 01672 02181 Preço do Trabalho 01422 02474 00026 02594 05526 Market Share 394900 66010 01288 285800 504100 Nível de capital pró prio 00136 00051 00001 00220 00052 Nível de atraso 00098 00095 00001 00058 00253 Notas significante a 1 significante a 5 significante a 10 Fonte Elaborado pelos autores 2020 Para as cooperativas quanto maior o nível de depósitos totais maior é o custo de interme diação financeira isso se deve pois diferente dos bancos que a principal obrigação com terceiros vem de valores obtidos com outras instituições bancárias as cooperativas possuem como sua prin cipal fonte de obrigação as captações obtidas por seus cooperados que são classificados como de pósitos totais Maia et al 2013 Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 37 Em relação ao market share as cooperativas de crédito apresentaram resultados estatistica mente diferentes daqueles apresentados pelos bancos e mais parecidos com os encontrados por Barros e Wanke 2014 para os bancos durante o período de crise financeira de 2008 Enquanto eles identificaram que bancos que possuíam agências na maior parte do Brasil tinham uma maior efici ência em relação aos seus custos esse resultado também pôde ser encontrado em relação às coope rativas de crédito O motivo pelo qual o porte das cooperativas influencia na eficiência dos custos se deve ao fato dos clientes ativos nas cooperativas de crédito serem os seus cooperados Assim a variável market share é a mais significativa uma vez que quanto maior for a quantidade de coope rados também será maior a quantidade de clientes que a cooperativa terá que atender Os bancos e cooperativas de crédito em relação ao nível de capital próprio apresentaram resultados semelhantes isso se deve ao fato de que quanto maior forem os recursos próprios em pregados na operação bancária em linhas gerais menor será os custos de intermediação financei ra uma vez que não terá que remunerar nenhum terceiro Esse resultado está de acordo com o encontrado por Tecles e Tabak 2010 que identificaram que no Brasil quanto maior o nível de capital próprio mais eficiente será a instituição financeira Figura 2 Fronteira de eficiência das cooperativas Fonte Elaborado pelos autores 2020 Para as 559 cooperativas de crédito foi considerada uma fronteira de eficiência de aproxi madamente 00642 em uma escala que varia entre 0 e 1 Uma vez que essa escala se dá por meio de uma distribuição exponencial uma fronteira de eficiência mais próxima de 1 representa que as chances de se ter instituições ineficientes são maiores uma vez que a área que fica abaixo da efici ência necessária é maior Nesse sentido Griffin e Stell 2007 argumentam que a fronteira estocásti ca representa uma aproximação das melhores saídas com base em um menor custo Assim 102 instituições ou 1825 foram eficientes enquanto 457 instituições correspondendo a 8175 foram ineficientes Os resultados das fronteiras de eficiência demonstram que as cooperativas possuem uma fronteira de eficiência maior que a dos bancos sendo respectivamente 00642 e 00478 Nas duas amostras a maioria das instituições ficaram abaixo da fronteira de eficiência apesar do aumento de regulação implementado no setor com os acordos de Basileia Portanto 7714 dos bancos e 8175 das cooperativas ficaram abaixo da eficiência esperada para o setor Fronteira de eficiência Cooperativas 100E5 100E4 0001 001 01 Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 38 44 Fatores que influenciam a ineficiência das instituições financeiras Uma vez que a maioria dos bancos e cooperativas de crédito ficaram abaixo da fronteira de eficiência buscouse analisar quais fatores obrigados pelos acordos de Basileia reduzem a inefici ência das instituições financeiras Foram considerados em uma regressão o índice de Basileia o nível de Imobilização e o grau de Alavancagem todos propostos por Basileia como medida que poderia reduzir a ineficiência das instituições financeira Para a regressão foi utilizado o modelo logit ordinal sendo o coeficiente de ineficiência o fator de ordem entre as instituições A tabela abaixo foi dividida em dois grupos de modo a con templar os bancos e as cooperativas Em ambos os casos a segunda coluna contém a probabilidade de ocorrência que é a chance que determinada variável possui em afetar o termo de ineficiência a terceira coluna contém os coeficientes do modelo logit para cada variável a quarta traz o VIF que é o fator de inflação das variáveis a quinta e a sexta coluna respectivamente trazem o erropadrão e o pvalue de cada variável Tabela 3 Índices obrigatórios e ineficiência Bancos1 Probabilidade de Ocorrência2 Coeficiente VIF Erro padrão Significância3 64424 0594 1080 1177 29630 0865 1080 2630 59632 0390 1160 1957 Cooperativas1 Probabilidade de Ocorrência2 Coeficiente VIF Erro padrão Significância3 0089 7018 1150 1508 99998 11796 1050 4190 14273 1793 1100 2599 Notas 1 O número de observação para os bancos foi de 70 instituições financeiras igual ao número de cooperativas que também foram de 70 instituições Essas 70 cooperativas foram escolhidas por apresenta rem o termo de ineficiência igual ou similar ao apresentado pelos bancos na técnica de correspondência chamada de vizinho mais próximo 2 A probabilidade de ocorrência PO se deu por PO exp coeficiente 1 exp coeficiente 3 significante a 1 significante a 5 significante a 10 Fonte Elaborado pelos autores 2022 Os resultados das regressões acima demonstram que o índice de Basileia Imobilização e Alavancagem não possuem correlação com a ineficiência de custos das instituições bancárias por outro lado para as cooperativas os índices de Basileia e Alavancagem tiveram uma correlação significativa a 1 com o termo de ineficiência de custos O fator de inflação das variáveis ou VIF demonstrou que para ambas regressões as variáveis foram consistentes ou seja não inflacionadas o que permite a inferência caso ela seja significativa Em relação aos bancos embora as probabilidades de ocorrência tenham apresentado per centuais acima de 50 para os índices de Basileia e Imobilização tal percentual não possui respal do uma vez que essas variáveis não foram significativas Entre os possíveis motivos estão a dife rença que os bancos da amostra possuem entre si seja em relação ao porte onde apenas 5 bancos juntos possuíam market share superior a 65 do mercado em dezembro de 2018 enquanto os ou tros 65 bancos possuíam juntos 18 de market share os outros 17 de market share correspondem as cooperativas de crédito da amostra e a outros bancos e cooperativas que por algum dos motivos listados na metodologia não fizeram parte da amostra Além do porte os bancos poderiam ser di gitais físicos ou híbridos com representação digital e física o que influenciaria na sua estrutura de custos Um terceiro possível motivo está nos principais produtos e públicoalvo que os bancos pos suíam o que poderia influenciar a sua estrutura da carteira de crédito Assim existiram na amos Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 39 tra bancos cuja principal atividade era a exploração de sua carteira comercial outros possuíam como atividade principal sua carteira de investimento e outros bancos cuja finalidade era a de fi nanciar produtos para terceiros de outras subsidiarias não financeiras como financiamento de veí culos e máquinas agrícolas Desse modo uma vez que os bancos variavam em estrutura porte públicoalvo perfil da carteira de crédito além de outros fatores não observáveis a sua ineficiência de custos não se correlaciona diretamente com os índices de Basileia Alavancagem e Imobilização Em relação ao índice de Basileia e Alavancagem os bancos podem optar a quem vai conce der seus empréstimos bem como pedir garantias reais se esse crédito for para o financiamento de bens duráveis e uma vez que são poucos os bancos que concentram a maior parte do mercado logo devem ser eles os mais lembrados por clientes que possuem um perfil de crédito arriscado Em relação ao índice de Imobilização os bancos digitais tendem a possuir menos estruturas físicas enquanto os bancos físicos com maior capilaridade no mercado tendem a possuir maior número de agências unidades de atendimento e consequentemente de pessoal alocado nesses lugares po rém esses bancos são minoria No tocante as cooperativas de crédito o índice de Basileia e o índice de Alavancagem estão correlacionados com a ineficiência dos custos No entanto embora o índice de Basileia esteja corre lacionado sua probabilidade é inferior a 1 o que significa dizer que sua chance de ocorrência é quase nula o que é justificável visto que o índice de Basileia tem como objetivo promover a efici ência e estabilidade das instituições financeiras no mercado Por outro lado uma cooperativa de crédito mais alavancada possui uma probabilidade superior a 99 de afetar o índice de ineficiên cia dos custos Assim em linhas gerais quanto mais alavancada é uma cooperativa mais sua estru tura de custos será pressionada o que pode aumentar sua ineficiência de custos por ter gastos ex tras No entanto em relação as cooperativas uma vez que sua atividade principal é suprir as ne cessidades dos seus cooperados sua estrutura de custos por vezes pode ser pressionada em prol desse objetivo Outros fatores que podem justificar esses achados em relação as cooperativas são que elas de forma geral se mostraram mais homogêneas entre si haja vista que o número de cooperativas ativas no Sistema Financeiro Nacional em dezembro de 2018 excedia ao de bancos em mais de sete vezes enquanto o total de bancos era de 135 instituições a quantidade de cooperativas era de 962 instituições Alinhado a isso nenhuma cooperativa de crédito possuía market share suficiente de modo a dominar o mercado sendo antes instituições consideradas de pequeno porte para o Siste ma Financeiro Nacional E um terceiro fator que deve ser levado em consideração é o fato de as cooperativas existirem primordialmente para atenderem as necessidades dos seus cooperados o que em muitos casos pode limitar sua cesta de produtos e serviços financeiros além de pressionar sua estrutura de custos 5 CONCLUSÃO Este artigo teve como objetivo analisar a forma com que os índices obrigatórios por Basileia afetam a eficiência de instituições financeiras brasileiras no período entre 2015 e 2018 Para atingir tal propósito foram coletados dados de 629 instituições financeiras das quais 70 instituições são bancárias e 559 são cooperativas de crédito A modelagem utilizada para encontrar a ineficiência e a eficiência bancária seguiu a mesma utilizada pelos estudos de Griffin e Stell 2007 e Tecles e Ta bak 2010 que criaram uma fronteira de eficiência estocástica e determina um fator de ineficiência Os principais resultados encontrados sobre quais fatores influenciam os custos das entida des financeiras apontam que obrigações com empréstimo proporção das despesas operacionais e preço pago pelo trabalho aumentam o custo dos bancos enquanto um maior nível de capital pró prio o reduz Para as cooperativas de crédito os fatores que elevaram seus custos foram os depósi tos totais e o market share e de modo negativo o nível de capital próprio Esses resultados demons Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 40 tram que bancos e cooperativas apesar de fazerem parte do SFN possuem diferentes fatores res ponsáveis por elevar seus custos mas que em ambos os casos quanto maior for o nível do capital próprio menor será o custo dessas entidades No que diz respeito à fronteira de eficiência foi demonstrado que as cooperativas de crédi to possuem uma zona de ineficiência maior que os bancos Também foi encontrado que os bancos foram proporcionalmente mais eficientes que as cooperativas sendo 7714 dos bancos e 8175 das cooperativas Esses resultados concordam de certo modo com o que foi encontrado nas regres sões logísticas para os bancos e cooperativas em que se evidenciou que uma maior alavancagem das cooperativas de crédito possui elevada probabilidade de influenciar no termo de ineficiência dos custos Ou seja as cooperativas de crédito que possuíam ineficiência de custos similares àque las apresentadas pelos bancos apesar de similares eram mais sensíveis a fatores que influencia vam seu nível de alavancagem como volume de depósito e nível de capital próprio o que afetava diretamente a eficiência em relação aos custos enquanto para os bancos não houve uma relação clara entre a gestão dos índices propostos por Basileia e a redução ou aumento do termo de inefici ência dos custos Esses resultados corroboram o que foi exposto por Pinheiro Savóia e Securato 2015 Goes Sheng e Schiozer 2016 e Cardoso et al 2019 sobre a gestão dos recursos de posse das entidades bancárias Desse modo concluise que o índice de imobilização não influencia a ineficiência dos bancos e das cooperativas enquanto o grau de alavancagem aumenta a ineficiência das cooperati vas de crédito Em relação ao índice de Basileia em si seus resultados não foram conclusivos uma vez que para os bancos ele apresentou elevada probabilidade mas não foi significativo enquanto para as cooperativas embora significativo seu efeito era mínimo em relação à ineficiência Vale ressaltar o que foi exposto por Maia et al 2013 que de forma geral as cooperativas de crédito são menos alavancadas uma vez que parte do seu capital é subsidiada pelos seus cooperados Nesse sentido Ventura Fontes Filho e Soares 2009 argumentam que o contexto das cooperativas é complexo uma vez que seus cooperados possuem uma relação de cliente e de fornecedor com as instituições fazendo com que para eles o lucro não seja o objetivo principal mas que apesar desse contexto desafiador as cooperativas não devem se acomodar em relação as suas ineficiências No mais por meio desses resultados é possível concluir que de modo geral para as coope rativas de crédito a observação e a boa gestão dos Índices de Basiléia Alavancagem Imobilização e Basiléia produzem um efeito positivo nos custos das cooperativas de modo a aumentar sua efi ciência Por outro lado para os bancos de forma geral a gestão dos Índices de Basiléia não reper cute diretamente no grau de ineficiência dos custos Possivelmente fatores não observados por este estudo como o seguimento do banco o tipo do consolidado bancário o tipo da carteira de crédito e a presença física eou digital da instituição são fatores que influenciam na sua estrutura de custos e consequentemente na ineficiência bancária Por fim este estudo possuiu como limitação o fato de a fronteira de eficiência bayesiana pressupor uma série de distribuições de probabilidade como o argumentado por Meeusen e Bro eck 1977 para este tipo de estudo Como sugestão para pesquisa futura podese verificar a forma que o cenário econômico local interfere na eficiência bancária REFERÊNCIAS Aigner D Lovell CAK Schmidt P 1977 Formulation and estimation of stochastic frontier production fuction models Journal of Econometrics 6 p 2137 httpsdoiorg1010160304 407677900525 Araújo A M H B Lustosa P R B Paulo E 2018 A ciclicidade da provisão para créditos de liquidação duvidosa sob três diferentes modelos contábeis Reino Unido Espanha e Brasil Re vista de Contabilidade e Finanças USP 29 76 p 97113 https1015901808057x201804490 Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 41 Armstrong C S Jagolinzer A D Larcker D F 2010 Chief Executive Officer Equity Incentives and Accounting Irregularities Journal of Accounting Research 48 2 p 225271 https 101111j1475679X200900361x Angrist J D Pischke J S 2009 Mostly Harmless Econometrics An Empiricistís Companion Princeton University Press Assaf A G Matousek R Tsionas E G 2013 Turkish bank efficiency Bayesian estimation with undesirable outputs Journal of Banking Finance 37 p 506517 httpsdxdoiorg10116jbankfin201209009 Barros C P Wanke P 2014 Banking efficiency in Brazil Journal of International Financial Mar kets Institutions Money 28 p 5465 httpsdxdoiorg101016jintfin201310004 Batesse G E Coelli T J 1992 Frontier production functions technical efficiency and panel data With application to paddy farmers in India Journal of Productivity analysis 3 12 p 153 169 httpsdoiorg101007BF00158774 BCB 2019a Organização do Sistema Financeiro Disponível em httpswwwbcbgovbrestabilidadefinanceiraorganizacao Acesso em 24 de set de 2019 BCB 2019b Dados selecionados de instituições financeiras supervisionadas Disponível em httpswww3bcbgovbrifdata Acesso em 24 de set de 2019 Behr A 2010 Quantile regression for robust bank efficiency score estimation European Journal of Operational Reseach 200 p 568581 httpsdoiorg101016jejor200812033 Beneditto E A N Silva R N S 2008 Análise do nível de transparência das instituições fi nanceiras brasileiras em relações ao acordo com o novo acordo de capitais Basiléia II Um es tudo de caso múltiplo Brazilian Business Review 5 3 p 192208 httpsdxdoiorg1015728bb2008532 Broeck J D Koop G Osiewalski J Steel MFJ 1994 Stochastic frontier models A Bayesian perspective Journal of Econometrics 61 p 273303 httpsdoiorg1010160304407694900876 Brito G A Lopes A B Coelho A C D 2012 Conservadorismo nos lucros contábeis dos bancos no Brasil Influência do controle estatal Revista Universo Contábil 8 3 p 1939 httpsdoiorg104270ruc2012429 Cardoso V R S Campos J A Dantas J A Medeiros O R 2019 Fatores relacionados à li quidez estrutural dos bancos no Brasil Revista de Contabilidade e Finanças USP 30 80 p 252 267 https1015901808057x201806350 Carvalho F L Diaz M D M Bialoskorski neto S Kalatzis A E G 2015 Saída e insucesso das cooperativas de crédito no Brasil Uma análise do risco Revista de Contabilidade e Finanças USP 26 67 p 7084 https1015901808057x201411390 Ferreira J H Zanini F A M Alves T W 2019 A diversificação de receitas bancárias seu impacto sobre risco e retorno dos bancos brasileiros Revista de Contabilidade e Finanças USP 30 79 p 91106 https1015901808057x201805810 Goes K C Sheng H H Schiozer R F 2016 Contingentes conversíveis e seus impactos na otimização da estrutura de capital de bancos brasileiros sob Basileia III Revista de Contabilidade e Finanças USP 27 70 p 8097 https1015901808057x201501350 Greene W H 2018 Econometric Analysis Pretince Hall Griffin J Stell M 2004 Semiparametric Bayesian inference for stochastic frotier models Jour nal of Econometrics 123 p 121152 httpsdoiorg101016jeconom200311001 Griffin J Stell M 2007 Bayesian stochastic frontier analysis using winbugs Journal of Produc tivity Analysis 27 p 163176 httpsdoiorg101007s111230070033y Guedhami O Pittman J A Saffar W 2014 Auditor Choice in Politically Connected Firms Jour nal of Accounting Research 52 nº1 p 107167 httpsdoiorg1011111475679X12032 Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 42 Kinas P G Andrade H A 2010 Introdução à análise bayesiana com R Porto Alegre maisQnada Editora Koop G Osiewalski J Steel M 1997 Bayesian efficiency analysis through individual effects Hospital cost frontiers Journal of Econometrics 76 77105 httpsdoiorg1010160304 407695017836 Kumbhakar S C Tsionas G T 2005 Measuring technical and allocative inefficiency in the translog cost system a Bayesian approach Journal of Econometrics 126 p 355384 httpsdoiorg101016jjeconom200405006 Lunn DJ Thomas A Best N Spiegelhalter 2020WinBUGS A Bayesian modelling frame work Concepts structure and extensibilityStatistics and Computing10pp 325337 httpsdoiorg101023A1008929526011 Macedo M A S Kelly V L A 2016 Gerenciamento de resultado em instituições financeiras no Brasil Uma análise com base em provisões para crédito de liquidação duvidosa Revista E videnciação Contábil finanças 4 2 p 8296 httpsdoiorg1018405recfin20160206 Maia S C Bressan V G F Lamounier W M Braga M J 2013 Gerenciamento de resultado em cooperativas de crédito no Brasil Brazilian Business Review 10 4 p 96116 httpsdxdoiorg1015728bb20131044 Malaquias F Malaquias R Hwang Y 2018 Understanding the determinants of mobile bank ing adoption a longitudinal study in Brazil Electronic Commerce Research and Applications 30 p 17 httpsdoiorg101016jelerap201805002 Meeusen W Broeck J D 1977 Efficiency estimation from CobbDouglas production func tions with composed error International Economic Review 18 2 p 435444 httpsdoiorg1023072525757 Ntzoufras I 2009 Bayesian Modeling Using WinBUGS New Jersey 1ª ed John Wiley Sons https 1010029780470434567 Pinheiro F A PSavóia J R F Securato J R 2015 Basileia III Impacto para os Bancos no Brasil Revista de Contabilidade e Finanças USP 26 69 p 345361 https1015901808 057x201500720 Sanches F Silva Júnior D Srisuma S 2018 Banking privatization and market strutucture in Brazil a dynamic structural analysis Journal of Economics 49 4 p 936963 httpsdoiorg1011111756217112257 Santos L J Macedo M A S Rodrigues A 2014 Determinantes do nível de divulgação das recomendações do Pilar 3 do acordo de Basileia 2 nas demonstrações financeiras de institui ções bancárias que atuam no Brasil Brazilian Business Review 11 1 p 2650 httpsdxdoiorg1015728bb20141112 Santos WP Rodrigues R N Anjos L C M Tavares M F N 2019 Análise da influência do cenário econômico interno e externo na prática de income smoothing em bancos brasileiros Re vista Evidenciação Contábil Finanças 7 1 p 7893 httpsdoiorg1022478ufpb2318 10010v0n037439 Santos W P 2022 Adeus Floresta Amazônica Um Estudo Sobre o Desmatamento na Amazônia Legal à Luz do Gasto Público Estadual e dos Fatores Socioeconômicos Locais Revista Evidenci ação Contábil Finanças 102 4263 httpsdoiorg1022478ufpb231810012022v10n257591 Sousa LO 2018 Gerenciamento de resultado em bancos Indícios relacionados à aversão à di vulgação de prejuízos Revista Evidenciação Contábil Finanças 6 1 p 83100 httpsdoiorg1018405recfin20180105 Tecles P L Tabak B M 2010 Determinants of bank efficiency the case Brazil European Jour nal of Operational Reseach 207 p 15871598 httpsdoiorg101016jejor201006007 Santos W P Bezerra E S Pereira M P Revista Evidenciação Contábil Finanças ISSN 23181001 João Pessoa v 11 n1 p 2243 JanAbril 2023 43 Ventura E C F Fontes Filho J R Soares M M 2009 Governança Corporativa Diretrizes e me canismos para fortalecimento da governança em cooperativas de crédito Brasília BCB 257 Wooldridge J M 2010 Econometric analysis of cross section and panel data Massachusetts Institute of Technology United States 2ª Ed Zhag X 2000 A Monte Carlo study on the finite sample properties of the Gibbs sampling meth od for a stochastic frontier model Journal of Productivity Analysis 14 p 7183 httpsdoiorg101023A1007895912705 Citation Shen Bingbing Aleksandr Aleksandrovich Perfilev Lidiya Pavlovna Bufetova and Xueyan Li 2023 Bank Profitability Analysis in China Stochastic Frontier Approach Journal of Risk and Financial Management 16 243 https doiorg103390jrfm16040243 Academic Editor Yong Tan Received 13 March 2023 Revised 7 April 2023 Accepted 13 April 2023 Published 16 April 2023 Copyright 2023 by the authors Licensee MDPI Basel Switzerland This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution CC BY license https creativecommonsorglicensesby 40 Journal of Risk and Financial Management Article Bank Profitability Analysis in China Stochastic Frontier Approach Bingbing Shen 1 Aleksandr Aleksandrovich Perfilev 1 Lidiya Pavlovna Bufetova 1 and Xueyan Li 2 1 Department of Economics Novosibirsk State University 2 Pirogov Str 630090 Novosibirsk Russia aperfilevgnsuru AAP lidabufgmailcom LPB 2 School of Economics and Business Administration Heilongjiang University Harbin 150080 China lixueyan0451aliyuncom Correspondence shenbingbing724gmailcom Abstract Chinas banking system has a relatively high level of state control while an important task in regulating the banking system is to manage the profitability of banks Using the stochastic frontier approach to assess the profitability of commercial banks not only allows for the banks ability to generate profits relative to the leading banks in the industry to be assessed but also takes into account the specifics of the management technologies used and the influence of the market environment This article analyzes the profitability of the Chinese banking system for the period 20122020 using the stochastic frontier approach from the position of the central bank The specifics of the analysis from the banks perspective imply a focus on the position of most banks regarding the level of best practices and trends in changing the overall level of profitability Analysis may be of interest to banking regulators and researchers In general the Chinese banking system demonstrates a high level of profit efficiency and cost efficiency although the dynamics of these indicators are negative The reason for the negative dynamics is a decrease in the economic growth rate of the economy the instability of the financial market and ongoing reforms Stateowned commercial banks are becoming highly profitable while national jointstock commercial banks are facing increasing competition and reducing efficiency of profitability City and rural commercial banks maintain a high level of profitability due to state support Keywords bank profitability stochastic frontier approach commercial bank ROA ROE profit efficiency cost efficiency JEL Classification G21 G31 1 Introduction The banking systems of developing countries tend to have a relatively high level of state control Many scholars believe that this inhibits competition and the efficient redistribution of financial resources according to Lee et al 2020 and Le and Ngo 2020 According to Huang et al 2021 and Mai et al 2022 developing countries should pursue a policy of deregulation and develop market mechanisms However the 2008 crisis and subsequent prolonged stagnation in the banking sector of developed countries against the background of Chinas significant economic success force a new look at the regulation of financial institutions competently combining administrative and market methods One important condition that central banks observe when regulating the banking system is the need to keep commercial banks interested in making profits Control over the profitability of credit institutions can help in understanding how effectively banks of different types and sizes function whether they need help and whether changes in the existing mechanisms of market regulation are necessary Central banks evaluate the profitability of individual banks using indicators such as return on assets and return on J Risk Financial Manag 2023 16 243 httpsdoiorg103390jrfm16040243 httpswwwmdpicomjournaljrfm J Risk Financial Manag 2023 16 243 2 of 20 capital and also pay attention to the share of profitable banks and the profitability of banks of different sizes and types as indicated by Capraru and Ihnatov 2014 and Lin and Zhang 2009 Statistical and regression methods are used to assess the degree of influence of banking determinants on profitability Djalilova and Piesse 2016 Saona 2016 Tan 2017 This allows for us to understand the extent to which the central banks regulatory methods and changes in the economy will affect the activities of credit institutions It is important for the central bank to make sure that commercial banks are able to generate the expected profits for their owners and are interested in developing the banking business To organize profitability control the central bank needs to develop appropriate methods of analysis The profitability of the banking sector has a direct impact on the effective use of financial resources and the rational allocation of public resources Athanasoglou et al 2008 The profitability of banks is a key element in ensuring the sustainability of the banking sector de Guindos 2019 Profitability measures the overall level of development and sustainable efficiency of commercial banks VeraGilces et al 2020 The need to be profitable is spelled out in the Commercial Bank Law of China1 the need to control the profitability of banks follows from the law on the Peoples Bank of China2 In recent years the profitability of commercial banks in China has significantly decreased as a result of slowing economic growth increased competition from nonfinancial organizations falling rates on financial markets and tightening capital adequacy requirements The banking sector regulator the Peoples Bank of China began to tighten capital adequacy requirements in 2011 and another liberalization of interest rates occurred in 2015 This complicates the activities of banks and requires efforts to ensure that commercial banks comply with regulatory requirements on the one hand and on the other hand retain the ability to operate on market principles and above all create profits The need to assess profitability from the position of regulatory and supervisory au thorities and identify trends and reasons for changing the behavior of banks requires the development of scientific approaches that allow for these problems to be solved As the results of a study by Chinese and foreign authors show the stochastic frontier approach helps to significantly improve the assessment of the profitability of commercial banks from the perspective of their technological capabilities and the influence of managerial and environmental factors A feature of the stochastic frontier approach is the possibility to construct the produc tion possibilities frontier of companies and determine each companys distance from this frontier Thus it becomes possible to evaluate the effectiveness of all or most firms in the industry The application of the stochastic frontier approach to assess the profitability of the Chinese banking sector requires that local features are taken into account This is due to the allocation of groups of banks that focus on certain tasks and types of clients and have different levels of access to resources and different opportunities for the placement of assets These groups are the conductors of the monetary policy of the state in certain areas of the economy We chose the period from 2012 to 2020 for this study This period is characterized by important changes in the economy and the banking sector In 2013 the Chinese government announced the launch of projects related to the export of capital to the countries of Asia Africa and Latin America3 During this period there was a slowdown in economic growth a significant decrease in interest rates in financial markets the development of banking technologies on a digital basis and the emergence of shadow banking The noted changes affected the activities of banks and forced them to adapt to new conditions which inevitably affected their profitability The movement of capital especially within the framework of international and na tional infrastructure projects contributed to the development of systemic banks and in fluenced their growth Jointstock banks also received the opportunity to increase the volume of their operations by servicing the growth of large companies the growth of household incomes and their operations Urban and rural banks were stimulated to grow J Risk Financial Manag 2023 16 243 3 of 20 by participating in programs to promote the development of urban and rural areas Foreign banks mainly paid attention to the growing integration of the Chinese economy into the world economy Chinese and foreign scientists have paid attention to assessing the effectiveness of the banking system in different periods of its development Fu and Heffernan 2007 One can single out works Horrocks et al 2008 that assess the success of Chinese reforms in the period up to 2012 They draw conclusions regarding the beginning of the process of transformation of the national banking system to market conditions We did not find works on a later period although serious changes occurred at this time both in the PRC and in the world economy One feature of our analysis is an attempt to analyze the profitability of Chinese banks in the period from 2012 to 2020 taking into account internal specifics from the position of the central bank of China This specificity is associated with the following features the presence of different groups of banks with different levels of involvement in state regulation the implementation of financial market reforms which contributed to a radical reduction in market rates the secrecy of data on the financial markets of China Therefore in order to optimize the commercial profitability and profit efficiency of the Chinese banking sector it is necessary to manage and improve the quality of banking assets Therefore this work is of great importance when studying the performance of listed banks in China and the factors influencing this The stochastic frontier approach can significantly improve the analysis of the prof itability of the banking system For China this is of particular importance because of the need to evaluate the results of both reforms and the results of economic regulation which may affect the interests of commercial banks The analysis of the profitability of the banking system should be based on its current level of profitability which is determined by the current frontal boundary The deviation of banks from this boundary allows for profitability trends to be determined When analyzing profitability it is important to consider the state of costs which will allow us to determine the reason for the change in the state of profitability 2 Characteristics of the Chinese Banking System from 2012 to 2020 21 The Structure of the Chinese Banking System Chinas banking system is characterized by a high degree of centralization and the presence of banks of various types Mai et al 2022 Half of the assets of the banking system belong to large banks The share of foreign banks is insignificant Big banks also hold more than 60 of deposits and up to 50 of loans The big banks are the core of the banking system and they should be the focus of economic analysis Basically these are large national jointstock commercial banks which are owned by the state and are called state commercial banks4 SO Their shares are traded on various world stock exchanges The peculiarity of stateowned commercial banks is their focus on the financing of large infrastructure and projects that are important to the country However they successfully compete with other banks carrying out banking operations with households and firms Political banks PB5 are directly managed by the state and mainly provide banking services to stateowned companies Their purpose is to stimulate the development of specific industries Political banks do not operate with the aim of maximizing profits National jointstock commercial banks NJ6 meet the financing and savings needs of enterprises and households they contribute to the formation of a competitive banking system These banks are majorityowned by the state either through government agencies or stateowned enterprises These are relatively large credit institutions that have been corporatized by government order with the aim of creating marketoriented organizations without government administration for the benefit of shareholders Foreign commercial banks FB established as branches subsidiaries or independent banks with foreign participation focus on servicing projects involving the combination of domestic and foreign enterprises they can help the customer develop a joint lending J Risk Financial Manag 2023 16 243 4 of 20 structure for domestic and foreign enterprises and provide services to customers related to foreign trade operations Rural commercial banks RB are jointstock local financial institutions established by farmers rural industrial and commercial enterprises and corporate entities Their main objective is to provide financial services for local farmers agriculture and rural economic development and to promote the coordinated development of the urban and rural economy All the rural cooperative banks have been converted into rural commercial banks City commercial banks CB used to exist as urban credit cooperatives Since 1995 the state council of China decided to establish city commercial banks in large and medium sized cities Most city commercial banks show flexibility and responsiveness to market changes compared to large banks creating a strong attraction for enterprises with urgent capital needs City commercial banks have the advantage that they are closely linked to the local economy and have a more detailed and accurate understanding of the lending capacity and performance of local customers The government can stimulate certain groups of banks to develop Infrastructure projects state commercial banks play an important role here to increase the wealth of rural regions rural commercial banks play an important role here and to develop business in cities urban commercial banks Therefore the profitability of some banks can be high not only because of their efficient operation but also as a result of incentives from the state to promote their development Between 2012 and 2020 Chinas banking system underwent significant changes The period began with a slowdown in economic growth and a search for new ways to develop the economy These paths were related to capital exports and implementation of large scale infrastructure projects Li et al 2023 The crash of 20152016 revealed problems that had accumulated in the financial sector which required decisive intervention by the government7 In 2018 trade relations with the US were deteriorating leading to a decline in exports Yang et al 2023 At the same time foreign commercial banks were sharply reducing service revenues During this period the mobile internet became increasingly popular and mobile payments and mobile ecommerce were becoming a way of life The financial innovations represented by the mobile internet are having an impact on the banking industry The work of Scotta et al 2017 showed that digital innovation has had a significant impact on the longterm profitability of banks The authors of Zhao et al 2022 investigated the impact of fintech innovation on the performance of Chinese banks They found that fintech innovation reduces bank profitability and overall asset quality which is more pronounced among large stateowned commercial banks but improves capital adequacy and governance efficiency 22 Profitability of Chinese Banks in the Period 20122020 Chinas economic growth has slowed since 2012 and longstanding structural con tradictions continue to emerge declining export prices for manufactured goods lower real enterprise profits lower fiscal revenue growth and increased economic risk potential Li 2020 In July 2013 the lending rates for banks and other financial institutions were fully liberalized Banks were allowed to set their own lending rates and on 22 November 2014 the Peoples Bank of China reduced the RMB lending and deposit rates of finan cial institutions using the base rate In 2015 deposit insurance rules were announced Lee and Hsieh 2013 and China formally established a deposit insurance system Despite a slowdown in economic growth Chinas interest rate liberalization reform was completed with further expansion of floating range of deposit rates and the introduction of a deposit insurance system We show in Figure 1 average ROA and ROE for Chinas banking sector decline over the period 20132020 As a result of interest rate liberalization bank profitability has declined Sun 2020 but the level of decline is in line with the standards of developed countries J Risk Financial Manag 2023 16 243 5 of 20 J Risk Financial Manag 2023 16 x FOR PEER REVIEW 5 of 21 Figure 1 Evolution of average ROA and ROE in Chinas banking sector over the period 20132020 Source authors preparation httpswwwtheglobaleconomycomChina accessed on 28 Decem ber 2022 Notes abscissa axisyears left ordinate axisaverage return on assets of Chinas banking system right ordinate axisaverage return on equity of Chinas banking system Interest rate liberalization Li et al 2021 has allowed banks the autonomy to set their own deposit and lending rates In a highly competitive environment this has narrowed the spreads on the yield differential between attracted and lent resources and negatively affected banks profitability Since 2012 the Chinese government has understood that the further development of the economy and society cannot be confined to the production of goods and must move to a new level the export of capital The Belt and Road Initiative is a largescale project to improve regional cooperation by improving connectivity between countries along the an cient Silk Road and beyond Li et al 2023 As of February 2019 the Belt and Road Initia tive has triggered an increase in transportation infrastructure investment in 96 countries along the Belt and Road since its inception in 2013 To service the construction of the One Belt One Road initiative Chinese banks are strengthening their financial support by de veloping relevant strategic plans improving the structure of their institutions and service network facilitating credit support for major projects along the One Belt One Road re gion and implementing business innovation institutional innovation and managerial in novation During the period under study banks in China had competitors in the form of com panies that along with their core business offer a range of banking services to customers We are talking primarily about the Alipay platform and Yue Bao which have had a pro found impact on the financial industry but there are also other competitors Zhao et al 2022 The growth of internet financing has attracted a significant portion of depositors funds to the companies in question which will undoubtedly put significant pressure on banks whose main form of income is interest For central banks ROA and ROE are the most important indicators of profitability as they show how banks generate profits under regulatory conditions In the chart below we analyze the average return on assets ROA how much net profit a bank generates per unit of assets and average the return on equity ROE the banks ability to generate net profit from its net assets for commercial banks in China over the period 20122020 As noted there has been a steady decline in bank asset returns over the past decade but this has varied across bank types There is a spread between rural banks urban banks and jointstock banks possibly related to the stimulus processes undertaken by the gov ernment National jointstock commercial banks show a slower rate of decline than others which suggests a relative increase in return on assets relative to other types of banks City commercial banks show a steeper decline in returns than others they are perhaps nega tively influenced by the processes related to shadow banking and intrabank competition as discussed above The Central Bank should perhaps pay attention to the development 166 155 134 117 115 103 104 096 2667 2431 1884 1649 1599 1369 1311 1173 0 05 1 15 2 0 5 10 15 20 25 30 2013 2014 2015 2016 2017 2018 2019 2020 ROAin percent ROEin percent Year Bank return on assetsROA Bank return on equityROE Figure 1 Evolution of average ROA and ROE in Chinas banking sector over the period 2013 2020 Source authors preparation httpswwwtheglobaleconomycomChina accessed on 28 December 2022 Notes abscissa axisyears left ordinate axisaverage return on assets of Chinas banking system right ordinate axisaverage return on equity of Chinas banking system Interest rate liberalization Li et al 2021 has allowed banks the autonomy to set their own deposit and lending rates In a highly competitive environment this has narrowed the spreads on the yield differential between attracted and lent resources and negatively affected banks profitability Since 2012 the Chinese government has understood that the further development of the economy and society cannot be confined to the production of goods and must move to a new level the export of capital The Belt and Road Initiative is a largescale project to improve regional cooperation by improving connectivity between countries along the ancient Silk Road and beyond Li et al 2023 As of February 2019 the Belt and Road Initiative has triggered an increase in transportation infrastructure investment in 96 countries along the Belt and Road since its inception in 2013 To service the construction of the One Belt One Road initiative Chinese banks are strengthening their financial support by developing relevant strategic plans improving the structure of their institutions and service network facilitating credit support for major projects along the One Belt One Road region and implementing business innovation institutional innovation and managerial innovation During the period under study banks in China had competitors in the form of compa nies that along with their core business offer a range of banking services to customers We are talking primarily about the Alipay platform and Yue Bao which have had a profound impact on the financial industry but there are also other competitors Zhao et al 2022 The growth of internet financing has attracted a significant portion of depositors funds to the companies in question which will undoubtedly put significant pressure on banks whose main form of income is interest For central banks ROA and ROE are the most important indicators of profitability as they show how banks generate profits under regulatory conditions In the chart below we analyze the average return on assets ROA how much net profit a bank generates per unit of assets and average the return on equity ROE the banks ability to generate net profit from its net assets for commercial banks in China over the period 20122020 As noted there has been a steady decline in bank asset returns over the past decade but this has varied across bank types There is a spread between rural banks urban banks and jointstock banks possibly related to the stimulus processes undertaken by the government National jointstock commercial banks show a slower rate of decline than others which suggests a relative increase in return on assets relative to other types of banks City commercial banks show a steeper decline in returns than others they are perhaps negatively influenced by the processes related to shadow banking and intrabank competition as discussed above The Central Bank should perhaps pay attention to the J Risk Financial Manag 2023 16 243 6 of 20 development of banks in this sector Foreign commercial banks show the movement of asset yields around the average value of 05 This can be regarded as the lower boundary within which the profitability of local banks can be reduced without affecting the financial stability of the credit institution Return on equity ROE considers the level of financial leverage and characterizes the profitability of bank owners reflecting their level of interest in business development The Central Bank is interested in creating the necessary conditions for the banking sector to be attractive to investors The right part of Figure 2 shows the dynamics of ROE for the observed period It can be noted that foreign commercial banks show a level in the range of 3 Chinese banks show a decline in ROE which as noted was affected by the general decline in interest rates in the economy The highest level of return on capital is shown by national jointstock commercial banks which seems to be due to their more effective management of the capital structure This shows the success of measures to introduce market mechanisms into the banking sector Rural commercial banks demonstrate good adaptation to market conditions They manage to compensate for the overall decline in return on assets and show a weaker decline in ROE than other banks Urban commercial banks show similar ROA and ROE dynamics J Risk Financial Manag 2023 16 x FOR PEER REVIEW 6 of 21 of banks in this sector Foreign commercial banks show the movement of asset yields around the average value of 05 This can be regarded as the lower boundary within which the profitability of local banks can be reduced without affecting the financial stabil ity of the credit institution Return on equity ROE considers the level of financial leverage and characterizes the profitability of bank owners reflecting their level of interest in business development The Central Bank is interested in creating the necessary conditions for the banking sector to be attractive to investors The right part of Figure 2 shows the dynamics of ROE for the ob served period It can be noted that foreign commercial banks show a level in the range of 3 Chinese banks show a decline in ROE which as noted was affected by the general decline in interest rates in the economy The highest level of return on capital is shown by national jointstock commercial banks which seems to be due to their more effective man agement of the capital structure This shows the success of measures to introduce market mechanisms into the banking sector Rural commercial banks demonstrate good adapta tion to market conditions They manage to compensate for the overall decline in return on assets and show a weaker decline in ROE than other banks Urban commercial banks show similar ROA and ROE dynamics a b Figure 2 a Average return on assets by bank type b average return on equity by bank type Source authors preparation Notes left axis of ordinatesaverage return on assets for a group of banks abscissa axisyears right axis of ordinatesaverage return on equity for a group of banks abscissa axisyears categoriesgroups of banks state commercial banks SO national joint stock commercial banks NJ foreign commercial banks FB rural commercial banks RB urban com mercial banks CB Thus we can observe a decline in profitability indicators in the banking system of China over the observed period Most experts Huang 2020 see this as positive associated with economic development and risk reduction In this case the use of SFA which allows us to understand how banks are positioned relative to the production capacity boundary allows for us to assess the effectiveness of banks in generating profits more accurately than traditional methods 3 Literature Review on Banking Profit Efficiency and Cost Efficiency In practice two boundary methods are the most popular the stochastic frontier ap proach SFA Aigner et al 1997 and data envelopment analysis DEA Charnes et al 1978 The SFA method is parametric it describes the inputoutput relationship using pro duction functions such as the CobbDouglas function The DEA method determines the production possibility boundary by constructing linear combinations of input and output parameters Both methods show similar results and the choice of one of them can be re lated both to the preference of the researcher and the specifics of the data Avkiran 2011 Figure 2 a Average return on assets by bank type b average return on equity by bank type Source authors preparation Notes left axis of ordinatesaverage return on assets for a group of banks abscissa axisyears right axis of ordinatesaverage return on equity for a group of banks abscissa axisyears categoriesgroups of banks state commercial banks SO national joint stock commercial banks NJ foreign commercial banks FB rural commercial banks RB urban commercial banks CB Thus we can observe a decline in profitability indicators in the banking system of China over the observed period Most experts Huang 2020 see this as positive associated with economic development and risk reduction In this case the use of SFA which allows us to understand how banks are positioned relative to the production capacity boundary allows for us to assess the effectiveness of banks in generating profits more accurately than traditional methods 3 Literature Review on Banking Profit Efficiency and Cost Efficiency In practice two boundary methods are the most popular the stochastic frontier approach SFA Aigner et al 1997 and data envelopment analysis DEA Charnes et al 1978 The SFA method is parametric it describes the inputoutput relationship using production functions such as the CobbDouglas function The DEA method determines the production possibility boundary by constructing linear combinations of input and output parameters Both methods show similar results and the choice of one of them can be related both to the preference of the researcher and the specifics of the data Avkiran 2011 Wang et al 2014 Henriques et al 2020 Since the late 1980s several studies have J Risk Financial Manag 2023 16 243 7 of 20 been conducted to measure the profit efficiency and cost efficiency of banks Estimates of banks profit efficiency and cost efficiency are important and the results of the studies have important implications for bank practitioners and policymakers Silva et al 2017 DEA is a nonparametric approach and there is no concept of a production function in DEA analysis but there is the concept of a production frontier The DEA method determines the production possibility frontier by constructing linear combinations of input and output parameters Avkiran 2011 Using the DEA method to study the efficiency of 16 major commercial banks during banking reform in China 20032011 Wang et al 2014 concluded that the overall efficiency of stateowned commercial banks was higher than that of jointstock commercial banks before the reform while the efficiency gap between stateowned and jointstock commercial banks narrowed after the reform The SFA method is parametric and uses production functions such as the Cobb Douglas function to describe the inputoutput relationship Using the stochastic frontier approach authors Pasiouras et al 2009 studied 615 commercial banks operating in 74 countries over the period 20002004 The research showed that stronger banking su pervision increases the cost and profit efficiency of banks The authors of Ngan 2014 measured the cost and profit efficiency of 45 Vietnamese commercial banks in the period 20072012 using the stochastic frontier analysis SFA method and the results showed that stateowned commercial banks SCOBs are more efficient in terms of profit efficiency than other domestic commercial banks JSCBs and foreign banks while foreign banks are ahead of stateowned banks in terms of cost efficiency The authors Aielloa and Bonanno 2016 using the stochastic frontier analysis SFA method showed that cost efficiency was not related to market risk in Italian banks in 20062011 while profit efficiency was related to market risk The cost efficiency of commercial banks is inversely related to loan diversification The profits of Italian commercial banks are negatively affected by size The authors Dong et al 2016 examined the cost and profit efficiency of four types of commercial banks in China over the period from 2002 to 2013 The study shows that cost and profit efficiency increased for stateowned banks jointstock banks foreign banks and city banks and that banks were more profitefficient than costefficient Foreign banks are the most costefficient but the least profitefficient Using the stochastic frontier approach Bitar et al 2018 analyzed 1992 banks in 39 OECD countries over the period 19992013 and found that higher capital ratios may harm the banking efficiency and profitability of highly liquid banks In summary our review of the literature on profit efficiency and cost efficiency in the banking sector reveals some mixed and contradictory results and research gaps Since 2012 rural commercial banks have played an increasingly important role in the Chinese banking system but existing studies have underresearched rural commercial banks in China Commercial banks in China vary in their ability to effectively apply banking technol ogy The choice of methodology may be related to the specifics of the data In our study we added the classification of Chinese commercial banks in China to the existing research analysis of rural commercial banks and used the SFA methodology because its results are illustrative and consider the inefficiency term 4 Methodology 41 Empirical Method In our study we used the SFA method because its results are quite clear it is robust to outliers and it allows for us to build a boundary based on the most widespread use of the technology Commercial banks in China differ in their effective application of banking technology and in their access to banking product markets as well as having different sizes In this case focusing on the most widespread use of technology would provide a fairer estimate Note that this assessment would be fair to banks when considered as separate entities without taking into account their importance to the banking sector This is important when creating a fair environment for all participants in the banking sector The stochastic frontier approach originally relied on a neoclassical theory of the firm which attempts to explain the relationship between production profits and costs Aigner et al 1997 Coelli et al 2005 Firms were assumed to operate in perfect outputcost markets and input prices were defined as exogenous Competition leads to the price equalization of inputs and outputs Subsequently it was argued that profitability cannot be explained by production factors alone but also by factors related to management and the behavior of key stakeholders Silva et al 2017 Lee and Hsieh 2013 The application of frontier methods requires compliance with a number of requirements that are difficult to capture in real practice all firms in the analysis are competing the products are homogenous all firms produce identical products and use identical resources and firms have an identical function in converting resources into final products The stochastic frontier approach considers the set of firms as a set of technologies which are characterized by the terms input and output where inputs are resources and outputs are the products obtained by using these resources They can be used to determine the boundary of production capacity at which firms use the minimum amount of resources to obtain the maximum volume of production Battese and Coelli 1995 Berger and Humphrey 1997 emphasized the advantages of effective frontier approaches compared to traditional performance measures such as return on assets because they take into account both the banks respective inputs and outputs and differences in the prices of inputs The SFA method assumes that not all components of the residuals of the production function built on the panel data are of a random nature and distinguishes nonrandom u and random v components in the residuals yit fxit β vit uit 1 vit N0 σ2v uit N0 σ2u 2 where yit denotes the production at observation t 1 2 T xit is a 1k vector of the values of known functions of production inputs and other explanatory variables associated with the ith firm at the tth observation β is a k1 vector of unknown parameters to be estimated8 vit are random errors which are distributed independently of the uit uit are nonnegative random variables associated with the technical inefficiency of production which are assumed to be independently distributed In most works on similar topics it is assumed that the random component v has a normal distribution with a mean of zero and component u has a seminormal distribution both quantities are independent The parameter u is used to measure the production efficiency of individual organizations as part of their cumulative total namely to estimate the difference between the output of an individual company and the maximum possible output The efficiency score is estimated using a stage stochastic frontier model Battese and Coelli 1995 Coelli et al 2005 in which the effect of environmental variables on inefficiency is modeled Historically the scientific basis for the construction of production functions was the neoclassical theory of the firm according to which in a perfect market firms had equal access to both resources and sales and competed with each other by reducing individual costs As was shown later this was not consistent with practice in particular the role of agents was not taken into account The development of management theory led to an understanding of the role that managerial factors play in firm performance The behavioral theory results led to an understanding of the importance of the harmonious development of an organization consideration of the interests of all interested groups and influence of external factors This is especially important for banks because the activity of these organizations is socially important it is based on the trust of clients is regulated and requires careful consideration of stakeholders opinions9 The impact of management and environmental factors can be accounted for in two ways by adding determinants to the function Aiello and Bonanno 2013 by estimating J Risk Financial Manag 2023 16 243 9 of 20 the efficiency derived from the function using the production parameters and then esti mating the impact of management determinants and environmental factors on efficiency Dong et al 2016 Based on our data we adopted the first approach adding management parameters Z and environmental factors E to the function The effect of environmental variables eg GDP inflation rate on inefficiency were modeled together with a profit or cost frontier model The profit and cost frontier models are represented below Profitit Fpy w β Z Eevpeup 3 Costit Fcy w β Z Eevceuc 4 where y is the product service produced w is the price β is the function parameters Profitit is the profit of object i at time t and Costit is the total cost of object i at time t Z denotes the control parameters vp and vc represent random components in the statistical error of Equations 3 and 4 up and uc represent nonrandom components in the statistical error of Equations 3 and 4 associated with inefficiency E denotes environmental parameters The SFA method distinguishes between the nonrandom u and random v components of the residuals The random component v has a normal distribution and the component u has a seminormal distribution with both quantities being independent There are two different versions of the concept of profit efficiency namely standard profit efficiency SPE and alternative profit efficiency APE Berger and Mester 1997 Standard profit efficiency indicates how close a bank is to maximizing profits at a specified input and output price It assumes that output markets are perfectly competitive Alterna tive profit efficiency assumes that there is imperfect competition and therefore banks have some market power APE measures how close a bank is to earning maximum profit by adjusting output prices and input quantities given a specific level of output and a specific input price Due to the absence of reliable data on banks output prices and the existence of imperfect competition in the Chinese banking sector we used alternative profit efficiency APE analysis in our study From Equations 3 and 4 profit efficiency is the ratio of actual observed bank profits to the maximum level of potential efficient bank profits Cost efficiency is the ratio of the minimum level of potentially efficient banks costs to the actual observed level of banks costs Profit efficiency and cost efficiency have a value between 0 and 1 Profit efficiency APE Fpy w β Z Eevp Fpy w β Z Eevpeup eup 5 Cost efficiency Fcy w β Z Eevc Fpy w β Z Eevcec euc 6 42 Variable Definitions There is no single approach to deciding which parameters should be used in the profitability function and the cost function Ariff and Luc 2008 Foroutan et al 2015 Gaganis et al 2020 Aiello and Bonanno 2013 Ding and Sickles 2018 Almanidis et al 2019 Le et al 2020 The researcher agrees that the set of these parameters can be the same As a rule there are basic parameters that characterize the technology control parameters and parameters of the environment external environment In the literature on bank efficiency there are many ways to measure the inputs and outputs of banks the production approach the intermediation approach and the operating approach Coelli et al 2005 Our study uses the intermediation approach The intermediation approach proposed by Sealey and Lindley 1977 views the bank as an intermediary that collects customer deposits and grants loans The bank acts as an intermediary collecting funds from savers and converting these funds into profitgenerating items loans etc The approach is J Risk Financial Manag 2023 16 243 10 of 20 consistent with the concept of the bank as a financial intermediary and the data required to implement the approach are readily available Berger and Humphrey 1997 highlighted the advantages of stochastic frontier methods compared to traditional measures of efficiency eg return on assets because they take into account both relevant bank inputs and outputs and differences in input prices they believe that the intermediary approach is more appropriate for measuring the efficiency of the bank as a whole Profit before tax P is used as a measure of profit Total costs C ie the sum of overheads interest costs and other operating expenses are used as a measure of costs There is no consensus as to which parameters should be used in the profitability function and the cost function The opinion of the researchers is that some basic parameters characterize the production parameters the management parameters and the environmental parameters Basic parameters production parameters The basic parameters outline the process of turning resources into final products In general the profitability and cost functions use loans investments and noninterest income as output parameters The input parameters are labor prices and deposit prices Three groups of earning assets were used loans other earning assets and noninterest income comprising mainly fee and commission income and investment income which is consistent with current practice These variables serve as the output parameters of the model The input parameters input prices the price of loanable funds the price of capital described in the model are as follows Table 1 shows the selected variables outputs and inputs total costs C and total profits P Table 1 Definition of the variables included in the cost and profit functions Variables Name Description Outputs y1 Loans Loans y2 Other earning assets Total earning assets minus loans y3 Noninterest income Other operating income Inputs w1 Price of loanable funds Interest expensetotal assets equity w2 Price of capital Overheadsfixed assets Cost C Total costs Overheads interest expense other operating expenses Profit P Total profits Profit before tax Source authors preparation We believe that the stochastic frontier approach to assessing the profitability of com mercial banks not only assesses the banks profitability relative to the leading banks in the industry but also takes into account the specific circumstances of the management techniques used and the market environment Control parameters Risk management is a fundamental element of commercial banks management Central banks regulate risks in the activities of commercial banks using capital adequacy risk as recommended by BASEL 3 Wanke et al 2022 provided a new perspective on the performance of banks in ASEAN member countries arguing that capital adequacy and asset quality especially credit quality are key determinants of banks financial position and profitability Bank management may expose the organization to high risk to provide higher profitability to investors Bank liquidity and bank profitability interact Nicolae et al 2015 AbuAlkheil et al 2018 found that liquidity risk affects the placing of some assets in a nonrecoverable form and reducing liquidity risk leads to lower profitability Klein and Weill 2022 argued that banks profitability is affected by their asset size which promotes economies of scale and larger banks have access to cheaper resources and more reliable funding VeraGilces et al 2020 Table 2 shows the control parameters Table 2 Definition of control variables Variables Name Description z1 CAR Capital adequacy ratio capitalrisk weighted assets z2 RQA Assets quality ratios loan lossesgross loans z3 LA Liquid assets ratio liquid assetsdeposits and short funding z4 TA Bank size total assets in logs Source authors preparation Environmental parameters Environmental parameters are not controlled by bank management but they affect a banks profitability Le and Ngo 2020 and Bitar et al 2018 argued that GDP growth leads to increased demand for production financing and affects the profitability of banks VeraGilces et al 2020 and Nicolae et al 2015 argued that inflation and deregulation affect bank profitability Davis et al 2022 argued that market competition affects bank profitability Dong et al 2016 used time trends to reflect the technological change Table 3 shows the environmental parameters used in the analysis of the banking system Table 3 Definition of environmental variables Variables Name Description e1 GDP GDP in logs e2 INF Inflation rate e3 SMCapitalizationGDP Stock market capitalizationGDP e4 T Year2011 Source authors preparation 43 Model Specification We used the transcendental logarithm translog from Christensen et al 1973 which is the most commonly used functional form in terms of bank efficiency Kumbhakar and Lovell 2000 Our profit efficiency frontier model is as follows lnP β0 i13 βi ln yi γ1 ln W 12 i j13 φij ln yi ln yj 13 μ11 ln W ln W i13 αi ln yi n14 θn zn ε14 εn en u v 7 where y1 y2 y3 are outputs w1 w2 are the price of inputs βi γ1 φij μ11 αi θn εn are the parameters to be estimated Using a translog linear homogeneity also requires standard symmetry φij φji u is the inefficiency of resource use v is random error lnP is the logarithm of total profits The total profits P and input price terms are normalized by one of the input prices w2 in order to impose a linear homogeneity of degree one on the input prices W w1w2 When using the SFA model to analyze profitability the observations in the panel data should not contain negative values To ensure this the profit values in the sample were adjusted based on the results of the study Aielloa and Bonanno 2016 Profit10 figures were transformed by adding the absolute value of the minimum profit Profitmin for each bank plus one After the transformation the right side of Equation 7 looks as follows lnP ln Profit Profitmin 1 w2 8 By substituting profit for cost11 in the formula we can obtain a cost function formula 44 Data Our sample covers 239 Chinese banks from 2012 to 2020 with annual data of 1719 observations in total All banks with complete data for at least eight years were included J Risk Financial Manag 2023 16 243 12 of 20 in the analysis We checked our data against the theoretical requirements and ensured that they were correct incorrect data were removed The sample included 6 stateowned commercial banks 12 national jointstock commercial banks 61 rural commercial banks 130 urban commercial banks and 30 foreign commercial banks see Table 4 The data for China for 20122020 were representative covering more than 95 of the banking sector by asset size The data were mainly taken from the ORBIS international database All financial variables were measured in thousands of US dollars Data on GDP in flation and stock market capitalization were taken from the Global Economy website httpswwwtheglobaleconomycomChinaeconomicgrowth accessed on 28 Decem ber 2022 Table 4 Data characteristics by type of bank Types Number of Banks Abbreviation Average Assets as of 2020 USD Billion State commercial banks 6 SO 3 447 National jointstock commercial banks 12 NJ 759 Foreign commercial banks 30 FB 17 Rural commercial banks 61 RB 34 City commercial banks 130 CB 60 Total 239 Source authors calculations The sample included all stateowned commercial banks national jointstock commer cial banks and most foreign and urban banks Rural commercial banks were represented by the most characteristic banks The sample included banks whose total asset value was more than 95 of the total asset value of the banking system PRC banks vary greatly in size The average size of assets of systemic banks is 3447 billion US dollars many of them top the list of the largest banks in the world Jointstock banks are much smaller but are nevertheless large and play an important role in the financial market Foreign commercial banks are small in size they are not involved in the transformation of assets but mainly focus on service functions The main focus of our study was to analyze the efficiency of the profitability and costs of Chinas commercial banks from the perspective of a regulator The regulator is interested in the number of banks that are close to the production possibilities frontier and banks that are far from this frontier It is important that as many banks as possible can generate profits at the level of the best technologies and management practices Outsiders are the object of analysis of the determinants that affect their weak position The regulator is interested in the dynamics of profitability and cost efficiency which can help to reveal negative or positive trends as well as the reaction of banks to reforms and external influences We did not consider issues related to the fulfillment of the theoretical requirements for the cost function and profitability function in detail We checked our data for compliance with theoretical requirements and made sure they were correct Outliers and incorrect data were removed however this did not significantly affect the sample size 5 Results Constructed regression equations showed high significance for the majority of de terminants see Table 5 which allows us to conclude that we correctly determined the input and output production parameters as well as the control parameters and external parameters12 J Risk Financial Manag 2023 16 243 13 of 20 Table 5 Parameter estimates for stochastic frontier profit function and cost function Variables Parameters Profit Function Cost Function Coefficient Signif Codes Coefficient Signif Codes Constant β0 88906 15309 lny1 β1 2482 0149 lny2 β2 1220 0098 lny3 β3 0143 0017 lnw1w2 γ1 1034 1097 05 lny1ˆ2 ϕ11 0285 0008 lny1 lny2 ϕ12 0409 0006 lny1 lny3 ϕ13 0008 0006 05 lny2ˆ2 ϕ22 0382 0010 lny2 lny3 ϕ23 0000 0005 05 lny3ˆ2 ϕ33 0001 0003 05 lnw1w2ˆ2 µ11 0023 0016 lnw1w2 lny1 α11 0026 0003 lnw1w2 lny2 α21 0043 0002 lnw1w2 ny3 α31 0006 0006 z1 CAR θ1 0001 0002 z2 RQA θ2 0018 0008 z3 LA θ3 0002 0001 z4 logTotalAssets θ4 2955 1268 e1 logGDP ε5 1774 0426 e2 Inflation ε6 0006 0000 e3 Stock marketGDP ε7 0005 0101 e4 T ε8 0120 0002 σ2 σ2v σ2 u σ2 0148 0051 Gamma γ 0651 0740 Mean efficiency 0797 0868 Total number of observations 1719 Notes the Variables column presents the determinants and their combinations that make up Equation 7 the Parameters column presents the coefficient notation of Equation 7 and the corresponding variables the profitability functions and cost functions blocks present the coefficient values of the parameters column Coefficient and the significance13 level estimates Significance respectively Signif codes 0 0001 001 005 01 1 Source authors calculations To assess the significance of efficiency factors we used the parameter γ Battese and Coelli 1992 1995 which is the ratio of the change due to inefficiency to the total change in the residuals of the regression model This indicates that the banks deviation from the efficiency frontier is due to inefficiency in the use of available resources γ σ2 u σ2v σ2 u σ2 u σ2 9 Σ2 v and σ2 u represent the random error term v and the nonnegative error term u with a value γ between 0 and 1 Le et al 2020 Aiello and Bonanno 2013 The closer the rate of change γ is to 1 the smaller the effect of the random error term v which means that γ is determined by the nonnegative error term u Conversely the closer the rate of change γ is to 0 the more that the profit function and the cost function fail the test The high value of the estimated γ parameter 065 for the profit function and 074 for the cost function indicates that the vast majority of the residual variation is due to profit inefficiency and cost inefficiency effects and strongly supports the use of the stochastic frontier model rather than the standard ordinary least squares OLS model The average profit efficiency of the banks in the entire sample was 08 indicating that most Chinese banks were close to the maximum possible profitability However the average cost efficiency of banks was 087 This suggests that banks are managing their costs efficiently The difference in average profitability and cost efficiency can be explained J Risk Financial Manag 2023 16 243 14 of 20 by the particularities of the period under study during which interest rates and capital adequacy requirements changed dramatically Deposits which account for the majority of commercial banks resources approximately 70 are classified as riskfree Interest rates on loans which also accounted for a large proportion of operating assets included a risk premium During this period reducing financial risk would have had a greater impact on bank profits than on costs To analyze the dynamics of profit efficiency and cost efficiency graphs were drawn up to show how these indicators change over time for different types of banks This is important because different types of banks have different abilities regarding the use of banking technology different levels of access to resources and different asset allocation capabilities For each group of banks the average values14 of the profit and cost efficiency indicator in a given year were estimated Figure 3 on the left shows the dynamics of average profit efficiency by bank type and on the right the average cost efficiency by bank type As can be seen from the left panel the dynamics of profit efficiency can be divided into two phases The first phase from 2012 to 2016 is characterized by growth and then the profit efficiency of Chinese banks starts to decline In our view this can be explained by the 20152016 stock market crisis in the Chinese financial market and the interest rate liberalization reforms Foreign banks were less affected by this process Before 2014 Chinese banks showed a multidirectional change in terms of cost efficiency Between 2014 and 2016 cost efficiency declined sharply Then as the financial markets stabilized an increase in cost efficiency could be seen in Chinese banks J Risk Financial Manag 2023 16 x FOR PEER REVIEW 14 of 21 closer the rate of change γ is to 0 the more that the profit function and the cost function fail the test The high value of the estimated γ parameter 065 for the profit function and 074 for the cost function indicates that the vast majority of the residual variation is due to profit inefficiency and cost inefficiency effects and strongly supports the use of the stochastic frontier model rather than the standard ordinary least squares OLS model The average profit efficiency of the banks in the entire sample was 08 indicating that most Chinese banks were close to the maximum possible profitability However the average cost efficiency of banks was 087 This suggests that banks are managing their costs efficiently The difference in average profitability and cost efficiency can be explained by the particularities of the period under study during which interest rates and capital adequacy requirements changed dramatically Deposits which account for the majority of commercial banks resources approximately 70 are classified as riskfree Interest rates on loans which also accounted for a large proportion of operating assets included a risk premium During this period reducing financial risk would have had a greater impact on bank profits than on costs To analyze the dynamics of profit efficiency and cost efficiency graphs were drawn up to show how these indicators change over time for different types of banks This is important because different types of banks have different abilities regarding the use of banking technology different levels of access to resources and different asset allocation capabilities For each group of banks the average values14 of the profit and cost efficiency indicator in a given year were estimated Figure 3 on the left shows the dynamics of average profit efficiency by bank type and on the right the average cost efficiency by bank type As can be seen from the left panel the dynamics of profit efficiency can be divided into two phases The first phase from 2012 to 2016 is characterized by growth and then the profit efficiency of Chinese banks starts to decline In our view this can be explained by the 20152016 stock market crisis in the Chinese financial market and the interest rate liberalization reforms Foreign banks were less affected by this process Before 2014 Chinese banks showed a multidirectional change in terms of cost efficiency Between 2014 and 2016 cost efficiency declined sharply Then as the financial markets stabilized an increase in cost efficiency could be seen in Chinese banks a b Figure 3 a Average profit efficiency by bank type b average cost efficiency by bank type Source authors preparation Notes axis of ordinatesaverage efficiency per group of banks axis of abscissayears categoriesgroups of banks as in Table 4 Stateowned commercial banks which had the highest profit efficiency relative to other types of banks during the study period are steadily declining Furthermore as can be seen in Figure 4 there is a clear distinction in size between stateowned commercial banks and other banks which are significantly larger and do not deviate much from the average efficiency values except for the Postal Savings Bank of China the youngest state Figure 3 a Average profit efficiency by bank type b average cost efficiency by bank type Source authors preparation Notes axis of ordinatesaverage efficiency per group of banks axis of abscissayears categoriesgroups of banks as in Table 4 Stateowned commercial banks which had the highest profit efficiency relative to other types of banks during the study period are steadily declining Furthermore as can be seen in Figure 4 there is a clear distinction in size between stateowned commercial banks and other banks which are significantly larger and do not deviate much from the average efficiency values except for the Postal Savings Bank of China the youngest stateowned bank The decline in profit efficiency intensified after 2018 which can be attributed to rising problems in Chinas international investments Belt and Road Initiative both for political reasons such as trade issues between the US and China and the general volatility of global markets The cost efficiency of stateowned commercial banks has been relatively stable but not the best Stateowned commercial banks do not need to improve their cost management and competitive resources They have a high reputation for household and corporate activities and can also rely on national policy support J Risk Financial Manag 2023 16 243 15 of 20 J Risk Financial Manag 2023 16 x FOR PEER REVIEW 15 of 21 owned bank The decline in profit efficiency intensified after 2018 which can be attributed to rising problems in Chinas international investments Belt and Road Initiative both for political reasons such as trade issues between the US and China and the general volatility of global markets The cost efficiency of stateowned commercial banks has been relatively stable but not the best Stateowned commercial banks do not need to improve their cost management and competitive resources They have a high reputation for household and corporate activities and can also rely on national policy support Figure 4 Profit efficiency of different types of banks Source authors preparation Notes ordinate axisthe average value of efficiency for the group of banks abscissa axisyears categoriesgroups of banks according to Table 4 where the size of the marker characterizes the size of bank assets and its representation corresponds to the scale Total Assets at the bottom of the figure National jointstock commercial banks showed the lowest level of profit efficiency and negative dynamics for 20122020 overall the decline has intensified since 2016 It is difficult for jointstock banks which cannot count on serious state support to adapt to changes in a short time moreover they are forced to compete for markets with both systemic and city banks as well as with shadow banking However these banks have the highest cost efficiency It can be concluded that national jointstock commercial banks are able to effectively manage their business but due to drastic changes in financial markets and a slowdown in economic growth they cannot fully control profitability Figure 5 shows that this trend is not typical for all national jointstock commercial banks the general trend was affected by the decrease in the efficiency of some banks It can be seen that over time the variability in both the profit and cost efficiency of national jointstock commercial banks increases relative to the average Figure 5 Cost efficiency of different types of banks Source authors preparation Notes ordinate axisthe average value of efficiency for the group of banks abscissa axisyears categoriesgroups of banks according to Table 4 where the size of the marker characterizes the size of bank assets and its representation corresponds to the scale Total Assets at the bottom of the figure Figure 4 Profit efficiency of different types of banks Source authors preparation Notes ordinate axisthe average value of efficiency for the group of banks abscissa axisyears categoriesgroups of banks according to Table 4 where the size of the marker characterizes the size of bank assets and its representation corresponds to the scale Total Assets at the bottom of the figure National jointstock commercial banks showed the lowest level of profit efficiency and negative dynamics for 20122020 overall the decline has intensified since 2016 It is difficult for jointstock banks which cannot count on serious state support to adapt to changes in a short time moreover they are forced to compete for markets with both systemic and city banks as well as with shadow banking However these banks have the highest cost efficiency It can be concluded that national jointstock commercial banks are able to effectively manage their business but due to drastic changes in financial markets and a slowdown in economic growth they cannot fully control profitability Figure 5 shows that this trend is not typical for all national jointstock commercial banks the general trend was affected by the decrease in the efficiency of some banks It can be seen that over time the variability in both the profit and cost efficiency of national jointstock commercial banks increases relative to the average J Risk Financial Manag 2023 16 x FOR PEER REVIEW 15 of 21 owned bank The decline in profit efficiency intensified after 2018 which can be attributed to rising problems in Chinas international investments Belt and Road Initiative both for political reasons such as trade issues between the US and China and the general volatility of global markets The cost efficiency of stateowned commercial banks has been relatively stable but not the best Stateowned commercial banks do not need to improve their cost management and competitive resources They have a high reputation for household and corporate activities and can also rely on national policy support Figure 4 Profit efficiency of different types of banks Source authors preparation Notes ordinate axisthe average value of efficiency for the group of banks abscissa axisyears categoriesgroups of banks according to Table 4 where the size of the marker characterizes the size of bank assets and its representation corresponds to the scale Total Assets at the bottom of the figure National jointstock commercial banks showed the lowest level of profit efficiency and negative dynamics for 20122020 overall the decline has intensified since 2016 It is difficult for jointstock banks which cannot count on serious state support to adapt to changes in a short time moreover they are forced to compete for markets with both systemic and city banks as well as with shadow banking However these banks have the highest cost efficiency It can be concluded that national jointstock commercial banks are able to effectively manage their business but due to drastic changes in financial markets and a slowdown in economic growth they cannot fully control profitability Figure 5 shows that this trend is not typical for all national jointstock commercial banks the general trend was affected by the decrease in the efficiency of some banks It can be seen that over time the variability in both the profit and cost efficiency of national jointstock commercial banks increases relative to the average Figure 5 Cost efficiency of different types of banks Source authors preparation Notes ordinate axisthe average value of efficiency for the group of banks abscissa axisyears categoriesgroups of banks according to Table 4 where the size of the marker characterizes the size of bank assets and its representation corresponds to the scale Total Assets at the bottom of the figure Figure 5 Cost efficiency of different types of banks Source authors preparation Notes ordinate axisthe average value of efficiency for the group of banks abscissa axisyears categoriesgroups of banks according to Table 4 where the size of the marker characterizes the size of bank assets and its representation corresponds to the scale Total Assets at the bottom of the figure The profit and cost efficiency of city commercial banks are similar to those of national jointstock commercial banks There is a plausible explanation for this These banks have similar customers and provide similar services However city commercial banks and national jointstock commercial banks have different custom sizes City banks work with small companies and individuals in smaller cities where there is less competition from jointstock banks and systemic banks The volatility of profit efficiency increases for city commercial banks Figure 4 while the volatility of cost efficiency decreases Figure 5 The profit and cost efficiency of agricultural commercial banks remained high and stable In our opinion the reason for this stability is related to the fact that the customers of J Risk Financial Manag 2023 16 243 16 of 20 these banks are linked to the domestic market and do not depend on external factors Rural commercial banks use rather conservative products such as loans and deposits and less often securities depending on market factors Rural development schemes are subsidized by the state and this provides a stable income for rural commercial banks Figures 4 and 5 show that the assets of rural banks significantly increased over the observation period with little change in the volatility of profitability or cost performance The efficiency of foreign commercial banks was stable and has been increasing since 2015 This can be explained by the fact that their core business is not tied to Chinas lending rates they do not need to adapt and the demand for their services from foreign companies is growing As can be seen from the data in Figures 4 and 5 for foreign banks the variability in both profit efficiency and cost efficiency significantly increased This indicates that foreign banks show different behavioral strategies in the Chinese market and have different competencies related to the characteristics of their customers Thus different types of Chinese banks depending on their characteristics show different behaviors as well as common variations Profitability declined for both external economic and political reasons as well as internal factors In terms of the operation of the market mechanism and the development of competition increased efficiency should be linked to both higher profitability and lower costs Sound regulatory policies should work in tandem with market mechanisms Therefore if we want to assess the quality of banks and the quality of regulatory action in terms of profitability it is necessary to examine the link between efficiency and related indicators eg ROA Figure 6 shows the relationship between return on assets ROA and profit efficiency performance There is a strong positive correlation between return on assets ROA and profit efficiency for national jointstock commercial banks National jointstock commercial banks were established as an important part of Chinas marketbased banking system to measure the role of marketbased approaches in financial reform and banking sector reform Stateowned commercial banks also show high correlation between bank return on assets ROA and profit efficiency However rural commercial banks foreign commercial banks and city commercial banks show no such relationship It can be argued that foreign banks concentrate on service functions such as foreign and domestic trade settlement and do not pursue the goal of increasing profitability Rural and city banks are quite heterogeneous and numerous and operate in distinctive niches cities small towns and rural settlements The profitability of these banks depends on a variety of unique reasons including innovation in banking financial products and government support These banks have a short history and are undergoing a development process the results will manifest in the future J Risk Financial Manag 2023 16 x FOR PEER REVIEW 17 of 21 Figure 6 The relationship between profit efficiency and return on assets ROA for different types of banks Source authors preparation Notes ordinate axisaverage value of return on assets for the group of banks abscissa axisvalue of efficiency categoriesgroups of banks according to Table 4 each point characterizes the position of a particular bank in the coordinate system used while the blue line determines the statistical relationship of data 6 Conclusions The Chinese banking system includes different types of banks with different resource acquisitions asset allocations and technological capabilities This specificity must be taken into account when analyzing profitability The studied period is characterized by instability due to the slowing economic growth financial market instability ongoing banking reforms and reorientation of the economy toward capital exports It is important to understand how the banking system functioned during this period and whether regulatory intervention was necessary The findings suggest that Chinese commercial banks are adapting to the changes brought about by external changes and banking reforms Liu 2017 Li and Liu 2019 and Mott 2022 have shown that interest rate policy affects the profitability of banks When characterizing the dynamics of profitability and cost efficiency by type of bank the following points can be emphasized Stateowned commercial banks lead in terms of profit efficiency cost efficiency is also high After 2015 national jointstock commercial banks had to compete for the market with systemic municipal and shadow banks This harmed their profitability It can be concluded that national jointstock commercial banks were able to manage their business efficiently but could not fully control their profitability due to dramatic changes in financial markets and the slowdown in economic growth The profitability and economic efficiency of city banks are similar to those of joint stock banks City banks work with small companies and individuals in smaller cities where there is less competition from jointstock and stateowed commercial banks Chinas cities have a large concentration of manufacturing logistics and financial institutions all of which are increasing City residents are financially literate and responsive to new banking products This has forced city banks to improve their customer service techniques which has affected their efficiency The profitability and economic efficiency of city commercial banks are similar to that of jointstock banks Urban commercial banks work with small companies and individuals in smaller cities where there is less competition from jointstock and system banks Foreign banks performance has remained stable and has been increasing since 2015 This can be explained by the fact that their core business is not tied to Chinese lending rates they do not need to adapt to Chinese reforms and the demand for services from foreign companies is increasing The profit efficiency and cost efficiency of foreign commercial banks increased between 2012 and 2020 Figure 6 The relationship between profit efficiency and return on assets ROA for different types of banks Source authors preparation Notes ordinate axisaverage value of return on assets for the group of banks abscissa axisvalue of efficiency categoriesgroups of banks according to Table 4 each point characterizes the position of a particular bank in the coordinate system used while the blue line determines the statistical relationship of data J Risk Financial Manag 2023 16 243 17 of 20 6 Conclusions The Chinese banking system includes different types of banks with different resource acquisitions asset allocations and technological capabilities This specificity must be taken into account when analyzing profitability The studied period is characterized by instability due to the slowing economic growth financial market instability ongoing banking reforms and reorientation of the economy toward capital exports It is important to understand how the banking system functioned during this period and whether regulatory intervention was necessary The findings suggest that Chinese commercial banks are adapting to the changes brought about by external changes and banking reforms Liu 2017 Li and Liu 2019 and Mott 2022 have shown that interest rate policy affects the profitability of banks When characterizing the dynamics of profitability and cost efficiency by type of bank the following points can be emphasized Stateowned commercial banks lead in terms of profit efficiency cost efficiency is also high After 2015 national jointstock commercial banks had to compete for the market with systemic municipal and shadow banks This harmed their profitability It can be concluded that national jointstock commercial banks were able to manage their business efficiently but could not fully control their profitability due to dramatic changes in financial markets and the slowdown in economic growth The profitability and economic efficiency of city banks are similar to those of jointstock banks City banks work with small companies and individuals in smaller cities where there is less competition from jointstock and stateowed commercial banks Chinas cities have a large concentration of manufacturing logistics and financial institutions all of which are increasing City residents are financially literate and responsive to new banking products This has forced city banks to improve their customer service techniques which has affected their efficiency The profitability and economic efficiency of city commercial banks are similar to that of jointstock banks Urban commercial banks work with small companies and individuals in smaller cities where there is less competition from jointstock and system banks Foreign banks performance has remained stable and has been increasing since 2015 This can be explained by the fact that their core business is not tied to Chinese lending rates they do not need to adapt to Chinese reforms and the demand for services from foreign companies is increasing The profit efficiency and cost efficiency of foreign commercial banks increased between 2012 and 2020 Thus different types of Chinese banks depending on their characteristics show different behaviors but follow some common trends The decline in profitability is due to both external economic and political reasons as well as internal factors related to changes in the structure of the stock market and financial sector reforms From the regulators perspective the focus should be on the performance of joint stock banks as well as on the factors of deterioration in the efficiency of different types of banks Given the increasing volatility of profitability and cost efficiency it is necessary to understand the position of outsiders and the reasons for their declining efficiency Author Contributions Conceptualization AAP and LPB methodology AAP software BS and AAP validation BS and AAP formal analysis BS and AAP investigation BS resources BS data curation BS writingoriginal draft preparation BS writingreview and editing BS and AAP visualization BS and LPB supervision XL funding acquisition BS All authors have read and agreed to the published version of the manuscript Funding This research was funded by China Scholarship Council grant number 202008230139 The APC was paid by China Scholarship Council grant number 202008230139 Data Availability Statement Data sources are contained within the article Data are freely available and can be accessed by everyone J Risk Financial Manag 2023 16 243 18 of 20 Conflicts of Interest The authors declare no conflict of interest Notes 1 Commercial Banking Law of the Peoples Republic of China httpwwwgovcntest20050628content10576htm accessed on 4 January 2023 2 Law of the Peoples Republic of China on the Peoples Bank of China httpwwwgovcntest20050628content10577htm accessed on 4 January 2023 3 Belt and Road Initiative httpswwwyidaiyilugovcn accessed on 4 January 2023 4 Bank of China China Construction Bank Industrial and Commercial Bank China Agricultural Bank and China Communications Bank PostalSavings Bank of China 5 Agriculture Development Bank of China China Development Bank The ExportImport Bank of China 6 There are 12 national jointstock commercial banks in China China Merchants Bank Pudong Development Bank CITIC Bank China Everbright Bank Huaxia Bank China Minsheng Bank Guangfa Bank Industrial Bank Ping An Bank Zhejiang Commercial Bank Hengfeng Bank и Bohai Bank 7 In 2015 the collapse of the Shanghai and Shenzhen stock exchanges evaporated about 5 trillion USD in market value Investors suffered heavy losses each losing more than 15000 USD on average Shi et al 2022 In 2015 the stock market problems pushed the Peoples Bank of China to raise its deposit and lending rates Rathnayake et al 2022 8 β Vector dimension number of resources used 1 9 Berger and Mester 1997 showed that it is necessary to take into account the methods of attracting and allocating resources the levels of interest rates and the risks taken In addition for banks of different sizes and specializations it is necessary to take into account the available opportunities for creating banking products A number of researchers believe that it is important for banks to take into account not only internal but also external factors such as market influences and banking regulation Bhattacharyya et al 1997 This influence extends to all banks and will lead to a parallel change in the border production function 10 According to Table 1 Profit Total profits Profit before tax 11 According to Table 1 Cost Total costs overheads interest expense other operating expense 12 The significance of individual factors characterizes the quality of the created regression model which can explain the estimated parameter This does not allow for us to draw unambiguous conclusions about the real influence of individual factors on the estimated parameter but characterizes only the statistical relationship Conclusions about the real influence of individual factors require additional reasoning and analysis 13 The criteria for determining significance are given in the last row of Table 5 14 The arithmetic mean References AbuAlkheil Ahmad Ghadeer Khartabiel and Nuradli Ridzwan Shah Mohd Dali 2018 A TwoStage Parametric Stochastic Frontier Analysis SFA of the Efficiency Performance of Shariah Compliant Banks A Global Evidence American Journal of Finance and Accounting 5 85110 CrossRef Aiello Francesco and Graziella Bonanno 2013 Profit and Cost Efficiency in the Italian Banking Industry 20062011 Economics and Business Letters 2 190205 CrossRef Aielloa Francesco and Graziella Bonanno 2016 Bank Efficiency and Local Market Conditions Evidence from Italy Journal of Economics and Business 83 7090 CrossRef Aigner Dennis C A Knox Lovell and Peter Schmidt 1997 Formulation and Estimation of Stochastic Frontier Production Function Models Journal of Econometrics 6 2137 CrossRef Almanidis Pavlos Mustafa Karakaplan and Levent Kutlu 2019 A Dynamic Stochastic Frontier Model with Threshold Effects US Bank Size and Efficiency Journal of Productivity Analysis 52 116 CrossRef Ariff Mohamed and Can Luc 2008 Cost and Profit Efficiency of Chinese Banks A NonParametric Analysis China Economic Review 19 26073 CrossRef Athanasoglou Panayiotis P Sophocles N Brissimis and Matthaios D Delis 2008 BankSpecific IndustrySpecific and Macroeconomic Determinants of Bank Profitability Journal of International Financial Markets Institutions and Money 18 12136 CrossRef Avkiran Necmi K 2011 Association of DEA SuperEfficiency Estimates with Financial Ratios Investigating the Case for Chinese Banks Omega 39 32334 CrossRef Battese George Edward and Tim J Coelli 1992 Frontier Production Functions Technical Efficiency and Panel Data With Application to Paddy Farmers in India Journal of Productivity Analysis 3 15369 Available online httpswwwjstororgstable41770578 accessed on 12 April 2023 CrossRef Battese George Edward and Tim J Coelli 1995 A Model for Technical Inefficiency Effects in a Stochastic Frontier Production Function for Panel Data Empirical Economics 20 32532 CrossRef J Risk Financial Manag 2023 16 243 19 of 20 Berger Allen N and David B Humphrey 1997 Efficiency of Financial Institutions International Survey and Directions for Future Research European Journal of Operational Research 98 175212 CrossRef Berger Allen N and Loretta J Mester 1997 Inside the Black Box What Explains Differences in the Efficiencies of Financial Institutions Board of Governors of the Federal Reserve System Working Paper 21 1059 CrossRef Bhattacharyya Arunava C A Knox Lovell and Pankaj Sahay 1997 The Impact of Liberalization on the Productive Efficiency of Indian Commercial Banks European Journal of Operational Research 98 33245 CrossRef Bitar Mohammad Kuntara Pukthuanthong and Thomas Walker 2018 The Effect of Capital Ratios on the Risk Efficiency and Profitability of Banks Evidence from OECD Countries Journal of International Financial Markets Institutions and Money 53 22762 CrossRef Capraru Bogdan and Iulian Ihnatov 2014 Banks Profitability in Selected Central and Eastern European Countries Procedia Economics and Finance 16 58791 CrossRef Charnes Abraham William W Cooper and Edwardo Rhodes 1978 Measuring the Efficiency of Decision Making Units European Journal of Operational Research 2 42944 CrossRef Christensen Laurits Dale Jorgenson and Lawrence Lau 1973 Transcendental Logarithmic Production Frontiers The Review of Economics and Statistics 55 2845 CrossRef Coelli Timothy J D S Prasada Rao Christopher J ODonnell and George E Battese 2005 An Introduction to Efficiency and Productivity Analysis 2nd ed New York Springer CrossRef Davis Philip Dilruba Karim and Dennison Noel 2022 The Effects of Macroprudential Policy on Banks Profitability International Review of Financial Analysis 80 127 CrossRef de Guindos Luis 2019 Challenges for Bank Profitability OMFIF City Lecture Available online httpswwwecbeuropaeupress keydate2019htmlecbsp1905017733ecc1a9enhtml accessed on 28 December 2022 Ding Dong and Robin Sickles 2018 Frontier Efficiency Capital Structure and Portfolio Risk An Empirical Analysis of US Banks Business Research Quarterly 21 715 CrossRef Djalilova Khurshid and Jenifer Piesse 2016 Determinants of Bank Profitability in Transition Countries What Matters Most Research in International Business and Finance 38 6982 CrossRef Dong Yizhe Michael Firth Wenxuan Hou and Weiwei Yang 2016 Evaluating the Performance of Chinese Commercial Banks A Comparative Analysis of Different Types of Banks European Journal of Operational Research 252 28095 CrossRef Foroutan Farnaz Homayoun Ranjbar and Mehdi HampaeyanMiandoab 2015 Estimation of Technical Efficiency in Banking Industry Using Stochastic Frontier Analysis International Journal of Productivity and Quality Management 16 5469 CrossRef Fu Xiaoqing and Shelagh Heffernan 2007 Cost XEfficiency in Chinas Banking Sector China Economic Review 18 3553 CrossRef Gaganis Chrysovalantis Emilios Galariotis Fotios Pasiouras and Christos Staikouras 2020 Bank Profit Efficiency and Financial Consumer Protection Policies Journal of Business Research 118 98116 CrossRef Henriques Iago Cotrim Vinicius Amorim Sobreiro Herbert Kimura and Enzo Barberio Mariano 2020 TwoStage DEA in Banks Terminological Controversies and Future Directions Expert Systems with Applications 161 131 CrossRef Horrocks Michael Venkataramanan Ramaswamy and Keri Rupp 2008 Financial Reforms in Chinese Banking The Impact on Personal Lending and Operational Efficiency Business Horizons 51 51117 CrossRef Huang Qiubin 2020 Ownership Concentration and Bank Profitability in China Economics Letters 196 13 CrossRef Huang Wen Ying Wu and Li Deng 2021 Does Banking Competition Stimulate Regional Innovation Evidence from China Pacific Basin Finance Journal 70 115 CrossRef Klein PaulOlivier and Laurent Weill 2022 Bank Profitability and Economic Growth Quarterly Review of Economics and Finance 84 18399 CrossRef Kumbhakar Subal C and C A Knox Lovell 2000 Stochastic Frontier Analysis Cambridge Cambridge University Press pp 279307 CrossRef Le Minh VietNgu Hoang Clevo Wilson and Shunsuke Managi 2020 Net Stable Funding Ratio and Profit Efficiency of Commercial Banks in the US Economic Analysis and Policy 67 5566 CrossRef Le Tu and Thanh Ngo 2020 The Determinants of Bank Profitability A CrossCountry Analysis Central Bank Review 20 6573 CrossRef Lee Chien Chiang and Meng Fen Hsieh 2013 The Impact of Bank Capital on Profitability and Risk in Asian Banking Journal of International Money and Finance 32 25181 CrossRef Lee Chien Chiang Chih Wei Wang and Shan Ju Ho 2020 Financial Innovation and Bank Growth The Role of Institutional Environments North American Journal of Economics and Finance 53 123 CrossRef Li Fangmin Tianle Yang Min Du and Miao Huang 2023 The Development Fit Index of Digital Currency Electronic Payment between China and the One Belt One Road Countries Research in International Business and Finance 64 114 CrossRef Li Jingya and MingHua Liu 2019 Interest Rate Liberalization and Passthrough of Monetary Policy Rate to Bank Lending Rates in China Frontiers of Business Research in China 13 219 CrossRef Li XiaoLin DengKui Si and Xinyu Ge 2021 Chinas Interest Rate Passthrough after the Interest Rate Liberalization Evidence from a Nonlinear Autoregressive Distributed Lag Model International Review of Economics and Finance 73 25774 CrossRef Li Yang 2020 Analyzing Efficiencies of City Commercial Banks in China An Application of the Bootstrapped DEA Approach Pacific Basin Finance Journal 62 112 CrossRef J Risk Financial Manag 2023 16 243 20 of 20 Lin Xiaochi and Yi Zhang 2009 Bank Ownership Reform and Bank Performance in China Journal of Banking and Finance 33 2029 CrossRef Liu Kerry 2017 Chinas Interest Rate Passthrough to Commercial Banks Before and After Interest Rate Liberalisation Economic Affairs 37 27987 CrossRef Mai Vo Phuong Kent Matthews David Meenagh Patrick Minford and Zhiguo Xiao 2022 Regulatory Arbitrage Shadow Banking and Monetary Policy in China Journal of International Financial Markets Institutions and Money 80 113 CrossRef Mott Carey 2022 China Reserve Requirements 20152016 Journal of Financial Crisis 4 393428 Available online httpselischolar libraryyaleedujournaloffinancialcrisesvol4iss419 accessed on 12 April 2023 Ngan Le Thi Thanh 2014 Profit and Cost Efficiency Analysis in Banking Sector A Case of Stochastic Frontier Approach for Vietnam Journal of Knowledge Management Economics and Information Technology IV 117 Available online wwwscientificpapersorg accessed on 28 December 2022 Nicolae Petria Bogdan Capraru and Iulian Ihnatov 2015 Determinants of Banks Profitability Evidence from EU 27 Banking Systems Procedia Economics and Finance 20 51824 CrossRef Pasiouras Fotios Sailesh Tanna and Constantin Zopounidis 2009 The Impact of Banking Regulations on Banks Cost and Profit Efficiency CrossCountry Evidence International Review of Financial Analysis 18 294302 CrossRef Rathnayake Dilesha Yang Bai Pierre Louembé and Li Qi 2022 Interest Rate Liberalization and Commercial Bank Performance New Evidence From Chinese AShare Banks SAGE Open 12 112 CrossRef Saona Paolo 2016 Intra and ExtraBank Determinants of Latin American Banks Profitability International Review of Economics and Finance 45 197214 CrossRef Scotta Susan V John Van Reenenb and Markos Zachariadis 2017 The LongTerm Effect of Digital Innovation on Bank Performance An Empirical Study of SWIFT Adoption in Financial Services Research Policy 46 9841004 CrossRef Sealey Calvin W and James T Lindley 1977 Inputs Outputs and a Theory of Production and Cost at Depository Financial Institutions Journal of Finance 32 125166 CrossRef Shi Yang Shu Chen and Ruiming Liu 2022 Fund Renaming and Fund Flows Evidence from Chinas Stock Market Crash in 2015 Economic Modelling 108 113 CrossRef Silva Thiago Christiano Benjamin Miranda Tabak Daniel Oliveira Cajueiro and Marina Villas Boas Diasd 2017 A Comparison of DEA and SFA Using Micro and MacroLevel Perspectives Efficiency of Chinese Local Banks Physica A Statistical Mechanics and Its Applications 469 21623 CrossRef Sun Rongrong 2020 Monetary Policy Announcements and Market Interest Rates Response Evidence from China Journal of Banking and Finance 113 105766 CrossRef Tan Yong 2017 The Impacts of Competition and Shadow Banking on Profitability Evidence from the Chinese Banking Industry North American Journal of Economics and Finance 42 89106 CrossRef VeraGilces Paul Segundo CaminoMogro Xavier OrdẽnanaRodríguez and Gino CornejoMarcos 2020 A Look inside Banking Profitability Evidence from a Dollarized Emerging Country Quarterly Review of Economics and Finance 75 14766 CrossRef Wang Ke Wei Huang Jie Wu and Yingnan Liu 2014 Efficiency Measures of the Chinese Commercial Banking System Using an Additive TwoStage DEA Omega 44 520 CrossRef Wanke Peter Amir Yazdi and Ramona Birau 2022 Revisiting CAMELS Rating System and the Performance of ASEAN Banks A Comprehensive MCDMZNumbers Approach IEEE Access 10 54098109 CrossRef Yang Shanran Benye Shi and Fujia Yang 2023 Research in International Business and Finance Macroeconomic Impact of the SinoU S Trade Frictions Based on a TwoCountry TwoSector DSGE Model Research in International Business and Finance 65 117 CrossRef Zhao Jinsong Xinghao Li ChinHsien Shi Chen and ChiChuan Lee 2022 Riding the FinTech Innovation Wave FinTech Patents and Bank Performance Journal of International Money and Finance 122 102552 CrossRef DisclaimerPublishers Note The statements opinions and data contained in all publications are solely those of the individual authors and contributors and not of MDPI andor the editors MDPI andor the editors disclaim responsibility for any injury to people or property resulting from any ideas methods instructions or products referred to in the content Este artigo está licenciado com uma Licença Creative Commons Atribuição Não Comercial 40 Internacional EFICIÊNCIA BANCÁRIA NO BRASIL 20102019 UMA ANÁLISE DE DOIS ESTÁGIOS Fábio Lucas Takahashi Graduado em Ciências Econômicas pela Universidade Estadual de Maringá UEM e mestrando em Economia da mesma instituição Email flucastakahashioutlookcom httpsorcidorg0000000171323876 Marcos Roberto Vasconcelos Graduado em Ciências Econômicas pela Universidade Federal do Paraná UFPR e doutor em Ciência Econômica pela Universidade Estadual de Campinas Unicamp Professor do Departamento de Economia DCO da Universidade Estadual de Maringá UEM Email mrvasconcelosuembr httpsorcidorg0000000314764899 Como citar este artigo Takahashi F L Vasconcelos M R 2022 Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Revista de Economia Mackenzie 192 165187 doi10593518082785remv19n2p165187 Recebido em 14042022 Aprovado em 16082022 166 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 Resumo O objetivo deste trabalho é analisar o desempenho do setor bancário brasileiro e seus determinantes A presente pesquisa foi dividida em duas partes Na primei ra buscase determinar a eficiência técnica pelo modelo DEABCC nos principais bancos brasileiros no período de 2010 a 2019 Na segunda parte são avaliados os determinantes que podem influenciar o desempenho dos bancos As estimati vas indicam que os grandes bancos são em média mais eficientes do que os bancos de pequeno e médio portes Além disso quando se observa o tipo de propriedade os bancos estrangeiros tendem a apresentar melhor desempenho em comparação com os bancos nacionais Ademais os determinantes apontam que a margem de lucro líquido e a renda sem juros são significantes e contribuem posi tivamente para o nível de eficiência Por sua vez o índice de adequação de capi tal os empréstimos inadimplentes e a inflação têm impacto negativo sobre a efi ciência técnica Palavraschave Análise Envoltória de Dados determinantes eficiência técnica instituições financeiras setor bancário Classificação JEL B23 B26 C24 G20 INTRODUÇÃO O aumento da incidência de crises bancárias e financeiras nas últimas dé cadas desencadeou uma agenda de pesquisas sobre o comportamento do setor bancário não apenas sobre as causas subjacentes das crises mas também so bre o impacto na economia real Kose et al 2010 Ahmad et al 2019 Nesse sentido a literatura de eficiência bancária tem recebido grande atenção prin cipalmente pelo impacto das reformas regulatórias da competição bancária e das novas tecnologias de informação que foram inseridas no sistema bancário Diallo 2018 A importância do setor bancário baseiase no fato de que os bancos são os principais canais de poupança e alocação de crédito em muitas economias entre elas a brasileira Tecles Tabak 2010 Isto é o setor bancário desem penha importante função ao converter depósitos em empréstimos e investi mentos Claessens Laeven 2005 Em razão do crescente volume e da com plexidade das transações e dos ativos negociados tornouse cada vez mais necessário avaliar os riscos da atividade bancária Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 167 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 É razoável esperar que um setor bancário eficiente proporcione maior esta bilidade ao sistema financeiro e conduza para o melhor desenvolvimento eco nômico do país Levine 1997 Dessa forma o conhecimento dos fatores que influenciam o desempenho do setor bancário é essencial para que os gestores das instituições financeiras reguladores e supervisores possam formular polí ticas mais apropriadas aos problemas de ineficiência e fragilidade bancária Abreu et al 2019 Parte considerável da eficiência das instituições financeiras depende dire tamente do gerenciamento eficiente da combinação de insumos inputs e pro dutos outputs nas operações bancárias Entretanto os fatores que determi nam a eficiência das instituições financeiras podem estar associados tanto a fatores internos por exemplo gestão ineficiente e fraudes como a fatores externos por exemplo mudanças regulatórias e variáveis macroeconômicas Berger DeYoung 1997 Diante do exposto o presente trabalho examina a eficiência técnica das instituições financeiras do setor bancário brasileiro no período de 2010 a 2019 e analisa seus determinantes Em particular concentrase em explorar as diferenças nas eficiências médias entre os tipos de propriedade público na cional privado nacional e estrangeiro e os diferentes tamanhos de banco pe queno médio e grande portes Para tanto utilizouse a metodologia de Análise Envoltória de Dados Data Envelopment Analysis DEA para estimar o nível de eficiência técnica das instituições financeiras Adicionalmente para avaliar os determinantes da efi ciência técnica foi empregada a regressão Tobit que permite mensurar variá veis com dados censurados ou limitados Nesse aspecto esperase que os re sultados permitam agregar novas evidências sobre o desempenho do setor bancário brasileiro Este trabalho tem implicações importantes Em primeiro lugar adicionamse evidências sobre a eficiência bancária e seus determinantes Em segundo lu gar ampliase a compreensão de como os indicadores de performance bancária e as variáveis macroeconômicas moldam o nível de eficiência das instituições financeiras Vários estudos investigaram a evolução do setor bancário brasilei ro e seus possíveis determinantes Staub et al 2010 Tecles Tabak 2010 Wanke Barros 2014 Embora a literatura de eficiência bancária no Brasil aborde as diferenças de eficiência entre o tipo de propriedade e o tamanho dos bancos notase que há poucas evidências do impacto dos empréstimos inadimplentes no desempe nho bancário Os resultados deste trabalho sugerem que os bancos brasileiros 168 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 melhoram seus desempenhos ao aumentarem as margens de lucros das recei tas provindas de juros e sem juros Entretanto podese observar que a menor eficiência do setor bancário brasileiro está associada à presença de maior capi talização inflação e empréstimos inadimplentes O artigo prossegue da seguinte forma a seção 1 apresenta o referencial teó rico a seção 2 contém a revisão de literatura de eficiência bancária a seção 3 exibe os dados utilizados neste trabalho a seção 4 expõe os métodos de pes quisa a seção 5 relata os resultados e por fim há as considerações finais 1 REFERENCIAL TEÓRICO Nos últimos anos as preocupações com a mensuração de desempenho das instituições financeiras têm adquirido maior atenção dos pesquisadores Ahmad et al 2019 Os motivos são diversos como preocupações com os riscos de falência bancária problemas de estabilidade financeira fatores que influen ciam o desenvolvimento financeiro etc Abreu et al 2019 Ahmad et al 2019 Nesses estudos temse utilizado com frequência a DEA para avaliar a eficiên cia das instituições financeiras A metodologia não paramétrica da DEA utiliza a combinação de insumos inputs e produtos outputs sem precisar necessariamente de suposições so bre sua função de produção Os modelos DEA podem avaliar a eficiência rela tiva de um conjunto de entidades denominadas Unidades Tomadoras de De cisão DecisionMaking Units DMU sendo uma DMU qualquer entidade que deve ser avaliada em termos de sua capacidade de converter múltiplos insumos em múltiplos produtos Cooper et al 2011 Inicialmente o modelo DEA introduzido por Charnes Cooper e Rhodes 1978 baseado principalmente no trabalho de Farrell 1957 proporcio nou a formação do modelo CCR ou Constant Return to Scale CRS Nesse as pecto o modelo CCR CRS assume que as DMU devem operar com retornos constantes de escala isto é se um aumento nos insumos inputs resultar em um aumento proporcional nos níveis do produto outputs Esse modelo men sura a eficiência relativa das DMU e produz resultados consistentes em termos de avaliação global das unidades produtivas Charner et al 1978 Entretanto os modelos DEA do tipo CCR CRS requerem a suposição de total proporcionalidade entre insumos e produtos o que pode não ser apro Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 169 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 priado em alguns contextos Para superar essa dificuldade Banker Charnes e Cooper 1984 desenvolveram o modelo BCC ou Variable Return to Scale VRS que mede as DMU com retornos variáveis de escala que ajuda a estimar se um aumento ou uma diminuição nos insumos e produtos não resulta em mudança proporcional dos insumos e produtos respectivamente Cooper et al 2011 Nesse sentido o modelo BCC pode ser útil para avaliar DMU com diferen tes tamanhos de economias de escala cujas operações são proporcionalmente divergentes entre as DMU Ou seja a escala das operações pode ter impacto nos níveis de eficiência criando economias ou deseconomias de escala Por isso o modelo proposto por Banker et al 1984 com retornos de variáveis de escala poderia apresentar resultados mais seguros A partir do desenvolvimento dos modelos CCR CRS e BCC VRS hou ve um crescimento extraordinário na metodologia DEA Como resultado surgiram trabalhos em quantidade elevada sobre as aplicações da DEA para avaliar a eficiência de unidades produtivas conforme Emrouznejad et al 2008 e Emrouznejad e Yang 2018 Um dos fatores para o elevado volume de trabalhos na literatura sobre DEA é a capacidade de os modelos inovarem tanto na teoria como nos resultados empíricos o que reforça sua aplicabili dade Emrouznejad Yang 2018 Na literatura de eficiência bancária por exemplo observamse diversos trabalhos sobre a eficiência relativa que podem envolver também temas sobre rentabilidade produtividade e custos entre as instituições financeiras Ahmad et al 2019 Abreu et al 2019 Esses estudos permitiram investigar o com portamento das instituições no setor bancário aprimorar a metodologia e pro porcionar evidências sobre as ineficiências dessas instituições Nesse sentido parte da literatura de eficiência bancária utiliza estudos ba seados em duas etapas pois os resultados fornecidos pela DEA podem não ser suficientes para fornecer diagnósticos mais precisos sobre o comportamento do setor bancário Entre esses modelos podese utilizar a regressão Tobit para estimar os efeitos de variáveis explicativas a fim de fornecer evidências sobre os possíveis fatores que influenciam a eficiência das instituições financeiras 2 LITERATURA DE EFICIÊNCIA BANCÁRIA Na literatura de eficiência bancária há estudos que aplicam para os Estados Unidos a Europa e outros países técnicas econométricas e não paramétricas 170 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 com o propósito de examinar a eficiência e a evolução de produtividade dos bancos Claessens et al 2001 Miccoel et al 2007 Nguyen 2018 Embora existam vastos estudos sobre eficiência bancária poucos se referem aos mode los de dois estágios DEA e análise dos determinantes no Brasil Staub et al 2010 Tecles Tabak 2010 Wanke Barros 2014 Como o foco do pre sente trabalho é o sistema bancário brasileiro são discutidos a seguir os estu dos referentes a economias emergentes Por exemplo GarzaGarcía 2012 investiga os determinantes da eficiência bancária no México durante o período de 2001 a 2009 O resultado da DEA indica que o setor bancário mexicano experimentou baixos níveis de eficiên cia técnica 15 eficiência técnica pura 29 e eficiência de escala 14 Além disso os principais determinantes do aumento da eficiência bancária estão associados aos aumentos dos empréstimos ao crescimento econômico variação do produto interno bruto PIB e à participação das instituições estrangeiras Os resultados de Defung et al 2016 fornecem uma análise da eficiência técnica do setor bancário da Indonésia no período de 1993 a 2011 A partir do modelo DEABCC foi possível verificar que as instituições financeiras dos tipos público nacional e estrangeiro por exemplo são consideradas mais efi cientes do que qualquer outro tipo de instituição Na análise dos determinan tes realizada pela regressão Tobit pôdese identificar que o impacto regulatório é geralmente positivo e estatisticamente significativo Para Kutlar et al 2017 que avaliaram os bancos comerciais da Turquia para o período de 2003 a 2012 por meio da metodologia DEA CCR e BCC os resultados observados foram mistos Na análise de eficiência tanto do modelo CCR quanto do BCC houve ao longo do período examinado uma redução do número de bancos tecnicamente eficientes no entanto a eficiência alocati va dos bancos turcos aumentou nesse período Complementarmente no mo delo de regressão Tobit os autores determinaram qual tipo de eficiência téc nica ou alocativa representava melhor o comportamento do produto output De acordo com os resultados as pontuações de eficiência alocativa represen taram melhor a reação do produto output Jayaraman e Srinivasan 2019 examinaram a eficiência de custo receita e lucro dos bancos da Índia de 2004 a 2013 também por meio da abordagem de dois estágios DEA e regressão Tobit Os resultados sugerem que a eficiên cia de custos e lucros dos bancos indianos está correlacionada positivamente e revelam que se os bancos são eficientes em termos de custos também o são em termos de lucro Da mesma forma os autores evidenciaram que a eficiência Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 171 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 de lucro é o tipo de eficiência que diferencia melhor o desempenho entre as instituições analisadas Por fim concluíram que os principais determinantes da eficiência bancária foram o tamanho total de ativos e a variável da razão de despesas operacionais em relação ao total de ativos Por sua vez Staub et al 2010 analisaram as eficiências técnica alocativa e de custos para as instituições financeiras do Brasil de 2000 a 2007 com a metodologia DEA Inicialmente os resultados indicaram que em média os bancos no Brasil apresentam níveis baixos de eficiência de custo em compara ção com bancos da Europa e dos Estados Unidos Para o período de alta vola tilidade macroeconômica de 2000 a 2002 a ineficiência dos bancos brasilei ros pode ser atribuída à eficiência técnica e não à eficiência alocativa Além disso os bancos públicos brasileiros mostraramse mais eficientes em termos de custo do que os privados nacionais e estrangeiros Tecles e Tabak 2010 por exemplo ao avaliarem a eficiência do setor ban cário brasileiro no período de 2000 a 2007 optaram por utilizar a abordagem de fronteira estocástica bayesiana para estimar a eficiência de custo e lucro Os resultados sugerem que os grandes bancos são mais eficientes em custo e lucro o que pode estar associado com a crescente concentração bancária observada no período de análise Por fim quando se analisaram os determinantes veri ficouse que a capitalização tem efeito positivo na eficiência de custo e lucro O estudo de Wanke e Barros 2014 para o setor bancário brasileiro con siderando dados relativos ao ano de 2012 estimou medidas de eficiência em um processo de dois estágios 1 a eficiência de custos e a produtiva e 2 os determinantes de cada uma dessas medidas de eficiência Os resultados su gerem que os bancos brasileiros são heterogêneos ou seja alguns são consi derados mais eficientes em termos de custo e outros em eficiência produtiva O tamanho do ativo bancário mostrouse o principal determinante da eficiên cia de custo enquanto a eficiência produtiva parece ser impactada positiva mente por eventos de fusão e aquisição bem como pela presença de um controlador privado Nesse sentido os modelos propostos em DEA podem apresentar diferentes resultados conforme o tamanho da amostra das variáveis escolhidas nos insu mos inputs e produtos outputs da orientação etc Da mesma forma os de terminantes da eficiência podem apresentar resultados mistos ou seja não há consensos definitivos Abreu et al 2019 Portanto o presente estudo tem como objetivo agregar novas evidências à literatura de eficiência bancária no Brasil e a seus possíveis determinantes de ineficiência 172 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 3 ANÁLISE DE DADOS A amostra foi composta pelas 39 maiores instituições financeiras que ope raram no Brasil no período de 2010 a 2019 Especificamente as instituições financeiras foram selecionadas de acordo com o tamanho total de ativos e o tipo de consolidado bancário tipo B1 que possuem carteira comercial As sim coletaramse os dados contábilfinanceiros das instituições financeiras no banco de dados IFdata disponibilizado pelo Banco Central do Brasil BCB Dessa forma a Tabela 1 mostra as variáveis utilizadas pelo presente trabalho Tabela 1 Descrição das variáveis do modelo de eficiência Variável Descrição Insumos x x1 Capital x1 patrimônio líquido x2 Depósito x2 total de depósitos Produtos y y1 Empréstimos e recebíveis y1 Operações de crédito líquidas de provisão y2 Total de títulos y2 aplicações interfinanceiras de liquidez Títulos e Valores Mobiliários TVM e instrumentos financeiros derivativos y3 Receita não financeira y3 rendas de prestação de serviços rendas tarifárias bancárias resultado de participações outras receitas operacionais Fonte Elaborada pelos autores com base nos dados obtidos no site do Banco Central do Brasil BCB 4 MÉTODOS DE PESQUISA A aplicação da metodologia neste trabalho está dividida em dois estágios No primeiro estágio a DEA é utilizada para medir o nível de eficiência opera cional das DMU representada pelas instituições financeiras No segundo estágio o modelo Tobit é usado para calcular os determinantes da eficiência técnica Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 173 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 41 Análise envoltória de dados Como já exposto melhorar a gestão das instituições financeiras tornouse um objetivo cada vez mais importante para diminuir os riscos no sistema fi nanceiro Uma das ferramentas mais utilizadas para mensuração de eficiência no setor bancário é a DEA A principal vantagem desse método é sua capaci dade de acomodar múltiplas entradas inputs e saídas outputs Também é útil porque permite avaliar a eficiência relativa comparando cada DMU com outras DMU com melhores pontuações de eficiência Nesse sentido a solução pode produzir unidades tecnicamente eficientes ou unidades com melhores práticas benchmark que são identificadas pelas DMU que se encontram na fronteira de melhores práticas As unidades tecni camente eficientes possuem pontuação igual a 1 ou 100 Por sua vez uni dades tecnicamente ineficientes ou inferiores à fronteira de melhores práticas são inferiores a 1 menor que 100 Cabe ressaltar que um dos objetivos do trabalho é entender a capacidade das instituições financeiras em transformar os insumos em produtos isto é como converter capital e depósito em títulos receitas não financeiras e em préstimos As pontuações de eficiência nos modelos DEA são definidos es sencialmente pela orientação à entrada insumos ou ao produto output A orientação para entrada representa a redução equiproporcional máxima permitida de suas entradas que ainda é capaz de produzir o mesmo nível de produtos outputs Por sua vez a orientação para produto output reflete a expansão equiproporcional máxima de seus produtos outputs que pode ser feita pelo mesmo nível de insumos inputs O sentido da orientação é especi ficar a forma de otimização do modelo de retorno de escala CRS ou VRS Cooper et al 2011 Para avaliar a eficiência bancária o presente trabalho utiliza o modelo DEA BCC ou VRS orientado para o produto output Nesse sentido a motivação para a escolha do modelo DEABCC decorre do fato de que há a necessidade de considerar os retornos variáveis de escala entre as instituições financeiras Além disso o modelo de eficiência do tipo CCR ou CRS que permite avaliar retornos constantes de escala só deve ser usado em contextos em que todas as instituições financeiras operem em condições semelhantes Desse modo dado um conjunto de DMU i n DMU i 12 i no qual cada unidade consome x insumos inputs 1i 2i ei x x x para produzir y produtos outputs 1i 2i fi y y y Assim de acordo com Cooper et al 2011 a eficiência do modelo BCC orientado para produto output da unidade θ k é dada pela seguinte programação linear 174 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 θ k max λ n j ij io j 1 x x i λ θ n j rj ro j 1 y y r λ j 0 j λ n j 1 j 1 1 O modelo BCC VRS introduzido por Banker et al 1984 mede a eficiên cia técnica considerando a restrição de convexidade que garante que a DMU tenha o tamanho de escala proporcional à unidade que está sendo medida Ao contrário do modelo CCR CRS o modelo BCC VRS permite retornos variá veis de escala Como as instituições no Brasil variam drasticamente de tama nho e mix de produtos e serviços financeiros uma estimativa com retornos constantes de escala seria inadequada para nossa análise e por isso o presen te trabalho optou pelo modelo BCC VRS 42 Análise de regressão Tobit Tendo em vista os resultados de eficiência técnica é possível avaliar a par tir do modelo de regressão Tobit as variáveis que influenciaram a variável dependente Isto é as pontuações de eficiência obtidas no modelo DEABCC são usadas como variável dependente E como a pontuação de eficiência é li mitada ao intervalo entre 0 e 1 tornase o uso da técnica de regressão dos mínimos quadrados um modelo inadequado pois pode produzir resultados tendenciosos e inconsistentes Por conta disso a regressão Tobit permite o uso de variável dependente de alcance limitado ou seja variável dependente censurada O modelo Tobit tem sido amplamente utilizado nos estudos para verificar os fatores que afetam o nível de eficiência das unidades de produção Defung et al 2016 A diferen ça de eficiência entre as instituições financeiras não é afetada apenas pelos insumos e produtos mas também por outros fatores condicionais para a ges tão delas Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 175 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 Assim o presente trabalho apresenta um modelo de regressão Tobit para entender melhor a influência dos indicadores de qualidade bancária na eficiên cia técnica dessas instituições financeiras GarzaGarcía 2012 O modelo Tobit do presente trabalho é descrito da seguinte forma α β β β β β β ε 1 2 3 4 5 6 i it it it it it it it ET CAR NIM NII ROA INF PIB 1 39 1 i t n 2 A variável dependente i ET na Equação 2 é a pontuação de eficiência téc nica obtida no modelo DEABCC orientado ao produto output Assim i re presenta a observação instituição financeira t referese ao período α é o termo constante β reflete o coeficiente de cada variável e ε é o termo de erro da regressão As variáveis explicativas estão descritas na Tabela 2 Tabela 2 Descrição das variáveis explicativas da regressão Tobit Variável Descrição CAR Índice de adequação de capital CAR patrimônio líquido total de ativos NIM Margem de lucro líquida NIM receita de intermediação financeira total de depósitos NII Renda sem juros NII rendas de prestação de serviços rendas de tarifas bancárias resultados de participação outras receitas operacionais total de ativos ROA Retorno sobre ativos ROA lucro líquido total de ativos NPLR Razão de índice de inadimplência NPLR provisão sobre operações de crédito operações de crédito líquidas de provisão INF Taxa de inflação anual INF Índice de Preço ao Consumidor Amplo IPCA PIB Taxa de crescimento econômico anual PIB taxa de crescimento econômico real Fonte Elaborada pelos autores com base nos dados obtidos nos sites do BCB Instituto Brasileiro de Geografia e Estatística Banco Mundial e Instituto de Pes quisa Econômica Aplicada 176 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 5 RESULTADOS EMPÍRICOS 51 Eficiência técnica do setor bancário brasileiro Os resultados de eficiência técnica do modelo DEA das 39 instituições fi nanceiras analisadas entre 2010 e 2019 foi em média de 07964 7964 O ano de melhor performance foi em 2018 com 08846 8846 e a pior performance ocorreu em 2017 com 06153 6153 As pontuações de eficiên cia técnica média de cada instituição financeira estão apresentadas na Tabela 3 Tabela 3 Eficiência técnica média das instituições financeiras de 2010 a 2019 Instituição financeira Tipo de propriedade Tamanho Eficiência técnica média Desvio padrão BB Público nacional Grande 10000 00000 CEF Público nacional Grande 10000 00000 Alfa Privado nacional Pequeno 10000 00000 Sicredi Privado nacional Pequeno 10000 00000 Itaú Unibanco Privado nacional Grande 10000 00000 Bradesco Privado nacional Grande 09931 00169 Votorantim Privado nacional Médio 09756 00770 Bancoob Privado nacional Pequeno 09650 00564 Safra Privado nacional Médio 09537 00850 CSF Estrangeiro Pequeno 09302 02207 Rabobank Estrangeiro Pequeno 09277 02125 John Deere Estrangeiro Pequeno 09241 01946 Volkswagen Estrangeiro Pequeno 08943 02229 Santander Estrangeiro Grande 08812 01564 BNP Paribas Estrangeiro Pequeno 08714 01078 Nordeste Público nacional Médio 08529 01119 J P Morgan Estrangeiro Pequeno 08481 01617 HSBC Estrangeiro Grande 08342 01207 Pan Privado nacional Pequeno 08203 02019 continua Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 177 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 Instituição financeira Tipo de propriedade Tamanho Eficiência técnica média Desvio padrão MercedesBenz Estrangeiro Pequeno 08185 02630 BTG Pactual Privado nacional Grande 08129 02009 Banestes Público nacional Pequeno 08103 01540 CCB Estrangeiro Pequeno 07774 02116 BMG Privado nacional Pequeno 07580 02452 Société Générale Estrangeiro Pequeno 07326 01709 BofA Securities Estrangeiro Pequeno 07320 01813 Citibank Estrangeiro Médio 07181 01366 ING Estrangeiro Pequeno 06988 03069 BRB Público nacional Pequeno 06790 01369 Daycoval Privado nacional Pequeno 06684 01755 Banpará Público nacional Pequeno 06666 01749 GM Estrangeiro Pequeno 06474 01844 Banrisul Público nacional Médio 06133 01090 ABC Estrangeiro Pequeno 05913 00837 Pine Privado nacional Pequeno 05581 02035 Amazônia Público nacional Pequeno 05465 01480 Credit Suisse Estrangeiro Pequeno 05265 02191 Crédit Agricole Estrangeiro Pequeno 04966 01656 Original Privado nacional Pequeno 03328 01524 Fonte Elaborada pelos autores com base nos dados obtidos no site do BCB Os resultados apontam que as únicas instituições financeiras tecnicamente eficientes ET 1 em todo o período analisado de 2010 a 2019 foram Banco do Brasil BB Caixa Econômica Federal CEF Alfa Sicredi e Itaú Unibanco Cabe ressaltar que a eficiência técnica ET é relacionada à qualidade da ges tão ou seja as instituições financeiras tecnicamente eficientes ET 1 são consideradas as unidades com as melhores práticas ou o benchmark do setor analisado Entretanto avaliar cada instituição apenas revela quais unidades de produ ção ou DMU utilizam de forma eficiente seus recursos inputs na elaboração Tabela 3 Eficiência técnica média das instituições financeiras de 2010 a 2019 conclusão 178 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 de seus produtos outputs Ou seja apenas se avalia a qualidade de gestão de uma determinada instituição Assim para obtermos avaliações mais agrega das podemos categorizar as instituições financeiras o que permite examinar eficiência por agrupamentos bancários Nesse sentido na literatura de eficiência bancária há o interesse de verificar se o tamanho total de ativos influencia o desempenho das instituições finan ceiras A Tabela 4 mostra quais grupos bancários por tamanho pequeno mé dio e grande são em média mais eficientes no mercado bancário Tabela 4 Eficiência técnica média das instituições financeiras por tipo de tamanho Ano Pequeno Médio Grande Média DP 2010 08090 08923 09931 08981 00922 2011 07268 08328 09407 08335 01070 2012 07956 08453 09692 08700 00894 2013 07625 07915 09610 08384 01072 2014 07601 07709 09248 08186 00921 2015 06794 07952 08493 07747 00868 2016 07692 07973 09578 08414 01017 2017 05509 06978 08259 06915 01376 2018 08599 09002 09786 09129 00604 2019 08347 09038 09729 09038 00691 Média 07548 08227 09373 DP 00884 00656 00562 Fonte Elaborada pelos autores com base nos dados obtidos no site do BCB DP desvio padrão Em relação ao tamanho das instituições financeiras apontase que as de grande porte são em média mais eficientes 9373 do que aquelas de mé dio 8227 e pequeno 7548 portes Isso corrobora a ideia de que as instituições de grande porte por possuírem quantidade maior de agências bancárias produtos mais diversificados e melhor aplicação de tecnologias po dem alcançar níveis de eficiência mais elevados Defung et al 2016 Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 179 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 No período de desaceleração econômica em 2014 e com recessão econô mica de 20152016 notase redução de 537 da eficiência técnica média entre 2014 e 2015 puxada principalmente pelas instituições de pequeno e grande portes Essa queda da eficiência pode ser atribuída à queda do total de depósitos 744 no setor bancário que ao reduzir os insumos inputs ao mesmo tempo teve redução no volume total dos empréstimos 024 A crise econômica no Brasil entre 2014 e 2016 originouse principalmen te da redução da capacidade de crescimento econômico e da crescente eleva ção da dívida pública federal Vartanian Garbe 2019 Essas mudanças no ambiente macroeconômico podem ter impactado o setor bancário em diferen tes graus dependendo do tamanho das perdas esperadas dos empréstimos concedidos e da desaceleração das operações de crédito O pior desempenho do setor bancário em 2017 se deve sobretudo à queda de depósitos 2258 das instituições de grande porte Além disso os pro dutos outputs das instituições de pequeno médio e grande portes sofreram reduções expressivas Essa redução dos produtos outputs pode ser auferida pelas perdas nos empréstimos 232 nos investimentos de títulos e instru mentos derivativos 790 e nas receitas não financeiras 2327 Outra classificação de bancos que possui relevância nos estudos de eficiên cia bancária está associada ao tipo de propriedade ou tipo de controle Nessa classificação é comum debater se os gestores das instituições públicas são mais ineficientes em relação às instituições privadas e estrangeiras o que é normalmente justificado pela interferência do governo ou pelos incentivos inadequados para a gestão desse tipo de instituição Jayaraman Srinivasan 2019 Claessens et al 2001 Sturm Williams 2004 Na Tabela 5 apresentase a eficiência técnica média das instituições finan ceiras por tipo de propriedade público estadual e federal privado nacional e estrangeiro O grupo de bancos públicos federais BB CEF Banco do Nordeste Banco da Amazônia é considerado o tipo de propriedade com melhor desempenho ao longo do período amostral com pontuação média de 08498 8498 Apesar do baixo desempenho do Banco da Amazônia com média de 05465 5465 o BB a CEF e o Banco do Nordeste apresentaram pontuações ele vadas durante todo o período analisado o que proporcionou a alta eficiência técnica média para o grupo 180 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 Tabela 5 Eficiência técnica média das instituições financeiras por tipo de propriedade Ano Público estadual Público federal Privado nacional Estrangeiro Média DP 2010 07174 08463 09534 08216 08347 00969 2011 05834 08064 08626 07592 07529 01207 2012 07730 08313 08740 08201 08246 00415 2013 07630 08021 08777 07553 07995 00560 2014 06174 07868 08078 08258 07594 00961 2015 05309 07947 07444 07381 07020 01169 2016 07489 08272 08181 07995 07984 00350 2017 05822 09052 06556 05185 06654 01694 2018 08644 09405 09093 08556 08924 00397 2019 07423 09578 08819 08604 08606 00893 Média 06923 08498 08385 07754 DP 01070 00624 00864 00994 Fonte Elaborada pelos autores com base nos dados obtidos no site do BCB DP desvio padrão No entanto se os bancos públicos federais e estaduais fossem analisados conjuntamente as instituições mais eficientes seriam os bancos estrangeiros Isto é os bancos públicos estaduais e federais teriam pontuação de 07711 7711 que seria inferior à pontuação média de 08385 8385 dos ban cos estrangeiros Dessa forma notamse diferenças significativas entre insti tuições públicas estaduais e federais A possível justificativa para o desempenho inferior das instituições públi cas estaduais pode estar associada aos incentivos errados como remuneração ou bônus inadequados aos gestores falta de rigor na avaliação de desempe nho não transparência etc Esses problemas refletemse nas operações ban cárias como falhas na supervisão da avaliação de risco dos empréstimos e volume mais expressivo de inadimplências Quando se analisam os períodos de melhor ou pior desempenho por tipo de propriedade notase que as instituições públicas estaduais privadas nacio nais e estrangeiras são mais sensíveis aos ciclos econômicos Por exemplo no período de recessão de 2015 a 2016 as instituições financeiras exceto os Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 181 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 bancos públicos federais tiveram quedas significativas no desempenho de modo que a atividade econômica pode influenciar a composição de insumos inputs e produtos outputs do setor bancário Para tanto as variáveis macroeconômicas por exemplo taxa de cresci mento e inflação podem ter efeitos significativos em determinadas institui ções financeiras Delis Papanikolaou 2009 Assim não só os fatores inter nos relacionados à gestão dos bancos influenciam os níveis de eficiência mas também os fatores externos por exemplo reformas na regulação e variáveis macroeconômicas a depender do grau impactam o desempenho dos bancos Berger DeYoung 1997 O fato de os bancos públicos federais exceto o Banco da Amazônia terem apresentado melhor desempenho nos períodos de desaceleração econômica pode estar associado às políticas de fomento ao crédito pelo governo federal Isto é as instituições públicas federais impulsionam suas carteiras de crédito nos períodos de desaceleração econômica o que permite elevar seus produtos outputs O ano com o pior desempenho foi 2017 e isso ocorreu em grande medi da por causa da redução dos produtos outputs dos bancos públicos esta duais 903 bancos públicos federais 503 bancos privados nacionais 903 e bancos estrangeiros 769 Parte considerável das perdas no setor bancário nesse período está associada com as receitas não financeiras que em média tiveram uma redução de 2326 Os resultados do baixo desempenho em 2017 podem estar relacionados aos efeitos tardios da recessão econômica de 20152016 e para que pudes sem mitigar maiores perdas dos produtos outputs as instituições reformula ram suas operações o que reduziu seus níveis de eficiência Consequente mente em 2018 é possível observar que essa reestruturação das operações elevou expressivamente a eficiência técnica para todos os tipos de instituição financeira 52 Determinantes da eficiência técnica do setor bancário brasileiro Para ampliar a avaliação de eficiência bancária é comum complementar as análises com os determinantes da eficiência Esses determinantes podem in cluir indicadores relacionados com as informações contábeis das instituições financeiras ou estabelecer relações com variáveis micro e macroeconômicas 182 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 Neste trabalho avaliaramse os indicadores bancários para gestão interna como o índice de adequação de capital capital adequacy ratio CAR margem financeira líquida net interest margin NIM renda sem juros noninterest income NII retorno sobre ativos return on assets ROA proxy para perdas de crédito nonperformance loans ratio NPLR e variáveis macroeconômicas como a taxa de inflação INF e taxa de crescimento econômico PIB A Tabe la 6 apresenta os determinantes da eficiência técnica a partir da regressão Tobit que permite avaliar variáveis censuradas Tabela 6 Determinantes da eficiência técnica Coeficiente Estimativa Erro padrão CAR Índice de adequação de capital 18469 03166 NIM Margem financeira líquida 02333 00436 NII Renda sem juros 113612 23265 ROA Retorno sobre ativos 54678 31021 NPLR Proxy para perdas de crédito 17106 04420 INF Taxa de inflação anual 35595 14149 PIB Taxa de crescimento econômico anual 06208 09230 Fonte Elaborada pelos autores com base em dados obtidos no site do BCB e denotam significância aos níveis de 1 5 e 10 respectivamente O CAR foi negativamente significante para o modelo Essa variável é padrão nos trabalhos sobre determinantes de eficiência bancária e pode ter resultados mistos Defung et al 2016 Nesse caso o resultado negativo do CAR pode indicar que os bancos tendem a ser excessivamente conservadores com a con cessão das operações de crédito o que pode levar a uma menor eficiência Nesse contexto o trabalho de Altunbas et al 2007 indica que há uma rela ção inversa entre eficiência e capitalização isto é um alto índice de capitalização pode reduzir o risco na concessão dos empréstimos No entanto pode restrin gir a alavancagem e os retornos financeiros que consequentemente afetam negativamente o nível de eficiência A variável NIM foi considerada positivamente significativa assim como os resultados apresentados por GarzaGarcía 2012 A relação positiva da NIM Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 183 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 aponta que as receitas líquidas de juros são importantes para o nível de eficiên cia isto é os lucros que provêm de empréstimos e investimentos contribuem para o desempenho positivo da instituição financeira Podese sugerir que as instituições financeiras que procuram aumentar as margens de receitas com os empréstimos e investimentos tendem a ter maiores pontuações de eficiência Da mesma forma a NII também foi significativa para a eficiência técnica A variável NII indica como as instituições financeiras geram receitas não decor rentes de juros com o objetivo de ampliar suas receitas e melhorar os níveis de liquidez em caso de aumento de perdas esperadas da inadimplência e outros investimentos Nesse aspecto o resultado pode indicar que as instituições fi nanceiras com maior capacidade de geração de receitas sem juros podem as segurar melhor desempenho O ROA não foi significativo para o modelo apesar de ser um importante indicador para a rentabilidade das instituições financeiras Em relação aos empréstimos inadimplentes observase que a NPLR é ne gativamente significativa para o nível de eficiência do setor bancário brasileiro Esse resultado sinaliza que as instituições financeiras são sensíveis ao aumen to da inadimplência e prejudicam a rentabilidade dos negócios com emprésti mos Assim o volume elevado de empréstimos inadimplentes pode reduzir a capacidade das instituições financeiras de produzir novos empréstimos e pa gar os custos operacionais Berger DeYoung 1997 Por fim as variáveis macroeconômicas selecionadas neste trabalho como a taxa de inflação INF e taxa de crescimento econômico anual PIB alcança ram resultados distintos A inflação INF obteve resultado significativo e cor respondeu ao esperado isto é a inflação teve impacto negativo sobre as ope rações bancárias alterando os preços dos ativos e serviços o que aumentou os custos para manter a rentabilidade dos negócios bancários Tan 2016 Já a taxa de crescimento econômico PIB que na literatura comumente parece ter impacto significativo e positivo sobre a eficiência das instituições financeiras ver por exemplo GarzaGarcía 2012 no presente trabalho não apresentou significância estatística CONSIDERAÇÕES FINAIS O objetivo deste artigo foi comparar a eficiência entre instituições financeiras no Brasil durante o período de 2010 a 2019 As estimativas de eficiência téc nica para cada banco foram calculadas usandose o modelo BCCVRS orientado 184 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 ao produto output da DEA Complementarmente no presente trabalho utili zouse a regressão Tobit para encontrar estimativas sobre os determinantes da eficiência técnica obtidos na DEA Os resultados empíricos sobre o tamanho dos bancos sugerem que as ins tituições de grande porte são mais eficientes tecnicamente em comparação com as instituições de pequeno e médio portes Uma explicação possível pode estar associada a melhor uso da tecnologia maior escala de suas operações e produtos mais diversificados como destacado por Defung et al 2016 Em relação ao tipo de propriedade ou controle os bancos públicos fede rais em média apresentaramse como mais eficientes que os outros tipos de propriedade No entanto quando se comparam as instituições públicas ou seja estaduais e federais no mesmo conjunto os bancos estrangeiros passam a ser o grupo com maior nível de eficiência Parte considerável da queda de eficiência técnica dos bancos públicos nacionais está associada ao baixo de sempenho dos bancos estaduais Banestes BRB Banpará e Banrisul Os determinantes da eficiência técnica tiveram resultados mistos As variá veis internas como NIM e NII foram positivamente significativas e correspon deram à capacidade de os bancos obterem receitas de juros e sem juros res pectivamente Isso indica que a geração de receita é um importante atributo para melhorar o nível de eficiência das instituições financeiras O impacto negativo do CAR pode sugerir que o maior nível de capitaliza ção assegura maior solidez financeira contra desacelerações econômicas mas restringe a eficiência técnica das instituições financeiras A NPLR por sua vez foi negativamente significativa para o nível de eficiência que indica que o vo lume elevado de empréstimos inadimplentes reduz a capacidade de os bancos operarem de forma eficiente Quando se analisaram as variáveis macroeconômicas observouse que a inflação INF pode produzir efeitos negativos sobre a eficiência do setor ban cário brasileiro No entanto a taxa de crescimento econômico PIB não apre sentou significância em relação ao nível de eficiência o que não descarta a possibilidade de que as desacelerações na economia possam reduzir as opera ções bancárias e consequentemente a eficiência das instituições financeiras As evidências do presente trabalho são importantes para avaliação dos ges tores dos diferentes tipos de instituição financeira da mesma forma que apon tam os agrupamentos bancários com melhor e pior desempenho As pesquisas futuras por exemplo poderiam concentrarse nas reformas regulatórias sobre o setor bancário Esses resultados podem evidenciar se o nível de capital re querido tem influência no desempenho das instituições financeiras Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 185 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 BANKING EFFICIENCY IN BRAZIL 20102019 A TWOSTAGE ANALYSIS Abstract The objective of this work is to analyze the performance of the Brazilian banking sector and its determinants The present research was divided into two parts the first part seeks to determine the technical efficiency by the DEABCC model in the main Brazilian banks in the period from 2010 to 2019 In the second part the determinants that can influence the performance of banks are evaluated Esti mates indicate that large banks are on average more efficient than small and mediumsized banks In addition when looking at the type of ownership foreign banks tend to perform better compared to national banks In addition the deter minants point out that the net interest margin and noninterest income are sig nificant and contribute positively to the level of efficiency In turn the capital adequacy ratio nonperforming loans and inflation have a negative impact on technical efficiency Keywords Data Envelopment Analysis determinants technical efficiency financial institutions banking sector Referências Abreu E S de Kimura H Sobreiro V A 2019 What is going on with studies on banking efficiency Research in International Business and Finance 47 195219 httpsdoiorg101016j ribaf201807010 Ahmad N Naveed A Ahmad S Butt I 2019 Banking sector performance profitability and efficiency A citationbased systematic literature review Journal of Economic Surveys 341 185 218 httpsdoiorg101111joes12346 Altunbas Y Carbo S Garderner E P M Molyneux P 2007 Examining the relationship between capital risk and efficiency in European banking European Financial Management 131 4970 httpsdoiorg101111j1468036X200600285x Banker R D Charnes A Cooper W W 1984 Some models for estimating technical and scale inefficiencies in Data Envelopment Analysis Management Science 309 10311142 https doiorg101287mnsc3091078 Berger A N DeYoung R 1997 Problem loans and cost efficiency in comercial banks Journal of Banking and Finance 216 849870 httpsdoiorg101016S0378426697000034 186 Revista de Economia Mackenzie v 19 n 2 São Paulo SP juldez 2022 p 165187 ISSN 18082785 online Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 Charnes A Cooper W W Rhodes E 1978 Measuring the efficiency of decision making units European Journal of Operational Research 26 429444 httpsdoiorg10101603772217 78901388 Claessens S DemirgüçKunt A Huizinga H 2001 How does foreign entry affect domestic banking markets Journal of Banking and Finance 255 891911 httpsdoiorg101016S0378 426600001023 Claessens S Laeven L 2005 Financial dependence banking sector competition and eco nomic growth Journal of the European Economic Association 31 179207 httpsdoiorg101162 1542476053295322 Cooper W W Seiford L M Zhu J Orgs 2011 Handbook on Data Envelopment Analysis Springer ScienceBusiness Media Defung F Salim R Bloch H 2016 Has regulatory reform had any impact on bank efficiency in Indonesia A twostage analysis Applied Economics 4852 50605074 httpsdoiorg101080 0003684620161170934 Delis M D Papanikolaou N I 2009 Determinants of bank efficiency Evidence from a semiparametric methodology Managerial Finance 353 260275 httpsdoiorg101108030 74350910931771 Diallo B 2018 Bank efficiency and industry growth during financial crises Economic Modelling 68 1122 httpsdoiorg101016jeconmod201703011 Emrouznejad A Parker B Tavares G 2008 Evaluation of research in efficiency and produc tivity A survey and analysis of the first 30 years of scholarly literature in DEA SocioEconomic Planning Sciences 423 151157 httpsdoiorg101016jseps200707002 Emrouznejad A Yang G 2018 A survey and analysis of the first 40 years of schorlarly literature in DEA 19782016 SocioEconomic Planning Sciences 61 48 httpsdoiorg101016jseps2017 01008 Farrell M J 1957 The measurement of productivity efficiency Journal of the Royal Statistical Society Serie A 1203 253281 httpsdoiorg1023072343100 GarzaGarcía J G 2012 Determinants of bank efficiency in Mexico A twostage analysis Applied Economic Letters 1917 16791682 httpsdoiorg101080135048512012665589 Jayaraman A R Srinivasan M R 2019 Determinants of Indian banks efficiency A twostage approach International Journal of Operational Research 362 270291 httpsdoiorg101504 IJOR2019102414 Kose M A Nagle P Ohnsorge F Sugawara N 2010 Global waves of debt Causes and conse quences World Bank Kutlar A Kabasakal A Ekici M S 2017 Efficiency of commercial banks in Turkey and their comparison Application of DEA with Tobit analysis International Journal of Mathematics in Operational Research 101 84103 httpsdoiorg101504IJMOR2017080746 Eficiência bancária no Brasil 20102019 Uma análise de dois estágios Fábio Lucas Takahashi Marcos Roberto Vasconcelos 187 Revista de Economia Mackenzie São Paulo v 19 n 2 doi10593518082785remv19n2p165187 Levine R 1997 Financial development and economic growth Views and agenda Journal of Economic Literature 352 688726 httpwwwjstororgstable2729790 Micco A Panizza U Yañez M 2007 Bank ownership and performance Does politics matter Journal of Banking and Finance 311 219241 httpsdoiorg101016jjbankfin200602007 Nguyen T L A 2018 Diversification and bank efficiency in six ASEAN countries Global Finance Journal 37 5778 httpsdoiorg101016jgfj201804004 Staub R B Souza G S Tabak B 2010 Evolution of bank efficiency in Brazil A DEA approach European Journal of Operational Research 2021 204213 httpsdoiorg101016jejor2009 04025 Sturm J Williams B 2004 Foreign bank entry deregulation and bank efficiency Lessons from the Australian experience Journal of Banking and Finance 281 17751799 httpsdoiorg 101016jjbankfin200306005 Tan Y 2016 The impacts of risk and competition on bank profitability in China Journal of International Financial Markets Institutions and Money 40 85110 httpsdoiorg101016jintfin 2015 09003 Tecles P L Tabak B M 2010 Determinants of bank efficiency The case of Brazil European Journal of Operational Research 2073 15871598 httpsdoiorg101016jejor201006007 Vartanian P R Garbe H de S 2019 The Brazilian economic crisis during the period 20142016 Is there precedence of internal or external factors Journal of International and Global Economic Studies 121 6686 Wanke P Barros C 2014 Twostage DEA An application to major Brazilian banks Expert System with Applications 415 23372344 httpsdoiorg101016jeswa201309031 Copyright of Revista de Economia Mackenzie is the property of Universidade Presbiteriana Mackenzie Revista de Economia Mackenzie and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holders express written permission However users may print download or email articles for individual use Accepted Manuscript Efficiency Factors in OECD Banks A TenYear Analysis Peter Wanke Abul Kalam Azad CP Barros PII S0957417416303657 DOI 101016jeswa201607020 Reference ESWA 10766 To appear in Expert Systems With Applications Received date 25 January 2016 Revised date 13 June 2016 Accepted date 14 July 2016 Please cite this article as Peter Wanke Abul Kalam Azad CP Barros Efficiency Fac tors in OECD Banks A TenYear Analysis Expert Systems With Applications 2016 doi 101016jeswa201607020 This is a PDF file of an unedited manuscript that has been accepted for publication As a service to our customers we are providing this early version of the manuscript The manuscript will undergo copyediting typesetting and review of the resulting proof before it is published in its final form Please note that during the production process errors may be discovered which could affect the content and all legal disclaimers that apply to the journal pertain ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 1 Highlights This paper assesses performance of 128 OECD banks from 20042013 An analytical approach that emulates the CAMELS rating system is used Results reveal that contextual variables have a prominent impact on efficiency levels ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 2 Efficiency Factors in OECD Banks A TenYear Analysis Peter Wanke COPPEAD Graduate Business School Federal University of Rio de Janeiro Rua Paschoal Lemme 355 21949900 Rio de Janeiro Brazil Email petercoppeadufrjbr Phone 55 21 39389494 Corresponding author Md Abul Kalam Azad University of Malaya Department of Applied Statistics Faculty of Economics and Administration 50603 Kuala Lumpur Malaysia Email azadiiucgmailcom Phone60176181759 CP Barros ISEG Lisbon School of Economics and Management University of Lisboa and CEsA Research Centre on African Asian and Latin American Studies Rua Miguel Lupi 20 1249078 Lisboa Portugal Email address cbarrosisegutlpt Phone 351 213 016115 Fax 351 213 925 912 Research made with support of Calouste Gulbenkian Foundation Abstract This paper presents a performance assessment of 128 banks from 23 OECD countries from 20042013 using different financial criteria that emulate the CAMELS rating system A robust TOPSIS approach for assessing bank efficiency is also developed and presented First alternative variable reduction techniques are employed to extract the major factors within each CAMELS criterion This is done to mitigate collinearity issues Then TOPSIS is used to measure bank performance based upon these factors equally weighted A comprehensive analysis based on a weighted linear optimization model for multicriteria classification is also performed which detects any discrepancies from the original scores Lastly censored quantile regressions are combined with bootstrapped TOPSIS scores to produce a model for predicting the impact of different contextual variables on different efficiency quantiles Results reveal that the effects of ownership trend and origin of the bank may vary with respect to efficiency levels whether high or low Keywords Banks OECD factor extraction TOPSIS censored quantile regression ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 3 1 Introduction Since the inception of efficiency measurement tools in banking most studies have focused on developed economies at the country level see for instance Matousek et al 2014 and the reference therein In this sense Paradi et al 2011 listed the ten countries most targeted by researchers worldwide all were developed economies except for India Some studies following a different perspective focused on economic blocks and analyzed specific issues such as publicprivate ownership bank origin and merger and acquisitions MA processes considered relevant to understanding banking performance Sturm Williams 2010 Ebrahim Girma Shah Williams 2014 In this regard OECD banks are of particular interest A handful of papers have analyzed different aspects of banking activity between and within OECD countries Jaffee Levonian 2001 Levintal 2013 Müller Uhde 2013 However to the best of our knowledge comparative performance studies within the realm of OECD banks are still scarce Thus this paper aims to contribute to the current body of knowledge by exploring the use of a robust TOPSIS Technique for Order of Preference by Similarity to Ideal Solution approach applied to a set of financial criteria that emulates the CAMELS rating system for purposes of assessing the performance of 128 banks from 23 OECD countries from 20042013 Although numerous studies on bank efficiency performed at the country level have been produced they are still concentrated in two main approaches best known as parametric and nonparametric Berger Humphrey 1997 De Borger Ferrier Kerstens 1998 Athanassopoulos 1998 Brandouy Briec Kerstens Van de Woestyne 2010 Brissimis Delis Tsionas 2010 Kerstens Mounir Van de Woestyne 2011 Briec Liang 2011 Lampe Hilgers 2014 Cummins RubioMisas 2006 The most popular parametric method is the Stochastic Frontier Approach or SFA whereas the most popular nonparametric method is the Data Envelopment Analysis or DEA Gunay 2012 Yang 2009 Chen 2002 Tözüm 2002 Li Liu Liu Whitmore 2001 Deliktas Balcilar 2005 Nonparametric methods however are more widely used in different banking industries in various countries and regions around the globe Wanke et al 2015 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 4 An emerging trend in banking performance measurement entails the use of multi criteria decisionmaking MCDM methods such as TOPSIS The TOPSIS method suggested by Hwang and Yoon 1981 belongs to the group of linear ranking methods of multidimensional objects Broadly speaking the ranking of objects from best to worst considering an assumed synthetic measure which is not subjected to a direct measurement is a task of linear ordering Dudek and Jefmanski 2015 According to Behzadian Khanmohammadi Otaghsara Yazdani and Ignatius 2012 TOPSIS studies in banking are still scarce For example Shaverdi et al 2011 applied TOPSIS to rank the performance of banks based on criteria weights obtained using a fuzzy analytical hierarchy process FAHP Similarly Mandic Delibasic Knezevic and Benkovic 2014 examined Serbian bank performance by first applying FAHP and then TOPSIS in a two stage approach Among earlier studies İÇ and Yurdakul 2010 proposed a ranking model using financial ratios and fuzzy TOPSIS for computing credit risk Similarly Tansel İç 2012 proposed a model combining fuzzy TOPSIS with linear programming for determining credit risk Most recently Wanke Barros and Macanda 2015 studied Angolan bank efficiency using TOPSIS and neural networks in a twostate approach This paper differs from previous studies and contributes to the state of art of the methodology in efficiency measurement by developing and presenting a robust TOPSIS approach using 10 years of panel data of OECD banks The robust TOPSIS approach adopted here is anchored in three key aspects which are further detailed in the methodology section First of all two different variable reduction techniques Principal Component Analysis PCA and Independent Component Analysis ICA are alternatively used to extract the main factors embedded in the several financial ratios that compose the different CAMELS criteria This is necessary not only to reduce the number of criteria to a manageable number of factors but also to avoid collinearity issues between different financial criteria Second the weighted linear optimization model for multi criteria classification proposed by Ng 2007 used an auxiliary tool of which different financial criteria orders and therefore their respective set of weights are comprehensively tested In this case any discrepancies can be measured and compared against the original TOPSIS scores computed with equal weights for the PCA and ICA extracted factors In contrast with other MCDM methods decision makers are free to determine their own set of ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 5 criteria weights in TOPSIS Put differently because criteria weights are defined exogenously in TOPSIS sensitivity analyses can be performed on different weighting criteria to assess the robustness of the results Thus the role of the Ng model within the ambit of TOPSIS to provide TOPSIS the weights for each criteria computed during the optimization stage Third bootstrapped truncated regressions similarly to what is done with classical DEA models are used to resample original TOPSIS scores and then in a censored quantile regression model to explore the impact of different contextual variables on different efficiency quantiles Results presented in this research contribute to the growing literature on banking efficiency especially by shedding light on how country origin ownership and MA may affect financial performance These results are consistent with and extend the findings of recently conducted studies Dima Dincă Spulbăr 2014 Du Sim 2015 Fungáčová Poghosyan 2011 Gardener Molyneux Hoai 2011 GarzaGarcia 2012 CC Lee Hsieh 2014 Sufian 2007 thus adding to the body of knowledge The remainder of the paper is organized as follows Section 2 presents the contextual setting Section 3 covers the literature review and Section 4 presents the methodology Data is presented in Section 5 while empirical results are presented and discussed in terms of policy implications in Section 6 Conclusions follow in Section 7 2 Contextual Setting Efficiency in OECD banks appears to be substantially impacted at the country level thus suggesting that regulatory barriers cultural aspects and the economic dynamics within each country are possible drivers for understanding higher or lower levels of banking performance Furthermore temporal analyses of basic financial ratios suggest that differences in banking indicators between OECD countries seem to spread extending over several years especially after the world financial crisis This observation also indicates that each country responded to the financial crisis in an individual way Thus four key development indicators and bank information are presented in Figure 1 and Table 1 First an inspection of the nonperforming loans to total gross loans in Figure 1 makes it clear that not all the selected countries pattern similarly along the time ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 6 series According to world development indicators nonperforming loans usually express the ratio between the value of nonperforming loans and the total value of the loan portfolio including nonperforming loans before the deduction of specific loanloss provisions Figure 1 depicts a common pattern of increasing frequency of nonperforming loans among the main six OECD countries Greece has the highest percentage followed by Ireland Italy Hungary Slovakia and Portugal which range from 343 to 1052 On the other hand Poland has successfully managed to decrease its level of nonperforming loans from 215 in 2001 to 57 in 2014 Figure 1 also reveals that all the major shifts of non performing loans among the above countries started after the global financial crisis of 2008 For the other selected countries bank risk for nonperforming loans remains relatively unchanged compared with the period prior to the financial crisis Figure 1 Nonperforming loans to total gross loans Source World development indicators retrieved on 251015 On the other hand Figure 2 presents liquid reserves to assets ratio for the selected OECD countries The significant finding related to Figure 2 is that a reasonable number of anomalies can be observed after the global financial crisis of 2008 All of the countries except Greece 95 and Denmark 54 had been experiencing an increase of liquid reserves to bank assets ratio within a limit of 08 to 301 from 2001 to 2007 But after 2008 these countries experienced significant variation throughout the following ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 7 years and to date the range between 06 and 62 The exceptional countries with banks having a high percentage of liquid assets are United States with 167 and Poland with 132 This suggests that these countries have economic responses in place to enhance their ability to overcome the worse economic conditions in the postcrisis period Figure 2 Banks liquid reserves to bank assets ratio Source World development indicators accessed 10252015 The remaining two variables bank capital to assets ratio and GDP per capita growth annual are presented in detail in Table 1 The data pertaining to bank capital to assets ratio show that all the selected countries exhibit similar patterns over the years Interestingly the United States has achieved a comparatively higher rate in all years However for the remaining member countries results range from 117 Ireland to 43 Finland This particular issue has revealed that OECD banks have been achieving a standard ratio in all these years Last but not the least GDP growth has a positive correlation pattern among these countries as presented in Table 1 In the year 2015 the GDP growth stayed within a range of 45 for Ireland to 22 for Italy The world economic crisis of 2008 is still having an impact here The only exception is Greece which is still struggling to recover economically In summary these four indicators exemplify the importance of the country effect and its underlying variables regulation economic dynamics culture for comprehending certain aspects of banking efficiency in OECD countries Section 3 further explores i how these financial indicators and others are ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 8 combined to emulate the CAMELS rating system and ii the expected impact of other contextual variables such as ownership and merger and acquisitions on efficiency levels in light of the current literature Table 1 Country level data of OECD Countries Country Name 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Bank capital to assets ratio 1 Australia 5200 NA 6900 5200 5000 5400 6289 6286 6070 6140 6004 5880 Austria 4900 NA NA 5240 6484 6284 6977 7493 7186 7770 7986 6922 France 5400 NA NA NA NA 3728 4098 4878 4754 5201 5818 NA Greece 6900 NA NA 7000 6800 6400 6959 6665 NA 5758 7548 9114 Luxembourg 5800 NA NA 5000 5000 4796 5537 5234 4998 6284 6392 6099 New Zealand NA NA NA NA 5300 6400 5500 6000 6500 7100 7300 7400 Poland 8300 NA NA 7800 8000 7500 8084 8198 7818 8693 9098 9253 Norway 5900 NA NA 7000 6400 5900 6000 6700 6800 NA NA NA Korea Rep 7000 NA NA NA NA 6295 7294 7595 8114 8231 8273 8036 Italy 6400 NA 4600 4600 4600 4100 4816 4979 5448 5386 5404 5922 Ireland 5300 NA 4329 4508 4395 3746 5435 5323 6440 7279 7749 11486 Finland 10900 NA 9400 9400 8000 6200 6407 5525 4354 4417 5011 4279 Denmark 5900 NA 4900 5700 6200 5500 5493 5430 5054 5266 7061 7466 Belgium 3100 NA NA 3300 4300 3300 4628 4964 4600 5761 6390 6335 Germany 4200 NA 4070 4329 4267 4500 4807 4298 4362 4727 5453 5612 Spain 7300 NA 6300 6400 6700 5900 6418 6095 5924 5757 6792 7242 Slovenia 8300 NA 8500 8400 8400 8400 8300 8200 NA NA NA NA Sweden 5000 NA NA 4900 4800 4700 5000 NA NA NA NA NA Portugal 5800 NA 6152 6641 6526 5800 6494 6697 5327 6701 6929 4651 United Kingdom 6600 NA NA 6100 5500 4400 5393 5366 5100 5509 6345 5865 United States 9200 NA 10300 10500 10300 9300 12375 12739 12229 11963 11776 11719 Switzerland 5700 NA NA 4900 4600 4837 5551 5401 5548 5514 5707 NA GDP per capita growth2 Australia 1815 2956 1863 1482 3114 1645 0343 0389 0910 1956 0749 0883 Austria 0267 2071 1447 2841 3286 1230 4051 1635 2725 0425 0356 0347 France 0107 2032 0845 1664 1730 0363 3439 1463 1587 0271 0221 0246 Greece 6303 4597 0553 5482 3212 0653 4400 5164 8616 6315 3327 1416 Luxembourg 0032 3445 2543 3222 4832 1293 7071 3243 0354 2532 0340 NA New Zealand 2587 2285 2250 1519 2004 2447 1244 0320 1442 1619 1691 NA Poland 3632 5197 3593 6267 7215 3852 2553 4006 4713 1821 1775 3491 Norway 0330 3347 1928 1574 1870 0859 2855 0643 0333 1408 0469 1103 Korea Rep 2423 4507 3711 4668 4974 2092 0230 6005 2913 1832 2455 2892 Italy 0292 0928 0455 1700 0963 1703 5911 1398 0414 3032 2831 2216 Ireland 1294 2684 3397 2665 1942 4575 7317 0817 2399 0533 0075 4465 Finland 1751 3625 2429 3657 4738 0253 8707 2522 2096 1894 1775 0564 Denmark 0117 2374 2155 3456 0378 1300 5594 1175 0737 1028 0900 0693 Belgium 0468 2988 1335 1955 2247 0159 3399 1339 0446 0631 0198 0685 Germany 0775 1203 0764 3827 3408 1244 5399 4250 3564 2089 0167 1298 1 Bank capital to assets is the ratio of bank capital and reserves to total assets Capital and reserves include funds contributed by owners retained earnings general and special reserves provisions and valuation adjustments Capital includes tier 1 capital paidup shares and common stock which is a common feature in all countries banking systems and total regulatory capital which includes several specified types of subordinated debt instruments that need not be repaid if the funds are required to maintain minimum capital levels these comprise tier 2 and tier 3 capital Total assets include all nonfinancial and financial assets Definition retrieved from httpdatabankworldbankorg on 10252015 2 Annual percentage growth rate of GDP per capita based on constant local currency Aggregates are based on constant 2005 US dollars GDP per capita is gross domestic product divided by midyear population GDP at purchasers prices is the sum of gross value added by all resident producers in the economy plus any product taxes and minus any subsidies not included in the value of the products It is calculated without making deductions for depreciation of fabricated assets or for depletion and degradation of natural resources Definition retrieved from httpdatabankworldbankorg on 10252015 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 9 Spain 1338 1402 1986 2428 1866 0484 4424 0446 0970 2152 0906 1860 Slovenia 2780 4285 3823 5320 6345 3137 8627 0781 0404 2844 1133 2524 Sweden 2005 3911 2408 4101 2641 1329 5989 5089 1892 1021 0428 1337 Portugal 1305 1568 0580 1370 2291 0055 3071 1852 1682 3638 1063 1472 United Kingdom 3816 1873 2103 2287 1760 1113 5032 1116 0854 0038 1020 1911 United States 1927 2830 2397 1681 0815 1230 3624 1678 0827 1542 1449 1631 Switzerland 0690 2138 2380 3362 3214 0986 3345 1887 0679 0041 0750 NA Source World development indicators World bank data retrieved 10252015 3 Literature Review This section is subdivided into two subsections The first reports on previous studies that used financial criteria to emulate rating systems and compute efficiency scores The second presents the findings of the previous studies that focused on the impact of exogenous contextual variables on efficiency scores The major research proposition is presented at the end of this section 31 Financial Indicators and Banking Performance Considering that the different efficiency measurement methods as discussed in the introduction can provide efficiency scores it is crucial to establish the linkages between banking efficiency and financial indicators Wanke Barros Faria 2015 Indeed these methods should be capable of indicating how effective a bank is in minimizing financial indicators related to decreasing performance and in maximizing others related increasing performance Tsai Chang 2010 This finetuning between efficiency scores and financial indicators is often accomplished by selecting a suitable set of criteria and the attendant impacts positive or negative on banking performance Several financial criteria or indicators are thought to be associated with banking performance For instance the prediction of failures poor performance using firmspecific characteristics and financial indicators dates back to the seminal works of Altman 1968 and Altman Haldeman and Narayanan 1977 which applied discriminant analysis to financial ratios in order to derive the Zscore approach More recently Männasoo and Mayes 2009 presented a comprehensive literature review on this subject According to these authors although there is no universal set of indicators across previous studies the ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 10 CAMELS criteria appear to have a significant capacity for detecting distress CAMELS stands for capital adequacy C asset quality A management efficiency M earnings E liquidity L and sensitivity to market risk S In recent decades several studies reported on the use of the CAMELS variables in risk measurement and monitoring Examples can be found in Cole and Gunther 1995a DeYoung 1998 Oshinsky and Olin 2006 Kumar and Ravi 2007 Poghosyan and Cihak 2011 and Ravisankar Ravi and Bose 2010 More recently Wanke Barros and Faria 2015 presented a practical application that emulated the CAMELS rating systems in the Brazilian banking industry using DEA dynamic slacks The fundamental ideas behind this practical application are embedded in the close relationship between efficiency levels and distance to the frontier of best practices the latter may be considered as proxies of looming financial distress In other words a consistent movement towards lower efficiency levels over the course of time may constitute a leading distress indicator It should be noted however that the original criteria used to determine the CAMELS ratings are not disclosed to the general public Jin Kanagaretnam Lobo 2011 Therefore proxies are often selected accordingly based both on prior studies and availability of data Table 2 contains a list of the major financial criteria or indicators commonly used to emulate the CAMELS rating system as found in recent papers Table 2 CAMELS sub criteria commonly used Betz Oprică Peltonen and Sarlin 2014 Maghyereh and Awartani 2014 Wang Lu and Wang 2013 Wang Lu and Lin 2012 Doumpos and Zopounidis 2010 Secme Bayrakdaroglu and Kahraman 2009 Zhao Sinha and Ge 2009 Cole and Gunther 1995 Cole and Gunther 1998 Capital adequacy ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 11 Total Regulatory Capital Ratio Equitiestotal assets Equities Fixed assetsTotal assets BASEL ratio Assets Quality Loan Loss Res Gross Loans Loan Loss Provision Net Interest Rev Loan Loss Res Impaired Loans Impaired Loans Gross Loans NCO Average Gross Loans Impaired Loans Equity Tier 1 Ratio Management Net Interest Margin Net Interest Revenue Average Assets Other Operating Income Average Assets NonInterest Expense Average Assets Earning quality Return on Average Assets ROAA Return on Average Equity ROAE NonOperating Items Net Income Cost to Income Ratio Liquidity Interbank ratio Net Loans Total Assets Net Loans Deposits ST Funding Liquid Assets Total Deposit Borrowings Sensitivity of Market risk Rate sensitive assetsrate sensitive liabilities Average earning asset Rate sensitive assetsrate sensitive liabilities Risk weighted asset II Risk weighted asset I II Share of trading income ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 12 32The impact of exogenous contextual variables on efficiency scores Exogenous contextual variables are thought to affect banking efficiency in a unidirectional way and therefore are directly linked to policy formulation Put differently exogenous contextual variables affect efficiency levels without being affected by them As a matter of fact exogenous contextual variables represent decision variables based on the banking discretion rather than endogenous variables generated within the ambit of an efficiency model or a production process such as the country effect discussed in Section 2 Past studies in banking have explored these exogenous contextual variables and their impacts on efficiency levels Assaf Barros Matousek 2011 Assaf Barros Ibiowie 2012 Assaf Matousek Tsionas 2013 We next discuss the impacts of these exogenous contextual variables on banking efficiency Here the comprehensive theoretical framework presented in Barros and Wanke 2014 and Wanke et al 2015 is adopted Broadly speaking merger and acquisition processes in the financial sector are well known for leveraging banks financial and operational indicators The same basic economic principles apply to smaller banks in emerging countries which are the first to suffer the consequences of systemic financial crises On the other hand private instead of public ownership may be accountable for differences in production technology and managerial style thus affecting efficiency levels More specifically public banks are common in some countries and whether they are more efficient than private banks is a matter of debate Dinc 2005 Galindo and Micco 2004 Although they are not usually tested in efficiency analyses in the OECD context public banks are representative justifying therefore the testing of whether or not they are less efficient than private banks This hypothesis was previously examined by Kumar and Gulati 2010 and Mohan and Ray 2004 On the other hand mergers and acquisitions between banks are known as horizontal mergers Andrade et al 2001 and aim to improve cost performance and synergy through a larger market share In the former case the merged ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 13 companies reduce operating costs but retain the facilities of the merged or acquired company Garette and Dussauge 2000 This discussion leads us to the major proposition of this study efficiency levels in the OECD banks are significantly affected by endogenous contextual variables related to the bank origin country effect which may reflect regulatory issues cultural barriers and economic dynamics as discussed in Section 2 and by exogenous contextual variables that is subject to managerial discretion such as ownership publicprivate and MAs In this regard in examining the US banking industry AlSharkas Hassan and Lawrence 2008 found that merging has a positive influence on banks cost and profit efficiency They posited the technological advantage of allocative efficiency as the leading driver of bank mergers in the US Moreover Pancurova and Lyocsa 2013 examined the banking sector of 11 Central and Eastern European Countries in order to find the determinants bank efficiency Their study revealed that both bank ownership and country of origin have significant influence on bank efficiency performance A similar study by Sufian and Kamarudin 2015 examined determinants of revenue efficiency of Islamic banks among the Southeast Asian countries and showed that bank ownership has significant influence on banks revenue efficiency An analogous result is also available in some influential studies including Lee Kim 2013 and Minh Long Hung 2013 4 Methodology This section describes the building blocks upon which the robust TOPSIS analysis is developed in this research The fundamentals of the TOPSIS method are presented in Section 41 then a robust approach for criteria weighting in TOPSIS is presented in Section 42 using the Ng model 2007 as cornerstone Section 43 discusses why data reduction techniques such as PCA and ICA are deemed necessary to extract the major factors from a set of several financial indicators so that collinearity can be properly handled in TOPSIS Lastly the fundamentals of censored quantile regressions and the underlying role of bootstrapped truncated regressions similar to those applied to traditional DEA models are discussed in Section 44 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 14 41 TOPSIS The design and application of Multi Criteria Decision Making MCDM methods has been a prolific research field in business analytics and engineering As a matter of fact over the course of the last three decades a variety of different MCDM have been developed and empirically employed in field research This list includes the AHP Saaty 1980 Ramanathan 2013 Promethee Brans Vincke 1985 Corrente Greco Slowinski 2013 Electre HatamiMarbini Tavana 2011 Corrente et al 2013 Dematel Tsay Chou Hsu 2009 Vikor Opricovic 1998 Opricovic Tzeng 2007 and Uta Siskos Askounis Psarras 2014 TOPSIS is one of the MCDM approaches used when the weighting criteria for choosing the most adequate method is exogenously defined Wang Nistor Murty Wei 2014 BilbaoTerol et al 2014 Feng Wang 2001 Pirdavani Brijs Geert 2010 Chen Leng Mao Liu 2014 Hassan Hawas Ahmed 2013 Barros Wanke 2015 Its basic principle assumes that the chosen alternative should simultaneously be closest to the positiveideal solution and farthest from the negativeideal solution Hwang Yoon 1981 Ertugrul Karakasoglu 2009 More precisely the positiveideal solution is the one that maximizes the benefit and simultaneously minimizes the total costs while the negativeideal solution is the one that minimizes the benefit and simultaneously maximizes the cost Lai Liu Hwang 1994 Wu Lin Lee 2010 In spite of TOPSIS general resemblance to DEA objectives where outputs may be maximized and or inputs minimized in nonradial radial models the determination of the weights relative to the importance of each criteria is a key step in TOPSIS In DEA in contrast these weights are defined within the ambit of the model As a matter of fact while DEA models rely on the optimization of the distance from each firm to the convexefficient production frontier by finding a proper set of weights which are endogenously defined TOPSIS employs purely analytical methods of Euclidean distance functions on normalized vectors of positive outputs and negative inputs criteria given that the weights have been previously defined by the decisionmaker Barros Wanke 2015 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 15 The TOPSIS technique is built on an evaluation matrix consisting of m alternatives and n criteria with the intersection of each alternative and criteria given as Therefore one obtains a matrix This matrix should be firstly normalized from a regulated matrix as demonstrated in Eq 1 1 After normalization the weighted normalized decision matrix for efficiency assessment should be calculated observing Eq 2 2 where is the weight given to the criteria and Once the weighting criteria are defined the worst alternative the negative ideal assessment unit and the best alternative the positive ideal assessment unit should be defined using Eqs 3 and 4 3 4 where and which are a set of positive benefit and negative cost attributes respectively Given the best and the worst alternatives the distance between the target alternative i and the worst condition should calculated using Eq 5 5 and the distance between the alternative i and the best condition Ab by observing Eq 6 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 16 6 where dia and dib are the Euclidean distance from the target alternative i to the worst and best conditions respectively Then the similarity of alternative i to the worst condition the inefficient best conditions should be computed respectively as 7 where 0 Si 1 i 12 m if and only if the alternative solution has the worst condition if and only if the alternative solution has the best condition According to Jahanshahloo Lotfi and Davoodi 2009 represents the efficiency scores for each alternative that is for each bank of the sample over the course of the years as determined by the decision making criteria Lastly the alternatives should be ranked according to a higher value of indicates a better solution with respect to higher efficiency levels within the environment of 128 OECD banks thereby enabling the subsequent assessment of the impact of the contextual variables 42 Robust TOPSIS weighting Applying the Ng model With respect to the definition of the weighing criteria several different methods can be found in literature thus there is not a single methodological procedure to be followed Madeira Junior Cardoso Junior Belderrain Correia Schwanz 2012 Hemmati et al 2013 for instance used the Entropy technique Seçme et al 2009 and Shaverdi et al 2011 in contrast used fuzzy AHP for determining the weights of main and subcriteria In our study the 21 financial indicators or criteria presented in Table 3 were initially given the same weight since the literature review provides no clear indication as to which criteria should be prioritized for ranking banks in the OECD context ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 17 Importantly this study conducts a robustness analysis based on the guidelines provided in Madeira Junior et al 2012 The underlying idea is to assess how different weighting criteria would impact the computation of TOPSIS scores compared to those derived from the initial ones where the equal weighting assumption was considered The steps taken in generating different weightings are further detailed in the next paragraphs and presented in the discussion section Readers should note however that despite the weighting criteria adopted the contextual variables are further used as performance predictors in the Censored Quantile Regression CQR analysis for the base case where all weights are assumed to be equal The Ng weighting model applied to this research assumes that there are I bank observations and that they should be ranked in terms of J criteria or financial indicators Further let the performance of i th bank observation in terms of each of the j th criteria or financial indicator be denoted as ij y The purpose is to aggregate multiple performance scores of a bank observation with respect to different criteria or financial indicators into a single score for the subsequent ranking In the Ngmodel the author first transforms all measures to a baseline for purposes of comparison Using the transformation min max min ij ij ij ij y y y y 8 The Ngmodel converts all measurement in a 01 scale for all items To facilitate the bank ranking under multiple criteria Ng defines a nonnegative weight w ij which is the weight of contribution of performance of the i th item under the j th criteria to the score of the item The criteria are assumed to be ranked in a descending order such that 1 2 i i iJ w w w for all item i The Ng 2007 for aggregation purposes is given as follows max 1 J i ij ij j S y w st 1 1 J ij j w 9 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 18 1 2 i i iJ w w w ij 0 w 1 1 i I j J 43 Techniques for Factor Extraction Efficiency measurement techniques often employ some kind of weighting on the inputs and outputs of each observation This being the case whenever inputs and outputs are strongly correlated the accuracy of the scores computed for each observation may be biased Madeira Junior et al 2012 Another frequent problem found in some efficiency measurement techniques resides on their reduced discriminatory capability that is sometimes efficiency scores are concentrated towards one Bian 2012 Jenkins Anderson 2003 To mitigate these problems a comparative PCAICA factor extraction is proposed in this research within the ambit of the robust TOPSIS Kao Lu and Chiu 2011 verified that ICA stands next to PCA in terms of effectiveness as a solution to the discrimination problem of the DEA model Pioneering works that applied factor extraction and efficiency measurement techniques include Adler and Golany 2001 2002 and Adler and Yazhemsky 2010 who used PCA to increase the discriminatory capability of DEA scores and therefore reduce the so called curse of dimensionality Jenkins Anderson 2003 Recent applications of PCA in conjunction with TOPSIS can be found for instance in Wanke Barros and Macanda 2015 It is worth mentioning within the realm of the innovative applications of this research that ICA is a statistical signal process technique for finding independent sources given only observed data that are mixtures of the unknown sources without any prior knowledge of the mixing mechanisms Lee 1998 Hyvarinen Karhunen Oja 2001 The aim is to extract the hidden information from the observed multivariate statistical data where no relevant signal mixture mechanisms are available This hidden information is called the independent components ICs of the observable data The ICs are properly selected based on their statistical independency and are regarded as the key factors affecting the efficiency measurement results Shi Liu Sun 2006 Zheng Huang ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 19 Shang 2006 There are few papers that integrate ICA with efficiency measurement techniques such as DEA Kao et al 2011 So far to the best of our knowledge ICA has not been used in conjunction with TOPSIS in previous research The estimated ICs are regarded as more meaningful variables representing the independent sources or hidden information of the inputoutput variables These issues are further explored below 431 Principal Component Analysis PCA PCA explains the variance structure of a data matrix through linear combinations of variables consequently reducing the data to a few principal components which generally account for 8090 of data variance If most of the population variance can be attributed to the first few components then they can replace the original variables with minimum loss of information Per Hair Anderson Tatham and Black 1995 let the random vector ie the original inputsoutputs chosen to be aggregated have the correlation matrix with eigenvalues and normalized eigenvectors Consider the linear combinations where the superscript t represents the transpose operator 10 11 12 The principal components PCs are the uncorrelated linear combinations ranked by their variances in descending order The PCA used here is based on correlation rather than on covariance because the variables used in TOPSIS are often quantified in different units of measure In general the inputs and outputs of a TOPSIS need to be strictly positive however PCs can contain negative values Readers should note that PCA identify underlying variables that are uncorrelated with each other Intuitively this is desirable because the underlying variables that account for a set of measured variables should correspond to physically different processes which ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 20 in turn should have outputs that are uncorrelated with each other However specifying that the underlying variables should be uncorrelated imposes quite weak constraints on the form these variables take It is the weakness of these constraints that ensures the factors extracted can be rotated in order to find a more interpretable set of factors without affecting the zero correlations between factors Most factor rotation methods yield a statistically equivalent set of uncorrelated factors The variety of factor rotation methods available is regarded with relative skepticism by some researchers This is because it is sometimes possible to use factor rotation to obtain new factors that while easily interpreted are statistically no more significant than results obtained with other factor rotations By contrast any attempt to rotate the factors independent components extracted by ICA would yield nonindependent factors Thus the independent components of ICA do not permit postICA rotations because such factors are statistically independent as further detailed in the next section 432 Independent Component Analysis ICA is based on the assumption that source signals are not only uncorrelated but also statistically independent Essentially if two variables are independent then the value of one variable provides absolutely no information about the value of the other By contrast even though two variables are uncorrelated the value of one variable can still provide information about the value of the other ICA seeks a set of statistically independent signals amongst a set of signal mixtures on the assumption that such statistically independent signals are derived from different physical processes The objective of finding such a set of statistically independent signals is achieved by maximizing a measure of the joint entropy of the extracted signals In the basic ICA model the observed random variables are generally assumed to be a linear mixture of the unknown sources from a mixing matrix It can be expressed as Hyvarinen et al 2001 13 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 21 where is the vector of observed random variables is the vector of the unknown latent variables meaning that they cannot be directly observed from the mixture variables is an unknown mixing matrix of size n n The problem is then to estimate both the mixing matrix A and latent variables b using observations of z alone In order to estimate the unknown latent variables b the de mixing matrix W is used to transform the observed random variables z to yield the independent components ICs The independent components g are used as the estimates of the latent variables b The ICA modeling is formulated as an optimization problem by setting up the measure of independence of ICs as an objective function and using some optimization techniques to solve for the demixing matrix W There exist several algorithms performing ICA Bell Sejnowski 1995 Hyvarinen Oja 2000 In general the ICs are obtained by using the demixing matrix W to multiply the mixture variables The de mixing matrix W can be determined using an unsupervised learning algorithm with the objective of maximizing the statistical independence of ICs The ICs with nonGaussian distributions imply the statistical independence Bell Sejnowsky 1995 Hyvarinen Oja 2000 The nonGaussianity of the ICs can be measured by the negentropy Hyvarinen et al 2001 14 where is a Gaussian random vector of the same covariance matrix as g H is the entropy of a random vector g with density pg η defined as 15 The negentropy is always nonnegative and is zero if and only if g has a Gaussian distribution Since the problem in using negentropy is computationally difficult an approximation of negentropy is proposed as follows Hyvarinen Oja 2000 16 where o is a Gaussian variable of zero mean and unit variance and g is a random variable of zero mean and unit variance F is a nonquadratic function and is given by ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 22 in this study The FastICA algorithm proposed by Hyvärinen and Oja 2001 is adopted in this paper to solve for the demixing matrix W and independent components 44 Censored Quantile Regression Traditional regression models cannot answer an important question do contextual variables influence efficiency levels differently for low efficiency OECD banks than for those with high or average efficiency levels A more comprehensive picture of the effect of the contextual variables on the efficiency levels can be obtained via CQR Quantile regression models the relation between a set of contextual variables and specific percentiles or quantiles of the response variable it specifies changes in the quantiles of the efficiency For example a median regression 50th percentile of efficiency on a banks contextual variables specifies the changes in the median efficiency as a function of these predictors In this sense the effect of private ownership on median efficiency levels can be compared to its effect on other quantiles of efficiency in OECD banks As a matter of fact the quantile regression parameter estimates the change in a specified quantile of the response variable produced by a one unit change in the predictor variable This allows one to compare how some percentiles of the efficiency levels may be more affected by certain bank characteristics or contextual variables than other percentiles This is reflected in the change in the size of the regression coefficient According to Leng and Tong 2013 the quantile regression was introduced by Koenker and Bassett 1978 and has become an increasingly important tool in statistical analysis Indeed their general quantile regression QR estimation became the most popular approach Chernozhukov Hong 2002 Contrary to the usual model for the conditional mean it provides distributional information on the dependence of T on Z The Ƭth conditional quantile function of the dependent variable T given covariates Z QT ƬZ is defined as QT ƬZ infv F0vZ Ƭ where F0 is the cumulative conditional distribution function of T given Z Correspondingly a quantile regression model for QT ƬZ with Ƭ 0 1 can be denoted as QT ƬZ Z 15 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 23 When data are subject to censoring statistical estimation and inference for quantile regression is more involved Indeed a naive procedure which completely ignores censoring may give highly biased estimates Koenker dOrey 1987 Equivariance to monotone transformations is an important property of quantile regression models Powell 1986 Powell 1986 1984 first studied Censored Quantile Regression CQR with fixed censoring For random censoring Ying Jung and Wei 1995 proposed a semiparametric median regression model Despite the simplicity of their method this procedure requires the unconditional independence of the survival time and censoring time This assumption is often restrictive as conditional independence given the covariates is more natural Kalbfleisch Prentice 2002 In addition the estimating equation approach proposed by Ying et al 1995 involves solving nonmonotone discrete equations creating optimization difficulty As a consequence inferential procedures such as the resampling approach by Jin Ying and Wei 2001 or the bootstrap method can be computationally unattainable Chernozhukov and Hong 2002 argue that the CQR models allow covariates to shift location scale and the entire shape of the distribution and permit distributionfree specifications As such CQR models compare favorably to the normal AmemiyaTobin Cox BuckleyJames and other approaches According to the authors in this model the latent variable is left censored by the observable possibly random censoring points Ci and we observe Yi Ci Xi Ci σi 1 Yi Ci 16 is assumed to be conditionally independent of the censoring point Ci that is for all y IR P Y yXi Ci PY yXi so that TZ Z 17 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 24 Conditioning on Ci equation 5 and the equivariance transformation yield the following CQR model Powell 1986 QYi Xi Ci TZ Z Ci 18 More recently Bose and Chatterjee 2003 and Portnoy 2013 offered alternatives for resampling xy pairs based on either conventional multinomial resampling or using exponential weighting and on a deleted jackknife respectively Similarly to what is done with bootstrapped truncated regressions for DEA scores please refer to Wanke et al 2015 for a review on the subject the bootstrapping approach is adopted here uses the original TOPSIS scores within the ambit of censored quantile regressions Resamples with a size of 200 replications are adopted here Readers should refer to Koenker 2008 for more details on how implementing this procedure in R 5 The Data The data on OCDE banks was obtained from BankScope database3 As previously mentioned the TOPSIS alternatives consisted of each one of the 1280 samples formed by the combination of 128 banks over 10 years The underlying idea is to develop a model for assessing banking performance in OECD countries based on the criteria or financial indicators that could emulate the CAMELS rating system The model is based on the premise that banking performance should be viewed as a result of capital adequacy C assets quality A management M earning quality E liquidity L and sensitivity to market risk R The criteria observed the financial indicators commonly found in the literature review presented in Section 31 and data availability Financial indicators have been grouped observing the major CAMELS criteria as presented in Table 3 which also gives their major descriptive statistics and their expected impact on efficiency levels Negative impacts represent criteria that should be minimized while positive impacts represent 3 Available at httpsbankscopebvdinfocomversion0151021homeservproductscope2006 retrieved 05282015 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 25 criteria that should be maximized when deriving the ideal and antiideal solutions Figure 3 depicts the correlations between the financial ratios used to emulate the CAMELS rating system as presented in Table 3 The data source also provided information on contextual variables for each bank such as ownership public or private country of origin and whether the bank underwent an MA process All variables were coded as dummies Table 3 Descriptive statistics for the TOPSIS criteria and the contextual variables TOPSIS Criteria Subcriteria Impact Min Max Mean SD C Total Capital Ratio 9860 223800 16455 13687 Equity Tot Assets 9588 99461 14718 19255 Equity Net Loans 22116 847475 29831 70886 Equity Cust Short Term Funding 10298 965997 41039 117182 Equity Liabilities 9284 984648 31385 100536 A Loan Loss Res Gross Loans 0005 44915 3171 4189 Loan Loss Prov Net Int Rev 959933 580576 23865 60821 NCO Net Inc Bef Ln Lss Prov 682609 612582 21479 105358 Tier 1 Ratio 9860 223000 12767 9938 M Net Interest Margin 18003 71334 3447 7025 Net Int Rev Avg Assets 2744 67069 3130 6274 Oth Op Inc Avg Assets 30116 92011 3601 10480 Non Int Exp Avg Assets 0151 73629 5087 8324 E Return on Avg Assets ROAA 32517 31553 1163 4396 Return on Avg Equity ROAE 992293 185708 3773 47291 Cost to Income Ratio 2585 934541 72592 57085 Recurring Earning Power 25896 39906 2284 5285 L Net Loans Tot Assets 0466 94423 59299 21614 Net Loans Dep ST Funding 3354 974381 99286 72694 Liquid Assets Dep ST Funding 0852 997421 48584 118879 S Net Income Risk Weighted Assets 30470 17420 0507 2504 Contextual Variables Trend 1000 10000 5500 2873 Private bank 1yes0no 0000 1000 0327 0469 Merger and acquisition 1yes0no 0000 1000 0992 0088 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 26 Legend C1 Total Capital Ratio C2 Equity Tot Assets C3 Equity Net Loans C4 Equity Cust Short Term Funding C5 Equity Liabilities A1 Loan Loss Res Gross Loans A2 Loan Loss Prov Net Int Rev A3 NCO Net Inc Bef Ln Lss Prov A4 Tier 1 Ratio M1 Net Interest Margin M2 Net Int Rev Avg Assets M3 Oth Op Inc Avg Assets M4 Non Int Exp Avg Assets E1 Return On Avg Assets ROAA E2 Return On Avg Equity ROAE E3 Cost To Income Ratio E4 Recurring Earning Power L1 Net Loans Tot Assets L2 Net Loans Dep ST Funding L3 Liquid Assets Dep ST Funding S1 Net Income Risk Weighted Assets Figure 3 Graphic correlogram for the financial ratios 6 Results and Discussion The efficiency levels calculated for 128 selected OECD banks from 2004 to 2013 using the TOPSIS approach and considering different grouping criteria are given in Figs 2 3 and 4 The plot of the kernel density of the TOPSIS efficiency scores not only reflects the fact that these scores are censored at 0 and 1 but also suggests a Gaussian shape around C1 C2 C3 C4 C5 A1 A2 A3 A4 M1 M2 M3 M4 E1 E2 E3 E4 L1 L2 L3 S1 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 27 the mean see Figure 4 left especially when looking at the scores computed from PCA and ICA factor extraction As a matter of fact the KullbackLeibler KL divergence Cobb 2013 Langseth Nielsen PerezBernabe Salmeron 2014 between the inverse cumulative distribution function of the standardized TOPSIS scores obtained from the original set of criteria and the standard Gaussian is 0295 thus suggesting a considerable overlap between both distributions see Figure 4 right A KL divergence value of 0 indicates both distributions are equal Although the median efficiency levels are quite stable over the analyzed period the KruskalWallis test for differences between medians showed no significant results there are substantial differences when efficiency levels are grouped by country see Figures 5 and 6 For instance efficiency tends to be highly dispersed in emerging countries such as Greece Mexico Slovenia and South Korea and higher and less dispersed in developed countries such as the US Germany and the UK A possible explanation for this phenomena may be found in Ebrahim Girma Shah and Williams 2014 who examined the perspective of the Brazil Turkey India and China group Since banking in emerging markets is striving to create premium value for their customers banks tend to have substantial cash to acquire other institutions and invest in other possible growth opportunities Similar results are obtained by Faria and McAdam 2015 who examined growth under regime shift in Arab economies and their macroeconomic stabilization Although this result cannot be taken as a general rule French and Italian banks present high efficiency dispersion distinct from German and British ones the eventual impact of contextual variables may be embedded within these grouping schemes As a matter of fact differences between countries may involve both differences in regulatory standards in finance as well as diverse cultural aspects with respect to business practices ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 28 Figure 4 Kernel densities of the TOPSIS efficiency scores left and the inverse cumulative distributions for the standardized TOPSIS scores and for the standard Gaussian Figure 5 Efficiency levels grouped by year ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 29 Figure 6 Efficiency levels grouped by country ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 30 In regards to the factor extraction Figure 3 reveals that the criteria used to emulate the CAMELS system are strongly correlated Considering that correlations are not taken into account in several efficiency measurement techniques including TOPSIS factor extraction techniques such as PCA and ICA should be alternatively performed on the data as suggested for instance in Madeira Junior et al 2012 Adler and Yazhemsky 2010 and Kao et al 2011 This would allow the extraction of factors that are not only correlated with each other in the case of PCA results but also statistically independent from each other in the case of ICA results Therefore factor extraction was conducted on each of the CAMELS criteria using the PCA with a Varimax standardized rotation for data collected from the 128 OECD banks According to Tabachnick and Fidell 2001 only factor loads greater than 050 deserve to be interpreted and in these cases the variable represents a good factor measure Results presented in Table 4 cover only autovalues greater than 1 With the exception of the S criteria formed by a single variable for the sake of simplicity two main factors were extracted for the other criteria In a similar way factor extraction was conducted with ICA and their results are presented in Table 5 In order to select the most meaningful ICs their statistical independency is evaluated by computing kurtosis values Note that kurtosis is one of the methods for important ICs selection Hyvarinen et al 2001 Kao et al 2011 suggested that ICs with positive kurtosis values higher than three should be selected as the candidate variables ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 31 Table 4 Factor extraction from PCA C Capital Adequacy Variables Factors A Asset Quality Variables Factors L Liquidity Variables Factors PC1 PC2 PC1 PC2 PC1 PC2 Total Capital Ratio 024 068 Loan Loss Res Gross Loans 084 007 Net Loans Tot Assets 086 023 Equity Tot Assets 046 002 Loan Loss Prov Net Int Rev 083 012 Net Loans Dep ST Funding 001 097 Equity Net Loans 001 048 NCO Net Inc Bef Ln Lss Prov 013 071 Liquid Assets Dep ST Funding 084 029 Equity Cust Short Term Funding 019 026 Tier 1 Ratio 009 071 KMO Measure of Sampling Adequacy 0422 Equity Liabilities 054 025 KMO Measure of Sampling Adequacy 049 KMO Measure of Sampling Adequacy 0633 Bartletts Test of Sphericity Approx ChiSquare 377145 Bartletts Test of Sphericity Approx Chi Square 2212387 Bartletts Test of Sphericity Approx Chi Square 16605 df 3 df 6 Sig 0 df 10 Sig 0 Percent of variance explained by the factors 84 Sig 0 Percent of variance explained by the factors 60 Percent of variance explained by the factors 67 M Management Capacity Variables Factors E Earning Quality Variables Factors PC1 PC2 PC1 PC2 Net Interest Margin 099 006 Return on Avg Assets ROAA 079 051 Net Int Rev Avg Assets 099 006 Return on Avg Equity ROAE 008 095 Oth Op Inc Avg Assets 015 097 Cost to Income Ratio 072 009 Non Int Exp Avg Assets 036 091 Recurring Earning Power 086 024 KMO Measure of Sampling Adequacy 0503 KMO Measure of Sampling Adequacy 0505 Bartletts Test of Sphericity Approx Chi Square 76981 Bartletts Test of Sphericity Approx Chi Square 2285119 df 6 df 6 Sig 0 Sig 0 Percent of variance explained by the factors 98 Percent of variance explained by the factors 78 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 32 Table 5 Factor extraction from ICA C Capital Adequacy Variables Factors A Asset Quality Variables Factors L Liquidity Variables Factors IC1 IC2 IC1 IC2 IC1 IC2 Total Capital Ratio 000016 000005 Loan Loss Res Gross Loans 000001 000044 Net Loans Tot Assets 0000652 0001039 Equity Tot Assets 000078 000087 Loan Loss Prov Net Int Rev 000096 001653 Net Loans Dep ST Funding 0001333 0013856 Equity Net Loans 000227 000069 NCO Net Inc Bef Ln Lss Prov 000941 000169 Liquid Assets Dep ST Funding 0008183 0002174 Equity Cust Short Term Funding 000628 000545 Tier 1 Ratio 000001 000000 Kurtosis value 65303 28689 Equity Liabilities 000369 000952 Kurtosis value 16853 88823 Kurtosis value 53227 27609 M Management Capacity Variables Factors E Earning Quality Variables Factors IC1 IC2 IC1 IC2 Net Interest Margin 001021 006958 Return on Avg Assets ROAA 000061 000062 Net Int Rev Avg Assets 000925 006216 Return on Avg Equity ROAE 000609 002054 Oth Op Inc Avg Assets 005963 003564 Cost to Income Ratio 001591 000789 Non Int Exp Avg Assets 004834 001737 Recurring Earning Power 000055 000014 Kurtosis value 64363 34831 Kurtosis value 85933 80476 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 33 The computation of the Spearmans rank correlation between the TOPSIS scores calculated from the original inputs and outputs and those calculated from the factors extracted using PCA and ICA methods revealed significant adherence between each pair of results with a correlation score of 0321 p 00001 for the PCA extraction and 0370 p 00001 for the ICA extraction This indicates that both sets of TOPSIS scores computed from extracted factors are isotonic with respect to the original score set that is their trends are strictly increasing and the orders of distances between points in the embedding matches are low Kruskal 1964 Figure 7 depicts the results for the isotonic regression between both sets of TOPSIS scores in accordance with the fundamentals provided in De Leeuw Hornik and Mair 2009 Figure 7 Results for the isotonic regression Despite the correlation analysis and the isotonic regression between the original sets of TOPSIS scores and those computed from PCA and ICA factor extraction it is still necessary according to Madeira Junior et al 2012 to assess how different weighting criteria impact isotonicity considering equal weights as the base case In this research we use the weighted linear optimization model proposed by Ng 2007 and inspired in Pearman 1977 to generate various sets of weights for the inputs and the output extracted ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 34 factors before conducting the sensitivity analysis on TOPSIS scores with respect to the weighting criteria Under Ngs approach the weights are endogenous that is they are automatically generated when the model is optimized and are therefore not subjective Moreover Ngs approach can be used in a supplementary manner along with several MCDM techniques such as TOPSIS so that weights can be specified exogenously by decisionmakers regardless of some degree of subjectivity However as presented in Ng 2007 the decision maker has to rank the importance of criteria before solving the model We used combinatorial analysis to generate within the ambit of the R statistical package the full universe of combinations where the extracted factors are ranked in different orders of importance The model presented in Ng 2007 was run 9 362880 2 2 times for the full set of the extracted factors for their combinations in different orders of importance both for PCA and ICA extractions In every round the sum of the weights for the extracted factors was equal to one observing the constraints of the linear program Then as a final step this full set of extracted factors and their weight combination 92 725760 was used to recompute the TOPSIS scores Table 6 depicts the results for the mean TOPSIS criteria weights with respect to different groups of results in terms of the Spearmans rank correlations and their respective significances between the original TOPSIS scores and each one of the 725760 weight combination under the PCA and the ICA extractions Four groups were defined for each extraction Considering the PCA extraction in the first group the correlations are found to be positive and significant at 005 This group is not only the most numerous with 4139 of the total 725760 cases but also within this group it is possible to affirm that this set of weights is robustly represented by the original TOPSIS scores with equal weights On the other hand in the second group although correlations are still positive they are not significant With respect to the second largest group 4077 of cases it is possible to affirm that this set of weights is weakly represented by the original TOPSIS scores with equal weights Isotonicity still holds albeit weakly The third and the fourth groups however presented no isotonicity and corresponded to almost 18 of cases On the other hand considering the ICA extraction 100 of results fell within the first group where correlations are significant and isotonicity is high These results suggest that original TOPSIS scores are more robustly represented by those scores computed from the factors ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 35 derived by ICA than those obtained from the factors computed using PCA whenever a different set of weighting criteria is taken into consideration Readers should also observe that the aggregate weighing of factors obtained from each factor extraction differs substantially in response to a given factor extraction method Table 6 Mean TOPSIS criteria weights per group of result PCA TOPSIS Criteria Positive Correlation Negative Correlation Significant Not Significant Not Significant Significant Inputs C1 0090892 0042534 0069463 0119357 C2 0078247 0026512 0059668 0124385 A1 0103201 0257382 0078912 0127316 A2 0077841 0044819 0049765 0076716 M1 0074107 0055108 0036662 0042762 M2 0079608 0043263 005836 0086392 L1 0119384 0335251 0212249 0174819 L2 0076334 0012693 0250664 0065385 S1 0118568 0000621 0002439 0001049 Outputs E1 0051433 0057338 0047173 0035785 E2 0130385 012448 0134645 0146034 Number of Cases 300423 295902 41732 87703 of Cases 4139 4077 575 1208 ICA TOPSIS Criteria Positive Correlation Negative Correlation Significant Not Significant Not Significant Significant Inputs C1 0073393 NA NA NA C2 0061668 NA NA NA A1 0167604 NA NA NA A2 0062635 NA NA NA M1 0060395 NA NA NA M2 0064396 NA NA NA L1 0051711 NA NA NA L2 0130107 NA NA NA S1 02194 NA NA NA Outputs E1 0059089 NA NA NA E2 00496 NA NA NA Number of Cases 725760 0 0 0 of Cases 10000 000 000 000 NA not available ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 36 Results for the CQR of the efficiency scores on the contextual variables for selected efficiency percentiles are presented in Table 7 tau 020 040 060 and 080 Their significances are given in the Appendix The full set of results for the percentile range from 005 to 095 in intervals of 005 is depicted in Figure 8 Significances and bootstrapped lower and upper confidence intervals can also be found in the Appendix The standard errors and confidence limits for the CQR coefficient estimates were obtained with asymptotic and bootstrapping methods Both methods provide robust results Koenker Hallock 2001 though the bootstrap method is preferred for being more practical Hao Naiman 2007 As easily observed in Table 7 the signs and the magnitude of the relationships between the PCA extraction method and the efficiency levels remained quite unchanged for different quantiles These results indicate that efficiency levels computed from PCA factors are systematically higher than those computed from ICA extraction for different efficiency quantiles In conjunction with results presented in Table 6 these results also corroborate the robustness of ICA extraction over PCA extraction within the ambit of efficiency computation since their scores are lower and more discriminatory On the other hand for several countryrelated variables besides trend ownership and MA changes in magnitude and sign reversion were verified In fact this effect occurs because rather than predicting the mean of the dependent variable CQR examines the quantiles of the dependent variable By choosing tau 04 or 06 the 40th and 60th percentiles of the data are being used to compute the regression Therefore CQR can answer the question For which type of OECD bank high or low efficiency does the impact of a given contextual variable prevail Sign reversion may suggest that local cultural and regulatory aspects ie the specifics of each country may be impacting banking operations differently at distinct efficiency levels On the other hand sign reversions in trend ownership and MA suggest that the beneficial effects of privatization mergers and learning curves may or may not be present depending upon the banks efficiency level In other words in light of such exogenous contextual variables higher efficiency banks may be affected differently than lower efficiency ones ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 37 Table 7 Coefficients for the CQR for selected percentiles tau 02 tau 04 tau 06 tau 08 Intercept 4122926e01 04221409424 4175025e01 04244353720 Trend 2102790e03 00023308742 1588746e03 00011177511 Trend2 1336560e04 00001488903 9157415e05 00001904028 Private Bank 2119663e03 00008224629 9959784e04 00090363456 MA 9548725e03 00159760743 6810100e03 00041106571 Austria 7133545e03 00046903075 3914858e03 00021778313 Belgium 7263434e03 00069033255 7021548e03 00113366396 Denmark 1301956e03 00044493096 7326106e03 00083083090 Finland 1912597e06 00068622923 1405544e02 00151699111 France 1508672e03 00014877053 2344498e03 00010760138 Germany 2691382e04 00024007925 2696792e03 00061903361 Greece 1224619e03 00080806710 1306467e02 00402785660 Ireland 4015955e03 00053940041 2822797e02 00417585709 Italy 9326014e03 00045616133 2763637e03 00037587382 Luxembourg 2098646e02 00243525250 2958012e02 00470656408 Mexico 3675285e02 00329281231 2922368e02 00380989958 Netherlands 5443823e03 00010127235 1634528e03 00010748266 Norway 1194644e03 00005866565 2560771e03 00039022218 Poland 1865862e02 00168202125 1074817e02 00087141654 Portugal 4481474e03 00067766857 7775031e03 00289144122 S Korea 6186100e03 00068988726 1974462e02 00285323970 Slovenia 3640998e03 00042373254 5149724e03 00076564991 Spain 1218051e03 00027374663 4024112e03 00021278413 Sweden 1171713e03 00070868866 1003613e02 00081081652 Turkey 1693147e02 00137878866 7654560e03 00101859648 UK 1430675e02 00055746423 3122834e03 00018982412 USA 8695248e03 00118017731 1440764e02 00276463442 Method PCA 2214945e02 00221114694 2200379e02 00215869525 The CQR results presented in Table 7 indicate that the effect of private ownership is negative for lower efficiency banks and positive for higher efficiency banks thus suggesting that public OECD banks should be minimally competitive in terms of efficiency if willing to benefit from a privatization process In other words these banks should first ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 38 become more efficient then privatize and not vice versa Conversely the aggregate effects of trend are negative in higher quantiles of banking efficiency and positive in lower ones thus suggesting that less efficient banks take more advantage of learning curves than do more efficient banks Regarding MA the results presented in the Appendix indicate that this variable was not significant for different efficiency quantiles Figure 8 Censored Quantile Regression Coefficients Plots The solid blue line indicates the quantile regression point estimates the lighter blue region is a pointwise 95 confidence band Xaxis represents the quantiles and Yaxis represents the coefficients for each contextual variable It is important to notice in Figure 8 that CQR coefficients for the country endogenous variables cross the horizontal effect equals zero axis in some quantiles more than once Such crossovers are however sometimes quite plausible and an advantage of the CQR approach is that they are more easily revealed Koenker Geling 2001 In this research crossover occurs for different efficiency quantiles in different countries suggesting that cultural and regulatory aspects related to the country origin may be negative positive for high low efficiency banks and that their efficiency levels may even decrease increase possibly fostering a situation where decreasing increasing returns to scale prevails This appears to be the case of banks located in Germany Spain and Slovenia Greece USA and Ireland For several countries however no significant results were found for different quantile levels eg Portugal Norway Netherlands and France ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 39 7 Conclusions This article presented and developed a robust TOPSIS approach to assess the efficiency of OECD banks in light of different contextual variables related to merger and acquisition ownership publicprivate trend learning curve and country location Three distinctive aspects characterize the robust TOPSIS approach used here i the use of data reduction techniques to handle collinearity issues present in the financial indicators ii a comprehensive criteria weight testing based on linear programming and iii the use of bootstrapped censored quantile regressions to assess how efficiency drivers vary between highlow efficiency banks The methodology used in this research contributed to the efficiency literature in general not only by the enhancing the potential use of multicriteria decision making techniques such as TOPSIS but also by offering an alternative to the two major shortcomings of the state of the art inherent in twostage bootstrapped DEA models The first one is related to the fact that in DEA weights are endogenously defined within the ambit of the linear optimization model thus making it difficult for the analyst to perform comprehensive sensitivity analysis based on the inputoutput here called criteria of financial indicators weights The other one is the shorter computation time involved in computing TOPSIS scores compared to those verified when computing DEA scores especially under intensive sensitivity analysis of weights Results derived in this research also represent a contribution to the literature in banking efficiency They indicate that higher and lower efficiency banks require different courses of action highlighting the importance of market segmentation when analyzing performance Differently from traditional efficiency studies where size is considered the common ground for banking segmentation this research offers a perspective of segmentation based on efficiency quantiles Although the quantile results obtained for the less efficient banks are quite intuitive and corroborate previous literature positive impact of trend and negative impact of private ownership the paths for maintaining high efficiency levels in more efficient banks is not trivial ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 40 Additionally the results presented here suggest a number of policy implications for OECD banks implying different courses of action for banks with higher and lower efficiency levels Lower efficiency banks clearly benefit more from learning curves and are less impacted by privatization than higher efficiency banks although nothing conclusive was found with respect to mergers and acquisitions While less efficient banks should focus on ownership and continuous operational improvements more efficient banks should pay attention to the cultural and regulatory specifics of the country where they operate perhaps choosing efficiencyfriendly OECD countries for expanding their operations In this sense worth mentioning is the substantial endogenous impact of the country effect suggesting that cultural and regulatory issues can be as relevant as other exogenous factors in understanding efficiency Lower dispersion on efficiency results may suggest that banks in developed countries are regulated more tightly than those operating in developing ones Future research should target higher efficiency banks in OECD countries in order to carefully analyze the role of their cultural and regulatory specifics in efficiency levels Possible venues of future studies could also try to overcome some limitations of the current method such as the testing of an alternative set of financial criteria used to emulate the CAMELS rating system and that tends to be limited by the characteristics of the secondary database used Comparative research venues on DEA and TOPSIS scores could also be considered by authors in future studies References Adler N Golany B 2001 Evaluation of deregulated airline networks using data envelopment analysis combined with principle component analysis with an application to Western Europe European Journal of Operational Research 1322 260273 Adler N Golany B 2002 Including principle component weights to improve discrimination in data envelopment analysis Journal of the Operational Research Society 539 985991 Adler N Yazhemsky E 2010 Improving discrimination in data envelopment analysis PCADEA or variable reduction European Journal of Operational Research 202 1 273284 Albayrak E Erensal Y C 2005 A study bank selection decision in Turkey using the extended fuzzy AHP method Proceedings of the 35th International Conference on Computers and Industrial Engineering Istanbul Turkey AlSharkas A A Hassan M K Lawrence S 2008 The impact of mergers and acquisitions on the efficiency of the US banking industry Further evidence Journal of Business Finance Accounting 3512 5070 httpdxdoiorg101111j14685957200702059x Altman E I 1968 Financial ratios discriminant analysis and the prediction of corporate bankruptcy Journal of Finance 234 589609 doi101111j154062611968tb00843x Altman E I Haldeman R G Narayanan P 1977 ZETATM analysis a new model to identify bankruptcy risk of corporations Journal of Bank Finance 11 2954 doi1010160378426677900176 Andrade G Mitchell M Stafford E 2001 New evidence and perspectives on mergers Journal of Economic Perspectives ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 41 152 103120 Assaf AG Barros CP Ibiwoye A 2012 Performance assessment of Nigerian Banks prior and post consolidation Evidence from a Bayesian approach The services Industries Journal 322 215229 doi101080026420692010529135 Assaf AG Barros C P Josiassen A 2010 Hotel efficiency A bootstrapped metafrontier approach International Journal of Hospitality Management 293 468475 doi101016jijhm200910020 Assaf AG Barros C P Matousek R 2011 Technical efficiency in Saudi banks Expert Systems with Applications 385 57815786 doi101016jeswa201010054 Assaf A Barros CP and Matousek R 2011 Productivity and efficiency analysis of shinkin banks Evidence from bootstrap and Bayesian approaches Journal of Banking and Finance 35 331342 Assaf A G Matousek R Tsionas E G 2013 Turkish bank efficiency Bayesian estimation with undesirable outputs Journal of Banking Finance 372 506517 doi101016jjbankfin201209009 Athanasoglou P P Brissimis S N Delis M D 2008 Bankspecific industryspecific and macroeconomic determinants of bank profitability Journal of International Financial Markets Institutions and Money 182 121136 doihttpdxdoiorg101016jintfin200607001 Athanassopoulos A D 1998 Nonparametric frontier models for assessing the market and cost efficiency of largescale bank branch networks Journal of Money Credit and Banking 302 172192 Retrieved from Go to ISI000073439900003 httpwwwjstororgstablepdfplus2601208pdf Azam J P Biais B Dia M 2004 Privatisation versus regulation in developing economies The case of West African Banks Journal of African Economies 133 361394 doi101093jaeejh022 Barros CP Ferreira C Williams J 2007 Analysing the determinants of performance of best and worse European banks A mixed logit approach Journal of Banking and Finance 31 21892203 Barros C P Managi S Matousek R 2012 The technical efficiency of the Japanese banks Nonradial directional performance measurement with undesirable output Omega 401 18 doi101016jomega201102005 Barros C P Peypoch N 2009 An evaluation of European airlines operational performance International Journal of Production Economics 1222 525533 doi101016jijpe200904016 Barros C P Peypoch N Williams J 2010 A note on productivity change in European cooperative banks The Luenberger indicator approach International Review of Applied Economics 242 137147 doi10108002692171003590047 Barros CP and P Wanke 2014 Banking efficiency in Brazil Journal of International Financial Markets Institutions Money 28 5465 Barros CP Wanke P 2015 An analysis of African airlines efficiency with twostage TOPSIS and neural networks Journal of Air Transport Management httpdxdoiorg101016jjairtraman201503002 Bauer P W Berger A N Humphrey D B 1993 Efficiency and productivity growth in US banking In HO Fried CAK Lovell SS Schmidt Eds The measurement of productive efficiency Techniques and applications pp 386413 Oxford Oxford University Press Behzadian M Khanmohammadi Otaghsara S Yazdani M Ignatius J 2012 A stateof theart survey of TOPSIS applications Expert Systems with Applications 3917 1305113069 Bell JA Sejnowski TJ 1995 An InformationMaximization Approach to Blind Separation and Blind Deconvolution Neural Computation 7 11291159 Berger A N Hanweck GA Humphrey D B 1987 Competitive viability in banking Scale scope and product mix economies Journal of Monetary Economics 203 501520 doi1010160304393287900390 Berger A N Humphrey D B 1997 Efficiency of financial institutions international survey and directions for future research European Journal of Operational Research 982 175212 doi101016S0377221796003426 Berger A N Hasan I Zhou M 2009 Bank ownership and efficiency in China What will happen in the worlds largest nation Journal of Banking and Finance 331 113130 doi101016jjbankfin200705016 Betz F Oprică S Peltonen T A Sarlin P 2014 Predicting distress in European banks Journal of Banking Finance 45 225241 doi101016jjbankfin201311041 Bian Y 2012 A GramSchmidt process based approach for improving DEA discrimination in the presence of large dimensionality of data set Expert Systems with Applications 39 3 37933799 BilbaoTerol A Arenas Parra M Cañal Fernandes V 2014 Using TOPSIS for assessing the sustainability of government bond funds Omega 49 117 doi101016jomega201404005 Bose A and S Chatterjee 2003 Generalized bootstrap for estimators of minimizers of convex functions J Stat Planning and Inf 117 225239 Brandouy O Briec W Kerstens K Van de Woestyne I 2010 Portfolio performance gauging in discrete time using a Luenberger productivity indicator Journal of Banking Finance 348 18991910 doi101016jjbankfin200912015 Brans J P Vincke P H 1985 Note A preference ranking organization method Management Science 316 647656 doi101287mnsc316647 Briec W Liang Q L 2011 On some semilattice structures for production technologies European Journal of Operational Research 2153 740749 doi101016jejor201106012 Brissimis S N Delis M D Tsionas E G 2010 Technical and allocative efficiency in European banking European Journal of Operational Research 2041 153163 doi101016jejor200909034 Chen S Leng Y Mao B Liu S 2014 Integrated weightbased multicriteria evaluation on transfer in large transport terminals A case study of the Beijing South Railway Station Transportation Research Part A Policy and Practice 66 1326 Chen TY 2002 A comparison of chanceconstrained DEA and stochastic frontier analysis Bank efficiency in Taiwan Journal of the Operational Research Society 535 492500 doi101057palgravejors2601318 Chen X Skully M Brown K 2005 Banking efficiency in China Application of DEA to pre and postderegulations eras 19932000 China Economic Review 163 229245 doi101016jchieco200502001 Chernozhukov V Hong H 2002 Threestep censored quantile regression and extramarital affairs Journal of the American Statistical Association 97459 872882 Cobb B 2013 Mixture distributions for modelling demand during lead time Journal of the Operational Research Society 64 217 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 42 228 Cole R A Gunther J W 1995a A CAMEL ratings shelf life Federal Reserve Bank of Dallas Review 1320 doihttpdxdoiorg102139ssrn1293504 Cole R A Gunther J W 1995b Separating the likelihood and timing of bank failure Journal of Bank Finance 196 10731089 doi101016037842669598952M Cole R A Gunther J W 1998 Predicting bank failures A comparison of onand offsite monitoring systems Journal of Financial Services Research 132 103117 doihttpdxdoiorg101023A1007954718966 Corrente S Greco S Slowinski R 2013 Multiple criteria hierarchy process with ELECTRE and PROMETHEE Omega 415 820846 doi101016jomega201210009 Cummins J D RubioMisas M 2006 Deregulation consolidation and efficiency Evidence from the Spanish insurance industry Journal of Money Credit and Banking 382 323355 Retrieved from Go to ISI000236422700002 De Borger B Ferrier G D Kerstens K 1998 The choice of a technical efficiency measure on the free disposal hull reference technology A comparison using US banking data European Journal of Operational Research 1053 427446 doi101016S026322419800013X De Leeuw J K Hornik P Mair 2009 Isotone Optimization in R PoolAdjacentViolators Algorithm PAVA and Active Set Methods Journal of Statistical Software 325 124 Deliktaş E Balcilar M 2005 A Comparative Analysis of Productivity Growth CatchUp and Convergence in Transition Economies Emerging Markets Finance Trade 411 628 De Young R 1998 Management quality and Xinefficiency in national banks Journal of Financial Services Research 131 522 doi101023A1007965210067 Dima B Dincă M S Spulbăr C 2014 Financial nexus Efficiency and soundness in banking and capital markets Journal of International Money and Finance 47 100124 doihttpdxdoiorg101016jjimonfin201405002 Dinc S 2005 Politicians and Banks Political Influences on GovernmentOwned Banks in Emerging Countries Journal of Financial Economics Vol 77 No 2 pp 453479 Doumpos M Zopounidis C 2010 A multicriteria decision support system for bank rating Decision Support Systems 501 5563 doihttpdxdoiorg101016jdss201007002 Du K Sim N 2016 Mergers acquisitions and bank efficiency CrossCountry evidence from emerging markets Research in International Business and Finance 36 499510 doi101016jribaf201510005 Dudek A Jefmański B 2015Tthe Fuzzy TOPSIS method and its implementation in the r programme Business Informatics 135 Ebrahim M S Girma S Shah M E Williams J 2014 Rationalizing the value premium in emerging markets Journal of International Financial Markets Institutions and Money 29 5170 doi101016jintfin201311005 Ertugrul I Karakasoglu N 2009 Performance evaluation of Turkish cement firms with fuzzy analytic hierarchy process and TOPSIS methods Expert Systems with Applications 361 702715 doi101016jeswa200710014 Faria J R McAdam P 2015 Macroeconomic adjustment under regime change From social contract to Arab Spring Journal of International Money and Finance 56 122 doihttpdxdoiorg101016jjimonfin201504001 Feng CM Wang RT 2001 Considering the financial ratios on the performance evaluation of highway bus industry Transport Reviews 214 449467 Figueira C Nellis J Parker D 2006 Does Ownership affect the efficiency of African banks Journal of Developing Areas 401 3863 doi101353jda20070004 Frei F X Harker P T 1999 Measuring aggregate process performance using AHP European Journal of Operational Research 1162 436442 Fungáčová Z Poghosyan T 2011 Determinants of bank interest margins in Russia Does bank ownership matter Economic Systems 354 481495 doihttpdxdoiorg101016jecosys201011007 Galindo A and A Micco 2004 Do StateOwned Banks Promote Growth CrossCountry Evidence for Manufacturing Industries Economic Letters Vol 84 pp 371376 Gardener E Molyneux P Hoai N L 2011 Determinants of efficiency in South East Asian banking Service Industries Journal 3116 26932719 doihttpdxdoiorg101080026420692010512659 Garette B Dussauge P 2000 Alliances versus acquisitions Choosing the right option European Management Journal 181 6369 GarzaGarcia J G 2012 Determinants of bank efficiency in Mexico a twostage analysis Applied Economics Letters 1917 1679 1682 doihttpdxdoiorg101080135048512012665589 Gunay E 2012 Risk Incorporation and Efficiency in Emerging Market Banks During the Global Crisis Evidence from Turkey 20022009 Emerging Markets Finance Trade 485 91102 Hair JF Anderson RE Tatham RL Black WC 1995 Multivariate Data Analysis Englewood Cliffs NJ PrenticeHall Hao L Naiman D Q 2007 Quantile Regression Thousand Oaks Sage Publications Hassan M Hawas Y Ahmed K 2013 A multidimensional framework for evaluating the transit service performance Transportation Research Part A Policy and Practice 50 4761 HatamiMarbini A Tavana M 2011 An extension of the Electre 1 method for group decision making under a fuzzy environment Omega 394 37386 doi101016jomega201009001 Hemmati M Dalghandi S A Nazari H 2013 Measuring relative performance of banking industry using a DEA and TOPSIS Measurement Science Letters 32 499503 Hwang C L Yoon K 1981 Multiple attribute decision making methods and applications a state of the art survey Berlin New York SpringerVerlag Hyvarinen A Oja E 2000 Independent Component Analysis Algorithms and Applications Neural Computation 13 411430 Hyvarinen A Karhunen J Oja E 2001 Independent Component Analysis New York NY John Wiley and Sons İÇ Y T Yurdakul M 2010 Development of a quick credibility scoring decision support system using fuzzy TOPSIS Expert Systems with Applications 371 567574 doihttpdxdoiorg101016jeswa200905038 Ikhide S I 2008 Measuring the operational efficiency of commercial banks in Namibia South African Journal of Economics 764 586595 doi101111j18136982200800214x Jaffee D Levonian M 2001 The Structure of Banking Systems in Developed and Transition Economies European Financial ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 43 Management 72 161181 doi1011111468036X00151 Jahanshahloo GR Lotfi FH Davoodi AR 2009 Extension of TOPSIS for decisionmaking problems with interval data Interval efficiency Mathematical and Computer Modelling 4956 11371142 Jenkins L Anderson M 2003 A multivariate statistic approach to reducing the number of variables in data envelopment analysis European Journal of Operational Research 1471 5161 Jin JY Kanagaretnam K Lobo G J 2011 Ability of accounting and audit quality variables to predict bank failure during the financial crisis Journal of Banking Finance 3511 28112819 doi101016jjbankfin201103005 Jin Z Ying Z Wei LJ 2001 A simple resampling method by perturbing the minimand Biometrika 88 381390 Kalbfleisch J D and Prentice R L 2002 Analysis of Correlated Failure Time Data in The Statistical Analysis of Failure Time Data Second Edition John Wiley Sons Inc Hoboken NJ USA doi1010029781118032985ch10 Kao LJ Lu CJ Chiu CC 2011 Efficiency measurement using independent component analysis and data envelopment analysis European Journal of Operational Research 210 2 310317 Kebede H A Wassie B 2013 How efficient are the Ethiopian microfinance institutions in extending financial services to the poor A comparison with the commercial banks Journal of African Economies 221 112135 doi101093jaeejs012 Kerstens K Mounir A Van de Woestyne I 2011 Nonparametric frontier estimates of mutual fund performance using C and Lmoments some specification tests Journal of Banking Finance 355 11901201 doi101016jjbankfin201009030 Kirkpatrick C H Murinde V Tefula M 2008 The measurement and determinants of xinefficiency in commercial banks in SubSaharan Africa European Journal of Finance 147 625639 doi10108013518470701705769 Kiyota H 2009 Efficiency of Commercial Banks in SubSaharan Africa A Comparative Analysis of Domestic and Foreign Banks Proceedings of the CSAE Conference Oxford UK Koenker R Bassett G Jr 1978 Regression quantiles Econometrica 461 3350 Koenker R dOrey V 1987 Computing Regression Quantiles Applied Statistics 36 383393 Koenker R Geling O 2001 Reappraising medfly longevity A quantile regression survival analysis Journal of the American Statistical Association 96 458468 Koenker R Hallock K 2001 Quantile regression An introduction Journal of Economic Perspectives 15 143156 Koenker R 2008 Censored Quantile Regression Redux Journal of Statistical Software 27 125 Kontolaimou A Tsekouras K 2010 Are cooperatives the weakest link in European banking A nonparametric metafrontier approach Journal of Banking Finance 348 19461957 Kruskal J B 1964 Nonmetric Multidimensional Scaling A numerical method Psychometrika 292 115129 doi101007BF02289694 Kumar P R Ravi V 2007 Bankruptcy prediction in banks and firms via statistical and intelligent techniques a review European Journal of Operational Research 1801 128 doi101016jejor200608043 Kumar S Gulati R 2010 Measuring efficiency effectiveness and performance of Indian public sector banks International Journal of Productivity and Performance Management 591 5174 Kumbhakar SC Wang D 2007 Economic reforms efficiency and productivity of Chinese banking Journal of Regulatory Economics 322 105129 Lai Y J Liu T Y Hwang C L 1994 TOPSIS for MCDM European Journal of Operational Research 76 486500 doi1010160377221794902828 Lampe H W Hilgers D 2014 Trajectories of efficiency measurement A bibliometric analysis of DEA and SFA European Journal of Operational Research 2401 121 doi101016jejor201404041 Langseth H Nielsen TD PerezBernabe I Salmeron A 2014 Learning mixtures of truncated basis functions from data International Journal of Approximate Reasoning 55 940956 Lee J Y Kim D 2013 Bank performance and its determinants in Korea Japan and the World Economy 270 8394 doihttpdxdoiorg101016jjapwor201305001 Lee T W 1998 Independent Component Analysis Theory and Application Boston MA Kluwer Academic Publishers Lee CC Hsieh MF 2014 Bank reforms foreign ownership and financial stability Journal of International Money and Finance 400 204224 doihttpdxdoiorg101016jjimonfin201309001 Leng C Tong X 2013 A quantile regression estimator for censored data Bernoulli 191 344361 Levintal O 2013 The real effects of banking shocks Evidence from OECD countries Journal of International Money and Finance 32 556578 doihttpdxdoiorg101016jjimonfin201205028 Li S Liu F Liu S Whitmore G A 2001 Comparative performance of Chinese commercial banks findings and policy implications Review of Quantities Finance and Accounting 162 149170 Madeira Junior AG Cardoso Junior M Belderrain M N Correia AR Schwanz SH 2012 Multicriteria and multivariate analysis for port performance evaluation International Journal of Production Economics 140 1 450456 Matousek R Rughoo A Sarantis N Assaf GA 2014 Bank performance and convergence during the financial crisis Evidence from the old European Union and Eurozone J of Bank Finance doi101016jjbankfin201408012 Maghyereh A I Awartani B 2014 Bank distress prediction Empirical evidence from the Gulf Cooperation Council countries Research in International Business and Finance 30 126147 doihttpdxdoiorg101016jribaf201307001 Mandic K Delibasic B Knezevic S Benkovic S 2014 Analysis of the financial parameters of Serbian banks through the application of the fuzzy AHP and TOPSIS methods Economic Modelling 430 3037 doihttpdxdoiorg101016jeconmod201407036 Männasoo K Mayes D G 2009 Explaining bank distress in Eastern European transition economies Journal of Banking Finance 332 244253 doi101016jjbankfin200807016 Minh N K Long G T Hung N V 2013 Efficiency and SuperEfficiency of Commercial Banks in Vietnam Performances and Determinants AsiaPacific Journal of Operational Research 301 doihttpdxdoiorg101142S0217595912500479 Mohan TR Ray SC 2004 Technical efficiency in public and private sectors in India evidence from the postreform years Working Paper Nº 200322 Economics Working Paper University of Connecticut Department of Economics Müller O Uhde A 2013 Crossborder bank lending Empirical evidence on new determinants from OECD banking markets ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 44 Journal of International Financial Markets Institutions and Money 23 136162 doihttpdxdoiorg101016jintfin201209004 NgWL2007A simple classifier for multiple criteria ABC analysis Eur J Oper Res 177 1 344353 ODonnell G J Westhnizen G V D 2002 Regional comparison of banking performance in South Africa South African Journal of Economics 802 246263 doi101111j181369822002tb01301x Okeahalam C C 2008 Internationalisation and firm performance Evidence from estimates of efficiency in banking in Namibia and Tanzania Journal of International Development 207 942964 doi101002jid1455 Opricovic S 1998 Multicriteria optimization of civil engineering systems Serbia Belgrade Faculty of Civil Engineering Opricovic S Tzeng G H 2007 Extended Vikor method in comparison with outranking methods European Journal of Operational Research 1782 514529 doi101016jejor200601020 Oshinsky R Olin V 2006 Troubled Banks Why dont they all fail FDIC Banking Review 181 2344 Pancurova D Lyocsa S 2013 Determinants of Commercial Banks Efficiency Evidence from 11 CEE Countries Finance a UverCzech Journal of Economics and Finance 632 152179 Retrieved from Go to ISIWOS000319494900003 Paradi JC Yang Z Zhu H 2011 Assessing bank and bank branch performance In Cooper WW Seiford LM Zhu J Eds Handbook on Data Envelopment Analysis pp 315361 Springer New York Pearman AD 1977 A weighted maximin and maximax approach to multiple criteria decision making Operational Research Quarterly 28 3 584587 Poghosyan T Cihák M 2011 Distress in European banks An analysis based on a new dataset Journal of Financial Services Research 403 163184 doi101007s1069301101031 Portnoy S 2013 The Jackknifes Edge Inference for Censored Quantile Regression CSDA forthcoming Powell J L 1984 Least absolute deviations estimation for the censored regression model Journal of Econometrics 25 3 303 325 Powell J L 1986 Censored regression quantiles Journal of Econometrics 32 1 143155 Ramanathan U 2013 Aligning supply chain collaboration using analytical hierarchy Process Omega 412 431440 doi101016jomega201203001 Ravisankar P Ravi V Bose I 2010 Failure prediction of dotcom companies using neural networkgenetic programming hybrids Journal of Information Science 1808 12571267 doi101016jins200912022 Saaty T L 1980 The analytical hierarchy process New York MacGraw Hill Secme N Y Bayrakdaroglu A Kahraman C 2009 Fuzzy performance evaluation in Turkish Banking Sector using Analytic Hierarchy Process and TOPSIS Expert Systems with Applications 369 1169911709 doihttpdxdoiorg101016jeswa200903013 Shaverdi M Akbari M Tafti S F 2011 Combining fuzzy MCDM with BSC approach in performance evaluation of Iranian private banking sector Advances in Fuzzy Systems112 doi1011552011148712 Shi J Liu X Sun Y2006 Melt index prediction by neural networks based on independent component analysis and multi scale analysis Neurocomputing 70 13 280287 Siskos E Askounis D Psarras J 2014 Multicriteria decision making support for global egoverning evaluation Omega 40 5163 doi101016jomega201402001 Staub R B Souza G S Tabak B M 2010 Evolution of bank efficiency in Brazil A DEA approach European Journal of Banking and Finance 2021 204213 doi101016jejor200904025 Sturm JE Williams B 2010 What determines differences in foreign bank efficiency Australian evidence Journal of International Financial Markets Institutions and Money 203 284309 doi101016jintfin201003008 Sufian F 2007 Mergers and acquisitions in the Malaysian banking industry technical and scale efficiency effects International Journal of Financial Services Management 24 304326 Sufian F Kamarudin F 2015 Determinants of revenue efficiency of Islamic banks Empirical evidence from the Southeast Asian countries International Journal of Islamic and Middle Eastern Finance and Management 81 130 doi101108IMEFM1220120114 Tabachnick BG Fidell LS 2001 Using multivariate statistics Boston Allyn and Bacon Tansel İç Y 2012 Development of a credit limit allocation model for banks using an integrated Fuzzy TOPSIS and linear programming Expert Systems with Applications 395 53095316 doihttpdxdoiorg101016jeswa201111005 Tözüm H 2002 Performance evaluation of banks Journal of Banking and Finance 27 19 Tsai BH Chang CH 2010 Predicting financial distress based on the credit cycle index A twostage empirical analysis Emerging Markets Finance and Trade 463 6779 Tsay W H Chou WC Hsu W 2009 The sustainability balanced scorecard as a framework for selecting social responsible investments an effective MCDM model Journal of Operational Research Society 6010 1396410 doi101057jors200891 Wang B Nistor I Murty T Wei YM 2014 Efficiency assessment of hydroelectric power plants in Canada A multi criteria decision approach Energy Economics 46 112121 Wang W K Lu W M Lin Y L 2012 Does corporate governance play an important role in BHC performance Evidence from the US Economic Modelling 293 751760 doihttpdxdoiorg101016jeconmod201201021 Wang W K Lu W M Wang Y H 2013 The relationship between bank performance and intellectual capital in East Asia Quality Quantity 472 10411062 doihttpdxdoiorg101007s1113501195822 Wanke P Barros C 2014 Twostage DEA An application to major Brazilian banks Expert Systems with Applications 415 23372344 doi101016jeswa201309031 Wanke P Barros C Faria J R 2015 Financial distress drivers in Brazilian banks a dynamic slacks approach European Journal of Operational Research 2401 258268 doi101016jejor201406044 Wanke P Barros C Macanda N P J 2015 Predicting Efficiency in Angolan Banks A TwoStage TOPSIS and Neural Networks Approach South African Journal of Economics doi101111saje12103 Worthington A C 1999 Malmquist indices of productivity change in Australian financial services Journal of International Financial Markets Institutions and Money 93 303320 doihttpdxdoiorg101016S104244319900013X Wu CS Lin CT Lee C 2010 Optimal marketing strategy A decisionmaking with ANP and TOPSIS International Journal of ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 45 Production Economics 1271 90196 Yang Z 2009 Assessing the performance of Canadian bank branches using data envelopment analysis Journal of the Operational Research Society 606 771780 doi101057palgravejors2602619 Ying Z Jung SH Wei LJ 1995 Survival analysis with median regression models Journal of the American Statistical Association 90 178184 Yurdakul M Iç Y T 2004 AHP approach in the credit evaluation of the manufacturing firms in Turkey International Journal of Production Economics 883 269289 doi101016S0925527303001890 Zhao H Sinha A P Ge W 2009 Effects of feature construction on classification performance An empirical study in bank failure prediction Expert Systems with Applications 362 26332644 doihttpdxdoiorg101016jeswa200801053 Zheng CH Huang DS Shang L 2006 Feature selection in independent component subspace for microarray data classification Neurocomputing 69 1618 24072410 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 46 Appendix Results for the Censored Quantile Regressions tau 1 02 Coefficients Value Lower Bd Upper Bd Std Error T Value Prt Intercept 041229 038766 043693 001257 3280455 000000 Trend 000210 000012 000409 000101 207473 003801 Trend2 000013 000031 000004 000009 148947 013636 Private Bank 000212 000520 000096 000157 134991 017705 MA 000955 003153 001243 001121 085143 039453 Austria 000713 000321 001747 000527 135236 017626 Belgium 000726 000465 001918 000608 119499 023209 Denmark 000130 001009 001269 000581 022408 082270 Finland 000000 002845 002845 001452 000013 099989 France 000151 001473 001171 000674 022368 082301 Germany 000027 001211 001265 000632 004261 096601 Greece 000122 001159 001403 000654 018737 085137 Ireland 000402 000659 001462 000541 074196 045811 Italy 000933 002108 000243 000600 155470 012002 Luxembourg 002099 000498 003700 000817 256897 001020 Mexico 003675 006825 000526 001607 228709 002219 Netherlands 000544 001803 000714 000642 084769 039661 Norway 000119 001153 000914 000527 022652 082080 Poland 001866 003018 000714 000588 317482 000150 Portugal 000448 000575 001471 000522 085837 039069 S Korea 000619 000772 002009 000710 087176 038334 Slovenia 000364 000964 001692 000677 053743 059097 Spain 000122 001053 001297 000600 020315 083901 Sweden 000117 007572 007807 003923 002987 097617 Turkey 001693 003048 000338 000691 244863 001434 UK 001431 002553 000309 000573 249881 001246 USA 000870 000255 001994 000574 151580 012957 Method PCA 002215 002061 002369 000079 2811852 000000 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 47 tau 1 04 Coefficients Value Lower Bd Upper Bd Std Error T Value Prt Intercept 042214 041528 042901 000350 12053231 000000 Trend 000233 000108 000358 000064 366400 000025 Trend2 000015 000025 000005 000005 299107 000278 Private Bank 000082 000295 000131 000109 075741 044880 MA 001598 002024 001171 000218 733853 000000 Austria 000469 000335 000603 000068 687293 000000 Belgium 000690 002012 003393 001379 050066 061661 Denmark 000445 000057 000833 000198 224556 002473 Finland 000686 000125 001247 000286 239660 001655 France 000149 000171 000468 000163 091301 036124 Germany 000240 001099 000618 000438 054807 058365 Greece 000808 000239 001377 000290 278335 000538 Ireland 000539 000306 000772 000119 453906 000001 Italy 000456 000904 000009 000228 199759 004576 Luxembourg 002435 000166 005036 001327 183506 006650 Mexico 003293 004770 001816 000754 436926 000001 Netherlands 000101 000518 000315 000213 047628 063388 Norway 000059 000161 000279 000112 052290 060105 Poland 001682 002380 000984 000356 472079 000000 Portugal 000678 000104 001460 000399 169863 008939 S Korea 000690 000205 001175 000248 278605 000534 Slovenia 000424 000117 000731 000157 270645 000680 Spain 000274 000001 000547 000139 196399 004953 Sweden 000709 000492 000925 000111 640780 000000 Turkey 001379 001661 001097 000144 957334 000000 UK 000557 000785 000330 000116 480384 000000 USA 001180 000905 001455 000140 841840 000000 Method PCA 002211 002094 002328 000060 3711471 000000 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 48 tau 1 06 Coefficients Value Lower Bd Upper Bd Std Error T Value Prt Intercept 041750 041090 042411 000337 12391415 000000 Trend 000159 000054 000264 000053 297613 000292 Trend2 000009 000017 000001 000004 220184 002768 Private Bank 000100 000095 000294 000099 100360 031557 MA 000681 001171 000191 000250 272311 000647 Austria 000391 000204 000579 000096 408682 000004 Belgium 000702 000461 000943 000123 570609 000000 Denmark 000733 000493 000972 000122 600373 000000 Finland 001406 000719 002092 000350 401079 000006 France 000234 000010 000459 000114 204856 004051 Germany 000270 001166 001706 000733 036805 071284 Greece 001306 000898 001714 000208 627651 000000 Ireland 002823 002593 003052 000117 2411375 000000 Italy 000276 000042 000595 000162 170122 008890 Luxembourg 002958 001654 007570 002353 125712 020871 Mexico 002922 003391 002454 000239 1222347 000000 Netherlands 000163 000077 000404 000123 133060 018332 Norway 000256 000031 000543 000146 174879 008033 Poland 001075 002066 000084 000506 212623 003348 Portugal 000778 001000 002555 000907 085742 039121 S Korea 001974 001119 002830 000437 452185 000001 Slovenia 000515 000267 000763 000127 406747 000005 Spain 000402 000206 000599 000100 402063 000006 Sweden 001004 000824 001183 000091 1097864 000000 Turkey 000765 001063 000468 000152 504298 000000 UK 000312 000558 000066 000126 248600 001292 USA 001441 001202 001679 000122 1184853 000000 Method PCA 002200 002084 002317 000059 3709730 000000 ACCEPTED MANUSCRIPT ACCEPTED MANUSCRIPT 49 tau 1 08 Coefficients Value Lower Bd Upper Bd Std Error T Value Prt Intercept 042444 026556 058331 008106 523610 000000 Trend 000112 000382 000159 000138 081011 041787 Trend2 000019 000005 000043 000012 155552 011982 Private Bank 000904 000255 001552 000331 273049 000632 MA 000411 016150 015328 008030 005119 095918 Austria 000218 000307 000743 000268 081288 041629 Belgium 001134 011282 013549 006335 017897 085796 Denmark 000831 000355 001307 000243 342048 000063 Finland 001517 001052 001982 000237 638889 000000 France 000108 000476 000692 000298 036117 071797 Germany 000619 001918 000680 000663 093395 035033 Greece 004028 003005 005051 000522 771950 000000 Ireland 004176 003298 005054 000448 932039 000000 Italy 000376 000257 001009 000323 116333 024469 Luxembourg 004707 004053 013466 004469 105315 029227 Mexico 003810 004652 002968 000430 887045 000000 Netherlands 000107 000870 000655 000389 027643 078222 Norway 000390 000068 000712 000164 237621 001749 Poland 000871 001610 000133 000377 231174 002079 Portugal 002891 002063 003720 000423 684039 000000 S Korea 002853 000986 006692 001959 145661 014522 Slovenia 000766 000167 001364 000305 250860 001212 Spain 000213 000187 000612 000204 104431 029634 Sweden 000811 000490 002112 000664 122125 022199 Turkey 001019 002422 000385 000716 142276 015481 UK 000190 001019 000640 000423 044857 065374 USA 002765 001634 003896 000577 479056 000000 Method PCA 002159 001905 002412 000129 1667546 000000 OMEGA Int J of Mgmt Sci Vol 17 No 3 pp 237250 1989 03050483j89 300 000 Printed in Great Britain All rights reserved Copyright 1989 Pergamon Press plc An Application Procedure for DEA B GOLANY Y ROLL Technionlsrael Institute of Technology Israel Received October 1987 in revised form November 1988 Data Envelopment Analysis DEA has become an accepted approach for assessing efficiency in a wide range of eases The present paper suggests a systematic application procedure of the DEA methodology in its various stages Attention is focused on the selection of decision making units DMUs to enter the analysis as well as the choice and screening of factors The application of several DEA models in different versions and formulations is demonstrated in the process of determining relative eMciencies within the compared DMUs I INTRODUCTION I EFFICIENCY OR PRODUCTIVITY analyses are vital managerial control tools for assessing the degree to which inputs are utilized in the process of obtaining desired outputs The abundant available literature on productivity measure ment tackles this problem from various points of view applying different approaches The economic approach assumes the existence of specific mostly constant over time input output relations which can be identified by the analysis of a large body of data eg national accounts Productivity due to this approach is evaluated against production functions which formulate the assumed relations 27 An engin eering approach is one where productivity is measured by comparing performance to suit ably set engineering standards 29 Other ap proaches assess productivity by means of ratio analysis or by variations on accepted accounting techniques None of these approaches is satis factory for measuring productivity in the service and public sectors or in nonprofit organiz ations This statement holds also for private sector companies if some of the pertinent factors are not readily expressed in economic terms The main reasons for the failure of traditional productivity measurement approaches in such cases are the following Most traditional approaches for pro ductivity assessment are based on process measures with little or no attention to important outcome measures eg 31 2 Such outcome measures as well as some input factors are typically quali tative in nature In addition to the problems involved in quantifying such factors it is usually extremely difficult to assign them proper relative weights 3 It is very difficult to formulate an ex plicit functional relationship between inputs and outputs with fixed weights on the various factors 4 Averaging performance across many DMUs as in statistical regressions fails to explain the behavior of individ ual DMUs The Data Envelopment Analysis DEA approach 11 15 tries to overcome these diffi culties It provides a means for assessing relative efficiencies of Decision Making Units DMUs with minimal prior assumptions on input output relations in these units Such relative efficiencies can be evaluated among a group of 237 238 Golany RollmAn Application Procedure for DEA a single period or in a sequence of periods Alternatively efficiencies can be evaluated with respect to known performance by competitors or measured against predetermined goals or standards Objectives for carrying out DEA vary widely including typically the following Identification of the sources and amounts of relative inefficiency in each of the compared units in any of its inputoutput dimensions eg 8 Ranking of the units by their efficiency outcomes Evaluation of management heading the compared units Evaluating the effectiveness of pro grams or policies which are outside the control of the units and differentiating between program inefficiency and managerial inefficiency eg 16 Creating a quantitative basis for re allocating resources among the units under evaluation Reallocation poli cies may take into account the slacks found in some units as well as the virtual multipliers obtained for the various inputs at different units see Appendix The general purpose of such reallocation policies is to shift limited resources to units where they would be utilized more effectively in generating desired outputs eg 7 Identification of efficient units or efficient inputoutput relations for pur poses not directly related to compari son among the units An example for such an objective is the use of DEA outcomes to determine test markets for the trial of new products Other objectives are cited in 36 Analysis and scrutiny of prevailing standards on specific inputoutput relations against actual performance eg 32 w Comparison and contrasting of results from previous studies eg 3 9 Since first publications on DEA in 1978 the literature has expanded rapidly reporting further conceptual and theoretical develop ments as well as numerous applications 14 However in most cases DEA applications are reported as demonstrations of or additions to the theory Thus they tend to emphasize a particular point rather than give an overall view of the application procedure Furthermore some steps in the procedure are not dealt with in a methodical manner Typical examples of such steps are the choice of factors to enter the analysis and the quantification of qualitative factors 34 Also little is published on the comparison between various versions of the DEA approach and on the significance of differences in outcomes resulting from employ ing different models The purpose of this paper is to bring together some of the developments in DEA and organize the experience gained into a systematic applica tion procedure Because of the extremely wide and heterogeneous areas for which DEA may be appropriate the proposed procedure can only serve as a general guideline Where applicable some of the different options are presented and discussed An attempt is made towards minimal repetition of material published elsewhere for which frequent reference is made However the paper is designed to be to a degree selfcon tained so that the proposed procedure can be readily followed by users with little previous experience of DEA To this end the most common DEA models are presented in the Appendix The three main phases in carrying out an efficiency study by means of DEA are the following i Definition and selection of DMUs to enter the analysis ii Determination of input and output factors which are relevant and suitable for assessing the relative efficiency of the selected DMUs iii Application of the DEA models and analysis of outcomes Each of these phases comprises several steps The entire proposed procedure is presented in a flowchart form in Fig 1 The different phases are discussed in some detail in the next three sections Then a short illustrative example is worked out followed by some concluding remarks Omega Vol 17 No 3 239 2 SELECTION OF DMUs DEA is a technique for assessing the relative efficiency of comparable units with a view to improving their performance This implies a basic assumption that differences in perform ance among like units exist and are measurable 23 25 Even under quite similar conditions one always finds differences in the way units are managed if only because they are led by dif ferent decision makers Thus on the one hand we look for a homogeneous set of units where comparison makes sense and on the other we try to identify the differences between them These contradicting considerations accompany every step of a DEA application They are however most prominent during the stages of choosing the DMUs to be compared and identi fying the factors affecting them A homogeneous group of units for our purposes is one where the units under consideration perform the same tasks with similar objectives all the units perform under the same set of market conditions this is of special importance in the analysis of nonprofit organizations such as schools army units state hospitals courts etc the factors both inputs and outputs characterizing the performance of all units in the group are identical except for differences in intensity or magni tude The next step is to determine the size of the comparison group Again two conflicting con siderations present themselves On the one hand there is a tendency to increase the number of units because the larger the population the larger the probability of capturing high per formance units which would determine the efficiency frontier Also a large set of units enables a sharper identification of typical relations between inputs and outputs in the set Furthermore as the number of units increases it is possible to incorporate more factors into the analysis A rule of thumb established here see 1 8 is that the number of units should be at least twice the number of inputs and outputs considered On the other hand the larger the number of units in the analyzed set the lower the homogeneity within the set generally increasing the possibility that results may be affected by some exogeneous factors which are not of interest The determination of DMUs to enter the DEA evalualtion process is affected by two kinds of boundaries One comprises the organiz ational physical or regional boundaries which define the individual units The other relates to the time periods used in measuring the DMUs activities Preferably the time periods to be considered should be natural ones corre sponding to seasonal cycles and budgeting or auditing periods Regarding the length of such periods it should be borne in mind that long periods may obscure important changes occur ing within them while short periods may give an incomplete picture of the DMUs activities Another consideration is how far back to go without distorting the comparison 26 and whether to create an overlap of data by means of window analysis 10 A final step in deter mining the set of units is sifting out DMUs which may be considered as outliers ie units or time periods deviating from the general characterization of the group to be analyzed This should be carried out with utmost care and if needed with the advice and intervention of managers 13 One should bear in mind that efficiency is measured with respect to the DMUs and factors selected There is no guarantee that the initial selection is correct in the sense that it serves best the purposes of the analysis Thus the considerations mentioned in this section may require the application of parts of the proposed procedure in an iterative fashion as indicated in Fig 1 No practical stopping rule is suggested for these iterations and they depend entirely on the specific application Furthermore it may be useful in many applications to present more than one set of results derived from different selections of DMUs factors or models 3 SELECTION OF FACTORS The initial list of factors to be considered for assessing DMU performance should be as wide as possible Every dimension the changes in which may affect the DMUs to be evaluated should be included in the initial list Such factors could be either fully or partially controllable by 240 Golany RollmAn Application Procedure for DEA I J VL I Set Factors Measurement Scales I Ralationshil Formalize Inial Model Computer Environment I Analyze by Factor I Define Population of DMUs I Set Goals for Analysis I Select DMUs to be Compared m LIS Relevant Factors I I Examine Factors I Judgement I j Examine Factors II Correlations Examine Factors III Trial Runs Formalize Final Model I Presenl Initial Results 1 i Special Analyses I I I Analyze by Individual DMU Fig DEA application flow chart the DMUs or they may be environmental factors outside the control of the DMUs Some of the factors would be quantitative ie readily available while other factors may be qualita tive in nature with different degrees of difficulty to be accorded numerical values The factors could be inputs or outputs or factors placed on either side of a production relationship All the factors which may possibly have any bearing on the performance of the DMUs to be analyzed should be listed at this stage without any numerical treatment Clearly this may result in quite a large number of factors in the initial list for example Thomas 35 started his study with an initial list of 92 factors which was eventually brought down to only 14 see also 16 22 The introduction of a large number of factors into the analysis will result in explaining away a larger portion of the differences among DMUs This will tend to shift the compared units towards the efficiency frontier resulting in a relatively large number of units with high efficiency scores In the following procedure it is proposed to introduce initially only a limited number of carefully selected factors thus accen tuating the basic differences among units At a later stage while analyzing the outcomes additional factors can be brought into the model to examine whether they explain some of the differences Thus the next steps are directed towards the reduction of the typically lengthy initial list to one that includes only the most relevant factors The factors chosen should distinguish clearly among the compared units and serve effectively the objectives of the analysis This refinement of the list can be carried out in three stages i judgemental screening ii nonDEA quantitative analysis iiiDEA based analysis 1 Judgemental process The first stage in reducing the list of factors is its critical examination by expert decision makers in the field where the DMUs operate Omega Vol 17 No 3 241 Since a large list of relevant factors is usually compiled some may be repeating virtually the same information some may not be regarded as crucial while others may appear to be con flicting or confusing A problem often encountered at this stage is the proper distinction between factors determin ing efficiency and factors explaining efficiency gaps For example labor input may serve to determine efficiency while the scale at which the unit operates may be an explaining factor Entering such explaining factors into the initial analysis may blur the overall picture and reduce the distinction between compared units It may also obscure our understanding of the way in which some factors affect performance and generally diminish the usefulness of efficiency analyses as tools for improving performance In public sector units or in nonprofit organiz ations where factors governing performance are not always welldefined special care should be exercised to distinguish between inputs and explaining factors Judgement may be exercised inter alia along the following lines Is the factor related to or contributing to one or more of the objectives set for the application Is the factor conveying pertinent infor mation not included in other factors Does the factor contain elements eg price which interfere with the notion of technical efficiency 23 Are data on the factor readily available and generally reliable Possible systematic procedures to be applied at this stage for structuring the judgemental process are Delphilike techniques 35 or vari ations of the Analytic Hierarchy Process 30 Ad hoc methods using expert knowledge are mentioned in 9 where a panel of experts was used to formalize the model and later check its results and in 33 where an official committee was formed to scrutinize the factors 2 NonDEA Quantitative Methods The first step here is to assign numerical values to the various factors For many of these factors the natural choice would be the physical units by which they are measured Such measures can be in dollar terms for economic factors in number of persons KWH of gener ated electricity gallons of fuel etc A non trivial consideration in this regard is whether to aggregate all or some factors which can be measured in economic terms into one dollar factor in other words assign fixed relations among the weights of these factors see discus sion in section 4 Whether or not to do such aggregations depends strongly on the objectives of the analysis Another issue concerning quan titative factors is the handling of cases where zero values are encountered for some factors in some DMUs or some time periods This may happen if the periods chosen do not correspond to natural cycles of operation or for reasons stemming from the datagathering procedure In such cases periods could be redefined or data accumulated across several periods In principle the DEA models can handle cases with zero values for some of the factors as long as there exists at least one input and one output for each DMU which is nonzero see Appendix However such cases should be handled with care as the computational algorithms may be sensitive to zero values 19 A final remark concerns the isotonicity relations which are assumed for DEA ie an increase in any input should not result in a decrease in any output Consequently the values of some factors may have to be inverted before they are entered into the analysis see for example 10 Another group of factors is the qualitative ones Indeed the inclusion of such factors is one of the novelties of DEA Nevertheless they have to be asigned numerical values in order to participate in the mathematical evaluation of efficiency The usual practice here is to locate some measurable surrogate variable which is assumed to bear a known relation to varying levels of the qualitative factor Typically several possible surrogates may be tried out for each qualitative factor until a suitable one is located Criteria for the choice of surrogate factors are the degree of correspondence between vari ations in the surrogate data and the examined factor the ability to express this correspondence in a functional form and the general compliance of the results to the analysis objectives An example of the selection process is described in 32 242 Golany RollAn Application Procedure for DEA The next step within this stage is to describe the production relations governing the DMUs to be analysed and classify the factors into inputs and outputs Resources utilized by the units or conditions affecting their operation are typical inputs while measurable benefits gener ated constitute the outputs In most cases this distinction is straightforward However some factors may be interpreted in both ways their classification depending on the analysts point of view eg 10 A useful procedure here may be to carry out a series of regression analyses of such factors one at a time on the factors known to be inputs and outputs A weak relation to inputs and strong relation to outputs indicates a preference towards classifying the factor as an input while a reversed outcome will point towards viewing the factor as an output A weak relation to all the factors may indicate a need to reexamine the factor and possibly delete it Alternatively strong relations may indicate that the information contained in that factor is already represented by other factors and again its deletion should be considered Similar regression analyses should be carried out on factors clearly defined as inputs and outputs with a view to eliminating redun dancies and reducing the list of factors eg 26 However as observed in 28 one should not regard these oneatatime regression tests as reliable rules but merely as indicators for a need to examine some of the factors more closely see 9 for an overview of contrasts between DEA and regression analyses It may be useful at this stage to use each factor separately to rank order all the DMUs These rankings can then be aggregated into a single ranking which best describes them all eg a ranking as in 20 Factors with rankings which differ sharply from the others are again candi dates for possible deletion Throughout this stage attempts can be made to refine or replace the surrogates selected earlier for the qualitative factors 3 DEAbased analyses The last step in the process of examining and refining the list of factors consists of trial runs of DEA models Within the family of the DEA models the one initially proposed by Charnes Cooper and Rhodes CCR 15 is the model which points out differences among DMUs in the most critical way Some of the subsequent models notably the one introduced by Banker Charnes and Cooper 2 incorporate some of the explanations to efficiency differences into the models themselves In the procedure proposed here analysis is started with the most strict model namely CCR Other DEA models are introduced in the subsequent stage when out comes are analyzed in an attempt to explain the resulting efficiency differences Factors which have remained in the list so far are now entered into the model and outcomes are examined closely Factors which are consist ently associated with very small multipliers ie have little impact on the efficiency scores may be dropped Basically we seek the ability to discriminate between the DMUs using the selected factors Hence factors which do not contribute to this end are candidates for elimination To test the discriminating power of the different factors the model is run with a series of combinations of these factors Then various grouping techniques can be applied to the DMUs using the resulting efficiency scores Factors which do not alter such group ings significantly should be examined closely Special attention should be given to factors which could not be easily classified as inputs or outputs They can be tried at both sides of the efficiency ratio and classified finally according to the results Some factors typically the nondiscretionary ones can be used to rescale all other factors in the analysis Thus non discretionary factors are accounted for in an indirect manner and the total number of factors reduced for a different approach see 4 Following a series of such iterations the list of factors to enter the analysis is decided upon Outcomes of the final iteration provide at the same time the first and basic set of efficiency scores for the compared DMUs The following section describes the process of further analysis and examination of these results 4 PRESENTATION AND ANALYSIS OF RESULTS Further analysis of the relative efficiency values obtained from the basic CCR model can be performed inter alia by applying different models of DEA and introducing addi tional information into the respective models There are two possibilities of formulating the CCR model One puts the emphasis on input Omega Vol 17 No 3 243 reduction and the other on output enhance ment Both formulations yield identical results which is not the case with other DEA models such as the BCC The choice between the two formulations is made according to prevailing circumstances In some applications the inputs are rather inflexible eg determined to a certain extent by higher managerial levels in which case the output formulation would be more suitable In other applications outputs are matched closely with goals set by management or restricted by environmental conditions Then the input formulation is more appro priate The models can be presented in either of two versions the primal version highlighting the virtual multipliers of inputs and outputs or the dual version emphasizing the relations between the examined DMU and the other members in the analyzed set A brief summary of some of the DEA models is presented in the Appendix For the demonstration of possible efficiency analyses a small illustrative example is worked out Suppose that the screening process of the relevant factors for a group of ten DMUs resulted in locating two outputs and three inputs as the most meaningful for assessing the relative efficiencies within the group The first ten lines in Table 1 list the numerical values of the five factors for each of the ten DMUs Table 2 presents the outcomes from applying the CCR model to this set of data The outcomes contain four components a An efficiency score for the specific DMU relative to a subgroup of DMUs those in the reference facet to which the examined DMU is best compared b The slacks output shortfalls and input surpluses associated with the examined Table I Illustrative example data DMU No Outputs Y2 Inputs XI X 2 X 3 I 090 70 I0 08 540 2 100 95 15 10 480 3 080 75 12 21 510 4 090 90 I0 06 420 5 070 80 18 05 600 6 100 50 7 09 520 7 080 70 I0 03 500 8 075 75 12 15 550 9 065 55 14 18 570 10 085 90 8 09 450 I I 095 I00 6 03 450 12 100 100 7 03 420 13 100 95 8 03 400 DMU in addition to the increase of all outputs or the decrease in all inputs by a factor equal to the efficiency score For example observing the outcomes for DMU all inputs could be de creased by a factor of 0847 and in addition input 3 decreased by 92 units in order to bring the efficiency of DMU to that of DMUs 4 6 and 7 c The DMUs in the reference facet for the examined DMU d The virtual multipliers for the examined DMU These are the factors weights determined by the model which when applied to the said DMU will render its best relative efficiency score by the fol lowing expression Oo uyo 3 see the Appendix for the notation A similar set of results can be obtained by applying other DEA models Three examples are given i The BCC model 5 and 6 in the Appendix This model takes into account the effect of returns to scale Table 2 Summary of outcomes across DMUs the CCR model Slacks DMU Outputs Inputs DMUs No Elciency I 2 I 2 3 in facet Virtual multipliers Outputs Inputs 1 2 I 2 3 1 0847 92 4 6 7 2 0972 50 35 031 4 3 0734 40 10 4 6 4 10 4 5 0829 017 44 4 7 6 10 6 7 0 7 8 0660 041 4 6 I0 9 0532 I0 029 052 4 I0 10 I0 0695 00032 0064 044 00000 0971 0 0 0 00021 0917 0 0020 0 00015 II I 0 0024 0 00018 0 00104 0 093 00109 0897 00021 0043 0 00013 1249 0 0045 061 00007 0715 00016 0034 0 00011 0818 0 0 0 00018 0 00111 0026 0 00018 244 Golany RollAn Application Procedure for DEA Table 3 Summary of outcomes across models DMU Model Efficiency Slacks Outputs Inputs DMUs I 2 1 2 3 in facet Vimtal multipliers Outputs Inputs I 2 I 2 3 CCR inputs CCR outputs BCC inputs BCC outputs Multipl 8 Additive 9 0847 92 4 6 7 0847 92 4 6 7 0937 025 676 4 6 2 0870 48 4 6 7 I0 0064 073 651 2 4 6 20 02 120 4 0695 00032 0064 0445 0 0695 00032 0064 0445 0 1063 00010 0 0275 0 0050 0 0044 0211 0001 0 0 0 0 0 OOI I 0 0 0 0 within the analyzed group of DMUs attempting to point out the most pro ductive scale size for each DMU while simultaneously identifying its technical inefficiency 2 Note however that the BCC model yields different results with the input and output formulations ii The unit invariant multiplicative model 7 in the Appendix This model links the DEA approach to some accepted CobbDouglaslike production func tions where production relationships are characterized by the product of factors raised to appropriate powers rather than the weighted sum of factors as in the common engineering approaches 18 iiiThe additive model 8 in the Appendix This model constitutes a Paretooptimal empirical production Table4 Effciencyrankingsacross models DMU CCR BCC CCSS CCGSS I 6 7 6 7 2 5 I5 I5 I5 3 8 8 9 6 4 14 I5 I5 I5 5 7 6 7 9 6 14 I5 I5 I5 7 I I5 I5 I5 8 9 9 8 8 9 10 10 10 10 10 14 I5 I5 15 function for the given group of DMUs 12 As mentioned above not all DEA models should be attempted for every application Rather some models are suitable only for specific cases or purposes of the analysis see the Appendix Here however only one example is given and the different models are applied to the same set of data The outcomes of these models for DMU are presented in Table 3 No efficiency score is given for the additive and multiplicative models since in these models the objective values represent sums of slacks rather than efficiency ratios see the Appendix In order to compare the efficiency evaluation of all the DMUs across the different models in spite of the difficulty men tioned earlier a summary of results in terms of optimal rankings is given in Table 4 In general most DMUs were ranked similarly by the different models which is an indication of the robustness of the technique However different rankings due to differences among the models should be expected as evident for DMUs whose ranking according to the CCGSS model is worse than its rankings by the other models A sum mary of DMUs appearing in the facets with the different models is given in Table 5 The last column in this table sums up the number of times each DMU was selected by the models run in this example to form a reference set for Table 5 DMUs appearing in facets with the different models DMU No CCR DMUs in reference facet BCCin BCCout Multipl Additive No of times this DMU is in a facet I 467 2 46 46 7 10 246 4 0 2 4 2 2 2 2 12 3 46 246 4 10 26 4 0 4 4 4 4 4 4 26 5 47 47 47 467 47 0 6 6 6 6 6 6 17 7 7 7 7 7 7 12 8 4 6 I0 2 I0 4 I0 2 6 4 0 9 4 2 6 4 2 6 4 0 I0 I0 lO 10 I0 I0 I0 Omega Vol 17 No 3 Table 6 Summary of outcomes 245 Efficiency scores CCR CCR BCCout CCR within constrained CCR within with CCR BCCout DMU original output aggregated original standard within within No set weights inputs set set category category j I 0847 0712 0844 0870 0681 10 10 2 0972 0924 0972 10 0833 0972 10 3 0734 0689 0732 0826 0627 0741 0833 4 10 10 10 10 0900 10 10 5 0829 0829 0622 0900 0560 10 10 6 10 0642 10 10 0906 10 10 7 10 10 10 10 0800 10 10 8 0660 0651 0636 0781 0573 0721 0818 9 0532 0451 0532 0737 0456 0600 0789 10 10 10 0941 10 0840 10 10 DMUs I5 were assumed in category A and DMUs 610 in category B other DMUs The above are only examples of the different DEA models There are several more such models each with its specific charac terisations for example the multiplicative model 17 and the weighted additive model Another direction in the analysis of efficiency outcomes is partitioning the group of DMUs into categories according to some characteristic which was not entered into the model as a factor determining inputoutput relationships The purpose of such categorization is twofold one is to gain a better relative assessment of efficiency by comparing performance within subgroups of units operating under similar conditions eg the same geographical region see 5 6 for related developments The other is a compari son between categories such as in the case where a category signifies a program under which a subgroup of DMUs operates 16 Table 6 shows the results of such a partitioning into two categories Further insight into the relative efficiency scores within the examined group of DMUs can be gained by imposing additional constraints on the standard models Such constraints can be added if additional information is available or specific hypotheses are to be tested about relationships among the factors Following are several examples ii iii i A known relationship between the weights of inputs This may be the ease when some of the inputs have market values and it is assumed that the virtual multipliers should relate to each other as the respective prices The additional constraint here will take the form of 9 in the Appendix In the illustrative example a situation is tested where the price of one unit of input 2 is ten times as high as that of one unit of input 1 The outcome given in Table 6 is identical to the one obtained without the additional constraint but with in puts l and 2 replaced by a single input reflecting their combined cost An assumed directional relationship between outputs 24 Such a condition can be imposed for example when there is a clear preference of one quali tative output over another The corre sponding constraint will take the form 10 in the Appendix Table 6 illustrates the case where there is a preference of output 2 over output 1 in the example This was effected by setting the virtual multiplier of output 2 at least as high as that of output 1 Bounding the range within which factor weights may vary 21 The argument here is that when we let the weights vary freely some DMUs may be as sessed by only a subset of the inputs and outputs which are considered rele vant Moreover a completely free choice of weights can be viewed in some cases as covering up the most serious deficiencies low outputs andor high inputs of the DMU being analyzed The difficulty here is in locating such suitable bounds One suggestion is to set the bounds on factor weights by observing the weight variation when the unbounded model is run Another option is to impose bounds on ratios between weights reflecting the 246 Golany RollAn Application Procedure for DEA importance attached by the analyst to the various input and output factors It should be noted that adding constraints will result only in downward changes in the efficiency scores On the other hand reducing the number of compared DMUs will result in upward changes in efficiency scores An additional perspective of the efficiency picture can be gained by adding to the observed data some external reference data Such refer ence data can be obtained from comparable DMUs outside the analyzed group or through setting standards of performance Adding ex ternal reference data serves to overcome the limitation of having only relative efficiencies within the analyzed group In the illustrative example three theoretical sets of data were added lines 1113 in Table 1 and the resulting efficiencies evaluated Table 6 summarizes the efficiency scores for the ten DMUs in the example when applying several models and model modifications It may be seen that substantially different efficiencies are obtained under each set of different con ditions The appropriateness of the different approaches and the significance of the respective changes in the efficiency scores should be judged in light of the specific circumstances of each application Efficiency outcomes can be pre sented graphically in the form of a bar chart as shown in Fig 2 for DMU8 in the example The total efficiency gap between the performance of DMU8 and the set of standards is partitioned into portions explained in different ways and an unexplained portion Again partitioning of the efficiency gap and the explanations offered depend strongly on the particular circumstances and the purpose of the analysis Also it should be noted that in most cases not all the evalu ations presented in Table 6 are simultaneously relevant Hence Fig 2 demonstrates a situation where only the first and the last four columns in Table 6 are of interest Similar barcharts can be constructed for the other evaluations when necessary An important aspect of any DEA implemen tation is the use of appropriate computer codes for handling the heavy computation needed in the analysis Since this technical issue is outside the main scope of this paper we find it sufficient to propose a general scheme for a DEA decision support system as in Fig 3 5 SUMMARY AND CONCLUSIONS DEA is an effective approach to assess the efficiency of organizations in cases where tra ditional approaches to efficiency analysis or pro ductivity measurement either fail or are difficult or impossible to apply Such are the cases typical within the public sector or the non Total Effic lency Gap GAp Total Within Compa Cetegor rlson Gap 10 818 ½ IIIIllNHHIIIIII Fig 2 Pardfioning the eiciency llun accounted for portion portlon explained by assumed diseconomies of scale within category Ioportlon explained by difference betwlsn best overall performance between standard and actual best performance gap DMU No 8 Omega Vol 17 No 3 I Input File Repeated LP IIi J Data Base Handler I Input File t Stasoa J Solver regr essions etc Tasks input I nput Fie l l Generator I Data Base handling package carble of storing retrieving errorchecking etc 2 Repeated Linear Programming mehanizm 3 Statistical Package 4 Graphical package 5 Report generator Fig 3 DEA decision support system 247 profit organizations where no strict functional relationship can be formed a priori between factors of production and where the relative weights of these factors inputs and outputs in the production processes are not welldefined However tackling such vaguely defined situ ations has its problems Prominent among these is the fact that DEA yields relative efficiencies only within the examined group of DMUs Outcomes are also dependent on the factors entered into the analysis and the numerical values accorded to qualitative factors Con sequently attention should be focused on the differences between efficiency scores rather than their absolute values It is also important to remember that the typical system under exam ination is dynamic in nature undergoing changes continuously Therefore special atten tion should be given to the identification of changes or trends in efficiencies over time Finally sensitivity of outcomes to the various assumptions made eg the mix of pertinent factors and the version of the DEA model employed should be carefully tested The example given in the previous section does not exhaust the possibilities for special analyses based on the DEA outcomes Other aspects may include statistical evaluation of the efficiency scores multiphase DEA sensitivity analysis techniques etc A possible approach for obtaining a general efficiency picture within the analyzed group is to divide it into several broad efficiency categories Members of the highest category would be the DMUs most often present in the reference facets and relatively insensitive to the different assumptions The lowest category would con tain the DMUs with consistently low efficiency scores with as many intermediate categories as befit the circumstances A valuable measure of change could be the shift of particular DMUs from one such category to another In recent years a growing number of researchers have criticized the field of OR for solving only limited and unrealistic problems while ignoring wicked problems which are common in real life situations The DEA approach is perhaps a step in the right direction OME 7 D 248 Golany RollAn Application Procedure for DEA It attempts to address tough problems and offers a way of tackling them The procedure proposed in this paper is meant to assist in extending the use of this method to areas which have not been explored so far as well as to strengthen existing applications APPENDIX DEA Models z DEA measures efficiency by estimating an em pirical production function which represents the highest values of outputs that could be gener ated by relevant inputs as obtained from ob served inputoutput vectors for the analyzed DMUs The inefficiency of a DMU is then measured by the distance from the point repre senting its input and output values to the corre sponding reference point on the production function There are a number of mathematical formulations of DEA all sharing the principle of envelopment An output vector Yk for DMUk a specific unit under evaluation is enveloped from above when the model identifies a combi nation of other output vectors for the same input vector X whose values are equal to or greater than all the elements in Yk Similarly the input vector Xk is enveloped from below when the model finds a combination of other input vectors whose values are smaller than or equal to all the elements in Xk If the pair Xk Yk cannot be enveloped simultaneously by a com bination of other DMUs other than DMUk itself then DMUk is efficient In general the set of efficient DMUs selected for evaluating an analyzed DMU defines one facet of the piece wise linear empirical production function A linear combination of these DMUs serves as a reference point for the measurement of the inefficiency of DMUk The data which DEA uses are only the observed Yj vectors for all the DMUs in the analysis By solving a series of linear programming LP optimizations one for each DMU DEA is able to identify those DMUs that are efficient on the facets and the This Appendix is not a comprehensive survey of the DEA methodology and is intended for readers who have already had some experience with DEA It highlights important differences in the meaning of the models and their suitability for different scenarios Readers are referred to 11 for detailed presentations of the methodology remaining inefficient DMUs along with their efficient reference points DEA was first developed as an extension to the classical engineering ratio notion of efficiency It determines for each DMU the maximal ratio of the sum of its weighted outputs divided by the sum of its weighted inputs where the weights are determined by the model This fractional programming problem is transformed into an LP model 1 below 11 Applying the duality theory of LP a dual program 2 can be derived for the transformed program The con straints of the dual lead to the envelopment interpretation given above Common notation used in the sequel is sum marized below Indices Data Variables jDMUs j n routputs r I t inputs i ffi 1 m yjthe value of the rth output of the jth DMU xgthe value of the ith input for the jth DMU a small positive number s aslacks corresponding to input i out put r respectively t0 2jweight of DMUj in the facet for the evaluated DMU 0 vvirtual multipliers for output r input i respectively hrelative efficiency of DMUk a The CCR model 15 For each DMU solve Max hk Yk st vxk 1 i r i The objective here is to find the largest sum of weighted outputs of DMU while keeping the sum of its weighted inputs at unit value and forcing the ratio of the sum of weighted outputs to the sum of weighted inputs for any DMU to be less than one This ratio corresponds to the classical Engineering ratio definition of efficiency 15 The dual program solves for each DMU St E y2j o yk J Y xj Okxk s 0 2 1 Omega Vol The objective function of this model attempts to find a minimal value for an intensity factor 0k which indicates the potential of a proportional reduction in all the inputs of DMUk In addition the objective function seeks the largest slack values in all inputoutput dimensions In other words it finds the reference point on the empirical production function which portrays DMUk in the worst efficiency characterization The constraints of the model represent the envelopment principle explained earlier We started in 1 with an emphasis on output enhancement holding the inputs at a fixed unit value Similar formulations are written for the case where input reduction is emphasized Min h k vxk i st ta y 1 r tatj Y zo t o 3 i and Max St jii fl 0 J i 4 b The BCC model 121 This model adds another restriction to the envelopment requirements It requires that the reference point on the production function for DMUk will be a convex combination of the observed efficient DMUs The dual formulation for DMUk is written in this case as St Z ij ff t yr i Y xoOkxsO E 2j 1 5 The corresponding primal has a slightly different objective from I Max h taYk uk st viz l i ta yj v Zo uk 0 6 t i the term uk was interpreted by BCC as an indicator of returns to scale 17 No 3 249 C The CCSS model 181 In formulations 2 4 and 6 the measure of relative efficiency is derived from sums of weighted inputs and outputs Alternatively this measure can be generated by taking the ratio of a product of outputs raised to virtual weights divided by the product of inputs raised to the virtual weights The envelopment formulation which results from this definition was shown in 18 to be associated with the CobbDouglas type production functions The model solves for each DMU Min hkas r i st logyoA j a logyk J IogxA j s logxk 1 E 7 J d The CCGSS model 121 Here DEA can be used to test for Pareto Optimality of the analyzed DMUs In 12 it was shown that DMUk is Paretooptimal if and only if the objective value in the next model is zero Min hk o si i st Y yjjy J Xo j S Xk J Z J t 8 e Modified DEA models Example I Model 1 with additional con straints such as c2vt c t v 2 0 9 These constraints specify known relations between virtual multipliers Adding such a con straint is equivalent to combining inputs 1 and 2 into one aggregate input eg as when con sidering economic inputs and testing the effect of replacing the physical quantities x and x2 with their combined cost clx c2x Example 2 Model 1 with additional con straints such as ta t ta 0 10 A constraint of this kind imposes an order of importance on qualitative variables 250 Golany RollAn Application Procedure for DEA REFERENCES 1 Aii I Charnes A Cooper WW Divine D Klopp GA and Stutz J 1988 An application of data envelopment analysis to US Army recruitment districts Applics Mgmt Sci Res A Edited by Schultz RL JAI Press 2 Banker RD Charnes A and Cooper WW 1984 Models for estimating technical and scale efficiencies in data envelopment analysis Mgmt Sci 309 10781092 3Banker RD Conrad RF and Strauss RP 1986 A comparative application of DEA and translog methods An illustrative study of hospital production Mgmt Sci 32i 3044 4 Banker RD and Maindiratta A 1986 Piecewise log linear estimation of efficient production surfaces Mgmt Sci 321 126135 5 Banker RD and Marcy RC 1986 Data envelopment analysis with categorical inputs and outputs Mgmt Sci 3212 16131627 6 Banker RD and More RC 0986 Efficiency analysis for exogenously fixed inputs and outputs Ops Res 344 5035 2 7 lkssent A Bessent W Charnes A Cooper WW and Thorogood R 1983 Program review for planning decisions through data envelopment analysis Educ Adm Q 192 5775 8 Bowlin WF 1987 Evaluating the efficiency of US AirForce real property maintenance activities J Opl Res Sac 382 127135 9 Bowlin WF Charnes A Cooper WW and Sherman HD 1985 Data envelopment analysis and regression approaches to efficiency estimation and evaluation Ann Ops Res 2 113138 10 Charnes A Clark T Cooper WW and Golany B 1985 A developmental study of data envelopment analysis in measuring the efficiency of maintenance units in the US Air Force Ann Ops Res 2 95112 1 I Charnes A and Cooper WW 1985 Preface to topics in data envelopment analysis Ann Ops Res 2 5994 12 Charnes A Cooper WW Golany B Sciford L and Stutz J 1985 Foundations of data envelopment analy sis for ParetoKoopmans efficient empirical production functions J Econometrics 30 91107 13 Charnes A Cooper WW Learner DB and Phillips FY 1985 Management science and marketing manage ment J Marketing 493 14 Charnes A Cooper WW Lewin AY Marcy RC and Rousseau JJ 1985 Sensitivity and stability analysis in DEA Ann Ops Res 2 139156 15 Charnes A Cooper WW and Rhodes E 1978 Measur ing efficiency of decision making units Fur J Ops Res 26 429444 16 Charnes A Cooper WW and Rhodes E 0981 Data envelopment analysis as an approach for evalualting program and managerial efficiency with an illustrative application to the program follow through experiment in US public school education Mgmt Sci 276 668697 17 Charnes A Cooper WW Seiford L and Stutz J 1982 A multiplicative model for efficiency analysis Socio Econ Plan Sci 165 222224 18 Charnes A Cooper WW Seiford L and Stutz J 1983 Invarlant multiplicative efficiency and piecewise CobbDouglas envelopments Ops Res Let 23 101103 19 Charnes A Cooper WW and Thrall RM 1986 Classi fying and characterizing efficiencies and inefficiencies in data envelopment analysis Ops Res Let 3 105110 20 Cook W and Seiford L 1978 Priority ranking and consensus formation Mgmt Sci 2416 17211732 21 Dyson RG and Thanassoulis E 1988 Reducing weight flexing in data envelopment analysis Y Opl Res Sac 396 563576 22 Fire R and Loveli CAK 1978 Measuring the technical efficiency of production J Econ Theory 191 150162 23 Farrell MJ 1957 The measurement of productive efficiency J R Statist SocA 1203 25328 24 Golany B 1988 A note on including ordinal relations among multipliers in data envelopment analysis Mgmt Sci 348 10291033 25 Leibenstein H 1966 Allocative efficiency versus X efficiency Am Econ Rev 563 392415 26 Lewin AY Marcy RC and Cook TJ 1982 Evaluating the administrative efficiency of courts Omega I04 401411 27 Nelson RR 1981 Research on productivity growth and productivity differences Dead ends and new departures J Econ Lit 193 102964 28 Rhodes E 1978 Data envelopment analysis and related approaches for measuring the efficiency of decision making units with an application to program following through in US education PhD dissertation Carnegy Melon University 29 Roll Y and Sachish A 1981 Productivity measurement at the plant level Omega 91 3742 30 Saaty TL 1980 The Analytic Hierarchy Process Planning Priority Setting Resource Allocation McGrawHill New York 31 Scott WR 1979 Measuring outputs in hospitals Measuring and Interpretation of Productivity National Academy of Science Washington DC 32 Scrrussi D 1987 Applying DEA to the IDF Air Force maintenance units MSc thesis Technionlsrael Institute of Technology 33 Sherman HD 1981 Measurement of hospital technical efficiencyA comparative evaluation of data envelop ment analysis and other approaches for locating inefficiencies in health care organizations DBA thesis Harvard University Graduate School of Business Administration 34 Smith P and Mayston D 1987 Measuring efficiency in the public sector Omega 153 181189 35 Thomas DL 1985 Auditing the efficiency of regulated companies through the use of data envelopment analy sis An application to electric cooperatives PhD thesis University of Texas Graduate School of Business Austin 36 Thompson RG Singleton FD Thrall RM and Smith BA 1986 Comparative site evaluations for locating a highenergy physics lab in Texas Interfaces 166 3549 ADDRESS FOR CORRESPONDENCE Dr B Goany Faculty of Industrial Engineering and Management Technion Israel Institute of Technology Hails 32 070 Israel Contents lists available at ScienceDirect European Journal of Operational Research journal homepage wwwelseviercomlocateejor Invited Review Assessing bank efficiency and performance with operational research and artificial intelligence techniques A survey Meryem Duygun Fethi a Fotios Pasiouras b a School of Management University of Leicester UK b School of Management University of Bath UK ARTICLE INFO ABSTRACT Article history Received 25 February 2009 Accepted 5 August 2009 Available online 11 August 2009 This paper presents a comprehensive review of 196 studies which employ operational research OR and artificial intelligence AI techniques in the assessment of bank performance Several key issues in the literature are highlighted The paper also points to a number of directions for future research We first discuss numerous applications of data envelopment analysis which is the most widely applied OR technique in the field Then we discuss applications of other techniques such as neural networks support vector machines and multicriteria decision aid that have also been used in recent years in bank failure prediction studies and the assessment of bank creditworthiness and underperformance 2009 Elsevier BV All rights reserved Keywords Artificial intelligence Banks Data envelopment analysis Efficiency Operational research Literature review 1 Introduction Banks play a central role in the economy They keep the savings of the public and finance the development of business and trade Furthermore numerous studies argue that the efficiency of financial intermediation affects economic growth while others indicate that bank insolvencies can result in systemic crises which have adverse consequences for the economy as a whole Thus the performance of banks has been an issue of major interest for various stakeholders such as regulators customers investors and the general public While bank performance has been traditionally evaluated on the basis of financial ratios advances in operational research OR and artificial intelligence AI have resulted in a shift towards such quantitative techniques Of course this is not surprising since OR has been extensively used in other applications in finance during the last half century Board et al 2003 This paper presents a comprehensive review of the use of OR and AI techniques in the assessment of bank efficiency and performance The rest of the paper is structured as follows Section 2 positions the survey within the existing literature and discusses our framework Section 3 discusses applications of data envelopment analysis DEA in the estimation of bank efficiency and productivity growth Section 4 presents applications of other OR and AI techniques in the prediction of bank failure and the assessment of bank creditworthiness and underperformance Section 5 summarizes our conclusions Corresponding author Tel 44 0 116 252 5328 fax 44 0 116 252 5515 Email addresses mfethileacuk MD Fethi fpasiourasbathacuk F Pasiouras 03772217 see front matter 2009 Elsevier BV All rights reserved doi101016jejor200908003 2007 discusses more recent applications of frontier techniques but his survey focuses only on studies that provide international comparisons of bank efficiency We differentiate our review from the above surveys by discuss ing applications of OR and AI techniques over the period 1998 early 2009 while focusing on bank performance We searched for papers in Scopus which is considered to be one of the largest ab stract and citation databases We consider only journal articles and we do not include working papers monographs dissertations or other publication outcomes Furthermore our search is limited to articles written in English We use a combination of various key words such as bank efficiency bank and data envelopment anal ysis bank performance bank and neural networks bank and artificial intelligence bank and operational or operations re search Additional studies were identified from crossreferencing and were manually collected We reviewed a total of 196 studies DEA is by far the most commonly used ORAI technique in assessing bank performance and we identified 151 studies that use DEAlike techniques to estimate various measures of bank efficiency and productivity growth and 30 studies that provide similar estimates at the branch level2 We also identified 15 studies that use classification techniques such as neural networks support vector machines mul ticriteria decision aid decision trees nearest neighbours to predict bank failure or assess bank creditworthiness and bank underper formance These studies were published in a total of 73 journals however around 58 of them appeared in just 12 journals The most frequent sources of publication are the European Journal of Operational Research 19 and the Journal of Banking and Finance 15 followed by Applied Financial Economics 13 Managerial Fi nance 11 Applied Economics 9 Expert Systems with Applications 9 the Journal of Productivity Analysis 9 and the Journal of Eco nomics and Business 83 3 DEA and bank efficiency DEA is a mathematical programming technique for the develop ment of production frontiers and the measurement of efficiency relative to these frontiers Each bank is assigned an efficiency score between 0 and 1 with higher scores indicating a more efficient bank relatively to other banks in the sample One of the wellknown advantages of DEA is that it works rela tively well with small samples Other advantages of DEA are that it does not require any assumptions to be made about the distribu tion of inefficiency and it does not require a particular functional form on the data in determining the most efficient banks However DEA is also subject to few limitations Two of the bestknown shortcomings are that DEA assumes data to be free of measure ment error and that it is sensitive to outliers Coelli et al 2005 also point out that i having few observations and many inputs andor outputs will result in many firms appearing on the DEA frontier ii treating inputsoutputs as homogenous commodities when they are heterogeneous may bias the results iii not accounting for differences in the environment may give misleading results iv standard DEA does not control for multiperiod optimi sation or risk managerial decision making Our survey shows that recent DEA studies have examined al most all of the banking sectors around the world A few recent studies provide crosscountry evidence Most of them examine banks from the large EU banking sectors Casu and Molyneux 2003 Beccalli et al 2006 LozanoVivas et al 2002 examine 10 EU countries Bergendahl 1998 focuses on Nordic countries while Pasiouras 2008a and Tanna 2009 examine international datasets 31 Methodological issues 311 Efficiency measures Most of the studies focus on the technical efficiency of banks eg LozanoVivas et al 2002 Drake et al 2006 Pasiouras 2008ab This efficiency measure indicates whether a bank uses the minimum quantity of inputs to produce a given quantity of outputs or maximizes the output quantity given a certain quantity of inputs However when price data for the inputs andor outputs are available one can also estimate cost andor profit efficiency mea sures4 Cost efficiency is the product of technical efficiency and allocative efficiency The latter refers to the ability of a bank to use the optimum mix of inputs given their respective prices Con sequently cost efficiency shows the ability of a bank to provide services without wasting resources as a result of technical or allo cative inefficiency As can be seen in Table 2 of the Supplementary material Appendix A around 35 studies present measures of DEA cost efficiency eg TortosaAusina 2002a Isik and Hassan 2002 2003a Pastor and Serrano 2006 propose the decomposition of cost inefficiency into composition inefficiency and intraspecialisation inefficiency The first component indicates the part of inefficiency due to the composition of specialisations of the banks in each banking sector The second component reveals the inefficient use of resources within each of the specialisation selected Prior 2003 also deviates from the above studies by calculating mea sures of short and longrun cost inefficiency as well as capacity inefficiency for Spanish banks The first refers to the case that a subset of inputs are fixed and impossible to modify in the short run Longrun inefficiency estimates are obtained under the assumption that inputs are variable and under the control of the company Finally capacity inefficiency obtained by the ratio of longrun to shortrun inefficiency refers to excess in costs as a re sult of inappropriate level in fixed inputs Similar concepts along with an application in the Indian banking sector are discussed in Sahoo and Tone 2009 Estimations of profit efficiency with DEA are rather limited in the literature One potential reason is the difficulty in collecting reliable and transparent information for output prices Further more the decomposition of profit efficiency into technical and allocative efficiency is not straightforward Coelli et al 2005 Fare et al 2004 propose the solution of two sets of linear programmes In the first a profit maximizing DEA is solved to measure profit efficiency In the second DEA problem technical efficiency is mea sured on the basis of a directional distance function that allows the simultaneous adjustment of inputs and outputs Kirkwood and Nahm 2006 also estimate profit efficiency although they use in put prices only Therefore in a sense they calculate a measure of efficiency that is similar to Berger and Mester 1997 alternative profit efficiency which is commonly used in the stochastic frontier analysis literature The studies of Maudos and Pastor 2003 and 2 Our survey focuses on studies that examine banking institutions as a whole however we also discuss studies on branch efficiency in Section 327 as one could argue that the efficiency of individual branches can influence the performance of banks as a whole 3 In the sections that follow we discuss various issues surrounding these studies while additional information is available in the Supplementary material Appendix A 4 One can also estimate revenue efficiency which is similar to profit efficiency In both cases both inputs and output prices are required The difference is that in the former measure the aim is to maximize revenues rather than profits ie revenues minus costs We are no aware of DEA studies focusing on revenue efficiency so we do not discuss this issue further Readers interesting in revenue efficiency could see Coelli et al 2005 for further details 190 MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 Ariff and Can 2008 provide estimates of both standard and alter native profit efficiency Finally around 40 studies obtain estimates of total factor pro ductivity TFP growth eg Casu et al 2004 Tanna 20095 This measure is usually decomposed further into technological change ie shift in the best practice frontier and technical efficiency change Furthermore in most cases technical efficiency change is disaggregated into pure technical efficiency change and scale effi ciency change under the variable returns to scale assumption dis cussed below 312 Constant vs variable returns to scale DEA can be implemented by assuming either constant returns to scale CRS or variable returns to scale VRS In their seminal study Charnes et al 1978 proposed a model that had an input orientation and assumed CRS This model returns a score that indi cates the overall technical efficiency OTE of each bank Banker et al 1984 suggested the use of variable returns to scale VRS that decomposes OTE into product of two components pure tech nical efficiency PTE and scale efficiency SE The former relates to the ability of managers to utilize firms given resources while the latter refers to exploiting scale economies by operating at a point where the production frontier exhibits CRS In most of the recent papers DEA models are estimated using the assumption of VRS while arguing that CRS is only appropriate when all firms are operating at an optimal scale6 Nevertheless other studies argue in favour of CRS rather than VRS or mention that caution is necessary when using the VRS formulation eg Noulas 1997 Avkiran 1999 Soteriou and Zenios 1999a Consequently many studies report the results obtained under both CRS and VRS assumptions eg Canhoto and Dermine 2003 Casu and Molyneux 2003 313 Outputinput orientation Technical efficiency can be estimated under either an inputori ented or outputoriented approach7 As Coelli et al 2005 point out the inputoriented technical efficiency measures address the question By how much can input quantities be proportionally reduced without changing the output quantities produced p 137 In con trast the outputoriented measures of technical efficiency address the question By how much can output quantities be proportionally expanded without altering the input quantities used p 137 By far studies in banking obtain efficiency estimates under the inputoriented approach8 This is most likely due to the assumption that bank managers have higher control over inputs eg personnel expenses rather than outputs eg loans income etc However there are also some studies that adopt the outputoriented approach eg Ataullah et al 2004 Ataullah and Le 2006 or report the results from both eg Casu and Molyneux 2003 Beccalli et al 20069 Fi nally others adopt a nonoriented slackbased measure SBM that assumes the simultaneous contraction of inputs and expansion of outputs eg Avkiran 2009ab 314 Selection of inputs and outputs There is an ongoing discussion in the banking literature regard ing the proper definition of inputs and outputs Using the words of Bergendahl 1998 There have been almost as many assumptions of inputs and outputs as there have been applications of DEA p 235 Berger and Humphrey 1997 identify two main approaches for the selection of inputs and outputs These are the production ap proach and the intermediation approach The first assumes that banks produce loans and deposits account services using labour and capital as inputs and that the number and type of transactions or documents processed measure outputs The second approach perceives banks as financial intermediaries between savers and investors Berger and Humphrey 1997 argue that neither of these two approaches is perfect because they cannot fully capture the dual role of financial institutions as providers of transactionsdoc ument processing services and also being financial intermediaries They point out that the production approach may be somewhat better for evaluating the efficiencies of bank branches and the intermediation approach may be more appropriate for evaluating financial institutions as a whole Furthermore there are difficulties in collecting the detailed transaction flow information required in the production approach As a result the intermediation approach is the one favoured in the literature However there is a controversy even within this approach con cerning the role of deposits Berger and Humphrey 1997 Conse quently some studies use only earning assets as outputs a selection that is in line with the asset approach of Sealey and Lind ley 1977 while others consider deposits as an additional output a selection that is more closely related to the socalled valueadded approach We find around 95 applications in bank efficiency where the monetary value of deposits is part of the input vector and 20 applications where deposits are part of the output vector10 Around 30 studies use interest expenses as an input without using the stock of deposits eg Weill 2004 Avkiran 2009ab In another 7 applica tions the stock of deposits is used as an output and the interest ex pense paid on deposits constitutes an input eg Maudos et al 2002 Chen et al 2005 Furthermore 7 studies use time deposits and sav ing deposits as input and demand deposits as output eg Bauer et al 1998 Gilbert and Wilson 1998 Finally in a few applications the deposits are included as both an input and an output eg TortosaAusina 2002a More recently some studies adopted another variation of the intermediation approach This is the socalled profitoriented or operating approach which defines revenue components eg interest income noninterest income etc as outputs and cost components eg personnel expenses interest expenses etc as in puts11 Drake et al 2006 mention that from the perspective of an 5 Almost all studies use the DEAlike Malmquist index Zen and Baldan 2008 use the Luenberger Indicator that is a generalization of the Malmquist Index 6 Reasons that may not allow a firm to operate at optimal scale include among others imperfect competition government regulations constrains on finance etc Coelli et al 2005 7 If price data are available then under the under cost minimization objective ie cost efficiency one obtains inputoriented measures of technical efficiency and input mix allocate efficiency Revenue maximization ie revenue efficiency results is outputoriented technical efficiency and outputmix allocative efficiency measures In the case of profit maximization profit efficiency technical efficiency can be obtained under either the input or outputoriented assumption Coelli et al 2005 8 Additional information is available in the Supplementary material Appendix A In contrast to efficiency estimates productivity measures are in several cases obtained using an outputoriented Malmquist index However as discussed in Coelli et al 2005 p 80 although the values of the components and consequently their contribution to overall productivity change may differ The overall TFP change measure will be the same regardless of the orientation ie input or output that is imposed 9 It should be mentioned that the inputoriented and outputoriented measures always provide the same value under CRS but they are unequal when VRS is assumed However Coelli et al 2005 mention that since linear programming does not suffer from statistical problems the choice of an appropriate orientation is not as important as in the case of econometric approaches Furthermore in many instances the choice of orientation has only a minor influence upon the scores obtained Coelli and Perelman 1996 10 This discussion refers to banks as a whole and not branches The number of applications does not match the number of studies as in several cases there are numerous models developed in each study Furthermore in some cases deposits or noninterest expenses are not considered in the analysis In one case Prior 2003 the output was the number of current and saving accounts rather than the monetary value of deposits ie stock 11 See Chu and Lim 1998 Avkiran 1999 Sturm and Williams 2004 Das and Ghosh 2006 Drake et al 2006 Ataullah and Le 2006 and Pasiouras 2008ab MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 191 inputoriented DEA relative efficiency analysis the more efficient units will be better at minimizing the various costs incurred in generating the various revenue streams and consequently better at maximizing profits p 1451 They also argue that this approach can be more appropriate in capturing the diversity of strategic responses by financial firms in the face of dynamic changes in competitive and environmental conditions Furthermore Luo 2003 calculates a measure of marketability efficiency in an attempt to capture the va lue of the bank in the stock market In this case revenue and profits are considered inputs whereas market value earnings per share and stock price are outputs As discussed before with the exception of deposits there is a general agreement about the main categories of inputs and out puts however this does not necessarily imply that there is consis tency with respect to the specific inputsoutputs used in various studies For instance the traditional inputs are fixed assets person nel12 and in many cases they are deposits eg Isik and Hassan 2002 Maudos and Pastor 2003 Casu and Girardone 2004 Havryl chyk 2006 However some studies use branches eg Chen 2001 loan loss provisions eg Drake et al 2006 Pasiouras 2008b and equity eg Chu and Lim 1998 Mukherjee et al 2001 Sturm and Williams 2004 Pasiouras 2008a as additional or alternative inputs Chen 2001 disaggregates deposits into current deposits and time deposits while Das and Ghosh 2006 use demand savings and fixed deposits Casu and Girardone 2006 and Beccalli et al 2006 use to tal costs as a single input while Casu and Molyneux 2003 use two inputs namely total costs and total deposits ie customers and shortterm funding Several studies use two outputs usually loans and other earn ing assets eg Casu and Molyneux 2003 Casu and Girardone 2004 200613 However some studies disaggregate loans into var ious categories such as housing loans and other loans eg Sturm and Williams 2004 real estate loans commercial loans and per sonal loans eg Mukherjee et al 2001 Fare et al 2004 or shortterm and longterm loans Isik and Hassan 2002 Others disaggregate other earning assets into investments and liquid as sets Tsionas et al 2003 or investment in government securities and investments in public and private enterprises eg Chen 2001 Canhoto and Dermine 2003 use the number of branches as an additional output under the assumption that it represents an additional value for retail customers Finally recent studies in clude noninterest income or offbalancesheet items as additional outputs eg Isik and Hassan 2002 2003a Sturm and Williams 2004 TortosaAusina 2003 Havrylchyk 2006 Pasiouras 2008b14 Finally Halkos and Salamouris 2004 propose an approach that deviates from the above literature They use an output vector that consists of five financial ratios and no inputs Their underlying hypothesis is that inputs are considered similar and equal for all banks as they operate in the same markets for money and service The comparison with standard DEA models developed under either CRS or VRS assumptions shows that the DEA ratio model has the highest discrimination power 315 Adjusting for the environment Coelli et al 2005 discuss four approaches that can be used to incorporate environmental variables in DEA applications15 The first method by Banker and Morey 1986 requires the environmen tal variables to be ordered from the least to the most harmful ones for efficiency Then the efficiency of a given firm is compared with those firms in the sample that have a value of the environmental var iable which is less than or equal to the given firm This ensures that banks are not compared with peers operating in a more favourable environment The second method by Charnes et al 1981 requires the decisionmaker to i divide the sample into subsamples and solve DEA problems for each subsample ii project all observed data points into their prospective frontiers and iii solve a single DEA using the projected points and assess any difference in the mean efficiency of the two subsamples According to Coelli et al 2005 the following two problems are common in both methods i sam ple splitting reduces the comparison set and ii only one environ mental variable can be considered in each case thereby limiting the scope of the analysis Under the third method the environmental variables are in cluded directly into the DEA problem as nondiscretionary inputs if it is believed to have a positive effect on efficiency or out puts if they have a negative effect on efficiency The disadvan tage of this approach is that one must know a priori the direction of the influence a shortcoming that is also applicable in the case of the first method Alternatively the environmental variables can be included as nondiscretionary neutral variables using an equality form The shortcoming of this approach is that it can reduce the reference set for each firm Recent applications of the above approaches in banking can be found in Pastor 1999 and LozanoVivas et al 2001 2002 The fourth method that is discussed in Coelli et al 2005 is the twostage approach This involves a DEA problem with traditional inputs and outputs in the first stage In the second stage the effi ciency scores obtained are regressed on the environmental vari ables While this approach has been frequently used in the banking literature with numerous applications it fails to adjust the efficiency measures Therefore it is more suitable when the objective is to examine the correlations of efficiency with various factors see Section 321 rather than provide the basis for absolute comparisons across different environments Finally Pastor 2002 Drake et al 2006 Avkiran 2009b and Thoraneenitiyan and Avkiran 2009 adjust the bank efficiency scores for risk andor external environmental factors using a mul tistage DEA In this case the application starts with the estimation of a DEA model with traditional inputs and outputs Then using the slacks from the DEA model they quantify the effect of the operat ing environment and adjust the initial dataset inputs andor out puts Finally they rerun the initial DEA model using the adjusted data 32 Topics of interest 321 Determinants of efficiency Several studies attempt to investigate the factors that influence the efficiency of banks Some studies examine only bankspecific factors and others examine both bankspecific attributes and envi ronmental determinants Commonly found bankspecific factors are size profitability capitalisation loans to assets Casu and Molyneux 2003 Casu and Girardone 2004 Ataullah and Le 2006 Ariff and Can 2008 12 Some studies use the number of personnel eg Maudos and Pastor 2003 Pasiouras 2008b However others rely on personnel expenses due to data unavailability eg Bergendahl 1998 LozanoVivas et al 2002 Drake et al 2006 13 Other earning assets normally include various items such as government securities investment securities trading securities other securities equity invest ments and other investments 14 Isik and Hassan 2003a mention among others that ignoring such nontraditional outputs may penalize banks that are heavily involved in these activities The reason is that while the resources used to produce these nontraditional outputs are part of the input vector the outputs generated using these inputs are not included in the output vector 15 According to Coelli et al 2005 environmental variables can include ownership differences location characteristics labour union power and government regulations and in a sense any factors that can influence the efficiency of the firm without being traditional inputs or under the control of managers 192 MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 Isik and Hassan 2003a examine additional characteristics such as the educational profile of bank personnel and CEOChairman affiliation Some studies examine whether age is related to efficiency eg Isik and Hassan 2003a Canhoto and Dermine 2003 Isik 2008 Furthermore Isik and Hassan 2002 Casu and Girardone 2004 and Pasiouras 2008b examine the international presence of Turkish Italian and Greek banks respectively Countryspecific factors include market concentration presence of foreign banks ratio of private investments to GDP fiscal deficits to GDP GDP growth Hauner 2005 Ataullah and Le 2006 Pasiouras 2008a examined the relationship between technical efficiency and regulations related to capital adequacy private monitoring banks activities deposit insurance schemes supervisory power and bank entry into the industry More recently Tanna 2009 investigated the impact of foreign direct investments on TFP growth These studies use the twostage approach discussed above First they use DEA to obtain efficiency estimates Then in a second stage the DEA scores are regressed on a number of explanatory variables using Tobit eg Hauner 2005 OLS Ataullah and Le 2006 GMM Ataullah and Le 2006 or GLS Isik and Hassan 2003a regression The rationale to use Tobit lies on the fact that the efficiency scores are bounded between 0 and 1 and hence noncensored estimates will be biased Ataullah and Le 2006 among others mention that it is not necessary to use Tobit as long as the efficiency scores can be transformed by taking the natural logarithm of efficiency score1 efficiency score Whether or not censored regression is used the most important argument comes from Simar and Wilson 2007 who point out that the covariates in the secondstep regression are obviously correlated with the one side error terms from the firststep as otherwise there would be no need for the secondstep regression Furthermore the covariates in the secondstep are likely to be highly correlated with the covariates in the firststep This means that the errors and the covariates in the firststep cannot be independent Thus Simar and Wilson conclude that the likelihood that is maximized is not the correct one unless one takes account of the correlation structure Casu and Molyneux 2003 present an early attempt to account for the problems that can emerge in the twostage DEA method using a bootstrap approach However Simar and Wilson 2007 mention that the simple bootstrap is not enough to overcome the drawbacks pointed out above They propose an algorithm that uses a double bootstrap procedure and they present an application to US banks Brissimis et al 2008 and Delis and Papanikolaou 2009 adopt this approach to examine the determinants of efficiency in the new EU banking sectors 322 Stock returns and efficiency The relationship between stock returns and publicly available information has attracted considerable attention in the accounting and finance literature Whilst most of the studies examine whether earnings reflect some of the information in stock prices recent research has however shifted towards the use of additional data such as accruals revenues economic value added and efficiency to understand how they affect stock prices and returns DEA applications that were published over the period of our survey and investigate the relationship between bank efficiency and stock returns examine Australia Kirkwood and Nahm 2006 Greece Pasiouras et al 2008 Malaysia Sufian and Majid 2006 Singapore Chu and Lim 1998 Turkey Erdem and Erdem 2008 and Spain Guzman and Reverte 2008 Furthermore in a crosscountry setting Beccalli et al 2006 provide evidence from France Germany Italy Spain UK that is the five principal EU banking sectors In most cases the results of these studies indicate a positive relationship between stock returns and efficiency changes Furthermore Beccalli et al 2006 reveal that the explanatory power of the model with DEA scores is higher than that of a model that uses the return on equity ROE as a measure of performance 323 Bank ownership A number of studies compare the efficiency of banks across different ownership types One approach used in the literature is to split the sample and compare the means of the different ownership groups Another approach is to incorporate dummy variables in a second stage analysis as the one described in Section 321 Some studies compare domestic and foreign banks Havrylchyk 2006 finds that greenfield banks are more efficient than domestic banks in Poland whereas foreign banks that acquired domestic ones have not successfully increased their efficiency Sturm and Williams 2004 report that foreign banks in Australia are more efficient than domestic ones Isik and Hassan 2003a find that foreign banks are more efficient than private domestic Turkish banks while Isik 2008 reports similar results for TFP growth estimates However Ataullah and Le 2004 find that prior to the financial liberalization of 19911992 foreign banks were less efficient than domestic ones in India and Pakistan nevertheless the opposite picture emerged after this period Other studies examine the efficiency of stateowned banks Many of them find that stateowned banks are less efficient than other banks For instance GarciaCestona and Surroca 2008 find that Spanish banks controlled by insiders ie managers and workers are more efficient than the ones controlled by public administrations Ariff and Can 2008 find that jointstock banks in China are more cost and profitefficient than stateowned banks Chen 1998 in Taiwan and Mercan et al 2003 in Turkey also report that the efficiency of privatelyowned banks is higher than that of stateowned banks Similarly Isik 2008 focuses on TFP growth and finds that the growth of private banks is more than double that of state banks In contrast other studies report that the efficiency of private banks is lower than that of stateowned banks in Turkey Isik and Hassan 2003a India Sathye 2003 and Austria and Germany Hauner 2005 324 Corporate events and efficiency Another part of the literature relates DEA efficiency estimates to corporate events such as mergers and acquisitions andor bankruptcy Some of the studies compare the efficiency of acquirers and their targets and examine whether mergers improve efficiency Avkiran 1999 reports that acquiring banks are more efficient than acquired Australian banks However the results of this study also indicate that acquiring banks do not always maintain their premerger efficiency and there is mixed evidence on the extent to which the benefits of efficiency are passed to the public The results of AlSharkas et al 2008 indicate that merged US banks are on average more technically efficient and that they experience higher productivity growth than nonmerged banks Hahn 2007a finds evidence that Austrian banks which engaged in domestic merger deals achieved a higher productive efficiency level than banks which did not participate in such deals The results also show that merger gains remain significant over a longer period of time more than 5 years but there is slight tendency to level off Sherman and Rupert 2006 also focus on the efficiency gains from bank mergers but they concentrate on the branch rather than the banklevel They find that there are opportunities to reduce operating costs Such benefits however are not realised until 4 years after the merger Kohers et al 2000 follow a different approach to test whether Xefficiencies influence the markets assessment of bank mergers They find that the abnormal returns of bidders around the announcement period were associated with both the target banks profit efficiency and cost efficiency as well as the difference between the average peer cost efficiency score and the corresponding targets cost16 Wheelock and Wilson 2000 also relate efficiency with bank acquisitions However in their case DEA efficiency scores are used as an input in hazard models that capture the acquisition and failure likelihood They find that inefficiency increases the risk of failure while reducing the probability of a US banks being acquired Using a sample of 245 US banks Luo 2003 also reports that overall technical efficiency can be useful in predicting bank failures 325 Regulatory reformliberalization and efficiency A number of studies examine the impact of regulatory reform and liberalization initiatives on bank efficiency and productivity The studies that we review seem to indicate a positive relationship in most countries including India and Pakistan Ataullah et al 2004 Australia Avkiran 1999 Sturm and Williams 2004 US Mukherjee et al 2001 China Chen et al 2005 Greece Tsionas et al 2003 Rezitis 2006 Taiwan Wang and Huang 2007 Korea Gilbert and Wilson 1998 Turkey Isik and Hassan 2003b and the new EU countries Brissimis et al 2008 However Hauner 2005 finds no evidence that productivity is related to deregula tion in Germany and Austria while Sathye 2002 argues that there can be a limit of deregulation after which no further productivity gains could be possible due to deregulation Some other interesting observations can be summarized as fol lows The banking sector reform in the newly acceded EU countries had a positive impact on bank efficiency while the effect of reform on TFP growth was significantly only toward the end of the reform process Brissimis et al 2008 The implementation of financial liberalization programmes may enable foreign banks to overcome the liability of foreignness and enhance their resource utilization Ataullah and Le 2004 The impact of deregulation will not neces sarily be the same across different bank ownerships such as state domestic foreign Isik and Hassan 2003b 326 Comparison of frontier techniques A few studies compare alternative frontier techniques over the period of our survey Bauer et al 1998 use a sample of US banks to compare DEA SFA TFA thick frontier approach and DFA dis tributionfree approach over six consistency criteria They find that DEA yields much lower average efficiencies ranks the banks differently and identifies the best and worst banks differently from parametric methods Furthermore compared to DEA the parametric measures were more highly correlated with the tradi tional nonfrontier performance measures Huang and Wang 2002 provide another comparison of SFA DFA and DEA using data from Taiwanese banks They also argue that the choice of the frontier approach can result in different conclusions Weill 2004 provides more recent evidence using a sample from five European countries However this crosscountry setting does not alter the main conclusion of lack of robustness across different approaches confirming the results of the previous two studies Delis et al 2009 provide further support to these findings using a dataset of Greek banks The study of Beccalli et al 2006 also provides useful insights for the differences of SFA and DEA While this study does not focus on the comparison between these two methods it finds that while changes in the stock prices of banks reflect percentage changes in DEA cost efficiency scores this trend is less clear with SFA efficiency estimates In contrast Fiordelisi 2008 reports that SFA cost efficiency estimates explain better the variations in shareholder value creation than those derived from DEA Casu et al 2004 compare productivity growth estimates ob tained through parametric and nonparametric approaches Their study seems to indicate that in this case the differences are not as large as in the efficiency studies They conclude that although the alternative methodologies produce in some cases opposing findings as for the sources of productivity for individual years in general they do not yield noticeably different results in terms of identifying the components of the productivity growth of EU banks during the period of their study 327 Efficiency of bank branches While most of the above studies focus on the efficiency of bank ing institutions as a whole a related strand of the literature exam ines the efficiency of bank branches17 In general branches are predominantly regarded as production units see eg Camanho and Dyson 2005ab 2006 Soteriou and Zenios 1999b Golany and Storbeck 1999 Athanassopoulos and Giokas 2000 where output is measured by the number of various accounts andor transactions as contrast to the intermediation approach employed by Portela and Thanassoulis 2005 2007 and Portela et al 2003 where branches are seen as intermediaries and thus outputs are measured in mone tary terms There are studies which consider the changing role of bank branches from a predominantly transactionbased one to a sales oriented role Cook et al 2000 and Cook and Hababou 2001 dis tinguish sales and services transactions functions of bank branches in a Canadian bank where inputs are usually shared be tween these two functions To model the shared resources Cook et al 2000 extend the usual methodology to develop a model which incorporates the best resource split and optimises the aggre gate efficiency score Further Cook and Hababou 2001 model the shared resources concept by adopting the Additive DEA model In a recent novel application Portela and Thanassoulis 2007 assess the branches of a Portuguese bank in terms of their performance in their new roles by specifying efficiency measures which con sider sales transactional activities as well as bank branch profit efficiency Service quality is also a significant dimension of bank branches performance Two different ways are developed in empirical applications to consider service quality For instance Soteriou and Zenios 1999b Golany and Storbeck 1999 and Athanassopo ulos and Giokas 2000 include quality variables in efficiency anal ysis whereas Portela and Thanassoulis 2007 use a post hoc analysis to compare service quality index with efficiency measures They find that service quality is positively related with operational and profit efficiency The diversity of environments in which the branches are oper ating is also considered in the literature Paradi and Schaffnit 2004 considered the role of the environmental parameters that are outside the management control in the commercial branches of a large Canadian bank and incorporated risk and economic growth rate of the region as two nondiscretionary factors On the other hand Camanho and Dyson 2006 assess the impact of environmental factors and regional managerial policies on branches productivity through the construction of an index that reflects the relative performance of bank branches operating in four different regions in Portugal Das et al 2009 introduce the concept of spatial efficiency for each region relative to the nation and thus measure the effects of differences in the regional characteristics on the efficiency of bank branches across four 16 Only cost efficiency was estimated with DEA in this study Profit efficiency was estimated using stochastic frontier analysis As in previous studies Kohers et al 2000 also provide comparisons of efficiency across different groups They find that acquiring US bank holding companies BHCs are less costefficient compared to their targets Also industry peers operate with greater cost efficiencies that then bidder and target BHCs 17 See Table 3 in the Supplementary material Appendix A for information on branch efficiency studies 194 MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 195 MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 associated with nonfailed banks and the unsafe traits failed banks known as features Then these features are used to vote on each bank and classify it as failed or nonfailed Lanine and Vander Vennet 2006 argue that the general rule of classifying safe and unsafe cells and assigning default probabilities works poorly when the number of banks falling into mixed cells is high To avoid this problem they suggest a formula to compute the probability of default directly without preliminary classification of cells as safe or unsafe Some researchers use soft computing approaches with the aim of synthesizing the human ability to tolerate and process uncer tain imprecise and incomplete information during the decision making Tung et al 2004 Soft computing usually integrates var ious techniques that may originate from different disciplines such as fuzzy logic19 neural networks machine learning etc in various combinations that will allow the researcher to exploit the strengths of individual techniques Ravi Kumar and Ravi 2007 A popular ap proach is the integration of neural network and fuzzy sets to develop a neural fuzzy network For example Tung et al 2004 propose a Generic Selforganizing Fuzzy Neural Network GenSoFNN based on the compositional rule of inference CRI to predict bank failure Multiple discriminant analysis MDA and logistic regression analysis LRA are two traditional techniques that are commonly used as benchmarks in classification studies eg Swicegood and Clark 2001 Pasiouras et al 2007 In contrast Tung et al 2004 benchmark their models against Cox regression which estimates the dependence of the risk of failure on banks characteristics and of the evolution of that risk over time 42 Variables In most of the cases the variables are selected on the basis of the CAMEL model and include measures for Capital strength Asset quality Earnings and Liquidity20 The number of variables differs significantly among studies Some researchers start from a large list of variables and then use statistical screening eg KruskalWallis or dimension reduction eg factor analysis to end up with a re duced set of variables In addition to the financial variables some studies include additional characteristics related to ownership eg of shares held by the government market information eg stock price in previous year auditing and country eg Heritage index 43 Classification results Many of the studies review in this section report quite satisfac tory classification accuracies and in several cases the proposed OR and AI methods outperform the traditional techniques ie MDA LRA Cox regression Furthermore the studies that focus on bankruptcy prediction in general perform more accurately than those that deal with credit ratings However this is not sur prising for two reasons First while classifying the banks in more groups is more informative it becomes more difficult to discrimi nate among groups in the intermediate categories Second credit ratings may incorporate additional qualitative information and hu man judgment that cannot be captured by the conventional vari ables used in most studies In any case comparisons across different studies should be trea ted with extreme caution One reason is that different studies make different assumptions about the prior probabilities of group membership as well as misclassification costs A second reason is that different studies use different approaches to validate the developed models For instance some researchers simply split the total sample in training and holdout datasets without account ing for the stability of models over time others use resampling techniques eg kfold crossvalidation and others use a holdout sample from a future period However it should kept in mind that classification ability is likely to overstate predictive ability Thus a superior approach would require that the model be validated against a future period as this approach more closely reflects a real world setting 5 Concluding remarks We have presented a comprehensive review of applications of operational research and artificial intelligence techniques in the assessment of bank performance by discussing a total of 196 stud ies published between 1998 and early 2009 We have classified the studies in two main categories The first uses DEA and the DEAlike Malmquist index to esti mate the efficiency and productivity growth of banks and bank branches We have discussed various methodological issues such as the estimated measures of efficiency the underlying assump tions of the estimated models and the selection of inputs and out puts We also discussed the main topics of interest including the relationship between ownership and efficiency stock returns and efficiency the determinants of efficiency the efficiency of bank branches amongst others Three of the main conclusions are that i profit efficiency and capacity efficiency have received quite lim ited attention in DEA studies in banking ii most studies that use a twostage DEA do not employ appropriate bootstrapping tech niques and their results may be biased iii there is much diversity among studies with respect to the selection of input and outputs As in the case of banks as a whole cost and profit efficiency has re ceived considerably less attention in branch efficiency studies Fur thermore an area of research deserving attention would be the estimation of bank branches efficiency over successive time periods Studies falling into the second category attempt to develop classification models to predict failure the credit ratings of banks and identify underperformers We have found both countryspe cific and crosscountry studies and applications of techniques that originate from various disciplines Most of the studies rely heavily on financial information although in some cases nonfinancial vari ables are also used Given the differences in the approaches em ployed to validate the models comparisons of classification accuracies across studies should be treated with caution We find only a few studies that propose the combination of the predictions of individual models into integrated metaclassifiers and we be lieve that this is an area of research that is worthy of further attention Appendix A Supplementary data Supplementary data associated with this article can be found in the online version at doi101016jejor200908003 References AlSharkas AA Hassan MK Lawrence S 2008 The impact of mergers and acquisitions on the efficiency of the US banking industry Further evidence Journal of Business Finance and Accounting 35 5070 Ariff M Can L 2008 Cost and profit efficiency of Chinese banks A nonparametric analysis China Economic Review 19 260273 Ataullah A Cockerill T Le H 2004 Financial liberalization and bank efficiency A comparative analysis of India and Pakistan Applied Economics 36 19151924 Ataullah A Le H 2004 Financial repression and liability of foreignness in developing countries Applied Economics Letters 11 545549 19 The fuzzy set theory aims to classify subjective reasoning and assign degrees of possibilities in reaching conclusions Fuzzy logic can also be used to obtain fuzzy if then rules 20 M in CAMEL refers to management quality and it is rarely used in empirical studies due to difficulties in assessing management quality 196 MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 Ataullah A Le H 2006 Economic reforms and bank efficiency in developing countries The case of the Indian banking industry Applied Financial Economics 16 653663 Athanassopoulos AD Giokas D 2000 The use of data envelopment analysis in banking institutions Evidence from the commercial bank of Greece Interfaces 30 2 8195 Avkiran NK 1999 The evidence on efficiency gains The role of mergers and the benefits to the public Journal of Banking and Finance 23 9911013 Avkiran NK 2009a Opening the black box on efficiency analysis An illustration with UAE banks Omega 37 930941 Avkiran NK 2009b Removing the impact of environment with unitsinvariant efficient frontier analysis An illustrative case study with intertemporal panel data Omega 37 535544 Banker RD Charnes A Cooper WW 1984 Some models for estimating technical and scale inefficiencies in data envelopment analysis Management Science 30 10781092 Banker RD Morey RC 1986 Efficiency analysis for exogenously fixed inputs and outputs Operations Research 34 513521 Bauer PW Berger AN Ferrier GD Humphrey DB 1998 Consistency conditions for regulatory analysis of financial institutions A comparison of frontier efficiency methods Journal of Economics and Business 50 85114 Beccalli E Casu B Girardone C 2006 Efficiency and stock performance in european banking Journal of Business Finance and Accounting 33 245262 Bergendahl G 1998 DEA and benchmarks An application to Nordic banks Annals of Operations Research 82 233249 Berger AN 2007 International comparisons of banking efficiency Financial Markets Institutions and Instruments 16 119144 Berger AN Humphrey DB 1997 Efficiency of financial institutions International survey and directions for future research European Journal of Operational Research 98 175212 Berger AN Mester LJ 1997 Inside the black box What explains differences in the efficiencies of financial institutions Journal of Banking and Finance 21 895947 Board J Sutcliffe C Ziemba WT 2003 Applying operations research techniques to financial markets Interfaces 33 1224 Boyacioglu MA Kara Y Baykan OK 2009 Predicting bank financial failures using neural networks support vector machines and multivariate statistical methods A comparative analysis in the sample of savings deposit insurance fund SDIF transferred banks in Turkey Expert Systems with Applications 36 16851694 Brissimis SN Delis MD Papanikolaou NI 2008 Exploring the nexus between banking sector reform and performance Evidence from newly acceded EU countries Journal of Banking and Finance 32 26742683 Camanho AS Dyson RG 2005a Cost efficiency measurement with price uncertainty A DEA application to bank branch assessments European Journal of Operational Research 161 432446 Camanho AS Dyson RG 2005b Cost efficiency Production and valueadded models in the analysis of bank branch performance Journal of the Operational Research Society 56 483494 Camanho AS Dyson RG 2006 Data envelopment analysis and Malmquist indices for measuring performance Journal of Productivity Analysis 26 3549 Camanho AS Dyson RG 2008 A generalisation of the Farrell cost efficiency measure applicable to nonfully competitive settings Omega 36 147162 Canbas S Cabuk A Kilic SB 2005 Prediction of commercial bank failure via multivariate statistical analysis of financial structures The Turkish case European Journal of Operational Research 166 528546 Canhoto A Dermine J 2003 A note on banking efficiency in Portugal New vs Old banks Journal of Banking and Finance 27 20872098 Casu B Girardone C 2004 Financial conglomeration Efficiency productivity and strategic drive Applied Financial Economics 14 687696 Casu B Girardone C 2006 Bank competition concentration and efficiency in the single European Market The Manchester School 74 441468 Casu B Girardone C Molyneux P 2004 Productivity change in European banking A comparison of parametric and nonparametric approaches Journal of Banking and Finance 28 25212540 Casu B Molyneux P 2003 A comparative study of efficiency in European banking Applied Economics 35 18651876 Charnes A Cooper WW Rhodes E 1978 Evaluating program and managerial efficiency An application of data envelopment analysis to program follow through Management Science 27 668697 Charnes A Cooper WW Rhodes E 1981 Evaluating program and managerial efficiency An application of data envelopment analysis to program follow through Management Science 27 668697 Chen TY 1998 A study of bank efficiency and ownership in Taiwan Applied Economics Letters 5 613616 Chen TY 2001 An estimation of Xinefficiency in Taiwans banks Applied Financial Economics 11 237242 Chen WH Shih JY 2006 A study of Taiwans issuer credit rating systems using support vector machines Expert Systems with Applications 30 427 435 Chen X Skully M Brown K 2005 Banking efficiency in China Application of DEA to pre and postderegulation eras 19932000 China Economic Review 16 229245 Chu SF Lim GH 1998 Share performance and profit efficiency of banks in an oligopolistic market Evidence from Singapore Journal of Multinational Financial Management 8 155168 Coelli TJ Perelman SA 1996 Comparison of Parametric and Nonparametric Distance Functions With Application to European Railways CREPP Discussion Paper No 199611 University of Liege Liege Coelli TJ Prasada Rao DS ODonnell CJ Battese GE 2005 An Introduction to Efficiency and Productivity Analysis second ed Springer USA Cook WD Seiford LM 2009 Data envelopment analysis DEA Thirty years on European Journal of Operational Research 192 117 Cook WD Hababou M 2001 Sales performance measurement in bank branches Omega 29 299307 Cook WD Hababou M Tuenter HJH 2000 Multicomponent efficiency measurement and shared inputs in Data Envelopment Analysis An application to sales and service performance in bank branches Journal of Productivity Analysis 14 209224 Das A Ghosh S 2006 Financial deregulation and efficiency An empirical analysis of Indian banks during the post reform period Review of Financial Economics 15 193221 Das A Ray SC Nag A 2009 Labouruse efficiency in Indian banking A branch level analysis Omega 37 411425 Dekker D Post T 2001 A quasiconcave DEA model with an application for bank branch performance evaluation European Journal of Operational Research 132 296311 Delis MD KoutsomanoliFillipaki K Staikouras CK Gerogiannaki K 2009 Evaluating cost and profit efficiency A comparison of parametric and nonparametric methodologies Applied Financial Economics 19 191202 Delis MD Papanikolaou NI 2009 Determinants of bank efficiency Evidence from a semiparametric methodology Managerial Finance 35 260275 Dimitras AI Zanakis SH Zopounidis C 1996 A survey of business failures with an emphasis on prediction methods and industrial applications European Journal of Operational Research 90 487513 Drake L Hall MJB Simper R 2006 The impact of macroeconomic and regulatory factors on bank efficiency A nonparametric analysis of Hong Kongs banking system Journal of Banking and Finance 30 14431466 Erdem C Erdem MS 2008 Turkish banking efficiency and its relation to stock performance Applied Economics Letters 15 207211 Fare R Grosskopf S Weber WL 2004 The effect of riskbased capital requirements on profit efficiency in banking Applied Economics 36 1731 1743 Fiordelisi F 2008 Efficiency and shareholder return in banking International Journal of Banking Accounting and Finance 1 114132 Gaganis C Pasiouras F Zopounidis C 2006 A multicriteria decision framework for measuring banks soundness around the world Journal of MultiCriteria Decision Analysis 14 103111 Gaganis C Liadaki A Doumpos M Zopounidis C 2009 Estimating and analyzing the efficiency and productivity of bank branches Managerial Finance 35 202 218 GarciaCestona M Surroca J 2008 Multiple goals and ownership structure Effects on the performance of Spanish savings banks European Journal of Operational Research 187 582599 Gilbert RA Wilson PW 1998 Effects of deregulation on the productivity of Korean banks Journal of Economics and Business 50 133155 Giokas DI 2008a Assessing the efficiency in operations of a large Greek bank branch network adopting different economic behaviours Economic Modelling 25 559574 Golany B Storbeck JE 1999 A Data Envelopment Analysis of the operational efficiency of bank branches Interfaces 29 1426 Guzman I Reverte C 2008 Productivity and efficiency change and shareholder value Evidence from the Spanish banking sector Applied Economics 40 2033 2040 Hahn FR 2007a Domestic mergers in the Austrian banking sector A performance analysis Applied Financial Economics 17 185196 Halkos GE Salamouris DS 2004 Efficiency measurement of the Greek commercial banks with the use of financial ratios A data envelopment analysis approach Management Accounting Research 15 201224 Hauner D 2005 Explaining efficiency differences among large German and Austrian banks Applied Economics 37 969980 Havrylchyk O 2006 Efficiency of the Polish banking industry Foreign versus domestic banks Journal of Banking and Finance 30 19751996 Huang TH Wang MH 2002 Comparison of economic efficiency estimation methods Parametric and nonparametric techniques The Manchester School 70 682709 Huang Z Chen H Hsu CJ Chen WH Wu S 2004 Credit rating analysis with support vector machines and neural networks A market comparative study Decision Support Systems 37 543558 Isik I 2008 Productivity technology and efficiency of de novo banks A counter evidence from Turkey Journal of Multinational Financial Management 18 427 442 Isik I Hassan MK 2002 Technical scale and allocative efficiencies of Turkish banking industry Journal of Banking and Finance 26 719766 Isik I Hassan MK 2003a Efficiency ownership and market structure corporate control and governance in the Turkish banking industry Journal of Business Finance and Accounting 30 13631421 Isik I Hassan MK 2003b Financial deregulation and total factor productivity change An empirical study of Turkish commercial banks Journal of Banking and Finance 27 14551485 Kantor J Maital S 1999 Measuring efficiency by product group Integrating DEA with activitybased accounting in a large Mideast bank Interfaces 29 2736 MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 197 Kirkwood J Nahm D 2006 Australian banking efficiency and its relation to stock returns The Economic Record 82 253267 Kohers T Huang MH Kohers N 2000 Market perception of efficiency in bank holding company mergers The roles of the DEA and SFA models in capturing merger potential Review of Financial Economics 9 101120 Kolari J Glennon D Shin H Caputo M 2002 Predicting large US commercial bank failures Journal of Economics and Business 54 361387 Lanine G Vander Vennet R 2006 Failure prediction in the Russian bank sector with logit and trait recognition models Expert Systems with Applications 30 463478 LozanoVivas A Pastor JT Hasan I 2001 European bank performance beyond country borders What really matters European Finance Review 5 141165 LozanoVivas A Pastor JT Pastor JM 2002 An efficiency comparison of European banking systems operating under different environmental conditions Journal of Productivity Analysis 18 5977 Luo X 2003 Evaluating the profitability and marketability efficiency of large banks An application of data envelopment analysis Journal of Business Research 56 627635 Maudos J Pastor JM 2003 Cost and profit efficiency in the Spanish banking sector 19851996 A nonparametric approach Applied Financial Economics 13 112 Maudos J Pastor JM Perez F 2002 Competition and efficiency in the Spanish banking sector The importance of specialization Applied Financial Economics 12 505516 Mercan M Reisman A Yolalan R Emel AB 2003 The effect of scale and mode of ownership on the financial performance of the Turkish banking sector Results of a DEAbased analysis SocioEconomic Planning Sciences 37 185202 Mukherjee K Ray SC Miller SM 2001 Productivity growth in large US commercial banks The initial postderegulation experience Journal of Banking and Finance 25 913939 Noulas AG 1997 Productivity growth in the Hellenic banking industry State versus private banks Applied Financial Economics 7 223228 Paradi JC Schaffnit C 2004 Commercial branch performance evaluation and results communication in a Canadian bankA DEA application European Journal of Operational Research 156 719735 Parkan C Wu ML 1999 Measurement of the performance of an investment bank using the operational competitiveness rating procedure Omega 27 201217 Pasiouras F Gaganis C Doumpos M 2007 A multicriteria discrimination approach for the credit rating of Asian banks Annals of Finance 3 351367 Pasiouras F 2008a International evidence on the impact of regulations and supervision on banks technical efficiency An application of twostage data envelopment analysis Review of Quantitative Finance and Accounting 30 187 223 Pasiouras F 2008b Estimating the technical and scale efficiency of Greek commercial banks The impact of credit risk offbalance sheet activities and international operations Research in International Business and Finance 22 301318 Pasiouras F Liadaki A Zopounidis C 2008 Bank efficiency and share performance Evidence from Greece Applied Financial Economics 18 1121 1130 Pastor JM 1999 Efficiency and risk management in Spanish banking A method to decompose risk Applied Financial Economics 9 371384 Pastor JM 2002 Credit risk and efficiency in the European banking system A threestage analysis Applied Financial Economics 12 895911 Pastor JM Serrano L 2006 The effect of specialisation on banks efficiency An international comparison International Review of Applied Economics 20 125 149 Portela MCAS Borges PC Thanassoulis E 2003 Finding closest targets in non oriented DEA models The case of convex and nonconvex technologies Journal of Productivity Analysis 19 251269 Portela MCAS Thanassoulis E 2005 Profitability of a sample of Portuguese bank branches and its decomposition into technical and allocative components European Journal of Operational Research 162 850866 Portela MCAS Thanassoulis E 2007 Comparative efficiency analysis of Portuguese bank branches European Journal of Operational Research 177 12751288 Prior D 2003 Long and shortrun nonparametric cost frontier efficiency An application to Spanish savings banks Journal of Banking and Finance 27 655 671 Ravi Kumar P Ravi V 2007 Bankruptcy prediction in banks and firms via statistical and intelligent techniques A review European Journal of Operational Research 180 128 Ravi V Kurniawan H Thai PNK Ravi Kumar P 2008 Soft computing system for bank performance prediction Applied Soft Computing 8 305315 Rezitis AN 2006 Productivity growth in the Greek banking industry A non parametric approach Journal of Applied Economics IX 119138 Sahoo BK Tone K 2009 Decomposing capacity utilization in data envelopment analysis An application to banks in India European Journal of Operational Research 195 575594 Sathye M 2003 Efficiency of banks in a developing economy The case of India European Journal of Operational Research 148 662671 Sathye M 2002 Measuring productivity changes in Australian banking An application of Malmquist indices Managerial Finance 28 4859 Sealey C Lindley J 1977 Inputs outputs and a theory of production and cost at depository financial institutions Journal of Finance 32 12511266 Sherman HD Rupert TJ 2006 Do bank mergers have hidden or foregone value Realized and unrealized operating synergies in one bank merger European Journal of Operational Research 168 253268 Simar L Wilson PW 2007 Estimation and inference in twostage semi parametric models of production processes Journal of Econometrics 136 31 64 Smith KA Gupta JND 2000 Neural networks in business Techniques and applications for the operations research Computers and Operations Research 27 10231044 Soteriou AC Zenios SA 1999a Using data envelopment analysis for costing bank products European Journal of Operational Research 114 224234 Soteriou A Zenios SA 1999b Operations quality and profitability in the provision of banking services Management Science 45 12211238 Sturm JE Williams B 2004 Foreign bank entry deregulation and bank efficiency Lessons from the Australian experience Journal of Banking and Finance 28 17751799 Sufian F Majid MZA 2006 Banks efficiency and stock prices in emerging markets Evidence from Malaysia Journal of AsiaPacific Business 7 3553 Swicegood P Clark JA 2001 Offsite monitoring systems for predicting bank underperformance A comparison of neural networks discriminant analysis and professional human judgment International Journal of Intelligent Systems in Accounting Finance and Management 10 169186 Tanna S 2009 The impact of foreign direct investment on total factor productivity growth International evidence from the banking industry Managerial Finance 35 297311 Tay FEH Cao L 2001 Application of support vector machines in financial time series forecasting Omega 29 309317 TortosaAusina E 2002a Bank cost efficiency and output specification Journal of Productivity Analysis 18 199222 TortosaAusina E 2003 Nontraditional activities and bank efficiency revisited A distributional analysis for Spanish financial institutions Journal of Economics and Business 55 371395 Tsionas EG Lolos SEG Christopoulos DK 2003 The performance of the Greek banking system in view of the EMU Results from a nonparametric approach Economic Modelling 20 571592 Thoraneenitiyan N Avkiran NK 2009 Measuring the impact of restructuring and countryspecific factors on the efficiency of postcrisis East Asian banking systems Integrating DEA with SFA SocioEconomic Planning Sciences 43 240 252 Tung WL Queka C Cheng P 2004 GenSoEWS A novel neuralfuzzy based early warning system for predicting bank failures Neural Networks 17 567587 Vapnik VN 1995 The Nature of Statistical Learning Theory SpringerVerlag New York Wang MH Huang TH 2007 A study on the persistence of Farrells efficiency measure under a dynamic framework European Journal of Operational Research 180 13021316 Weill L 2004 Measuring cost efficiency in European banking A comparison of frontier techniques Journal of Productivity Analysis 21 133152 Wheelock DC Wilson PW 2000 Why do banks disappear The determinants of US bank failures and acquisitions Review of Economics and Statistics 82 127 138 Wu D Yang Z Liang L 2006 Using DEAneural network approach to evaluate branch efficiency of a large Canadian bank Expert Systems with Applications 31 108115 Zen F Baldan C 2008 The strategic paths and performances of Italian mutual banks A nonparametric analysis International Journal of Banking Accounting and Finance 1 189214 Zhao H Sinha AP Ge W 2009 Effects of feature construction on classification performance An empirical study in bank failure prediction Expert Systems with Applications 36 26332644 Zhou P Ang BW Poh KL 2008 A survey of data envelopment analysis in energy and environmental studies European Journal of Operational Research 189 1 18 198 MD Fethi F Pasiouras European Journal of Operational Research 204 2010 189198 UNIVERSIDADE DE SÃO PAULO FACULDADE DE ECONOMIA ADMINISTRAÇÃO E CONTABILIDADE DE RIBEIRÃO PRETO DEPARTAMENTO DE ADMINISTRAÇÃO PROGRAMA DE PÓSGRADUAÇÃO EM ADMINISTRAÇÃO DE ORGANIZAÇÕES CARLOS ROBERTO BONARETTI FILHO A eficiência bancária brasileira de 2000 a 2018 uma avaliação a partir da Fronteira Estocástica e do DEA de dois estágios RIBEIRÃO PRETO 2021 Prof Dr Vahan Agopyan Reitor da Universidade de São Paulo Prof Dr André Lucirton Costa Diretor da Faculdade de Economia Administração e Contabilidade de Ribeirão Preto Prof Dr Jorge Henrique Caldeira de Oliveira Chefe do Departamento de Administração Prof Dr João Luiz Passador Coordenador do Programa de PósGraduação em Administração de Organizações CARLOS ROBERTO BONARETTI FILHO A eficiência bancária brasileira de 2000 a 2018 uma avaliação a partir da Fronteira Estocástica e do DEA de dois estágios Versão Corrigida A original encontrase disponível na FEARPUSP Dissertação de Mestrado apresentada ao Programa de PósGraduação em Administração de Organizações da Faculdade de Economia Administração e Contabilidade de Ribeirão Preto da Universidade de São Paulo para a obtenção do título de Mestre em Ciências Orientador Prof Dr Alexandre Pereira Salgado Junior Ribeirão Preto 2021 metropolitan regions in India Gaganis et al 2009 examine the impact of regional conditions on the efficiency and productivity of bank branches in Greece Some interesting observations can be summarised as follows In general the existing studies of bank branch efficiency use data on a small number of branches of nonUS banks and mostly employ DEA There are studies however which evaluate bank branch performance with DEA and compare the results with other methodologies ie FDH Portela et al 2003 Malmquist indices Camanho and Dyson 2006 loglinear frontier method Giokas 2008a ABC Plus Kantor and Maital 1999 OCRA Parkan and Wu 1999 Neural Network Model Wu et al 2006 Quasiconcave DEA FDH and Parametric Frontier models Dekker and Post 2001 Most of the studies which were reviewed use technical and operational efficiencies as estimated measures Recent studies examine the allocative profit efficiency see Portela and Thanassoulis 2005 2007 and cost efficiency Camanho and Dyson 2005ab 2008 aspects of bank branches 4 Other OR and AI techniques and bank performance In this section we review 15 recent studies that develop classification models used in the prediction of bank failure 9 bank underperformance 2 and credit ratings 4 Canbas et al 2005 point out that the study of bank failure is important for at least two reasons First an understanding of the reasons beyond the failure allows regulatory authorities to manage and supervise banks more efficiently Second the ability to differentiate between healthy and troubled banks can reduce the expected cost of bank failure either by taking actions to prevent failure or by minimizing the costs to the public As discussed in Gaganis et al 2006 while many of the failure prediction models achieve very promising classification accuracies one common drawback is that they concentrate on the assignment of banks in two groups failed and nonfailed However the classification of banks as bad or good reduces the usefulness of the model Obviously the classification into more groups relates to another strand of the literature that deals with credit risk modelling and attempts to replicate the ratings assigned by the credit agencies In the following subsections we provide a brief description of the methodologies the employed variables and the classification results18 A summary of the main characteristics of these studies is available in Table 4 of the Supplementary material Appendix A 41 Methodologies Neural network NN modelling is an intelligence technique that follows a process similar to the human brain As in other classification models the parameters of an NN model need to be estimated before the network can be used for prediction purposes There are numerous NN architectures learning methods and parameters Generally speaking NN architectures can be either feedback or feedforward In a feedback network nodes can receive inputs from nodes in any other layer in the network whereas in a feedforward network inputs are only received from previous layers The multilayer preceptor MLPNN competitive learning neural networks CLNN selforganizing map neural networks SOMNN backpropagation neural networks BPNN and probabilistic neural networks PNN are alternative approaches employed in the studies under review eg Chen and Shih 2006 Boyacioglu et al 2009 FOLHA DE APROVAÇÃO Nome BONARETTIFILHO Carlos Roberto Título A eficiência bancária brasileira de 2000 a 2018 uma avaliação a partir da Fronteira Estocástica e do DEA de dois estágios Dissertação de Mestrado apresentada ao Programa de PósGraduação em Administração de Organizações da Faculdade de Economia Administração e Contabilidade de Ribeirão Preto da Universidade de São Paulo para a obtenção do título de Mestre em Ciências Aprovado em Banca examinadora Profa Dra Perla Calil Pongeluppe Wadhy Rebehy Instituição Faculdade de Administração Economia e Contabilidade de Ribeirão Preto FEARPUSP Assinatura Prof Dr Fabiano Guasti Lima Instituição Faculdade de Administração Economia e Contabilidade de Ribeirão Preto FEARPUSP Assinatura Prof Dr Andrei Aparecido de Albuquerque Instituição Universidade Federal de São Carlos Assinatura The approach of support vector machines SVMs used in Boyacioglu et al 2009 Chen and Shih 2006 and Huang et al 2004 was introduced by Vapnik 1995 It is based on the Structural Risk Minimization SRM principle from computational learning theory This seeks to minimise an upper bound of the generalisation error rather than minimise the training error SVMs use a linear structure to implement nonlinear class boundaries through extremely nonlinear mapping of the input vectors into the highdimensional space If the data are linearly separated SVM uses a special kind of linear model the optimal separating hyperplane that provides the maximum separation between the classes The training points that are closest to the maximum margin hyperplane are called support vectors All other training examples are irrelevant for determining the binary class boundaries From the implementation point of view Vapnik 1995 showed how training a SVM and finding the parameters leads to a quadratic optimisation problem with bound constraints and one linear equality constraint This means that the solution of SVMs is unique optimal and absent from local minima Tay and Cao 2001 Nearest neighbours is a nonparametric estimation method that has been applied in various problems in finance The nearest neighbour rule classifies an object ie bank to the class of its nearest neighbour in the measurement space using some kind of distance measure like the local metrics the global metrics the Mahalanobis or the Euclidean distance The modification of the nearest neighbour rule the knearest neighbour kNN method that is employed in Zhao et al 2009 classifies an object ie bank to the class ie failed or nonfailed more heavily represented among its knearest neighbours In the case of decision trees that originate from machine learning instead of developing classification functions or network architectures a binary decision tree is developed This can be accomplished through a set of ifthen split conditions that lead to the accurate classification of cases eg banks Commonly used algorithms are CART eg Ravi et al 2008 and C45 eg Zhao et al 2009 The UTilités Additives DIScriminantes UTADIS multicriteria decision aid MCDA method used in Gaganis et al 2006 employs the framework of preference disaggregation analysis for the development of an additive utility function that is used to score the firms and decide upon their classification The estimation of the additive utility model is performed through mathematical programming techniques The MultiGroup Hierarchical Discrimination MHDIS method used in Pasiouras et al 2007 is another MCDA approach that uses utility functions for discrimination purposes MHDIS distinguishes the groups progressively starting by discriminating the first group from all the others and then proceeds to the discrimination between the alternatives belonging to the other groups To accomplish this task instead of developing a single additive utility function that describes all alternatives as in UTADIS two additive utility functions are developed in each one of the n 1 steps where n is the number of groups At each stage of the hierarchical discrimination procedure two linear and a mixedinteger programming problems are solved to estimate the utility thresholds and the two additive utility functions in order to minimise the classification error As discussed in Kolari et al 2002 and Lanine and Vander Vennet 2006 trait recognition is another nonparametric approach under which individual traits are initially developed from different segments of the distribution of each variable and the interactions of these segments with one or more other variables segmented distributions When all possible traits of the variables are tabulated for all banks trait recognition uses a search routine to cull traits that do not discriminate between failed and nonfailed banks Trait recognition uses two sets of discriminators the safe traits AGRADECIMENTOS Primeiramente à minha mãe Irene exemplo de pessoa a ser seguido meu orgulho minha inspiração e meu alicerce sempre presente no meu desenvolvimento e em minhas conquistas Mãe e pai ao mesmo tempo Mulher forte guerreira e dedicada A melhor mãe que poderia ter Ao meu pai Carlos Bonaretti que mesmo não estando presente participa de certa forma de todas as minhas conquistas Meu exemplo a ser seguido pois se eu for metade da pessoa que ele foi serei um grande homem Um exemplo de pai de marido de filho de amigo de profissional enfim um exemplo de homem Dedico esse trabalho à ele meu pai que deve estar vibrando e todo orgulhoso com mais essa conquista Aos meus irmãos Luiz Otavio e André Luis pessoas essenciais em minha vida companheiros e cúmplices que com minha mãe são meus maiores orgulhos Presentes em toda a minha vida e que estarão sempre ao meu lado Aos grandes amigos com os quais a vida me presenteou Gabriel Riguetti Cesar Zambrano Murilo Kazon Paulo Camperoni Otávio Falconi e Giovani Magri Todos eles sem exceções companheiros a qualquer hora e amigos sempre presentes Amigos que levarei sempre comigo Ao meu orientador Prof Dr Alexandre Salgado que tanto me apoiou no desenvolvimento deste estudo e que compartilhou comigo toda sua experiência de vida acadêmica pessoa que muito admiro por todas suas conquistas e todo o seu conhecimento acumulado Também agradeço ao GREFIC grupo de estudos que sempre me apoiou À Profa Dra Perla pelo conhecimento e paciência Ao Marco Silva Allan Angelli e José Sardelari pela parceria Gostaria de agradecer também aos grandes amigos que a academia me trouxe Yago Marinzeck e Fabio Carlucci Para finalizar gostaria de agradecer à Deus que colocou todas essas pessoas maravilhosas em meu caminho e que me proporcionou experiências e oportunidades maravilhosas me deu o dom da vida e saúde para chegar aonde cheguei O presente trabalho foi realizado com apoio da Coordenação de Aperfeiçoamento de Pessoal de Nível Superior Brasil CAPES Código de Financiamento 001 RESUMO BONARETTI FILHO Carlos Roberto A eficiência bancária brasileira de 2000 a 2018 uma avaliação a partir da Fronteira Estocástica e do DEA de dois estágios 129 f Dissertação Mestrado em Administração de Organizações Faculdade de Economia Administração e Contabilidade de Ribeirão Preto da Universidade de São Paulo 2021 As mudanças no ambiente competitivo as políticas regulatórias e a inovação forma responsáveis nas últimas décadas por incentivarem uma significativa mudança no modelo de negócio das instituições bancárias Os bancos têm acrescido ao seu portifólio tradicional serviços como subscrição de títulos corretagem de seguros serviços de fundos mútuos serviços fiduciários e outras atividades geradoras de taxas o que incrementa sua receita além de estenderem suas fontes de passivos ao financiamento de atacado Devido à sua estrutura dinâmica e à complexidade do ambiente econômico em que operam os bancos estão sujeitos a muitos riscos dentre os quais destacase em relevância o risco de crédito responsável pelo aumento do custo marginal da dívida e do patrimônio líquido Assim à medida em que aumenta a exposição de um banco ao risco de crédito aumenta sua tendência de vivenciar uma crise financeira Daí a importância das instituições bancárias serem eficientes em suas atividades de intermediação e rentabilidade dois aspectos considerados pela academia como principais abordagens de análise de eficiência Nesse contexto o presente trabalho analisa a eficiência por meio de duas técnicas de mensuração a Análise Envoltória de Dados e a Análise da Fronteira Estocástica das instituições bancárias que compõem o sistema financeiro brasileiro São portanto analisados os principais fatores que influenciaram a eficiência bancária nacional entre os anos de 2000 e 2018 Foram considerados para análise os indicadores CAMELS das instituições financeiras Como resultado este estudo identificou as variáveis CAMELS que explicam a eficiência bancárias das instituições analisadas o que sugere que existem práticas relacionadas a referidas variáveis que influenciam negativamente os escores de eficiência Palavraschave eficiência bancária CAMELS DEA SFA Attract visitors to your showroom and get more customers Engaging content for your showroom with NFC With exhibition and showroom NFC capability customers can interact and access product information video or presentation without needing to handle anything Cover your video presentation or interesting pages with Hall of Posters Tags and get visitors to engage with your products You can use as many tags as you want ABSTRACT BONARETTI FILHO Carlos Roberto Brazilian banking efficiency from 2000 to 2018 an evaluation using Stochastic Frontier Analysis and twostage DEA 133 f Dissertation Masters Degree in Business Administration School of Economics Business Administration and Accounting at Ribeirão Preto University of São Paulo 2021 In the last decades changes in the competitive environment regulatory policies and innovation have been responsible for encouraging a significant change in banks business models Banks have added to their traditional portfolio services such as securities underwriting insurance brokerage mutual fund services fiduciary services and other feegenerating activities which increase their revenue in addition to extending their sources of liabilities to finance wholesale Due to their dynamic structure and the complexity of the economic environment in which they operate banks are subject to many risks among which stands out the credit risk responsible for the increase in the marginal cost of debt and equity Thus as the banks exposure to credit risk increases so does its tendency to experience a financial crisis Hence the importance of banking institutions being efficient in their intermediation and profitability activities two aspects considered by academia as the main approaches to analyzing efficiency In this context the present study analyzes the efficiency through two measurement techniques Data Envelopment Analysis and Stochastic Frontier Analysis of the banking institutions that make up the Brazilian financial system Therefore the main factors that influenced the national banking efficiency between 2000 and 2018 are analyzed The CAMELS indicators of financial institutions were considered for analysis As a result this study identified the CAMELS variables that explain the banking efficiency of the analyzed institutions which suggests that there are practices related to these variables that negatively influence the efficiency scores Keywords bank efficency CAMELS DEA SFA Share social media Share your Instagram TikTok Twitter to your visitors and get more followers Turn followers into buyers Share your Shop link on your favorite marketplace to get more regular customers Share music Share and play music in your shop or at your event Share QR codes Share several websites and QR codes from a single Meko Tag LISTA DE ABREVIATURAS E SIGLAS BACEN Banco Central do Brasil BCC Banker Charnes e Cooper CAMELS Capital Assets Management Earnings Liquidity Sensibility to market risk CCR Charnes Cooper e Rhodes CRS Constant Returns to Scale DEA Data Envelopment Analysis DMU Decision Making Units FFIEC Federal Financial Institutions Examination Council FMI Fundo Monetário Internacional FSI Financial Soundness Indicators iid independente e identicamente distribuída IMF International Monetary Fund OCC Office of the Comptroller of the Currency ROA Return on Assets ROE Return on Equity SFA Stochastic Frontier Analysis UFIRS Uniform Financial Institutions Rating System VRS Variable Returns to Scale Make it easy to connect Your visitors can scan a single tag with their phone to link to all your social media and clicktocall number Get more followers and share updates Motivation your customers to follow you on social media realtime with a single tap Link with all your customers Share your social media websites phone numbers product information and location from a single tag If You want to order single or bulk tags for your business please contact us Your logo your choice For bulk order you can select your logos for custom tags Or you can choose our design NFC Tag MekoTag contactmekotagcom 66 0 913460561 wwwmekotagcom mekotag mekotag showroom tags LISTA DE QUADROS Quadro 1 Principais funções determinísticas 42 Quadro 2 Quadro sinótico de revisão bibliográfica sobre utilização de DEA na avaliação de eficiência do sistema bancário 48 Quadro 3 Quadro sinótico de revisão bibliográfica sobre utilização da Fronteira Estocástica para a análise da eficiência bancária 50 Quadro 4 Literatura sobre variáveis representativas do acrônimo CAMELS C A M 63 Quadro 5 Literatura sobre variáveis representativas do acrônimo CAMELS E L S 67 Quadro 6 Variáveis independentes advindas do CAMELS 80 Quadro 7 Adequação à Teoria Financeira Abordagem de Intermediação e Rentabilidade 104 YOUR JOURNEY STARTS HERE A DIVERSE ABUNDANCE OF OUR SELECTED INDEPENDENT FILMMAKERS ANIMATION ART DOCUMENTARY DRAMA EXPERIMENTAL FICTION FRISCO IS A PROGRAM OF EXPERIMENTAL AND INDEPENDENT FILMMAKERS BASED IN SAN FRANCISCO THE PROGRAM PRESENTS LOCAL FILMMAKERS WORK IN CONTINUOUS MONTHLY PROGRAMS OFFERING A TRUE WAVER OF SOLID ART DIVERSE PERSPECTIVES AND A SENSE OF COMMUNITY FOR THE FILMMAKER AND FILMLOVING AUDIENCE chal chill 8 h mafiomaw hale lanoelemaris lONstiNVA S RWINKEYP Ty UDDATTI YAAAA 101 HOST 202 Installation Liberty SCIENCE FANTASY STARS CE LE RACY 2 SAYS LISTA DE TABELAS Tabela 1 Inputs e Outputs de Macoris Salgado Junior e Falsarella Junior 2015 75 Tabela 2 Inputs e Outputs utilizados em DEA e SFA com múltiplos inputs e outputs 75 Tabela 3 Inputs e Outputs utilizados em DEA e SFA com um único output 76 Tabela 4 Estatísticas descritivas dos escores de DEA e SFA 88 Tabela 5 Variáveis significantes para Intermediação model summary 89 Tabela 6 Variáveis significantes para Intermediação coefficients a 89 Tabela 7 Variáveis significantes para Rentabilidade model summary 90 Tabela 8 Variáveis significantes para Rentabilidade coefficients a 90 Tabela 9 Regressão para efeitos fixos de Intermediação 91 Tabela 10 Regressão para efeitos fixos de Rentabilidade 92 Tabela 11 Efeitos na eficiência SFA e DEA para Abordagem de Intermediação 99 Tabela 12 Efeitos na eficiência SFA e DEA para Abordagem de Rentabilidade 102 LISTA DE FIGURAS Figura 1 Fronteira de Eficiência 34 Figura 2 Minimização de Inputs e Maximização de Output CCR 35 Figura 3 Representação gráfica CCR Orientado a Inputs 1 e Outputs 2 36 Figura 4 Minimização de Inputs e Maximização de Output BCC 37 Figura 5 Representação gráfica BCC Orientado a Inputs 1 e Outputs 2 37 Figura 6 Classificação entre Ganhos de Escala e Orientação 38 Figura 7 Fronteira Estocástica de Produção 40 Figura 8 DEA e SFA dois estágios com regressão linear múltipla 79 Santo A Rookie But On His Way VIDEO MUSIC FEATURING J SEEK ON STAGE COLORADO SPRINGS CHRISTIAN ARTISTS SPREAD THE WORD OF CHRIST THROUGH MUSICAL EVENTS DUBLIN VIDEO JESSE SANTO TALKS ABOUT LEADING A CHRISTIAN HIP HOP CONCERT SERIES IN COLORADO SPRINGS VIDEOGRAPHY BY MIRIAM BRADY BY KENDRACOLE AMERICA IS KNOWN FOR ITS MANY EXTREMES FROM THE HOT SUMMERS TO THE COLD WINTERS FROM THE BUSINESS WORLD TO THE MUSICAL ARENA AND FROM BASKETBALL STARS TO SOLO SINGERSONGWRITERS GIVEN THAT OUR PLUTONIC MISSION IS TO HELP AND APPRECIATE BLACK AND LATINO CHRISTIAN ARTISTS ESPECIALLY THOSE WHO LIVE IN UNDERRESOURCED COMMUNITIES IT CAME AS NO SURPRISE THAT JESSE SANTO AN UP AND COMING CHRISTIAN HIPHOP ARTIST FROM COLORADO SPRINGS COLORADO STRUGGLES TO GET THE ATTENTION OF THE HIGHER UPS AT THE NATIONAL CHRISTIAN MUSIC INDUSTRY LEADERSHIP LEVEL FOR THE MOST PART HIS MUSIC AND ARTISTIC STYLES HAVE BEEN BRUSHED ASIDE FOR HAVING TOO MUCH URBAN CHRISTIAN SOUND INSTEAD OF THE TRADITIONAL FULL BAND SOUND THEY DONT WANT RAP OR HIPHOP AS CHRISTIAN MUSIC A LOT OF PEOPLE DONT GET IT SANTO SAID SANTO WANTED TO FIX THAT HE COULD NOT LET OTHER PEOPLE TELL HIM TO EMBRACE THE LEGACY OF ITS FORMALITY OF HOW TO BE A CHRISTIAN ARTIST THE PHILOSOPHY OF IF IT DOESNT FIT THE FORMULA WE DONT WANT IT IS EXACTLY WHY SANTOS MESSAGE TO YOUNG ARTISTS IS TO SEARCH FOR YOUR AUTHENTICITY IS THERE ROOM FOR US SANTO ASKED THERES ALWAYS ROOM FOR YOUR AUTHENTICITY IF THE FACT THAT YOURE BLACK IF THE FACT THAT YOURE LATINO IF THE FACT THAT YOURE YOUNG IF THE FACT THAT YOURE OLD IF THE FACT BE THAT YOU WERE RAISED IN A SINGLE MOTHER HOME IT DOESNT MATTER WE ALL HAVE THE WARRIORS IN OUR CIRCLES THAT HAVE BEEN CALLED TO PUSH OUR CULTURE FORWARD WE HAVE TO DO THAT WE HAVE TO PUSH FORWARD THE POWER OF WHATEVER GIFT YOU HAVE USE IT FOR THE GLORY OF GOD ON STAGE SANTOS PROJECT IS CALLED WARRIORS FOR LIGHT RECORDS AND ITS CONCEPT RPUSHES ARTISTS TO EMBRACE THE UNIQUE SOUND AND SOUNDS OF OUR CITIES AND COMMUNITIES SANTO IS INITIATING A CHRISTIAN HIPHOP CONCERT SERIES CALLED THE GOSPEL MOB TOUR THAT IS CURRENTLY DOING A WEST COAST RUN STARTING IN FEBRUARY 2024 HE PARTNERED WITH LOCAL COLORADO SPRINGS PROMOTERS TO BRING THE MUSIC AND DANCE CULTURE TO LOCAL CHURCHES AND VENUES THAT TYPICALLY SERVE BLACK AND LATINO COMMUNITIES THESE SHOWS ARE ABOUT CELEBRATING CHRISTIAN URBAN ARTISTS MUSIC AND DANCE CULTURE THERE IS A NEED IN THE URBAN CHURCHES FOR THIS SHOWCASE OF TALENTS HE SAID LISTA DE GRÁFICOS Gráfico 1 Abordagem Intermediação 84 Gráfico 2 Abordagem Rentabilidade 86 VIDEO MUSIC JESSE SANTO TALKS ABOUT LEADING A CHRISTIAN HIP HOP CONCERT SERIES IN COLORADO SPRINGS COLORADO CHRISTIAN ARTISTS SPREAD THE WORD OF CHRIST THROUGH MUSICAL EVENTS THATS WHERE MY PASSION COMES FROM I AM EQUIPPED TO SUPPORT THIS ART FORM AND TO RAISE UP ARTISTS TO THIS PLATFORM SANTO SAID I FEEL LIKE I AM DOING MY PART WITH THIS CONCERT SERIES THIS TYPE OF CONCERT SERIES ALSO HELPS LOCAL ARTISTS TO RECOGNIZE THAT THERE IS ROOM FOR THEM AS CHRISTIAN MUSICIANS IN THEIR COMMUNITIES ITS NOT JUST ABOUT MUSIC FOR ME ITS ABOUT POSITIVE COMMUNITY MENTAL HEALTH SPIRITUALITY SO THATS THE BASE OF WHAT I AM WORKING TO DO WITH THESE EVENTS SANTO SAID MORE INFO JESSESANTOJESSE IMINSTAGRAM JESSESANTOMUSICGMAILCOM 23 ÍNDICE 1 INTRODUÇÃO 25 11 JUSTIFICATIVA 27 12 PROBLEMATIZAÇÃO E OBJETIVOS DA PESQUISA 28 2 REFERENCIAL TEÓRICO 29 21 CONCEITO DE EFICIÊNCIA 29 22 TÉCNICAS DE ANÁLISE DA EFICIÊNCIA 30 221 Análise Envoltória de Dados 31 222 Análise da Fronteira Estocástica 37 23 EFICIÊNCIA BANCÁRIA 43 24 FERRAMENTAS DE AVALIAÇÃO DA EFICIÊNCIA BANCÁRIA OU MENSURAÇÃO DA EFICIÊNCIA BANCÁRIA 46 25 FATORES EXPLICATIVOS DO DESEMPENHO DE INSTITUIÇÕES FINANCEIRAS 49 26 INDICADORES DE RISCO DE INSTITUIÇÃO FINANCEIRA CAMELS 50 261 Adequação de capital capital adequacy 51 262 Qualidade dos ativos asset quality 53 263 Gestão administração management 54 264 Lucro earnings 54 265 Liquidez liquidity 56 266 Sensibilidade ao risco de mercado sensitivity to market risk 57 267 Variáveis CAMELS 58 3 PROCEDIMENTOS METODOLÓGICOS 70 31 CLASSIFICAÇÃO DA PESQUISA 70 32 ETAPAS DA PESQUISA 71 4 DESCRIÇÃO DOS RESULTADOS 80 41 DEA Segundo Estágio resultados da regressão linear múltipla 85 411 DEA Segundo Estágio para abordagem Intermediação 85 412 DEA Segundo Estágio para abordagem Rentabilidade 86 42 SFA e os fatores explicativos para eficiência bancária 87 43 Análise consolidada DEA e SFA nas abordagens Intermediação e Rentabilidade 89 431 Pressupostos dos indicadores CAMELS para as Abordagens de Intermediação e Rentabilidade 92 432 Intermediação 94 433 Rentabilidade 97 5 DISCUSSÃO DOS RESULTADOS 100 6 CONCLUSÕES 110 REFERÊNCIAS BIBLIOGRÁFICAS 112 24 25 1 INTRODUÇÃO Houve nas últimas décadas uma mudança significativa dos negócios dos bancos para áreas não tradicionais em resposta às mudanças no ambiente competitivo às políticas regulatórias e à inovação nas habilidades e tecnologias de gestão WU HSIA HENG 2006 Além dos serviços tradicionais de empréstimo e recebimento de depósitos os bancos têm se empenhado cada vez mais na subscrição de títulos corretagem de seguros serviços de fundos mútuos serviços fiduciários e muitas outras atividades geradoras de taxas para renda extra além de estenderem suas fontes de passivos ao financiamento de atacado FRIOSI et al 2017 Embora o grau de diversificação tenha diminuído relativamente em alguns países após a crise financeira global de 2008 e 2009 MARTEL VAN RIXTEL MOTA 2012 ROENGPITYA TARASHEV TSATSARONIS 2014 os passivos de fontes não depositárias ainda são amplamente consideradas como determinantes importantes de seu desempenho MERGAERTS VENNET 2016 A principal fonte de receita do setor bancário consiste em empréstimos concedidos por bancos comerciais Portanto o risco de crédito é um dos riscos mais importantes enfrentados pelos bancos O risco de crédito é definido pelo Comitê de Supervisão Bancária da Basiléia BASEL COMMITTEE ON BANKING SUPERVISION 2000 como a probabilidade de perda parcial ou total do empréstimo em aberto devido ao não pagamento do empréstimo no prazo Um aumento no risco de crédito aumenta o custo marginal da dívida e do patrimônio líquido Correspondentemente o custo de captação do banco aumenta À medida que aumenta a exposição de um banco ao risco de crédito aumenta a tendência dos bancos de vivenciar uma crise financeira À medida em que as transações e produtos financeiros se tornam mais complexos e a integração e liberalização bancária tornam o setor mais competitivo há sobre os bancos uma forte pressão para administrar os riscos de forma eficaz sem negociar por menor retorno ou eficiência Muitos estudos examinaram a relação entre risco bancário e eficiência usando diferentes medidas ALTUNBAS et al 2007 BERGER MESTER 1997 KWAN EISENBEIS 1997 PRUTEANUPODPIERA WEILL SCHOBERT 2008 Também está em questão se qualquer relação causal entre risco e eficiência é temporária ou de longo prazo Staikouras Mamatzakis e KoutsomanoliFillippaki 2009 descobriram que para os bancos europeus o impacto da ineficiência no risco é pequeno e breve enquanto o efeito reverso do risco na ineficiência é negativo e significativo 26 O impacto da competição sobre a lucratividade no setor bancário foi documentado no paradigma tradicional de estruturacondutadesempenho que argumenta que em uma indústria mais concentrada com um nível mais baixo de competição as empresas tendem a conspirar entre si para obter lucros mais elevados Além disso há uma grande quantidade de literatura investigando o impacto da concorrência na lucratividade do setor bancário MAUDOS FERNANDEZ DE GUEVARA 2004 TAN FLOROS 2014 TAN 2016 A compreensão das razões que impactam no fator de produção do setor bancário tem um papel de destaque no contexto empresarial e consiste numa ferramenta eficaz de obtenção de lucros maiores índices de eficiência nos processos e flexibilidade estrutural aspectos de grande relevância à análise da competitividade e consolidação das instituições no mercado BRAGA JUNIOR SILVA SILVA 2014 Para Macoris Salgado Junior e Falsarella Junior 2015 ao avaliarse a eficiência das instituições financeiras é necessário realizar a comparação do desempenho dos bancos que executam eficientemente determinada atividade em relação aos demais Sendo assim é de grande importância utilizar os bancos considerados eficientes como benchmark para as demais instituições bancárias nacionais e internacionais Portanto compreender os fatores que determinam a eficiência bancária e considerado um fator chave para o desenvolvimento de melhores estratégias de gestão e políticas públicas STAUB SOUZA TABAK 2010 Verificase desta forma que a mensuração da eficiência bancária é um índice de desempenho relevante e que a relação entre as entradas ou seja os recursos empregados com as saídas que são os resultados obtidos pode ser considerada uma ferramenta eficiente de gestão para decisões de investimento avaliação de recursos humanos redução de custos e despesas entre outros fatores BATISTA 2009 São diversas as abordagens que podem realizar mensuração e a avaliação da eficiência bancária porém há destaque para duas abordagens distintas intermediação e rentabilidade MACORIS SALGADO JUNIOR FALSARELLA JUNIOR 2015 Conhecer os impactos de variáveis sobre as óticas de intermediação e rentabilidade é a base para se entender o atual momento do setor bancário Na intermediação são avaliadas e mensuradas as instituições com foco na capacidade de atuarem como intermediadores financeiros captando fundos de agentes superavitários e repassandoos na forma de crédito para agentes deficitários YANG 2009 Para definir entradas e saídas bancárias sob a ótica da abordagem de intermediação baseada em Berger e Mester 1997 2003 e muitos outros LOZANOVIVAS PASIOURAS 27 2010 GAGANIS PASIOURAS 2013 MALIKOV HUMBHAKAR SUN 2016 KHAN et al 2018 Essa abordagem assume que os bancos coletam fundos principalmente depósitos como entradas e os transformam em empréstimos ou outros ativos Na abordagem da rentabilidade são levadas em consideração a capacidade de a instituição maximizar as riquezas frente à redução de custos como forma de ampliar a sua rentabilidade e a sua lucratividade DRAKE HALL SIMPER 2006 FETHI PASIOURAS 2010 As abordagens de intermediação e rentabilidade foram utilizadas no presente estudo por constituíremse nas principais abordagens utilizadas para avaliar a eficiência MACORIS SALGADO JÚNIOR FALSARELLA JÚNIOR 2015 A maior parte da literatura reconhece a relevância das técnicas não paramétrica da Análise Envoltória de Dados ou Data Envelopment Analysis DEA e paramétrica da Análise da Fronteira Estocástica ou Stochastic Frontier Analysis SFA como principais métodos de análise de eficiência TABACK TECLES 2010 STAIKOURAS MAMATZAKIS KOUTSOMANOLIFILIPPAKI 2008 Assim sendo contribuise com a literatura ao analisar a eficiência em duas principais abordagens por meio da comparação de duas técnicas de mensuração de eficiência das instituições bancárias que compõem o sistema financeiro nacional composto por um sistema bancário reconhecidamente consolidado e bastante competitivo Esta dissertação está organizada em seis capítulos sendo o primeiro a introdução o segundo o referencial teórico seguidos pelos capítulos da metodologia descrição dos resultados discussão dos resultados e conclusão 11 JUSTIFICATIVA A eficiência do setor bancário tem sido amplamente estudada e discutida em diversos trabalhos nacionais e internacionais Tratase de um setor de elevada importância e com impactos no crescimento e estabilidade das economias nacionais e mundial Frente a análise de diferentes países identificouse um alto grau de ineficiência que afetou principalmente os bancos pertencentes a economias em desenvolvimento TABAK TECLES 2010 A crise iniciada em 2007 nos Estados Unidos e que se espalhou para as demais economias do mundo demonstrou o impacto que uma instituição financeira pode devido a sua característica de organização complexa e interconectada causar na estabilidade do sistema financeiro global com consequências negativas para a economia real Diante disto órgãos reguladores e formuladores de políticas definiram a necessidade de estipular medidas e modelos 28 para medir o risco destas instituições e identificar as instituições que poderiam prejudicar os mercados financeiros e levar a um colapso de todo o sistema financeiro BONGINI NIERI PELAGATTI 2015 Levando em consideração o papel de fundamental importância dos bancos nas economias nacionais as análises e mensurações de seu desempenho são extremamente relevantes FETHI PASIOURAS 2010 Frente a isto esta dissertação auxilia na resolução deste problema ao analisar medidas de eficiência bancária que funcionam como uma ferramenta de monitoramento destas instituições O Brasil possui relevante importância na América Latina devido a sua economia e consequentemente devido ao seu sistema bancário Frente a relevância do Brasil seu mercado de capitais ou mercado de títulos ainda é pouco desenvolvido ficando o mercado de crédito responsável por desempenhar um papel principal na economia brasileira Este fato é comum em economias em desenvolvimento e o entendimento do sistema bancário se faz necessário para uma efetiva condução de estratégias por parte dos gestores bem como de órgãos públicos STAUB SOUZA TABAK 2010 Este estudo compara todos os bancos que constam no segmento B1 do Banco Central do Brasil Bacen ao longo de 18 anos 12 PROBLEMATIZAÇÃO E OBJETIVOS DA PESQUISA Com a finalidade de analisar a eficiência bancária brasileira e os seus fatores explicativos este estudo objetivou investigar o problema de pesquisa que se segue Quais foram os fatores que influenciaram a eficiência dos bancos brasileiros entre os anos de 2000 e 2018 Portanto o objetivo deste trabalho é identificar os fatores que influenciaram a eficiência dos bancos brasileiros nas abordagens de intermediação e rentabilidade Os objetivos específicos são i propor um modelo de inputs e outputs para a mensuração de eficiência bancária ii avaliar a eficiência e os fatores explicativos sob a ótica de duas técnicas de mensuração de eficiência DEA e SFA 29 2 REFERENCIAL TEÓRICO Considerando a eficiência um aspecto intrínseco para a competitividade do setor bancário a revisão da literatura realizada foi de fundamental importância para embasar este estudo e melhor compreender os conceitos de eficiência que impactam no setor bem como compreender a utilização de duas técnicas uma paramétrica Análise Envoltória de Dados e outra não paramétrica Análise da Fronteira Estocástica no setor bancário brasileiro O objetivo deste referencial teórico é o de apresentar os conceitos de eficiência no setor bancário analisados através das técnicas DEA e SFA em concordância com a proposta de avaliar quais as variáveis impactam nos escores de eficiência dentre os bancos analisados 21 CONCEITO DE EFICIÊNCIA Eficiência é objeto de estudo de diversas áreas do conhecimento que visam extrair o máximo de produtos ou outputs através da utilização mínima possível de recursos ou inputs Tal fato indica se uma instituição é econômica ao utilizar os seus recursos FARRELL 1957 A apresentação deste conceito pode ser relacionada com a condição de equilíbrio que representa a existência da maximização da capacidade econômica sem o efeito de prejuízo de outros fatores e que pode ser representada e calculada através do uso de expressões matemáticas PARETO 1897 Através desta análise Pareto 1897 descreve ser possível comparar e mensurar a eficiência de um sistema econômico considerando a divisão em três grupos O primeiro considera a eficiência na produção como sendo a capacidade de se maximizar a produção sem que haja alterações na produção de demais produtos deste sistema A segundo especifica a eficiência na cesta de produtos como a capacidade de produção para atendimento das preferências dos agentes econômicos Por fim o terceiro considerado como eficiência nas trocas é determinado como a capacidade de realizar a distribuição dos produtos produzidos da forma mais eficiente Ao conceituar eficiência Farrel 1957 considera que em ambientes produtivos a eficiência é obtida ao analisar os dados da produção de acordo com o coeficiente de utilização dos recursos Dentre as formas de verificar a eficiência a razão entre quantidade de insumos e produtos é uma das mais utilizadas e é matematicamente expressa por 30 𝐸𝑓𝑖𝑐𝑖ê𝑛𝑐𝑖𝑎 ƒ𝑚𝑎𝑥𝐷𝑎𝑑𝑜𝑠𝑃𝑟𝑜𝑑𝑢𝑧𝑖𝑑𝑜𝑠 ƒ𝑚𝑖𝑛𝐼𝑛𝑠𝑢𝑚𝑜𝑠 𝑈𝑡𝑖𝑙𝑖𝑧𝑎𝑑𝑜𝑠 Ainda segundo Farrel 1957 produzir o máximo de produtos através de uma certa quantidade de entradas ou da mesma forma produzir determinado produto utilizando quantidades mínimas de insumo é considerado eficiência técnica Sendo assim quando uma unidade é analisada e considerada eficiente dentro dos parâmetros anteriores ela é considerada eficiente e presente na fronteira de produção Dentre as diversas técnicas utilizadas para se medir a eficiência duas delas possuem reconhecimento comprovado em diversas áreas de estudo uma paramétrica DEA e outra não paramétrica SFA Os modelos não paramétricos baseiamse na construção de fronteiras determinísticas onde encontramse as unidades com as melhores práticas dentre todas as demais unidades presentes na amostra e devido a este fato sua vantagem decorre do fato da não necessidade de apresentar hipóteses sobre a função de produção desenvolvida por Farrel 1957 bem como a não necessidade de identificar a ineficiência ODECK BRATHEN 2012 Os modelos paramétricos por sua vez procedem com a construção da fronteira considerando os impactos que diferentes cenários podem causar na função da produção de cada unidade ou seja a possibilidade de um mesmo modelo ter resultados diferentes quando analisado em cenários diferentes Além disso o fato desta técnica considerar o termo de erro aleatório e o termo de ineficiência em conjunto bem como o impacto de cenários diferentes na função produção traduzemse na vantagem desta técnica KUMBHAKAR LOVELL 2003 22 TÉCNICAS DE ANÁLISE DA EFICIÊNCIA É de grande importância após a definição da técnica de mensuração identificar a ótica e abordagem de eficiência de acordo basicamente com o objeto do estudo bem como os seus inputs e outputs Este estudo apresentará a partir deste momento a revisão sobre as duas principais técnicas utilizadas nesta pesquisa e que foram empregadas para a geração dos resultados preliminares que se encontram aqui demonstrados Esperase frente ao conteúdo deste referencial teórico sobre as técnicas não paramétricas e paramétricas que os resultados obtidos a partir da Análise Envoltória de Dados 31 apresentem médias menores quando comparadas as médias obtidas pela Análise da Fronteira Estocástica e por outro lado uma maior volatilidade BERGER MESTER 1997 221 Análise Envoltória de Dados Ao realizar um estudo que analisava a produção agrícola norte americana Farrel 1957 preconizou a mensuração da eficiência da produtividade a fronteira de eficiência e a DEA ao no desenvolvimento do estudo medir a produtividade a partir da observação da relação existente entre produto e insumo dentro de um processo produtivo específico Considerase uma unidade produtiva eficiente quando se obtém a máxima produção quando aplicado determinada quantidade de insumo para o processo O termo Data Envelopment Analysis DEA ficou conhecido oficialmente quando Charnes Cooper e Rhodes 1978 apresentaram a técnica em um estudo Devido a inicial dos nomes dos autores esta metodologia passou a ser denominada CCR Este modelo também é conhecido como CRS sigla em inglês para Constant Returns to Scale ou retorno constante de escalas Tratase de uma técnica de programação matemática que possibilita avaliar o grau de eficiência produtiva entre várias unidades tomadoras de decisão ou Decision Making Units DMUs ao se considerar os insumos disponíveis inputs e os produtos produzidos outputs Para Fethi e Pasiouras 2010 esta técnica foi identificada em 181 do total de 196 artigos revisados consolidandose desta forma como a técnica mais utilizada para avaliação deste segmento econômico Esta é uma técnica não paramétrica e por este motivo não objetiva otimizar um plano de regressão simples Ou seja a DEA visa a otimização individual de cada uma das observações resultando desta forma uma fronteira de eficiência que é definida como a representação do nível máximo de produção para determinado nível de insumos e baseiase no conceito de ParetoKoopmans GUERREIRO 2006 GRIGORIAN MANOLE 2006 A Figura 1 demonstra a representação gráfica da relação entre a quantidade de insumos e a quantidade produzida por DMU e é útil para ilustrar e exemplificar o conceito de fronteira 32 Figura 1 Fronteira de Eficiência Fonte Guerreiro 2006 Percebese por meio da figura 1 que uma das características da técnica DEA é a união das unidades eficientes por uma linha que resulta em uma superfície côncava onde as unidades ineficientes se encontram abaixo desta superfície Segundo Guerreiro 2006 o índice de eficiência das DMUs ineficientes é calculado através da projeção destas até a linha da fronteira Portanto é a representação numérica da distância da DMU ineficiente até a fronteira Esta projeção determina a utilização dos dois modelos existentes na literatura sendo o modelo orientado ao output e o modelo orientado ao input Becker Lunardi e Maçada 2003 descreveram estas duas vertentes sendo o modelo orientado ao output o que apresenta a orientação que mantem os inputs constantes e busca maximizar os resultados obtidos portanto o fato de produzir mais outputs através da mesma quantidade de inputs Já o modelo orientado ao input foi classificado como sendo o que apresenta a orientação onde há como foco a redução dos insumos mantendo os outputs constantes portanto produzir a mesma quantidade de outputs com a menor utilização de inputs Este estudo apresentou até o momento o conceito inicial da técnica DEA e sua evolução a partir do estudo de Farrel 1957 e a apresentação do conceito de DEA por Charnes 33 Cooper e Rhodes 1978 como uma técnica que constrói fronteiras de eficiência através dos inputs e outputs das DMUs analisadas Porém essa técnica de fronteiras pode apresentar retornos constantes ou variáveis de escala O modelo CCR já apresentado traça fronteiras que apresentam retornos de escala constantes e há também um segundo modelo a ser abordado onde a técnica DEA foi aprimorada pelo estudo de Banker Charnes e Cooper 1984 recebendo o nome de modelo BCC também devido as iniciais dos nomes dos autores O modelo BCC apresenta a construção das fronteiras através de retornos variáveis de escala O modelo CCR que surgiu em 1978 tem a representação matemática para as duas formas de orientação apresentadas na Figura 2 Figura 2 Minimização de Inputs e Maximização de Output CCR Fonte Adaptado de Guerreiro 2006 Onde Eƒƒ0 eficiência da DMU0 Uj vi pesos de outputs e inputs respectivamente Xik yjk inputs i e outputs j da DMUk Xi0 yj0 inputs i e outputs j da DMU0 34 A representação gráfica para a técnica DEA CCR para as duas formas de orientação está apresentada na Figura 3 Figura 3 Representação gráfica CCR Orientado a Inputs e Outputs Fonte Adaptado de Guerreiro 2006 Banker Charnes e Cooper 1984 alteraram a o modelo CCR em duas óticas principais sendo uma a eficiência técnica e a outra a eficiência de escala Ou seja passaram a considerar que as DMUs estão sujeitas a variações de escalas resultando em ganhos ou perdas ou seja retornos variáveis de escala sendo em inglês Variable Returns to Scale que dá origem a sigla VRS Para Freaza Madeiro e Gomes 2004 o modelo BCC identifica a correta utilização dos recursos de acordo com o grau de escala de operação da DMU observada Ou seja a eficiência da escala representa o quociente da eficiência do modelo BCC com a eficiência do modelo CCR O resultado é a medida da distância de uma DMU em análise para uma DMU modelo que é ficticiamente operada com uma escala de tamanho mais produtivo O cálculo de minimização de inputs e maximização de outputs está estampado na Figura 4 35 Figura 4 Minimização de Inputs e Maximização de Output BCC Fonte Adaptado Guerreiro 2006 Onde Eƒƒ0 eficiência da DMU0 Uj vi pesos de outputs e inputs respectivamente Xik yjk inputs i e outputs j da DMUk Xi0 yj0 inputs i e outputs j da DMU0 A representação gráfica para a técnica DEA BCC para as duas formas de orientação está apresentada na Figura 5 Figura 5 Representação gráfica BCC Orientado a Inputs e Outputs Fonte Adaptado Guerreiro 2006 36 Para consolidar a apresentação do conceito de DEA e seus modelos bem como suas possíveis orientações podese apresentar a Figura 6 que sintetiza todo o conteúdo analisado nesta seção O modelo representa as possibilidades de análise e utilização da técnica Ganhos de escala constantes direcionando para a técnica DEA CCR e as suas duas orientações ou seja orientado ao input ou orientado ao output Ganhos de escala variáveis direcionando para a técnica DEA BCC e suas duas orientações ou seja orientado ao input ou orientado ao output Figura 6 Classificação entre Ganhos de Escala e Orientação Fonte Jubran 2006 Uma das técnicas que se destacam para a avaliação bancária é a análise envoltória de dados ou DEA Para Fethi e Pasiouras 2010 esta técnica foi identificada em 181 do total de 196 artigos revisados consolidandose desta forma como a técnica mais utilizada para avaliação deste segmento econômico A técnica DEA possui certa vantagem sobre os tradicionais índices contábeis e financeiros pois pode considerar múltiplos inputs e múltiplos outputs o que é bastante válido 37 no setor bancário NIGMONOVI 2010 GRIGORIAM MANOLE 2006 THANASSOULIS BOUSSOFIANE DYSON 1996 O fato de se considerar múltiplos inputs e outputs é essencial quando se pretende observar estratégias que podem influenciar os resultados em diversos setores ou níveis de organizações Para avaliar o impacto de variáveis independentes na eficiência das DMUs mensurada ao utilizar a DEA aplicase uma Análise de Regressão para modelos que apresentam múltiplos estágios conforme estudos realizados por Despotis e Smirlis 2002 e Zhu 2003 Wanke 2012 discorre sobre as abordagens de DEA em dois estágios que demandam a análise de fatores externos que impactem nos resultados obtidos como escores de eficiência A Análise de Regressão neste estágio possui a mesma eficácia de outros métodos paramétricos aplicados com o objetivo de identificar o impacto das variáveis independentes na variável dependente WANKE 2012 222 Análise da Fronteira Estocástica A SFA surgiu dos estudos feitos por Aigner Lovell e Schimidt 1977 e dos estudos dos pesquisadores Meeusen e van de Broeck 1977 Tratase de uma técnica capaz de reduzir alguns gargalos deixados por técnicas determinísticas como por exemplo o estudo da ineficiência REIFSCHNEIDER STEVENSON 1991 Estes estudos surgiram como contraponto às técnicas não paramétricas com o objetivo de analisar os desvios da fronteira de produção e são capazes de estimar a ineficiência global das unidades ou firmas através da decomposição do termo de erro em dois elementos O primeiro destes elementos referese ao termo que contabiliza os efeitos aleatórios e o segundo referese à ineficiência tecnológica da firma Battese e Corra 1977 também realizaram estudos a fim de identificar estes termos A principal vantagem dos modelos de SFA reside no fato deste tipo de modelo conseguir identificar que os desvios em relação a fronteira de produção ou de custo podem ocorrer devido a ineficiência técnica dos produtores ou seja pela incapacidade de transformar uma quantidade x de insumos em uma quantidade y de produto de maneira tecnologicamente viável fazendo com que este produtor desviese da fronteira ou no reconhecimento de que estes desvios podem ocorrer devido a choques aleatórios ou interferências externas fora do controle dos produtores unidades ou firmas ZANINI 2004 38 Dessa forma podese definir que as firmas localizadas na fronteira determinística são classificadas como tecnicamente eficientes enquanto as firmas que estão abaixo desta fronteira de produção são consideradas tecnicamente ineficientes Greene 2002 sintetiza a proposta do modelo de SFA como um modelo de regressão estimado por verossimilhança que apresenta um distúrbio assimétrico e não normal Kumbhakar e Lovell 2003 classificam as fronteiras de produção as fronteiras de custo as fronteiras de receita e as fronteiras de lucro como estocásticas por apresentarem uma variação aleatória em seu ambiente operacional ou seja a ineficiência causa desvios unilaterais nas fronteiras estocásticas A fim de capturar os efeitos da variação aleatória que ocorre no ambiente operacional e identificar os efeitos das ineficiências técnicas ou alocativas utilizase o erro simétrico e erro unilaterais respectivamente A Figura 7 representa o padrão da SFA de Produção onde as firmas i e j estão inseridas A firma i está apresentada acima da fronteira determinística e a firma j abaixo da fronteira determinística Podese observar que o efeito da eficiência encontrase partindo da quantidade de produtos produzidos até o ponto em que a firma se encontra Também se observa na figura 7 que o efeito do ruído ou seja fatores que estão fora do controle das firmas estão representados e representam a distância do ponto representado pela firma até a fronteira determinística onde encontrase as firmas consideradas eficientes tecnicamente Figura 7 Fronteira Estocástica de Produção Fonte Pontes 2017 39 Segundo estudos de Aigner Lovell e Schimidt 1977 os modelos de SFA por considerarem a ineficiência técnica e os choques aleatórios também chamados de ruído podem ser representados da seguinte maneira 𝑦𝑖 𝑇𝐸𝑖 𝑓𝑥𝑖 𝛽 𝑒𝑣𝑖 Onde 𝑓𝑥𝑖 𝛽 𝑒𝑣𝑖 fronteira de produção estocástica 𝑣𝑖 variável aleatória irrestrita de sinal 𝑇𝐸𝑖 estimativa de eficiência da iésima firma produtora 𝛽 vetor de parâmetros da função de produção a ser estimada 0 𝑇𝐸𝑖 1 Em que a parte que representa a fronteira determinística ou seja 𝑓𝑥𝑖 𝛽 que é comum a todas as firmas e a parte que representada por 𝑒𝑣𝑖 a captura os efeitos de choques aleatórios que afetam a iésima firma produtora Sabese ainda que 𝑇𝐸𝑖 é uma medida de eficiência técnica representada por um valor entre zero e um e que se refere a medida entre a saída da iésima firma em relação à saída que poderia ser produzida por uma firma totalmente eficiente usando o mesmo vetor de entrada de modo que 𝑇𝐸𝑖 𝑦𝑖 𝑓𝑥𝑖 𝛽 𝑒𝑣𝑖 Relativo à forma da função 𝑓𝑥𝑖 𝛽 conforme estudo realizado por Coelli Rao e Battese 2005 esta deve preferencialmente possuir as seguintes características 1 Flexível permitindo aproximações diferenciais de primeira ordem no caso das formas flexíveis de primeiraordem ou de segunda ordem no caso das flexíveis de segundaordem sabendo que essas últimas podem aumentar o número de parâmetros a estimar e a possibilidade de multicolinearidade por exemplo 40 2 Linear em parâmetros tornandoas favoráveis às estimativas usando técnicas de regressão 3 Regular satisfazendo as condições de regularidade econométrica a depender se tratase de uma função produção custo etc Exemplo não negatividade essencialidade fraca convexidade e nãodecrescente em relação aos inputs para uma função de produção 4 Parcimoniosa sendo nas palavras de Coelli Rao e Battese 2005 a forma funcional mais simples que realize o trabalho adequadamente As principais funções determinísticas são retratadas pelo Quadro 1 Quadro 1 Principais funções determinísticas Fonte Adaptado de Coelli Rao e Battese 2005 Com a finalidade de desenvolver a última equação apresentada de uma maneira mais prática diversos autores em seus estudos sobre SFA consideraram a função 𝑓𝑥𝑖 𝛽 como 41 linear e aplicaram uma transformação logarítmica com a finalidade de isolar o termo 𝜇𝑖 como o erro que captura os efeitos da ineficiência técnica AIGNER LOVELL SCHMIDT 1977 MEEUSEN VAN DEN BROECK 1977 KUMBHAKAR GHOSH MCGUKIN 1991 REIFSCHNEIDER STEVENSON 1991 BATTESE COELLI 1995 KUMBHAKAR LOVELL 2003 COELLI RAO BATTESE 2005 Desta forma ao aplicarse uma transformação logarítmica a equação 𝑦𝑖 𝑇𝐸𝑖 𝑓𝑥𝑖 𝛽 temos o seguinte desenvolvimento 𝐿𝑛𝑦𝑖 𝐿𝑛𝑓𝑥𝑖 𝛽 𝐿𝑛𝑇𝐸𝑖 Sendo que 𝐿𝑛𝑦𝑖 𝐿𝑛𝑓𝑥𝑖 𝛽 𝑢𝑖 Onde 𝑢𝑖 é uma medida de ineficiência técnica 𝑢 𝐿𝑛𝑇𝐸𝑖 0 𝑢 1 𝑇𝐸𝑖 Então 𝑇𝐸𝑖 𝑒𝑢 Desta forma a equação acima pode ser reescrita conforme abaixo 𝑦𝑖 𝑓𝑥𝑖 𝛽 𝑒𝑣𝑖𝑢𝑖 Segundo Coelli Rao e Battese 2005 a análise de SFA em sua maioria é direcionada para a previsão dos efeitos da ineficiência A medida mais comum de eficiência técnica orientada para o produto é a relação entre a produção observada e a SFA correspondente 𝑇𝐸𝑖 𝑦𝑖 𝑦𝑖 𝑒𝑥𝑝𝛽0 𝛽𝑖𝑥𝑖 𝑣𝑖 𝑢𝑖 𝑒𝑥𝑝𝛽0 𝛽𝑖𝑥𝑖 𝑣𝑖 exp 𝑢𝑖 42 Onde 𝑇𝐸𝑖 estimativa de eficiência da iésima firma produtora 𝑦𝑖 quantidade de um único produto 𝛽 vetor de parâmetros da função de produção a ser estimada 𝑣𝑖 variável aleatória irrestrita de sinal 𝑢𝑖 é uma medida de ineficiência técnica Ainda de acordo com as contribuições de Coelli Rao e Battese 2005 o resultado obtido da fórmula acima mensura a eficiência técnica com seu valor entre zero e um e a relação entre a produção que poderia ser produzida por uma empresa totalmente eficiente e utilizando o mesmo input em relação a iésima empresa Devido à característica do modelo que apresenta um erro simétrico 𝑢𝑖 e uma variável aleatória não negativa 𝑣𝑖 há a necessidade de se estimar os parâmetros de análise com atenção a estes dois termos Devese levar em consideração que cada 𝑣𝑖 é distribuído independentemente de cada 𝑢𝑖 COELLI RAO BATTESE 2005 Aigner Lovell Schimidt 1977 através do modelo HalfNormal utilizaramse de duas premissas A premissa de que a variável aleatória 𝑣𝑖 é considerada uma iid independente e identicamente distribuída com distribuição normal que assume média zero e variância 𝜎𝑣 2 E a premissa que considera que a medida de ineficiência técnica 𝜇𝑖 também considerada uma iid com distribuição halfnormal e variância 𝜎𝑢 2 Segundo Coelli Rao e Battese 2005 outros modelos também podem ser utilizados em substituição ao modelo halfnormal como o modelo de fronteira normal truncada e o modelo gamma No contexto desta dissertação e considerando a geração de fronteiras por meio da técnica de SFA é importante identificar os impactos das variáveis independentes nos escores ou variável dependente Com isso a heterocedasticidade que ocorre quando a variância dos resíduos não é constante atua como ferramenta na identificação da ineficiência das firmas Brito 2016 aborda o comportamento heterocedástico do fator de ineficiência e o comportamento homocedástico do termo aleatório afirmando com isto que não considerar este fato na análise da fronteira de produção ocasiona em uma perda de eficiência dos estimadores Para o autor o modelo heterocedástico permite identificar quais fatores ou características de cada firma estimulam a eficiência ou a ineficiência 43 Para considerar esse aspecto conforme indicado por Danelon 2018 detalhase o termo 𝑢𝑖 das fórmulas anteriores de modo que 𝑢𝑖 𝛿𝑧𝑖 𝑤 Em que 𝑧𝑖 referese a cada uma das varíaveis que afetam a ineficiência técnica e 𝛿 é um vetor de parâmetros a ser determinado A utilização da heterocedasticidade foi considerada neste estudo a fim de mensurar os impactos que as variáveis independentes podem ocasionar nos escores gerados através da técnica de SFA Segundo Danelon 2018 outros fatores do ambiente de produção podem afetar os resultados da eficiência técnica tornandoos componentes importantes para a análise da produtividade 23 EFICIÊNCIA BANCÁRIA Os estudos de eficiência bancária podem ser divididos em dois níveis de acordo com as partes interessadas stakeholders Em um nível macro há evidências de que o crescimento econômico está positivamente relacionado à eficiência do setor bancário ABEDIFAR HASAN TARAZI 2016 BERGER et al 2004 Em um nível micro os estudos de eficiência fornecem informações de benchmarking que serão do interesse de gerentes de bancos e formuladores de políticas a fim de melhorar o desempenho dos bancos A eficiência bancária e a dinâmica relacionada estão intimamente ligadas à tomada de risco e capitalização do banco enquanto a competição cada vez maior no setor de serviços financeiros torna a eficiência uma consideração primária para os gestores bancários reguladores e formuladores de políticas FIORDELISI MARQUESIBANEZ MOLYNEUX 2011 Altunbas et al 2007 encontravam evidências em bancos europeus que analisaram que mostram relação positiva entre risco no nível de capital e liquidez e que a solidez financeira do setor influencia positivamente na redução da assunção de riscos bancários e dos níveis de capital De igual forma um efeito positivo da ineficiência na assunção de riscos foi encontrado em trabalho de Kwan e Eisenbeis 1997 que identificaram que o efeito positivo verificado da ineficiência sobre o nível de capital pode ser atribuído à pressão regulatória sobre as 44 instituições de baixo desempenho Os autores também constataram a relação direta entre a ineficiência e o crescimento dos empréstimos indicativo de que a eficiência operacional melhora a uma taxa decrescente à medida em que a taxa de crescimento dos empréstimos aumenta o que apoia a hipótese de que os gestores que buscam crescimento tendem a operar de forma ineficiente KWAN EISENBEIS 1997 Por meio da SFA Lensink Meesters e Naaborg 2008 analisaram 2095 bancos comerciais em 105 países no período de 1998 a 2003 e também descobriram que a propriedade estrangeira afeta negativamente a eficiência do banco efeito menos pronunciado em países com boa governança Seu trabalho também apurou que a qualidade das instituições no país de origem e a maior similaridade entre a qualidade institucional do país de origem e a do país anfitrião reduzem a ineficiência dos bancos estrangeiros LENSINK MEESTERS NAABORG 2008 Já Nigmonov 2010 utilizandose da DEA concluiu que houve diminuição dos níveis gerais de eficiência de bancos do Uzbequistão analisados no período de 2004 a 2006 tendo como principal fonte de ineficiência se a eficiência técnica que varia conforme o porte das instituições bancárias Yin 2021 em seu estudo que contempla um conjunto de dados abrangente cobrindo 148 países no período de 19952015 investigou a relação entre concorrência e eficiência no setor bancário As evidências mostram que a competição bancária é prejudicial à eficiência de custos O ambiente regulatório e institucional do banco no qual os bancos operam não apenas influencia a eficiência do banco mas também afeta o vínculo entre competição e eficiência Especificamente regulamentações restritivas sobre atividades bancárias e requisitos de capital rigorosos levam à ineficiência de custos mas supervisões e compartilhamento de informações eficazes dos registros de crédito são mais propícios à eficiência do banco Além disso este estudo conclui que o impacto adverso da concorrência sobre a eficiência pode ser mitigado ou mesmo revertido por regulamentações rigorosas supervisões bem implementadas eou mecanismos eficazes de compartilhamento de informações Para Duan e Zhang 2001 a pesquisa sobre a eficiência dos bancos está crescendo naturalmente e a literatura anterior enfatiza dois conceitos diferentes de eficiência econômica para medir a eficiência do banco eficiência de custo e eficiência de lucro A eficiência de custos é o critério mais utilizado em relação ao outro eficiência de lucro na literatura Berger 1995 encontra uma relação negativa entre eficiência de custos e empréstimos inadimplentes e ele também revela que a eficiência de custos pode ser um indicador importante do problema futuro dos bancos BERGER DEYOUNG 1997 FRIES TACI 2005 Recentemente Tabak Fazio e Cajueiro 2012 e Silva et al 2016 descobriram que os bancos grandes parecem superar os 45 bancos não grandes em termos de eficiência de custo e lucro Assaf et al 2019 examinam os efeitos da eficiência do banco em tempos normais sobre o desempenho do banco eles descobriram que a eficiência de custo durante tempos normais ajuda os bancos a reduzir o risco e aumentar a lucratividade durante as crises financeiras institucionais subsequentes enquanto a eficiência do lucro tem benefícios limitados Yin Yang e Lu 2020 discutem as associações entre globalização bancária e eficiência Eles descobriram que a entrada de bancos estrangeiros está associada a uma menor eficiência nos países anfitriões As instituições financeiras possuem elevada importância e geram diversos impactos a economia real YANG et al 2020 Por isso a mensuração da eficiência bancária deve assumir uma parcela das análises destas instituições tanto em tempos de crise financeira onde auxiliam na avaliação dos prováveis impactos do sofrimento de uma determinada instituição financeira na estabilidade do sistema financeiro quanto em tempos normais onde a utilização dessas ferramentas é crucial para calibrar instrumentos prudenciais como capital requisitos e prêmios de seguro de acordo com a contribuição relativa de diferentes instituições ao risco sistêmico GRAVELLE LI 2013 Para Shamshur e Weill 2019 a eficiência de custos dos bancos é uma medida ampla de desempenho bancário que tem sido frequentemente utilizada na literatura bancária empírica nas últimas duas décadas Ele mede a capacidade de um banco de operar com custos mais baixos comparando sua estrutura de custos com a de um banco com as melhores práticas Uma vasta literatura sobre eficiência bancária tem se concentrado em medir o nível e os determinantes da eficiência bancária em todo o mundo BERGER HASAN ZHOU 2009 FUJI MANAGI MATOUSEK 2014 GODDARD MOLYNEUX WILSON 2004 com o objetivo de melhorar o desempenho das instituições financeiras ALTUNBAS et al 2007 BONIN HASAN WACHTEL 2005 BERGER BONACCORSI DI PATTI 2006 BARTH CAPRIO LEVINE 2013 Para Thanassoulis 2003 existem motivações que levam ao estudo e a análise da eficiência Tais motivações podem ser descritas na forma de incentivos que levam a melhorias no processo produtivo Alguns dos principais incentivos descritos são focados na produção e em seu resultado considerando os ganhos com as taxas marginais de substituição de fatores de produção a redução de custos na utilização dos recursos avaliação da variação de produtividade entre outros 46 24 FERRAMENTAS DE AVALIAÇÃO DA EFICIÊNCIA BANCÁRIA OU MENSURAÇÃO DA EFICIÊNCIA BANCÁRIA A eficiência bancária é objeto de estudo relevante e uma das formas de mensurála baseiase na utilização de técnicas estatísticas que frente a cálculos predefinidos geram fronteiras de eficiência BERGER HUMPHREY 1997 Sabese que os escores de eficiência gerados por ambas as técnicas oferecem a possibilidade de críticas frente a técnica escolhida e por este motivo este trabalho utiliza conjuntamente ambas as técnicas para a análise de uma mesma base de dados com o intuito de comparar os resultados geradas e auxiliar no entendimento da eficiência bancária brasileira de 2000 a 2018 Através da definição de Cooper Seiford e Zhu 2011 a técnica de DEA desenvolvida por Charnes Cooper e Rhodes 1978 é uma das técnicas de maior relevância e reconhecimento para a mensuração de eficiência e aceita pela comunidade científica como modelo de processos operacionais A utilização desta técnica é amplamente empregada e encontrada na literatura como ferramenta para a mensuração de eficiência bancária devido ao seu caráter de técnica não paramétrica Esta característica a classifica como uma técnica mais flexível em relação as variáveis utilizadas PERICO REBELATTO SANTANA 2008 Outro fator de grande relevância é de que a utilização da ferramenta DEA ocorreu em praticamente todos os sistemas bancários internacionais FETHI PASIOURAS 2010 É possível identificar que a metodologia de análise empregada pela técnica DEA possui ampla utilização quando se trata de estudos que avaliam o sistema bancário O Quadro 2 sintetiza relevante literatura seminal acerca da utilização da DEA na avaliação do sistema bancário Quadro 2 Quadro sinótico de revisão bibliográfica sobre utilização de DEA na avaliação de eficiência do sistema bancário Autores Resumo do trabalho Paradi e Schaffnit 2004 Os autores avaliaram o desempenho de filiais de uma grande rede canadense de bancos utilizandose de DEA Dois modelos foram considerados no estudo um direcionado à observação do uso dos recursos e o outro direcionado à análise da gestão sênior da instituição 47 Mostafa 2009 O estudo investigou a eficiência de bancos árabes utilizando duas metodologias quantitativas DEA e redes neurais Os resultados indicam que a precisão da análise dos modelos preditivos é similar à de métodos estatísticos tradicionais O estudo destacou a importância econômica de se incentivar o aumento da eficiência no setor bancário árabe Yang et al 2010 O trabalho utilizouse de uma análise híbrida minimax e DEA aplicada a um estudo de caso realizado em quatorze filiais de um banco internacional de Manchester a fim de avaliar a eficiência de instituições bancárias Os resultados serviram ao apoio à tomada de decisão sobre definição de metas e alocação de recursos Siriopoulos e Tziogkidis 2010 O trabalho analisou a eficiência de bancos comerciais gregos no período de 1995 a 2003 por meio da análise DEA a duas abordagens utilizadas para medir eficiência Os resultados da pesquisa sugerem que o setor bancário grego operou com eficiência na média durante períodos de desestabilização Avkiran 2011 Este trabalho investiga a extensão a análise DEA de eficiência bancária está associada aos principais índices financeiros A utilização da DEA para análise de eficiência surgiu como combinação mais significativa e explica a variação NOPAT e ROE Luo Bi e Liang 2012 Neste trabalho foi proposto um novo método de escolha de variáveis DEA baseado na concepção de valor adicionado ao dinheiro Este método trouxe como vantagens maior objetividade por evitar a influência de fatores subjetivos subsequente aos cálculos e o fato de proporcionar aos gestores e pesquisadores a medida de variáveis e sua exata classificação O estudo analisou quatorze bancos comerciais chineses com testes estatísticos e de regressão satisfatórios Paradi Zhu e Edelstein 2012 O trabalho desenvolveu uma abordagem que utiliza a DEA como framework para identificar grupos gerenciais e grupos líderes em desempenho em cerca de mil filiais de um grande banco canadense As vantagens desta nova abordagem foram demonstradas por meio da comparação dos resultados obtidos com algoritmos tradicionais de clustering Chortareas Girardoni e Ventouri 2012 O trabalho investigou a dinâmica entre políticas regulatórias e de supervisão e vários aspectos de eficiência e desempenho de bancos comerciais de 22 países da União Europeia entre 2000 e 2008 O primeiro estágio do trabalho utilizou a DEA para análise de eficiência Foi ainda conduzida uma análise de sensibilidade sobre a robustez do modelo Os resultados demonstraram que o fortalecimento das restrições de capital e do poder de supervisão das operações melhora a eficiência bancária Fonte elaborado pelo autor 48 O método SFA desenvolveuse a partir dos estudos simultâneos de Aigner Lovell e Schmidt 1977 e dos estudos de Meeusene e Van Den Broeck 1977 Recentemente a quantidade de estudos sobre eficiência bancária através da utilização da técnica de SFA vem ganhando força dentre os estudos existentes e frente aos estudos que envolvem a DEA Trabalhos envolvendo eficiência bancária e SFA têm demonstrado a importância da técnica e do fato de se levar em consideração o termo de ineficiência como se observa na compilação de trabalhos apresentada no Quadro 3 Quadro 3 Quadro sinótico de revisão bibliográfica sobre utilização da Fronteira Estocástica para a análise da eficiência bancária Autores Resumo do trabalho Dong et al 2016 O trabalho examinou a eficiência de custo e lucratividade de quatro tipos de bancos comerciais chineses no período de 2002 a 2013 Os autores descobriram que as eficiências de custo e lucratividade dos bancos melhoraram ao longo dos anos e que as instituições são mais eficientes em custo que em lucratividade O gap de eficiência de lucro entre os bancos estrangeiros e os bancos domésticos aumentou no período de 2007 a 2013 Estrutura de ativos competição de mercado e porte são os principais determinantes da eficiência dos bancos chineses Boucinha Ribeiro e WeymanJones 2013 O artigo analisou analisa a tecnologia de produção dos bancos portugueses entre 1992 e 2006 período que engloba a adesão de Portugal à área do Euro Como resultado o estudo identificou que o avanço tecnológico deslocou a fronteira de custos para baixo ao longo do período considerado enquanto a distância em que os bancos operaram parece ter permanecido constante Notouse ainda que aumentos na produção em economias de escala também contribuíram para o aumento registrado na produtividade Jiang Yao e Feng 2013 O artigo combinou o efeito estático da propriedade e o efeito dinâmico da privatização com o desempenho de bancos chineses no período de 1995 a 2010 Foi observado um desempenho significativamente maior por intermediários privados bancos comerciais com ação conjunta com bancos comerciais municipais em relação aos bancos estatais Os bancos de capital aberto sujeitos a monitoramento do mercado apresentam melhor desempenho independentemente do status de propriedade A privatização de bancos melhorou o desempenho no que diz respeito ao ingresso de receitas e ganhos de eficiência no curto ou longo prazo Os resultados sugerem que os bancos chineses são muito mais eficientes na geração de receita de juros do que na captação de receita não financeira Tabak Fazio e Cajueiro 2012 Este artigo estudou o efeito da competição de bancos latino americanos na assunção de riscos e analisou se a capitalização e o porte da instituição interferem nessa relação Concluiuse que a competição afeta o risco de forma não linear a intensidade da 49 competição está relacionada à estabilidade o porte do banco explica a vantagem da competição enquanto a capitalização só é positiva para bancos maiores o índice de capital explica a vantagem da concorrência mais baixa Fiordelisi MarquesIbanez e Molyneux 2011 Foi avaliada a relação intertemporal entre eficiência capital e risco bancário em uma amostra de bancos comerciais europeus empregando várias definições de eficiência risco e capital Os resultados sugerem que a menor eficiência do banco em relação aos custos e receitas causa maior risco e que aumentos no capital do banco precedem melhorias na eficiência de custo Também descobriuse que bancos mais eficientes eventualmente se tornam mais capitalizados e que níveis de capital mais altos tendem a ter um efeito positivo sobre os níveis de eficiência Rossi Schwaiger e Winkler 2009 O artigo analisou como a diversificação dos bancos de acordo com o tamanho e o ramo de atividade afeta o risco a eficiência de custo e lucro e a capitalização bancária Foram analisados os grandes bancos comerciais austríacos durante os anos de 1997 a 2003 Foram testados vários tipos diferentes de hipóteses gerenciais formalizadas de acordo com uma versão modificada do modelo de Berger e DeYoung Descobriuse que embora a diversificação afete negativamente a eficiência de custos ela aumenta a eficiência dos lucros e reduz o risco realizado dos bancos Por fim a diversificação parece ter um impacto positivo na capitalização dos bancos Berger Hasan e Zhou 2009 O artigo analisou a eficiência dos bancos chineses no período de 1994 a 2003 As descobertas sugerem que os quatro grandes bancos estudados são de longe os menos eficientes os bancos estrangeiros são mais eficientes e a propriedade estrangeira minoritária está associada a uma eficiência significativamente melhorada Foram apresentadas verificações de robustez que corroboram os resultados Essas descobertas sugerem que a propriedade estrangeira minoritária dos quatro grandes bancos em via de privatização provavelmente melhorará seu desempenho de forma significativa Fu e Heffernan 2009 Este artigo investigou a relação entre a estrutura de mercado e o desempenho do sistema bancário chinês de 1985 a 2002 O modelo foi estendido para considerar questões como o impacto do tamanho propriedade do banco e se os quatro grandes bancos desfrutam de uma vida tranquila Em média a eficiência diminuiu significativamente e a maioria dos bancos estava operando abaixo dos níveis de eficiência de escala Não houve evidência para apoiar a hipótese da vida tranquila provavelmente porque os controles rígidos das taxas de juros impediram os bancos estaduais de obter lucros de monopólio Fonte Elaborado pelo autor 25 FATORES EXPLICATIVOS DO DESEMPENHO DE INSTITUIÇÕES FINANCEIRAS Algumas peculiaridades do sistema financeiro tornamno suscetível à uma combinação de riscos incomum aos demais players econômicos 50 Especialmente face à sua estrutura dinâmica e à complexa natureza do ambiente econômico em que operam os bancos enfrentam uma diversidade de riscos EKINCI POYRAZ 2019 Koch e Macdonald 2014 argumentam que os riscos enfrentados pelos bancos podem ser categorizados em seis risco de crédito risco de liquidez risco de mercado risco operacional risco nominal e riscos legais Cada um tem potencial de impactar negativamente a lucratividade o valor de mercado o passivo e o patrimônio líquido das instituições financeiras KOCH MACDONALD 2014 Como a principal fonte de receita do setor bancário consiste em empréstimos concedidos por bancos comerciais o risco de crédito defendem Ekinci e Poyraz 2019 é um dos mais importantes riscos enfrentados pelos bancos Risco de crédito assim definido pelo Comitê de Supervisão Bancária da Basiléia BASEL COMMITTEE ON BANKING SUPERVISION 2000 é a probabilidade de perda parcial ou total do empréstimo em aberto devido ao não pagamento do empréstimo no prazo Em tempos de crise a reação dos bancos menos capitalizados que compram títulos com rendimento mais baixo consiste no aumento dos títulos sobre a oferta de crédito em busca de mais liquidez e portanto segurança PEYDRÓ POLO SETTE 2021 O aumento do risco de crédito impacta no aumento do custo marginal da dívida e do patrimônio líquido e consequentemente no aumento do custo de captação da instituição e na tendência dos bancos de vivenciarem uma crise EKINCI POYRAZ 2019 A propósito disso De Brandt Lecarpentier e Pouvelle 2020 encontraram forte interação entre a liquidez do mercado e a liquidez do financiamento bancário durante os períodos de crise resultado impulsionado pelos bancos menos líquidos 26 INDICADORES DE RISCO DE INSTITUIÇÃO FINANCEIRA CAMELS Um sistema financeiro equilibrado e fortalecido deve ser obtido em conjunto a uma economia saudável e um setor bancário sólido e eficiente O sistema bancário desempenha um papel essencial na criação de ativos produtivos e geração de superávits As instituições bancárias utilizam estratégias e procedimentos para prevenir possíveis deteriorações early warning system que são projetados para gerar os ratings de solidez individual daquele banco JEELANBASA JEELANBASHA 2018 PIMENTA 2014 Diversos modelos de classificação de risco foram desenvolvidos a fim de funcionar como preditores de crises bancárias auxiliando na identificação das instituições cuja condição 51 merece atenção especial na supervisão de seus regimes de riscos Essas classificações são baseadas em avaliações subjetivas de vários aspectos do funcionamento de uma instituição bancária onde o examinador pode incluir outros fatores pertinentes à sua avaliação Geralmente resultados obtidos são compartilhados apenas com a administração do banco em questão e não são divulgadas publicamente SAHAJWALA VAN DEN BERGH 2000 Em 1979 o Federal Financial Institutions Examination Council FFIEC um comitê interno do Federal Reserve System criou o UFIRS Uniform Financial Institutions Rating System cujo objetivo era avaliar de maneira uniforme os riscos e a solidez das instituições bancárias por meio de análise dos critérios de supervisão e previsão de riscos das instituições financeiras de maneira individual A estratégia utilizada foi conhecida como rating CAMEL SAHAJWALA VAN DEN BERGH 2000 Inicialmente o acrônimo CAMEL possuía cinco componentes de avaliação porém com a mudança nos mercados globais políticas e procedimentos de supervisão em 1997 os órgãos norteamericanos viram a necessidade de incluir mais um indicador dessa forma surgiu o termo CAMELS O rating CAMELS avalia seis medidas de desempenho sendo C Adequação de capital A Qualidade dos ativos M Gestão Administração E Lucro e L Liquidez e S Sensibilidade ao risco de mercado SAHAJWALA VAN DEN BERGH 2000 261 Adequação de capital capital adequacy Conforme citado anteriormente o CAMELS avalia seis indicadores de maneira distinta O primeiro item avaliado é a adequação de capital representada pela letra C Capital adequacy Esse item envolve pontos essenciais de análise do capital como a relação entre o nível e a qualidade do capital com a condição financeira geral da instituição necessidade de capital emergente adicional característica e volume dos ativos problemáticos e sua adequação para as possíveis perdas com os empréstimos e arrendamentos balanço patrimonial detalhado contendo os ativos intangíveis riscos de mercado riscos de concentração e riscos associados à atividades não tradicionais riscos por atividades extrapatrimoniais perspectivas e planos de crescimento acessibilidade a mercados de capitais e outras fontes de capital FEDERAL RESERVE SYSTEM 1996 Para Dincer et al 2011 índices que compõem a adequação de capital demonstram a capacidade financeira de uma instituição em cumprir suas funções frente aos e riscos 52 envolvidos como o risco de crédito de mercado e risco operacional assim sendo estabelecem qual a parcela do capital é utilizado para suportar os ativos de risco dos bancos Pile e Paradi 2002 afirmam tratarse de um índice que pode revelar a possibilidade de falhas de uma instituição financeira em lidar com o risco Canbas et al 2005 afirmam ser um importante índice para mensurar a possibilidade de risco de default de uma instituição Mili et al 2017 defendem que para as autoridades reguladoras o problema é definir as ferramentas para controlar e fiscalizar a saúde financeira de um banco isso gerou o conceito de adequação de capital em meados da década de 1970 A expansão dos empréstimos dos bancos sem aumento paralelo de seu capital obrigou os reguladores a definir diversos procedimentos de controle e a lançar novas reformas para evitar a insolvência do setor bancário No contexto de proteção contra choques os requisitos de capital são a regulamentação prudencial mais frequentemente citada A importância do capital regulatório para garantir a estabilidade bancária tem motivado diversos pesquisadores a estudar os determinantes dos bancos de capital regulatório O exame das relações entre as classificações de crédito e a estrutura de capital dos bancos em um contexto de vários países é oportuno e importante para a lucratividade estabilidade e regulamentação dos bancos pelas seguintes razões Em primeiro lugar os ativos e passivos dos bancos são fundamentalmente diferentes daqueles das empresas não financeiras FAN et al 2012 e as informações embutidas em uma classificação de crédito constituem uma base de confiança e segurança dos clientes na qualidade dos ativos de um banco o que é importante para a estabilidade do banco MILLON THAKOR 1985 Em segundo lugar várias regulamentações relacionadas ao capital dos bancos e investimentos em títulos estão diretamente vinculadas às classificações de crédito ressaltando assim a importância das classificações para a lucratividade e estabilidade dos bancos KISGEN 2019 De fato Kisgen 2019 apropriadamente observa que os reguladores financeiros contam com classificações de crédito e níveis de classificação para decidir se permitem que determinados grupos de investidores como bancos invistam em títulos de outros bancos e em que medida os grupos de investidores atendem aos requisitos de capital específicos ao emprestar para outros bancos Terceiro a nova regulamentação de capital Basileia III impõe que os bancos tenham pelo menos 3 do capital em relação ao total de seus ativos Este novo requisito de índice de alavancagem coincide com o índice do Lehman Brothers em sua última demonstração de posição financeira antes de ir à falência ADMATI HELLWIG 2019 53 262 Qualidade dos ativos asset quality O segundo item qualidade dos ativos Asset quality avalia se há potencial risco de crédito associados às carteiras de empréstimos e investimentos tanto imóveis quanto outros ativos Nesse item são avaliados todos os riscos que afetam o valor ou a comercialização dos ativos incluindo os riscos de mercado de imagem operacionais e estratégicos FEDERAL RESERVE SYSTEM 1996 Este item demonstra o nível de risco dos ativos e a capacidade financeira bancária sendo utilizado na avaliação da condição financeira atual e futura ao auxiliar com a previsão de ameaças a lucratividade e a viabilidade da instituição financeira DINCER et al 2011 Para Curry Elmer e Fissel 2003 os índices de qualidade dos ativos são utilizados para identificar os ativos que podem gerar dificuldades financeiras as instituições bancárias Já Affes e HentatiKaffel 2017 indicam que estes índices demonstram quanto capital é utilizado para apoiar os ativos de risco de uma instituição bancária mensurando a alavancagem e a resistência a choques financeiros Para Beltrame Previtali e Sclip 2018 o aumento dos requisitos de capital impostos pelas autoridades reguladoras no âmbito da estrutura de Basileia ou adicionalmente exigidos para situações específicas contribuem para apoiar os bancos na absorção de perdas em empréstimos inadimplentes e de forma mais geral promover a resiliência do sistema bancário a potenciais choques negativos Apesar de sob o ponto de vista dos banqueiros levantar novo capital social ser caro a evidência empírica sugere uma relação inversa entre capital próprio e risco sistemático em todos os principais mercados bancários Ahamed 2017 afirma que a qualidade dos ativos é considerada uma das principais questões no caminho da solidez e saúde financeira de um banco a inadimplência em empréstimos e adiantamentos e a baixa qualidade dos ativos tem repercussão significativa para toda a economia de um país Embora haja evidências do efeito adverso da baixa qualidade dos ativos sobre a lucratividade há evidências reconhecidamente limitadas ou inexistentes sobre a ligação entre a diversificação da receita e a lucratividade em bancos com diferentes qualidades de ativos Na categoria de indicadores de qualidade de ativos considerase o índice de Nonperforming Loan NPL o índice de cobertura abaixo do padrão e os empréstimos depreciados não reservados Este grupo inclui informações úteis para analisar a credibilidade financeira dos bancos e suas exposições ao risco que são derivadas do negócio principal dos bancos ou seja suas atividades de empréstimo O índice de inadimplência é a proporção de 54 empréstimos inadimplentes brutos pelo para o total de empréstimos e contas a receber brutos com clientes BERGER KLAPPER TURKARISS 2009 Para Forgione e Migliardo 2018 o índice define empréstimos inadimplentes como empréstimos que estão vencidos por mutuários em dificuldades Como esse índice é uma medida do valor dos empréstimos duvidosos uma parte do índice de inadimplência provavelmente se tornará perdas no período subsequente Quanto menor a proporção melhor é a qualidade do ativo A este respeito a recente crise financeira degradou o desempenho financeiro das empresas e por sua vez expandiu a quantidade de empréstimos inadimplentes 263 Gestão administração management No item gestãoadministração representado pela letra M Management mensurase a capacidade dos gestores em identificar monitorar e controlar os riscos de atividades gerais da instituição a fim de garantir operação segura sólida eficiente e respeitando as leis e regimentos do país Os riscos de crédito de mercado de imagem conformidade e liquidez são itens avaliados nesse tópico mediante a postura e tomada de decisão do gestor FEDERAL RESERVE SYSTEM 1996 Para Dincer et al 2011 é um componente que abrange um grande número de variáveis como por exemplo a experiência da gerência o que torna difícil a mensuração quando comparado aos demais índices Assim como para Affes e HentatiKaffel 2017 que complementam o abordado anteriormente ao afirmar que este indicador abrange a qualidade dos gestores bem como da estrutura de gerenciamento das instituições financeiras Para DeYoung 1998 os resultados obtidos em seus estudos demonstram que a comum afirmação de que uma eficiência nos custos reflete a qualidade da gestão é verdadeira e traduz o fato de que uma instituição bem gerida utiliza os seus recursos de maneira mais eficiente 264 Lucro earnings O lucro earnings é medido não somente pela quantidade dos ganhos propriamente ditos mas também pelos fatores que podem influenciar a sustentabilidade dos ganhos como os riscos de créditos excessivos ou inadequados Os lucros podem ser influenciados pela dependência indevida de ganhos extraordinários eventos não recorrentes ou efeitos fiscais favoráveis FEDERAL RESERVE SYSTEM 1996 55 Tratase do índice mais utilizado nos estudos nacionais e internacionais e por retratar a rentabilidade das instituições financeiras possui uma relação inversamente proporcional a falhas e ineficiência AVKIRAN CAI 2012 Pennacchi e Santos 2021 citam trabalhos que reconheceram o foco dos bancos no ROE mas principalmente para criticar o ROE como uma meta de desempenho devido ao seu fracasso em contabilizar o risco de patrimônio Alessandri e Haldane 2009 e Drake 2001 mostram que para os bancos do Reino Unido a década de 1970 sinalizou um mar de mudança à medida que o ROE médio dos bancos saltou de cerca de 7 para cerca de 20 e atribuem isso ao menor capital e maior risco de ativos bancários Begenau e Stafford 2016 Eles mostram empiricamente que os bancos com baixo retorno sobre ativos ROA tentam manter um alto ROE usando maior alavancagem Além disso as medidas do valor contábil do mercado de ações do patrimônio do banco estão altamente correlacionadas com o ROE Juntando esses dois fatos eles sugerem que os bancos manipulam o ROE para cima por meio da alavancagem porque os investidores do mercado de ações ineficientemente se concentram no ROE Pennacchi e Santos 2021 argumentam que escolhida uma meta de ROE o modelo criado prevê que os bancos seriam especialmente resistentes à regulamentação póscrise financeira como Basileia III que gradualmente os forçou a aumentar seu capital Quando o valor do contrato é baixo padrões de capital mais elevados reduzem o excesso de patrimônio líquido de um banco levandoo a escolher o padrão de capital mínimo Assim um subproduto dos requisitos de capital mais elevados é a pressão para baixo no ROE do banco o que agora pode tornálo uma métrica de desempenho pior do que o EPS Consequentemente podemos esperar que os bancos diminuam a ênfase do ROE em favor do EPS Finalmente embora nosso artigo forneça uma teoria positiva para o foco dos bancos no ROE pesquisas futuras podem considerar se há uma justificativa normativa para que a regulamentação imponha uma métrica de desempenho diferente Ekinci e Poyraz 2019 analisam o impacto do risco de crédito no desempenho dos bancos estatais bancos privados e bancos estrangeiros que operavam na Turquia entre 2005 e 2017 a fim de comparar os bancos de acordo com sua estrutura de propriedade O Retorno sobre o Ativo ROA e o Retorno sobre o Patrimônio Líquido ROE foram usados como indicadores de desempenho financeiro enquanto o NonPerforming Loans NPLs foi usado como indicadores de risco de crédito Os resultados da estimativa mostraram que existe uma relação negativa entre o risco de crédito e o ROA bem como entre o risco de crédito e o ROE sugerindo então que existe uma relação entre a gestão do risco de crédito e a rentabilidade bancária 56 265 Liquidez liquidity No item liquidez liquidity devemse comparar os níveis atuais e as principais fontes de liquidez da instituição com as necessidades de financiamento levando em consideração a gestão de fundos em relação ao tamanho complexidade e perfil de risco da instituição Ou seja é importante garantir que o nível de liquidez seja suficiente para cumprir as obrigações financeiras e atender as necessidades bancárias da comunidade FEDERAL RESERVE SYSTEM 1996 Conforme observado pelo Comitê de Basileia BASEL COMMITTEE ON BANKING SUPERVISION 2000 alguns bancos faliram durante a fase inicial de estresse de liquidez da última crise financeira apesar de terem níveis de patrimônio robustos De acordo com o Comitê um dos motivos para essas falhas parece ter sido o fato de os bancos não administrarem sua liquidez de forma prudente Para Meriläinena e Junttila 2020 uma solução proeminente dos problemas de liquidez no setor bancário deflagrados pela crise financeira de 20082009 a estrutura de Basileia III introduz duas melhorias regulatórias o índice de financiamento estável líquido NSFR e o índice de cobertura de liquidez ICL O ICL visa garantir que os bancos tenham um estoque adequado de ativos líquidos para atender às suas necessidades de liquidez de curto prazo O NSFR por sua vez é um índice de liquidez estrutural que trata dos descasamentos de prazos entre ativos e passivos O novo regulamento visa reduzir os riscos nestas funções estabelecendo requisitos mínimos para os dois índices O regulamento tenta evitar problemas que possam ocorrer devido a choques de liquidez Esta visão prevê que a criação de liquidez deixa os bancos vulneráveis à insolvência quanto maior a criação de liquidez maior a probabilidade de falência do banco CANBAS CABUK KILIC 2005 Mas outra visão prevê que a criação de liquidez está associada negativamente à falência de bancos por dois motivos Primeiro a criação de liquidez é uma função primária e crucial das instituições bancárias Uma vez que a criação de liquidez como uma medida chave da produção bancária total contém informações sobre a capacidade de um banco de apoiar a macroeconomia e facilitar as transações entre os agentes econômicos por meio da transformação de maturidade a incapacidade do banco de gerenciar seu balanço e realizar uma criação de liquidez é provável que a função seja um sinal de alerta precoce de problemas FUNGACOVA PESSAROSSI WEILL 2013 No novo regime de Basileia III foi acordado pela primeira vez um quadro regulamentar de liquidez a nível internacional com a introdução de dois rácios de liquidez O 57 Índice de Cobertura de Liquidez e o Índice de Financiamento Estável Líquido NSFR buscam objetivos complementares que são promover a resiliência de curto prazo do perfil de liquidez dos bancos e manter um perfil de financiamento estável respectivamente Embora o último ainda não tenha entrado em vigor o primeiro já foi implementado progressivamente desde 2015 O ICL é projetado para garantir que os bancos resistam a um cenário de estresse de liquidez de 30 dias Mais recentemente o impacto da regulamentação de liquidez de Basileia III foi avaliado em termos de prevenção de risco de liquidez bem como seu impacto macroeconômico geral Teoricamente End e Kruidhof 2013 tentam simular as implicações sistêmicas do Liquidity Coverage Ratio No entanto empiricamente a maioria dos estudos enfoca proxies dos índices de liquidez regulatória como depósitos sobre empréstimos para Tabak e Tecles 2010 266 Sensibilidade ao risco de mercado sensitivity to market risk Por último localizase a sensibilidade ao risco de mercado sensitivity to market risk nesse item são avaliadas as alterações nas taxas de juros taxas de câmbio preços de commodities ou preços de ações Esses itens podem afetar diretamente os lucros eou o capital econômico bancário FEDERAL RESERVE SYSTEM 1996 Nesse modelo Dincer et al 2011 avaliam o nível de liquidez dos bancos sob a ótica da capacidade da instituição financeira em pagar seus ativos e de cobrir retiradas de dinheiro inesperadas Peydró Polo e Sette 2021 demonstram a importância dos títulos de valores mobiliários principalmente em tempos de crise com condições de política monetária mais amenas os bancos menos capitalizados reagem aumentando os títulos sobre a oferta de crédito com efeitos reais associados ao nível da empresa É importante ressaltar que os bancos menos capitalizados compram títulos com rendimento mais baixo mesmo dentro de títulos com pesos de risco de capital regulatório idênticos alcançando assim segurança e liquidez Os resultados sugerem que a liquidez e a capacidade de assumir riscos são os principais motores do comportamento dos bancos devido às condições de política monetária Ao contrário de tempos de crise em tempos de précrise normais quando os atritos financeiros são limitados à medida que as condições de política monetária se tornam mais suaves os bancos menos capitalizados não expandem os títulos em vez da oferta de crédito Além disso em tempos de 58 crise com condições monetárias mais amolecidas a substituição do crédito por títulos ajuda a restaurar a rentabilidade e o capital dos bancos menos capitalizados e esta melhoria no balanço ajuda a reiniciar a oferta de crédito com defasagem de um ano Affinito Albareto e Santioni 2016 exploram os determinantes do aumento dos títulos de dívida soberana nas carteiras dos bancos durante a crise geralmente adotou uma perspectiva macroeconômica Os autores adotam uma abordagem microeconômica e analisam os principais determinantes das compras banco a banco investigando as condições do balanço dos bancos italianos de 2007 a 2013 Os resultados mostram que as características específicas do balanço dos bancos são importantes e os bancos compram títulos do governo para sustentar suas condições financeiras A alta liquidez dos títulos do governo os altos rendimentos e a conveniência em termos de encargos de capital os tornam adequados para satisfazer as necessidades dos bancos em períodos de intensa demanda de liquidez declínio da lucratividade dos bancos e da qualidade dos empréstimos e aumento das restrições de capital 267 Variáveis CAMELS Os escores do rating CAMELS são avaliados separadamente onde escores mais altos representam desempenho mais fraco práticas inadequadas de gerenciamento de risco em relação ao tamanho complexidade e perfil de risco da instituição portanto maior grau de preocupação de supervisão SAHAJWALA VAN DEN BERGH 2000 Após a identificação dos riscos cabe à gestão conduzir e implementar mudanças no setor como a criação de produtos ou início de novas atividades Com a ampliação na gama de produtos financeiros oferecidos pelos bancos as instituições perceberam a necessidade de utilizar sistemas sólidos para o gerenciamento de riscos devido à aceleração no ritmo das transações que afetam o risco de mercado Dessa forma o sistema CAMELS revisado auxilia a administração a identificar mensurar monitorar e controlar os possíveis riscos financeiros PIMENTA 2014 Para avaliar as práticas de gerenciamento de riscos devese levar em consideração o tamanho a complexidade e o perfil de risco de uma instituição financeira Por isso é esperado que bancos menos complexos não utilizem sistemas elaborados e altamente formalizados de gerenciamento de riscos para receber classificações fortes ou satisfatórias de componentes ou compostos pois é esperado que eles gerenciem seus riscos de maneira eficaz SAHAJWALA VAN DEN BERGH 2000 PIMENTA 2014 59 Conforme citado anteriormente a literatura nos traz diferentes cálculos para mensurar individualmente os seis componentes do CAMELS SAHAJWALA VAN DEN BERGH 2000 PIMENTA 2014 podendo variar de acordo com o ponto de vista do examinador eou do investigador dos órgãos reguladores como a Office of the Comptroller of the Currency OCC Gabinete de Controladoria da Moeda e o FSI Financial Soundness Indicators Indicadores de Solidez Financeira criado pelo IMF International Monetary Fund Fundo Monetário Internacional De acordo com International Monetary Fund 2006 os seis itens de avaliação do CAMELS possuem suas particularidades individuais para cada categoria O indicador para a Adequação de Capital C deve ser calculado por meio dos valores de empréstimos inadimplentes Nonperforming loans menos os valores das provisões específicas dividido pelo o capital O resultado permite identificar precocemente os possíveis ativos problemáticos INTERNATIONAL MONETARY FUND 2006 Para Qualidade dos ativos A o cálculo do indicador envolve o valor de inadimplência da carteira dividido pelo valor total da carteira de crédito que deve incluir a inadimplência antes da dedução das provisões de liquidações duvidosas específicas Dessa forma é possível prever problemas relacionados com a qualidade dos ativos da carteira de crédito INTERNATIONAL MONETARY FUND 2006 O indicador de gestãoadministração M mensura a qualidade gerencial da instituição financeira ou seja é esperado que bancos com maior eficiência em sua gestão estejam menos propensos a riscos financeiros Para avaliar essa variável devemos utilizar o resultado de intermediação financeira e somar com as receitas com serviços esse valor deve ser dividido pela soma das despesas pessoais com as despesas administrativas INTERNATIONAL MONETARY FUND 2006 Os lucros E devem ser calculados por meio da soma entre o maior retorno de ativos ROA com os resultados de intermediação financeira e divididos pela receita bruta da intermediação financeira Os lucros indicam um bom desempenho financeiro para a instituição que consegue dessa forma investir na expansão se mantenha competitiva no mercado e aumente seu capital INTERNATIONAL MONETARY FUND 2006 O indicar para a liquidez L avalia a capacidade da gestão em balancear a lucratividade com os riscos de perda ou seja para calcular a liquidez é necessário dividir o total de ativos líquidos pelo total de ativos INTERNATIONAL MONETARY FUND 2006 Para a sensibilidade ao risco de mercado S último indicador a ser calculado há que se considerar a posição aberta de câmbio líquida dividida pelo capital O resultado obtido 60 permite com que o avaliador análise o poder da gestão em controlar o nível de exposição da instituição frente ao risco das taxas de juros flutuação do preço patrimonial e da volatilidade cambial INTERNATIONAL MONETARY FUND 2006 Diante da literatura nacional e internacional e a fim de embasar a escolha das fórmulas para cada uma das letras do acrônimo CAMELS utilizadas neste estudo e apresentadas na seção do método deste trabalho e ampliar o que foi posto anteriormente tendo apenas o FMI como base os Quadros 4 e 5 com os autores seus trabalhos e as fórmulas utilizadas por cada um deles apresenta o resultado da revisão bibliográfica deste tema e elucida a quantidade de índices que podem ser utilizados para cada novo autor A realização desta revisão sustenta a escolha das fórmulas ou métodos de mensuração que são utilizados neste trabalho e apresenta contribuição para a literatura e para futuros estudos 61 Quadro 4 Literatura sobre variáveis representativas do acrônimo CAMELS C A M C Capital A Asset M Management Literatura Empréstimos Vencidos NPLs Provisão para Créditos de Liquidação Duvidosa Capital Empréstimos Vencidos NPLs Crédito Total Resultado de Intermediação Financeira Receitas com Serviços Despesas Pessoais Despesas Administrativas IMF 2006 Empréstimos Vencidos NPLs Provisão para Créditos de Liquidação Duvidosa Capital Empréstimos Vencidos NPLs Crédito Total Resultado de Intermediação Financeira Receitas com Serviços Despesas Pessoais Despesas Administrativas PIMENTA 2014 Capital Próprio Ativo Total Empréstimos Vencidos NPLs Crédito Total Despesas Operacionais Total de Receitas GOMES 2012 Ativo Total Capital Próprio ROA Lucro Líquido Ativo Total x 100 Capital Tier 12 e3 Ativos Ponderados pelo Risco Provisões para Empréstimos Vencidos Empréstimos Vencidos NPLs ROE Lucro Líquido Patrimônio Líquido x 100 Patrimônio Líquido Títulos e Valores Mobiliários e Instrumentos Financeiros Derivativos Relações Interfinanceiras Relações Interdependências Operações de Crédito Operações de Arrendamento Mercantil Outros Créditos Outros Valores e Bens Provisões para Operações de Crédito Total da Carteira Despesas de CaptaçãoCirculante e Exigível de Longo Prazo ROSA GARTNER 2018 Créditos vencidos total das operações de crédito X KORONTAI 2016 62 C Capital A Asset M Management Literatura Capital total regulamentar para cobertura dos ativos ponderados pelo risco Créditos vencidos líquidos de provisão capital Capital nível 1 para cobertura de ativos ponderados pelo risco Distribuição Setorial dos créditos total das operações de crédito X X X GODOI et al 2016 Operações de Crédito e Arrendamento Mercantil Outros Créditos Patrimônio Líquido Imobilizado de Arrendamento Permanente Ajustado x 100 Patrimônio Líquido X MARCUSSO 2017 Capital de Terceiros Patrimônio Líquido Operações de Crédito e Arrendamento Mercantil Outros Créditos Ativo Total Relações Interdependência Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos Patrimônio Líquido Tier I Tier II Ativos Ponderados pelo Risco Total de Empréstimos Vencidos Provisões Empréstimos Total Despesas Administrativas Vendas CHRISTOPOULOS MYLONAKIS DIKTAPANIDIS 2011 Patrimônio Líquido Ativo Total Empréstimos em Inadimplência Empréstimos Bruto Despesas Operacionais Ativo Total ROMAN SARGU 2013 63 C Capital A Asset M Management Literatura Loan Loss Provisions Net Interest Revenues Despesas com Juros Depósitos Total loans total asset Patrimônio líquido Ativos Sujeito a Risco de Crédito Risco de Mercado Risco Operacional Ativos Financeiros Líquidos Ativo Total Despesa de Juros Despesa Total DINCER et al 2011 Patrimônio Líquido Ativo Total Empréstimo Total e Recebíveis Ativo Total Receita de Juros Receita Total Patrimônio Líquido Depósitos Fundos Não Depósitos Ativo Permanente Ativo Total Renda Total Despesa Total Patrimônio Líquido Ativo Total Empréstimo Não RecebidoAtivo Total Despesa com PessoalDespesa Operacional Total MAGRO MICHELS SILVA 2017 Empréstimo Não Recebido Crédito Bruto Despesas GeraisAtivo Total Patrimônio Líquido Empréstimo Total Credito LíquidoAtivo Total Receita totalAtivo Total Provisão de PerdasAtivo Total Total de Despesas OperacionaisReceita Total Crédito TotalPatrimônio Líquido Total Equity Total Assets Nonperforming loanstotal assets X AFFES HENTATI KAFFEL 2017 Total Equity Total Loans Nonperforming loansgross loans 64 C Capital A Asset M Management Literatura Loan loss reservestotal assets Loan loss reservesgross loans Patrimônio Líquido Ativo Total Empréstimos Vencidos NPLs Empréstimo Total Despesa não Financeira Receita Financeira Líquida Receita Financeira AVKIRAN LIN 2012 Empréstimos Vencidos NPLs Ativo Total Despesa com PessoalMédia dos Ativos Provisões para Empréstimos Ativo Total Custos Receita Provisões para Empréstimos Empréstimo Total Fonte Elaborado pelo autor 65 Quadro 5 Literatura sobre variáveis representativas do acrônimo CAMELS E L S E Earning L Liquidity S Sensitivity to Market Risk Literatura ROA Lucro Líquido Ativo Total x 100 Total de Ativos Líquidos Total de Ativos Posição Aberta de Câmbio Líquida Capital IMF 2006 Resultado de Intermediação Financeira Receita Bruta de Intermediação Financeira ROA Lucro Líquido Ativo Total x 100 Total de Ativos Líquidos Total de Ativos Posição Aberta de Câmbio Líquida Capital PIMENTA 2014 Resultado de Intermediação Financeira Receita Bruta de Intermediação Financeira X Crédito a Clientes Líquido Depósitos de Clientes Value at Risk VaR GOMES 2012 Disponibilidades Crédito a Instituições Financeiras Total de Ativos Receitas Operacionais Despesas Operacionais Circulante e Realizável de Longo Prazo Permanente Disponibilidades Aplicações Interfinanceiras de Liquidez Índice do Segmento Financeiro de Valores Mobiliários Livres Depósitos Obrigações por Operações Compromissadas X ROSA GARTNER 2018 ROA Lucro Líquido Ativo Total x 100 Ativos Líquidos Ativo Total Posição Líquida em Moeda Estrangeira Capital KORONTAI 2016 ROE Lucro Líquido Patrimônio Líquido x 100 Margem Financeira Resultado de Intermediação Receitas Ativos Líquidos Dívidas de Curto Prazo Despesas Administrativas Receitas ROE Lucro Líquido Patrimônio Líquido x 100 X Passivo Circulante Passivo Exigível à Longo Prazo Patrimônio Líquido GODOI et al 2016 66 E Earning L Liquidity S Sensitivity to Market Risk Literatura Lucro Líquido Receita Líquida da Intermediação Financeira Dívidas com Vencimento a Longo Prazo superiores há 1 ano Patrimônio Líquido Número de ações X Preço das ações Preço de Mercado da Ação Lucro Contábil por Ação Despesas de Intermediação Financeira de Captações no Mercado Despesas Empréstimos e Repasses Despesas Arrendamento Mercantil Despesas de Operações de Vendas ou Transferências de Ativos Financeiros Previsão para CL Ativo Total Ativo Permanente Relações Interdependências Imobilizado de Arrendamento Disponibilidades Aplicações Interfinanceiras Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos Outras Obrigações e Exercícios Futuros Obrigações por Empréstimos e Repasses X MARCUSSO 2017 Atividade Bancária x 100 Ativo Total Ativo Permanente Relações Interdependência Imobilizado de Arrendamento Aplicações Financeiras Líquidas Depósitos Interfinanceiros Depósitos a Prazo ROA Lucro Líquido x 100 Ativo Total Depósitos à Vista Depósitos de Poupança Passivo Exigível x100 ROE Lucro Líquido x 100 Patrimônio Líquido Depósito Total Depósito Total Recursos de Aceites Total Obrigações por Empréstimos e Repasses Patrimônio Líquido Receita Operações Arrendamento Mercantil Receitas Operações Vendas ou Transferência de Ativos Financeiros Receita de Intermediação financeira Operação de Crédito e Arrendamento Mercantil Receitas Operações com TVM Receitas Operações com Instrumentos Financeiros Derivativos Receitas Operações de Câmbio Receitas Aplicações Compulsórias Outras Receitas ou Despesas Operacionais Receitas de Prestação de Serviços Outras Receitas ou Despesas Operacionais de Renda e Rendas de Tarifas Bancárias Ativo Total Relações de Interdependência Permanente 67 E Earning L Liquidity S Sensitivity to Market Risk Literatura Imobilizado de Arrendamento Despesas Intermediação Financeira e Captações no Mercado Despesas Empréstimos e Repasses Despesas Arrendamento Mercantil Despesas Operações Câmbio Provisão para CL Despesas Operações de Vendas ou Transferências de Ativos Financeiros Ativo Total Permanente Imobilizado de Arrendamento Relações Interdependências x100 ROA Lucro Líquido Ativo Total x 100 Empréstimos Total Depósitos Total Total Títulos Valores Mobiliários Ativo Total CHRISTOPOULOS MYLONAKIS DIKTAPANIDIS 2011 ROE Lucro Líquido Patrimônio Líquido x 100 Cativo Circulante Ativo Total ROA Lucro Líquido Ativo Total x 100 Ativo Total Depósitos Líquidos e Financiamento de Curto Prazo Ativo Total Ativo Total do Setor ROMAN SARGU 2013 ROE Lucro Líquido Patrimônio Líquido x 100 Empréstimos Depósitos Líquidos e Financiamento de Curto Prazo Lucro Líquido Ativo Total Ativo LíquidoAtivo Total Ativo Total Ativo do Setor DINCER et al 2011 Lucro Líquido Patrimônio Líquido Ativo Líquido Passivo Curto Prazo Total de Empréstimos e Recebíveis Setor Empréstimos e Recebíveis Ativo Líquido Depósito Total Fundos Não Depósito Depósito Total Depósitos Setoriais ROE Sobra LíquidaPatrimônio Líquido Ativo LíquidoAtivo Total Valores Mobiliários em CarteiraAtivo Total MAGRO MICHELS SILVA 2017 ROA Sobra LíquidaAtivo Total Deposito TotalAtivo Total Receita TotalReceita Operacional Depósito TotalPatrimônio Líquido Valores Mobiliários em CarteiraEmpréstimos Total Provisão para Credito de Difícil LiquidaçãoAtivo Total Crédito TotalDepósito total 68 E Earning L Liquidity S Sensitivity to Market Risk Literatura Financiamento a Curto PrazoPassivo Total ROA Net Incometotal assets Total loanstotal customer deposits X AFFES HENTATI KAFFEL 2017 ROE Net IncomeTotal Equity Total customer depositstotal assets Lucro Líquido Ativo Total Compras Líquidas de Fundos Federais Ativo total Títulos Negociados Ativo Total AVKIRAN LIN 2012 Ativos Estrangeiros Passivos Estrangeiros Ativos Líquidos Ativo Total Receita Financeira Líquida Ativo Médio Fonte Elaborado pelo autor 69 Baseado no conceito de eficiência e sua aplicação no setor bancário bem como no embasamento das técnicas de mensuração de eficiência sendo uma paramétrica Fronteira Estocástica e outra não paramétrica Análise Envoltória de Dados além do estudo do acrônimo CAMELS para a mensuração de riscos este estudo foi estruturado de modo a analisar o impacto de cada um destes itens no setor bancário brasileiro ao longo dos anos 2000 a 2018 trazendo novo conhecimento a respeito das técnicas aplicadas em conjunto e do sistema de early warnings CAMELS aplicado na identificação de impactos nas variáveis dependentes deste estudo 70 3 PROCEDIMENTOS METODOLÓGICOS Nesta seção serão apresentados a caracterização da pesquisa e os procedimentos metodológicos com as etapas necessárias para se atingir os objetivos propostos 31 CLASSIFICAÇÃO DA PESQUISA Este estudo utilizou a abordagem metodológica quantitativa com o intuito de empregar métodos estatísticos e não estatísticos para verificar o relacionamento entre as variáveis analisadas após o levantamento da base de dados utilizada A abordagem quantitativa procura quantificar os dados e busca encontrar uma evidência conclusiva É baseada em amostras grandes e representativas e realiza a aplicação e análise estatística Os resultados obtidos de uma pesquisa qualitativa podem ser tratados como conclusivos e utilizados para recomendar um curso de ação final MALHOTRA et al 2005 A pesquisa quantitativa se centra na objetividade Influenciada pelo positivismo considera que a realidade só pode ser compreendida com base na análise de dados brutos recolhidos com o auxílio de instrumentos padronizados e neutros A pesquisa quantitativa recorre à linguagem matemática para descrever as causas de um fenômeno as relações entre variáveis etc FONSECA 2002 p 36 Uma pesquisa além da definição de sua abordagem metodológica também deve ser classificada quanto aos seus objetivos e procedimentos de investigação Este projeto sob a ótica de seus objetivos pode ser classificado como descritivo uma vez que procura especificar as propriedades as características e os perfis importantes de pessoas grupos comunidades ou qualquer outro fenômeno que se submeta a análise Eles medem avaliam ou coletam dados sobre diversos aspectos dimensões ou componentes do fenômeno a ser pesquisado Do ponto de vista científico descrever é coletar dados para pesquisadores quantitativos medir para os qualitativos coletar informações SAMPIERI COLLADO LUCIO 2006 p 101 Vergara 2006 considera que a pesquisa descritiva possui o objetivo de descrição das características de populações ou fenômenos e o estabelecimento de relações entre variáveis 71 Finalmente quanto aos procedimentos de investigação este projeto pode ser classificado como bibliográfico e documental pois as informações necessárias para sua elaboração são obtidas de fontes secundárias desde periódicos e artigos até a base de dados e relatórios divulgados em órgãos de regulação e controle Segundo Vergara 2006 a investigação documental é realizada em documentos mantidos no interior de órgãos públicos ou privados Para Gil 1999 a pesquisa bibliográfica utilizase principalmente das contribuições da comunidade científica sobre determinado tema de estudo já a pesquisa documental utiliza se de materiais que ainda não receberam um tratamento analítico ou que podem ser utilizados ou moldados de acordo com os objetivos da pesquisa A partir deste delineamento tornase possível atingir os objetivos propostos nesta dissertação de maneira a garantir que as normas de pesquisa sejam respeitadas 32 ETAPAS DA PESQUISA Para a realização da presente pesquisa foram utilizados os Balanços Patrimoniais e os Demonstrativos de Resultado do Exercício de cada banco consolidados e referentes aos anos de interesse desta pesquisa coletados no site do Banco Central do Brasil A base de dados disponível no site do BACEN iniciase em 2000 e possui dados até os fechamentos anteriores a esta pesquisa Esta base de dados de caráter público é amplamente utilizada no sistema financeiro e por pesquisadores Após o levantamento das informações financeiras através do Balanço Patrimonial e Demonstrativo de Resultado de cada banco foram selecionadas as variáveis analisadas em cada uma das três abordagens que orientam este estudo A metaanálise realizada por Macoris Salgado Junior e Falsarella Junior 2015 obteve como resultado as variáveis mais utilizadas pelas abordagens nos estudos envolvendo DEA e serviu de base para a escolha dos inputs e outputs a serem utilizados neste projeto de pesquisa tanto para a técnica DEA quanto para a técnica SFA conforme exposto na Tabela 1 72 Tabela 1 Inputs e Outputs de Macoris Salgado Junior e Falsarella Junior 2015 INTERMEDIAÇÃO RENTABILIDADE Inputs Depósitos Número de Pessoal Despesas com Pessoal Despesas Operacionais Despesas Operacionais não relacionadas a Pessoal Ativos Patrimônio Líquido Passivo Outputs Empréstimos Investimentos e Outros Ativos Realizáveis ROE ROI Lucro Fonte Adaptado Macoris Salgado Junior e Falsarella Junior 2015 O trabalho de Macoris Salgado Junior e Falsarella Junior 2015 foi adaptado à realidade atual das instituições financeiras a base de dados do Banco Central do Brasil e as necessidades de obtenção de escores que demonstrem de maneira mais eficaz a eficiência em cada uma das abordagens para as instituições financeiras nacionais O resultado desta adaptação e aprimoramento está demonstrado na Tabela 2 Tabela 2 Inputs e Outputs utilizados em DEA e SFA com múltiplos inputs e outputs INTERMEDIAÇÃO RENTABILIDADE Input Captações Ativo Total Input Despesas de Pessoal Patrimônio Líquido Output Aplicações Interfinanceiras Lucro Líquido Output Operações de Crédito e Arrendamento Mercantil Total Fonte Elaborado pelo autor A tabela 2 demonstra as variáveis utilizadas como inputs e outputs para a técnica DEA e SFA quando considerada a utilização de múltiplos inputs e múltiplos outputs e que foi desenvolvida em estudos anteriores a este que se apresenta A fim de se utilizar as duas técnicas em conjunto DEA e SFA este estudo realizou uma adaptação aos inputs e outputs apresentados na tabela anterior com o objetivo de atender também trabalhos que utilizem um único output face à necessidade do grupo de trabalho de rodar um modelo mais simples posto que se trata do primeiro estudo do grupo com a técnica SFA De fato está sendo apresentado modelo simplificado de análise de eficiência com um único output que considerou 73 i a não utilização de Lucro Líquido ii a utilização de Despesas Administrativas nos inputs da Abordagem de Intermediação em virtude da facilidade tecnológica iii a revisão dos inputs e outputs da Abordagem de Produção vez que tal indicador contempla praticamente todas as despesas fixas da instituição porém apenas uma parte da receita vez que as receitas não relacionadas a juros abrangem principalmente a receita com títulos e câmbio iv a revisão da Abordagem de Rentabilidade vez que Patrimônio Líquido contempla o patrimônio líquido efetivamente estampado no passivo o que resulta numa sobreposição de valores entre esses inputs por isso a utilização apenas de Ativo como input e Receita como output v que Receita de Serviços é muito similar a Receita com Juros vez que existem reciprocidades nas operações de crédito o que deve ser considerado como custo ou seja grande parte da Receita de Serviços advém diretamente dos juros O resultado de referidas adaptações é apresentado na Tabela 3 Tabela 3 Inputs e Outputs utilizados em DEA e SFA com um único output INTERMEDIAÇÃO RENTABILIDADE Input Captações Ativo Total Input Despesas Administrativas e Despesa de Pessoal Soma Despesas de Intermediação Financeira Despesas de Pessoal Despesas Administrativas Despesas Tributárias e Outras Despesas Operacionais Output Soma de Aplicações Interfinanceiras de Liquidez e Operação de Crédito e Arrendamento Mercantil Receitas Fonte Elaborada pelo autor Como primeiro estágio desta etapa os inputs e outputs acima foram aplicados aos bancos selecionados na categoria B1 A categoria B1 de acordo com a classificação de Tipo de Consolidado Bancário do Banco Central do Brasil compreende os bancos comerciais bancos múltiplos com carteira comercial ou caixas econômicas 74 Os bancos que apresentaram prejuízos no ano ou inputs e outputs sem informações foram excluídos chegandose dessa forma a amostra final analisada Os números negativos não foram considerados devido a uma restrição do modelo DEA Além disso os outliers que são unidades que apresentam desvios significativos em relação ao comportamento médio frente as demais DMUs ou unidades foram retirados da base de dados antes da geração da fronteira através dos softwares para as técnicas DEA e SFA Posteriormente os bancos foram divididos pelo porte pela estrutura de capital e pelo país de origem do controlador Para a classificação de porte que considerou os ativos totais de cada banco os bancos foram classificados em grande porte médio porte e pequeno porte Na sequência os bancos foram analisados frente aos escores obtidos por ambas as técnicas DEA e SFA para as três abordagens já apresentadas nesta seção As séries temporais contendo o Balanço Patrimonial e Demonstrativo de Resultado dos bancos analisados foram corrigidas pela inflação do período para representar fielmente as evoluções nos escores frente ao tempo Dessa forma tornouse possível a comparação efetiva dos bancos através da homogeneidade dos valores financeiros e consequentemente dos escores da técnica DEA e SFA Após a geração dos escores para ambas as técnicas gráficos contendo os escores para cada uma das abordagens bem como para cada uma das técnicas foram formulados para que movimentos relevantes fossem observados de forma visual Utilizouse para isto as medianas dos escores Após o primeiro estágio caracterizado pela aplicação das técnicas DEA e SFA e geração dos escores bem como a fronteira de eficiência a fim de buscar um maior entendimento e profundidade sobre os resultados obtidos aplicouse o segundo estágio baseado em um modelo de regressão linear múltipla Hair et al 2005 define a regressão linear múltipla como a técnica mais utilizada e mais versátil e que deve ser aplicada quando há apenas uma única variável dependente e duas ou mais variáveis independentes relacionadas a ela Esta dissertação utiliza a análise de regressão múltipla cujo objetivo é utilizar técnicas estatísticas para construir modelos que descrevem as relações entre várias variáveis explicativas de um determinado processo Proposto por Galton 1885 o conceito de regressão teve seu desenvolvimento através dos anos e por diversos autores Neter e Wasserman 1974 seguidos por Draper e Smith 1981 desenvolveram os estudos anteriores e auxiliaram na formulação que representa a regressão ao considerar novas aplicabilidades e conceitos Bowerman e OConnel 1990 75 também foram autores importantes neste desenvolvimento a partir do estudo realizado sobre modelos lineares de estatística A fórmula utilizada da equação de Regressão Linear Múltipla é comumente apresentada como 𝑌 𝛽0 𝛽1𝑋1 𝛽𝑝𝑋𝑝 𝑒 Onde Y Variável dependente X Variável independente 𝛽 coeficiente associado a variável e erro que apresenta distribuição normal com média zero e variância 𝜎2 A apresentação dos conceitos de Regressão Linear Múltipla se faz necessária neste estudo uma vez que será utilizada como o segundo estágio da DEA para verificar os efeitos das variáveis independentes em seus escores Para tanto foi realizado teste de heterocedasticidade para identificar os impactos das variáveis independentes nas variáveis dependentes Para o segundo estágio os escores obtidos anteriormente serão usados como variável dependente e as variáveis CAMELS utilizadas como as variáveis independentes conforme Figura 8 76 Figura 8 DEA e SFA dois estágios com regressão linear múltipla Fonte Elaborado pelo autor 77 O segundo estágio é utilizado para verificar quais variáveis podem influenciar significativamente nos escores de eficiência gerados no primeiro estágio Para o segundo estágio as variáveis CAMELS serão utilizadas como as variáveis independentes e os escores gerados pela técnica DEA e SFA serão para cada caso utilizados como as variáveis dependentes A escolha do acrônimo CAMELS já embasada pelo referencial teórico deste estudo após pesquisa na literatura nacional e internacional possibilitou a identificação de diversas fórmulas e índices financeiros Destas fórmulas para fins de adequação ao mercado brasileiro e a forma de publicação das demonstrações financeiras bem como a base utilizada IFBACEN as fórmulas que possuem contas tanto de balanço patrimonial quanto de demonstrativo de resultado representadas nas demonstrações financeiras das instituições financeiras brasileiras foram consideradas e possibilitaram a formulação do Quadro 6 Após avaliação das disponibilidades de informações utilizouse o seguinte critério utilização das equações apresentadas pelo Fundo Monetário Internacional presentes no banco de dados do Bacen e as que complementam a representação da definição de capital ativos gestão lucro liquidez e sensibilidade ao risco de mercado com outro conceito O Quadro 6 apresenta as equações a serem utilizadas por este estudo para os seis componentes do acrônimo CAMELS essas serão utilizados como variáveis independentes da regressão linear realizada no segundo estágio Quadro 6 Variáveis independentes advindas do CAMELS Inicial Equação Autores C Patrimônio Líquido Ativo Total Roman Sargu 2013 Dincer et al 2011 Magro Michels e Silva 2017 A Provisões para Operações de Crédito Total da Carteira Rosa e Gartner 2018 Ativo Circulante Ativo Total Dincer et al 2011 M Resultado de Intermediação Financeira Receitas com Serviços Despesas Pessoais Despesas Administrativas International Monetary Fund 2006 Pimenta 2014 E ROA Lucro Líquido Ativo Total x 100 International Monetary Fund 2006 Christopoulos Mylonakis e Diktapanidis 2011 Dincer et al 2011 Roman Sargu 2013 Pimenta 2014 Korontai 2016 Magro Michels Silva 2017 Marcusso 2017 ROE Lucro Líquido Patrimônio Líquido x 100 Korontai 2016 Christopoulos Mylonakis e Diktapanidis 2011 Dincer et al 2011 78 Roman Sargu 2013 Magro Michels Silva 2017 Marcusso 2017 L Ativo Circulante Ativo Total International Monetary Fund 2006 Christopoulos Mylonakis e Diktapanidis 2011 Dincer et al 2011 Pimenta 2014 Korontai 2016 Magro Michels Silva 2017 Total Créditos Total Depósitos Christopoulos Mylonakis e Diktapanidis 2011 Magro Michels Silva 2017 S Valores Mobiliários em Carteira Ativo Total Magro Michels Silva 2017 Valores Mobiliários em Carteira Empréstimos Total Fonte Elaborado pelo autor A análise de adequação de capital com o objetivo de mensurar a qualidade do capital da instituição frente ao seu cenário financeiro teve a fórmula selecionada que utiliza as contas de Patrimônio Líquido e Ativo Total com a finalidade de demonstrar quanto do Ativo da instituição financeira é financiado pelo seu Patrimônio Líquido assim sendo qual a dependência a instituição financeira possui frente aos recursos de terceiros para a continuidade de suas operações Para qualidade dos ativos a fim de verificar o risco da instituição de não recebimento de seus ativos duas fórmulas foram selecionadas e consideram as contas de Provisão sobre Operações de Crédito que considera a provisão para perdas com as operações de acordo com o nível de risco destas Operações de Crédito que contabiliza todas as operações sendo elas classificadas como normais em atraso ou em liquidação e Ativo Total a fim de mensurar quanto da provisão sobre estas operações impacta em sua carteira de crédito especificamente e de forma mais ampla em seu ativo total O componente de gestão que visa identificar a capacidade dos gestores em acompanhar e mitigar os riscos aos quais a instituição financeira está exposta teve a equação que considera as contas de Resultado de Intermediação Financeira Rendas com Prestação de Serviço Despesas de Pessoal e Despesas Administrativas selecionada com a finalidade de identificar o impacto das despesas com colaboradores e administrativas em geral na geração de resultados e rendas oriundas de sua atividade principal de intermediação de recursos e prestação de serviços Cabe destacar que a geração de resultados e rendas demonstra eficiência na mitigação de riscos e contenção de perdas sendo um bom indicativo de gestão de riscos Para o componente de lucro os índices financeiros de lucratividade ROA e ROE foram selecionados Estes dois índices possuem uma aplicabilidade comprovada cientificamente em diversos setores e organizações As instituições financeiras medem os retornos obtidos lucro 79 líquido sobre a sua carteira de ativos e sobre o seu patrimônio como forma de mensurar a sua eficiência Para liquidez foram consideradas as contas de Ativo Circulante Ativo Total Operações de Crédito e Depósito Total com o objetivo de quantificar o nível de liquidez da instituição financeira e a sua capacidade de cumprir suas obrigações financeiras e de gestão de ativos de terceiros e clientes 80 4 DESCRIÇÃO DOS RESULTADOS Este capítulo apresenta os resultados encontrados através da utilização das técnicas estatísticas DEA e SFA diante de uma mesma base de dados Alinhado ao objetivo deste trabalho as abordagens de intermediação e rentabilidade foram analisadas através dos gráficos gerados após a obtenção dos escores de eficiência Para que as análises e comparações entre as duas técnicas pudessem ocorrer considerouse a mesma base de dados contendo os mesmos bancos e períodos Todos os gráficos apresentam dados pontuais porém em cada uma das apresentações as linhas são contínuas apenas para melhor visualização dos pontos e de suas tendências Os gráficos contendo as medianas dos escores das técnicas DEA e SFA para as instituições financeiras brasileiras para as abordagens de intermediação e rentabilidade estão representados abaixo O Gráfico 1 apresenta a mediana dos escores obtidos pelas técnicas DEA e SFA para a abordagem de intermediação onde os inputs e o output utilizados estão representados na Tabela 3 81 Gráfico 1 Abordagem Intermediação Fonte Elaborado pelo autor 82 Percebese que a mediana dos escores obtidos a partir da técnica SFA apresentou uma menor volatilidade em comparação à técnica DEA Da mesma forma seus valores absolutos são maiores e concentrados na faixa de escores de 07 enquanto os resultados DEA são menores e encontramse distribuídos entre as faixas de 03 e 05 Mesmo com a menor volatilidade analisada no gráfico da técnica de SFA é possível identificar que ambas as técnicas apresentam momentos de elevações e diminuições dos índices porém não se pode afirmar que há uma relação entre estes movimentos das duas técnicas ou seja existem momentos em que os movimentos são opostos ou momentos em que apenas a técnica DEA demonstra variações O Gráfico 2 representa a mediana dos escores obtidos pelas técnicas DEA e SFA para a abordagem de rentabilidade onde os inputs e o output utilizados estão representados na Tabela 3 83 Gráfico 2 Abordagem Rentabilidade Fonte Elaborado pelo autor 84 É possível identificar que assim como a abordagem anterior já analisada a abordagem de rentabilidade apresenta os menores escores para a técnica DEA enquanto a técnica SFA apresenta a menor variabilidade no decorrer do período analisado Foi possível encontrar padrões de altas e baixas com o mesmo movimento nos anos 2002 2004 2005 2008 2011 e 2013 Observase neste trabalho ao analisarse os Gráficos 1 e 2 que as técnicas DEA e SFA apresentam compatibilidade contudo é importante salientar que as duas técnicas possuem vieses e a combinação das técnicas permite uma melhor reflexão sobre o desempenho das variáveis conforme será apresentado na seção discussão dos resultados onde ao considerar bases de dados diferentes e países diferentes os resultados são diversos devido aos modelos matemáticos que suportam cada um dos estudos analisados Ao analisarse as abordagens de intermediação e rentabilidade é possível identificar que a técnica SFA foi a que acusou menor volatilidade representada nos gráficos 1 e 2 A abordagem de intermediação apresentou maior oscilação comparada com a abordagem de rentabilidade Para ambas as técnicas e ambas as abordagens os gráficos gerados através dos escores obtidos pela técnica SFA foram os que apresentaram os maiores valores para os escores Nesta visão destacase a abordagem de rentabilidade sendo bastante superior e consistente quando comparada a abordagem de intermediação Os gráficos gerados através da técnica DEA para ambas as abordagens apresentaram maior volatilidade mas ao mesmo tempo mantiveramse sempre nos menores patamares e muito próximos aos escores centrais de eficiência ou seja próximos a faixa de 05 É importante salientar que ambas as técnicas possuem modelos matemáticos diferentes e por este motivo são esperados resultados diferentes entre as técnicas e sua aplicação para as abordagens deste estudo Com o objetivo de embasar ainda mais esta análise na Tabela 4 encontramse apresentadas as estatísticas descritivas em relação as duas técnicas e as abordagens deste estudo Assim como exposto no referencial teórico de acordo com Berger e Mester 1997 comprovouse o cenário de maior volatilidade e médias menores obtidas pela técnica de Análise Envoltória de Dados frente a técnica de Análise de Fronteira Estocástica 85 Tabela 4 Estatísticas descritivas dos escores de DEA e SFA DEA SFA Intermediação Rentabilidade Intermediação Rentabilidade Máximo 10000 10000 09799 09694 Mínimo 00007 00263 01065 00346 Média 04302 05273 08434 07295 Mediana 04186 05064 08546 07465 Desvio Padrão 02287 01698 00706 01019 Variância 00523 00288 00050 00104 Fonte Elaborada pelo autor Através das estatísticas descritivas é possível confirmar que a técnica SFA é menos sensível aos outliers e por este motivo apresenta valores absolutos maiores para as médias quando comparada aos valores gerados para as médias da técnica DEA O mesmo ocorre para a análise da mediana Seguindo o mesmo racional da análise de média e mediana podese verificar que os desviopadrão para a técnica SFA são menores bem como os valores de variância confirmando a menor sensibilidade desta em relação aos outliers 41 DEA Segundo Estágio resultados da regressão linear múltipla A seguir são transcritos os resultados das análises realizadas em software IBM SPSS com destaque para as variáveis que deram significantes para 411 DEA Segundo Estágio para abordagem Intermediação As variáveis significantes para Intermediação estão presentadas nas Tabelas 5 e 6 86 Tabela 5 Modelo para Intermediação model summary R Square Adjusted R Square Std Error of the Estimate DurbinWatson 461 459 157362111 466 Fonte elaborada pelo autor Tabela 6 Variáveis significantes para Intermediação coefficients a Constantes Beta Sig Tolerance VIF Patrimônio Líquido Ativo Total 570 000 767 1303 Provisão sobre Operações de Crédito Operações de Crédito 099 000 872 1147 Operações de Crédito Ativo Total 413 000 387 2583 Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas 253 000 631 1585 ROE Lucro Líquido Patrimônio Líquido x 100 042 027 916 1092 Ativo Circulante Ativo Total 393 000 441 2266 TVM e instrumentos Financeiros Derivativos Ativo Total 094 001 395 2530 TVM e Instrumentos Financeiros Derivativos Operações de Crédito 086 000 670 1492 Fonte elaborada pelo autor Onde Dependent Variable DEAIntermPredictors Constant Patrimônio Líquido Ativo Total Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas TVM e instrumentos Financeiros Derivativos Ativo Total Ativo Circulante Ativo Total Operações de Crédito Ativo Total Provisão sobre Operações de Crédito Operações de Crédito TVM e Instrumentos Financeiros Derivativos Operações de Crédito ROE Lucro Líquido Patrimônio Líquido x 100 412 DEA Segundo Estágio para abordagem Rentabilidade As variáveis que deram significativas para Rentabilidade na análise via SPSS estão especificadas nas Tabelas 7 e 8 87 Tabela 7 Modelo para Rentabilidade model summary R Square Adjusted R Square Std Error of the Estimate DurbinWatson 286 285 136337268 755 Fonte elaborada pelo autor Onde Predictors Constant ROE Lucro Líquido Patrimônio Líquido x 100 Patrimônio Líquido Ativo Total Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas Provisão sobre Operações de Crédito Operações de Crédito ROA Lucro Líquido Ativo Total x 100 TVM e Instrumentos Financeiros Derivativos Operações de Crédito Tabela 8 Variáveis significantes para Rentabilidade coefficients a Constantes Beta Sig Tolerance VIF Patrimônio Líquido Ativo Total 327 000 879 1137 Provisão sobre Operações de Crédito Operações de Crédito 156 000 904 1107 Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas 201 000 691 1448 ROA Lucro Líquido Ativo Total x 100 155 000 342 2927 ROE Lucro Líquido Patrimônio Líquido x 100 275 000 339 2947 TVM e Instrumentos Financeiros Derivativos Operações de Crédito 065 000 720 1388 Fonte elaborada pelo autor Onde Dependent Variable DEARent 42 SFA e os fatores explicativos para eficiência bancária O modelo SFA CobbDouglas halfnormal com heterocedasticidade que separa ineficiência persistente e ineficiência variável no tempo KUMBHAKAR LIEN HARDAKER 2012 é apresentado a seguir 88 No primeiro passo o Teste de Especificação de Hausman para Intermediação e Rentabilidade rejeitou a hipótese nula de que o modelo de painel com efeitos aleatórios seja mais apropriado considerandose desta forma efeitos fixos confirmados por Prob chi2 00000 Também no primeiro passo foi executada a regressão para efeitos fixos resultando nos dados estampados nas Tabelas 9 e 10 para Intermediação e Rentabilidade respectivamente Tabela 9 Regressão para efeitos fixos de Intermediação Variável Coef Std Err Pt Patrimônio Líquido Ativo Total 112686 0173637 0000 Provisão sobre Operações de Crédito Operações de Crédito 0315239 020155 0118 Operações de Crédito Ativo Total 3880897 0172254 0000 Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas 0013102 0004834 0007 ROA Lucro Líquido Ativo Total x 100 0021496 0008114 0008 ROE Lucro Líquido Patrimônio Líquido x 100 0003349 0001423 0019 Ativo Circulante Ativo Total 406171 0174784 0000 Operações de CréditoDepósito Total 288e06 209e06 0168 TVM e instrumentos Financeiros DerivativosAtivo Total 2275202 0163384 0000 TVM e Instrumentos Financeiros DerivativosOperações de Crédito 615e06 226e06 0007 Fonte elaborada pelo autor O teste retornou u 11474461 e 07553889 P 69764795 89 Tabela 10 Regressão para efeitos fixos de Rentabilidade Variável Coef Std Err Pt Patrimônio Líquido Ativo Total 0014794 013121 0910 Provisão sobre Operações de Crédito Operações de Crédito 047714 0152768 0002 Operações de Crédito Ativo Total 1222386 013028 0000 Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas 0045969 0003675 0000 ROA Lucro Líquido Ativo Total x 100 0113539 0006581 0000 ROE Lucro Líquido Patrimônio Líquido x 100 0007362 0001194 0000 Ativo Circulante Ativo Total 0641651 013215 0000 Operações de CréditoDepósito Total 220e06 158e06 0164 TVM e instrumentos Financeiros DerivativosAtivo Total 0665468 0123492 0000 TVM e Instrumentos Financeiros DerivativosOperações de Crédito 628e06 171e06 0000 Constante 7555242 0088925 0000 Fonte elaborada pelo autor O teste retornou u 05054876 e 05692987 P 44083762 43 Análise consolidada DEA e SFA nas abordagens Intermediação e Rentabilidade Os resultados deste estudo trazem os efeitos das variáveis independentes CAMELS na variável dependente escores DEA e escores SFA encontrados através da regressão linear 90 múltipla aplicada Essas avaliações são realizadas em duas principais abordagens utilizadas para o estudo da eficiência bancária intermediação e rentabilidade Ao considerar a base de dados de bancos brasileiros no período que contempla os anos de 2000 a 2018 podemos identificar as variáveis que apresentaram resultados estatisticamente significantes e seus impactos Desta forma para este estudo entre as variáveis estudadas observaramse cinco situações i variáveis estatisticamente relevantes e com impacto positivo sobre a eficiência bancária ii variáveis estatisticamente relevantes e com impacto negativo sobre a eficiência bancária iii variáveis não estatisticamente relevantes e com impacto positivo sobre a eficiência bancária iv variáveis não estatisticamente relevantes e com impacto negativo sobre a eficiência bancária e por último v variáveis sem relação com a eficiência bancária Alinhado ao objetivo deste estudo que busca identificar o impacto de variáveis CAMELS na eficiência bancária considerouse para análise apenas as variáveis estatisticamente relevantes e com impactos positivos ou negativos sobre a eficiência No contexto brasileiro abrangido pelo período deste estudo podese afirmar que as variáveis que apresentam impactos positivos podem ser interpretadas como quanto maior o valor da variável analisada maior a eficiência DEA e SFA para a abordagem analisada Da mesma forma ao analisar as variáveis que apresentam impacto negativo a interpretação é quanto maior o valor da variável menor a eficiência DEA e SFA para a abordagem analisada Entendese que a variável Patrimônio Líquido sobre Ativo Total demonstra quanto de recursos próprios é utilizado no financiamento e manutenção de sua atividade Assim sendo ao considerar o setor bancário tanto sob a ótica da abordagem de intermediação quanto sob a ótica da abordagem de rentabilidade entendese que quanto menor o resultado deste índice melhor podemos considerar seu impacto para as instituições bancárias O índice representado pela razão entre Provisões sobre Operações de Crédito sobre Operações de Crédito e o índice representado por Operações de Crédito sobre Ativo Total buscam mensurar o risco de a instituição bancária não receber seus ativos Esperase para o primeiro índice abordado que quanto menor este índice melhor a eficiência da instituição 91 financeira para ambas as abordagens aplicadas nesta dissertação Já para o segundo índice abordado esperase para a abordagem de intermediação que quanto maior melhor a eficiência bancária Por outro lado sob a ótica de intermediação esperase que quanto maior o índice melhor a sua eficiência e esperase o contrário sob a ótica de rentabilidade ou seja quanto menor este índice melhor a eficiência da instituição analisada Para estudar o impacto do indicador de gestão na eficiência bancária este estudo utilizouse do índice da soma do Resultado de Intermediação Financeira com Renda de Prestação de Serviços sobre a soma de Despesas com Pessoal com Despesas Administrativas Através deste índice mensurase o impacto das despesas com colaboradores e administrativas na geração de renda na atividade bancária Para isso ao considerarse a abordagem de intermediação quanto maior o valor encontrado neste índice melhor a eficiência bancária da instituição Já para a abordagem de rentabilidade podese considerar que não há uma relação conclusiva em relação a este indicador e a variável independente de gestão do acrônimo CAMELS utilizada neste estudo Um dos índices mais utilizados para justificar a eficiência bancária encontrase nos indicadores de rentabilidade Dentre eles o presente estudo utilizou os mais encontrados na literatura nacional e internacional o ROA e o ROE A utilização dos retornos obtidos por instituições financeiras sobre sua carteira de ativos e patrimônio são fundamentais para identificar a eficiência bancária Esperase portanto tanto para ROA quanto para ROE que quanto maiores os índices encontrados melhor para ambas as abordagens Para mensurar o impacto da liquidez na eficiência bancária utilizouse a razão encontrada entre o Ativo Circulante sobre o Ativo Total sendo utilizado para quantificar o nível de liquidez da instituição bancária ou seja a sua capacidade de pagamento de suas obrigações junto a terceiros e clientes Frente ao exposto considerase que quanto maior este índice mais eficiente a instituição financeira em ambas as abordagens A sensibilidade ao risco de mercado foi mensurada através de duas variáveis a variável Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos sobre o Ativo Total e a variável TVM e Instrumentos Financeiros Derivativos sobre Operações de Crédito Essas variáveis identificam a exposição da instituição financeira ao mercado de crédito e o risco de mercado de suas operações mantidas pelas instituições financeiras Para a abordagem de intermediação esperase para ambas as variáveis analisadas que quanto menor melhor a sua eficiência Em contrapartida para a abordagem de rentabilidade esperase que quanto maior os índices obtidos melhor a eficiência bancária das instituições devido ao maior retorno que estas operações podem desenvolver para compensar o risco 92 Com o objetivo de entender os impactos gerados pelas variáveis independentes sobre a eficiência bancária brasileira dos anos de 2000 a 2018 sob a ótica das abordagens de produção e rentabilidade este estudo realizou a análise das regressões retro item 42 segundo estágio DEA e heterocedasticidade SFA para identificar se o impacto das variáveis independentes sobre a eficiência bancária ocorreu neste estudo de acordo com a teoria financeira apresentada ou de forma contrária Além disso esta análise possibilita entender o comportamento de ambas as técnicas DEA e SFA bem como identificar semelhanças e diferenças entre elas A fim de atingir este objetivo as variáveis são analisadas uma a uma para cada uma das técnicas e comparadas com os achados sobre a teoria financeira 431 Pressupostos dos indicadores CAMELS para as Abordagens de Intermediação e Rentabilidade A abordagem de intermediação trata sobre a capacidade que a instituição financeira possui em intermediar a transação de recursos financeiros entre agentes superavitários e agentes deficitários A eficiência bancária sob a ótica de intermediação representa a quantidade de recursos que a instituição financeira consegue intermediar de maneira a receber o maior retorno sobre o maior volume de transações A abordagem de rentabilidade mensura a capacidade da instituição financeira de gerar receita orientada ao lucro Dessa forma analisase a eficiência sob a ótica de rentabilidade de uma instituição financeira pela sua capacidade de minimizar os custos resultantes da geração de receitas Buscase portanto lucratividade Diante destes fatos ao analisarmos as afirmações da teoria financeira podemos considerar sob a ótica da abordagem de intermediação e rentabilidade frente as variáveis independentes CAMELS que apresentaram impactos significativos nas variáveis dependentes tanto positivos quanto negativos o que se segue Entendese que a variável Patrimônio Líquido sobre Ativo Total demonstra quanto de recursos próprios é utilizado no financiamento e manutenção de sua atividade Assim sendo ao considerar o setor bancário tanto sob a ótica da abordagem de intermediação quanto sob a ótica da abordagem de rentabilidade entendese que quanto menor o resultado deste índice melhor podemos considerar seu impacto para as instituições bancárias O índice representado pela razão entre Provisões sobre Operações de Crédito sobre Operações de Crédito e o índice representado por Operações de Crédito sobre Ativo Total 93 buscam mensurar o risco de a instituição bancária não receber seus ativos Esperase para o primeiro índice abordado que quanto menor este índice melhor a eficiência da instituição financeira para ambas as abordagens aplicadas nesta dissertação Já para o segundo índice abordado esperase para a abordagem de intermediação que quanto maior melhor a eficiência bancária Por outro lado sob a ótica de intermediação esperase que quanto maior o índice melhor a sua eficiência e esperase o contrário sob a ótica de rentabilidade ou seja quanto menor este índice melhor a eficiência da instituição analisada Para estudar o impacto do indicador de gestão na eficiência bancária este estudo utilizouse do índice da soma do Resultado de Intermediação Financeira com Renda de Prestação de Serviços sobre a soma de Despesas com Pessoal com Despesas Administrativas Através deste índice mensurase o impacto das despesas com colaboradores e administrativas na geração de renda na atividade bancária Para isso ao considerarse a abordagem de intermediação quanto maior o valor encontrado neste índice melhor a eficiência bancária da instituição Já para a abordagem de rentabilidade podese considerar que não há uma relação conclusiva em relação a este indicador e a variável independente de gestão do acrônimo CAMELS utilizada neste estudo Um dos índices mais utilizados para justificar a eficiência bancária encontrase nos indicadores de rentabilidade Dentre eles o presente estudo utilizou os mais encontrados na literatura nacional e internacional o ROA e o ROE A utilização dos retornos obtidos por instituições financeiras sobre sua carteira de ativos e patrimônio são fundamentais para identificar a eficiência bancária Esperase portanto tanto para ROA quanto para ROE que quanto maiores os índices encontrados melhor para ambas as abordagens Para mensurar o impacto da liquidez na eficiência bancária utilizouse a razão encontrada entre o Ativo Circulante sobre o Ativo Total sendo utilizado para quantificar o nível de liquidez da instituição bancária ou seja a sua capacidade de pagamento de suas obrigações junto a terceiros e clientes Frente ao exposto considerase que quanto maior este índice mais eficiente a instituição financeira em ambas as abordagens A sensibilidade ao risco de mercado foi mensurada através de duas variáveis a variável Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos sobre o Ativo Total e a variável TVM e Instrumentos Financeiros Derivativos sobre Operações de Crédito Essas variáveis identificam a exposição da instituição financeira ao mercado de crédito e o risco de mercado de suas operações mantidas pelas instituições financeiras Para a abordagem de intermediação esperase para ambas as variáveis analisadas que quanto menor melhor a sua eficiência Em contrapartida para a abordagem de rentabilidade esperase que quanto maior os 94 índices obtidos melhor a eficiência bancária das instituições devido ao maior retorno que estas operações podem desenvolver para compensar o risco Com o objetivo de entender os impactos gerados pelas variáveis independentes sobre a eficiência bancária brasileira dos anos de 2000 a 2018 sob a ótica das abordagens de produção e rentabilidade este estudo realizou a análise das regressões segundo estágio DEA e heterocedasticidade SFA para identificar se o impacto das variáveis independentes sobre a eficiência bancária ocorreu neste estudo de acordo com a teoria financeira apresentada ou de forma contrária Além disso esta análise possibilita entender o comportamento de ambas as técnicas DEA e SFA bem como identificar semelhanças e diferenças entre elas A fim de atingir este objetivo as variáveis serão analisadas uma a uma para cada uma das técnicas e comparadas com os achados sobre a teoria financeira 432 Intermediação Neste estudo ao analisar o item de adequação do capital a variável Patrimônio Líquido sobre Ativo Total apresentou efeito negativo sobre a eficiência para a técnica DEA Desta forma considerase adequada à teoria utilizada como base deste estudo Ou seja a técnica DEA para este estudo para a base de dados deste estudo e para o período abrangido por ele demonstra suportar as teorias de finanças utilizadas para embasar este estudo Por outro lado o índice apresentou impacto positivo sobre a eficiência para a técnica SFA e portanto podese identificar uma não conformidade com a teoria apresentada anteriormente Para analisar a qualidade dos ativos o índice representado por Operações de Crédito sobre Ativo Total apresentou resultado positivo para a técnica DEA e para a técnica SFA por este motivo pode ser considerado como uma das variáveis que estão alinhadas a teoria financeira Além desse índice a razão entre Provisões sobre Operações de Crédito e Operações de Crédito também apresentou resultado compatível com a teoria financeira aplicada nesta dissertação Este fato para a variável em questão ocorreu apenas para a técnica DEA Ao estudar o impacto do indicador de gestão na eficiência bancária este estudo utilizouse do índice da soma do Resultado de Intermediação Financeira com Renda de Prestação de Serviços sobre a soma de Despesas com Pessoal com Despesas Administrativas a técnica DEA e SFA apresentaram um impacto negativo deste índice sobre a eficiência Neste caso sugerese sugere não adequação a teoria financeira 95 Um dos índices mais utilizados para justificar a eficiência bancária encontramse nos indicadores de rentabilidade Dentre eles o presente estudo utilizou os mais encontrados na literatura nacional e internacional o ROA e o ROE Os resultados obtidos demonstram que para a técnica DEA somente o ROE apresentou impacto a ser considerado onde encontrouse um efeito positivo na eficiência bancária e adequação a teoria financeira já abordada Os resultados obtidos nos indicadores de rentabilidade demonstram que o ROA apresentou impactos negativos para a técnica SFA resultado que indica neste estudo a não adequação a teoria fato que deve ser comprovado através de estudos futuros sobre o tema Já para os resultados do índice de lucratividade ROE encontrouse uma relação positiva para a técnica SFA e consequentemente o status de adequada a teoria financeira O índice utilizado para mensurar o impacto da liquidez na eficiência bancária e que apresentou efeito positivo tanto para a técnica DEA quanto para SFA e adequação à teoria financeira foi a razão encontrada entre o Ativo Circulante sobre o Ativo Total Com o objetivo de mensurar a sensibilidade ao risco de mercado duas variáveis foram analisadas A variável Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos sobre o Ativo Total que apresentou efeito positivo para DEA sendo classificado como adequado ao estipulado pela teoria financeira e que apresentou impacto negativo para SFA e foi classificada como não adequada à teoria financeira aplicada A variável Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos sobre Operações de Crédito apresentou efeito positivo para ambas as técnicas e portanto foi classificada como adequada a teoria financeira A Tabela 11 compila os resultados obtidos para a abordagem de intermediação para ambas as técnicas Análise Envoltória de Dados e Fronteira Estocástica 96 Tabela 11 Efeitos na eficiência SFA e DEA para Abordagem de Intermediação CAMELS Variável SFA Efeitos fixos e aleatórios R2 048 DEA R2 046 Efeito na eficiência Significância Coef Efeito na eficiência Significância Coef C Patrimônio Líquido Ativo Total POSITIVO 0000 1 13E01 NEGATIVO 0000 570E01 A Provisão sobre Operações de Crédito Operações de Crédito Sem relação Sem relação Sem relação POSITIVO 0000 990E02 Operações de Crédito Ativo Total POSITIVO 0000 388E01 POSITIVO 0000 413E01 M Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas NEGATIVO 0007 131E03 NEGATIVO 0000 25E01 E ROA Lucro Líquido Ativo Total x 100 NEGATIVO 0008 215E03 Sem relação Sem relação Sem relação ROE Lucro Líquido Patrimônio Líquido x 100 POSITIVO 0019 335E04 POSITIVO 0027 420E02 L Ativo Circulante Ativo Total POSITIVO 0000 406E01 POSITIVO 0000 393E01 S TVM e instrumentos Financeiros Derivativos Ativo Total NEGATIVO 0000 228E01 POSITIVO 0001 940E02 TVM e instrumentos Financeiros Derivativos Operações de Crédito POSITIVO 0007 615E06 POSITIVO 0000 860E02 Fonte Elaborada pelo autor 97 433 Rentabilidade Da mesma forma que a abordagem de Intermediação a abordagem de Rentabilidade também considerou as variáveis estatisticamente relevantes e com impactos positivos ou negativos sobre a eficiência É importante ressaltar que de acordo com o período base de dados e modelos matemáticos das técnicas DEA e SFA utilizados neste estudo podemos considerar que as variáveis que apresentam impactos positivos podem ser interpretadas como quanto maior o valor da variável analisada maior a eficiência DEA E SFA para a abordagem analisada Da mesma forma ao analisar as variáveis que apresentam impacto negativo a interpretação é quanto menor o valor da variável maior a eficiência DEA e SFA para a abordagem analisada Para a abordagem de Rentabilidade a variável Patrimônio Líquido sobre Ativo Total apresentou significância apenas para a técnica DEA identificando um impacto negativo desta variável na eficiência e consequentemente sendo classificada como adequada a teoria financeira abordada neste estudo Neste estudo a razão entre Provisões sobre Operações de Crédito e Operações de Crédito apresentou um impacto negativo para a técnica DEA o que condiz com a teoria de finanças apresentada Já para a técnica SFA o efeito observado foi positivo desenquadrandoa do embasamento teórico utilizado Já a variável Operações de Crédito sobre Ativo Total apresentou um efeito positivo sobre a eficiência apenas para a técnica SFA e portanto desenquadrandoa do grupo de variáveis que se adequam a teoria financeira deste estudo Ao analisar a soma do Resultado de Intermediação Financeira com Renda de Prestação de Serviços sobre a soma de Despesas com Pessoal com Despesas Administrativas encontrou se um impacto negativo deste índice sobre a eficiência para ambas as técnicas utilizadas neste estudo Este fato diante a teoria financeira utilizada por este estudo para a abordagem de rentabilidade evidenciou a impossibilidade de se concluir se um efeito negativo ou positivo no efeito da eficiência seria o mais adequado Os índices mais utilizados ao se mensurar eficiência bancária ROA e ROE são índices indispensáveis para a análise da eficiência bancária sob a ótica de rentabilidade Ambos os índices apresentaram efeito positivo na eficiência bancária para as técnicas DEA e SFA Desta forma considerase ambos os índices adequados a teoria financeira presente nesta dissertação O índice Ativo Circulante sobre o Ativo Total apresentou apenas para a técnica SFA 98 impacto positivo sobre a eficiência bancária alinhandose desta forma aos indicadores que se relacionam adequadamente a teoria financeira A variável Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos sobre o Ativo Total apresentou impacto positivo e adequação a teoria financeira apenas para a técnica SFA Já a variável Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos sobre Operações de Crédito apresentou impacto negativo para a técnica DEA e positivo para a técnica SFA Desta forma considerase adequado a teoria financeira apenas para a técnica SFA A Tabela 12 compila os resultados obtidos para a abordagem de intermediação para ambas as técnicas Análise Envoltória de Dados e Fronteira Estocástica 99 Tabela 12 Efeitos na eficiência SFA e DEA para Abordagem de Rentabilidade CAMELS Variável SFA Efeitos fixos e aleatórios R2 038 DEA R2 028 Efeito na eficiência Significância Coef Efeito na eficiência Significância Coef C Patrimônio Líquido Ativo Total Sem relação Sem relação Sem relação NEGATIVO 0000 327E01 A Provisão sobre Operações de Crédito Operações de Crédito POSITIVO 0002 477E02 NEGATIVO 0000 156E01 Operações de Crédito Ativo Total POSITIVO 0000 122E01 Sem relação Sem relação Sem relação M Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas NEGATIVO 0000 460E03 NEGATIVO 0000 201E01 E ROA Lucro Líquido Ativo Total x 100 POSITIVO 0000 114E02 POSITIVO 0000 155E01 ROE Lucro Líquido Patrimônio Líquido x 100 POSITIVO 0000 736E04 POSITIVO 0000 275E01 L Ativo Circulante Ativo Total POSITIVO 0000 642E02 Sem relação Sem relação Sem relação S TVM e instrumentos Financeiros Derivativos Ativo Total POSITIVO 0000 665E02 Sem relação Sem relação Sem relação TVM e instrumentos Financeiros Derivativos Operações de Crédito POSITIVO 0000 628E06 NEGATIVO 0000 65E02 Fonte Elaborada pelo autor 100 5 DISCUSSÃO DOS RESULTADOS A utilização combinada das técnicas DEA e SFA pode ser justificada pela diferença em seus modelos matemáticos que são mais sensíveis cada um com suas peculiaridades a determinadas alterações das variáveis Desta forma criase uma harmonia entre a utilização combinada das mesmas o que fortalece a análise dos fatores que influenciaram a eficiência bancária brasileira entre os anos de 2000 e 2018 ponto central desta dissertação A fim de aprofundar a análise dos resultados estudos internacionais que possuem resultados sobre a influência de índices CAMELS na eficiência bancária foram analisados e relacionados É importante ressaltar que além de modelos matemáticos diferentes estes estudos também possuem diferenças entre locais onde foram aplicados bem como os períodos de aplicação O quadro sinótico dos pressupostos teóricos relativos à abordagem de Intermediação e de Rentabilidade está estampado no Quadro 7 101 Quadro 7 Adequação à Teoria Financeira Abordagem de Intermediação e Rentabilidade CAMELS Variável Abordagem de intermediação Abordagem de rentabilidade DEA SFA DEA SFA Adequação à Teoria Financeira Adequação à Teoria Financeira Adequação à Teoria Financeira Adequação à Teoria Financeira C Patrimônio Líquido Ativo Total CORROBORA DIVERGE CORROBORA X A Provisão sobre Operações de Crédito Operações de Crédito X CORROBORA CORROBORA DIVERGE Operações de Crédito Ativo Total CORROBORA CORROBORA X DIVERGE M Resultado de Intermediação Financeira Rendas de Prestação de Serviços Despesas de Pessoal Despesas Administrativas DIVERGE DIVERGE INCONCLUSIVO INCONCLUSIVO E ROA Lucro Líquido Ativo Total x 100 CORROBORA X CORROBORA CORROBORA ROE Lucro Líquido Patrimônio Líquido x 100 CORROBORA CORROBORA CORROBORA CORROBORA L Ativo Circulante Ativo Total CORROBORA CORROBORA X CORROBORA S TVM e instrumentos Financeiros Derivativos Ativo Total DIVERGE CORROBORA X CORROBORA TVM e instrumentos Financeiros Derivativos Operações de Crédito DIVERGE DIVERGE DIVERGE CORROBORA Fonte Elaborado pelo autor 102 Para a análise da Adequação de Capital através da variável Patrimônio Líquido sobre Ativo Total ao considerar a aplicação da técnica DEA para a abordagem de Intermediação e realização do segundo estágio Adesina 2019 buscou preencher uma lacuna importante na literatura de eficiência bancária investigando a influência do IC Capital Intelectual nas eficiências técnicas alocativas e de custo dos bancos na África A amostra consistiu de 339 bancos comerciais em 31 países africanos no período de 2005 a 2015 Rosman Wahab e Zainol 2014 e Sufiana e Habibullah 2012 evidenciaram o impacto positivo que esta variável possui sobre a eficiência bancária para bancos islâmicos No caso dos bancos em países do Oriente Médio e da Ásia em média podem ser caracterizados como tecnicamente eficientes ao longo dos períodos Isso indica que o a gestão dos bancos foi capaz de controlar de forma eficiente os custos e usar a diversidade de entradas para produzir saídas independentemente dos efeitos de escala Ainda para a abordagem de Intermediação Bittar Pukthuanthong e Walker 2019 através da utilização da técnica DEA ao analisarem a adequação de capital encontraram efeitos positivos da capitalização sobre a eficiência Os autores exploraram o efeito de manter maiores índices de capital e liquidez sobre a eficiência dos bancos convencionais e islâmicos bem como se esse efeito difere entre bancos menos eficientes e mais eficientes usando regressões de quantis condicionais Os resultados sugerem que os índices de capital e liquidez estão positivamente associados à eficiência existente dos dois tipos de banco No entanto este efeito positivo parece ser menos pronunciado para os bancos islâmicos indicando que maiores índices de capital e liquidez ampliam a eficiência diferença entre os dois tipos de banco em favor dos bancos convencionais Por fim os resultados sugerem que bancos mais capitalizados e mais líquidos também foram mais eficientes durante a crise financeira global e a Primavera Árabe Da mesma forma porém com a aplicação da técnica SFA Otero et al 2020 evidenciaram impacto positivo desta variável sobre a eficiência das instituições financeiras Este estudo investigou a eficiência de custos do sistema bancário MENA indústria para o período de 20052012 usando um modelo de fronteira estocástica com variáveis específicas do país Os resultados da análise mostraram que a pontuação média de eficiência de custo foi de 77 e que houve uma melhoria nas pontuações de eficiência durante o período coberto A variação em termos de eficiência de custos é de 19 por cento entre os países Geograficamente Israel 86 é o mais econômico enquanto os bancos do Kuwait 67 são os menos 103 Os resultados também mostraram que os bancos em Jordânia 82 Tunísia 82 e Omã 80 em média tiveram maior pontuações comparadas com outros países Diante desses resultados parece que ainda há espaço para melhorias quando se trata de banco eficiência nesta região Em relação aos determinantes da eficiência o artigo mostra a importância da estrutura de mercado como principal determinante da eficiência de custos para os bancos que operam na região MENA Em particular o nível de concentração e participação de mercado tiveram uma influência negativa na técnica eficiência apoiando a hipótese de vida silenciosa Também encontrouse suporte para a hipótese de competiçãoineficiência que pode ser devido ao relação menos estável entre clientes e bancos e quanto maior probabilidade de buscarem melhores condições em outro lugar No nível individual o tamanho do banco mostra a importância das economias de escala e o nível de capital também produz um efeito positivo Finalmente nenhuma diferença foi encontrada em termos de eficiência entre bancos convencionais e islâmicos Os efeitos na estrutura do mercado são mais importantes no pós crise período do que o précrise Na verdade o efeito do desempenho na eficiência é só positivo no período de crise mostrando que bancos em expansão fase têm outras fontes disponíveis não dependem da eficiência para ser rentáveis e menos preocupados com os custos O mesmo acontece com competição e é em tempos de crise que a rivalidade tem um impacto negativo na eficiência de custos Crise pode forçar bancos a serem mais agressivos para tentar capturar depósitos e este comportamento aumenta a instabilidade do a relação entre os bancos e os clientes A crise também pode afetar negativamente a solvência de certos bancos incentivando os clientes a mudar para outra instituição financeira devido à amplificação da informação assimétrica Portanto os resultados mostram a importância de se considerar a situação econômica na análise da eficiência de custos Nesse sentido as políticas bancárias devem promover o desempenho do banco porque como apontamos está positivamente associado ao custo eficiência Além disso as autoridades monetárias e os decisores políticos devem adotar políticas que podem aumentar o tamanho dos bancos MENA mas ao mesmo o tempo controla o nível de concentração e competição isto é porque de acordo com nossa análise altos níveis de ambas as características de mercado conduzem a um efeito negativo na eficiência Ademais as autoridades devem aumentar a eficiência em uma fase de expansão do mercado porque é quando os bancos estão menos comprometidos com a eficiência de custos Finalmente capital 104 requisitos bem como contribuindo para a solvência dos bancos têm mostrou um impacto positivo na eficiência apoiando regras e regulamentos como Basileia III Saeed et al 2020 encontraram em seu estudo resultado importante ao concluírem que para a abordagem de Intermediação geralmente um aumento da capitalização apresenta impacto negativo na eficiência contudo para os bancos islâmicos da amostra utilizada o resultado é inverso e o aumento da capitalização possui impacto positivo sobre a eficiência e resulta em um aumento da eficiência bancária onde a capitalização é representada pela relação entre o patrimônio líquido e os ativos O que evidencia que a capitalização mais alta reduz a exposição ao risco e oferece uma afirmação da hipótese regulatória segundo a qual o capital deve aumentar proporcionalmente com a exposição ao risco do banco O efeito da capitalização para o risco representa as consequências não intencionais de ações regulatórias onde os bancos respondem aos requisitos de capitalização aumentados assumindo mais risco para compensar os benefícios perdidos da alavancagem Sarmiento e Galán 2017 através da aplicação da técnica SFA também evidenciaram a relação positiva desta variável sobre a eficiência Em geral os bancos colombianos exibem uma tendência de queda no risco de crédito e de mercado juntamente com níveis estáveis de capitalização e liquidez crescente No entanto os autores observaram diferenças importantes no nível de exposição ao risco de bancos com diferentes características de porte e propriedade que coincidem com as alterações regulatórias mencionadas incluindo aquelas adotadas em 20072008 para atender aos padrões de Basileia II Os autores consideram a medida de capitalização como baseada em duas características importantes Em primeiro lugar a regulamentação colombiana que estabelece que os bancos estrangeiros devem manter o mesmo capital mínimo que os bancos locais para operar Isso ocorre porque os bancos estrangeiros operam como subsidiárias em vez de agências na Colômbia e por sua vez precisam manter seu próprio capital Desta forma a medida de capitalização apresentada é comparável entre bancos com diferentes propriedades Em segundo lugar os autores argumentam que as diferenças nos níveis de capitalização podem sinalizar o apetite de risco dos bancos e influenciar seu desempenho Foi identificado no estudo realizado que níveis mais altos de capitalização levam a uma maior eficiência de custo e lucro As razões para esses resultados podem ser derivadas dos problemas de agência entre acionistas e administradores Os acionistas de bancos altamente capitalizados têm mais incentivos para controlar melhores custos e alocação de capital do que os de bancos 105 pouco capitalizados Isso incentiva melhores mecanismos de governança corporativa que podem levar a melhorias de eficiência Os resultados indicam que o efeito da capitalização sobre a eficiência difere entre bancos com diferentes tamanhos e participações e que os bancos pequenos e domésticos se beneficiam mais de índices de capital mais elevados em termos de eficiência de custo e lucro Para Djalilov e Piesse 2019 ao utilizarem a técnica SFA para a análise de adequação de capital encontraram impacto negativo sobre a eficiência Para os autores o nível de requisitos de capital impostos pelos reguladores também é controverso na literatura De acordo com a visão do interesse público a política de requisitos de capital reduz significativamente o nível de risco moral quando os proprietários dos bancos são obrigados a ter mais capital em risco Isso eventualmente leva a empréstimos mais cuidadosos e melhor desempenho O efeito dos requisitos de capital prudencial sobre a estabilidade bancária parece ser positivo nos setores bancários com supervisão e monitoramento relativamente fracos e instituições subdesenvolvidas características presentes nos setores bancários de países em transição Por outro lado a maioria desses países experimentou um crescimento sustentável e baixas taxas de inflação nas últimas duas décadas Isso causou aumentos na demanda por empréstimos bancários e portanto a presença de requisitos de capital prudencial pode limitar as oportunidades de crescimento dos bancos Sufian e Habibullah 2009 ao aplicar a técnica DEA corrobora com este resultado e considera o efeito negativo desta variável sobre a eficiência bancária Os autores demonstram que os bancos nacionais foram atingidos nas últimas décadas em simultâneo em várias frentes nomeadamente no levantamento de fundos estrangeiros perdas cambiais um aumento acentuado dos NPLs e perdas em participações acionárias A erosão da base de capital do sistema bancário devido a todos esses fatores restringiu severamente sua capacidade de emprestar até mesmo para empresas solventes em meio à crise devido à necessidade de cumprir as regras internacionais de adequação de capital Othman AbdulMajid e AbdulRahman 2017 ao aplicarem a técnica SFA também encontraram impacto negativo desta variável sobre a eficiência bancária na abordagem de Intermediação De acordo com a hipótese de risco bancos altamente capitalizados podem ter menos incentivos de risco e portanto são mais propensos a adotar práticas de redução de custos Por exemplo os acionistas podem ser mais ativos no controle dos custos dos bancos ou na alocação de capital por meio da disciplina de mercado 106 Em outras palavras esperase que os bancos altamente capitalizados que oferecem grande quantidade de financiamento de parceria sejam mais eficientes Para os autores o estudo apoia a ideia de que o financiamento de parceria pode ser usado como uma estratégia para melhorar a eficiência bancária especialmente para bancos com alta capitalização portanto baixo risco de capital Portanto ao monitorar os bancos que oferecem financiamento de parceria as autoridades podem precisar examinar seu nível de financiamento de parceria bem como seu risco de capital O estudo de Silva et al 2017 apesar de não ter realizado o segundo estágio para verificar o impacto de CAMELS na eficiência considerou positivamente mensurar a eficiência bancária através dos índices escolhidos e objetivou analisar índices que representam a possibilidade de aumento do comprometimento da gestão de crédito das cooperativas e a relação com a necessidade de mudanças nos instrumentos de controle gerencial Já para a abordagem de Rentabilidade ao estudar a variável Patrimônio Líquido sobre Ativo Total Fernandes Stasinakis e Bardarova 2018 encontraram impactos positivos gerados por este indicador CAMELS na eficiencia bancária de países europeus através da técnica DEA Da mesma forma Chen et al 2019 ao estudar os bancos Taiwaneses através da técnica DEA também encontrou impactos positivos sobre a eficiência bancária Ao considerar a razão entre Provisões sobre Operações de Crédito e Operações de Crédito Sufian e Habibullah 2009 ao considerar a técnica DEA e a abordagem de Intermediação encontrou que a eficiência bancária está negativamente relacionada ao comportamento de preferência de despesa e às condições econômicas enquanto a eficiência do banco está positivamente relacionada à intensidade de empréstimos em bancos da Malásia Sufiana e Habibullah 2012 chegaram ao mesmo resultado ao utilizar a técnica DEA em bancos da Indonésia onde os bancos relativamente melhor capitalizados com níveis de liquidez mais baixos exibem níveis mais elevados de eficiência enquanto os bancos com alto risco de crédito e despesas gerais tendem a ser relativamente ineficientes em sua função de intermediação O resultado sobre o impacto do crescimento do PIB dá suporte ao argumento da associação positiva entre crescimento econômico e desempenho do setor financeiro Ao estudar bancos chineses através da técnica DEA Avkiran 2011 ao analisar o impacto da variável da razão entre Provisões sobre Operações de Crédito e Operações de Crédito não encontrou relação de significância na abordagem de Rentabilidade Nesse estudo a rentabilidade teve associação com dois índices de lucratividade lucros pósimpostos sobre 107 ativos totais médios e retorno sobre o patrimônio líquido médio indicados como significativos em associação com estimativas de eficiência AVKIRAN 2011 DONG et al 2016 Ainda em relação a Qualidade dos Ativos sob a ótica da abordagem de Intermediação Saeed et al 2020 não encontraram resultados estatisticamente significativos para mensurar o impacto das Operações de Crédito sobre o Ativo Total dos bancos estudados ao utilizarem a técnica SFA Sufian e Habibullah 2009 através da aplicação da técnica DEA encontrou a relação positiva entre esta variável e o seu impacto sobre a eficiência bancária Da mesma forma Rosman Wahab e Zainol 2014 também chegaram a mesma conclusão ao aplicar a técnica DEA para o estudo de bancos islâmicos Por outro lado ao estudar as instituições bancárias da Indonésia Sufian e Habibullah 2012 encontram impactos negativos sobre a eficiência de instituições financeiras Sob a ótica da abordagem de Rentabilidade não houve estudos no levantamento realizado neste trabalho que analisaram os impactos da variável Operações de Crédito sobre Ativo Total a fim de analisar a qualidade dos ativos Ainda sobre a análise das variáveis CAMELS sob a ótica de Intermediação para Gestão a variável que traz a soma do Resultado de Intermediação Financeira com Renda de Prestação de Serviços sobre a soma de Despesa com Pessoal com Despesa Administrativa não foi encontrada nos estudos analisados Contudo o estudo de Shaddady e Moore 2018 identifica através do estudo das variáveis CAMELS e utilização da técnica DEA a relevância da supervisão sobre a gestão com um impacto positivo para economias avançadas e negativo para economias emergentes Também não houve estudos sob a ótica de Rentabilidade que analisaram o impacto desta variável a fim de analisar a Gestão das instituições bancárias As variáveis ROA e ROE são as variáveis mais encontradas em estudos sobre rentabilidade e eficiência bancária Nos estudos utilizados nesta pesquisa a variável ROA apresentou a totalidade dos casos tanto para a abordagem de Intermediação quanto para a abordagem de Rentabilidade Para os estudos analisados neste trabalho Djalilov e Piesse 2019 através da aplicação da técnica SFA identificaram impactos positivos da variável ROA sobre a eficiência para a abordagem de Intermediação ao considerar que a lucratividade é importante porque contribui para os níveis de eficiência dos bancos Portanto o Retorno sobre Ativos ROA é usado como uma variável de lucratividade Isso pode significar que os bancos com maior ROA e maiores em tamanho são mais eficientes Além disso também pode indicar que o crescimento 108 positivo do PIB e as restrições de atividade melhoram a eficiência bancária na amostra utilizada no estudo Da mesma forma Otero et al 2020 também encontraram impactos positivos da variável ROA sobre a eficiência bancária através da utilização da técnica SFA Para os autores o efeito do ROA é estatisticamente significativo e confirma a noção geral de que a lucratividade está positivamente relacionada à eficiência de custos Consequentemente os bancos com maior lucro tendem a ser mais eficientes e afirmam que resultado encontrado demonstra que os bancos com melhores práticas de gestão e controle de custos obtiveram maior lucratividade destacando a importância da eficiência de custos para o desempenho dos estabelecimentos nos países estudados Além disso os bancos mais rentáveis dispõem de recursos para investir em tecnologia processos e recursos humanos para aumentar sua eficiência de custos Ao analisar o impacto da variável ROA sobre a eficiência bancária através da técnica DEA Rosman Wahab e Zainol 2014 encontraram efeito também positivo e demonstraram que as instituições financeiras se tornam mais eficientes de acordo com o aumento de sua rentabilidade Isso ocorre pois a variável ROA incluída no modelo de regressão como proxy da lucratividade auxilia na comprovação de que os bancos que relatam índices de lucratividade mais altos são geralmente preferidos pelos clientes e portanto atraem a maior parte dos depósitos e os melhores tomadores de crédito potenciais que por sua vez criam um ambiente favorável para os bancos lucrativos serem mais eficientes Assim como para a abordagem de Intermediação para a abordagem de Rentabilidade foram encontrados estudos que analisaram apenas a variável ROA Nestes estudos Theodoridis e Anwar 2011 através da aplicação da técnica SFA encontrou um efeito positivo sobre a razão entre Lucro Líquido e Ativo Total sobre a eficiência bancária O ROA como proxy para a lucratividade do banco tem um impacto positivo na eficiência de custos dos bancos indonésios sobre os três modelos utilizados e com um nível de confiança de 99 Estes resultados implicam que a rentabilidade obtida pelos bancos indonésios impactam significativamente na melhoria do nível de eficiência de custos dos bancos comerciais durante o período Da mesma forma porém através da aplicação da técnica DEA Fernandes Stasinakis e Bardarova 2018 também encontraram efeitos positivos desta variável sobre as instituições bancárias estudadas Para os autores o ROA ainda exerce um efeito positivo e estatisticamente significativo em duas medidas de eficiência dos bancos No entanto o efeito é menor durante a crise financeira do que em períodos sem crises Este é um resultado esperado uma vez que os 109 bancos tendem a ser mais produtivos quando conseguem atrair níveis mais elevados de depósitos e encontrar tomadores de crédito com boa capacidade de crédito Por outro lado Chen et al 2019 ao estudar bancos taiwaneses encontrou impacto negativo desta variável sobre a eficiência bancária diferindo das expectativas gerais o que reflete a circunstância em que os bancos negligenciam a eficiência operacional ao tentar expandir suas operações e itens de negócios para criar um ROA mais alto Para a abordagem de Intermediação Sarmiento e Galán 2017 analisaram através da aplicação da técnica SFA que a razão entre o Ativo Circulante e Ativo Total tem um efeito negativo na eficiência bancária A proporção de ativos líquidos sobre ativos totais tem aumentado gradualmente ao longo do tempo especialmente para bancos grandes e estrangeiros A liquidez é medida como a razão entre os ativos líquidos sobre os ativos totais em que os ativos líquidos incluem disponibilidades de caixa negociáveis e disponíveis para vender instrumentos de dívida pública e privada e garantias dadas em operações compromissadas Ativos com maior liquidez impedem os bancos de descasamentos de vencimento embora mantêlos não tenha custo pois eles têm vencimentos mais curtos e portanto retornos mais baixos Por outro lado Bitar Pukthuanthong e Walker 2019 identificaram que os índices de liquidez estão positivamente associados à eficiência bancária analisada Para os autores tanto os depositantes quanto os investidores preferem negociar com bancos que tenham uma proporção saudável de ativos líquidos em relação aos depósitos e portanto os bancos mais eficientes tendem a manter índices de liquidez mais altos que podem servir como um mecanismo de segurança para proteger contra baixas de capital vendas dispendiosas e necessidades repentinas de aumentar o patrimônio líquido caro o que poderia afetar negativamente suas pontuações de eficiência Para a abordagem de Rentabilidade não houve estudos no levantamento realizado neste trabalho que analisaram os impactos da variável Ativo circulante sobre Ativo Total a fim de analisar a liquidez Também não houver estudos para ambas as abordagens que analisaram os impactos das variáveis de Títulos de Valores Mobiliários e Instrumentos Financeiros Derivativos sobre Ativo Total ou sobre Operações de Crédito a fim de analisar a sensibilidade ao risco de mercado 110 6 CONCLUSÕES Buscouse nesta pesquisa identificar os fatores que influenciaram a eficiência dos bancos brasileiros nas abordagens de intermediação e rentabilidade no período abordado sob a ótica de duas principais técnicas de mensuração de eficiência DEA e SFA Para isto foram tratados os dados de 221 instituições financeiras pertencentes ao grupo B1 extraídas do endereço eletrônico do BACEN atraves do relatório denominado IFdata Através da avaliação dos indicadores de eficiência pelo modelo proposto para as técnicas DEA e SFA em conjunto com a Análise de Regressão Linear Múltipla e do teste de heterocedasticidade foi possível identificar as variáveis CAMELS que explicam a eficiência bancária das instituições analisadas Para a abordagem de intermediação dos dez indicadores CAMELS nove apresentaram resultados significativos dentre os quais apenas dois deles do grupo Management e Sensitivity to Market Risk apesar de significativos divergem para ambas as técnicas da teoria financeira que suporta este estudo Da mesma forma para a abordagem de rentabilidade dos dez indicadores CAMELS nove apresentaram resultados significativos dentre os quais apenas um deles do grupo Management apesar de significativo foi inconclusivo para ambas as técnicas em relação a teoria financeira que suporta este estudo Notase uma relação entre os indicadores deAsset Quality Earnings eLiquidity com a eficiência na abordagem de intermediação Por outro lado uma relação inversa entre os indicadores de Management e Sensibility to Market Risk relacionados a mesma abordagem O indicador de Capital Adequacy ficou inconclusivo Na abordagem de rentabilidade percebeu se uma relação positiva entre os indicadores de Capital Adequacy Earnings e Liquidity com a eficiência e resultados inconclusivos nos demais Os resultados sugerem que existem práticas relacionadas a essas variáveis que influenciam positivamente os escores de eficiência bem como práticas que influenciam negativamente os escores de eficiência o que também foi observado na revisão bibliográfica realizada Diversos pontos deste estudo cumpriram os objetivos propostos e permitiram concluir que os modelos escolhidos para suportar as técnicas DEA e SFA apoiados na adaptação do modelo de inputs e outputs realizado bem como os indicadores CAMELS aplicados na regressão são suficientes para explicarem a eficiência bancária brasileira no período deste estudo 111 Esta pesquisa tem como limitações a utilização conjunta das técnicas vez que ambas possuem modelos matemáticos diferentes Entretanto aprofundarse a essa análise seria adentrar em uma discussão de modelos matemáticos e não na análise da eficiência bancária foco primário deste estudo Embora o objetivo não tenha sido comparar as técnicas notase uma similaridade nos resultados obtidos por ambas técnicas Para este estudo considerando o país a base de dados disponível e o período analisado não se identificou diferenças significativas entre as técnicas Por isso optouse por focarse no resultado da eficiência e não nos modelos matemáticos possíveis Foi utilizado o modelo que mais se adequava à base de dados disponível Outro fator limitante concentrase nas características individuais de cada país e de seu sistema financeiro ou seja houve índices CAMELS utilizados em trabalhos internacionais cuja aplicação não foi possível no presente estudo face à divergência de padrão de informações financeiras disponível no site do Bacen O mesmo se deu para os inputs e outputs utilizados nas abordagens de eficiência Futuros trabalhos poderão aprofundarse nas análises de eficiência por meio da revisão dos modelos matemáticos utilizados para embasamento das técnicas DEA e SFA e dos inputs e outputs utilizados em cada uma das abordagens Ademais a base de dados poderá ser segmentada entre instituições públicas ou privadas em combinação entre elas ou ainda entre instituições de diferentes portes a fim de que se possam inferir conclusões sobre suas comparações ao contrário do que foi feito no presente trabalho que analisou a base que engloba todas as instituições nacionais Sugerese também para próximos estudos a revisão das variáveis CAMELS utilizadas vez que não existe um padrão para tanto ou seja diversos outros índices que se encaixam em CAMELS poderiam ser incluídos na análise de seu impacto nas variáveis dependentes que no caso consistem nos escores de eficiência calculados para DEA e SFA 112 REFERÊNCIAS BIBLIOGRÁFICAS ABEDIFAR P HASAN I TARAZI A Financegrowth nexus and dualbanking systems Relative importance of Islamic banks Journal of Economic Behavior Organization v 132 p 198215 2016 ADESINA K S Bank technical allocative and cost efficiencies in Africa The influence of intellectual capital The North American Journal of Economics and Finance v 48 p 419 433 2019 ADMATI A R HELLWIG M F Bank Leverage Welfare and Regulation Rock Center for Corporate Governance at Stanford University Working Paper No 235 Stanford University Graduate School of Business Research Paper No 3729 MPI Collective Goods Discussion Paper No 201813 httpdxdoiorg102139ssrn3257957 2019 AFFES Z HENTATIKAFFEF R Predicting US Banks Bankruptcy Logit Versus Canonical Discriminant Analysis Computational Economics v54 n1 p199244 2017 AFFINITO M ALBARETO G SANTIONI R Purchases of sovereign debt securities by Italian banks during the crisis the role of balancesheet conditions Bank of Italy Economic Research and International Relations Area n 330 p 541 2016 AHAMED M M Asset quality noninterest income and bank profitability Evidence from Indian banks Economic Modelling v 63 june 2017 p 114 2017 AIGNER D J LOVELL A K SCHMIDT P Formulation and estimation of stochastic frontier production function models Journal of Econometrics v6 n1 p2137 1977 ALESSANDRI P HALDANE A G Banking on the State Bank of England working paper Disponível em httpswwwbankofenglandcoukspeech2009bankingonthestate nov 2009 ALTUNBAS Y et al Examining the Relationships between Capital Risk and Efficiency in European Banking European Financial Management v 13 n 1 4970 2007 ASSAF NETO A Mercado Financeiro 4ed São Paulo Atlas 2011 AVKIRAN N K Association of DEA SuperEfficiency Estimates with Financial Ratios Investigating the case for Chinese banks Omega v39 n3 p323334 2011 AVKIRAN N K CAI L C Predicting bank financial distress prior to crises New Zealand Finance Colloquium 2012 AVKIRAN N LIN C Predicting bank financial distress prior to crises UQ Business School The University of Queensland Australia Working paper Disponível em httpwwwnzfcacnzarchives2012papersupdated24pdf 2012 BANKER R CHARNES A COOPER W Some Models for Estimating Technical and Scale Inefficiencies in Data Envelopment Analysis Management Science v30 n9 p10781092 1984 113 BARTH J R CAPRIO G LEVINE R Bank Regulation and Supervision in 180 Countries from 1999 to 2011 httpsdxdoiorg102139ssrn2203516 2013 BASEL COMMITTEE ON BANKING SUPERVISION Principles for the management of credit risk Basel Switzerland Bank for International Settlements 2000 Disponível em httpswwwbisorgpublbcbs75pdf Acesso em 14 fev 2021 BATISTA F D Metodologia para o Uso da Análise por Envoltória de Dados no Auxilio a Decisão Dissertação Mestrado em Engenharia de Produção Universidade Federal de Itajubá Itajubá 2009 BATTESE G E COELLI T J A Model for Technical Inefficiency Effects in a Stochastic Frontier Production Function for Panel Data Empirical Economics v20 p325332 1995 BATTESE G E CORRA G S Estimation of a production frontier model With application to the pastoral zone of Estern Australia Australian Journal of Agricultural Economics v21 n3 p169179 1977 BECKER J L LUNARDI G L MACADA A C G Análise de Eficiência dos Bancos Brasileiros Um enfoque nos investimentos realizados em Tecnologia de Informação TI Producao v13 n2 2003 BEGENAU J STAFFORD E Inefficient Banking Harvard Business School working paper p 49 dec 2016 BELTRAME F PREVITALI D SCLIP A Systematic risk and banks leverage The role of asset quality Finance Research Letters v 27 dec 2018 p 113117 2018 BERGER A N et al Bank Concentration and Competition An Evolution in the Making Journal of Money Credit and Banking v 36 n 3 pt 2 p 433451 jun 2004 BERGER A N The relationship between capital and earnings in banking Journal of Money Credit and Banking v 27 p 432456 1995 BERGER A N DEYOUNG R Problem loans and cost efficiency in commercial banks Journal of Banking Finance v 21 n 6 jun 1997 p 849870 1997 BERGER A N HUMPHREY D B Efficiency of financial institutions International survey and directions for future research European Journal of Operational Research v98 n2 p175212 1997 BERGER A N MESTER L J Inside the Black Box What Explains Differences in the Efficiencies of Financial Institutions Journal of Banking and Finance v21 p895947 1997 BERGER A BONACCORSI DI PATTI E Capital structure and firm performance A new approach to testing agency theory and an application to the banking industry Journal of Banking Finance v 30 n 4 p 10651102 2006 BERGER A HASAN I ZHOU M Bank ownership and efficiency in China What will happen in the worlds largest nation Journal of Banking and Finance v33 p113130 2009 114 BERGER A KLAPPER L TURKARISS R Bank Competition and Financial Stability Journal of Financial Services Research v 35 n 2 p 99118 apr 2009 BITTAR M PUKTHUANTHONG K WALKER T J Efficiency in Islamic vs Conventional Banking The Role of Capital and Liquidity jul 58 p httpdxdoiorg102139ssrn3423228 2019 BONGINI P NERI L PELEGATTI M The importance of being systemically important financial institutions Journal of Banking and Finance v50 n1 p562574 2015 BONIN J P HASAN I WACHTEL P Bank performance efficiency and ownership in transition countries Journal of Banking Finance v 29 n 1 p 3153 2005 BOUCINHA M RIBEIRO N WEYMANJONES T An assessment of Portuguese banks efficiency and productivity towards euro area participation Journal of Productivity Analysis v39 n2 177190 2013 BOWERMAN B L OCONNEL R T Linear statistical models An applied approach 2ed Boston PWSKent 1990 BRAGA JUNIOR J SILVA C A SILVA D E P Auditoria da Gestão da Inovação no Setor Bancário O caso do BANESE Navus Revista de Gestão e Tecnologia v5 n1 2014 BRITO R A Há espaços para melhora do setor leiteiro Uma análise de fronteira estocástica de produção e regressão quantílica utilizando dados do Censo Agropecuário 2006 IBGE Tese Doutorado em Ciências Economia Aplicada Escola Superior de Agricultura Luiz de Queiroz Universidade de São Paulo Piracicaba 2016 CANBAS S CABUK A KILIC S B Predicting of commercial bank failure via multivariate statistical analysis of financial structures The Turkish case European Journal of Operational Research v166 n2 p528546 2005 CHARNES A COOPER W W RHODES E Measuring the Efficiency of Decision Making Units European Journal of Operational Research v2 p429444 1978 CHEN S et al Bank Competition Foreign Bank Entry and RiskTaking Behavior Cross Country Evidence Journal of Risk and Financial Management v 12 n 106 p 126 2019 CHORTAREAS G E GIRARDONE C VENTOURI A Bank supervision regulation and efficiency Evidence from the European Union Journal of Financial Stability v8 n4 p292 302 2012 CHRISTOPOULOS A G MYLONAKIS J DIKTAPANIDIS P Could Lehman Brothers Collapse Be Anticipated An Examination Using CAMELS Rating System International Business Research v4 n2 p 1119 2011 COELLI T J RAO D S P BATTESE G E An introduction to efficiency and productivity analysis 3ed London Kluwer Academic Publishers 2005 COOPER W W SEIFORD L M ZHU J Handbook on Data Envelopment Analysis 2ed New York Springer US p498 2011 115 CURRY T J ELMER P J FISSEL G S Using Market information to help identify distressed institutions A regulatory perspective FDIC Bank Review v15 n3 p116 2003 DANELON A F Determinantes da eficiência técnica na agroindústria sucroenergética brasileira Uma abordagem por fronteira estocástica Dissertação Mestrado em Ciências Economia Aplicada Escola Superior de Agricultura Luiz de Queiroz Universidade de São Paulo Piracicaba 2018 DE BRANDT O LECARPENTIER S POUVELLE C Determinants of banks liquidity a French perspective on interactions between market and regulatory requirements Journal of Banking Finance v 124 working papers 782 2020 DESPOTIS D K SMIRLIS Y G Data envelopment analysis with imprecise data European Journal of Operational Research v140 n1 p2436 2002 DeYOUNG R Management quality and Xinefficiency in National Banks Journal of Financial Services Research v13 n1 p522 1998 DINCER H et al A performance evaluation of the Turkish banking sector after the global crisis via CAMELS ratios Procedia Social and Behavioral Sciences v24 n1 p 15301545 2011 DJALILOV K PIESSE J Bank regulation and efficiency Evidence from transition countries International Review of Economics Finance v 64 p 308322 2019 DONG Y et al Evaluating the performance of Chinese commercial banks A comparative analysis of different types of banks European Journal of Operational Research v252 n1 p280295 2016 DRAKE L Efficiency in UK Banking Applied Financial Economics v11 pp557571 2001 DRAKE L HALL M J B SIMPER R The Impact of Macroeconomic and Regulatory Factors on Bank Efficiency A nonparametric analysis of Hong Kongs banking system Journal of Banking and Finance v30 n5 p14431466 2006 DRAPER N R SMITH H Applied regression analysis 2ed New York John Wiley and Sons 1981 DUAN J ZHANG H Pricing Hang Sang Index Options Around The Asian Financial Crises A GARCH Approach Journal of Banking and Finance v 25 n 11 p 19892014 2001 EKINCI R POYRAZ G The Effect of Credit Risk on Financial Performance of Deposit Banks In Turkey Procedia Computer Science v 158 p 979987 2019 END J W KRUIDHOF M Modelling the liquidity ratio as macroprudential instrument Journal of Banking Regulation v 14 n 2 p 106 2013 FARRELL M The Measurement of Productive Efficiency Journal of the Royal Statistical Society v120 n3 p253290 1957 116 FEDERAL RESERVE SYSTEM Board of Governors of the Federal Reserve System Division of Banking Supervision and Regulation Uniform Financial Institutions Rating System SR 9638 SUP 26 de dezembro 1996 FERNANDES F D S STASINAKIS C BARDAROVA V Twostage DEATruncated Regression Application in Banking Efficiency and Financial Development Expert Systems with Applications v 96 n 15 apr p 284301 2018 FETHI M D PASIOURAS F Assessing Bank Efficiency and Performance with Operational Research and Artificial Intelligence Techniques A survey European Journal of Operational Research v204 n2 p189198 2010 FIORDELISI F MARQUESIBANEZ D MOLYNEUX P Efficiency and risk in European banking Journal of Banking and Finance v35 n5 p13151326 2011 FONSECA J J S Metodologia da Pesquisa Científica Fortaleza UEC 2002 FORGIONE A F MIGLIARDO C Forecasting distress in cooperative banks The role of asset quality International Journal of Forecasting v 34 n 4 p 678695 2018 FREAZA F P MADEIRO G L E GOMES L F A M Analise de Eficiência do Mercado Bancário Brasileiro Utilizando a Metodologia da Analise Envoltoria de Dados XXXVIII Simpósio Brasileiro Pesquisa Operacional Anais 2004 FRIES S TACI A Cost efficiency of banks in transition Evidence from 289 banks in 15 postcommunist countries Journal of Banking Finance v 29 n 1 p 5581 2005 FRIOSI J et al Análise Exploratória da Inovação Bancária Brasileira e as Tendências Para o Setor Revista de Gestão Empresarial v1 n1 2017 FU X M HEFFERNAN S The effects of reform on Chinas bank structure and performance Journal of Banking and Finance v33 n1 p3952 2009 FUJI H MANAGI S MATOUSEK R Indian Bank Efficiency and Productivity Changes with Undesirable Outputs A Disaggregated Approach Journal of Banking and Finance v 38 p 4150 2014 FUNGACOVA Z PESSAROSSI P WEILL L Is Bank Competition Detrimental to Efficiency Evidence from China China Economic Review v 27 dec BOFIT Discussion Paper No 312012 httpdxdoiorg102139ssrn2206860 2013 GAGANIS C PASIOURAS F Financial supervision regimes and bank efficiency International evidence Journal of Banking Finance v 37 n 12 p 54635475 2013 GALTON F Regression toward mediocrity in heredity stature Journal of Anthropological Institute v15 p246263 1885 GIL A C Métodos e Técnicas de Pesquisa Social 5 ed São Paulo Atlas 1999 GODDARD J MOLYNEUX P WILSON J O S The profitability of european banks a crosssectional and dynamic panel analysis The Manchester School v 72 p 363381 2004 117 GODOI A F et al Análise das Variáveis Macroeconômicas e do acrônimo CAMELS sobre o retorno das ações ordinárias nas Instituições Financeiras nacionais de grande porte Revista de Administração de Roraima v6 n1 p4770 2016 GOMES R M M P Análise de Crédito a Instituições Financeiras A metodologia CAMELS Dissertação Mestrado em Economia e Gestão Instituto Superior de Economia e Gestão Universidade Técnica de Lisboa Lisboa 2012 GRAVELLE T LI F Measuring systemic importance of financial institutions An extreme value theory approach Journal of Banking and Finance v 37 n7 p21962209 2013 GREENE W H Econometric analysis 5ed New Jersey Prentice Hall 2002 959p GRIGORIAN D A MANOLE V Determinants of Commercial Bank Performance in Transition An application of Data envelopment analysis Comparative Economic Studies v48 n3 p497522 2006 GUERREIRO A S Analise da Eficiência de Empresas de Comercio Eletronico Usando Tecnicas da Analise Envoltoria de Dados Dissertação Mestrado em Engenharia Industrial Pontifica Universidade Católica do Rio de Janeiro Rio de Janeiro 2006 HAIR J F et al Análise Multivariada de Dados 5 ed Porto Alegre Bookman 2005 INTERNATIONAL MONETARY FUND Guia de Compilación dos Indicadores de Solidez Financiera Washington FMI 2006 JEELANBASHA V JEELANBASHA V Comparative performance analysis of selected banks using CAMEL model Journal of Business and Management v20 n6 p8793 2018 JIANG C YAO S FENG G Bank ownership privatization and performance Evidence from a transition country Journal of Banking and Finance v37 n9 p33643372 2013 JUBRAN A J Modelo de Analise de Eficiência na Administracao Publica Estudo aplicado as prefeituras Brasileiras usando a analise envoltoria de dados Tese Doutorado em Engenharia Escola Politécnica Universidade São Paulo São Paulo 2006 KHAN M N et al Efficiency measurement of Islamic and conventional banks in Saudi Arabia an empirical and comparative analysis AlShajarah Journal of the International Institute of Islamic Thought and Civilization ISTAC dec p 11134 2018 KISGEN D J The impact of credit ratings on corporate behavior Evidence from Moodys adjustments Journal of Corporate Finance v 58 n C p 567582 2019 KOCH T W MACDONALD S S Bank Management 7 ed Boston SouthWestern Cengage Learning 2014 KORONTAI J N Governança corporativa dos bancos e sua relação com indicadores de desempenho e risco Dissertação Mestrado em Contabilidade Universidade do Paraná Curitiba 2016 118 KUMBHAKAR S C GHOSH S MCGUCKIN J T A generalized production frontier approach for estimating determinants of inefficiency in US Dairy farms Journal of Business and Economic Statitics v9 n1 p279286 1991 KUMBHAKAR S C LIEN G HARDAKER J B Technical efficiency in competing panel data models a study of Norwegian grain farming Journal of Productivity Analysis v 41 p 321337 2012 KUMBHAKAR S C LOVELL C A K Stochastic Frontier Analysis 2ed Cambridge University Press Cambridge 2003 KWAN S EISENBEIS R Bank Risk Capitalization and Operating Efficiency Journal of Financial Services Research v 12 n 2 p 117131 1997 LENSINK R MEESTERS A NAABORG I Bank efficiency and foreign ownership Do good institutions matter Journal of Banking and Finance v32 n5 p834844 2008 LOZANOVIVAS A PASIOURAS F The impact of nontraditional activities on the estimation of bank efficiency International evidence Journal of Banking Finance v 34 n 7 p 14361449 2010 LUO Y BI G LIANG L Inputoutput indicator selection for DEA efficiency evaluation An empirical study of Chinese commercial banks Expert Systems With Applications v39 n1 p11181123 2012 MACORIS L S SALGADO JUNIOR A P FALSARELLA JUNIOR E The different approaches of banking efficiency a MetaAnalysis International Conference on Data Envelopment Analysis 13th edition 2015 Braunschweig Germany MAGRO C B D MICHELS A SILVA T P Análise da eficiência no desempenho financeiro das cooperativas de crédito brasileiras Revista Brasileira de Gestão e Desenvolvimento Regional v13 n2 p73102 2017 MALHOTRA N K ROCHA I LAUDISIO M C ALTHEMAN E BORGES F M Introdução a Pesquisa de Marketing São Paulo Pearson 2005 MALIKOV E HUMBHAKAR S C SUN Y Varying coefficient panel data model in the presence ofendogenous selectivity and fixed effects Journal of Econometrics v 190 p 233 251 2016 MARCUSSO M A Conflito entre as abordagens de rentabilidade intermediação e produção em bancos corporate brasileiros uma análise DEA dois estágios entre 1996 e 2015 Dissertação Mestrado em Administração das Organizações Faculdade de Economia Administração e Contabilidade de Ribeirão Preto Universidade de São Paulo Ribeirão Preto 2017 MARTEL M VAN RIXTEL A MOTA E G Business models of international banks in the wake of the 20072009 global financial crisis Banco de España Revista de Estabilidad Financiera n 22 p 99121 2012 119 MAUDOS J FERNANDEZ DE GUEVARA J Factors explaining the interest margin in the banking sectors of the European Union Journal of Banking Finance v 28 N 9 p 2259 2281 sept 2004 MEEUSEN W VAN DEN BROECK J Efficiency Estimation from CobbDouglas Production Functions with Composed Error International Economic Review v18 n2 p435444 1977 MERGAERTS F VENNET R V Business models and bank performance A longterm perspective Journal of Financial Stability v 22 n C p 5775 2016 MERILÄINEN JM JUNTTILA J The relationship between credit ratings and asset liquidity Evidence from Western European banks Journal of International Money and Finance v 108 november 2020 p 121 2020 MILI M et al Determinants of the capital adequacy ratio of foreign banks subsidiaries The role of interbank market and regulation Research in International Business and Finance v 42 p 442453 2017 MILLON M H THAKOR A V Moral Hazard and Information Sharing A Model of Financial Information Gathering Agencies The Journal of Finance v 40 p 14031422 1985 MOSTAFA M Modeling the Efficiency of Top Arab Banks A DEAneural network approach Expert Systems with Applications v 6 n1 p 309320 2009 NETER J WASSERMAN W Applied linear statistical models1ed Homewood III Richard D Irwin 1974 NIGMONOV A Bank Performance and Efficiency in Uzbekistan Eurasian Journal of Business and Economics v3 n5 p125 2010 ODECK J MOLDE S B A metaanalysis of DEA and SFA studies of the technical efficiency of seaports A comparison of fixed and randomeffects regression models Transportation Research Part A Policy and Practice v46 n10 p15741585 2012 OTERO L et al What determines efficiency in MENA banks Journal of Business Research v 112 p 331341 2020 OTHMAN N ABDULMAJID M ABDULRAHMAN A Partnership financing and bank efficiency PacificBasin Finance Journal v 46 n PA p 113 2017 PARADI J C SCHAFFNIT C Commercial Branch Performance Evaluation and Results Communication in a Canadian Bank A DEA application European Journal of Operational Research v156 n3 p719735 2004 PARADI J C ZHU H EDELSTEIN B Identifying Managerial Groups in a Large Canadian Bank Branch Network with a DEA Approach European Journal of Operational Research v219 n1 p178187 2012 PARETO V The New Theories of Economics Journal of Political Economy v5 n 4 p 485502 1897 120 PENNACCHI G G SANTOS J A C Why do banks target ROE Journal of Financial Stability v 54 june 2021 in press 2021 PERICO A E REBELATTO D A DO N SANTANA N B Eficiência Bancária Os maiores bancos são os mais eficientes Uma análise por envoltória de dados Gestão Produção v15 n2 p421431 2008 PEYDRÓ JL Ç POLO A SETTE E Monetary policy at work Security and credit application registers evidence Journal of Financial Economics in press doiorg101016jjfineco202101008 2021 PILLE P PARADI J C Financial performance analysis of Ontario Canada Credit Unions An application of DEA in the regulatory environment European Journal of Operational Research v139 p339350 2002 PIMENTA A C Contribuição para a análise do desempenho das instituições financeiras a partir dos indicadores da carteira de crédito Dissertação Mestrado em Administração de Empresas Pontifícia Universidade Católica São Paulo 2014 PONTES T T S Determinantes da eficiência técnica no setor de supermercados brasileiro uma análise utilizando métodos de fronteira de produção Tese Doutorado em Administração das Organizações Faculdade de Economia Administração e Contabilidade de Ribeirão Preto Universidade de São Paulo Ribeirão Preto 2017 PRUTEANUPODPIERA A WEILL L SCHOBERT F Banking Competition and Efficiency A MicroData Analysis on the Czech Banking Industry Comparative Economic Studies v 50 n 2 p 253273 2008 REIFSCHNEIDER D STEVENSON R Systematic departures from the frontier A framework for the analysis of firm inefficiency International Economic Review v32 n3 p715723 1991 ROENGPITYA R TARASHEV N TSATSARONIS K Bank business models BIS Quarterly Review dec 2014 p 5565 2014 ROMAN A SARGU A C Analysing the financial soundness of the commercial banks in Romania An approach based on the CAMELS framework Procedia Economics and Finance v6 n1 p703712 2013 ROSA P GARTNER I Financial distress em bancos brasileiros Um modelo de alerta antecipado Revista Contabilidade Finanças v29 n77 p312331 2018 ROSMAN R WAHAB N A ZAINOL Z Efficiency of Islamic banks during the financial crisis An analysis of Middle Eastern and Asian countries PacificBasin Finance Journal v 28 p 7690 2014 ROSSI S P SCHWAIGER M S WINKLER G How loan portfolio diversification affects risk efficiency and capitalization A managerial behavior model for Austrian banks Journal of Banking and Finance v33 n12 p22182226 2009 SAEED M et al The intertemporal relationship between risk capital and efficiency The case of Islamic and conventional banks PacificBasin Finance Journal v 62 n C p 122 2020 121 SAHAJWALA VAN DEN BERGH P Supervisory Risk Assessment and Early Warning Systems Bank for International Settlements Basel Switzerland BIS Press Office 2000 SAMPIERI R H COLLADO C F LUCIO P B Metodologia de pesquisa 3 ed São Paulo MacGrawHill 2006 SARMIENTO M GALÁN J E The influence of risktaking on bank efficiency Evidence from Colombia Emerging Markets Review v 32 n C p 5273 2017 SHADDADY A MOORE T Investigation of the effects of financial regulation and supervision on bank stability The application of CAMELSDEA to quantile regressions Journal of International Financial Markets Institutions and Money v 58 2018 SHAMSHUR A WEILL L Does bank efficiency influence the cost of credit Journal of Banking Finance v 105 n C p 6273 2019 SILVA T P et al Financial and economic performance of major Brazilian credit cooperatives Contaduría y Administración v 62 n 5 p 14421459 2017 SIRIOPOULOS C TZIOGKIDIS P How do Greek banking institutions react after significant events A DEA approach Omega v38 n5 p294308 2010 STAIKOURAS C MAMATZAKIS E KOUTSOMANOLIFILIPPAKI A Cost efficiency of the banking industry in the South Eastern European region Journal of International Financial Markets v18 n5 p483497 2008 STAUB R B SOUZA G TABAK B M Evolution of Bank Efficiency in Brazil A DEA approach European Journal of Operational Research v202 n1 p204213 2010 SUFIAN F HABIBULLAH M S Bank specific and macroeconomic determinants of bank profitability Empirical evidence from the China banking sector Frontiers of Economics in China v 4 p 274291 2009 SUFIAN F HABIBULLAH M S Globalizations and bank performance in China Research in International Business and Finance v 26 n 2 may p 221239 2012 TABAK B M FAZIO D M CAJUEIRO D O The relationship between banking market competition and risktaking Do size and capitalization matter Journal of Banking and Finance v36 n12 p33663381 2012 TABAK B TECLES P Estimating a bayesian stochastic frontier for the Indian banking system International Journal of Production Economics v125 p96110 2010 TAN Y The impacts of risk and competition on bank profitability in China Journal of International Financial Markets Institutions and Money v 40 p 85110 2016 TAN Y FLOROS C Risk profitability and competition evidence from the Chinese banking industry Journal of Developing Areas v 48 n 3 p 303319 2014 THANASSOULIS E Introduction to the Theory and Application of Data Envelopment Analysis A Foundation Text with Integrated Software Springer 2003 122 THANASSOULIS E BOUSSOFIANE A DYSON R G A Comparison of Data Envelopment Analysis and Ratio Analysis as Tools for Performance Assessment Omega International Journal of Management Science v24 n3 p229244 1996 THEODORIDIS A M ANWAR M A comparison of DEA and SFA methods a case study of farm households in Bangladesh Journal of Developing Areas v 45 N 1 p 95110 2011 VERGARA S C Projetos e Relatórios de Pesquisa em Administração São Paulo Atlas 2006 WANKE P Determinants of scale efficiency in the Brazilian thirdparty logistics industry from 2001 to 2009 Brazilian Administration Review v9 n1 p 6687 2012 WU JH HSIA TL HENG M S H Core capabilities for exploiting electronic banking Journal of Electronic Commerce Research v 7 n 2 p 112122 2006 YANG J B et al Integrated Bank Performance Assessment and Management Planning Using Hybrid Minimax Reference Point DEA approach European Journal of Operational Research v207 n3 p15061518 2010 YANG X et al Systemic importance of financial institutions A complex network perspective Physica A Statistical Mechanics and its Applications v545 n1 p111 2020 YANG Z Bank Branch Operating Efficiency A DEA Approach International Multi Conference of engineers and Computer Scientists Vols I and II Anais 2009 YIN H The impact of competition and bank market regulation on banks cost efficiency Journal of Multinational Financial Management 100677 in press 2021 YIN H YANG J LU X Bank globalization and efficiency Host and homecountry effects Research in International Business and Finance v 54 dec 2020 ZANINI A Regulação econômica no setor elétrico brasileiro uma metodologia para definição de fronteiras de eficiência e cálculo do fator x para empresas distribuidoras de energia elétrica Tese Doutorado em Engenharia Elétrica Pontifícia Universidade Católica do Rio de Janeiro Rio de Janeiro 2004 ZHU J Imprecise data envelopment analysis IDEA A review and improvement with an application European Journal of Operational Research v144 n3 p513529 2003 Electronic copy available at httpssrncomabstract1350544 Electronic copy available at httpssrncomabstract1350544 Assessing Bank Performance with Operational Research and Artificial Intelligence Techniques A Survey Meryem DuygunFethi School of Management University of Leicester Fotios Pasiouras School of Management University of Bath University of Bath School of Management Working Paper Series 200902 This working paper is produced for discussion purposes only The papers are expected to be published in due course in revised form and should not be quoted without the authors permission Electronic copy available at httpsssrncomabstract1350544 Electronic copy available at httpssrncomabstract1350544 Electronic copy available at httpssrncomabstract1350544 2 University of Bath School of Management Working Paper Series School of Management Claverton Down Bath BA2 7AY United Kingdom Tel 44 1225 826742 Fax 44 1225 826473 httpwwwbathacukmanagementresearchpapershtm 2009 200901 Androniki Apostolakou Gregory Jackson Corporate Social Responsibility in Western Europe An Institutional Mirror or Substitute 200902 Meryem Duygun Fethi Fotios Pasiouras Assessing Bank Performance with Operational Research and Artificial Intelligence Techniques A Survey Electronic copy available at httpsssrncomabstract1350544 3 Assessing Bank Performance with Operational Research and Artificial Intelligence Techniques A Survey Meryem DuygunFethi1 Fotios Pasiouras2 1School of Management University of Leicester UK 2School of Management University of Bath UK Abstract This paper presents a comprehensive review of 179 studies which employ operational research OR and Artificial Intelligence AI techniques in the assessment of bank performance We first discuss numerous applications of data envelopment analysis which is the most widely applied OR technique in the field Then we discuss applications of other techniques such as neural networks support vector machines and multicriteria decision aid that have also been used in recent years in bank failure prediction studies and the assessment of bank creditworthiness and underperformance Keywords Artificial Intelligence Banks Data Envelopment Analysis Operational Research Literature review Copyright The Authors 2009 Author for correspondence Tel 44 0 116 252 5328 Fax 44 0 116 252 5515 Emails mfethileacuk M DuygunFethi fpasiourasbathacuk F Pasiouras Electronic copy available at httpsssrncomabstract1350544 4 1 Introduction Banks play a central role in the economy They keep the savings of the public and finance the development of business and trade Furthermore numerous studies argue that the efficiency of financial intermediation affects economic growth while others indicate that bank insolvencies can result in systemic crises which have adverse consequences for the economy as a whole1 Thus the performance of banks has been an issue of major interest for various stakeholders such as depositors regulators customers and investors While bank performance has been traditionally evaluated on the basis of financial ratios advances in operational research OR and artificial intelligence AI have resulted in a shift towards the use of such stateoftheart techniques Of course this is not surprising since OR has been extensively applied to finance during the last half century Board et al 2003 This paper presents a comprehensive review of the use of OR and AI techniques in the assessment of bank performance The rest of the paper is structured as follows Section 2 positions the survey within the existing literature and discusses our framework Section 3 discusses applications of data envelopment analysis DEA in the estimation of bank efficiency and productivity growth Section 4 presents applications of other OR and AI techniques in the prediction of bank failure and the assessment of bank creditworthiness and underperformance Section 5 summarizes our conclusions 2 Scopus and framework There are several interesting reviews that are related to our survey For example Cook and Seiford 2009 review the methodological developments of DEA over the last thirty years However they do not discuss applications of DEA Zhou and Poh 2008 provide a recent survey of DEA applications but they focus on energy and environmental studies Dimitras et al 1996 discuss applications of various techniques in the prediction of business failures but they focus on industrial firms Smith and Gupta 2000 provide a discussion of the application of neural networks in business problems Board et al 2003 survey OR applications in the financial markets Thus the above surveys are either quite general or they do not focus on applications in banking 1 See Levine 2005 for a discussion of the theoretical and empirical discussion of the literature on finance and growth See Caprio and Klingebiel 2003 for banking crises and the associated costs Electronic copy available at httpsssrncomabstract1350544 5 Berger and Humphrey 1997 review studies that examine the efficiency of financial institutions However their coverage is limited to efficient frontier techniques eg DEA stochastic frontier analysis Furthermore the survey is now more than ten years old and since that time numerous papers have been published Berger 2007 discusses more recent applications of frontier techniques but his survey focuses only on studies that provide international comparisons of bank efficiency Finally Ravi Kumar and Ravi 2008 discuss applications of statistical and AI techniques in bankruptcy prediction The applications that they survey were published until 2005 and although a few of them focus on the banking sector most of the studies deal with nonfinancial firms We differentiate our review from the above surveys by discussing applications of OR and AI techniques over the period 19982008 while focusing on bank performance2 We searched for papers in Scopus which is considered to be one of the largest abstract and citation databases We consider only journal articles and we do not include working papers monographs dissertations or other publication outcomes Furthermore our search is limited to articles written in English We use a combination of various keywords such as bank efficiency bank and data envelopment analysis bank performance bank and neural networks bank and artificial intelligence bank and operational or operations research A few additional studies were identified from crossreferencing and were manually collected We reviewed a total of 179 studies DEA is by far the most commonly used ORAI technique in assessing bank performance and we identified 136 studies that use DEAlike techniques to estimate various measures of bank efficiency and productivity growth and 28 studies that provide similar estimates at the branch level3 We also identified 15 studies that use classification techniques such as neural networks support vector machines multicriteria decision aid decision trees nearest neighbours to predict bank failure or assess bank creditworthiness and bank underperformance These studies were published in a total of 67 journals however around 56 of them appeared in just eleven journals As shown in Table 1 the most 2 Some of the studies were still in press at the time of the writing of this review but we included them in our discussion as they were already publicly available since 2008 ie 2008 DOI 3 Our survey focuses on studies that examine banking institutions as a whole however we also discuss studies on branch efficiency in section 327 as one could argue that the efficiency of individual branches can influence the performance of banks as a whole Electronic copy available at httpsssrncomabstract1350544 6 frequent sources of publication are the European Journal of Operational Research 18 and the Journal of Banking and Finance 15 followed by Applied Financial Economics 13 Applied Economics 9 Expert Systems with Applications 9 and the Journal of Productivity Analysis 9 In the sections that follow we discuss various issues surrounding these studies while additional information is available in Appendices I to III Insert Table 1 Around Here 3 DEA and bankefficiency DEA is a mathematical programming technique for the development of production frontiers and the measurement of efficiency relative to these frontiers Charnes et al 1978 The bestpractice production frontier for a sample of banks is constructed through a piecewise linear combination of actual inputoutput correspondence set that envelops the inputoutput correspondence of all banks in the sample Thanassoulis 2001 Each bank is assigned an efficiency score between 0 and 1 The scores are only relative to the banks in the sample with higher scores indicating a more efficient bank One of the wellknown advantages of DEA is that it works relatively well with small samples Other advantages of DEA are that it does not require any assumptions to be made about the distribution of inefficiency and it does not require a particular functional form on the data in determining the most efficiency banks However DEA is also subject to few limitations Two of the best known shortcomings are that DEA assumes data to be free of measurement error and that it is sensitive to outliers Coelli et al 2005 also point out that i having few observations and many inputs andor outputs will result in many firms appearing on the DEA frontier ii treating inputsoutputs as homogenous commodities when they are heterogeneous may bias the results iii not accounting for differences in the environment may give misleading results iv standard DEA does not control for multiperiod optimization or risk managerial decision making Our survey shows that recent DEA studies have examined almost all the banking sectors around the world A few recent studies provide crosscountry evidence Most of them examine banks from the large EU banking sectors Pastor 2002 Casu and Molyneux 2003 Beccalli et al 2006 LozanoVivas et al 2002 Electronic copy available at httpsssrncomabstract1350544 7 examine ten EU countries Bergendahl 1998 focuses on Nordic countries while Pasiouras 2008a examines an international dataset 31 Methodological issues 311 Efficiency measures Most of the studies focus on the technical efficiency of banks eg LozanoVivas et al 2002 Drake et al 2006 Pasiouras 2008ab This efficiency measure indicates whether a bank uses the minimum quantity of inputs to produce a given quantity of outputs or maximizes the output quantity given a certain quantity of inputs However when price data for the inputs andor outputs are available one can also estimate cost andor profit efficiency measures4 Cost efficiency is the product of technical efficiency and allocative efficiency The latter refers to the ability of a bank to use the optimum mix of inputs given their respective prices Consequently cost efficiency shows the ability of a bank to provide services without wasting resources as a result of technical or allocative inefficiency Appendix I shows that around 35 studies present measures of DEA cost efficiency eg TortosaAusina 2002abc Isik and Hassan 2002 2003a Pastor and Serrano 2006 propose the decomposition of cost inefficiency into composition inefficiency and intraspecialisation inefficiency The first component indicates the part of inefficiency due to the composition of specialisations of the banks in each banking sector The second component reveals the inefficient use of resources within each of the specialisation selected Prior 2003 also deviates from the above studies by calculating measures of short and longrun cost inefficiency as well as capacity inefficiency for Spanish banks The first refers to the case that a subset of inputs are fixed and impossible to modify in the shortrun Longrun inefficiency estimates are obtained under the assumption that inputs are variable and under the control of the company Finally capacity inefficiency obtained by the ratio of longrun to shortrun inefficiency refers to excess in costs as a result of inappropriate level in fixed inputs Similar concepts along with an application in the Indian banking sector are discussed in Sahoo and Tone 2008 4 One can also estimate revenue efficiency which is similar to profit efficiency In both cases both inputs and output prices are required The difference is that in the former measure the aim is to maximize revenues rather than profits ie revenues minus costs We are no aware of DEA studies focusing on revenue efficiency so we do not discuss this issue further Readers interesting in revenue efficiency could see Coelli et al 2005 for further details Electronic copy available at httpsssrncomabstract1350544 8 Estimations of profit efficiency with DEA are rather limited in the literature One potential reason is the difficulty in collecting reliable and transparent information for output prices Furthermore the decomposition of profit efficiency into technical and allocative efficiency is not straightforward Coelli et al 2005 Fare et al 2004 propose the solution of two sets of linear programmes In the first a profit maximising DEA is solved to measure profit efficiency In the second DEA problem technical efficiency is measured on the basis of a directional distance function that allows the simultaneous adjustment of inputs and outputs Kirkwood and Nahm 2006 also estimate profit efficiency although they use input prices only Therefore in a sense they calculate a measure of efficiency that is similar to Berger and Mesters 1997 alternative profit efficiency which is commonly used in the stochastic frontier analysis literature The studies of Maudos and Pastor 2003 and Ariff and Can 2008 provide estimates of both standard and alternative profit efficiency Finally around 30 of the studies obtain estimates of total factor productivity TFP growth eg Sathye 2002 Casu et al 20045 This measure is usually decomposed further into technological change ie shift in the best practice frontier and technical efficiency change Furthermore in most cases technical efficiency change is disaggregated into pure technical efficiency change and scale efficiency change under the variable returns to scale assumption discussed below 312 Constant vs Variable Returns to scale DEA can be implemented by assuming either constant returns to scale CRS or variable returns to scale VRS In their seminal study Charnes et al 1978 proposed a model that had an input orientation and assumed CRS This model returns a score that indicates the overall technical efficiency OTE of each bank Banker et al 1984 suggested the use of variable returns to scale VRS that decomposes OTE into product of two components pure technical efficiency PTE and scale efficiency SE The former relates to the ability of managers to utilize firms given resources while the latter refers to exploiting scale economies by operating at a point where the production frontier exhibits CRS 5 Almost all the studies use the DEAlike Malmquist index Zen and Baldan 2008 use the Luenberger Indicator that is a generalization of the Malmquist Index Electronic copy available at httpsssrncomabstract1350544 9 In most of the recent papers DEA models are estimated using the assumption of VRS while arguing that CRS is only appropriate when all firms are operating at an optimal scale6 Nevertheless other studies argue in favour of CRS rather than VRS For example Noulas 1997 points out that the assumption of CRS allows the comparison between small and large banks He claims that in a sample where a few large banks are present the use of VRS framework raises the possibility that these large banks will appear as being efficient for the simple reason that there are no truly efficient banks Berg et al 1991 Avkiran 1999 also mentions that under VRS each unit is compared only against other units of similar size instead of against all units Hence the assumption of VRS may be more suitable for large samples Soteriou and Zenios 1999a argue that caution is necessary when using the VRS formulation First because the model orientation ie input minimization or output maximization becomes important Second because the use of the weights restriction in the VRS assessment may lead to some other problematic results Allen 1997 Consequently many studies report the results obtained under both CRS and VRS assumptions eg Canhoto and Dermine 2003 Casu and Molyneux 2003 313 Outputinput orientation Technical Efficiency can be estimated under either an inputoriented or output oriented approach7 As Coelli et al 2005 point out the inputoriented technical efficiency measures address the question By how much can input quantities be proportionally reduced without changing the output quantities produced p 137 In contrast the outputoriented measures of technical efficiency address the question By how much can output quantities be proportionally expanded without altering the input quantities used p 137 By far studies in banking obtain efficiency estimates under the inputoriented approach8 This is most likely due to the assumption that bank managers have higher 6 Reasons that may not allow a firm to operate at optimal scale include among others imperfect competition government regulations constrains on finance etc Coelli et al 2005 7 If price data are available then under the under cost minimization objective ie cost efficiency one obtains inputoriented measures of technical efficiency and inputmix allocate efficiency Revenue maximization ie revenue efficiency results is outputoriented technical efficiency and outputmix allocative efficiency measures In the case of profit maximization profit efficiency technical efficiency can be obtained under either the input or outputoriented assumption Coelli et al 2005 8 Additional information is available in Appendix I In contrast to efficiency estimates productivity measures are in several cases obtained using an outputoriented Malmquist index However as discussed in Coelli et al 2005 p 80 although the values of the components and consequently their Electronic copy available at httpsssrncomabstract1350544 10 control over inputs eg personnel expenses rather than outputs eg loans income etc However there are also some studies that adopt the outputoriented approach eg Ataullah et al 2004 Ataullah and Le 2006 or report the results from both eg Casu and Molyneux 2003 Beccalli et al 2006 It should be mentioned that the inputoriented and outputoriented measures always provide the same value under CRS but they are unequal when VRS is assumed However Coelli et al 2005 mention that since linear programming does not suffer from statistical problems the choice of an appropriate orientation is not as important as in the case of econometric approaches Furthermore in many instances the choice of orientation has only a minor influence upon the scores obtained Coelli and Perelman 1996 314 Selection of Inputs and Outputs There is an ongoing discussion in the banking literature regarding the proper definition of inputs and outputs In the words of Bergendahl 1998 There have been almost as many assumptions of inputs and outputs as there have been applications of DEA p 235 Berger and Humphrey 1997 identify two main approaches for the selection of inputs and outputs These are the production approach and the intermediation approach The first assumes that banks produce loans and deposits account services using labour and capital as inputs and that the number and type of transactions or documents processed measure outputs The second approach perceives banks as financial intermediaries between savers and investors Berger and Humphrey 1997 argue that neither of these two approaches is perfect because they cannot fully capture the dual role of financial institutions as providers of transactionsdocument processing services and also being financial intermediaries They point out that the production approach may be somewhat better for evaluating the efficiencies of bank branches and the intermediation approach may be more appropriate for evaluating financial institutions as a whole Furthermore there are difficulties in collecting the detailed transaction flow information required in the production approach As a result the intermediation approach is the one favoured in the literature However there is a controversy even within this approach concerning the role of deposits Berger and Humphrey 1997 Consequently some studies use only contribution to overall productivity change may differ the overall TFP change measure will be the same regardless of the orientation ie input or output that is imposed Electronic copy available at httpsssrncomabstract1350544 11 earning assets as outputs a selection that is in line with the asset approach of Sealey and Lindley 1977 while others consider deposits as an additional output a selection that is more closely related to the socalled valueadded approach We find around eighty applications in bank efficiency where the monetary value of deposits is part of the input vector and twenty applications where deposits are part of the output vector9 Around thirty studies use interest expenses as an input without using the stock of deposits eg Sathey 2002 Weill 2004 In another eight applications the stock of deposits is used as an output and the interest expense paid on deposits constitutes an input eg Maudos et al 2002 Saha and Ravisankar 2000 Chen et al 2005 Furthermore around five studies use time deposits and saving deposits as input and demand deposits as output eg Bauer et al 1998 Gilbert and Wilson 1998 Sathye 2001 Finally in a few applications the deposits are included as both an input and an output eg TortosaAusina 2002a More recently some studies have adopted another variation of the intermediation approach This is the socalled profitoriented or operating approach which defines revenue components eg interest income noninterest income etc as outputs and cost components eg personnel expenses interest expenses etc as inputs10 Drake et al 2006 mention that from the perspective of an inputoriented DEA relative efficiency analysis the more efficient units will be better at minimizing the various costs incurred in generating the various revenue streams and consequently better at maximizing profits p 1451 They also argue that this approach can be more appropriate in capturing the diversity of strategic responses by financial firms in the face of dynamic changes in competitive and environmental conditions Furthermore Luo 2003 calculates a measure of marketability efficiency in an attempt to capture the value of the bank in the stock market In this case revenue and profits are considered inputs whereas market value earnings per share and stock price are outputs As discussed before with the exception of deposits there is a general agreement about the main categories of inputs and outputs however this does not 9 This discussion refers to banks as a whole and not branches The number of applications does not match the number of studies as in several cases there are numerous models developed in each study Furthermore in some cases deposits or non interest expenses are not considered in the analysis In one case Prior 2003 the output was the number of current and saving accounts rather than the monetary value of deposits ie stock 10 See Chu and Lim 1998 Avkiran 1999 Sturm and Williams 2004 Das and Ghosh 2006 Drake et al 2006 Ataullah and Le 2006 Pasiouras 2008b Pasiouras et al 2008 Electronic copy available at httpsssrncomabstract1350544 12 necessarily imply that there is consistency with respect to the specific inputsoutputs used in various studies For instance the traditional inputs are fixed assets personnel11 and in many cases deposits eg Isik and Hassan 2002 Maudos and Pastor 2003 Casu and Girardone 2004 Havrylchyk 2006 However some studies use branches eg Chen 2001 loan loss provisions eg Drake 2001 Drake et al 2006 Pasiouras 2008b and equity eg Chu and Lim 1998 Mukherjee et al 2001 Sturm and Williams 2004 Pasiouras 2008a as additional or alternative inputs Chen 2001 disaggregates deposits into current deposits and time deposits while Das and Ghosh 2006 use demand savings and fixed deposits Casu and Girardone 2006 and Beccalli et al 2006 use total costs as a single input while Casu and Molyneux 2003 use two inputs namely total costs and total deposits ie customers and short term funding Several studies use two outputs usually loans and other earning assets eg Casu and Molyneux 2003 Casu and Girardone 2004 200612 However some studies disaggregate loans into various categories such as housing loans and other loans eg Sturm and Williams 2004 real estate loans commercial loans and personal loans eg Mukherjee et al 2001 Fare et al 2004 or shortterm and long term loans Isik and Hassan 2002 Others disaggregate other earning assets into investments and liquid assets Tsionas et al 2003 or investment in government securities and investments in public and private enterprises eg Chen 2001 Canhoto and Dermine 2003 use the number of branches as an additional output under the assumption that it represents an additional value for retail customers Finally recent studies include noninterest income or offbalancesheet items as additional outputs eg Isik and Hassan 2002 2003a Sturm and Williams 2004 TortosaAusina 2003 Havrylchyk 2006 Pasiouras 2008b13 Finally Halkos and Salamouris 2004 propose an approach that deviates from the above literature They use an output vector that consists of five financial ratios and no inputs Their underlying hypothesis is that inputs are considered similar and equal 11 Some studies use the number of personnel eg Drake 1999 Sathye 2001 Maudos and Pastor 2003 Pasiouras 2008b However due to data unavailability others rely on personnel expenses eg Bergendahl 1998 TortosaAusina 2002 LozanoVivas et al 2002 Drake et al 2006 12 Other earning assets normally include various items such as government securities investment securities trading securities other securities equity investments and other investments 13 Isik and Hassan 2003a mention among others that ignoring such nontraditional outputs may penalize banks that are heavily involved in these activities The reason is that while the resources used to produce these nontraditional outputs are part of the input vector the outputs generated using these inputs are not included in the output vector Electronic copy available at httpsssrncomabstract1350544 13 for all banks as they operate in the same markets for money and service The comparison with standard DEA models developed under either CRS or VRS assumptions shows that the DEA ratio model has the highest discrimination power 315 Adjusting for the environment Coelli et al 2005 discuss four approaches that can be used to incorporate environmental variables in DEA applications14 The first method by Banker and Morey 1986 requires the environmental variables to be ordered from the least to the most harmful ones for efficiency Then the efficiency of a given firm is compared with those firms in the sample that have a value of the environmental variable which is less than or equal to the given firm This ensures that banks are not compared with peers operating in a more favourable environment The second method by Charnes et al 1981 requires the investigator to i divide the sample into subsamples and solve DEA problems for each subsample ii project all observed data points into their prospective frontiers and iii solve a single DEA using the projected points and assess any difference in the mean efficiency of the two subsamples According to Coelli et al 2005 the following two problems are common in both methods i by splitting up the sample they reduce the comparison set and ii only one environmental variable can be considered in each case thereby limiting the scope of the analysis Under the third method the environmental variables are included directly into the DEA problem as nondiscretionary inputs if it is believed to have a positive effect on efficiency or outputs if they have a negative effect on efficiency The disadvantage of this approach is that one must know a priori the direction of the influence a shortcoming that is also applicable in the case of the first method Alternatively the environmental variables can be included as nondiscretionary neutral variables using an equality form The shortcoming of this approach is that it can reduce the reference set for each firm Recent applications of the above approaches in banking can be found in Pastor 1999 and LozanoVivas et al 2001 2002 14 According to Coelli et al 2005 environmental variables can include ownership differences location characteristics labour union power and government regulations and in a sense any factors that can influence the efficiency of the firm without being traditional inputs or under the control of managers Electronic copy available at httpsssrncomabstract1350544 14 The fourth method that is discussed in Coelli et al 2005 is the twostage approach This involves a DEA problem with traditional inputs and outputs in the first stage In the second stage the efficiency scores obtained are regressed on the environmental variables While this approach has been frequently used in the banking literature with numerous applications it fails to adjust the efficiency measures Therefore it is more suitable when the objective is to examine the correlations of efficiency with various factors see Section 321 rather than provide the basis for absolute comparisons across different environments Finally Pastor 2002 and Drake et al 2006 adjust the bank efficiency scores for risk and external environmental factors respectively using a multistage DEA see Fried et al 1999 2002 In this case the application starts with the estimation of a DEA model with traditional inputs and outputs Then using the slacks from the DEA model they quantify the effect of the operating environment and adjust the initial dataset inputs andor outputs Finally they rerun the initial DEA model using the adjusted data 32 Topics of interest 321 Determinants of efficiency Several studies attempt to investigate the factors that influence the efficiency of banks Some studies examine only bankspecific factors and others examine both bankspecific attributes and environmental determinants Commonly found bankspecific factors are size profitability capitalisation loans to assets Casu and Molyneux 2003 Casu and Girardone 2004 Ataullah and Le 2006 Ariff and Can 2008 Isik and Hassan 2003a examine additional characteristics such as the educational profile of bank personnel and CEOChairman affiliation Some studies examine whether age is related to efficiency eg Isik and Hassan 2003a Canhoto and Dermine 2003 Isik 2008 Furthermore Isik and Hassan 2002 Casu and Girardone 2004 and Pasiouras 2008b examine the international presence of Turkish Italian and Greek banks respectively Countryspecific factors include market concentration presence of foreign banks ratio of private investments to GDP fiscal deficits to GDP GDP growth Hauner 2005 Ataullah and Le 2006 More recently Pasiouras 2008a examined the relationship between technical efficiency and regulations related to capital Electronic copy available at httpsssrncomabstract1350544 15 adequacy private monitoring banks activities deposit insurance schemes supervisory power and bank entry into the industry These studies use the twostage approach discussed above First they use DEA to obtain efficiency estimates Then in a second stage the DEA scores are regressed on a number of explanatory variables using Tobit eg Hauner 2005 OLS Ataullah and Le 2006 GMM Ataullah and Le 2006 or GLS Isik and Hassan 2003a regression The rationale to use Tobit lies on the fact that the efficiency scores are bounded between 0 and 1 and hence noncensored estimates will be biased Ataullah and Le 2006 among others mention that it is not necessary to use Tobit as long as the efficiency scores can be transformed by taking the natural logarithm of efficiency score 1efficiency score Whether or not censored regression is used the most important argument comes from Simar and Wilson 2007 who pont out that the covariates in the second step regression are obviously correlated with the one side error terms from the first step as otherwise there would be no need for the second stepregression Furthermore the covariates in the secondstep are likely to be highly correlated with the covariates in the firststep This means that the errors and the covariates in the first step cannot be independent Thus Simar and Wilson conclude that the likelihood that is maximized is not the correct one unless one takes account of the correlation structure Casu and Molyneux 2003 present an early attempt to account for the problems that can emerge in the twostage DEA method using a bootstrap approach However Simar and Wilson 2007 mention that the simple bootstrap is not enough to overcome the drawbacks pointed out above They propose an algorithm that uses a double bootstrap procedure and they present an application to US banks Brissimis et al 2008 adopt this approach to examine the determinants of efficiency in the new EU banking sectors 322 Stock returns and efficiency Following the work of Ball and Brown 1968 the relationship between stock returns and publicly available information has attracted considerable attention in the accounting and finance literature Whilst most of the studies examine whether earnings reflect some of the information in stock prices recent research has however shifted towards the use of additional data such as accruals revenues economic value added and efficiency to understand how they affect stock prices and returns DEA Electronic copy available at httpsssrncomabstract1350544 16 applications that were published over the period of our survey and investigate the relationship between bank efficiency and stock returns examine Australia Kirkwood and Nahm 2006 Greece Pasiouras et al 2008 Malaysia Sufian and Majid 2006 Singapore Chu and Lim 1998 Sufian and Majid 2007 Turkey Erdem and Erdem 2008 and Spain Guzman et al 2008 Furthermore in a crosscountry setting Beccalli et al 2006 provide evidence from France Germany Italy Spain UK that is the five principal EU banking sectors In most cases the results of these studies indicate a positive relationship between stock returns and efficiency changes Furthermore Becalli et al 2006 reveal that the explanatory power of the model with DEA scores is higher than that of a model that uses the return on equity ROE as a measure of performance 323 Bank ownership A number of studies compare the efficiency of banks across different ownership types One approach used in the literature is to split the sample and compare the means of the different ownership groups Another approach is to incorporate dummy variables in a second stage analysis as the one described in Section 321 Some studies compare domestic and foreign banks Havrylchyk 2006 finds that greenfield banks are more efficient than domestic banks in Poland whereas foreign banks that acquired domestic ones have not successfully increased their efficiency Sturm and Williams 2004 report that foreign banks in Australia are more efficient than domestic ones Isik and Hassan 2003a find that foreign banks are more efficient than private domestic Turkish banks while Isik 2008 reports similar results for TFP growth estimates However Ataullah and Le 2004 find that prior to the financial liberalization of 19911992 foreign banks where less efficient than domestic ones in India and Pakistan nevertheless the opposite picture emerged after this period Other studies examine the efficiency of stateowned banks Many of them find that stateowned banks are less efficient than other banks For instance Garcia Cestona and Surroca 2008 find that Spanish banks controlled by insiders ie managers and workers are more efficient than the ones controlled by public administrations Ariff and Can 2008 find that jointstock banks in China are more cost and profitefficient than stateowned banks Chen 1998 in Taiwan and Mercan et al 2003 in Turkey also report that the efficiency of privatelyowned banks is Electronic copy available at httpsssrncomabstract1350544 17 higher than that of stateowned banks Similarly Isik 2008 focuses on TFP growth and finds that the growth of private banks is more than double that of state banks In contrast other studies report that the efficiency of private banks is lower than that of stateowned banks in Turkey Isik and Hassan 2003a India Sathye 2003 and Austria and Germany Hauner 2005 324 Corporate events and efficiency Another part of the literature relates DEA efficiency estimates to corporate events such as mergers and acquisitions andor bankruptcy Some of the studies compare the efficiency of acquirers and their targets and examine whether mergers improve efficiency Avkiran 1999 reports that acquiring banks are more efficient than acquired Australian banks However the results of this study also indicate that acquiring banks do not always maintain their premerger efficiency and there is mixed evidence on the extent to which the benefits of efficiency are passed to the public The results of AlSharkhas et al 2008 indicate that merged US banks are on average more technically efficient and that they experience higher productivity growth than nonmerged banks Hahn 2007a finds evidence that Austrian banks which engaged in domestic merger deals achieved a higher productive efficiency level than banks which did not participate in such deals The results also show that merger gains remain significant over a longer period of time more than five years but there is slight tendency to level off Sherman and Rupert 2006 also focus on the efficiency gains from bank mergers but they concentrate on the branch rather than the bank level They find that there are opportunities to reduce operating costs Such benefits however are not realised until four years after the merger Kohers et al 2000 follow a different approach to test whether Xefficiencies influence the markets assessment of bank mergers They find that that the abnormal returns of bidders around the announcement period were associated with both the target banks profit efficiency and cost efficiency as well as the difference between the average peer cost efficiency score and the corresponding targets cost15 Wheelock and Wilson 2000 also relate efficiency with bank acquisitions However in their 15 Only cost efficiency was estimated with DEA in this study Profit efficiency was estimated using stochastic frontier analysis As in previous studies Kohers et al 2000 also provide comparisons of efficiency across different groups They find that acquiring US bank holding companies BHCs are less costefficient compared to their targets Also industry peers operate with greater cost efficiencies that then bidder and target BHCs Electronic copy available at httpsssrncomabstract1350544 18 case DEA efficiency scores are used as an input in hazard models that capture the acquisition and failure likelihood They find that inefficiency increases the risk of failure while reducing the probability of a US banks being acquired Using a sample of 245 US banks Luo 2003 also reports that overall technical efficiency can be useful in predicting bank failures 325 Regulatory reformliberalization and efficiency A number of studies examine the impact of regulatory reform and liberalization initiatives on bank efficiency and productivity The studies that we review seem to indicate a positive relationship in most countries including India and Pakistan Ataullah et al 2004 Australia Avkiran 1999 Sturn and Williams 2004 US Mukherjee et al 2001 China Chen et al 2005 Greece Tsionas et al 2003 Rezitis 2006 Taiwan Wang and Huang 2007 Korea Gilbert and Wilson 1998 Turkey Isik and Hassan 2003b and the new EU countries Brissimis et al 2008 However Hauner 2005 finds no evidence that productivity is related to deregulation in Germany and Austria while Sathye 2002 argues that there can be a limit of deregulation after which no further productivity gains could be possible due to deregulation Some other interesting observations can be summarized as follows The banking sector reform in the newly acceded EU countries had a positive impact on bank efficiency while the effect of reform on TFP growth was significantly only toward the end of the reform process Brissimis et al 2008 The implementation of financial liberalization programmes may enable foreign banks to overcome the liability of foreignness and enhance their resource utilization Ataullah and Le 2004 The impact of deregulation will not necessarily be the same across different bank ownerships such as state domestic foreign Isik and Hassan 2003b 326 Comparison of frontier techniques A few studies compare alternative frontier techniques over the period of our survey Bauer et al 1998 use a sample of US banks to compare DEA SFA TFA thick frontier approach and DFA distributionfree approach over six consistency criteria They find that DEA yields much lower average efficiencies ranks the banks differently and identifies the best and worst banks differently from parametric methods Furthermore compared to DEA the parametric measures were more highly Electronic copy available at httpsssrncomabstract1350544 19 correlated with the traditional nonfrontier performance measures Huang and Wang 2002 provide another comparison of SFA DFA and DEA using data from Taiwanese banks They also argue that the choice of the frontier approach can result in different conclusions Weill 2004 provides more recent evidence using a sample from five European countries However this crosscountry setting does not alter the main conclusion of lack of robustness across different approaches confirming the results of the previous two studies Delis et al 2008 provide further support to these findings using a dataset of Greek banks The study of Beccalli et al 2006 also provides useful insights for the differences of SFA and DEA While this study does not focus on the comparison between these two methods it finds that while changes in the stock prices of banks reflect percentage changes in DEA cost efficiency scores this trend is less clear with SFA efficiency estimates In contrast Fiordelisi 2008 reports that SFA cost efficiency estimates explain better the variations in shareholder value creation than those derived from DEA Casu et al 2004 compare productivity growth estimates obtained through parametric and nonparametric approaches Their study seems to indicate that in this case the differences are not as large as in the efficiency studies They conclude that although the alternative methodologies produce in some cases opposing findings as for the sources of productivity for individual years in general they do not yield noticeably different results in terms of identifying the components of the productivity growth of EU banks during the period of their study 327 Efficiency of bank branches While most of the above studies focus on the efficiency of banking institutions as a whole a related strand of the literature examines the efficiency of bank branches16 In general branches are predominantly regarded as production units see Camanho and Dyson 1999 2005 2006 Zenios et al 1999 Soteriou and Zenios 1999b Golany and Storbeck 1999 Athanassopoulos and Giokas 2000 and Hartman et al 2001 where output is measured by the number of various accounts andor transactions as contrast to the intermediation approach employed by Portela and Thanassoulis 2005 2007 Portela et al 2003 Giokas 2008a where branches are seen as intermediaries and thus outputs are measured in monetary terms 16 See Appendix II for information on branch efficiency studies Electronic copy available at httpsssrncomabstract1350544 20 There are studies which consider the changing role of bank branches from a predominantly transactionbased one to a salesoriented role Cook et al 2000 and Cook and Hababou 2001 distinguish sales and services transactions functions of bank branches in a Canadian Bank where inputs are usually shared between these two functions To model the shared resources Cook et al 2000 extend the usual methodology to develop a model which incorporates the best resource split and optimises the aggregate efficiency score Further Cook and Hababou 2001 model the shared resources concept by adopting the Additive DEA model In a recent novel application Portela and Thannasoulis 2007 assess the branches of a Portuguese bank in terms of their performance in their new roles by specifying efficiency measures which consider sales transactional activities as well as bank branch profit efficiency Service quality is also a significant dimension of bank branches performance Two different ways are developed in empirical applications to consider service quality For instance Soteriou and Zenios 1999b Golany and Storbeck 1999 Athanassopoulos and Giokas 2000 include quality variables in efficiency analysis whereas Portela and Thannasoulis 2007 use a posthoc analysis to compare service quality index with efficiency measures They find that service quality is positively related with operational and profit efficiency The diversity of environments in which the branches are operating is also considered in the literature Paradi and Schaffnit 2004 considered the role of the environmental parameters that are outside the management control in the commercial branches of a large Canadian bank and incorporated risk and economic growth rate of the region as two nondiscretionary factors On the other hand Camanho and Dyson 2006 assess the impact of environmental factors and regional managerial policies on branches productivity through the construction of an index that reflects the relative performance of bank branches operating in four different regions in Portugal Das et al 2007 introduce the concept of spatial efficiency for each region relative to the nation and thus measure the effects of differences in the regional characteristics on the efficiency of bank branches across four metropolitan regions in India Some interesting observations can be summarised as follows In general the existing studies of bank branch efficiency use data on a small number of branches of nonUS banks and mostly employ DEA There are studies however which evaluate bank branch performance with DEA and compare the results with other methodologies ie FDH Portela et al 2003 Malmquist indices Camanho and Electronic copy available at httpsssrncomabstract1350544 21 Dyson 2006 loglinear frontier method Giokas 2008b ABC Plus Kantor and Maital 1999 OCRA Parkan and Wu 1999 Neural Network Model Wu et al 2006 Quasiconcave DEA FDH and Parametric Frontier models Dekker and Post 2001 Most of the studies which were reviewed use technical and operational efficiencies as estimated measures Recent studies examine the allocative profit efficiency see Portela and Thanassoulis 2005 2008 and cost efficiency Camanho and Dyson 2005 2008 aspects of bank branches 4 Other OR and AI techniques and bank performance In this section we review 15 recent studies that develop classification models used in the prediction of bank failure 9 bank underperformance 2 and credit ratings 4 Canbas et al 2005 point out that the study of bank failure is important for at least two reasons First an understanding of the reasons beyond the failure allows regulatory authorities to manage and supervise banks more efficiently Second the ability to differentiate between healthy and troubled banks can reduce the expected cost of bank failure either by taking actions to prevent failure or by minimizing the costs to the public Thomson 1991 As discussed in Gaganis et al 2006 while many of the failure prediction models achieve very promising classification accuracies one common drawback is that they concentrate on the assignment of banks in two groups failed and nonfailed However the classification of banks as bad or good reduces the usefulness of the model Obviously the classification into more groups relates to another strand of the literature that deals with credit risk modelling and attempts to replicate the ratings assigned by the credit agencies In the following subsections we provide a brief description of the methodologies the employed variables and the classification results17 A summary of the main characteristics of these studies is available in Appendix III 41 Methodologies Neural network NN modelling is an intelligence technique that follows a process similar to the human brain As in other classification models the parameters of a NN model need to be estimated before the network can be used for prediction purposes There are numerous NN architectures learning methods and parameters 17 To conserve space we keep our discussion on methodologies short Detailed information is available in the corresponding applications andor devoted textbooks Electronic copy available at httpsssrncomabstract1350544 22 Generally speaking NN architectures can be either feedback or feedforward In a feedback network nodes can receive inputs from nodes in any other layer in the network whereas in a feedforward network inputs are only received from previous layers The multilayer preceptor MLPNN competitive learning neural networks CLNN selforganizing map neural networks SOMNN backpropagation neural networks BPNN and probabilistic neural networks PNN are alternative approaches employed in the studies under review eg Chen and Shih 2006 Boyacioglu et al 2008 The approach of Support Vector Machines SVMs used in Boyacioglu et al 2008 Chen and Shih 2006 and Huang et al 2004 was introduced by Vapnik 1995 It is based on the Structural Risk Minimization SRM principle from computational learning theory This seeks to minimise an upper bound of the generalisation error rather than minimise the training error SVMs use a linear structure to implement nonlinear class boundaries through extremely nonlinear mapping of the input vectors into the highdimensional space If the data are linearly separated SVM uses a special kind of linear model the optimal separating hyperplane that provides the maximum separation between the classes The training points that are closest to the maximum margin hyperplane are called support vectors All other training examples are irrelevant for determining the binary class boundaries From the implementation point of view Vapnik 1995 showed how training a SVM and finding the parameters leads to a quadratic optimisation problem with bound constraints and one linear equality constraint This means that the solution of SVMs is unique optimal and absent from local minima Tay and Cao 2001 Nearest Neighbours is a nonparametric estimation method that has been applied in various problems in finance The nearest neighbour rule classifies an object ie bank to the class of its nearest neighbour in the measurement space using some kind of distance measure like the local metrics the global metrics the Mahalanobis or the Euclidean distance The modification of the nearest neighbour rule the knearest neighbour kNN method that is employed in Zhao et al 2008 classifies an object ie bank to the class ie failed or nonfailed more heavily represented among its k nearest neighbours In the case of decision trees that originate from machine learning instead of developing classification functions or network architectures a binary decision tree is developed This can be accomplished through a set of ifthen split conditions that lead Electronic copy available at httpsssrncomabstract1350544 23 to the accurate classification of cases eg banks Commonly used algorithms are CART eg Ravi et al 2008 and C45 eg Zhao et al 2008 The UTilités Additives DIScriminantes UTADIS multicriteria decision aid MCDA method used in Gaganis et al 2006 employs the framework of preference disaggregation analysis for the development of an additive utility function that is used to score the firms and decide upon their classification The estimation of the additive utility model is performed through mathematical programming techniques The MultiGroup Hierarchical Discrimination MHDIS method used in Pasiouras et al 2007 is another MCDA approach that uses utility functions for discrimination purposes MHDIS distinguishes the groups progressively starting by discriminating the first group from all the others and then proceeds to the discrimination between the alternatives belonging to the other groups To accomplish this task instead of developing a single additive utility function that describes all alternatives as in UTADIS two additive utility functions are developed in each one of the n1 steps where n is the number of groups At each stage of the hierarchical discrimination procedure two linear and a mixedinteger programming problems are solved to estimate the utility thresholds and the two additive utility functions in order to minimise the classification error As discussed in Kolari et al 2002 and Lanine and Vander Vennet 2006 trait recognition is another nonparametric approach under which individual traits are initially developed from different segments of the distribution of each variable and the interactions of these segments with one or more other variables segmented distributions When all possible traits of the variables are tabulated for all banks trait recognition uses a search routine to cull traits that do not discriminate between failed and nonfailed banks Trait recognition uses two sets of discriminators the safe traits associated with nonfailed banks and the unsafe traits failed banks known as features Then these features are used to vote on each bank and classify it as failed or nonfailed Lanine and Vander Vennet 2006 argue that the general rule of classifying safe and unsafe cells and assigning default probabilities works poorly when the number of banks falling into mixed cells is high To avoid this problem they suggest a formula to compute the probability of default directly without preliminary classification of cells as safe or unsafe Some researchers use soft computing approaches with the aim of synthesizing the human ability to tolerate and process uncertain imprecise and incomplete Electronic copy available at httpsssrncomabstract1350544 24 information during the decisionmaking Tung et al 2004 Soft computing usually integrates various techniques that may originate from different disciplines such as fuzzy logic18 neural networks machine learning etc in various combinations that will allow the researcher to exploit the strengths of individual techniques Ravi Kumar and Ravi 2008 A popular approach is the integration of neural network and fuzzy sets to develop a neural fuzzy network For example Tung et al 2004 propose a Generic Selforganizing Fuzzy Neural Network GenSoFNN based on the compositional rule of inference CRI to predict bank failure Multiple discriminant analysis MDA and logistic regression analysis LRA are two traditional techniques that are commonly used as benchmarks in classification studies eg Swicegood and Clark 2001 Pasiouras et al 2007 Kosmidou and Zopounidis 2008 In contrast Tung et al 2004 benchmark their models against Cox regression which estimates the dependence of the risk of failure on banks characteristics and of the evolution of that risk over time 42 Variables In most of the cases the variables are selected on the basis of the CAMEL model and include measures for Capital strength Asset quality Earnings and Liquidity19 The number of variables differs significantly among studies Some researchers start from a large list of variables and then use statistical screening eg KruskalWallis or dimension reduction eg factor analysis to end up with a reduced set of variables In addition to the financial variables some studies include additional characteristics related to ownership eg of shares held by the government market information eg stock price in previous year auditing and country eg Heritage index 43 Classification results Many of the studies in Appendix III report quite satisfactory classification accuracies and in several cases the proposed OR and AI methods outperform the traditional techniques ie MDA LRA Cox regression Furthermore the studies that focus on bankruptcy prediction in general perform more accurately than those that deal with 18 The theory of fuzzy sets proposed by Zadeh 1965 provides a mathematical approach to emulate human cognitive process The fuzzy set theory aims to classify subjective reasoning and assign degrees of possibilities in reaching conclusions Fuzzy logic can also be used to obtain fuzzy ifthen rules 19 M in CAMEL refers to management quality and it is rarely used in empirical studies due to difficulties in assessing management quality Electronic copy available at httpsssrncomabstract1350544 25 credit ratings However this is not surprising for two reasons First while classifying the banks in more groups is more informative it becomes more difficult to discriminate among groups in the intermediate categories Second credit ratings may incorporate additional qualitative information and human judgment that cannot be captured by the conventional variables used in most studies In any case comparisons across different studies should be treated with extreme caution One reason is that different studies make different assumptions about the prior probabilities of group membership as well as misclassification costs A second reason is that different studies use different approaches to validate the developed models For instance some researchers simply split the total sample in training and holdout datasets without accounting for the stability of models over time others use resampling techniques eg kfold crossvalidation and others use a holdout sample from a future period However it should kept in mind that classification ability is likely to overstate predictive ability Thus a superior approach would require that the model be validated against a future period as this approach more closely reflects a real world setting 5 Concluding remarks We have presented a comprehensive review of applications of operational research and artificial intelligence techniques in the assessment of bank performance by discussing a total of 179 studies published between 1998 and 2008 We have classified the studies in two main categories The first uses DEA and the DEAlike Malmquist index to estimate the efficiency and productivity growth of banks and bank branches We have discussed various methodological issues such as the estimated measures of efficiency the underlying assumptions of the estimated models and the selection of inputs and outputs We also discussed the main topics of interest including the relationship between ownership and efficiency stock returns and efficiency the determinants of efficiency the efficiency of bank branches amongst others Three of the main conclusions are that i profit efficiency and capacity efficiency have received quite limited attention in DEA studies in banking ii most studies that use a twostage DEA do not employ appropriate bootstrapping techniques and their results may be biased iii there is much diversity among studies with respect to the selection of input and Electronic copy available at httpsssrncomabstract1350544 26 outputs As in the case banks as a whole cost and profit efficiency has received considerably less attention in branch efficiency studies Furthermore an area of research deserving attention would be the estimation of bank branches efficiency over successive time periods Studies falling into the second category attempt to develop classification models to predict failure the credit ratings of banks and identify underperformers We have found both countryspecific and crosscountry studies and applications of techniques that originate from various disciplines Most of the studies rely heavily on financial information although in some cases nonfinancial variables are also used Given the differences in the approaches employed to validate the models comparisons of classification accuracies across studies should be treated with caution We find only a few studies that propose the combination of the predictions of individual models into integrated metaclassifiers and we believe that this is an area of research that is worthy of further attention References Akhtar MH 2002 Xefficiency analysis of commercial banks in Pakistan a preliminary investigation Pakistan Development Review 41 4 567580 Alam P Booth D Lee K Thordarson T 2000 The use of fuzzy clustering algorithm and selforganizing neural networks for identifying potentially failing banks an experimental study Expert Systems with Applications 18 2000 185 199 Allen R 1997 Incorporating value judgements in data envelopment analysis PhD Dissertation Warwick Business School University of Warwick UK AlSharkas AA Hassan MK Lawrence S 2008 The Impact of Mergers and Acquisitions on the Efficiency of the US Banking Industry Further Evidence Journal of Business Finance and Accounting 35 5070 Ariff M Can L 2008 Cost and profit efficiency of Chinese banks A non parametric analysis China Economic Review 19 260273 Asmild M Paradi JC Aggarwall V Schaffnit C 2004 Combining DEA Window Analyis with the Malmquist Index Approach in a Study of the Canadian Banking Industry Journal of Productivity Analysis 21 6789 Electronic copy available at httpsssrncomabstract1350544 27 Ataullah A Cockerill T Le H 2004 Financial liberalization and bank efficiency a comparative analysis of India and Pakistan Applied Economics 36 1915 1924 Ataullah A Le H 2004 Financial repression and liability of foreignness in developing countries Applied Economics Letters 11 545549 Ataullah A Le H 2006 Economic reforms and bank efficiency in developing countries the case of the Indian banking industry Applied Financial Economics 16 653663 Athanassopoulos AD Giokas D 2000 The use of data envelopment analysis in banking institutions evidence from the commercial bank of Greece Interfaces 30 2 8195 Avkiran NK 1999 The evidence on efficiency gains The role of mergers and the benefits to the public Journal of Banking and Finance 23 9911013 Aysan AF Ceyhan SP 2008 What determines the banking sector performance in globalized financial markets The case of Turkey Physica A 387 15931602 Bala K Cook WD 2003 Performance measurement with classification information an enhanced additive DEA model Omega 31 439450 Ball R Brown P 1968 An Empirical Evaluation of Accounting Income Numbers Journal of Accounting Research 6 159178 Banker RD Morey RC 1986 Efficiency Analysis for Exogenously Fixed Inputs and Outputs Operations Research 34 513521 Banker RD Morey RC 1986 Efficiency Analysis for Exogenously Fixed Inputs and outputs Operations Research 34 513521 Barr RS 2002 Evaluating the productive efficiency and performance of US commercial banks Managerial Finance 28 8 325 Bauer PW Berger AN Ferrier GD Humphrey DB 1998 Consistency Conditions for Regulatory Analysis of Financial Institutions A Comparison of Frontier Efficiency Methods Journal of Economics and Business 50 85114 Beccalli E Casu B Girardone C 2006 Efficiency and Stock Performance in European Banking Journal of Business Finance and Accounting 33 245 262 Berg S Forsund F Jansen E 1991 Malmquist indices of productivity growth during the deregulation of Norwegian banking 198089 Scandinavian Journal of Economics 94 211228 Electronic copy available at httpsssrncomabstract1350544 28 Bergendahl G 1998 DEA and benchmarks an application to Nordic banks Annals of Operations Research 82 233249 Bergendahl G Lindblom T 2008 Evaluating the performance of Swedish savings banks according to service efficiency European Journal of Operational Research 185 16631673 Berger AN 2007 International Comparisons of Banking Efficiency Financial Markets Institutions Instruments 16 119 144 Berger AN Humphrey DB 1997 Efficiency of financial institutions International survey and directions for future research European Journal of Operational Research 98175212 Berger AN Mester LJ 1997 Inside the black box What explains differences in the efficiencies of financial institutions Journal of Banking and Finance 21 895 947 Board J Sutcliffe C Ziemba WT 2003 Applying Operations Research Techniques to Financial Markets INTERFACES 33 1224 Boyacioglu MA Kara Y Baykan OmK 2008 Predicting bank financial failures using neural networks support vector machines and multivariate statistical methods A comparative analysis in the sample of savings deposit insurance fund SDIF transferred banks in Turkey Expert Systems with Applications doi101016jeswa200801003 Brissimis SN Delis MD Papanikolaou NI 2008 Exploring the nexus between banking sector reform and performance Evidence from newly acceded EU countries Journal of Banking and Finance 32 26742683 Camanho AS Dyson RG 1999 Efficiency size benchmarks and targets for bank branches An application of data envelopment analysis Journal of the Operational Research Society 50 9 903915 Camanho AS Dyson RG 2005 Cost efficiency measurement with price uncertainty a DEA application to bank branch assessments European Journal of Operational Research 161 432446 Camanho AS Dyson RG 2006 Data envelopment analysis and Malmquist indices for measuring performance Journal of Productivity Analysis 26 3549 Camanho AS Dyson RG 2008 A generalisation of the Farrell cost efficiency measure applicable to nonfully competitive settings Omega 36 147162 Electronic copy available at httpsssrncomabstract1350544 29 Canbas S Cabuk A Kilic SB 2005 Prediction of commercial bank failure via multivariate statistical analysis of financial structures the Turkish case European Journal of Operational Research 166 528546 Canhoto A Dermine J 2003 A note on banking efficiency in Portugal New vs Old banks Journal of Banking and Finance 27 20872098 Caprio G Klingebiel D 2003 Episodes of Systemic and Borderline Financial Crises World Bank January Available at httpgoworldbankorg5DYGICS7B0 Casu B Girardone C 2004 Financial conglomeration efficiency productivity and strategic drive Applied Financial Economics 14 687696 Casu B Girardone C 2005 An analysis of the relevance of offbalance sheet items in explaining productivity change in European banking Applied Financial Economics 15 10531061 Casu B Girardone C 2006 Bank competition concentration and efficiency in the single European Market The Manchester School 74 441468 Casu B Girardone C Molyneux P 2004 Productivity change in European banking A comparison of parametric and nonparametric approaches Journal of Banking and Finance 28 25212540 Casu B Molyneux P 2003 A comparative study of efficiency in European banking Applied Economics 35 18651876 Chang TC Chiu YH 2006 Affecting factors on riskadjusted efficiency in Taiwans banking industry Contemporary Economic Policy 24 4 634648 Charnes A Cooper WW Rhodes E 1978 Evaluating Program and Managerial Efficiency An Application of Data Envelopment Analysis to Program Follow Through Management Science 27 668697 Charnes A Cooper WW Rhodes E 1981 Evaluating Program and Managerial Efficiency An Application of Data Envelopment Analysis to Program Follow Through Management Science 27 668697 Chen TY 1998 A study of bank efficiency and ownership in Taiwan Applied Economics Letters 5 613616 Chen TY 2001 An estimation of Xinefficiency in Taiwans banks Applied Financial Economics 11 237242 Chen TY 2002a Measuring firm performance with DEA and prior information in Taiwans banks Applied Economics Letters 9 201204 Electronic copy available at httpsssrncomabstract1350544 30 Chen WH Shih JY 2006 A study of Taiwans issuer credit rating systems using support vector machines Expert Systems with Applications 30 427435 Chen X Skully M Brown K 2005 Banking efficiency in China Application of DEA to pre and postderegulation eras 19932000 Chen CW Liou TS Wu YS 2004 The performance measurement of local banks in the era of the financial liberalization on Taiwan Information and Management Sciences 15 4 89104 Chen TY 2002b A comparison of chanceconstrained DEA and stochastic frontier analysis bank efficiency in Taiwan Journal of the Operations Research Society 53 5 492500 Chen TY 2004 A study of cost efficiency and privatisation in Taiwans banks the impact of the Asian financial crisis Service Industries Journal 24 5 137 151 Chen TY Yeh TL 2000 A measurement of bank efficiency ownership and productivity changes in Taiwan Service Industries Journal 20 1 95109 Chiu YH Jan C Shen DB Wang PC 2008 Efficiency and capital adequacy in Taiwan banking BCC and superDEA estimation Service Industries Journal 28 4 479496 Chu SF Lim GH 1998 Share performance and profit efficiency of banks in an oligopolistic market evidence from Singapore Journal of Multinational Financial Management 8 155168 Coelli TJ Perelman SA 1996 Comparison of Parametric and Nonparametric Distance Functions With Application to European Railways CREPP Discussion Paper no 199611 University of Liege Liege Coelli TJ Prasada Rao DS ODonnell CJ Battese GE 2005 An Introduction to Efficiency and Productivity Analysis 2nd Ed Springer USA Cook WD Seiford LM 2009 Data envelopment analysis DEA Thirty years on European Journal of Operational Research 192 117 Cook WD Seiford LM Zhu J 2004 Models for performance benchmarking measuring the effect of ebusiness activities on banking performance Omega 32 313322 Cook WD Hababou M 2001 Sales performance measurement in bank branches Omega 29 299307 Electronic copy available at httpsssrncomabstract1350544 31 Cook WD Hababou M Liang L 2005 Financial liberalization and efficiency in Tunisian banking industry DEA test International Journal of Information Technology Decision Making 4 3 455475 Cook WD Hababou M Tuenter HJH 2000 Multicomponent efficiency measurement and shared inputs in Data Envelopment Analysis an application to sales and service performance in bank branches Journal of Productivity Analysis 14 209224 Damar HE 2006 The effects of shared ATM networks on the efficiency of Turkish banks Applied Economics 38 683697 Das A Ghosh S 2006 Financial deregulation and efficiency An empirical analyis of Indian banks during the post reform period Review of Financial Economics 15 193221 Das A Ray SC Nag A 2007 Labouruse efficiency in Indian banking a branchlevel analysis Omega doi101016jomega200705002 Debasish SS Mishra B 2007 Performance Evaluation of Indian BanksA Study using Data Envelopment Analysis European Journal of Economics Finance and Administrative Sciences9 8694 Dekker D Post T 2001 A quasiconcave DEA model with an application for bank branch performance evaluation European Journal of Operational Research 132 296311 Delis MD KoutsomanoliFillipaki K Staikouras CK Gerogiannaki K 2008 Evaluating cost and profit efficiency a comparison of parametric and nonparametric methodologies Applied Financial Economics doi 10108009603100801935370 Devaney M Weber WL 2000 Productivity growth market structure and technological change evidence from the rural banking sector Applied Financial Economics 10 587595 Dimitras AI Zanakis SH Zopounidis C 1996 A survey of business failures with an emphasis on prediction methods and industrial applications European Journal of Operational Research 90 487513 Dogan E Fausten DK 2003 Productivity and Technical Change in Malaysian Banking 19891998 AsiaPacific Financial Markets 10 205237 Drake L 2001 Efficiency and productivity change in UK banking Applied Financial Economics 11 557571 Electronic copy available at httpsssrncomabstract1350544 32 Drake L Hall MJB 2003 Efficiency in Japanese banking An empirical analysis Journal of Banking and Finance 27 891917 Drake L Hall MJB Simper R 2006 The impact of macroeconomic and regulatory factors on bank efficiency A nonparametric analysis of Hong Kongs banking system Journal of Banking and Finance 30 14431466 Erdem C Erdem MS 2007 Turkish banking efficiency and its relation to stock performance Applied Economics Letters 15 207211 Fare R Grosskopf S Weber WL 2004 The effect of riskbased capital requirements on profit efficiency in banking Applied Economics 36 1731 1743 Fiordelisi F 2008 Efficiency and shareholder return in banking International Journal of Banking Accounting and Finance 1 114 132 Fried HO Lovell CAK Schmidt SS Yaisawarng S 2002 Accounting for Environmental Effects and Statistical Noise in Data Envelopment Analysis Journal of Productivity Analysis 17 157174 Fried HO Schmidt SS Yaisawarng S 1999 Incorporating the operating environment into a nonparametric measure of technical efficiency Journal of Productivity Analysis 12 249267 Fukuyama H Weber WL 2005 Estimating output gains by means of Luenberger efficiency measures European Journal of Operational Research 164 535547 Fukuyama H Weber WL 2002 Estimating output allocative efficiency and productivity change Application to Japanese banks European Journal of Operational Research 137 177190 Fung MK 2006 Scale economies Xefficiency and convergence of productivy among bank holding companies Journal of Banking and Finance 30 2857 2874 Gaganis C Pasiouras F Zopounidis C 2006 A Multicriteria Decision Framework for Measuring Banks Soundness Around the World Journal of MultiCriteria Decision Analysis 14 103111 GarciaCestona M Surroca J 2008 Multiple goals and ownership structure Effects on the performance of Spanish savings banks European Journal of Operational Research 187 582599 Gilbert RA Wilson PW 1998 Effects of Deregulation on the Productivity of Korean Banks Journal of Economics and Business 50 133155 Electronic copy available at httpsssrncomabstract1350544 33 Giokas DI 2008a Cost efficiency impact of bank branch characteristics and location Managerial Finance 34 3 172185 Giokas DI 2008b Assessing the efficiency in operations of a large Greek bank branch network adopting different economic behaviors Economic Modelling 25 559574 Golany B Storbeck JE 1999 A Data Envelopment Analysis of the operational efficiency of bank branches Interfaces 29 3 1426 Griffiths B Beynon MJ 2005 Expositing stages of VPRS analysis in an expert system Application with bank credit ratings Expert Systems with Applications 29 879888 Guzman I Reverte C 2008 Productivity and efficiency change and shareholder value evidence from the Spanish banking sector Applied Economics 40 2033 2040 Habibullah M S Muhd Iqbal Makmur M I AzmanSaini WNW Radam A Boon OH 2005 Bank Efficiency and the Efficient Market Hypothesis The Case for Bank Stock Prices in KLSE Savings and Development 29 4 263290 Hahn FR 2007a Domestic mergers in the Austrian banking sector a performance analysis Applied Financial Economics 17 185196 Hahn FR 2007b Environmental determinants of banking efficiency in Austria Empirica 34 231245 Halkos GE Salamouris DS 2004 Efficiency measurement of the Greek commercial banks with the use of financial ratios a data envelopment analysis approach Management Accounting Research 15 201224 Hartman TE Storbeck JE Byrnes P 2001 Allocative efficiency in branch banking European Journal of Operational Research 134 232242 Hauner D 2005 Explaining efficiency differences among large German and Austrian banks Applied Economics 37 969980 Havrylchyk O 2006 Efficiency of the Polish banking industry Foreign versus domestic banks Journal of Banking and Finance 30 19751996 Howland M Rowse J 2006 Measuring bank branch efficiency using data envelopment analysis managerial and implementation issues INFOR 44 1 49 63 Electronic copy available at httpsssrncomabstract1350544 34 Hu WC Lai MC Huang HC 2008 Rating the relative efficiency of financial holding companies in an emerging economy A multiple DEA approach Expert Systems with Applications doi 10 1016jeswa200806080 Huang TH Wang MH 2002 Comparison of economic efficiency estimation methods parametric and nonparametric techniques The Manchester School 70 682709 Huang LH Hsiao HC Cheng MA Chang SJ 2008 Effects of financial reform on productivity change Industrial Management Data Systems 108 7 867886 Huanga Z Chena H Hsua CJ Chenb WH Wu S 2004 Credit rating analysis with support vector machines and neural networks a market comparative study Decision Support Systems 37 543 558 Isik I 2007 Bank ownership and productivity developments evidence from Turkey Studies in Economics and Finance 24 115139 Isik I 2008 Productivity technology and efficiency of de novo banks A counter evidence from Turkey Journal of Multinational Financial Management 18 427442 Isik I Akcaoglu E 2006 An Empirical Analysis of Productivity Developments in Traditional Banks The Initial PostLiberalization Experience Central Bank Review 1 135 Isik I Hassan MK 2002 Technical scale and allocative efficiencies of Turkish banking industry Journal of Banking and Finance 26 719766 Isik I Hassan MK 2003a Efficiency Ownership and Market Structure Corporate Control and Governance in the Turkish Banking Industry Journal of Business Finance and Accounting 30 1363 1421 Isik I Hassan MK 2003b Financial deregulation and total factor productivity change An empirical study of Turkish commercial banks Journal of Banking and Finance 27 14551485 Isik I Hassan MK 2003c Financial disruption and bank productivity The 1994 experience of Turkish banks The Quarterly Review of Economics and Finance 43 291320 Isik I Uysai D 2006 The Sources of Productivity Growth in Turkish Banks by Ownership The Initial Responses of Public Private and Foreign Banks to Liberalization Turkish Studies 7 365403 Electronic copy available at httpsssrncomabstract1350544 35 Kantor J Maital S 1999 Measuring efficiency by product group integrating DEA with activitybased accounting in a large Mideast bank Interfaces 29 3 2736 Kao C Liu ST 2004 Predicting bank performance with financial forecasts A case of Taiwan commercial banks Journal of Banking and Finance 28 2353 2368 Kirkwood J Nahm D 2006 Australian Banking Efficiency and Its Relation to Stock Returns The Economic Record 82 253267 Kohers T Huang Mh Kohers N 2000 Market perception of efficiency in bank holding company mergers the roles of the DEA and SFA models in capturing merger potential Review of Financial Economics 9 101120 Kolari J Glennon D Shin H Caputo M 2002 Predicting large US commercial bank failures Journal of Economics and Business 54 361387 Kosmidou K Zopounidis C 2008 Predicting US commercial bank failures via a multicriteria approach International Journal of Risk Assessment and Management 9 2643 Krishnasamy G Ridzwa AH Perumal V 2003 Malaysian Post Merger Banks Productivity Application of Malmquist Productivity Index Managerial Finance 30 63 74 Kumar S Gulati R 2008 Evaluation of technical efficiency and ranking of public sector banks in India International Journal of Productivity Performance Management 57 7 540568 Kuosmanen T Post T 2001 Measuring economic efficiency with incomplete price information With an application to European commercial banks European Journal of Operational Research 134 4358 Kyj L Isik I 2008 Bank xefficiency in Ukraine An analysis of service characteristics and ownership Journal of Economics and Business 60 369393 Lanine G Vander Vennet R 2006 Failure prediction in the Russian bank sector with logit and trait recognition models Expert Systems with Applications 30 463478 Laurenceson J Qin F 2008 Has minority foreign investment in Chinas banks improved their cost efficiency China and World Economy 16 3 5774 Leightner JE Knox Lovell CA 1998 The impact of financial liberalization on the performance of Thai banks Journal of Economics and Business 50 115 131 Electronic copy available at httpsssrncomabstract1350544 36 Levine R 2005 Finance and Growth Theory and Evidence Chapter 12 in Handbook of Economic Growth vol 1 Part A Springer pp 865934 Lim GH Randhawa DS 2005 Competition liberalization and efficiency Evidence from a twostage banking model on banks in Hong Kong and Singapore Managerial Finance 31 1 5277 Lin YH Hsu GJY Hsiao CK 2007 Measuring efficiency of domestic banks in Taiwan application of data envelopment analysis and Malmquist index Applied Economics Letters 14 821827 Liu ST 2008 Slacksbased efficiency measures for predicting bank performance Expert Systems with Applications doi 101016jeswa200801032 Liu ST Kao C 2008 Stochastic data envelopment analysis in measuring the efficiency of Taiwan commercial banks European Journal of Operational Research doi 101016jejor200802023 LozanoVivas A Pastor JT Hasan I 2001 European Bank Performance Beyond Country Borders What Really Matters European Finance Review 5 141165 LozanoVivas A Pastor JT Pastor JM 2002 An Efficiency Comparison of European Banking Systems Operating under Different Environmental Conditions Journal of Productivity Analysis 18 5977 Luo X 2003 Evaluating the profitability and marketability efficiency of large banks An application of data envelopment analysis Journal of Business Research 56 627635 Maghyereh A 2004 The effect of financial liberalization on the efficiency of financial institutions the case of Jordanian commercial banks Savings and Development 124 Maudos J Pastor JM 2003 Cost and profit efficiency in the Spanish banking sector 19851996 a nonparametric approach Applied Financial Economics 13 112 Maudos J Pastor JM Perez F 2002 Competition and efficiency in the Spanish banking sector the importante of specialization Applied Financial Economics 12 505516 Mercan M Reisman A Yolalan R Emel AB 2003 The effect of scale and mode of ownership on the financial performance of the Turkish banking sector results of a DEAbased analysis SocioEconomic Planning Sciences 37 185202 Electronic copy available at httpsssrncomabstract1350544 37 Mester LJ 1996 A Study of Bank Efficiency Taking into Account RiskPreferences Journal of Banking and Finance 20 102545 Mostafa M 2007a Benchmarking top Arab banks efficiency through efficient frontier analysis Industrial Management Data Systems 107 6 802823 Mostafa M 2007b Modeling the efficiency of GCC banks a data envelopment analysis approach International Journal of Productivity and Performance Management 56 7 623643 Mukherjee K Ray SC Miller SM 2001 Productivity growth in large US commercial banks The initial postderegulation experience Journal of Banking and Finance 25 913939 Mukherjee A Nath P Pal M 2003 Resource service quality and performance triad a framework for measuring efficiency of banking services Journal of Operations Research Society 54 723735 Neal P 2004 Xefficiency and productivity change in Australian banking Australian Economic Papers June 174191 Ng GS Quek C Jiang H 2008 FCMACEWS A bank failure early warning system based on a novel localized pattern learning and semantically associative fuzzy neural network Expert Systems with Applications 34 9891003 Noulas AG 2001 Deregulation and operating efficiency the case of the Greek banks Managerial Finance 27 8 3547 Noulas AG 1997 Productivity growth in the Hellenic banking industry state versus private banks Applied Financial Economics 7 223228 Noulas AG Glaveli N Kiriakopoulos I 2008 Investigating cost efficiency in the branch network of a Greek bank empirical study Managerial Finance 34 3 160171 Olgu O WeymanJones TG 2008 Parametric and nonparametric Malmquist productivity decomposition a case study of European commercial banks International Journal of Business Performance Management 10 4 344365 Paradi JC Schaffnit C 2004 Commercial branch performance evaluation and results communication in a Canadian banka DEA application European Journal of Operational Research 156 719735 Parkan C Wu ML 1999 Measurement of the performance of an investment bank using the operational competitiveness rating procedure Omega 27 201217 Electronic copy available at httpsssrncomabstract1350544 38 Pasiouras F Gaganis Ch Doumpos M 2007 A multicriteria discrimination approach for the credit rating of Asian banks Annals of Finance 3 351367 Pasiouras F 2008b Estimating the technical and scale efficiency of Greek commercial banks The impact of credit risk offbalance sheet activities and international operations Research in International Business and Finance 22 301318 Pasiouras F 2008a International evidence on the impact of regulations and supervision on banks technical efficiency an application of twostage data envelopment analysis Review of Quantitative Finance and Accounting 30 187 223 Pasiouras F Liadaki A Zopounidis C 2008 Bank efficeincy and share performance evidence from Greece Applied Financial Economics 18 1121 1130 Pastor JM 1999 Efficiency and risk management in Spanish banking a method to decompose risk Applied Financial Economics 9 371384 Pastor JM 2002 Credit risk and efficiency in the European banking system A threestage analysis Applied Financial Economics 12 895911 Pastor JM Serrano L 2006 The Effect of Specialisation on Banks Efficiency An International Comparison International Review of Applied Economics 20 125 149 Paul S Kourouche K 2008 Regulatory policy and the efficiency of the banking sector in Australia Australian Economic Review 41 3 260271 Porembski M Breitenstein K Alpar P 2005 Visualising efficiency and reference relations in data envelopment analysis with an application to the branches of a German bank Journal of Productivity Analysis 23 203221 Portela MCAS Borges PC Thanassoulis E 2003 Finding closest targets in nonoriented DEA models the case of convex and nonconvex technologies Journal of Productivity Analysis 19 251269 Portela MCAS Thanassoulis E 2005 Profitability of a sample of Portuguese bank branches and its decomposition into technical and allocative components European Journal of Operational Research 162 850866 Portela MCAS Thanassoulis E 2007 Comparative efficiency analysis of Portuguese bank branches European Journal of Operational Research 177 12751288 Electronic copy available at httpsssrncomabstract1350544 39 Post T Nonparametric efficiency estimation in stochastic environments Noiseto signal estimation finite sample performance and hypothesis testing Journal of Banking and Finance 20652080 Prior D 2003 Long and shortrun nonparametric cost frontier efficiency An application to Spanish savings banks Journal of Banking and Finance 27 655 671 Ramanathan R 2007 Performance of banks in countries of the Gulf Cooperation Council International Journal of Productivity and Performance Management 56 2 137154 Ravi Kumar P Ravi V 2007 Bankruptcy prediction in banks and firms via statistical and intelligent techniques A review European Journal of Operational Research 180 128 Ravi V Kurniawan H Thai PNK Ravi Kumar P 2008 Soft computing system for bank performance prediction Applied Soft Computing 8 305315 Ravi V Pramodh C 2008 Threshold accepting trained principal component neural network and feature subset selection Application to bankruptcy prediction in banks Applied Soft Computing doi101016jasoc200712003 Ray SC Mukherjee K 1998 A study of size efficiency in US banking identifying banks that are too large International Journal of systems Science 29 11 12811294 Resti A 1998 Regulation can foster mergers can mergers foster efficiency The Italian case Journal of Economics and Business 50 157169 Rezitis AN 2006 Productivity growth in the Greek banking industry A non parametric approach Journal of Applied Economics IX 119138 Rizvi SFA 2001 Postliberalisation efficiency and productivity of the banking sector in Pakistan Pakistan Development Review 40 4 605632 Saha A Ravisankar TS 2000 Rating of Indian commercial banks A DEA approach European Journal of Operational Research 124 187203 Sahoo BK Tone K 2008 Decomposing capacity utilization in data envelopment analysis An application to banks in India European Journal of Operational Research doi 101016jejor200802017 Sathye M 2003 Efficiency of banks in a developing economy The case of India European Journal of Operational Research 148 662671 Electronic copy available at httpsssrncomabstract1350544 40 Sathye M 2001 Xefficiency in Australian banking An empirical investigation Journal of Banking and Finance 25 613630 Sathye M 2002 Measuring Productivity Changes in Australian Banking An Application of Malmquist Indices Managerial Finance 28 48 59 Sealey C Lindley J 1977 Inputs outputs and a theory of production and cost at depository financial institutions Journal of Finance 32 12511266 Seiford LM Zhu J 1999 Profitability and Marketability of the Top 55 US Commercial Banks Management Science 45 12701288 Sherman HD Rupert TJ 2006 Do bank mergers have hidden or foregone value Realized and unrealized operating synergies in one bank merger European Journal of Operational Research 168 253268 Sherman HD Zhu J 2006 Benchmarking with qualityadjusted DEA QDEA to seek lowercost high quality service Evidence from a US bank application Annals of Operations Research 145 301309 Simar L Wilson PW 2007 Estimation and inference in twostage semi parametric models of production processes Journal of Econometrics 136 31 64 Smith KA Gupta JND 2000 Neural networks in business techniques and applications for the operations research Computers and Operations Research 27 10231044 Soteriou AC Zenios SA 1999a Using data envelopment analysis for costing bank products European Journal of Operational Research 114 23424 Soteriou A Zenios SA 1999b Operations quality and profitability in the provision of banking services Management Science 45 9 12211238 Sowlati T Paradi JC 2004 Establishing the practical frontier in data envelopment analysis Omega 32 261272 Stavárek D 2006 Banking efficiency in the context of European integration Eastern European Economics 44 4 531 Sturm JE Williams B 2004 Foreign bank entry deregulation and bank efficiency Lessons from the Australian experience Journal of Banking and Finance 28 17751799 Sueyoshi T Kirihara Y 1998 Efficiency measurement and strategic classification of Japanese banking institutions International Journal of Systems Science 29 11 12491263 Electronic copy available at httpsssrncomabstract1350544 41 Sufian F 2008 Determinants of bank efficiency during unstable macroeconomic environment Empirical evidence from Malaysia Research in International Business and Finance doi 101016jribaf200807002 Sufian F Majid MZ A 2006 Banks Efficiency and Stock Prices in Emerging Markets Evidence from Malaysia Journal of AsiaPacific Business 7 3553 Sufian F Majid MZ A 2007 Xefficiency and share prices in the Singaporean banking sector A DEA window analysis approach Investment Management and Financial Innovations 4 73 90 Sufian F 2007 Trends in the efficiency of Singapores commercial banking groups a nonstochastic frontier DEA window analysis approach International Journal of Productivity Performance Management 56 2 99136 Swicegood P Clark JA 2001 Offsite Monitoring Systems for Predicting Bank Underperformance A Comparison of Neural Networks Discriminant Analysis and Professional Human Judgment International Journal of Intelligent Sytems in Accounting Finance and Management 10 169186 Tay FEH Cao L 2001 Application of support vector machines in financial time series forecasting Omega The International Journal of Management Science 29 309317 Thanassoulis E 2001 Introduction to the Theory and Application of Data Envelopment Analysis A Foundation Text with Integrated Software Kluwer Academic Publishers USA Thomson JB 1991 Predicting bank failure in the 1980s Economic Review Federal Reserve Bank of Cleveland Q1 920 TortosaAusina E 2002a Bank Cost Efficiency and Output Specification Journal of Productivity Analyis 18 199222 TortosaAusina E 2002b Cost Efficiency and Product Mix Clusters across the Spanish Banking Industry Review of Industrial Organization 20 163181 TortosaAusina E 2002c Exploring efficiency differences over time in the Spanish banking industry European Journal of Operational Research 139 643664 TortosaAusina E 2003 Nontraditional activities and bank efficiency revisited a distributional analysis for Spanish financial institutions Journal of Economics and Business 55 371395 TortosaAusina E GrifellTatje E Armero C Conesa D 2008 Sensitivity analysis of efficiency and Malmquist productivity indices An application to Electronic copy available at httpsssrncomabstract1350544 42 Spanish savings banks European Journal of Operational Research 184 1062 1084 Tsionas EG Lolos SEG Christopoulos DK 2003 The performance of the Greek banking system in view of the EMU results from a nonparametric approach Economic Modelling 20 571592 Tunga WL Queka C Cheng P 2004 GenSoEWS a novel neuralfuzzy based early warning system for predicting bank failures Neural Networks 17 567587 Valverde SC Humphrey DB Paso RL 2007 Opening the black box Finding the source of cost inefficiency Journal of Productivity Analysis 27 209220 Vapnik VN 1995 The nature of statistical learning theory SpringerVerlag New York Wang MH Huang TH 2007 A study on the persistence of Farrells efficiency measure under a dynamic framework European Journal of Operational Research 180 13021316 Webb RM 2003 Levels of efficiency in UK retail banks a DEA window analysis International Journal of the Economics of Business 10 305322 Weill L 2004 Measuring Cost Efficiency in European Banking A Comparison of Frontier Techniques Journal of Productivity Analyis 21 133152 Wheelock DC Wilson PW 2000 Why do banks disappear The determinants of US bank failures and acquisitions The Review of Economics and Statistics 82 127138 Wu D Yang Z Liang L 2006 Using DEAneural network approach to evaluate branch efficiency of a large Canadian bank Expert Systems with Applications 31 108115 Yildirim C 2002 Evolution of banking efficiency within an unstable macroeconomic environment the case of Turkish commercial banks Applied Economics 34 22892301 Zadeh L A 1965 Fuzzy sets Information and Control 8 338353 Zen F Baldan C 2008 The strategic paths and performances of Italian mutual banks a nonparametric analysis International Journal of Banking Accounting and Finance 1 189 214 Zenios CV Zenios SA Agathocleous K Soteriou AC 1999 Benchmarks of the efficiency of bank branches Interfaces 29 3 3751 Electronic copy available at httpsssrncomabstract1350544 43 Zhao H Sinha AP Ge W 2008 Effects of feature construction on classification performance An empirical study in bank failure prediction Expert Systems with Applications doi101016jeswa200801053 Zhao T Casu B Ferrari A 2008 Deregulation and productivity growth a study of the Indian commercial banking industry International Journal of Business Performance Management 10 4 318343 Zhou p Ang BW Poh KL 2008 A survey of data envelopment analysis in energy and environmental studies European Journal of Operational Research 189 118 Electronic copy available at httpsssrncomabstract1350544 44 Table 1 Surveyed studies by publication outlet Bank Efficiency productivity Branch Efficiency productivity Bank Bankruptcy ratings underperformance Total European Journal of Operational Research 12 7 0 19 Journal of Banking and Finance 15 0 0 15 Applied Financial Economics 13 0 0 13 Applied Economics 9 0 0 9 Journal of Productivity Analysis 5 4 0 9 Expert Systems with Applications 2 1 6 9 Journal of Economics and Business 6 0 1 7 Managerial Finance 5 2 0 7 Omega 0 5 0 5 Applied Economics Letters 4 0 0 4 Interfaces 0 4 0 4 Other 65 5 8 78 Total 136 28 15 179 Electronic copy available at httpsssrncomabstract1350544 45 Appendix I Studies in bank efficiency and productivity change Authors Publication year Country Sample Period InputOutput Oriented Treatment of deposits Estimated measures 2nd Stage regression Akhtar 2002 Pakistan 40 banks 1998 Input Input OTE AE OE na AlSharkas et al 2008 US 440 bank mergers peer group of non merged banks 19862002 Input Input OTE PTE SE AE CE Productivity and APE using SFA na Ariff Can 2008 China 28 banks 230 obs 19952004 Input CE Output PE Input CE APE SPE Tobit Asmild et al 2004 Canada 5 banks 19812000 Input Output OTE Productivity na Ataullah et al 2004 India Pakistan na 19881998 Output Interest expense input OTE PTE SE na Ataullah Le 2006 India 566 Obs 19921998 Output Interest expense input PTE OLS GMM Ataullah Le 2004 India Pakistan na 19871998 Input Deposits Output Financial expenses input PTE na Avkiran 1999 Australia 1619 banks 19861995 Input Interest expense input in Model 1 Deposits input in Model 2 OTE na Aysan Ceyhan 2008 Turkey 466 obs 19902006 Input Input PTE Productivity FE regression Barr et al 2002 US na 19841998 Input Interest expense input OTE na Bauer et al 1998 US 683 banks 19771988 Input Demand deposits output Time saving deposits input CE na Beccalli et al 2006 France Germany Italy Spain UK 1129 banks per country 90 banks in total 19992000 Both Interest expenses part of single input ie total cost PTE and CE with SFA na Bergendahl 1998 Denmark Finland Norway Sweden 48 banks 19921993 Input Output OTE PTE na Electronic copy available at httpsssrncomabstract1350544 46 Bergendahl Lindblom 2008 Sweden 8588 banks 19972001 Input Output OTE na Brissimis et al 2008 10 new EU countries 364 banks 4368 obs 19942005 Input Input PTE Productivity double bootstrap two stage least squares truncated Canhoto Dermine 2003 Portugal 20 banks 19901995 Input Output OTE PTE SE Productivity na Casu Girardone 2004 Italy 3648 banks 168 obs 19961999 Input Ef Output Prod Input PTE AE CE Productivity Logistic Casu Molyneux 2003 France Germany Italy Spain UK 530 banks 19931997 Both Input OTE PTE BootstrapTobit Casu et al 2004 France Germany Italy Spain UK 2086 banks 19942000 Output Cost of deposits as input Productivity na Casu Girardone 2006 EU15 11000 obs 19972003 Input Interest expenses part of a single input ie total cost PTE na Casu Girardone 2005 France Germany Italy Spain UK 2086 banks 19942000 Output Cost of deposits input Productivity na Chang Chiu 2006 Taiwan 26 banks 19962000 Input Input OTE AE CE Tobit Chen 2001 Taiwan 41 banks 19881997 Both Input OTE na Chen 1998 Taiwan 34 banks 1996 Input Deposits input in 2 Models Interest expenses input in 6 models OTE PTE SE Type of regression not specified Chen 2002a Taiwan 41 banks 1998 Output na OTE na Chen 2002b Taiwan 39 banks 273 obs 19942000 Input Input OTE Regression Electronic copy available at httpsssrncomabstract1350544 47 Chen 2004 Taiwan 44 banks 19942000 Input Input CEOTEPTESEAE na Chen Yeh 2000 Taiwan 34 banks 19951996 Input Ef Output Prod Input OTEPTESE na Chen et al 2005 China 43 banks 19932000 Input Price of deposits input Deposits output PTE AE CE na Chen et al 2004 Taiwan 49 banks 20002002 Input Ef Output Prod Input OTE PTESE Productivity na Chiu et al 2008 Taiwan 46 banks 20002002 Input Ef Output Prod Input PTE productivity Chu Lim 1998 Singapore 6 banks 19921996 Both Interest expenses input OTE PTE SE na Cook et al 2005 Tunisia 10 banks 69 obs 19921997 Input Interm App Interest expenses Input Product App Deposits Input OTE Regression Damar 2006 Turkey 3538 banks 20002003 Input Interest expense input Deposits output OTE PTE SE Type of regression not specified Das Ghosh 2006 India 7498 banks 19922002 Input 3 models in total Interest expense input in 2 models Deposits output in 1 model OTE PTE SE Tobit Debasish Mishra 2007 India 93 banks 20002007 Output Input OTE na Delis et al 2008 Greece 1423 banks 244 obs 19932005 Input Input CE and PE using SFA na Devaney Weber 2000 US 3391 4146 banks 19901993 Output Input OTE PTE SE Productivity SUR Dogan Fauesten 2003 Malaysia 1618 banks 19891998 Output Deposits output Interest expenses input OTE Productivity FE linear model Drake 2001 UK 9 banks 19841995 Input Input in Models 1 and 1a Output in Models 2 and 2b OTE PTE SE Productivity na Drake et al 2006 Hong Kong 4766 banks 413 observations in total 19952001 Input na in model 1 output in model 2 PTE Tobit Drake Hall 2003 Japan 149 banks 1997 Input Input OTE PTE SE na Electronic copy available at httpsssrncomabstract1350544 48 Erdem Erdem 2008 Turkey 10 banks 19982004 Input Input OTE AE CE na Fare et al 2004 US 858938 banks 19901994 Both Transaction deposits Output Non transaction deposits input TE AE SPE na Fiordelisi 2008 France Germany Italy UK Varies from 68 UK 1997 to 2183 Germany 2002 19972002 Input Output PTE SE AE CE and PE using SFA na Fukuyama Weber 2005 Japan 110141 banks 1073 obs 19921999 Output TE Input AE Input Luenberger Farrell PTE AE na Fukuyama Weber 2002 Japan 138141 banks 19921996 Both Input TE Output AE Output TE Input AE Input part of funds from customers Productivity na Fung 2006 US 135 banks 1080 obs 19962003 Input Input PTE SE Productivity OLS GarciaCestona Surroca 2008 Spain 226 Obs 19982002 Output Output IP AE IPR na Gilbert Wilson 1998 Korea 1525 banks 19801994 Output Demand deposits output Time savings deposits input TE Productivity na Guzman Reverte 2008 Spain 14 banks 20002004 Both Input OTE PTE Productivity na Habibullah et al 2005 Malaysia 37 banks 19881993 Input Input OTESEPTE Congestion Efficiency Granger causality Hahn 2007a Austria Around 800 banks 19962002 Input na in Model 1 Input in Model 2 PTE Tobit Bootstrapped Hahn 2007b Austria More than 800 banks 19952002 Input Profit oriented na Int App Input PTE Tobit Halkos Salamouris 2004 Greece 1518 banks 19971999 Output na in basic model Interest expense input in benchmark model OTE PTE SE and superefficiency measures na Hauner 2005 Germany Austria 97 banks 485 obs 19951999 Input Input PTE SE AE CE Productivity Tobit Havrylchyk 2006 Poland 247 Obs 19972001 Input Input OTE PTE SE AE CE Tobit Hu et al 2008 Taiwan 14 banks 2007 Input na OTE PTE SE na Electronic copy available at httpsssrncomabstract1350544 49 Huang et al 2008 Taiwan 42 banks 20012004 Output Input Productivity Regression Huang Wang 2002 Taiwan 22 banks 19821997 Input Input CE na Isik 2008 Turkey 794 obs 19811996 Input Input OTE PTE SE Productivity na Isik 2007 Turkey 51 banks 439 obs 19811990 Output Input Productivity GLS Isik Hassan 2003c Turkey 54 banks 19921996 Input Ef Output Prod Input part of loanable funds OTE PTE SE Productivity na Isik Akcaoglu 2006 Turkey 28 banks 19811990 Output Input part of loanable funds Productivity na Isik Hassan 2002 Turkey 36 banks in 1998 50 banks in 1992 53 banks in 1996 1988 1992 1996 Input Input OTE PTE SE AE CE also CE PE from SFA GLS Isik Hassan 2003a Turkey 39 54 56 banks 1988 1992 1996 Input Input OTE PTE SE AE CE GLS Tobit Isik Hassan 2003b Turkey 3856 banks 458 obs 19811990 Input Ef Output Prod Input OTE PTE SE Productivity na Isik Uysal 2006 Turkey 439 obs 19811990 Output Input part of loanable funds Productivity na Kao Liu 2004 Taiwan 24 banks 2000 Input Input OTE na Kao Liu 2008 Taiwan 25 banks 19972001 Both min max problems Purchased funds input Demand deposits output OTE na Kirkwood Nahm 2006 Australia 10 banks 19952002 Input Input PTE SE AE CE APE Productivity na Kohers et al 2000 US 94 bidder 94 targets 8960 obs peers 19901995 Input Input Output CE na Krishnasamy et al 2003 Malaysia 10 banks 20002001 Output Output Productivity na Kumar Gulati 2008 India 27 banks 20042005 Input Loanable funds input OTE Super eff Regression Kuosmanen Post 2001 EU 453 banks 1997 Input Debt capital input TE AE CE na Kyj Isik 2008 Ukraine 150 banks 883 obs 19982003 Input Input part of loanable funds OTE PTE SE GLS Laurenceson Qin 2008 China 65 banks 20012006 Input Interest expenses input Deposits output CE Tobit Leightner Lovell 1998 Thailand 31 banks 19891994 Input na Productivity na Electronic copy available at httpsssrncomabstract1350544 50 Lim Randhawa 2005 Hong Kong Singapore 19 banks 19951999 Input Production output Intermediation input OTE PTESE na Lin et al 2007 Taiwan 37 banks 20022003 Output Interest expenses input OTE Productivity na Liu 2008 Taiwan 24 banks 2000 Input Input OTE na LozanoVivas et al 2002 10 European 612 banks 1993 Input Output PTE na LozanoVivas et al 2001 10 EU countries 612 banks 1993 Input Output PTE na Luo 2003 US 245 banks 2000 Input na OTE PTE SE na Maghyereh 2004 Jordan 14 banks 19842001 Input Input OTE PTE SE Productivity Tobit Maudos et al 2002 Spain 1666 Obs 19851996 Input Price of loanable funds Input Loanable funds output CE Tobit Maudos Pastor 2003 Spain 5077 savings 7598 national 19851996 Input CE Output PE Input CE APE SPE na Mercan et al 2003 Turkey 545 obs 19891999 Input na OTE na Mostafa 2007b GCCGulf Coopn Council 43 banks 2005 Output na PTEOTESE na Mostafa 2007a Arab banks 85 banks 2005 Output na PTEOTESE na Mukherjee et al 2003 India 27 banks 19972000 Output Output PTE na Mukherjee et al 2001 US 201 banks 19841990 Input Input OTE PTE GLS Neal 2004 Australia 1226 banks 19951999 Input Ef Output Prod Demand deposits output Term deposits certificates of deposits other deposits input PTE SE AE CE Productivity na Noulas 2001 Greece 19 banks 19931998 Input Interest expense input OTE na Olgu WeymanJones 2008 EU12 and EU10 164 banks 19972001 Output Input Productivity na Pasiouras 2008b Greece 1218 banks 78 obs 20002004 Input Deposits Input in Models 14 Interest expense input in Model 5 PTE SE Tobit Electronic copy available at httpsssrncomabstract1350544 51 Pasiouras 2008a 95 countries 715 bank 2003 Input Input OTE PTE SE Tobit Pasiouras et al 2008 Greece 10 banks 2002005 Input Interest expense input OTE PTE SE na Pastor 1999 Spain na 19851995 Input na OTE PTE SE na Pastor 2002 Spain Italy Frane Germany 2598 Obs 19881994 Input Output CE Tobit Logistic Pastor Serrano 2006 10 European 540 banks 3780 obs 19921998 Input Input CE COMPE ISPE na Paul Kourouche 2008 Australia 10 banks 90 obs 19972005 Input Interest expense input OTEPTESE na Prior 2003 Spain na 1986 1990 1995 Input Number of current savings accounts output SRCE LRCECAPE na Ramanathan 2007 GCC 55 banks 20002004 Input Input OTE PTESE Productivity na Ray Mukherjee 1998 US 201 banks 19841990 Input Transaction deposits Input PTESESZE Tobit Resti 1998 Italy 67 merger deals 19871995 Input Output PTE Extra efficiency na Rezitis 2006 Greece 6 banks 19821997 Output Input OTE PTE SE Productivity Tobit Rizvi 2001 Pakistan 37 banks 19931998 Both Output OTEPTESE Productivity na Saha Ravisankar 2000 India 25 banks 19921995 Input Interest expense input Deposits output OTE na Sahoo Tone 2008 India 78 banks 19972001 Both Input PCU TE RECU OCI na Sathye 2003 India 94 banks 19971998 Input Interest expense input in Model 1 Deposits input in Model 2 PTE na Sathye 2001 Australia 29 banks 1996 Input Time saving deposits input Demand deposits output OTE AE CE Type of regression not specified Sathye 2002 Australia 17 banks 19951999 Output Interest expenses input Productivity Type of regression not specified Seiford Zhu 1999 US 55 banks 1995 Output oriented Input na OTE PTE Electronic copy available at httpsssrncomabstract1350544 52 congestion Simar Wilson 2007 US 322 banks 6955 banks Q4 2002 na Input TE Double Bootstrap truncated regression Stavarek 2006 EU11 countries 122126 banks 20012003 Input Input PTE SE na Sturm Williams 2004 Australia 1026 banks 273 obs In total 19882001 Input Input OTE PTE SE Productivity na Sueyoshi Kirihara 1998 Japan 136 banks na Input na OTE na Sufian 2008 Malaysia 3336 banks 171observations 19951999 Input Model 1 input Model 2 output with interest expense as input Model 3 interest expense input OTE PTE SE Tobit Sufian 20007 Singapore 141 banks 19932003 Input Input OTEPTESE na Sufian Majid 2007 Singapore 6 banks 19932003 Input Input PTE na Sufian Majid 2006 Malaysia 9 banks 20022003 Both Interest expenses input OTE PTE SE na TortosaAusina 2002c Spain 116121 banks 19861995 Input Model 1 Input Model 2 savings time deposits output savings deposits other deposits and interbank deposits input CE na TortosaAusina 2002a Spain 104 banks 19851997 Input Model 1 Input Model 2 Input Output CE na TortosaAusina 2002b Spain 104 banks 19851997 Input Input CE na TortosaAusina 2003 Spain 120162 banks 19861997 Input Input CE na TortosaAusina et al 2008 Spain na 19921998 Output Both OTE Productivity na Tsionas 2003 Greece 1719 banks 19931998 Input Ef Output Prod Input OTE PTE SE AE CE Productivity na Valverde et al 2007 Spain 77 banks 1540 obs 19922001 Input Input OCE ICE na Electronic copy available at httpsssrncomabstract1350544 53 Wang Huang 2007 Taiwan 22 banks 19822001 Input Input OTE OTEQFI AE CE AR GMM Correlation Markov Webb 2003 UK 68 banks 19821995 Input Input OTE PTE SE na Weill 2004 France Germany Italy Spain Switzerland 688 banks 19921998 Input Interest expense input CE Correlation Wheelock Wilson 2000 US 13162555 banks 19841993 Both Input PTE and CE with SFA na Yildirim 2002 Turkey 3859 banks 594 obs 19881999 Input Input OTE PTE SE na Zen Baldan 2008 Italy 121 banks 20012005 Input Output PTE Productivity na Zhao et al 2008 India 65 banks 845 obs 19922004 Input Input part of loanable funds OTE PTESE Productivity na Notes Classified in 2008 on the basis of doi OTE Overall Technical Efficiency ie CRS PTE Pure Technical Efficiency ie VRS SE Scale Efficiency AE Allocative Efficiency CE Cost Efficiency APE Alternative Profit Efficiency SPE Standard Profit Efficiency Productivity Various Productivity Growth Measures SRCE Shortrun cost efficiency LRCE Longrun cost efficiency CAPE Capacity efficiency PCU Physical capacity utilization RECU Ray economic capacity utilization OCI Optimal capacity idleness IP Performance index of PTE result of private negotiations among stakeholders AE A proxy of Allocative Efficiency calculated without including input prices IPR Index of Efficiency with preferences revealed by the legislator IPR AE x IP COMPE Composition efficiency ISPE Intraspecialization efficiency OCE Operating cost efficiency ICE Interest cost efficiency OTEQFI OTE in the presence of quasifixed inputs SZE Size efficiency PTE OTE these studies report measures which they label as Pefficiency or cost efficiency simply on the basis of some inputsoutputs that they use but they do not include inputsoutput price inputs so in a sense they present measures of technical efficiency Electronic copy available at httpsssrncomabstract1350544 54 Appendix II Studies in branch efficiency and productivity growth Authors Inputs Outputs InputOutput Oriented Approach for inputoutput selection Estimated measures Sample size Country Portela Thannasoulis 2007 INPUT SET1 TRANSCT EFF 1 Number ETMs ATMs CATs 2 Rent 3 No clients not registered INPUT SET2OPRTNL EFF 1 Number of staff 2 Rent INPUT SET3PROFIT EFF 1 Number of staff 2 Supply costs OUTPUT SET1 TRANSCT EFF1 No new registrations for internet use 2 No transactions in CATs 3 No deposits in ETMs OUTPUT SET2 OPRTNL EFF 1number of clients 2 Value current accounts 3 Value other resources 4 Value titles deposited 5 Value credit by bank 6 Value credit by associates 7 Number transactions OUTPUT SET3PROFIT EFF1 Value current accounts 2 Value other resources 3 Value credit over bank 4 Value credit associates Output except profit efficiency nonoriented DEA Intermediation Transactional Operational Profit 57 Portugal Portela Thannasoulis 2005 1Number of staff 2Supply costs 1Value current accounts 2Value other resources 3Value credit by bank 4Value credit associates Nonoriented Intermediation Overall profit efficiency technical profit efficiency allocative profit efficiency scale and mix effects 57 Portugal Portela et al 2003 1staff costs 2other operating costs 1 Value of current accounts 2Value of credit 3 Interest revenues Nonoriented Intermediation PTE 24 Portugal Camanho Dyson 2008 1Number of branch and account managers 2Number of administrative and commercial staff 3Number of tellers 4Operational costs excluding staff costs And input prices 1 Total value of deposits 2 Total value of loans 3 Total value of off balance sheet business 4 Number of general service transactions Input DEA Production OTE Allocative efficiency and Farrell CE measures market efficiency and economic efficiency 39 Portugal Electronic copy available at httpsssrncomabstract1350544 55 Camanho Dyson 2006 1Number of branch employeesincluding branch and account managers administrative and commercial staff and tellers2 Operational costs excluding staff costsAnd input prices 1 Total value of deposits 2 Total value of loans 3 Other revenue 4 Number of general service transactions Input DEA Production Overall regional performance within region efficiency spread and frontier productivity na Portugal Camanho Dyson 2005 1 Number of branch and account managers 2 Number of administrative and commercial staff 3Number of tellers 4Operational costs excluding staff costs 1Number of general service transactions Input DEA Production OTE Farrell CE Optimistic CE and Pessimistic CE 144 Portugal Camanho Dyson 1999 1 Number of employees in the branch 2 Floor space of the branch 3 Operational costs costs of supplies and other services 4 Number of external ATMs 1Number of general service transactions performed by branch staff 2Number of transactions in external ATMs 3 Number of all types of accounts at the branch 4 Value of savings 5Value of loans Both Production PTE SE 168 Portugal Das et al 2007 1 of different categories of employees 1value of deposits 2value of credit 3noninterest income Input DEA Intermediation Labor use efficiency Regional or spatial efficiency 222 India Giokas 2008a 1Personnel costs 2Running costs 3Operating expenses 1Value of loans 2Value of deposits 3Noninterest income Input DEA Intermediation Operating efficiency 171 Greece Giokas 2008b INPUT SET1Production efficiency 1Personnel costs 2Running and other operating costs INPUT SET2 Transaction efficiency 1Personnel costs 2Running costs and other operating costs INPUT SET3 Intermediation efficiency 1Interest costs 2Non interest costs OUTPUT SET1 1Value of loan portfolio 2Value of deposits 3Non interest income OUTPUT SET 2 1Loan transactions 2Deposit transactions 3Remaining transactions OUTPUT SET 3 1Interest income 2Noninterest income Input DEA Production Intermediation Production transaction Intermediation specific efficiency 44 Greece Electronic copy available at httpsssrncomabstract1350544 56 Porembski et al 2005 1 of Employees 2 Office space 1Private demand deposits 2Business demand deposits 3Time deposits 4Saving deposits 5Credits 6Bearer securities 7Recourse guarantees 8Bonds 9Investment deposits10 Insurances contracts 11Contributions to a building society contracts Input DEA Production PTE SE 141 German Paradi Schaffnit 2004 INPUT SET 1 Model 1 Production model 1staff 2information technology 3premises 4other non interest expenses INPUT SET 2 Model 2 Strategic model 1staff 2information technology 3premises 4other non interest expenses 5nonaccrual loans Environmental factors Growth factor risk rating incorporated into Models OUTPUT SET1 1Deposits2Loans3Fee income4Maintenance activities OUTPUT SET2 1Deposits 2Loans 3Fee income4 Deposit spread 5 Loan spread Both Intermediation OTE PTE Effectiveness Cost effectiveness 90 Canada Sowlati Paradi 2004 1FTE sales 2FTE support 3FTE other 1Loans 2Mortgages 3RRSPs 4Letters of Credit Input DEA Production PTE 79 Canada Athanassopoulos Giokas 2000 INPUT SET 1 1labor hrs 2branch size 3 computer terminals 4operating expenditure INPUT SET 2 1labor costs 2operating expenses 3running costs of the building OUTPUT SET1 1st Stage1Group A transactions 2 Group B transactions 3Group C transactions 4Group D transactions 2nd stage 1Credit transactions 2 Deposit trans 3 Foreign receipts OUTPUT SET 2 1savings deposits2current deposits 3demand deposits4time deposits 5total loans 6noninterest income Input DEA Intermediation Production Intermediation market efficiency 47 Greece Cook Hababou Tuenter 2000 1FSE service staff2 FSA sales staff 3FSU support staff 4FOT other staff 1MDP counter level deposits 2 MTR transfers between accounts 3RSP retirement savings plan openings 4MOR mortgage accounts opened Production Aggregate Sales Service efficiencies 20 Canada Electronic copy available at httpsssrncomabstract1350544 57 Cook Hababou 2001 Service 1FSEtotal number of full time equivalent service staff Sales 1FSAtotal number of full time equivalent sales staff Shared 1FSU total number of full time equivalent support staff 2FSTtotal number of full time equivalent other staff Service 1TOTMENUtotal number of menu account transactions 2VISAnumber of Visa cash advances 3 CADnumber of commercial deposit transactions Sales 1RSPnumber of RSP account openings 2MORTnumber of mortgages transacted 3BPLnumber of variable rate consumer loans transacted Input DEA Production Sales service efficiency 20 Canada Golany Storbeck 1999 1Teller Labor 2NonTeller 3Labor Retail Sq Feet 4Marketing 5Employment rate 1Total Loans 2Total Deposits 3Depth 4Satisfaction Output DEA Production Operational efficiency 182 USA Hartman et al 2001 1No of staff 2No of computer terminals3square meters of premises 1Amount of deposits 2amount of loans 3Amount of house mortgages Nondiscretionary variables no of customers Input DEA Production OTE PTE AE 50 Sweden Kantor Maital 1999 INPUT SET 1 Customer services 1Labor costs MSincluding social benefits 2services 3area square meters for services only INPUT SET 2 Bank transactions 1Labor costs MSincluding social benefits2 transactions only 3area square meters OUTPUT SET 1 1number of demand deposit accounts 2weighted output indexcustomer service transactions 3queuereplacing actions OUTPUT SET 2 1Credit cards regular and gold 2weighted output index transactions3 commissions on import and export commercial accounts 4savings accounts activity DEA Production customer service efficiency transaction efficiency 250 na Parkan Wu 1999 Cost 1salaries 2 staff benefits 3 electronic data processing expenses 4 occupancy 5 temporary staff expenses6 printing and stationary Revenue 1 LC letter of credit advice 2 LC confirmation 3 LC issuance 4 guarantee 5 acceptance 6 negotiation and 7 net interest revenue Input DEA Intermediation DEA scaled inefficiency 24 Hong Kong Wu et al 2006 1Personnel 2Other general expenses 1Deposits 2 Revenues 3 Loans Input DEA Intermediation DEA efficiency DEA NN efficiency 142 Canada Electronic copy available at httpsssrncomabstract1350544 58 Dekker Post 2001 1Front office personnel 2Facilitating personnel Total revenue Input DEA Intermediation Revenue generating efficiency 314 Netherlands Soteriou Zenios 1999 INPUT SET 1 Operational efficiency 1st Set of Inputs 1Managerial personnel 2Clerical personnel 3Computer terminals 4Space 2ndset of inputs1Current accounts 2Savings accounts 3Foreign currency and commercial accounts 4Credit applications 5 Commissions INPUT SET2 Service Quality Efficiency Model 1st Set of Inputs 1Managerial personnel 2Clerical personnel 3Computer terminals 4Space 2ndset of inputs1Current accounts 2Savings accounts 3Foreign currency and commercial accounts 4Credit applications 5 Commissions INPUT SET3 Profitability efficiency 1st Set of Inputs 1Managerial personnel 2Clerical personnel 3Computer terminals 4Space 2ndset of inputs1Current accounts 2Savings accounts 3Foreign currency and commercial accounts 4Credit applications 5 Commissions OUTPUT SET 1 Total amount of work produced OUTPUT SET 2 The level of service quality achieved ie Operational Internal MarketingPerceived measures OUTPUT SET 3 Profit Both Production Operational efficiency Service Quality Efficiency Profitability Efficiency 144 Cyprus Zenios et al 1999 INPUT SET1 1Managerial personnel 2Clerical personnel 3Computer terminals 4Space INPUT SET 2 1Current accounts 2Savings accounts 3Foreign currency 4commercial accounts Credit applications Total amount of work produced Input DEA Production Operating efficiency 144 Cyprus Sherman Rupert 2006 Personnel FTEs 1 Platform 2Teller3Manager 4Operating Expenses Supplies Telephone Travel Postage 1 Teller Transactions Deposits Withdrawals Checks Cashed Bank Checks Loan Payments Bonds Sold Redeemed Coupons 2MarketingNew Accounts 3Night Deposits 4Safe Deposit Visits 5ATMs Serviced Input DEA Production OTE 217 branches from 4 merged banks USA Electronic copy available at httpsssrncomabstract1350544 59 6Loans Mortgage Consumer Referrals Applications Closings Howland Rowse 2006 1Nonsales FTE2sales FTE 3 Size in total square feet 4City employment rate 1Loan volume 2 Deposit volume 3 Average number of productscustomer 4 customer loyalty Both Production PTE 162 Canada Sherman Zhu 2006 1Platform FTSe 2 Teller FTEs 3Management FTEs 4 Postage supplies telephone travel expenses 1Deposits withdrawals checks cashed 2Bank checks 3Bond transactions 4Night deposits 5Safe deposits visits 6New accounts 7Mortgage and consumer loans 8 ATMs Quality measuremystery shopper scores Input DEA Production OTE 225 US Noulas et al 2008 Value of 1Personnel costs 2Other operating expenses Value of 1Deposits 2Financial products 3Loans 4 Other loans Input DEA Intermediation OTE 58 Greece Electronic copy available at httpsssrncomabstract1350544 60 Appendix III Studies in bank bankruptcy credit rating and underperformance Part A Authors Subject Country Training sample Period covered Techniques Used Alam et al 2000 BF US Training 3 failed 17 extreme performance 80 healthy Validation not specified although the authors mention that they repeat their analysis in a second sample of another 100 obs with qualitatively similar results 1992 FC CLNN SOMNN Boyacioglu et al 2008 BF Turkey Training 14 failed 29 nonfailed Validation 7 failed 15 nonfailed 19972003 MLPNN CLNN SOMNN LVQNN SVMs MDA CA LRA Chen Shih 2006 CR Taiwan Training twAA higher ratings 16 TwA 25 TwBBB 26 tw BB lower ratings 6 Validation twAA higher ratings 6 TwA 9 TwBBB 9 tw BB lower ratings 2 19982003 SVMs BPNN LR Gaganis et al 2006 CR 79 countries Training Fitch ratings A B 466 C 214 D E 214 Validation 10fold crossvalidation 2004 UTADIS MDA OLR Huang et al 2004 CR US Taiwan Training Taiwan AAA 8 AA 11 A 31 BBB 23 BB 1 US AA 20 A 181 BBB 56 BB 7 B 1 Validation 10fold cross validation leaveoneout crossvalidation Taiwan 19982002 US 1991 2000 SVMs BPNN LRA Kolari et al 2002 BF US Training 2 samples from 1989 Failed banks 18 Nonfailed banks 1012 1061 Validation Various samples from 1990 1991 1992 Failed banks 6 18 Nonfailed banks 1053 1178 19891992 TR LRA Kosmidou Zopounidis 2008 BF US Training 4 samples from 19931999 Failed banks 1989 Nonfailed banks 90455 Validation 20002003 Failed banks 23 Nonfailed banks 112 19932003 UTADIS MDA Lanine Vander Vennet 2006 BF Russia Training 4 samples from January 1997March or August 2000 Failed banks 5878 Nonfailed banks 290390 Validation 4 samples from April or September 2000 to November 2003 or January or June 2004 19972004 MTR TR LRA Ng et al 2008 BF US Training 20 of total sample Failed 358548 Nonfailed 25552585 Validation 80 of total sample Failed 358 548 Nonfailed 2555 2585 19802000 FCMACCRISS GenSoFNNCRISS Coxs hazard model Pasiouras et al 2007 CR 9 Asian countries Training Fitch ratings A AB 4 B BC 31 C CD 69 D DE 77 E 34 Validation 10fold crossvalidation 2004 MHDIS MDA LRA Electronic copy available at httpsssrncomabstract1350544 61 Ravi et al 2008 BU US Training 640 adequate performance 160 poor performance banks the 160 poor performance banks where then duplicated to produce a total of 640 poor performance cases Validation 200 banks 1993 MLFFBP PNN RBFN SVM CART FRBC PCAMLFFBP Ravi Pramodh 2008 BF Turkey Spain Training Turkey 40 Spain 66 Validation 10fold crossvalidation na PCNN PCATANN PCABPNN TANN BPNN Swicegood Clark 2001 BU US Smallcommunity banks Adequate performance 480 training 320 validation Underperformance 120 training 80 validation Large regional banks Adequate perf 356 training 237 validation Underperformance 89 training 59 training 1993 NN MDA Tung et al 2004 BF US Failed 548 Nonfailed 2555 Training 20 Validation80 19802000 GenSoFNNCRI Cox Zhao et al 2008 BF US Training 240 failed 240 nonfailed Validation 10fold crossvalidation 19911992 LRA C45 decision tree BPNN kNN Notes BF Bank Failure CR Credit ratings BU Bank underperformance FC Fuzzy clustering MLPNN Multilayer perception neural networks CL NNCompetitive learning NN SOMNN Selforganizing map NN LVQNN Learning vector quantization NN SVMs Support Vector Machines MDA Multivariate discriminant analysis CA kmeans Cluster Analysis LRA Logistic Regression Analysis TR Trait Recognition MTR Modified Trait Recognition BPNN Back propagation NN LR Linear regression GenSoFNNCRI Generic Selforganising Fuzzy Neural NetworksCRI Scheme kNN knearest neighbour UTADIS UTilités Additives DIScriminantes MHDIS MultiGroup Hierarchical Discrimination Electronic copy available at httpsssrncomabstract1350544 62 Appendix III Studies in bank bankruptcy credit rating and underperformance Part B Authors Variables considered Classification resuls Alam et al 2000 1 Net income to total assets 2 Net Loan losses to Adjusted assets 3 Net loan losses to total loans 4 nonperforming loans to total assets 5 net loan losses provisions to net income na Boyacioglu et al 2008 1 Shareholders equity total assets 2 shareholders equity total loans 3 shareholders equity net profit total assets offbalance sheet commitments 4 permanent assets total asses 5 total loanstotal assets 6 loans under followuptotal loans 7 specific provisiontotal loans 8 specific provisionloans under followup 9 personnel expenses average assets 10 net profitaverage assets 11 net profitaverage shareholders equity 12 income before taxesaverage assets 13 interest incometotal operating income 14 noninterest expensestotal operating income 15 liquid assetstotal assets 16 total loanstotal deposits 17 trading securitiestotal assets 18 FX assetsFX liabilities 19 net interest incomeaverage assets 20 net on balance sheet positiontotal shareholders equity Note reduced or normalized sets of variables were used in some models Correct accuracies MLPNN 9550 CLNN 6818 SOMNN 6363 LVQ NN 100 SVMs 9090 MDA 6818 CA 8181 LRA 8181 Chen Shih 2006 1 of shares held by the government 2 of shares held by major shareholders above 10 3 average stock price in previous year 4 total assets 5 liabilities 6 shareholders equity 7 net sales 8 operating income loss 9 liabilities to assets 10 deposits to net worth 11 fixed assets to net worth 12 current reverse ratio 13 total assets turnover times 14 deposit to loan 15 nonperforming loan ratio 16 interest payments to annual average balance of deposits ratio 17 interest income to annual average balance of credit extension ratio 18 average amount of business income per employee 19 average amount of profit per employee 20 return on total assets 21 return on shareholders equity 22 operating income to paidin capital 23 net profit to sales 24 EPS 25 cash flow adequacy ratio Note some models use subsets of these variables Correct accuracies SVMs 73088462 BPNN 506538 LR 2308 50 Gaganis et al 2006 1 EquityTotal assets 2 Loan loss provisions net interest revenue 3 Return on average assets 4 Cost to income ratio 5 Liquid assets Customer short term funding 6 Log total assets 7 Number of subsidiaries 8 Number of institutional shareholders 9 Whether the banks is listed on a stock exchange or not 10 Whether the country is developed or developing UTADIS 6891 MDA 6506 OLR 6288 Huang et al 2004 1 Total assets 2 Total liabilities 3 Longterm debts total invested capital 4 Debt ratio 5 Current ratio 6 EBITinterest 7 Operating profit margin 8 Shareholders equity longterm debt fixed assets 9 Quick ratio 10 Return on total assets 11 Return on equity 12 Operating incomereceived capitals 13 Net income before tax received capitals 14 Net profit margin 15 EPS 16 Gross profit margin 17 Nonoperating income sales 18 Net income before tax sales 19 Cash flow from operating activities current liabilities 20 cash flow from operating activities capital expenditures increase in inventory cash dividends in last 5 years 21 cash flow from operating activities cash dividends fixed assets other assets working capitals Note Some models were developed with subsets of these variables Correct accuracies SVMs Taiwan 7568 7973 US7887 8038 BP NN Taiwan 7432 7568 US 7568 8075 LRA Taiwan 7027 7568 US 7509 7698 Kolari et al 2002 1 Total assets millions 2 Net interest incometotal assets 3 Net income after taxes total assets 4 Total equitytotal assets 5 Allowance for loan lossestotal assets 6 Provisions for loan lossestotal assets 7 Net loan chargeoffs total assets Correct accuracies TR around 85 LRA slightly Electronic copy available at httpsssrncomabstract1350544 63 8 Total securities total assets 9 Nondeposit liabilities total liabilities 10 Certificates of deposit total deposits 11 Total loans leases total assets 12 Sum of key asset accounts total assets 1324 maximum change in each preceding ratio mean of corresponding ratio 25 Maximum change in loans past due at least 90 daysmean of numerator 26 Bank Holding Co total assets millions 27 Bank total assets BHC total assets 28 Maximum change in total assets mean change in total assets better than chance Kosmidou Zopounidis 2008 26 variables were initially considered After a combination of correlation analysis ANOVA KruskalWallis MannWhitney the following 9 variables were used 1 Return on current assets 2 Financing of current assets 3 Operating income current assets 4 Income before taxes exceptional outcome provisions for loan losses net depreciations 5 Asset per employee 6 Provisions for precarious loans overdue loans 7 loans internal deposits 8 Tier 1 ratio 9 Ratio of precarious capital Correct accuracies UTADIS 7575 8544 DA 75757935 Combined model 836 Lanine Vander Vennet 2006 1 Net incometotal assets 2 Liquid assetstotal assets 3 Government debt securitiestotal assets 4 Capital total assets 5 Overdue loans overdue promissory notes total loans 6 Total loanstotal assets 7 Log total assets Mean Square Error TR 0091 0167 MTR 0072 0149 LRA 0098 0138 AUCROC TR 0570 0745 MTR 0539 0906 LRA 0511 0752 Ng et al 2008 1 Average total equity capital average total assets 2 Average accumulated loan loss allowance average gross total loans and lease 3 Average accumulated loans 90 days late average gross total loans leases 4 Annual loan loss provision average gross total loans and leases 5 Noninterest expense operating income 6 total interest income interest expense average total asset 7 net after tax income applicable income taxes average total equity capital 8 average cash average federal funds sold average total deposit average fed funds purchased average banks liability on acceptance average other liabilities 9 change in Annual Gross total loans leases Note Additional models with 3 selected variables were also developed FCMACCRISS9 inputs 8207 year 3 to 9111 year 1 FCMACCRISS 3 inputs 8376 year 3 to 9528 year 1 GenSoFNNCRISS Cox vs FCMACCRISvarious scenarios as for misclassification costs of failed vs survived banks FCMAC better in minimizing Error I ie classiyfing a failed bank as survived FCMAC vs GenSoFNN no difference in year 1 error rate of GenSoFNN almost double in years 2 3 Pasiouras et al 2007 19 financial variables were initially considered Using factor analysis the following 5 financial variables were selected 1 Equity Customer short term funding 2 Net interest margin 3 Liquid assets Total deposits borrowings 4 Net loans Total deposits borrowings 5 Return on average equity In addition the following 5 nonfinancial variables were used 6 Number of institutional shareholders 7 Number of subsidiaries 8 Auditors opinion 9 Whether the bank is listed or not 10 Heritage Correct accuracies MHDIS 6603 MDA 5373 LRA 4755 Electronic copy available at httpsssrncomabstract1350544 64 bankingfinance index Ravi et al 2008 Ravi Pramodh 2008 Turkey 1 Interest expenses average profitable assets 2 Interest expenses average nonprofitable assets 3 Shareholders equity total incomedeposits nondeposit funds 4 Interest income interest expenses 5 Shareholders equity total income total assets 6 Shareholders equity total incomeTotal assets contingencies commitments 7 Networking capital total assets 8 Salary employees benefits reserve for retirement no of personnel 9 Liquid assets deposits nondeposit funds 10 Interest expense total expenses 11 Liquid assets total assets 12 Standard capital ratio Spain 1 Current assets total assets 2 Current assets cash total assets 3 Current assets loans 4 Reserves loans 5 Net income total assets 6 Net income total equity capital 7 Net income loans 8 Cost of sales sales 9 Cash flow loans Note in the case of Spain Turkey 6 7 variables are selected when feature subset selection algorithm is used Correct accuracies PCNN WOFSLTF Spain 966 Turkey 975 PCNN WFSLTF Spain 925 Turkey100 TANN Spain 916 Turkey 925 PCATANN Spain 975 Turkey 975 PCABPNN Spain 841 Turkey 85 BPNN Spain 8167 Turkey 875 PCNNWOFSSTF Spain 824 Turkey 925 PCNNWFSSTF Spain 866 Turkey 90 Swicegood Clark 2001 1 Earning assets Total assets 2 Total securities Total assets 3 Total loans Total asses 4 Nonperforming assets Total assets 5 Allowance for loan losses Total loans 6 Asset growth 7 Core deposits Total assets 8 Volatile liabilities Total liabilities 9 Total equity Total assets 9 Noninterest income Total assets 10 Interest income Total assets 11 Gains Losses from sale of securities Total assets 12 Salary Total assets 13 Noninterest expense SalaryTotal assets 14 Total interest expense Total assets 15 Provision expense Total loans 16 Off balance sheet commitments Total assets 17 Regional geographic location 18 Charter 19 Holding company affiliation 20 Federal Reserve Bank member 21 Branch or Unit bank 22 Deposit insurance Correct accuracies NN Large Regional Underperformance 661 Adequate performance 852 Small Community Underperformance 600 Adequate performance 828 MDA Large Regional Underperformance 508 Adequate performance 953 Small Community Underperformance 287 Adequate performance 921 Tung et al 2004 Average total equity capital average total assets 2 Average Accumulated loan loss allowance average gross total loans and leases 3 Average accumulated loans 90 plus days late average gross total loans and leases 4 Annual loan loss provisions Equal error rates GenSoFNNCRI 817 Electronic copy available at httpsssrncomabstract1350544 65 average gross total loans and leases 5 Noninterest expense operating income 6 Total interest income interest expense average total assets 7 Net income After tax applicable income taxes average total equity capital 8 Average cash average federal funds soldaverage total deposits average fed funds purchased average banks liability on acceptances average other liabilities 9 change in annual Gross total loans and leases 3610 Errors under different misclassification costs Cox Type I error 6 54 Cox Type II error 310 836 GenSoFNN CRI Type I error 0 069 GenSoFNNCRI Type II error 4545 100 Zhao et al 2008 93 raw accounting variables 26 fianancial ratios 1 Total equity total assets 2 Gross loans total assets 3 Commercial loans gross loans 4 Individuals loans gross loans 5 Real estate loans gross loans 6 Late loans 90 days gross loans 7 Not accruing loans gross loans 8 Loan loss provisions gross loans 9 Chargeoff loans gross loans 10 Loan allowance gross loans 11 Total loans netoff unearned total equity 12 Net income operating income 13 Net income total assets 14 Operating expenses total assets 15 Operating income total assets 16 Interest income interest expenses 17 Operating income operating expenses 18 Net income total assets 19 Net income total equity 20 Interest expenses for deposits Total deposits 21 Interest income from Loans Gross loans 22 Interest income operating income 23 Cash total assets 24 Cash Securities Total assets 25 Cash Federal funds sold US Treasury US Government Obligations total assets 26 Gross loans total deposits Misclassifications costs under different prior probabilities and cost ratios With raw accounting items LRA00870287 Tree 00760298 BPNN 0091 0417 kNN 0071 0297 With financial ratios LRA 0053 0199 Tree 0063 0257 BPNN 0057 0240 kNN 0057 0272 Electronic copy available at httpsssrncomabstract1350544 66 Electronic copy available at httpsssrncomabstract1350544 Production 263 551561 julset 2016 httpdxdoiorg10159001036513185714 1 Introduction The Brazilian banking sector has experienced vast changes over the last years This provoked a wave of banking mergers and acquisitions MAs and the penetration of some foreign banks into the Brazilian retail banking market Such processes focus on providing the institutions greater power in this competitive environment Coupled to this there are numerous innovative processes related to management which also allow banks to not only further improve customer service but also to offer a greater variety of services at lower costs The objective of this paper is to analyze by applying the bootstrap DEA the efficiency of banks in Brazil using a financial intermediation To reach the proposed goal we use the database entitled The Largest Banks from 2010 to 2013 which is periodically published by the Central Bank of Brazil The banking literature recognizes the vast progress made by banks with regard to its main functions and responsibilities Over time practically everywhere in the world the changes seen in the banks function are remarkable with new products and services increasingly available to customers However the financial intermediary role still prevails In fact all the other features incorporated into the functions of a bank are still the result of its primary function which is financial intermediation The structure of the paper is as follows Section two presents a brief literature review on bank efficiency Section three describes the data the employed methodology and the proposed DEAbased Section four discusses the results obtained from the application of the model and the final remarks are in section five Received Oct 23 2014 Accepted Nov 12 2015 Estimating the efficiency from Brazilian banks a bootstrapped Data Envelopment Analysis DEA Ana Elisa Péricoa Naja Brandão Santanab Daisy Aparecida do Nascimento Rebelattoc aDepartamento de Economia Faculdade de Ciências e Letras Universidade Estadual Paulista Araraquara SP Brasil bCentro de Ciências da Natureza Universidade Federal de São Carlos Buri SP Brasil cDepartamento de Engenharia de Produção Escola de Engenharia de São Carlos Universidade de São Paulo São Carlos SP Brasil anaelisafclarunespbr Abstract The Brazilian banking sector went through several changes in its structure over the past few years Such changes are related to fusions and acquisitions as well as the largest market opening to foreign banks The objective of this paper is to analyze by applying the bootstrap DEA the efficiency of banks in Brazil in 20102013 The methodology was applied to 30 largest banking organizations in a financial intermediation approach In that model the resources entering a bank in the form of deposits and total assets are classified as inputs and besides these manual labor is also considered as a resource capable of generating results For the output variable credit operations represent the most appropriate alternative considering the role of the bank as a financial intermediary In this work the matter of the best classification among retail banks and banks specialized in credit has little relevance The low relevance in this type of comparison is a result of analysis by segments segments were analyzed separately The results presented here point to an average level of efficiency for the large Brazilian banks in the period This scenario requires efforts to reduce expenses but also to increase revenues Keywords Bank efficiency Data Envelopment Analysis DEA Financial intermediation Bootstrap Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 552 Périco A E et al 2 Efficiency studies of the banking systems The literature offers two main approaches to evaluate the efficiency of banks financial intermediation and results The intermediation approach Sealey Lindley 1977 originates directly from the traditional role of financial institutions as financial intermediaries in which the banks main activity is related to raising funds from savings and redirecting it to borrowers deficit agents Thus in that model the resources entering a bank in the form of deposits and capital are classified as inputs and besides these manual labor is also considered as a resource capable of generating results For the output variable credit operations represent the most appropriate alternative considering the role of the bank as a financial intermediary For the resultoriented approach a set of expenditures is considered on the input side and a set of revenues on the output side Giokas 2008 This work regards the recent literature that investigates the characteristics of banks influences on credit as well as factors related to the performance of these institutions On the subject of banking efficiency Halkos Tzeremes 2013 analyzed the efficiency of fortyfive Greek banks that participated in merger or acquisition processes The results showed that in periods of crisis most of these banks were unable to generate operational efficiency gains yet in less turbulent periods the efficiency gains were observed Halkos Salamouris 2004 analyzed the Greek commercial banks with the use of financial ratios They find a wide variation in performance and show that the increase in efficiency was accompanied of a reduction in the number of small banks due to mergers and acquisitions Several authors have investigated banking efficiency taking into consideration different aspects Paradi et al 2011 for example argue that the management related aspects are crucial in determining the efficiency of a bank However by contrast much of the work conducted on bank efficiency Gaganis et al 2009 Giokas 2008 Portela Thanassoulis 2007 Pastor et al 2006 Stavarek 2005 among others does not even consider using management variables to measure efficiency Gaganis et al 2009 considering the variables used to measure efficiency and the results found point to risk nonpayment as the main element to determine the effectiveness of the Greek banks analyzed The results of Pastor et al 2006 for European banks are in line with the results of Gaganis et al 2009 Determining efficiency will depend greatly on the variables used and the efficiency approach chosen Some authors claim that the size of the bank is a determining factor for its efficiency as for instance in Macedo Barbosa 2009 which clearly defends the middlemarket segment These authors observed a relationship between bank size and performance since the results show that in this segment it is not possible for a small institution to have high performance Périco et al 2008 claim that the size of a bank was not a decisive factor to attribute efficiency given that in Brazil many medium and small banks had greater efficiency than the bigger banks The literature contains a wide range of models within different efficiency approaches always striving for greater performance by the decision makers Ultimately the driving force of this work is to contribute to this discussion 3 Data and methodology This section is divided into three parts The first part is a summary of main information concerning the data used in this research the composition of the investigated sample specification of variables used temporal delimitation as well as the data source The second and third parts represent brief descriptions and characterization of the techniques used in the manipulation of data Data Envelopment Analysis and the technique of Bootstrap resampling 31 Data In this study we used a sample of the 30 largest banks in accordance with the classification criteria used by the Central Bank for all periods It should be noted that only Commercial Bank institutions or Multiple Banks with commercial portfolios were selected We sought to work with institutions of all sizes large medium and small and service sectors retail and loan specialized specific niches to enable the analysis to also consider these aspects In our analysis we use the intermediation approach considering that financial institutions act as intermediates between depositors and borrowers Therefore we define our DEA formulation based on the adaptation of several other studies Sealey Lindley 1977 Berger Humphrey 1992 1997 Drake Hall 2003 Fukuyama Matousek 2011 by using total deposits labor and total assets as inputs and credit operations as output It is worth noting that the selection of these variables followed mainly two parameters The first one is related to the variables used in similar studies developed in other countries Sealey Lindley 1977 as Berger Humphrey 1992 1997 are important references of international banking literature the Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 553 Périco A E et al first studies on bank efficiency were carried out by them The other authors followed a similar pattern The second parameter which justifies the use of these variables is related to the recognition of the fundamental role of a bank financial intermediation The bank raises funds deposits to generate credit operations To capture the deposits are required employees and fixed investments total assets Table 1 presents details about the variables The data used were the banking institutions annual financial records from January 2010 to December 2013 obtained from the website of the Central Bank of Brazil in the report Top 50 Banks see Table 2 for the descriptive statistics of the variables used 32 Data Envelopment Analysis DEA is an operational research technique which is based on linear programming with the objective to comparatively analyze independent units in terms of their relative performance As it does not use a predefined production function identical to all organizations in the inputoutput relationship analysis it is classified as nonparametric Therefore the data envelopment analysis not require preparing a fixed weighted formula to measure the efficiency of the units under analysis because the weights of each variable are determined by the technique itself DEA can be regarded as a body of concepts and methodologies incorporated into a collection of models with different possible interpretations Charnes et al 1994 The envelopment surface will differ depending on the scale assumptions that underpin the model Two scale assumptions are generally employed constant returns to scale CRS and variable returns to scale VRS CRS reflects the fact that output will change in the same proportion as inputs are changed VRS reflects the fact that production technology may exhibit increasing constant and decreasing returns to scale The outputoriented model with variable returns to scale Banker et al 1984 is given below Equation 1 Table 1 The Description of variables Variable Description Type of variable Deposits Total funds raised by the bank R Input Total Assets The sum of current and longterm investment owned by a bank R Input Employees Number of bank employees Input Credit Operations Credit operations granted by the bank R Output Table 2 Descriptive statistics of the inputs and output Year 2010 2011 2012 2013 Deposits Mean 456628150333 521115219000 542254312000 558905980333 Median 61957005000 77399675000 80379365000 82346000000 Std 894408037244 1044387421259 1121331790300 1159606894802 Min 4328340000 8959830000 13446400000 5566110000 Max 3774464830000 4427709130000 4728728180000 4712436530000 Total Assets Mean 101324005667 114807047333 131259972000 131071639000 Median 20668655000 23710745000 27442270000 27501885000 Std 189417267802 211361958139 238948290176 237526293726 Min 2026390000 4021390000 4636810000 5739680000 Max 653224550000 725284140000 828252210000 864668980000 Employees Mean 195841000 204111667 204484667 202552667 Median 13935000 14485000 14385000 14475000 Std 382685442 397364027 402363999 398818453 Min 420000 390000 350000 150000 Max 1264260000 1312990000 1306380000 1253190000 Credit Operations Mean 444317015000 544711561333 639246160667 715029469333 Median 65432925000 72121090000 84388715000 97058130000 Std 837404182047 1032839825529 1247355762180 1427959213199 Min 1453180000 1871370000 4672510000 6393090000 Max 3341930460000 3975211610000 4905323020000 5479481140000 Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 554 Périco A E et al n DEA i ki k i 1 1 1 1 Min v x v 1 0 0 ˆ m r rk r n m i jr r jr k i r r i Subjetctto u y v x u y v u v θ 1 where y output x inputs u v weights r 1 m i 1 n j1n The variables uk and vk are introduced representing the variable returns to scale These variables should not meet the constraint of positivity and may be negative values Among the DEA models it is possible and recommended to choose the most suitable one for the sample through the testing hypotheses of returns to scale presented in Banker 1996 which verifies that scale return hypothesis constant or variable is most plausible for the data set used Banker 1996 suggests applying the nonparametric test of two KolmogorovSmirnov samples based on the maximum distance of the cumulative distributions of the efficiency scores of the DEACRS and DEAVRS models The test evaluates the null hypothesis of constant returns to scale against the alternative hypothesis of variable returns to scale This test is based on the maximum vertical distance between ˆ c ln ˆ c j F θ and ˆ v ln ˆ v j F θ the empirical distributions of ln ˆc θ j and ln ˆv θ j are used The statistic takes values between 0 and 1 Values near 1 tend to reject the null hypothesis and accept the alternative hypothesis Banker Natarajan 2004 33 Statistical Inference by bootstrap Considering that DEA is a deterministic approach a different result from the full efficiency can be interpreted as inefficiency Among other factors this inefficiency or pseudo efficiency may be due to data collection errors or factors attributed to chance compromising estimates made about the scores Dong Featherstone 2004 Aiming to correct this weakness several studies Efron 1987 Xue Harker 1999 Löthgren Tambour 1999 have suggested the use of the bootstrap for more consistent results Bootstrap procedures produce confidence limits on the efficiencies of units to capture the true efficient frontier within the specified interval Dyson Shale 2010 To correct the efficiency values in view of the inherent random data error the approach proposed by Simar Wilson 1998 2000 was used Through this proposal the Bootstrap technique is applied to the DEA methodology to proceed with the statistical inference of the efficiency results achieved by the DEA model Appendix A Thus for each DMU the confidence interval of efficiency the bias and the corrected efficiency can be estimated which will be considered for the performance evaluation of the banks The bootstrap bias estimate for the original DEA estimator 0 ˆDEA x 0 y θ can be calculated as Equation 2 B 1 B DEA 0 0 DEAb 0 0 DEA 0 0 b 1 BIAS x y B x y ˆ ˆ y ˆ x θ θ θ 2 Furthermore 0 0 ˆ DEA b x y θ are the bootstrap value and B is the number of bootstrap replications In this way a biased corrected estimator of 0 x 0 y θ can be calculated as Equation 3 0 0 0 0 0 0 1 0 0 0 0 1 ˆ ˆ ˆ ˆ ˆ 2 DEA B DEA DEA B DEA DEA b b x y x y BIAS x y x y B x y θ θ θ θ θ 3 According to Simar Wilson 2000 this bias correction may create additional noise the sample variance of bootstrap values 0 ˆDEA b x 0 y θ need to be calculated The calculation of the variance of the bootstrap values is illustrated below Equation 4 2 2 1 1 0 0 0 0 1 1 ˆ ˆ ˆ B B DEA b DEA b b b B x y B x y σ θ θ 4 The bias correction illustrated in Equation 3 needs to be avoided unless Equation 5 0 0 ˆ ˆ 1 3 B DEA BIAS x y θ σ 5 4 Financial performance of banks 41 Data envelopment analysis Table 3 summarizes the distribution of efficiency scores of Brazilian banks considering the combined period 20102013 for the two returns to scale models constant returns to scale and variable returns to scale With regards to the definition of the DEA model constant or variable scale literature suggests performing the KolmogorovSmirnov test KS since the choice of technology is a key issue and if decided arbitrarily it can produce biased results In the test procedure the value of this statistic 07197 was obtained α1 allowed accepting the assumption of variable returns to scale Accordingly Table 4 shows Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 555 Périco A E et al the efficiencies of the banks investigated considering the variable returns to scale for each year as well as for the combined period In order to fix efficiency values considering the random error inherent in the data we used the classification proposed by Simar Wilson 1998 2000 Table 5 shows the original efficiency indicators set on the bias the bias and indicators corrected by the bootstrap technique 2000 pseudosamples generated the bootstrap results From the results shown in Table 5 the significant influence of the efficiency rates on the changes in the samples was found reducing the average efficiency by 1443 Corrected average efficiency values in most cases are considerably different in magnitude when compared with the original values Note for example the case of one of the largest Brazilian banks Banco do Brasil which after the bootstrap technique was applied had its index corrected from 998 to 5731 Table 3 Distribution of banks per class Performance Class Banks Banks 0020 3 10 2 667 2040 4 1333 2 667 4060 7 2333 4 1333 6080 9 30 5 1667 8099 2 667 4 1333 100 5 1667 13 4333 Total 30 100 30 100 DEA CRS DEA VRS Minimum 668 729 Average 6197 7572 Maximum 100 100 Table 4 Efficiency indicators for banks Bank Total 2010 2011 2012 2013 Average ABCBrasil 9830 100 9860 7987 7117 8741 Alfa 100 100 9365 9047 100 9603 Bancoob 100 4776 100 100 100 8694 Banestes 4250 4147 42 3730 4108 4046 Banrisul 5249 5624 5493 5208 4510 5208 Bansicredi 100 7862 100 100 100 9465 Basa 2740 3268 3398 3236 2305 3051 Banco do Brasil 100 992 100 100 100 998 BIC 6967 7436 6566 7012 6069 677 BMG 8718 7762 6314 9556 7942 7893 BNB 6016 6784 6580 5502 4845 5927 BNP Paribas 4759 6165 5754 3819 2636 4593 Bradesco 9607 9742 9627 9542 9421 9583 BRB 6894 4675 6192 7107 7567 6385 BTG Pactual 1777 1696 1403 1572 2246 1729 CEF 100 100 994 100 100 9985 Citibank 2799 3226 3169 2689 2297 2845 Credit Suisse 100 100 100 100 5446 8861 Daycoval 6228 6583 6642 5452 5094 5942 Deutsche 100 1230 100 4017 1867 4278 Fibra 100 8703 9308 9077 100 9272 HSBC 5148 5625 5451 5246 4969 5322 Itaú 100 100 100 100 9629 9907 JP Morgan Chase 729 100 567 662 313 2885 Mercantil do Brasil 8521 6541 7238 8444 9785 8002 Panamericano 6929 100 6157 5347 5541 6761 Safra 100 9507 100 100 7977 9371 Santander 100 100 100 100 9849 9962 Societe Generale 100 100 100 993 100 9982 Votorantim 100 100 991 100 100 9977 Average 7572 7375 7438 7139 6718 7168 Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 556 Périco A E et al This enables verify that by conducting a purely deterministic analysis disregarding the statistical bias influence the results found may be overestimated 42 Analysis of results Table 6 summarizes the descriptive efficiency statistics of 30 banks In the analysis of the disaggregated periods a decrease in average efficiency was observed of approximately 55 per cent 20102013 However it was concluded that analyzing the efficiency of all banks together could be a mistake because of the different types of operations performed by them Accordingly the banks were grouped into two segments retail banking and banks specialized in credit This grouping followed the criteria of the Central Bank To observe the efficiency differences by service segment the banks average efficiency indicators were calculated for two service subdivisions in 2010 2011 2012 and 2013 It should be noted that of the 30 banks analyzed 14 are retail banks large public retail banks large foreign retail banks large private domestic retail banks and regional public retail banks and the rest are banks specialized in credit focused on providing few credit modalities Table 7 shows the results Table 7 shows that the average efficiency of the sample decreased by 55 Among the banks classified as retailers an average decrease of approximately 33 was observed between 2010 and 2013 As for banks specialized in loans there was a decrease in average efficiency From 2010 to 2013 the decrease was of 7 Tables 8 and 9 show performance indicators for two bank segments It can be observed that almost all banks specialized in loans between 2011 and 2013 had their average efficiency indicators reduced In early 2012 the Federal Government concluded that Brazilian bank spreads were at much higher levels than those practiced internationally After this in April 2012 two major Brazilian public banks Banco do Brasil and Caixa Econômica Federal initiated a retraction movement in interest rates charged to loan operations In this sense although at much lower Table 5 Efficiency indicators for banks after applying Bootstrap Bank Average efficiency Bias Corrected average efficiency ABCBrasil 8741 219 8522 Alfa 9603 1360 8287 Bancoob 8694 2472 6222 Banestes 4046 240 3806 Banrisul 5208 287 4921 Bansicredi 9465 1950 7515 Basa 3051 325 2726 Banco do Brasil 998 4249 5731 BIC 677 234 6536 BMG 7893 358 7535 BNB 5927 1521 4406 BNP Paribas 4593 172 4421 Bradesco 9583 4043 5540 BRB 6385 937 5448 BTG Pactual 1729 123 1606 CEF 9985 4322 5663 Citibank 2845 295 2550 Credit Suisse 8861 2082 6779 Daycoval 5942 084 5858 Deutsche 4278 265 4013 Fibra 9272 1041 8231 HSBC 5322 908 4414 Itaú 9907 4283 5624 JP Morgan Chase 2885 873 2012 Mercantil do Brasil 8002 1541 6461 Panamericano 6761 075 6686 Safra 9371 1803 7568 Santander 9962 4364 5598 Societe Generale 9982 1102 8880 Votorantim 9977 1798 8179 Average 7167 1443 5725 Table 6 Descriptive statistics of efficiency indicators of 30 banks Descriptive Statistics 2010 2011 2012 2013 Average 5770 5988 5688 5453 Mean 5669 5833 5671 5677 StandardDeviation 2039 2172 2121 2272 Maximum 9287 9221 9003 8973 Minimum 619 723 774 583 Table 7 Efficiency average by segment 2010 2011 2012 2013 Total sample 5770 5988 5688 5453 Retail banks 5070 5113 5044 4902 Banks specialized in credit 6381 6753 6250 5935 Table 8 Efficiency indicators of retail banks Banks 2010 2011 2012 2013 Banco do Brasil 5687 5730 5698 5809 Bradesco 5689 5535 5395 5540 CEF 5619 5662 5630 5741 Itaú 5633 5676 5644 5544 Santander 5576 5619 5587 5612 Banrisul 5602 5495 4788 3798 BNB 4859 4774 4152 3839 Citibank 3007 2575 2603 2016 HSBC 4559 4486 4354 4255 Safra 7592 8022 7846 6812 Banestes 4313 3753 3450 3706 Basa 2835 2921 2820 2327 Mercantil do Brasil 5651 5936 6739 7517 BRB 4367 5397 5911 6117 Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 557 Périco A E et al intensity private banks retail and specialized in credit followed suit in this retraction movement of rates However it should be noted that Brazil has a large bank concentration in which six banks hold over 80 of the market share Departamento Intersindical de Estatística e Estudos Socioeconômicos 2013 of which two are the major Brazilian public banks Between April 2012 and January 2013 the operations for natural persons underwent significant changes in their rates Two major Brazilian public banks reduced the overdraft rate by an average of 43 while for the personal credit line the average reduction by private banks was of 375 the mean reduction in two public banks was of 26 and in the private banks the average reduction was of 145 and for the auto loan financing the average reduction was of 315 for public banks and 22 for the banks investigated Departamento Intersindical de Estatística e Estudos Socioeconômicos 2013 With regards to the credit lines for companies the mean rate reductions also differed between public and private banks For promissory note discounts the public banks offered average reductions of 345 while the average reduction by private banks was of 135 For working capital finance the average reduction of public banks was of 40 and in private banks it was of 195 Departamento Intersindical de Estatística e Estudos Socioeconômicos 2013 By monitoring the retractions in interest rates there was as expected an increase in credit operations throughout the Brazilian banking system According to the data from the Departamento Intersindical de Estatística e Estudos Socioeconômicos 2013 it can be observed that the operations of the public financial system grew by 282 in 12 months Dec2011 to Dec2012 In the same period the operations of the domestic private financial system and foreign institutions grew only 67 and 85 respectively It can then be understood why so many of the loan specialized banks underwent the greatest reduction in average efficiency during the period investigated They increased their resources the number of employees deposits and capital and the credit operation growth fell short of what was feasible This result is due to the decrease in interest rates initiated by two major Brazilian public banks which virtually dominated the expansion of loan operations within this period The loan specialized banks had higher average efficiency indicators than that of the retail banks and this is the result of their higher input productivity in relation to the loan operations generated by these institutions In other words public banks and retail banks may generate greater loan operation activities both in the number of operations as well as financial resources but their resources employees deposits and capital are less productive than the resources of banks specialized in credit The DEA analysis does not consider the size of a unit to classify it as efficient what is considered in this type of analysis is the use of resources inputs to achieve the product outputs Efficiency is the relationship between the results obtained and the resources used Accordingly managerial aspects resource allocation decisions are more relevant in the DEA analysis than the size of the decision making unit Tables 10 and 11 identify the less efficient banks considering the segment in which they operate For the Retail Bank category the least efficient bank was Citibank and for the Banks Specialized in Credit category JP Morgan was the most inefficient For Citibank the DEA suggests another bank as a benchmark the Safra bank both midsized Comparing Citibank to Safra enables identify the Table 9 Efficiency indicators for banks specialized in credit Banks 2010 2011 2012 2013 ABCBrasil 9287 9221 7974 7608 Alfa 8224 8000 7976 8949 Bancoob 4419 8366 6178 5925 BIC 7100 6484 6485 6075 BMG 7388 6109 8559 8082 BNP Paribas 5436 5312 3962 2972 BTG Pactual 1579 1422 1512 1912 Credit Suisse 6471 7418 8023 5203 Deutsche 619 8183 4514 2738 Fibra 7907 8461 7892 8665 JP Morgan Chase 5970 723 774 583 Panamericano 8457 6666 5806 5815 Votorantim 7975 7815 8275 8650 Bansicredi 6572 8485 7504 7500 Daycoval 6152 6392 5574 5313 Societe Generale 8542 9003 9003 8973 Table 10 Citibank Safra Citibank Safra Average efficiency 2550 7568 Deposits R 62579957 52290920 Net equity R 27623372 26014704 Number of employees 6190 5753 Credit operations R 51511037 156610079 Table 11 JP Morgan Alfa JP Morgan Alfa Average efficiency 2012 8287 Deposits R 6504849 8141772 Net equity R 11353675 7619346 Number of employees 737 888 Credit operations R 1439015 26979487 Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 558 Périco A E et al underutilization of some resources by the first bank Table 10 describes some of the data Using the DEA technique enabled estimate the growth potential of Loan Operations from the observed input levels of Citibank This bank could increase its output by 215 considering its resources available each year Table 11 displays the comparison between JP Morgan and its benchmark the Alpha Bank both midsized The JP Morgan bank given the resources employed had the potential to expand its lending operations by more than 1000 in the overall period Table 12 was prepared from these results in order to establish the following relationships credit operations deposits credit operationstotal assets and credit operationsemployees Table 12 enables identify the productivity of the variables of the model and verify the reason why some banks are considered more efficient than others For example if the efficiency of a bank were exclusively dependent on the relationship between deposits and operations the Alpha Bank would be more efficient since each deposit unit of this bank is converted into 331 units of loan operations Accordingly the least efficient would be JP Morgan 022 In the Credit OperationTotal Assets relation the most efficient bank was Sicredi since each total assets unit produced 1583 units of loan operations and again the least efficient was JP Morgan 013 And finally if bank efficiency were to be defined only by the productivity of employees the Credit Suisse bank would have been the most efficient with each employee having produced 390806 units of credit operations JP Morgan would again be the least efficient 1954 Of the top five retail banks operating in the country Banco do Brasil Caixa Econômica Federal Bradesco Itaú and Santander the most productive bank considering only the deposits feature was Santander considering only the total assets feature it was Caixa Econômica Federal and considering only the employees feature it was Banco do Brasil By demonstrating that a bank is more efficient in some variables Table 11 than others we do not state that the situation of the bank model is ideal since there is no consensus in the literature about Table 12 Data banks Banks Credit OperationsDeposits Credit Operations Total Assets Credit OperationsEmployees ABCBrasil 224 471 5581 Alfa 331 354 3039 Bancoob 048 1116 4515 Banestes 051 395 375 Banrisul 085 463 713 Bansicredi 093 1583 9545 Basa 083 115 220 Banco do Brasil 100 738 1378 BIC 127 526 4579 BMG 196 454 9406 BNB 116 456 346 BNP Paribas 117 224 3857 Bradesco 116 404 1022 BRB 083 651 524 BTG Pactual 044 081 2393 CEF 108 1424 1075 Citibank 082 186 832 Credit Suisse 116 118 39080 Daycoval 177 327 2675 Deutsche 043 103 1749 Fibra 137 739 4873 HSBC 073 516 631 Itaú 120 385 969 JP Morgan Chase 022 013 195 Mercantil do Brasil 097 875 726 Panamericano 131 479 5307 Safra 299 602 2722 Santander 139 264 1310 Societe Generale 242 295 2303 Votorantim 295 673 13193 Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 559 Périco A E et al the optimal capacity of such specific production factors Therefore when we declare for example that the Credit Suisse bank is the most efficient considering only the credit operations by employees we are only saying that the employees of this bank are more productive in comparison to the employees of the other 29 banks Bank efficiency is related to the balance between the input resources employees financial resources structure and equipment necessary for a banks operation and etc and output resources loan operations income from financial intermediation and etc for the quality of such services among other factors The analysis proposed in this work was purely about financial intermediation which showed that all the banks investigated can and should improve the performance of their production resources However we delimit the conclusions to the sample investigated quite aware that if the comparison undertaken encompassed large banks from different countries Brazilian banks would admittedly be more distant from the efficient frontier 5 Final Remarks This paper proposes measuring the efficiency of 30 Brazilian banks using the data envelopment analysis methodology This technique was applied to a set of banks in 2010 2011 2012 and 2013 The concept of efficiency has traditionally been related to the reason output input of investigated units The unit that generates more output with fewer resources is considered the most efficient In configurations of multiple inputs and output the allocation of weights to the inputs and output is required in order to calculate the efficiency of production units The DEA is a nonparametric technique that uses linear programming to determine optimal weights that minimize the distance between the efficient frontier and the unit investigated The main advantage of DEA is that it does not require the specification of a production function The DEA uses a set of inputs that the unit wants to minimize and a set of output that the unit wants to maximize The disadvantage of the technique is that the statistical inference is very difficult to apply in efficiency scores Therefore the DEA bootstrap procedure Simar Wilson 1998 allows extraction of the sensitivity of scores stemming from sample distribution inefficiency In this direction we seek to benefit from the advantages of using DEA and take care of what we consider the fragility of it through the use of DEA bootstrap It should be emphasized that the results obtained do not refer to absolute efficiency The banks considered efficient are only classified this way for the group analyzed In this work the matter of the best classification among retail banks and banks specialized in credit has little relevance The low relevance in this type of comparison is a result of analysis by segments segments were analyzed separately The data envelopment analysis allows to calculate the efficiency indicators of homogeneous units and of different sizes However even for banks of different sizes it was understood that they were not effectively homogeneous since they operate in different niches and therefore perform different operations Accordingly the classification of efficient retail banks is completely set apart from the classification of banks specialized in credit Therefore in this analysis the efficiency increase gains more emphasis to each segment investigated In September 2012 the Valor Econômico newspaper published an article entitled With falling interest rates banks pursue efficiency gains Mandl 2012 and reported some efficiency results of large Brazilian banks The paper discusses the results reported by Goldman Sachs declaring that compared to their peers around the world Brazilian banks were the least efficient The results presented here although only addressing Brazilian reality point to an average level of efficiency for the large Brazilian banks in the period analyzed which somehow corroborates the results reported by Goldman Sachs when referring to the efficiency of large Brazilian banks This is a new scenario which not only requires efforts to reduce expenses but also to increase revenues even taking into consideration reduced interest rates which result in lower earnings for banks References Banker R D 1996 Hypothesis tests using data envelopment analysis Journal of Productivity Analysis 723 139159 httpdxdoiorg101007BF00157038 Banker R D Natarajan R 2004 Statistical tests based on DEA efficiency scores In W Cooper L Seiford J Zhu Eds Handbook on data envelopment analysis pp 265298 New York Kluwer Academic Publishers Banker R D Charnes A Cooper W W 1984 Some models for estimating technical and scale inefficiencies in data envelopment analysis Management Science 309 10781092 httpdxdoiorg101287mnsc3091078 Berger A N Humphrey D B 1992 Measurement and efficiency issues in commercial banking In Z Griliches Output measurement in the service sectors pp 175212 Chicago University of Chicago Press Berger A N Humphrey D B 1997 Efficiency of financial institutions international survey and directions for future Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 560 Périco A E et al research European Journal of Operational Research 982 175212 httpdxdoiorg101016S0377221796003426 Charnes A Cooper W Lewin A Y Seiford L M 1994 Data envelopment analysis theory methodology and application Boston Kluwer Academic Publishers Departamento Intersindical de Estatística e Estudos Socioeconômicos DIEESE 2013 Um novo cenário para o setor financeiro no Brasil Nota Técnica 123 pp 115 São Paulo DIEESE Retrieved in 23 October 2013 from httpwwwdieeseorg brnotatecnica2013notaTec123CenarioSetorFinanceiropdf Dong F Featherstone A 2004 Technical and scale efficiencies for Chinese rural credit cooperatives a bootstrapping approach in data envelopment analysis Working Paper 04 WP366 Iowa Center for Agricultural and Rural Development 26 p Retrieved in 23 October 2013 from httpwww researchgatenetpublication24085100Technicaland ScaleEfficienciesforChineseRuralCreditCooperativesA BootstrappingApproachinDataEnvelopmentAnalysis Drake L Hall M J B 2003 Efficiency in Japanese banking an empirical analysis Journal of Banking Finance 275 891917 httpdxdoiorg101016S0378426602002406 Dyson R G Shale E A 2010 Data envelopment analysis operational research and uncertainty The Journal of the Operational Research Society 611 2534 httpdxdoi org101057jors2009145 Efron B 1987 Better bootstrap confidence intervals Journal of the American Statistical Association 82397 171185 httpdxdoiorg10108001621459198710478410 Fukuyama H Matousek R 2011 Efficiency of Turkish banking twostage network system Variable returns to scale model Journal of International Financial Markets Institutions and Money 211 7591 httpdxdoi org101016jintfin201008004 Gaganis C Liadaki A Doumpos M Zopounidis C 2009 Estimating and analyzing the efficiency and productivity of bank branches evidence from Greece Managerial Finance 352 202218 httpdxdoiorg10110803074350910923518 Giokas D 2008 Assessing the efficiency in operations of a large Greek bank branch network adopting different economic behaviors Economic Modelling 253 559 574 httpdxdoiorg101016jeconmod200710007 Halkos G E Salamouris D S 2004 Efficiency measurement of the Greek commercial banks with the use of financial ratios a data envelopment analysis approach Management Accounting Research 152 201224 httpdxdoi org101016jmar200402001 Halkos G E Tzeremes N G 2013 Estimating the degree of operating efficiency gains from a potential bank merger and acquisition A DEA bootstrapped approach Journal of Banking Finance 375 16581668 httpdxdoi org101016jjbankfin201212009 Löthgren M Tambour M 1999 Testing scale efficiency in DEA models a bootstrapping approach Applied Economics 3110 12311237 httpdxdoiorg101080000368499323445 Macedo M A S Barbosa A C M 2009 Eficiência no sistema bancário Brasileiro uma análise do desempenho de bancos de varejo atacado middlemarket e financiamento utilizando DEA Revista de Informação Contábil 3 124 Mandl C 2012 September 3 Com juros em queda bancos perseguem ganhos de eficiência Valor Econômico Caderno de Finanças Retrieved in 23 October 2013 from http wwwvalorcombrfinancas2814064comjurosemqueda bancosperseguemganhosdeeficienciacomments Paradi C P Rouatt S Zhu H 2011 Twostage evaluation of bank branch efficiency using Data Envelopment Analysis Omega 391 99109 httpdxdoiorg101016j omega201004002 Pastor J T Lovell C A K Tulkens H 2006 Evaluating the financial performance of bank branches Annals of Operations Research LouvainlaNeuve 1451 321337 httpdxdoiorg101007s1047900600383 Périco A E Rebelatto D A N Santana N B 2008 Eficiência bancária os maiores bancos são os mais eficientes Uma análise por envoltória de dados Gestão Produção 152 421431 httpdxdoiorg101590 S0104530X2008000200016 Portela M Thanassoulis E 2007 Comparative efficiency analysis of Portuguese bank branches European Journal of Operational Research 1772 12751288 httpdxdoi org101016jejor200601007 Sealey C Jr Lindley J T 1977 Inputs outputs and a theory of production and cost at depository financial Institutions The Journal of Finance 324 12511266 httpdxdoiorg101111j154062611977tb03324x Simar L Wilson P 2000 A general methodology for bootstrapping in nonparametric frontier models Journal of Applied Statistics 276 779802 httpdxdoi org10108002664760050081951 Simar L Wilson P W 1998 Sensitivity analysis of efficiency scores How to bootstrap in nonparametric frontier models Management Science 441 4961 http dxdoiorg101287mnsc44149 Stavarek D 2005 Estimation of Factors Influencing Efficiency of Banks in New Member States before Joining the European Union Ekonomicky Casopis 53 593610 Xue M Harker P T 1999 Overcoming the inherent dependency of DEA efficiency scores a bootstrap approach Working Paper 9917 The Wharton Financial Institutions Center Estimating the efficiency from Brazilian Data Envelopment Analysis DEA Production 263 551561 julset 2016 561 Périco A E et al Appendix A Bootstrap procedures This appendix illustrates the bootstrapbased algorithm introduced by Simar Wilson 1998 2000 The procedure is as follows Step 1 Transform the inputoutput vectors using the original efficiency estimates 1 ˆ ˆ ˆ l in i i i in i i n as x y x y θ θ Step 2 Generate smoothed resampled pseudoefficiencies iy as follow Step 21 Given a set of estimated efficiencies ˆin θ use the rule of thumb Silverman 1986 p 4748 to obtain the bandwidth parameter h as h 09 15 13 ˆ min 4 ˆ 1 3 σθ R where ˆˆθ σ is the standard deviation of ˆin θ and 13 R is the interquartile range of the empirical distribution of ˆin θ Step 22 Generate iδ by replacing with replacement from the empirical distribution of ˆin θ estimated efficiencies Step 23 Generate the sequence iδ using 1 2 i i i i i i i h if h h otherwise δ ε δ ε δ δ ε where iε is drawn iid from a standard normal distribution Step 24 Generate smoothed pseudoefficiencies iγ using the following formula ˆ 2 2 1 ˆ i i i i h θ δ δ δ γ σ where 1 n i i i n δ δ which is the average of the resampled original efficiencies Step 3 Let the pseudodata be given by ˆ l i i i i i x y x y γ Step 4 Estimate the bootstrap efficiencies using the pseudodata as 1 1 ˆ n SW n in z i i i i min y Yz x X z z z R θ θ θ θ Step 5 Repeat steps 24 2000 times to create a set of 2000 bankspecific bootstrapped efficiency estimates 1 1 2000 ˆSW b in i n b θ References Silverman B W 1986 Density estimation for statistics and data analysis monographs on statistics and applied probability London Chapman Hall Simar L Wilson P W 1998 Sensitivity analysis of efficiency scores How to bootstrap in nonparametric frontier models Management Science 441 4961 httpdxdoiorg101287mnsc44149 Simar L Wilson P 2000 A general methodology for bootstrapping in nonparametric frontier models Journal of Applied Statistics 276 779802 httpdxdoiorg10108002664760050081951