Policy Research Working Paper 10904 Regional Convergence in Brazil Kaleb Abreha Rafael Ornelas Gabriel Zaourak Macroeconomics, Trade and Investment Global Practice September 2024 Policy Research Working Paper 10904 Abstract This paper examines whether labor productivity converged agriculture, extractives, and manufacturing. These findings across Brazil’s states (“departments”) between 2002 and of the regional convergence are robust to controlling for 2018. The results show strong evidence of unconditional state and industry fixed effects, states’ initial poverty rates, convergence in which states with lower levels of initial labor human capital, tax collection per capita, and infrastructure. productivity experienced substantially faster labor produc- Given the high disparity in labor productivity across Brazil’s tivity growth. The convergence rate was faster over 2002–10 states, such regional convergence has the potential to raise compared to 2010–18 period and particularly strong in aggregate productivity and per capita income. This paper is a product of the Macroeconomics, Trade and Investment Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at gzaourak@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Regional Convergence in Brazil∗ Kaleb Abreha Rafael Ornelas Gabriel Zaourak JEL Classification: O11, O47, O54, R11, R12 Keywords: Regional inequality, regional convergence, labor productivity, Brazil ∗ The World Bank, 1818 H Street, NW Washington, DC 20433, USA. Corresponding email addresses: gza- ourak@worldbank.org (G. Zaourak), kaleb.abreha@gmail.com (K. Abreha), and ramaralornelas@worldbankgroup.org (R. Ornelas). The views expressed herein are solely our own and should not be attributed to the World Bank, its executive directors, the countries they represent, the Mosbacher Institute, the Bush School, or Texas A&M University. All errors and omissions are ours. 1 Introduction Evidence shows a substantial and widening gap in per capita income across countries although income distribution became roughly stable in the early 2000s (Hall and Jones 1999; Jones 2016). Jones (2016, p. 37-38) reports that “the poorest countries in 1960 such as Ethiopia were only about 32 times poorer than the United States. By 2011, there are many countries with relative incomes below this level, and both Niger and the Central African Republic were more than 64 times poorer than the United States.” Relatedly, regional inequalities are as equivalently high and ubiquitous as in cross-country cases, especially among low-income countries. Gennaioli et al. (2014) find considerable regional dispari- ties within countries for a large sample of 107 countries in 2005. Comparing the per capita income of the richest to the poorest regions, they find an average ratio of 4.41. This ratio is 3.77 for Africa, 5.63 for Asia, 3.74 for Europe, 4.60 for North America, and 5.61 for South America. These results on regional inequalities hold even after excluding extremely poor and rich regions: the average stan- dard deviation of per capita income in the 20 poorest countries is 1.64 times larger than the average dispersion of per capita income in the 20 richest countries (Gennaioli et al. 2014).1 Against this backdrop of significant spatial (across countries and regions) per capita income gaps, a large body of research has tested for evidence of convergence and the drivers of economic growth. In this respect, the convergence hypothesis states that the distribution of per capita income across countries converges in the long run, and hence, it will be independent of countries’ initial conditions. Under the Solow-Swan growth model, capital exhibits diminishing returns, which leads to the reallocation of capital to low-income countries. These countries will then experience faster economic growth and eventually catch up with their high-income counterparts. The speed of convergence, however, depends on several factors, including geography, institu- tions, policy, and technological progress. In addition, although spatial (across countries and re- gions) convergence occurs, initial differences in per capita income and productivity persist because of agglomeration economies. These small differences compound over time to render high inequali- ties. For instance, if all regions in Brazil have the same steady-state per capita income, a region that 1 Price differences across the regions partly explain the regional disparities observed. However, it is unlikely for the price differences to fully explain the significant regional variations in nominal incomes. The regional variations also depend on how regions are weighted in these computations given differences in regions’ resource endowments, population sizes and age structures, and employment rates. 2 is currently one standard deviation below steady-state income will require about 23 years to close the gap at a convergence rate of 2 percent per year (Gennaioli et al. 2014). In this paper, we investigate whether labor productivity converged across Brazil’s states (“de- partments”) between 2002 and 2018. During the sample period, our findings show strong evidence of unconditional regional convergence in which states with lower initial levels of labor productiv- ity experienced substantially faster labor productivity growth. Additionally, the convergence rate is faster for the period 2002-10 compared with the 2010-18 period. The convergence rate is particularly strong in agriculture, extractives, and manufacturing. These findings also hold when controlling state and industry fixed effects as well as states’ initial poverty rate, human capital, per capita tax collection, and infrastructure. This paper is related to the literature on convergence, especially regional convergence. It is re- lated to Gennaioli et al. (2014), who show that regional convergence is relatively faster than cross- country convergence but generally slower than expected using a large sample of countries, including Brazil. We complement their findings by focusing on labor productivity (rather than per capita in- come) and using a matched employer-employee dataset. Unlike Azzoni and Castro (2023) who provide evidence of stronger convergence across states of Brazil only during periods of economic crisis (2008 and 2014) and show that periods of better aggregate growth are linked with increasing regional dispersion, our results generally show evidence of regional convergence between 2002 and 2018. Our results also contribute to country case studies on regional inequality that account for the sectoral dimension (e.g., Kinfemichael and Morshed 2019a; Iacovone et al. 2015). The rest of this paper is organized as follows. Section 2 provides a brief survey of the evidence on the patterns and drivers of convergence from cross-country and country case studies. Section 3 presents the main data sources and summary statistics on labor productivity, poverty rate, and ge- ographic concentration of different sectors. It then introduces the test used to detect if there is any form of regional convergence. Section 4 discusses the results, and section 5 concludes. 3 2 Literature review A common approach to testing the convergence hypothesis is the β -convergence that relies on whether poorer countries grow faster. It can be tested by using country-level cross-sectional or panel data: yi,t log = α + βlog (yi, t−τ ) + εi,t (1) yit−τ where y represents per capita income in country i and time t,2 τ > 0, and ε is an error term. Em- pirically, testing for convergence entails conducting a test of the null hypothesis β = 0 against the alternative β < 0 (Johnson and Papageorgiou 2020). This test presumes that initial per capita income contains sufficient information to capture countries’ initial conditions. If the null hypothesis is re- jected, this provides support to the β -convergence hypothesis. Specifically, it implies the occurrence of unconditional convergence in that there is no effect of initial conditions in the long run. However, several studies show that initial per capita income is not a sufficient statistic to deter- mine convergence, and therefore, other factors such as geography, human capital, and institutions need to be accounted for in characterizing income growth.3 This is called conditional convergence and can be tested as in equation 1 after introducing controls (Ci,t−τ ) and country fixed effects (λi ): yi,t log = α + β log (yi,t−τ ) + δCi,t−τ + λi + εi,t (2) yi,t−τ Another approach to testing for convergence is the σ -convergence depending on whether cross- country dispersion in per capita income falls over time. That is, between two different points in time 2 > σ 2 , where σ 2 is the cross- t and t + τ where τ > 0, the presence of σ -convergence yields σt t+τ country dispersion in per capita income. It is important to note that β -convergence is a necessary but not sufficient condition for σ -convergence.4 Empirically, testing for convergence is commonly done using linear models (as in equations (1) and (2)). Studies have also used non-linear models to allow for possibilities of club convergences and divergence. Johnson and Papageorgiou (2020) categorize these models into those clustering the data 2 Although the specification in equation 1 is based on per capita income, it can be generalized to productivity or any other economic development indicators. 3 See, for example, Durlauf and Johnson (1995), Fiaschi et al. (2019), and Tan (2010). 4 Using U.S. county-level data on real per capita income, Young et al. (2008) show that β -convergence does not necessarily imply σ -convergence. 4 into groups of countries, models having smoothly varying parameters, and models characterizing the cross-country per capita income distribution (not just a few moments). On the other hand, a large number of studies that apply linear or non-linear models use primarily cross-sectional or panel data of countries. However, recent papers have started using time-series methods. Consistent with results from non-linear models, which strongly reject unconditional convergence in favor of club convergence, results based on time-series methods mostly find evidence in support of club convergence (Johnson and Papageorgiou 2020). Cross-country convergence In one of the recent studies, Bakker et al. (2020) find no support for unconditional convergence and show that differences in human capital and total factor productivity (TFP)—not the capital- labor ratio—drive cross-country differences in GDP per capita in a sample of emerging economies in Europe and Asia and advanced countries between 1996 and 2017. Their findings are in line with the occurrence of conditional convergence. Unlike the rapid rate of convergence in emerging Asia ad and Europe, their findings uncover weaker growth performance and a lack of convergence in Latin America although most of the countries in the region experienced less macroeconomic volatility and favorable terms of trade by comparison. Similarly, Kollias et al. (2021) document the absence of labor productivity convergence in Latin America between 1970 and 2014. In fact, they find countries’ productivity contracted, including the ineffectiveness of the structural reforms of the 1990s. Additionally, Barrios et al. (2019) find no evidence of overall convergence but evidence of club convergence among a sample of 17 Latin American countries during 1990-2014, which does not nec- essarily depend on economic ties or geographical proximity.5 Applying the same method to a larger set of 63 countries during 1960-2011, Tebaldi (2016) also finds evidence against overall convergence in TFP but that supports club convergence and show that initial conditions (TFP, institutional quality, and economic openness) matter for TFP growth. Similarly, Borsi and Metiu (2015) show a long-run difference between new and old EU members and a lack of overall convergence between 1970 and 2010, but find evidence of club convergence depending on geographic region, and not necessarily related to EU membership. 5 Barrios et al. (2019) apply Phillips and Sul’s (2007) method, which characterizes the transition path of individual coun- tries and tests if they have a common growth path. Recent methods do not require a prior grouping of countries; clusters of countries would emerge endogenously (Beylunioglu et al. 2020; Phillips and Sul 2007). 5 Sectoral convergence Given the patterns of cross-country convergence are reflections of and shaped by countries’ sectoral composition, it is insightful to investigate sectoral convergence. Using data on more than 100 coun- tries, Rodrik (2013) shows strong evidence of unconditional convergence in manufacturing labor productivity, as well as σ -convergence for a sub-sample of countries. However, there is no aggregate convergence because of a rather small share of formal manufacturing employment and the slow pace of industrialization in low-income countries. In addition, Kinfemichael and Morshed (2019a), using 95 countries, find evidence of unconditional convergence for aggregate labor productivity, primarily driven by the fast convergence in services. Kinfemichael (2019) also reports similar results. By contrast, Herrendorf et al. (2022), based on a new dataset for low-income countries during 1990-2018, uncover that cross-country differences in productivity gaps (with high-income countries) are larger in manufacturing than in the aggregate, and no observed tendency for productivity to con- verge in manufacturing. They also find a positive correlation between manufacturing labor produc- tivity growth and aggregate and other sectors’ labor productivity growth, suggesting weak growth of manufacturing employment is not necessarily the force behind the lack of convergence in low- income countries. The seemingly sharp contrast findings between Rodrik (2013) and Herrendorf et al. (2022) are partly attributed to the role and size of the informal sector in developing coun- tries. Rodrik (2013)’s findings in support of unconditional convergence are based on data on larger establishments in the formal manufacturing sector.6 When the sample coverage extends to small and informal establishments and workers, which are highly prevalent in developing countries, as in Herrendorf et al. (2022), the data does not reveal unconditional convergence in manufacturing. An earlier study by Wong (2006) decomposes the pattern of β -convergence into within- and between-sector contributions and show that both the within-sector growth and between-sector re- allocations. Among the OECD countries during 1970-1990, Wong (2006) shows sizable contribu- tions of productivity growth in agriculture and services to aggregate convergence, but limited role of labor reallocation and productivity growth in manufacturing. Relatedly, Dieppe and Matsuoka (2021) apply the same technique on a larger sample of advanced and developing countries and more recent years 7 and show that both the within-sector growth and between-sector reallocations 6 Rodrik (2013) clearly indicates the context in which to interpret the findings; they apply to formal manufacturing. 7 Their results are based on sectoral data for 91 countries over 1995-2018, including a longer horizon (between 1975 and 2018) for a sub-sample of 60 countries. 6 have contributed significantly toward convergence in aggregate labor productivity. Interestingly, productivity growth in agriculture was the main driver of aggregate convergence. Nevertheless, agricultural productivity in low-income countries still markedly lags and is weakly converging rel- ative to advanced countries. As for manufacturing, they find evidence in support of unconditional convergence consistent with Rodrik (2013) albeit with a slower convergence rate. In line with Bakker et al. (2020), their findings also imply addressing barriers to sectoral reallocation through human capital development, good governance, and a favorable business environment would boost produc- tivity growth and accelerate convergence. Regional convergence Studies have also examined the patterns of regional inequalities and convergence within countries. Gennaioli et al. (2014), based on regional data from 83 countries, show that rich countries experi- enced faster regional convergence than poor countries, mainly attributed to better capital markets in the former and limited factor mobility in the latter. Their results further show that the pace of convergence across regions within countries is relatively faster than the one observed across coun- tries.Yet, the regional convergence rate is generally slow, as indicated by lower resource mobility. This is despite regions being more likely to exhibit homogeneity in technology, productivity, and institutions compared to countries. Examining the determinants of regional growth, Gennaioli et al. (2014) show that geography (e.g., resource endowment, temperature, proximity to the ocean) and human capital (e.g., educa- tional attainment) are important factors but not culture (e.g., ethnic heterogeneity, trust) and in- stitutions (e.g., business environment) factors. More importantly, they highlight the crucial role of human capital through the mechanisms of worker education, entrepreneurial/managerial educa- tion, and externalities while noting the ease with which firms, workers, and entrepreneurs move across regions than countries. Coming to country-specific studies, Kinfemichael and Morshed (2019b) argue that the slow- down of labor productivity convergence in recent years in the United States is possibly due to factors such as the decline in cross-state migration, increase in housing costs in major cities, and strengthen- ing of agglomeration economies. During the period 2000-2012 for Peru, Iacovone et al. (2015) report evidence of labor productivity cross-region convergence in manufacturing and mining sectors, but 7 there is no strong evidence of convergence in agriculture and services. More importantly, their re- sults show that the rate of convergence in manufacturing and mining has been strong enough to lead aggregate convergence across regions, albeit at a slower pace because of high employment shares of agriculture and services.8 In the case of Brazil, Azzoni and Castro (2023) examined regional disparities from 2002 to 2019. Their findings indicate a general increase in σ -convergence in per capita income across regions. Moreover, periods of economic crisis, such as the Great Recession of 2008 and the economic crisis of 2014, were linked to more pronounced convergence. The study also reveals evidence of conditional convergence in regional per capita income and wages (serving as a proxy for labor productivity) during these crisis periods. Overall, this observed (or lack thereof) convergence and its speed call for further evidence that accounts for the drivers of factor accumulation, sectoral composition, technology diffusion, and pro- duction externalities within and across regions and countries. 3 Empirical approach 3.1 Data Given data availability, we focus on the period from 2002 to 2018. The main dataset comes from the Relação Anual de Informações Sociais (RAIS), which is a matched employer-employee dataset put together by the Ministry of Labor and Employment (Ministério do Trabalho e Emprego). RAIS covers all registered firms and formal employees in which a firm is identified by its registration number (CNPJ) and a worker by its identification number (PIS). Most importantly, both the PIS and CNPJ do not change over time, even when a worker stops to work in the formal sector. These unique features allow us to track workers and firms over time. The dataset contains detailed information on firms’ geographic location, economic classification (CNAE 5-digit), employment size, and personnel records such as monthly wage, age, gender, and education, among others.9 This dataset is used to compute the key variable of interest: labor productivity at the sectoral and industry levels. Another data source is derived from the Pesquisa Nacional por Amostra de Domicílios (PNAD). 8 They also report lack of convergence in poverty rates across regions because of limited labor mobility toward converging sectors and high employment of the poor in sectors exhibiting weak convergence. 9 National Classification of Economic Activities (CANE) is Brazil’s system of economic activity classification in the pro- duction and dissemination of economic statistics. 8 This data is gathered through surveys conducted with individuals from 27 states ( ’departments’), covering aspects such as education, employment, income, and housing characteristics, among oth- ers.10 We primarily utilize this dataset to calculate the baseline poverty rate for each state, defined as the proportion of the population earning an income below half of the minimum wage. 3.2 Convergence test Using the RAIS dataset, we compute labor productivity for each sector and subsector in each state as follows: W age bill (i, j, t) yi,j, t = Employment(i, j, t) where i represents a state, j a sector or an industry, and t a time period. We use the following four sectors: agriculture, extractives, manufacturing, and services. Each industry in the CNAE classifi- cations (2-digit) is grouped into one of these sectors.11 We define labor productivity using the wage bill because data value-added are available only at the sector level. However, this will not be an issue given that the wage bill is a good proxy for labor productivity. Then, we test for the β -convergence across states using the estimating equation: yi,j,t log = α + β log (yi,j,t−τ ) + εi,j,t (3) yi,j,t−τ where τ > 0 and εi,j,t is an error term. We consider 8-year and 16-year windows. If data rejects the null hypothesis β = 0 against the alternative β < 0, it shows evidence of unconditional regional convergence. Alternatively, we test for conditional convergence by controlling the state (si ) and industry (λj ) fixed effects, and for states’ initial period poverty rate, indicators of human capital, per capita tax collection, and infrastructure access: yi,j, t log = α + βlog (yi,j,t−τ ) + δCi,j,t−τ + si + λj + εi,j,t (4) yi,j,t−τ We also run equations 3 and 4 regression at the aggregate levels; that is, all the sectors combined. 10 See the list of states in Appendix A. 11 See the list of sectors and industries within the CNAE classification in Appendix A. 9 4 Results 4.1 Unconditional convergence Considering all the sectors in the aggregate, Figure 1 plots the average annualized growth rates in la- bor productivity during 2002-10 and 2010-18 against the initial labor productivity levels in 2002 and 2010, respectively. The pattern clearly shows evidence of the unconditional convergence of produc- tivity across regions. States with initial lower productivity levels in 2002 experienced higher growth rates annually, catching up with the leading states. The convergence rate is particularly faster for the 2002-10 period compared with the 2010-18 period. Furthermore, all the sectors have experienced unconditional productivity convergence with the regional convergence pattern looking stronger in non-agriculture sectors (Figure 2). We see that initial levels of labor productivity are relatively lower in agriculture compared with the other sectors. Considering industry-level labor productivity within each sector, the coefficient estimates in Ta- ble 1 provide strong support for unconditional convergence across regions.12 Looking at the entire sample period (2002-18) and in the aggregate, the estimate is negative (-0.023) and statistically sig- nificant, representing a coverage rate of about 2.3 percent. The estimates for individual sectors are also negative and significant; entire sample period (2002-18), the estimates are -0.023, -0.031 (agri- culture), -0.030 (extractives), -0.028 (manufacturing), and -0.020 (services). These results imply that states with lower levels of labor productivity in 2002, on average, experienced faster produc- tivity growth over the sample period. This regional convergence holds when considering both the regional economies and disaggregated by sectors. 12 In the appendix, we report the estimation results based on labor productivity computed at different levels of aggrega- tions. Except for agriculture and services, regression results at broader levels of aggregation are comparable to estimates at the industry levels within each of the four sectors. The aggregation level seems to have an influence on the signif- icance and size of the estimate. These features suggest substantial variation in productivity growth across industries within these sectors. Some industries may have experienced rapid growth but their shares in the sectoral economy are probably small. 10 Figure 1: Aggregate labor productivity growth, 2002-2018 Source: World Bank staff calculations using RAIS. Figure 2: Average labor productivity growth by sector, 2002-2018 (a) Agriculture (b) Extractives (c) Manufacturing (d) Services Source: World Bank staff calculations using RAIS. 11 Table 2 compares the patterns of regional convergence in the first and second halves of the sample period and shows that the convergence coefficient is negative and significant in the two sub-periods as well as all the sectors. The sizes of the estimates are also systematically larger in the 2002-10 period compared with the estimates in 2010-18. In the 2002-10 period, the estimates are -0.038 (all sectors), -0.079 (agriculture), vs -0.059 (extractives), -0.041 (manufacturing), and -0.035 (services). The corresponding estimates in 2010-18 are -0.019, -0.020, -0.021, -0.030, and -0.017. The convergence rate is especially low for services. Table 1: Growth of Aggregate and Sectoral Labor Productivity 2002-18 (Labor Productivity Com- puted at the Industry Level) All Sectors Agriculture Extractives Manufacturing Services (1) (2) (3) (4) (5) Initial labor productivity (log) -0.023∗∗∗ -0.031∗∗∗ -0.030∗∗∗ -0.028∗∗∗ -0.020∗∗∗ (0.0011) (0.0076) (0.0053) (0.0020) (0.0012) Constant 0.153∗∗∗ 0.201∗∗∗ 0.210∗∗∗ 0.186∗∗∗ 0.137∗∗∗ (0.0069) (0.0447) (0.0353) (0.0124) (0.0073) Industry FEs No No No No No State FEs No No No No No Obs. 2,266 81 112 625 1,448 R2 0.294 0.228 0.291 0.440 0.255 Source: World Bank staff using RAIS. Note: Robust standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. 4.2 Conditional convergence A multiplicity of factors influences the pace of regional convergence in productivity. Some are geo- graphic (e.g., resource endowment, sea access), institutional (e.g., regulatory capacity, market dis- tortions), and policy. To this end, the regional convergence regressions are now estimated after con- trolling for industry fixed effects (the levels and productivity growth rates inherently differ due to differences in production and market structure) and state fixed effects (to capture the effect of mostly geographic factors on regional development). In addition, these regression account for states’ ini- tial level of development (approximated by poverty rates) and determinants of development (e.g., population size and age composition, labor market features, and infrastructure). Table 3 presents tests for the presence of regional conditional convergence. Column (1) shows that including the industry and state fixed effects accelerates productivity growth and promotes re- 12 Table 2: Growth of sectoral labor productivity 2002-10 versus 2010-18 (at the industry level) All sectors Agriculture Extractives Manufacturing Services 2002-10 2010-18 2002-10 2010-18 2002-10 2010-18 2002-10 2010-18 2002-10 2010-18 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Initial labor productivity (log) -0.038∗∗∗ -0.019∗∗∗ -0.079∗∗∗ -0.020∗ -0.059∗∗∗ -0.021∗∗ -0.041∗∗∗ -0.030∗∗∗ -0.035∗∗∗ -0.017∗∗∗ (0.0020) (0.0019) (0.0152) (0.0117) (0.0094) (0.0098) (0.0032) (0.0035) (0.0022) (0.0020) 13 Constant 0.253∗∗∗ 0.131∗∗∗ 0.486∗∗∗ 0.136∗ 0.408∗∗∗ 0.154∗∗ 0.265∗∗∗ 0.195∗∗∗ 0.228∗∗∗ 0.116∗∗∗ (0.0126) (0.0121) (0.0895) (0.0714) (0.0625) (0.0660) (0.0200) (0.0220) (0.0139) (0.0127) Industry FEs No No No No No No No No No No State FEs No No No No No No No No No No Obs. 2,293 2,247 80 80 113 107 637 622 1,463 1,440 R2 0.237 0.099 0.350 0.046 0.321 0.059 0.305 0.191 0.200 0.093 Source: World Bank staff using RAIS. Note: Robust standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. Table 3: Growth of aggregate and sectoral labor productivity, 2002-2018 All Agriculture Extractives Manufacturing Services (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Initial labor productivity (log) -0.040∗∗∗ -0.024∗∗∗ -0.042∗∗∗ -0.044∗∗∗ -0.035∗∗∗ -0.031∗∗∗ -0.042∗∗∗ -0.031∗∗∗ -0.045∗∗∗ -0.021∗∗∗ (0.0019) (0.0012) (0.0099) (0.0076) (0.0075) (0.0056) (0.0031) (0.0022) (0.0025) (0.0012) Initial poverty rate -0.001∗∗∗ -0.001∗∗∗ -0.001 -0.001∗∗∗ -0.001∗∗∗ (0.0001) (0.0004) (0.0010) (0.0002) 14 Constant - 0.165∗∗∗ - 0.286∗∗∗ - 0.227∗∗∗ - 0.208∗∗∗ - 0.147∗∗∗ (0.0075) (0.0460) (0.0405) (0.0143) (0.0078) Industry Fes Yes No Yes No Yes No Yes No Yes No State Fes Yes No Yes No Yes No Yes No Yes No Obs. 2,266 2,266 81 81 112 112 625 625 1,448 1,448 R2 0.624 0.306 0.817 0.341 0.580 0.301 0.634 0.467 0.706 0.264 Source: World Bank staff using RAIS. Note: Robust standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. Table 4: Growth of sectoral labor productivity 2002-10 versus 2010-18 Agriculture Extractives Manufacturing Services (1) (2) (3) (4) (5) (6) (7) (8) Initial labor productivity (log) -0.095∗∗∗ -0.037∗∗∗ -0.060∗∗∗ -0.022∗∗ -0.043∗∗∗ -0.034∗∗∗ -0.035∗∗∗ -0.018∗∗∗ (0.0156) (0.0124) (0.0101) (0.0098) (0.0035) (0.0041) (0.0022) (0.0021) Initial poverty rate -0.002∗∗ -0.002∗∗∗ -0.001 -0.001 -0.001∗∗∗ -0.001∗∗∗ -0.0004 -0.001∗∗∗ (0.0007) (0.0005) (0.0018) (0.0015) (0.0003) (0.0003) (0.0002) (0.0002) 15 Constant 0.597∗∗∗ 0.259∗∗∗ 0.416∗∗∗ 0.163∗∗ 0.288∗∗∗ 0.233∗∗∗ 0.235∗∗∗ 0.133∗∗∗ (0.0927) (0.0774) (0.0748) (0.0653) (0.0227) (0.0279) (0.0147) (0.0143) Industry FEs No No No No No No No No State FEs No No No No No No No No Obs. 80 80 113 107 637 620 1,463 1,440 R2 0.390 0.188 0.322 0.061 0.315 0.216 0.201 0.108 Source: World Bank staff using RAIS. Note: Robust standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. gional convergence. This result also holds when examining the regional convergence in the major sectors separately. We also see in column (2) that states with an initial high prevalence of poverty rate have experienced relatively higher productivity; this effect is considerably stronger in the agri- cultural sector suggesting the prevalence of poverty in agriculture-intensive regions. Note that the convergence rate in the services sector is about half when initial poverty is accounted for relative to the estimates when the industry and state fixed effects are included. A feature worth mentioning is the fact that after conditioning on departmental poverty rates or fixed effects, convergence ap- pears to be significant and faster when the data is disaggregated by industries. This suggests that productivity convergence would have been even faster if industries were more “similar”. Table 5: Growth of aggregate labor productivity, 2002-18 (1) (2) (3) (4) (5) (6) Initial labor productivity -0.024∗∗∗ -0.024∗∗∗ -0.023∗∗∗ -0.024∗∗∗ -0.023∗∗∗ -0.024∗∗∗ (log) (0.0012) (0.0012) (0.0012) (0.0012) (0.0011) (0.0012) Initial poverty rate -0.0006∗∗∗ (0.0001) Initial years of schooling (tot. 0.003∗∗∗ pop.) (0.0007) Initial working age popula- 0.000 tion (% of tot. pop.) (0.0002) Initial government tax col- 0.004∗∗∗ lected per capita (log) (0.0009) Initial HHs with access to 0.0002∗∗ electricity network (% total HHs) (0.0000) Initial HHs with landline or 0.0002∗∗∗ mobile telephone (% of total HHs) (0.0000) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Constant 0.165 0.144 0.151 0.138 0.136 0.152∗∗∗ (0.0075) (0.0068) (0.0115) (0.0071) (0.0100) (0.0068) Industry FEs No No No No No No State FEs No No No No No No Obs. 2,266 2,266 2,266 2,266 2,266 2,266 R2 0.306 0.301 0.294 0.301 0.296 0.302 ∗∗∗ footnote: World Bank staff using RAIS. Note: Robust standard errors parentheses. p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. 16 As in the case of unconditional convergence results, the convergence rate is markedly higher for the 2002-10 period than for the 2010-18 period. This is particularly evident for agriculture and extractives (Table 4). Furthermore, Table 5 introduces additional state controls in the convergence regressions. We see that there is a strong and significant convergence in productivity across the states, including identical sizes of the coefficient estimates. It is found that productivity growth is higher in states with better human capital development (measured by years of schooling in the state’s population, stronger institutional capacity (approximated by per capita tax collection), and better infrastructure access (electricity and telephone connection). However, the share of the work- ing population does not have an impact. 5 Conclusion Testing for regional convergence in productivity across Brazil’s states between 2002 and 2018, the findings provide evidence supporting unconditional convergence in which states with lower levels of initial labor productivity experienced substantially faster labor productivity growth. Addition- ally, the convergence rate is faster for the period 2002-10 compared with the 2010-18 period and particularly strong in agriculture, extractives, and manufacturing. These findings on regional con- vergence are robust to controlling for state and industry fixed effects, states’ initial poverty rate, human capital, tax collection per capita, and infrastructure. Given the high disparity in labor productivity across Brazil’s states, such regional convergence has the potential to raise aggregate productivity and per capita income. However, the extent to which regional convergence promotes aggregate productivity critically depends on the reallocation of factor inputs and output shares toward sectors and industries with higher levels of productivity and faster speeds of convergence. 17 References Azzoni, C., & Castro, G. (2023). Economic Crises and Regional Disparities in Brazil in The XXI Century. TD NEREUS, 1-2023. Bakker, B., Ghazanchyan, M., Ho, A., & Nanda, V. (2020). The Lack of Convergence of Latin-America Compared with CESEE: Is Low Investment to Blame? IMF Working Paper(WP/20/98). Barrios, C., Flores, E., & Martínez, M. (2019). Convergence Clubs in Latin America. Applied Economics Letters, 26(1), 16-20. Beylunioglu, F., Yazgan, M., & T. (2020). Detecting Convergence Clubs. Macroeconomic Dynamics, 24(3), 629-669. Borsi, M., & Metiu, N. (2015). The Evolution of Economic Convergence in the European Union. Empirical Economics, 48, 657-681. Dieppe, A., & Matsuoka, H. (2021). Sectoral Decomposition of Convergence in Labor Productivity: A Re-examination from a New Dataset [World Bank Policy Research Working Paper # 9767]. Durlauf, S., & Johnson, P. (1995). Multiple Regimes and Cross-Country Growth Behaviour. Journal of Applied Econometrics, 10(4), 365-384. Fiaschi, D., Lavezzi, A., & Parenti, A. (2019). Deep and Proximate Determinants of the World Income Distribution. Review of Income and Wealth, 66(3), 677-710. Gennaioli, N., La Porta, R., De Silanes, F., & Shleifer, A. (2014). Growth in Regions. Journal of Economic Growth, 19, 259-309. Hall, R. E., & Jones, C. (1999). Why Do Some Countries Produce So Much More Output per Worker Than Others? Quarterly Journal of Economics, 114(1), 83–116. Herrendorf, B., Rogerson, R., & Valentinyi, A. (2022). New Evidence on Sectoral Labor Productivity: Implications for Industrialization and Development. NBER Working Paper # 29834. Iacovone, L., Sánchez-Bayardo, L., & Sharma, S. (2015). Regional Productivity Convergence in Peru. World Bank Policy Research Working Paper # 7749. Johnson, P., & Papageorgiou, C. (2020). What Remains of Cross-Country Convergence? Journal of Economic Literature, 58(1), 129–175. Jones, C. I. (2016). The Facts of Economic Growth. In J. B. Taylor & H. Uhlig (Eds.), Handbook of Macroeconomics (Vol. 2, p. 3–69). Elsevier. 18 Kinfemichael, B. (2019). The Rise of Services and Convergence in Labor Productivity Among Coun- tries. Applied Economics Letters, 26(21), 1749-1755. Kinfemichael, B., & Morshed, A. (2019a). Unconditional Convergence of Labor Productivity in the Service Sector. Journal of Macroeconomics, 59, 217-229. Kinfemichael, B., & Morshed, A. (2019b). Convergence of Labor Productivity Across the US States. Economic Modelling, 76, 270–280. Kollias, C., Tzeremes, P., & Tzeremes, N. (2021). Productivity Convergence and Divergence in Latin America, 1970-2014. Review of Economic Analysis, 13, 315-337. Phillips, P., & Sul, D. (2007). Transition Modeling and Econometric Convergence Tests. Econometrica, 75(6), 1771-1855. Rodrik, D. (2013). Unconditional Convergence in Manufacturing. Quarterly Journal of Economics, 125(1), 165-204. Tan, C. (2010). No One True Path: Uncovering the Interplay between Geography, Institutions, and Fractionalization in Economic Development. Journal of Applied Econometrics, 25(7), 1100-1127. Tebaldi, E. (2016). The Dynamics of Total Factor Productivity and Institutions. Journal of Economic Development, 41(4), 1-25. Wong, W. (2006). OECD Convergence: A Sectoral Decomposition Exercise. Economics Letters, 39(2), 210-214. Young, A., Higgins, M., & Levy, D. (2008). Sigma Convergence versus Beta Convergence: Evidence from U.S. County-Level Data. Journal of Money, Credit and Banking, 40(5), 1083-1093. 19 Appendix A. Additional results Figure A.1: Agriculture - Labor productivity growth, 2002-10 versus 2010-18 Source: World Bank staff calculations using RAIS. Figure A.2: Extractives - Labor productivity growth, 2002-10 versus 2010-18 Source: World Bank staff calculations using RAIS. 20 Figure A.3: Manufacturing - Labor productivity growth, 2002-10 versus 2010-18 Source: World Bank staff calculations using RAIS. Figure A.4: Services - Labor productivity growth, 2002-10 versus 2010-18 Source: World Bank staff calculations using RAIS. 21 Table A.1: Real total productivity and GDP growth by state and sector, 2003-18 Year Real total productivity growth Real GDP per capita growth Min Mean Max Std. dev. Min Mean Max Std. dev. 2003 -25.25 -1.49 28.76 5.28 -37.91 1.39 7.21 3.44 2004 -15.42 0.32 14.00 5.47 -25.88 3.45 12.49 3.02 2005 -16.53 3.82 19.71 6.45 -32.41 1.67 10.61 2.63 2006 -17.28 3.67 32.65 8.42 -32.58 1.57 16.12 5.71 2007 -16.51 -3.38 13.89 4.63 -25.42 7.12 19.92 3.69 2008 -29.02 1.05 135.60 5.35 -35.35 1.85 203.90 4.77 2009 -12.18 1.35 28.39 6.97 -7.82 -0.07 14.09 2.98 2010 -24.47 1.29 16.06 7.18 0.90 6.93 31.15 3.29 2011 -8.77 0.85 25.32 5.69 -2.45 3.77 41.71 2.08 2012 -4.31 2.11 22.73 3.45 -2.45 2.25 48.70 2.97 2013 -11.81 0.98 12.92 3.99 -7.24 -0.46 41.68 3.18 2014 -5.90 1.15 16.40 3.79 -2.85 1.28 23.30 2.08 2015 -9.35 0.60 26.63 4.33 -8.00 -3.87 10.33 1.55 2016 -24.21 -1.41 11.47 5.97 -8.53 -4.62 13.78 1.80 2017 -6.03 1.42 15.98 4.05 -6.68 1.54 10.83 2.88 2018 -12.74 1.03 37.87 7.79 -15.24 1.37 6.07 2.56 Source: World Bank staff calculations using RAIS. 22 Table A.2: Real total productivity growth by state and sector, 2003-18 States Agriculture Extractives Manufacturing Services Mean Std. dev. Mean Std. dev. Mean Std. dev. Mean Std. dev. Acre 3.97 18.91 2.88 14.07 2.20 6.62 0.94 6.65 Alagoas 1.03 6.77 16.12 47.49 1.93 5.98 1.11 4.93 Amapá 3.68 16.39 4.54 27.67 0.42 11.34 1.42 12.40 Amazonas 4.54 14.26 12.42 71.75 0.54 5.18 1.02 4.68 Bahia 2.47 5.29 3.48 36.65 0.59 6.10 0.76 3.92 Ceará 1.75 7.50 5.57 35.39 0.93 4.01 0.34 3.25 Distrito Federal 3.98 18.97 11.25 52.96 -0.28 8.56 0.77 7.58 Espírito Santo 3.78 10.26 3.56 10.13 0.90 9.49 0.91 8.27 Goiás 3.17 4.09 1.21 9.35 2.47 4.57 1.38 4.24 Maranhão 2.16 6.49 10.23 22.46 0.09 8.81 1.00 4.87 Mato Grosso 3.14 8.37 4.12 17.94 2.04 9.07 2.15 8.96 Mato Grosso do Sul 1.24 4.40 0.88 11.56 1.94 5.33 0.56 5.52 Minas Gerais 2.36 5.84 2.81 22.09 0.13 9.31 0.63 13.08 Paraná 2.57 4.59 3.29 20.49 0.32 2.91 0.31 3.77 Paraíba 1.77 6.56 2.33 9.20 1.43 4.49 1.59 5.28 Pará 1.74 7.26 0.44 15.55 0.80 7.07 0.65 6.21 Pernambuco 0.90 2.35 7.21 45.77 0.33 3.72 0.24 2.20 Piauí 3.57 7.61 4.26 28.08 1.06 3.91 0.85 3.33 Rio Grande do Norte 0.50 3.70 7.45 44.37 -0.40 11.97 0.94 3.22 Rio Grande do Sul 1.77 4.00 0.49 4.84 0.01 4.11 0.29 3.67 Rio de Janeiro 2.72 8.83 1.45 12.74 0.75 8.63 0.12 5.01 Rondônia 2.26 4.84 1.17 7.65 1.61 5.84 0.44 5.61 Roraima 3.21 14.87 5.11 28.74 1.82 8.27 0.40 7.60 Santa Catarina 0.58 4.18 0.39 5.76 0.12 3.41 0.19 3.74 Sergipe 3.26 8.98 0.76 13.13 0.54 10.20 1.87 5.01 São Paulo 3.10 14.26 7.43 27.64 0.03 6.03 0.19 5.45 Tocantins 2.89 5.13 3.59 12.46 2.86 6.11 2.06 8.42 All 2.52 9.52 4.49 29.01 0.86 7.16 0.78 6.36 Source: World Bank staff calculations using RAIS. 23 Table A.3: Labor productivity convergence regressions, Agriculture Growth of Labor Productivity 2002-2018 Total aggregation Disaggregated by all sectors (1) (2) (3) Initial labor productivity (log) -0.00486 -0.0311*** -0.0415*** (0.00977) (0.00758) (0.00994) Constant 0.0497 0.201*** No (0.0573) (0.0447) Industry FEs No No Yes State FEs No No Yes Obs. 27 81 81 R2 0.009 0.228 0.817 ∗∗∗ Robust standard errors parentheses. p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. Table A.4: Labor productivity convergence controlling for initial poverty level, Agriculture Growth of Labor Productivity 2002-2018 Total aggregation Disaggregated by all sectors (1) (2) Initial labor productivity (log) -0.0155 -0.0439*** (0.0102) (0.00763) Initial poverty rate -0.000558 -0.00130*** (0.000428) (0.000375) Constant 0.116* 0.286*** (0.0610) (0.0460) Industry FEs No No State FEs No No Obs. 27 81 R2 0.075 0.341 ∗∗∗ Robust standard errors parentheses. p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. 24 Table A.5: Labor Productivity Convergence Regressions, Extractives Growth of Labor Productivity 2002-2018 Total Aggregation Disaggregated by all sectors (1) (2) (3) Log initial labor productivity −0.0175∗∗∗ −0.0298∗∗∗ −0.0349∗∗∗ (0.00555) (0.00529) (0.00753) Constant 0.134∗∗∗ 0.210∗∗∗ - (0.0391) (0.0353) Industry FEs No No Yes State FEs No No Yes Obs. 27 112 112 R2 0.227 0.291 0.580 ∗∗∗ Robust standard errors parentheses. p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. Table A.6: Labor Productivity Convergence Controlling for Initial Poverty Level, Extractives Growth of Labor Productivity 2002-2018 Total Aggregation (1) (2) Initial labor productivity (log) −0.0145∗∗ −0.0311∗∗∗ (0.00574) (0.00557) Initial poverty rate 0.000968 −0.00108 (0.00109) (0.000947) Constant 0.106∗∗ 0.227∗∗∗ (0.0431) (0.0405) Industry FEs No No State FEs No No Obs. 27 112 R2 0.255 0.301 ∗∗∗ Robust standard errors in parentheses. p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. 25 Table A.7: Labor Productivity Convergence Regressions, Manufacturing Growth of Labor Productivity 2002-2018 Total Aggregation (1) (2) (3) Initial labor productivity (log) −0.0183∗∗∗ −0.0284∗∗∗ −0.0415∗∗∗ (0.00442) (0.00200) (0.00306) Constant 0.122∗∗∗ 0.186∗∗∗ No (0.0282) (0.0124) Industry FEs No No Yes State FEs No No Yes Obs. 27 625 625 R2 0.398 0.440 0.634 ∗∗∗ Robust standard errors parentheses. p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. Table A.8: Labor Productivity Convergence Controlling for Initial Poverty Level, Manufacturing Total Aggregation (1) (2) Initial labor productivity (log) −0.0223∗∗∗ −0.0307∗∗∗ (0.00448) (0.00218) Initial poverty rate −0.000537 −0.000916∗∗∗ (0.000323) (0.000180) Constant 0.152∗∗∗ 0.208∗∗∗ (0.0295) (0.0143) Industry FEs No No State FEs No No Obs. 27 625 R2 0.448 0.467 ∗∗∗ Robust standard errors parentheses. p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. 26 Table A.9: Labor Productivity Convergence Regressions, Services Total Aggregation (1) (2) (3) Initial labor productivity (log) −0.0122∗∗ −0.0203∗∗∗ −0.0448∗∗∗ (0.00509) (0.00115) (0.00236) Constant 0.0880∗∗ 0.137∗∗∗ No (0.0338) (0.00729) Industry FEs No No Yes State FEs No No Yes Obs. 27 1,448 1,448 R2 0.193 0.255 0.706 ∗∗∗ Robust standard errors parentheses. p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. Table A.10: Labor Productivity Convergence Controlling for Initial Poverty Level, Services Total Aggregation (1) (2) Initial labor productivity (log) −0.00777 −0.0212∗∗∗ (0.00915) (0.00119) Initial poverty rate 0.000296 −0.000489∗∗∗ (0.000466) (0.000125) Constant 0.0558 0.147∗∗∗ (0.0645) (0.00781) Industry FEs No No State FEs No No Obs. 27 1,448 R2 0.211 0.264 ∗∗∗ Robust standard errors parentheses. p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. 27 Table A.11: Aggregate Labor Productivity Convergence Regressions Total Aggregation (1) (2) (3) (4) (5) Initial labor productivity (log) −0.0115∗∗ −0.0137∗∗∗ −0.0191∗∗∗ −0.0229∗∗∗ −0.0390∗∗∗ (0.00425) (0.00257) (0.00413) (0.00110) (0.00192) Constant 0.0835∗∗∗ 0.100∗∗∗ No 0.153∗∗∗ No (0.0282) (0.0163) (0.00689) Industry FEs No No Yes No Yes State FEs No No Yes No Yes Obs. 27 108 108 2,266 2,266 R2 0.215 0.222 0.734 0.294 0.624 ∗∗∗ Robust standard errors parentheses. p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. Table A.12: Aggregate Labor Productivity Convergence Controlling for Initial Poverty Level Total Aggregation (1) (2) (3) Initial labor productivity (log) −0.00773 −0.0134∗∗∗ −0.0240∗∗∗ (0.00800) (0.00259) (0.00115) Initial poverty rate 0.000270 9.86e − 05 −0.000614∗∗∗ (0.000434) (0.000315) (0.000108) Constant 0.0561 0.0975∗∗∗ 0.165∗∗∗ (0.0565) (0.0170) (0.00754) Industry FEs No No No State FEs No No No Obs. 27 108 2,266 R2 0.232 0.223 0.306 ∗∗∗ Robust standard errors parentheses. p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. 28 B. List of states and sector classification Table B.1: List of States and Their Abbreviations State Abbreviation Acre AC Alagoas AL Amazonas AM Amapá AP Bahia BA Ceará CE Distrito Federal DF Espírito Santo ES Goiás GO Maranhão MA Minas Gerais MG Mato Grosso do Sul MS Mato Grosso MT Pará PA Paraíba PB Pernambuco PE Piauí PI Paraná PR Rio de Janeiro RJ Rio Grande do Norte RN Rondônia RO Roraima RR Rio Grande do Sul RS Santa Catarina SC Sergipe SE São Paulo SP Tocantins TO 29 Table B.2: List of CNAE2 Sectors and Subsectors Sector Subsector CNAE 2-digit Agriculture Agriculture, livestock, forestry production, fishing, and aquaculture 01-03 Extractives Industry Extraction industries 05-09 Manufacturing Industry Manufacturing industries 10-33 Services Electricity and gas 35 Services Water, sewage, waste management, and decontamination activities 36-39 Services Construction 41-43 Services Trade; repair of motor vehicles and motorcycles 45-47 Services Transportation, storage, and postal services 49-53 Services Accommodation and food services 55-56 Services Information and communication 58-63 Services Financial activities, insurance, and related services 64-66 Services Real estate activities 68 Services Professional, scientific, and technical activities 69-75 Services Administrative and support services 77-82 Services Public administration, defense, and social security 84 Services Education 85 Services Human health and social services 86-88 Services Arts, culture, sports, and recreation 90-93 Services Other service activities 94-96 Services Domestic services 97 Services International organizations and other extraterritorial institutions 99 30