Convergence in Brazil: New Evidence Using a Multilevel Approach

Full Terms Conditions of access and use can be found at httpwwwtandfonlinecomactionjournalInformationjournalCoderaec20 Download by University of Newcastle Australia Date 09 March 2017 At 1433 Applied Economics ISSN 00036846 Print 14664283 Online Journal homepage httpwwwtandfonlinecomloiraec20 Convergence in Brazil new evidence using a multilevel approach Alberto Díaz Dapena Fernando Rubiera Morollón Mônica de Moura Pires Andréa da Silva Gomes To cite this article Alberto Díaz Dapena Fernando Rubiera Morollón Mônica de Moura Pires Andréa da Silva Gomes 2017 Convergence in Brazil new evidence using a multilevel approach Applied Economics DOI 1010800003684620171299101 To link to this article httpdxdoiorg1010800003684620171299101 Published online 09 Mar 2017 Submit your article to this journal View related articles View Crossmark data Convergence in Brazil new evidence using a multilevel approach Alberto Díaz Dapenaa Fernando Rubiera Morollón a Mônica de Moura Piresb and Andréa da Silva Gomesb aREGIOlab Regional Economics Laboratory University of Oviedo Oviedo Spain bUESC Universidade Estadual de Santa Cruz Ilheus Brazil ABSTRACT Empirical analysis of regional convergence does not focus its attention on the spatial level of the data Most of the time the analysis is made at the aggregated level by large regions or states where there are more data available However when there is a wide intraregional heterogeneity it is possible to have regional convergence coexisting with local processes of divergence This is our hypothesis for the case of Brazil where there are relevant different intrastate behaviours To capture these behaviours in this article an adaptation of multilevel techniques to the βConvergence analysis is proposed using data of Brazilian states and municipalities taken from the Economic Census 1991 to 2010 With this data we confirm the existence of convergence at the national level but we can observe that this general trend coexists with different intrastate behaviours across Brazilian geography The most industrialized or urbanized states of the Southeast usually present internal divergence while the less developed inland states normally show convergence which agrees with what we expect from the centralperiphery models KEYWORDS βConvergence multilevel econometrics economic growth Brazil JEL CLASSIFICATION R11 R15 O47 I Introduction Brazil is the largest South American economy and one of the most important emerging countries in the world The Gross Domestic Product hereafter GDP of São Paulo alone exceeds the entire production of Argentina Chile Colombia or Venezuela and is 139 of the total GDP in Latin America and the Caribbean in 2010 Comisión Económica para América Latina y el Caribe 2015 However Brazil is also one of the coun tries with larger intra and interregional disparities According to Comisión Económica para América Latina y el Caribe 2013 Brazil and Mexico the two main economies of Latin America present the highest poverty values in the continent Particularly the Northeastern and Southeastern regions of Brazil show the highest incidence of poverty mainly in Bahia where more than 5 million people almost onethird of total population are in poverty In Pará São Paulo Pernambuco and Rio de Janeiro there are more than 3 million inhabitants who are in this situation data from 2011 see Comisión Económica para América Latina y el Caribe 2013 The strong income disparities that are observable among individuals could also be observable among territories at different levels among the States but also within them The relevance of the Brazilian economy and the dimension of its internal disparities have motivated researchers to study the evolution in its territorial economic differences Obviously this implies the exis tence of a large body of literature about the most common way of measuring the evolution of spatial economic differences the βConvergence analysis However most of these works use data at the State level The number of contributions that analyze the βConvergence with more disaggregated information is lower For instance Resende 2011 and Resende Carvalho and De Sakowski 2016 studied the pro cess of convergence using different spatial levels but to our knowledge any previous research has com bined local municipal and regional states levels at the same time for the case of Brazil Our hypothesis is that the definition of the spatial unit in the betaconvergence analysis could signifi cantly affect the convergence conclusions This should be more relevant in cases where the intra regional intrastate disparities are especially large such as in the case of Brazil Consequently in this article a multilevel betaconvergence model is pro posed Multilevel techniques allow the researcher to identify the specific contribution of each level of CONTACT Fernando Rubiera Morollón frubierauniovies APPLIED ECONOMICS 2017 httpdxdoiorg1010800003684620171299101 2017 Informa UK Limited trading as Taylor Francis Group 2 A DIAZ DAPENA ET AL hierarchical information to the total variability oConvergence However the development of eco Chasco and Lopez 2009 use this approach to nomic growth theories to understand crosscountry evaluate the effect of decentralization policies in patterns triggered the attention of the economist In regional disparities using a multilevel convergence this context of neoclassic macroeconomic theories the model Fazio and Piacentino 2011 studied the con interesting concept of BConvergence was born in the vergence at micro firms and macro regions pioneer studies of Abramovitz 1986 and Baumol levels simultaneously by means of a multilevel con 1986 ditional model In this article a similar strategy is The initial approach consisted of estimating a followed We are going to estimate a conditional simple regression of the income per capita growth betaconvergence model for the Brazilian economy rate or similar variable in a spatial unit over its using data at local municipalities and regional initial level Therefore this approach is based in the States levels simultaneously This allows us to iden following equation tify possible convergence or divergence intrastate r behaviours that can coexist with global convergence log i yp Xo Blogy YXi i 1 or divergence behaviours among the states The T results of this new approach could help in the proper where log p s the logarithm of growth of Lo dtc ete income per capita during a period of time T in the analysis of the evolution of the spatial disparities in 0 the Brazilian economy as well as in the correct SP atial unit and y is the mncome Per capita the evaluation of the situation and possible future regio initial situation of the period Through this estima nal policies for this country tion we can see if poorer territories tend to grow With this objective in mind the article is struc faster or slower than the rich as neoclassic tured as follows First Section 2 reviews the previous PP roaches predict If the parameter By is negative literature on convergence in general but is mainly this indicates that lower levels of income per capita focused on applications to the Brazilian case In Produce higher growth rates leading to a process of Section 3 we present the multilevel approach and CORVergence in the long run A positive estimate of the potentialities in BConvergence analysis as well By would reveal a process of divergence x is a vector as a multilevel conditional 8Convergence specifica of control variables tion This model is estimated with the Brazilian data The initial studies see Baumol 1986 SalaI in Section 4 explaining the database and main Martin 1996 Coulombe and Lee 1993 Shioji 1992 results Main conclusions are summarized in De La Fuente 2002 among others were based in Section 5 crosssectional data and mostly with ordinary least square OLS regressions This implies important limitations of the analysis and potentially relevant Il BConvergence analysis for Brazilian estimation problems To structure the recent territories a reappraisal advances in the empirical estimation of convergence models Islam 2003 and Magrini 2004 presented Convergence studies evolution from simple linear regressions to spatial dependence models outstanding surveys of the evolution of this topic during recent years The authors classified three Since the early studies of Kuznets 1955 Easterlin types of approaches i panel data ii time series 1960 or Williamson 1965 among others conver and iii spatial dependence gence among territories has always attracted the atten One of the more relevant contributions was the tion of regional scientists or geographers In these introduction of panel data in the estimation proce initial studies the authors analyzed the evolution of dure The fBConvergence equation regresses the spatial disparities using some key variable such as per income per capita growth rate in a territory on the capita income or production An index similar to the initial level of that variable It is clear that this standard deviation was usually used in these measure regression should be controlled by the differential ments of dispersion commonly known as factors of each region When no other control Parameter in the expresion By 1 e would determine the average regional rate of convergence to the same income per capita in unconditional convergence equations or to steady state in conditional convergence ones variable is considered we talk about an analysis of absolute βConvergence whereas if another explana tory variable is included we conduct a conditional βConvergence analysis Nevertheless having wide databases that inform about the relevant factors of the region is not easy and lack of information usually restricts the possibilities of the researcher Panel data allows the researcher to introduce them as fix effects see HoltzEakin Newey and Rosen 1988 and Arellano and Bond 1991 Thanks to this tech nique it is possible to control the different initial characteristics of each spatial unit see Caselli Esquivel and Lefort 1996 Meanwhile time series analysis by Quah 1993 Bernard and Durlauf 1995 and Carlino and Mills 1996 developed a new concept of convergence They conducted a time series test for unit roots in order to find a stochastic convergence This concept tries to discover persistent differences in the series of income or total production Finally but more importantly the relevance of spill overs in economic growth could be considerable espe cially if local or regional data are used so if the spatial dynamics are ignored it can be an important source of model misspecification Anselin 1988 During the last decades there was a great development of techniques to consider spatial dependence see Anselin and Florax 1995 and Anselin and Rey 1997 and more recently see Lasarge and Fisher 2008 Abreu et al 2005 could also be consulted to review the relevance and integra tion of spatial effects in the empirical models of eco nomic growth More focused in the convergence models Rey and Montouri 1999 were the first in proposing a betaconvergence equation with spatial effects Since then many very relevant extensions were made integrating the advances obtained from the two other previous lines For instance Elhost 2003 proposes a model with spatial effects and panel data which is assumed to be the most complete and standard way of estimating betaconvergence Additionally the previous approach in Elhost et al 2010 was sophisticated incorporating spatial effects and timeseries dynamics All of these contributions specially the models with spatial effects solve most of the estimation problems significantly improving the quality of the conclusions especially when the researcher is work ing with crosscountry data or large regionsstates aggregately However when the studies are referred to regional or local data some problems are still present If we are working in a higher level of aggre gation we are not sure under what degree the spatial scale could affect our conclusions see DíazDapena FernándezVázquez and RubieraMorollón 2016 Finally the βConvergence concept or any one of the other crosssection convergence approaches does not provide evidence about the internal dynamics of regional income distribution The convergence analysis in Brazil literature revision As noted in the introduction Brazil is an economy with important spatial differences in development and growth Additionally the official statistics system of Brazil produces very precise information on GDP at the state and local level which allows for application of the different convergence models and techniques at different regional scales As a result many Brazilian and foreign authors have studied convergence in Brazil2 The general conclusion of all of these studies is that the Brazilian territories present a relatively low convergence rate Some examples are explained below Azzoni Filho and Menezes 2000 found a slow process of absolute income convergence in the per iod 19391996 Azzoni Filho and Menezes 2000 also included geographic characteristics such as cli mate public and private infrastructure as well as other important elements in order to explain the growth rates of states or regions In a second essay they showed that there is no absolute convergence on income though there is conditional convergence with high velocity They explained that Brazilian states approached a similar stable equilibrium in the period of analysis with regional inequalities This was one of the first essays to use the micro database from the Instituto Brasileiro de Geografia e Estatística hereafter IBGE aiming to analyze Brazilian regions pattern of per capita income In general results indicated regional inequality in Brazil might be reduced by investing in public infra structure and education 2See A Ferreira and Diniz 1995 Ferreira and Ellery 1996 A Ferreira 2000 Zini Junior 1998 Azzoni Filho and Menezes 2000 Azzoni Filho and Menezes 2000 APPLIED ECONOMICS 3 In Andrade et al 2004 a hypothesis of conver gence among Brazilian cities was tested during 19701996 and results displayed two forms of conver gence a lowincome convergence in cities from the North and Northeast and a highincome convergence in cities from the MiddleWest Southeast and South Magalhães and Hewings 2005 also estimated a convergence process in the 27 Brazilian states dur ing 197095 with a strong spatial correlation pat tern however such a phenomenon did not happen in a scattered manner for instance São Paulo had great influence on other regions for example the Brazilian Northeast region By analyzing conditional convergence in Brazil during 19852004 and focusing on the role of human capital Cravo and Soukiazis 2011 observed that different levels of human capital had different impacts on per capita income growth for a regions development Low human capital levels explained convergence among less developed states Souza and Osorio 2014 analyzed income con vergence between metropolitan and rural Brazilian regions during 19812009 They verified that household per capita incomes have grown faster in rural Brazilian regions since the beginning of the 1980s On the other hand in metropolitan regions around two thirds of the rising income resulted from Brazilian Social Welfare program expansion During 19812009 the inequality reduction between Brazilian major cities and other Brazilian regions was 51 of full drop 20 during 19952009 Without this convergence the income inequality in 2009 would be approxi mately 10 greater than the observed inequality according to these authors Finally Resende 2011 and Resende Carvalho and De Sakowski 2016 use the most advanced econometric approaches and more recent informa tion to estimate the equations of convergence for the case of Brazil Particularly in Resende Carvalho and De Sakowski 2016 a set of differ ent spatial models spatial error spatial lag and spatial Durbin are estimated with panel data They use different spatial scales until the maxi mum level of microregions which is an intermediate level of desegregation between states and municipalities They remark that the conclu sion depends clearly on the choice of spatial desegregation They also highlight the necessity of a new crosslevel framework in order to under stand why the process differs in the scales For this reason multilevel methodology could be the new step to understanding the process of convergence in Brazil using different spatial scales simultaneously III A new perspective the multilevel approach The multilevel technique has been widely used in different disciplines Goldstein 1986 for more information see Snijders and Bosker 1999 Hox 2010 Goldstein 2011 In Economics most of the applications have been made in labour market stu dies Andersson Hammarstedt and Hussain 2013 Cohen 1998 Now it is being successfully intro duced in other types of economic studies Li and Wei 2010 Srholec 2010 In multilevel methodology the focus is on the importance of the hierarchy in the data This approach splits the variability in the different levels of the analysis The different components of variability can be used to evaluate the impor tance of each level in the process of analysis Using spatial levels as hierarchy this methodology could be a suitable approach for the analysis of convergence This type of analysis is particularly efficient when our main interest is the effect of the hier archy By means of Restricted Maximum Likelihood REML3 this estimation introduces two additional parameters for the variance σ2 u and σ2 v σ2 u takes into account the variability in the intercepts while σ2 v accounts for variability in the slopes These components can be used to eval uate the importance of the groups in the sample Thus compared with an analysis with dummies the multilevel technique only needs two para meters to evaluate the hierarchy One traditional problem is the choice of the level of aggregation However the multilevel 3REML becomes more suitable estimation strategy when the number of groups is less than 30 This type of estimation is chosen because to the case of Brazil we are restricted to 27 States in the first level 4 A DÍAZ DAPENA ET AL APPLIED ECONOMICS 5 technique solves this problem by explicitly intro size As explained in Goldstein 1986 and appendix ducing the hierarchy in the model Despite this 22 of Goldstein 2011 the expected value of the advantage the choice of the geographical unit vector with the concrete countryspecific estimate still can affect the results if the areas are not p is given by Equation 5 appropriate The introduction to the multilevel analysis can be p EplYV RV y XB used to measure the divide in the variability of this type Y Xp 5 of analysis of each geographical level see the applica where V is the covariance matrix of the model and R tions of multilevel to the convergence analysis proposed jg g blockdiagonal matrix Each rth block R is by Chasco and Lopez 2009 or Fazio and Piacentino generated according to the following equation 2011 In our case our propose starts with the condi tional BConvergence Equation 1 of previous section R ZQ 6 but we can distinguish between two spatial levels N where Q is the covariance matrix of the random municipalities grouped into M states In Equation 2 effects at the municipal level and Z are the explana logyt y is the growth of the Income per capita In tory variables with random coefficients the right side of the equation logy represents the As a result the conditional covariance matrix of p initial situation of a territory and x isa group of control can be obtained using the expression Ep p variables with a vector of coefficients y p p as in Equation 7 log yir Yr or Boplogyi yx Ep p6p sRTV éir Vi Tyly7lyT 71 1N Wj 1M 2 LV xQTV xy XT VR In the multilevel model the parameters xo and 7 B can vary along the federal units They are In this equation S is also a block diagonal matrix assumed to vary according to a normal distribu Each rth block is the information of a country in tion with a constant variance This model is Matrix Q known as the Random Slope Model The optimi zation of the REML estimates these two additional parameters IV Application to the Brazilian case Xor Xo UBo By vr Ur N0 02 v Database Brazilian economic census N00 eir N0 02 3 The application of the multilevel approach the The evaluation of the hierarchy is made with the empirical estimation of Equations 1 and 2 Variance Partition Coefficient VPC This estima implies the necessity to use information at a highly tion indicates the percentage of the variation that isaggregate scale as well as an aggregate scale is caused by the group level see Equation 4 simultaneously In the case of Brazil the informa This equation can be easily modified to a general tion on different economic variables at different case with more random effects following Goldstein spatial scales States and municipalities is available 2011 thanks to the Economic Census provided by the Instituto Brasileiro de Geografia e Estatistica Varu logyv 02 2logy ow logy a IBGE 0 2logy ow log 9 o Brazil is administratively divided into 5565 munici we 02 2logyouy logy02 02 palities aggregated into 27 States including the federal mu ues district of Brasilia The average number of municipa 4 lities per State excluding the Federal District is 214 In Concrete intercepts slopes and their conditional the conditional analysis we use 4067 municipalities covariance matrix can be estimated in a second because for some variables we do not have information step by means of these variances and the sample for all of the observations 6 A DIAZ DAPENA ET AL From the time perspective we consider the longest Traditional migratory flows are introduced as the available period with coherent local municipalities share of immigrants in 19911995 over the popula and aggregated states information of the economic tion in 1991 The intraregional differences in the variables considered which is from 1991 to 2010 structure of sectors see Ferreira and Tatiwa 2014 Although the convergence analyses are more interest are provided with specific variables of the shares for ing when a long period of time is available two decades the different sectors in the employment of a munici could be considered as a long enough period of time to pality The agricultural sector is excluded and used reach interesting conclusions about the recent as the reference Table 1 summarizes all the variables dynamics of convergence considered The main variables of the model are built using income per capita in thousands of R at constant oo prices of 1991 The dependent variable logyy Estimation results is the cumulative growth rate from 1991 to 2010 and Table 2 summarizes the results of our estimation for the independent variable is the income per capita the period 19912010 To compare them with pre level at the beginning of the period 1991 As the vious studies a traditional OLS BConvergence IBGE does not provide any desegregated price index equation is estimated in columns one and two it is impossible to deflect at the state or municipal Second in columns three and four the random level Although it is usual in convergence literature slope model proposed in Equation 2 is also esti to work with notlocally deflected data considering mated In this specification there are two levels the size and spatial heterogeneity of the Brazilian municipalities 4067 and states 27 case this is an important data constraint that should In the OLS estimation a significant and negative be considered in the resulting analysis and 8 parameter of 0263 in the absolute model and conclusions 0386 in the conditional model are found These Control variables are introduced following the results indicate a significant process of convergence model of Mankiw Romer and Weil 1992 and the poorer a territory the higher the growth of Resende Carvalho and De Sakowski 2016 The income per capita These coefficients indicate a growth of the labour force is introduced through speed of convergence of 16 in the absolute the growth of the population of each municipality model and 257 in the conditional model Our between 1991 and 2010 The rest of the explanatory conclusions with the OLS approach for the B para variables are given in terms of the initial year of meter as well as for the rest of the control variables analysis and are provided by the IBGE The model are consistent with the conclusions obtained in pre also introduces the population density of each muni vious literature for the case of Brazil Particularly cipality Human capital is introduced through a the results are very similar to those obtained in proxy variable This proxy is the average years of Azzoni 2001 or more recently in Resende 2011 education of the population in each municipality and Resende Carvalho and De Sakowski 2016 Table 1 Summary of the data 19912010 Description Mean Std Dev Min Max Lin y2010 yi991 Difference in logarithms of the income per capita 0799 0282 0739 1989 Incomepcyg91 Income per capita in 1991 234831 143577 33240 1185280 Human Average number of years of education 7485 2068 0910 12120 capital 599 Density 9 Population in a square kilometre 82122 456353 0055 12220780 Population Difference in logarithms of the population 0166 0312 1199 4321 Growth 9612010 Inmigration991 Proportion of immigrants in 19911995 over the population in 1991 0073 0101 0 2 Mining99 Proportion of employment in the mining sector 0013 0061 0 1 Industry 99 Proportion of employment in the industry sector 0185 0218 0 1 Construction901 Proportion of employment in the construction sector 0019 0075 0 1 Services Proportion of employment in the services and trade sector 0330 0264 0 1 andtrade 499 Publicsectorj901 Proportion of employment in the public sector 0398 0332 0 1 Source own APPLIED ECONOMICS 7 Table 2 Results of the BConvergence with the absolute and conditional approaches and by the OLS and the Random Slope Model 19912010 OLS Multilevel Variables in logarithms Absolute Conditional Absolute Conditional LnIncomepcyso1 0263 0386 0413 0461 LnHumancapital991 0365 0227 LnDensity9 0016 0031 PopulationGrowthy9912010 014 0009 Inmigration99 0241 009 Mining99 0133 0089 Industry 0036 0012 Construction901 0121 0057 Servicesandtradej99 0032 0009 Publicsectorj991 0024 0003 Constant 0799 0014 0747 0214 02 var InIncomepcyos1 001 0006 o2 var Constant 0039 0023 o2 var Residual 0032 0026 VPC in the mean 5428 531 General speed of convergence 16 257 28 33 N 5565 4067 5565 4067 LR test versus linear regression 267106 11094 Significant at 10 significant at 5 significant at 1 Population growth is adjusted with an additional 3 of capital depreciation and increment of technology Source own They also agree with the conclusions that other account that we only need to estimate three addi authors obtained in other countries such as Barro tional parameters et al 1991 or De La Fuente 2002 among many The coefficients in this model are similar to others those obtained with OLS estimation and similar However the contribution of this article consists to the ones obtained in Resende Carvalho and in using the multilevel approach to measure the De Sakowski 2016 with a set of spatial models intrastate differences in this global process of con Both models conditional and absolute multilevel vergence As shown in Table 2 the multilevel convergence show a negative and significant 6 approach presents a pvalue of the likelihood ratio parameter of 0413 and 0461 respectively To test LR test close to 0 which is significantly differ compare these with the OLS estimation this coef ent from the OLS version This test is based on the ficient represents a speed of convergence of 28 difference of the estimated likelihood of the null and 33 In the conditional models we can also model and the alternative with r restrictions in observe that human capital has a positive coeffi parameters see Equation 8 cient that agrees with the model of Mankiw Romer and Weil 1992 According to this litera IR 2InL nun InLatternativeXy 8 ture higher education implies a higher level of productivity In addition the coefficient of the This result allows us to use the Random Slope growth of population is negative and significant Model which has a fixed part of a general intercept as expected by the Solow model However the xo and a coefficient of convergence B but aside gctimation of population effect is nonsignificant from this fixed part it has a random part where By in the multilevel version On the other hand and B are an intercept and a coefficient specific for jmmigration and density have a positive and sig each state as was explained in Section 3 Thereforea nificant effect which could be related to accumu negative positive would indicate a higher lower lative processes in the territory see Fujita process of convergence in that area This second part Krugman and Venables 2001 Nevertheless the is estimated through the variance of each component effect of the immigration becomes less significant 007 and the covariance between them The effi in the multilevel model which could indicate that ciency of this model becomes clear if we take into an important part of this effect is constant within 8 A DIAZ DAPENA ET AL the states The coefficients of the shares of different Table 3 Multilevel conditional convergence with variability in sectors have a positive and significant effect only in ll the coefficients 19912010 Coefficient the specific case of construction These coefficients oa an LnIncomepcio91 0473 seem to indicate that there are no significant dif LnHumancapitalo91 0277 feren in hnol n rs Ln Density 9 0035 erences techno O8y betwee secto s PopulationGrowth9912010 0006 The result of the multilevel estimation can be Inmigrations9 0141 interpreted using the VPC as in Equation 3 indus ot et 1991 This coefficient indicates the percentage of the Construction991 0091 sg Servicesandtrade j991 002 variation caused by the aggregate level Although Publicsectorso 0003 different points can be used to evaluate the VPC Constant 0747 o7 varLnIncomepci9s1 0009 evaluation in the mean of the independent vari of varLnHumancapitalser 0048 able is used in order to obtain a central and 0 varLnDensity99 00004 representative value This coefficient has a signif ow vartinnigation sae 5 1991 icant value of 531 for this sample when it is 62 varMiningjgo 0017 evaluated in the mean a Cnet 0 Og varConstruction Finally the robustness of this analysis could be 0 varServicesandtrade so 0003 improved using a model that allows variability not O19 Var Publicsectori991 797E13 aL o2var Constant 0021 only in the Bcoefficient but also within all of of var Residual oo24 them With this model we want to introduce VPC in the mean 466 i 9 variability in other variables to avoid a bias caused General speed of convergence 57 by the omission of variability in the rest of the Significant at 10 significant at 5 significant at 1 coefficients Population growth is adjusted with an additional 3 of capital depreciation and increment of technology Table 3 shows the model of multilevel conver source own gence with the inclusion of variability in all of the coefficients The results are similar in the coeffi cients of all the variables In fact the coefficient of Table 4 Internal convergence of each state the multilevel ys conditional model with variability in all the coefficients the variability of the convergence in states is 19912010 almost the same Using an LR test it could not RegionStateSCSRGOVSateSSCCCSS be rejected that the variability in the coefficients of Convergence v Divergence v convergence is the same found in the model pre CENTRAL WEST CENTRAL WEST Goias GO 222 Distrito Federal DF 0378 sented in Table 2 In addition the general coeffi NORTH Mato Grosso do Sul 0393 cient of convergence shows a speed of convergence Amazonas AM 3386 Man rosso iu 0574 of 34 This coefficient is also coherent with the Rondénia RO 2459 NORTHEAST ex i i previous results in this article obtaining a speed of are A TV oee cand 0062 convergence of 33 in the conditional model of Tocantins To 0831 Parana PR 1807 Table 2 d with lit t t Amapa AP 0526 Rio Grande do Sul RS 4552 able 2 and with previous literature to compare Roraima RR 0335 SOUTHEAST with the results using an standard Spatial Durbin tee ea os fio de nee 0073 ahia 4969 spirito Santo 1139 Model see Resende Carvalho and De Sakowski Alagoas AL 1633 Sa Paulo SP 3684 2016 Maranhdao MA 1364 Minas Gerais MG 3959 Ceara CE 0788 Thanks to our approach we can differentiate Pernambuco PE 0612 among the global process and the intrastate pro Persie By Ooa cesses Table 4 shows the different coefficients of Rio Grande do Norte 0079 convergence inside each state These coefficients somn are used for further analysis to avoid possible Santa Catarina SC 0219 Omission of other variables in the variability of Significant at 10 significant at 5 significant at 1 the sl 4 Source own e slopes VPC estimation in this model can also be obtained but it has to be evaluated at a point in all the variables As a result it may be difficult to compare in the same terms as those obtained in Table 2 Discussion of the results We are going to focus our comments on the final model presented in Tables 3 general estimation and 4 intra States slopes Figure 1 shows the parameters of conver gence obtained in Table 4 by States distinguishing the convergence and divergence areas as well as significant and insignificant ones They are also represented in a caterpillar plot The visual representation of the results allows us to observe spatial patterns in intrastate convergence behaviour The higher levels of internal divergence are concentrated in the southeast states while Figure 1 Structure of the slopes 90 confidence intervals 19912010 APPLIED ECONOMICS 9 significant internal convergence is usual in the inter ior states as well as some cases of the Northeast The Southeast area of Brazil is one of the richest wards inside the country It has almost 80 million inhabitants 421 of the national population They have the highest levels of urbanization of the coun try and they hold 554 of total Brazilian GDP Moreover its income per capita is R30163 which is higher than the countrys average R22868 They show one of the most important growths in GDP per capita during the analyzed period As Montibeller Filho and Gargioni 2014 noted the economy of the Southeast is based mainly on man ufacturing activities It holds approximately 25 of Brazils employment rate in the processing industry see Agência Nacional do Petróleo Gás Natural e Biocombustíveis 2015 Bahia and Alagoas are the two cases with internal significant convergence being located in the coast northeast part Both cases especially Alagoas pre sent very intensive growth during the last decade thanks to the oil and gas industry development However even considering this intensive economic growth both states as well as the entire north area of Brazil are clearly under the national levels of GDP per capita having economies that mainly depend on agricultural activities or agroindustry On the other hand the rest of the states with sig nificant internal convergence are in the interior of the country Goiás and have economies mainly focused on agricultural activities but with lower levels of occupation of the land or large forest areas or Amazonian zones see Agência Nacional do Petróleo Gás Natural e Biocombustíveis 2015 In general it is interesting to observe how the cases with higher levels of industrialization or urba nization are the ones that normally present a higher trend to diverge while those places in which their economies are mainly focused in agricultural or mining sectors present internal convergence or not significant differences with the general behaviour of global convergence These results are especially interesting if we put it in context with the theoretical confrontation between the Neoclassical framework and Regional Economic theories Under the Neoclassical framework all the activities and conse quently all territories independent of their economic structure will follow the rule of decreasing returns which is behind that of the expected negative sign of the β parameter However under Regional Economics or New Economic Geography see Krugman 1991 Krugman and Venables 1995 Fujita Krugman and Venables 2001 Fujita and Thisse 2013 among others agglomeration and scale economies could avoid or delay the apparition of decreasing returns in central places while the peripheral areas lost activity in favour of the central ones These theories forecast that this break up between the centre and the periphery inside of a territory will appear in the context of strong urban concentration see Ciccone 2002 Combes Duranton and Gobillon 2011 or industrial development Our results of convergence for the case of Brazil with this new multilevel approach provide evidence in this sense The Neoclassical framework could explain well the processes in economies based in agribusiness or traditional activities that need to occupy the entire territory However the expected convergence along the territory seems to be more difficult with processes of industrialization and urba nization These economies are based in activities that tend to concentrate Therefore they are sensible to the phenomenon of agglomeration and scale econo mies This means when considering Brazil that if the country continues its process of industrialization and urbanization it will have to address a probable tendency to enlarge its territorial disparities It may happen especially at intrastate levels Thus the ana lysis of convergence at aggregated spatial levels could hide local behaviours of divergence V Summary and policy implications Brazil is one of the most important emerging econo mies in the world With 2027 million people its development has the potential of influencing the global economy At the same time Brazil still main tains high levels of economic disparities among peo ple as well as territories For this reason measuring the patterns of convergence in a case such as this is especially relevant and it explains the vast previous literature about convergence in Brazil The contribution of this article is measuring con vergence at global and local levels simultaneously to observe if different intraregional patterns exist though hidden in the general trend This is possible due to the application of the multilevel approach in 10 A DÍAZ DAPENA ET AL convergence analysis We used this new perspective in the Brazilian case measuring simultaneously the convergence at the country level but also at the level of each one of the 27 States of Brazil We use dis aggregated data at the municipal level from 1991 to 2010 Our results are in line with previous studies of convergence for Brazil The conclusions of the glo bal analysis are very similar to those obtained by previous authors quoted in the article Nevertheless the multilevel approach allows us to differentiate intrastate behaviours obtaining interesting new conclusions We found that the more industrialized or urbanized States of the coast in the South tend to present patterns of divergence The more devel oped industrialized orand urbanized the economy is the higher the divergence On the other hand the States mainly rooted in agricultural activities especially those with a high degree of agroindustry development and wide occupation of the territory are those with the highest internal levels of convergence These particular internal behaviours could be hid den in a global analysis and have important political implications According to these results as the Brazilian economy grows more and develops its manufacturing activities and urban agglomerations we will see an enlargement of the territorial dispa rities This will happen at a local level more than among large areas or States A strong Regional Policy should be considered to avoid territorial rup tures and according to our results that show impor tant intrastate heterogeneity the spatial level of implementation of this policy should be local In this sense the recent experiences of local smart specialization recipes in the European Union could be very interesting to design an a la carte regional policy for Brazil Disclosure statement No potential conflict of interest was reported by the authors Funding This work was supported by the Secretaría de Estado de Investigación Desarrollo e Innovación MINECO13 ECO201348161R ORCID Fernando Rubiera Morollón httporcidorg00000002 48540802 References Abramovitz M 1986 Catching Up Forging Ahead and Falling Behind The Journal of Economic History 46 385406 doi101017S0022050700046209 Abreu M de Groot H and Florax R 2005 Space and Growth A Survey of Empirical Evidence and Methods Région et Développement 21 1344 Agência Nacional do Petróleo Gás Natural e Biocombustíveis 2015 Petróleo E EstadoBrasilia Brazil Andersson L M Hammarstedt and S Hussain 2013 Ethnic Origin Local Labour Markets and Self Employment in Sweden A Multilevel Approach The Annals of Regional 50 3 885910 Andrade E M Laurini R Madalozzo and P L Valls Pereira 2004 Convergence Clubs among Brazilian Municipalities Economics Letters 83 2 179184 doi101016jeconlet200311005 Anselin L 1988 Spatial Econometrics Methods and Models Amsterdam Springer Anselin L and R J G M Florax 1995 New Directions in Spatial Econometrics New York Springer Anselin L and S J Rey 1997 Introduction to the Special Issue on Spatial Econometrics International Regional Science Review 20 12 17 doi101177 016001769702000101 Arellano M and S Bond 1991 Some Tests of Specification for Panel Data Monte Carlo Evidence and an Application to Employment Equations The Review of Economic Studies 58 2 277297 doi1023072297968 Azzoni C R 2001 Economic Growth and Regional Income Inequality in Brazil The Annals of Regional Science 35 133152 doi101007s001680000038 Azzoni C R N M Filho and T D Menezes 2000 Geography and Income Convergence among Brazilian States Washington InterAmerica Development Bank Barro R J X SalaIMartin O J Blanchard and R E Hall 1991 Convergence across States and Regions Brookings Papers on Economic Activity 1991 1 107182 doi1023072534639 Baumol W J 1986 Productivity Growth Convergence and Welfare What the LongRun Data Show The American Economic Review 76 10721085 Bernard A B and S N Durlauf 1995 Convergence in International Output Journal of Applied Econometrics 10 2 97108 doi101002ISSN10991255 Carlino G and L Mills 1996 Convergence and the US States A TimeSeries Analysis Journal of Regional Science 36 4 597616 doi101111jors199636issue4 Caselli F G Esquivel and F Lefort 1996 Reopening the Convergence Debate A New Look at CrossCountry APPLIED ECONOMICS 11 Growth Empirics Journal of Economic Growth 1 3 363 389 doi101007BF00141044 Chasco C and AM López 2009 Multilevel Models An Application To The Betaconvergence Model Région Et Développement 30 3558 Ciccone A 2002 Agglomeration Effects in Europe European Economic Review 46 2 213227 doi101016 S0014292100000994 Cohen P N 1998 Black Concentration Effects on Black White and Gender Inequality Multilevel Analysis for US Metropolitan Areas Social Forces 77 1 207229 doi1023073006015 Combes P P G Duranton and L Gobillon 2011 The Identification of Agglomeration Economies Journal of Economic Geography 11 2 253266 doi101093jeglbq038 Comisión Económica para América Latina y el Caribe 2013 Panorama Del Desarrollo Territorial De América Latina Y El Caribe Pactos Para La Igualdad Territorial CEPAL Santiago de Chile Comisión Económica para América Latina y el Caribe 2015 Panorama Del Desarrollo Territorial En América Latina Y El Caribe Pactos Para La Igualdad Territorial CEPAL Santiago de Chile Coulombe S and F Lee 1993 Regional Economic Disparities in Canada Ottawa University of Ottawa Cravo T A and E Soukiazis 2011 Human Capital Thresholds and Economic Growth in Brazil RDE Revista De Desenvolvimento 23 106120 de la Fuente A 2002 On the Sources of Convergence A Close Look at the Spanish Regions European Economic Review 46 3 569599 doi101016S0014292101001611 DíazDapena A E FernándezVázquez and F Rubiera Morollón 2016 The Role of Spatial Scale in Regional Convergence The Effect of MAUP in the Estimation of BetaConvergence Equations The Annals of Regional Science 56 2 473489 doi101007s0016801607500 Easterlin R 1960 Interregional Differences in per Capita Income Population and Total Income 18401950 Trends in the American Economy in the Nineteenth The Conference on Research in Income and Wealth NBER Books Washington 73140 Elhorst JP 2003 Specification And Estimation Of Spatial Panel Data Models International Regional Science revue 26 3 244268 Elhorst JP G Piras and G Arbia 2010 Growth and conver gence in a multiregional model with spacetime dynamics Geographical Analysis 3 338355 doi1023071913103 Fazio G and D Piacentino 2011 Testing for Convergence from the Microlevel Discussion Papers Dipartimento di Studi Economici Salvatore Vinci Universitá degli Studi di Napoli Naples Italy Ferreira A 2000 Convergence in Brazil Recent Trends and LongRun Prospects Applied Economics 32 479490 doi101080000368400322642 Ferreira A and C C Diniz 1995 Convergência Entre as Rendas per Capita Estaduais No Brasil Revista De Economia Política 15 4 3856 Ferreira E A and R F Tatiwa 2014 Dinâmica Intrarregional Do Brasil Quem Dirige O Crescimento Industrial Das Regiões Brasileiras Economia Aplicada 18 2 243270 doi10159014138050ea404 Ferreira P C and R Ellery Jr 1996 Convergência Entre a Renda per Capita Dos Estados Brasileiros Revista De Econometria 16 5482 Fujita M P Krugman and A Venables 2001 The Spatial Economy Cities Regions and International Trade Bolston MIT press Fujita M and J Thisse 2013 Economics of Agglomeration Cities Industrial Location and Globalization Cambridge Cambridge University press Goldstein H 1986 Multilevel Mixed Linear Model Analysis Using Iterative Generalized Least Squares Biometrika 73 1 4356 doi101093biomet73143 Goldstein H 2011 Multilevel Statistical Models London John Wiley Sons HoltzEakin D W Newey and H S Rosen 1988 Estimating Vector Autoregressions with Panel Data Econometrica Journal of The Econometric Society 56 1371 doi1023071913103 Hox J 2010 Multilevel Analysis Techniques and Applications New York Routledge Islam N 2003 What Have We Learnt from the Convergence Debate Journal of Economic Surveys 17 3 309362 doi101111joes200317issue3 Krugman P 1991 Increasing Returns and Economic Geography The Journal of Political Economy 99 3 483499 doi101086261763 Krugman P and A J Venables 1995 Globalization and the Inequality of Nations The Quarterly Journal of Economics 110 4 857880 doi1023072946642 Kuznets S 1955 Economic Growth and Income Inequality The American Economic Review 45 1 128 Lasage JP and MM Fischer 2008 Spatial Growth Regressions Model Specification Estimation And Interpretation Spatial Economic Analysis 3 3 275 Li Y and Y H D Wei 2010 The SpatialTemporal Hierarchy of Regional Inequality of China Applied Geography 30 303316 doi101016japgeog200911001 Magalhães A and G J D Hewings 2005 Spatial Dependence and Regional Convergence in Brazil Investigaciones Regionales 6 520 Magrini S 2004 Regional Di Convergence In Handbook of Regional and Urban Economics 4 eds V Henderson P Nijkamp E S Mills P C Cheshire and J F Thisse 2741 2796 Amsterdam Elsevier Mankiw N G D Romer and D N Weil 1992 A Contribution to the Empirics of Economic Growth The Quarterly Journal of Economics 107 2 407437 doi1023072118477 Montibeller Filho G and S L Gargioni 2014 Desenvolvimento Da Região Sul Do Brasil Um Olhar Territorial Para O Desenvolvimento Sul Rio De Janeiro Rio de Janeiro Banco Nacional de Desenvolvimento Econômico e Social 12 A DÍAZ DAPENA ET AL Quah D 1993 Empirical CrossSection Dynamics in Economic Growth European Economic Review 37 23 426434 doi1010160014292193900315 Resende G M 2011 Multiple Dimensions of Regional Economic Growth The Brazilian Case 19912000 Papers in Regional Science 90 3 629662 doi101111 j14355957201000336x Resende G M A X Y Carvalho and P A M De Sakowski 2016 Evaluating Multiple Spatial Dimensions of Economic Growth in Brazil Using Spatial Panel Data Models The Annals of Regional Science 56 1 131 doi101007s0016801507069 Rey S J and B D Montouri 1999 US Regional Income Convergence A Spatial Econometric Perspective Regional Studies 33 2 143156 doi101080 00343409950122945 SalaIMartin X 1996 The Classical Approach to Convergence Analysis The Economic Journal 106 1019 1036 doi1023072235375 Shioji E 1992 Regional Growth in Japan New Haven US Yale University Snijders T A B and R Bosker 1999 Multilevel Modeling An Introduction to Basic and Advanced Multilevel Modeling Souza P H G F and R G Osorio 2014 A Redução Das Disparidades Regionais E A Queda Da Desigualdade Nacional De Renda Revista Planejamento E Políticas Públicas 44 9 12071220 Srholec M 2010 A Multilevel Approach to Geography of Innovation Regional Studies 44 9 12071220 doi10108000343400903365094 Williamson J G 1965 Regional Inequality and the Process of National Development A Description of the Patterns Economic Development and Cultural Change 13 4 184 doi101086450136 Zini Junior Á A 1998 O Acordo Sobre Compras Governamentais Da Organizaçao Mundial Do Comércio E a Economia Brasileira Estudos Econômicos 28 2 181198 APPLIED ECONOMICS 13