Pro-Poor Growth in Brazilian States and Groups of Municipalities during the 90 th Decade

Similar documents
Growth and Poverty Reduction: An Empirical Analysis Nanak Kakwani

Poverty Reduction and Economic Growth: The Asian Experience Peter Warr

Changes in Brazilian Rural Poverty and Inequality From 1991 to 2000: The Role of Migration

Poverty, Income Inequality, and Growth in Pakistan: A Pooled Regression Analysis

Patterns and determinants of wage inequality in the Brazilian territory 1

Economic Growth and Poverty Alleviation in Russia: Should We Take Inequality into Consideration?

Interrelationship between Growth, Inequality, and Poverty: The Asian Experience

When Job Earnings Are behind Poverty Reduction

Economic Growth and Poverty Reduction: Lessons from the Malaysian Experience

China s (Uneven) Progress Against Poverty. Martin Ravallion and Shaohua Chen Development Research Group, World Bank

Poverty in Uruguay ( )

Asian Development Bank Institute. ADBI Working Paper Series. Income Distributions, Inequality, and Poverty in Asia,

Income Distributions, Inequality, and Poverty in Asia,

PERSISTENT POVERTY AND EXCESS INEQUALITY: LATIN AMERICA,

The Role of Labor Market in Explaining Growth and Inequality: The Philippines Case. Hyun H. Son

CHAPTER 2 LITERATURE REVIEWS

Poverty, growth and inequality

Operationalising Pro- Poor Growth. A Country Case Study on Brazil

Education, cost of living and regional wage inequality in Brazil

Migration in Brazil in the 1990s 1

Spatial Inequality in Cameroon during the Period

A poverty-inequality trade off?

ESTIMATING INCOME INEQUALITY IN PAKISTAN: HIES TO AHMED RAZA CHEEMA AND MAQBOOL H. SIAL 26

Travel Time Use Over Five Decades

The labor market in Brazil,

Inflation and Income Inequality: Is Food Inflation Different?

How much of Brazilian Inequality can we explain?

Demographic Changes and Economic Growth: Empirical Evidence from Asia

ERD. Working Paper. No. Interrelationship between Growth, Inequality, and Poverty: The Asian Experience. Hyun H. Son ECONOMICS AND RESEARCH DEPARTMENT

Inequality in Indonesia: Trends, drivers, policies

REMITTANCES, POVERTY AND INEQUALITY

Impacts of Economic Integration on Living Standards and Poverty Reduction of Rural Households

Migration and Tourism Flows to New Zealand

Trade led Growth in Times of Crisis Asia Pacific Trade Economists Conference 2 3 November 2009, Bangkok. Session 10

HOW ECONOMIES GROW AND DEVELOP Macroeconomics In Context (Goodwin, et al.)

Poverty and Inequality

Ghana Lower-middle income Sub-Saharan Africa (developing only) Source: World Development Indicators (WDI) database.

POLICY OPTIONS AND CHALLENGES FOR DEVELOPING ASIA PERSPECTIVES FROM THE IMF AND ASIA APRIL 19-20, 2007 TOKYO

UGANDA S PROGRESS TOWARDS POVERTY REDUCTION DURING THE LAST DECADE 2002/3-2012/13: IS THE GAP BETWEEN LEADING AND LAGGING AREAS WIDENING OR NARROWING?

Handout 1: Empirics of Economic Growth

Understanding Subjective Well-Being across Countries: Economic, Cultural and Institutional Factors

Human Capital and Income Inequality: New Facts and Some Explanations

Dynamics of spatial inequality in the Brazilian labor market between 1980 and 2000: a fixed effect approach 1

Columbia University. Department of Economics Discussion Paper Series

Poverty Profile. Executive Summary. Kingdom of Thailand

Research Report. How Does Trade Liberalization Affect Racial and Gender Identity in Employment? Evidence from PostApartheid South Africa

Inequality is Bad for the Poor. Martin Ravallion * Development Research Group, World Bank 1818 H Street NW, Washington DC

Pro-Poor Growth and the Poorest

Latin American Immigration in the United States: Is There Wage Assimilation Across the Wage Distribution?

Inequality in Brazil

Does Inequality Matter for Poverty Reduction? Evidence from Pakistan s Poverty Trends

Economic Growth, Income Inequality, and Poverty Reduction in People s Republic of China BO Q. LIN

Household Income inequality in Ghana: a decomposition analysis

The Impact of the Interaction between Economic Growth and Democracy on Human Development: Cross-National Analysis

Human Capital and the Recent Decline of Earnings Inequality in Brazil *

INCLUSIVE GROWTH AND POLICIES: THE ASIAN EXPERIENCE. Thangavel Palanivel Chief Economist for Asia-Pacific UNDP, New York

A Comparative Perspective on Poverty Reduction in Brazil, China, and India

The interaction effect of economic freedom and democracy on corruption: A panel cross-country analysis

Emerging Market Consumers: A comparative study of Latin America and Asia-Pacific

Remittances and Poverty. in Guatemala* Richard H. Adams, Jr. Development Research Group (DECRG) MSN MC World Bank.

Application of PPP exchange rates for the measurement and analysis of regional and global inequality and poverty

High Technology Agglomeration and Gender Inequalities

Inclusion and Gender Equality in China

Real Convergence in the European Union

Poverty and Inequality Changes in Turkey ( )

Gender preference and age at arrival among Asian immigrant women to the US

Growth, Inequality, and Poverty: An Introduction Nanak Kakwani, Brahm Prakash, and Hyun Son

Ethnic Minorities in Northern Mountains of Vietnam: Poverty, Income and Assets

Wage Structure and Gender Earnings Differentials in China and. India*

Accounting for the role of occupational change on earnings in Europe and Central Asia Maurizio Bussolo, Iván Torre and Hernan Winkler (World Bank)

An Empirical Analysis of Pakistan s Bilateral Trade: A Gravity Model Approach

The Social Policy and Development Centre (SPDC)

Poverty of Ethnic Minorities in the Poorest Areas of Vietnam

Secondary Towns and Poverty Reduction: Refocusing the Urbanization Agenda

Unequal Provinces But Equal Families? An Analysis of Inequality and Migration in Thailand 1

Online Appendices for Moving to Opportunity

Gender and Ethnicity in LAC Countries: The case of Bolivia and Guatemala

Volume 35, Issue 1. An examination of the effect of immigration on income inequality: A Gini index approach

Does the G7/G8 Promote Trade? Volker Nitsch Freie Universität Berlin

There is a seemingly widespread view that inequality should not be a concern

New Evidence on the Urbanization of Global Poverty

Cornell University ILR School. Chen Zongsheng Nankai University. Wu Ting Party School of Communist Party of China

for Latin America (12 countries)

Patterns of Inequality in India

Is Corruption Anti Labor?

Tourist Arrivals in the APEC Region: Determinants and Inclusive Impacts By Emmanuel A. San Andres 1

Tiago Freire 1 Faculty of Business, Government and Law, University of Canberra Roberts Capital Advisors, LLC.

GENDER EQUALITY IN THE LABOUR MARKET AND FOREIGN DIRECT INVESTMENT

The Effect of International Trade on Wages of Skilled and Unskilled Workers: Evidence from Brazil

Accounting for Heterogeneity in Growth Incidence in Cameroon

Violent Conflict and Inequality

Is inequality an unavoidable by-product of skill-biased technical change? No, not necessarily!

Pro-poor Growth and Policies: The Asian Experience

Inequality in China: Selected Literature

A Structural Analysis of Growth and Poverty in the Short-Term

Southern Africa Labour and Development Research Unit

vi. rising InequalIty with high growth and falling Poverty

Asian Development Review

Skill Wage Gap in Brazil:

Does Learning to Add up Add up? Lant Pritchett Presentation to Growth Commission October 19, 2007

Transcription:

Pro-Poor Growth in Brazilian States and Groups of Municipalities during the 90 th Decade Lizia de Figueiredo 1 Raimundo de Sousa Leal Filho 2 Abstract: The pro-poor growth literature about Brazil is related to the analysis of poverty in Brazilian States. We propose, instead, to deal with groups of municipalities for which we have compatible data. We estimate the elasticity of poverty with respect to income and to an inequality measure, and differences in the response of poverty for units with better provision of human capital and other variables. We also test for the presence of regional effects, and investigate if the decrease in poverty was influenced by initial conditions. Our main conclusion was that groups of municipalities with more education were the ones with lower poverty rates at the beginning, and also more successful in alleviating poverty during the 90 s. Resumo: A literatura sobre crescimento pró-pobre no Brasil esteve focada na análise da pobreza entre os Estados da Federação. Pretendemos contribuir com esta agenda de pesquisa trabalhando informações organizadas por grupos de municípios para os quais dispomos de dados compatíveis. Estimamos a elasticidade da pobreza com relação à renda regional e a uma medida de desigualdade, e as diferenças nas respostas da pobreza para grupos com melhor provisão inicial de capital humano e outras variáveis. Também testamos a presença de efeitos regionais, e investigamos se a variação da pobreza foi influenciada pelas condições iniciais. Nossa principal conclusão foi que os grupos de municípios com maior estoque inicial de educação foram os que apresentaram menores taxas de pobreza no início do período, e também foram melhor sucedidos na redução da pobreza durante o período analisado. Key-Words: Pro-poor growth, poverty and inequality, human capital. Palavras-Chave: Crescimento pró-pobre, pobreza e desigualdade, capital humano. JEL Code: I32, O49 1 Corresponding author: Lizia@cedeplar.ufmg.br 2 Raimundo.Sousa@fjp.mg.gov.br 1

I. Introduction The concern about the effects of economic growth in alleviating poverty has been on the rise since the 90s, probably encouraged by the Millennium Goal of poverty reduction. The pro-poor growth literature has especially tried to define pro-poor growth and to identify periods where economic growth resulted in a decrease of poverty. Pro-poor growth has been either identified with a larger change in income of the lower classes of income distribution (Kakwani & Pernia, 2000) or with the decrease in a poverty index due to an increase in income (Ravallion, 2004). This paper is developed based on the latter definition. Several methodologies have been proposed to identify pro-poor growth, as the index develop by Pernia and Kakwani (2000) - based on the income-elasticity of poverty and on the elasticity of poverty with respect to the Gini coefficient, and the growth incidence curve of Ravallion & Chen (2003). Using their index, Kakwani & Pernia (2000) found that growth has been pro-poor in Korea, while it has not always favored the lower deciles of income in Lao PDR and Thailand. Growth was more pro-poor in rural than in urban areas, and the economic performance was worse for the poor during economic crises, in these countries. Ravallion & Chen (2003) applied their methodology to the economic performance of China in the 90s, where they found a pro-poor pattern of growth. Besides assessing the information provided by alternative poverty measures available, the literature has also been rich in developing alternative methodologies for estimating the elasticity of poverty with respect to income and inequality and for decomposing the poverty index within its components (income, inequality and poverty line). Zhang & Wan (2005) use a semi-parametric approach for estimating the income distribution and also introduce the Shapley value decomposition to estimate the relative contributions of economic growth, inequality and the poverty line to poverty reduction, founding, for urban provinces of China, that economic growth was the major component. Income distribution was also considered very important. Kraay (2004) analyze how poverty change has reacted to differences in average income growth, in income elasticity of poverty, and in the effects of growth trough the income 2

distribution. He found that, for a sample of developing countries in the 80 s and 90 s, the main source of poverty alleviation was average economic growth. Secondly comes the the poverty-reducing pattern of growth in relative income. Recently, the discussion is shifting towards the research of the determinants of pro-poor growth. Ravallion (2004) states that pro-poor growth is a function of initial conditions, especially the initial income inequality. The change in income inequality and sectoral and regional effects should also be investigated. Ravallion & Datt (1999) found that there were no differences in the magnitudes of the response of poverty to an increase in agricultural income and to an increase in government expenditures, among the 15 states of India from 1960 to 1994. On the other hand, the elasticity of poverty with respect to the non-agricultural output was different among the states, being higher in the ones with initial higher literacy rates, initial higher farm productivity and accordingly with the rural living standards. The pro-poor growth literature about Brazil is in its beginning, being related mainly to the analysis of poverty in Brazilian States, which form a sample with reduced number of observations. We propose, instead, to deal with an aggregation of municipalities called micro regions, for which we have compatible data from the 1991 and 2000 demographic censuses. We wish to estimate the elasticity of poverty with respect to income and to an inequality measure in 1991, and also to identify if, in this period, the response of poverty was different for units with better provision of human capital, different composition of economic activities and different degrees of congestion. We also test for the presence of regional effects. For the period 1991-2000, we try to investigate if the decrease in poverty was influenced by different initial conditions, and if their impact was direct on a poverty measure, or indirect, when affecting the income elasticity of poverty. Section II reviews the Brazilian literature in pro-poor growth; section III estimates the model for 1991, while section IV shows the results for the decade. Section V uses kernel estimation to find if units that were poor in 1991 have stronger probability of being poor in 2000, and also investigate the shape of the distribution of income and of poverty. 3

II. Pro-Poor Growth Literature in Brazil For Brazil, Paes de Barros, Henriques & Mendonça (2000) estimate the proportion of the poverty change due to the change in inequality, from each year since 1977 to 1997. The relative importance of economic growth is estimated through the residual of the decomposition. It was found that economic growth was the driving force of poverty change, although the inequality component had increased in importance by the end of the eighties, when inequality rose in Brazil. Barreto, Marinho & Soares (2003) show that there was no significant relationship between per capita income and poverty, from 1970 to 1999, considering the Brazilian States. Income growth and inequality are the main determinants of poverty. In order to quantify the relative strength of these forces over poverty, Barreto, Marinho & Soares (2003) estimated the elasticities of poverty with respect to income and to the gini coefficients, for a sample of 26 Brazilian States, in the years 1985 to 1999 3. Their results indicate that an increase in income has different impacts over the Brazilian States, being smaller in the poorest ones, and lower than one for the States of the Northeast. They also find that this elasticity has increased in the period for the majority of States, with the exception of the Northern States. Hoffmann (2004) found a similar pattern for the income-elasticity of poverty, although estimating this variable using the parameters of a log-normal specification for the distribution of income. The results of Barreto, Marinho & Soares (2003) are always higher in absolute terms. He also estimated the elasticity of poverty with respect to an inequality measure that increases, as expected, with the income level. The higher levels of income elasticity are found in Roraima, São Paulo, Rio de Janeiro and Santa Catarina, while the lower are found in Maranhão, Piauí and Ceará. Rio de Janeiro, São Paulo and Distrito Federal have the higher levels in the elasticity of poverty with respect to inequality, while Maranhão and Piauí have the lower ones in 1999. Both elasticities were estimated for 1999, 2001 and 2002. 3 2 ln Pit = α i + β i ln yit + β 2 ln Git + β 3 ln yit + β 4[ln yit ln Git ] + ηit. The elasticity of poverty ε it = β1 + β3 ln yit + β4 lngit, which is shown for 1985, 1992 with respect to income would then be and 1999. 4

In 1999, for Brazil, a 1% decrease in the Gini coefficient lessened the P 0 index in 1.81%, and a 1% increase in per capita average income diminished the proportion of poor in 0.84%, accordingly to Hoffmann (2004). Barreto, Marinho & Soares (2003) followed Datt & Ravallion (1992) in decomposing the poverty measure amongst its income and inequality components, and their results indicate that both elements had similar shares in explaining the evolution of poverty. The income share was found to be higher in the majority of the cases where poverty was abridged. In the Northern States, poverty was explained mainly by the inequality component (56.48%). Leal Filho (2004) found that a 1% increase in Brazilian States per capita GDP shortened extreme poverty by 1.20 %, in the period 1991 to 2000. Following Ravallion (2001), he estimated the income elasticity of poverty correct by an inequality measure, founding higher and significant values for this variable. A 1% increase in per capita GDP in a state with a Gini coefficient of 60% decreased poverty in 1.2%, while the same income increase in a state with Gini coefficient of 40% decreased poverty in 1.8%. The literature reviewed above tried to quantify the impact of income growth and inequality over some poverty measure, but did not try to identify, amongst the causes of economic growth, which ones would have been more significant for poverty reduction. Tocheto et al (2004) 4, using Ravallion & Datt (2000) approach, find that the income elasticity of poverty of the Brazilian states was higher where local government expenditures with education were lower, and where inflation was higher. More interestingly, they found that income elasticity varies with regional effects in panel estimation with 30 states, from 1985 to 1999 (excluding the years 91 and 94). The income elasticity of the states were usually positive, indicating that in the majority of the states income growth was not pro-poor (the authors use Kakwani definition of pro- 4 The estimated models are: 1) ln Pit = β 1 ln PIBISit + β2 ln AGRit + β3 lngovit + γinfit + δ t + ηi + υit ; and ln P = β [ln PIBIS η ] + β ln AGR + β3 lngov + γinf + δ t + η + υ 2) it i 1 it i 2 it it it i it. P = poverty measure; PIBIS = non-agricultural output; AGR = agricultural per capita income; GOV = state government expenditures with education, INF = inflation rate of the metropolitan areas. These models were estimated with fixed and random effects, respectively, for the Brazilian States in the period 1985 to 1999 (excluding 1991 and 1994). 5

poor growth), with the exception of Ceará and Distrito Federal. The levels of this elasticity vary from 5.94, in Bahia, to 0,42, in Rio de Janeiro. The authors use the growth curve of Son to observe the impact of economic growth in the different deciles of income, concluding that during the 80 s growth was usually not pro-poor in the majority of Brazilian States, contrary to what happens during the 90 s. During downturn economic periods, growth was usually not pro-poor. III. Determinants of Pro-Poor Growth: Groups of Municipalities in Brazil The results presented in the previous section should be viewed with caution, due to a small number of observations in the sample. Trying to avoid the problem of micronumerosity, we decided to discuss the determinants of poverty amongst groups of municipalities called micro regions in Brazil. It is also interesting to study this larger sample because the determinants of economic growth and of poverty may change accordingly to local characteristics, and also because it allows us to take account of within-region differences. Table 1 (see Appendix) shows the OLS results for a set of equations that try to explore the determinants of poverty amongst 558 Brazilian groups of municipalities, for which the data was obtained from IPEADATA (www.ipeadata.org.br). The dependent variable is the natural logarithm of the poverty index (the proportion of poor people over the entire population). The independent variables are all measured in 1991 levels. Table 1, column 1, shows the results of our basic equation, which simply aims to calculate the elasticity of poverty to its two main determinants: income and inequality. A 1% increase in average income (relative to the minimum wage) lessens the poverty index by 0.93%, while a decrease in 1% in the Theil index reduces poverty by 2%. These results are very similar to the ones obtained by Hoffmann (2004), for the Brazilian States in 1999, 2001 and 2002. 6

Column 2 shows how the results change when, following Ravallion (2001), we correct the income by the Theil coefficient, assuming that inequality affects poverty through the income elasticity of poverty 5. The estimated income elasticity of poverty is 1.96, implying that a 1% increase in the average income (with respect to the minimum age) decreases poverty by 1.96%, much higher than the ones estimated for the states, using the same correction, by Leal Filho (2004), for the change in poverty during the 90 s decade. It must be also observed that the elasticity of poverty with respect to the Theil coefficient has increased to 2.44%, as well as the fit of the econometric model, in which 91% of the variations in poverty are explained by these two variables. The results above indicate that, in the 90's decade, the decrease in inequality played a relatively stronger role to explain poverty reduction, compared to an increase in income. The raise in the relative importance of inequality was also established in the literature that dealt with the sample of Brazilian states. This is an interesting result that contrasts with the previous results for Brazil as a whole. Paes de Barros, Henriques & Mendonça (2000) indicate that economic growth was the major source of poverty reduction from 1977 to 1997. The explanation to this contrast resides in the slow economic growth during the period, combined with a decrease in inequality in Brazil. The Brazilian GDP grew 3.5% per annum from 1980 to 2000, while per capita GPD grew only 0.25% p.a. during the same period (or 1.05% p.a., considering only the 90 s decade). From 1980 to 2000, the Theil coefficient dropped 1%, due to the results during 90 s (in the 80 s, inequality increased 1%). Poverty rates showed a 3% decrease during the 90 s. Columns (3) to (6) attempt to investigate how poverty reacts to different local characteristics. Column (3) wishes to test if regions with higher human capital have lower poverty rates, using two human capital proxies: average years of schooling and life expectation. 5 The econometric model is: ln P0it = α + β1 [(1 THEILit ) lnym it ] + β2theilit + υit, where P0 is the poverty index and YM is the average income relative to the minimum wage. 7

Areas with higher years of schooling show lower poverty rates, as the negative and significant coefficient of this variable indicates. One year of extra schooling is associated with a poverty index 0.06% smaller. The coefficient of life expectation is not significant, what can be due to multicolinearity problems. Column (4) shows that the poverty rates are not different in regions with different urbanization rates, while groups of municipalities with higher percentage of population occupied in services or in the industrial sector show lower poverty rates, since the coefficient of these variables are negative and significant. Together with an insignificant coefficient on the commerce variable, the results indicate that poverty and agriculture are strongly associated. Introducing other control variables on the basic equation (2) has stronger impact over the income elasticity of poverty than on the poverty elasticity with respect to the inequality measure, what brought us to investigate the indirect impacts of human capital and of sectoral composition through the income-elasticity of poverty. Due to this concern, we interacted all our control variables with the average income. A significant coefficient for the interaction term would be evidence in favor of the indirect effect of the variable. Table 1, column (5) shows the results of the econometric model with the interaction terms. The F-test for the marginal contribution of the interaction terms does not allow us to exclude them from the econometric model and its fit has increased to 98%. Controlling for the interaction terms, the coefficients of years of schooling and of the industrial sector are no longer significant, while the negative coefficient of the urbanization rates now imply that more urbanized areas show smaller poverty rates. The negative coefficient of urbanization and the negative and significant coefficient of its interaction term allow us to say that a more urbanized area not only has small poverty rates, but that an increase in its income is strongly translated into a decrease in poverty, although the magnitudes of these coefficients are low. Although there are no more direct effects of human capital, areas better endowed with human capital also show higher income elasticity (in absolute terms). 8

Agricultural areas are the ones with lower levels of income elasticity, since the coefficients of INTCOM, INTSER and INTIND are negative and significant. The results of this exercise say that differences in urbanization, in sectoral composition and in human capital are not as important per se (although the F-test does not allow us to exclude the variables not interacted from the econometric model), but through different ways regions can respond to an increase in income. The effects of the inclusion of other controls over the levels of poverty elasticity are strong. The income elasticity drops in absolute terms from 1.99 to 1.10. The elasticity of poverty to the Theil coefficient drops from 2.44 (Equation (2)) to 1.6, implying that part of the importance of inequality was mistakenly capturing a correlation of the Theil index with the other variables. Even though, this clearly shows that inequality was an important determinant of poverty in 1991. It is clear that regions with higher human capital and higher urbanization rates had a higher income elasticity of poverty, the contrary being observed in predominantly agricultural regions. The direct effect of the local output s sectoral composition over poverty reduction could be observed, although it is not possible to identify one single sector of activity to be more pro-poor, since the significances of their coefficients, individually, were not robust. One exception was the services: areas with a large fraction of the service sector were significantly associated with smaller poverty rates in all specifications for this model. Finally, Table 1, Column (6) shows the results after introducing regional dummies. Although some of them are significant, the F-test for joint significance shows an extremely low marginal effect for these variables. Only groups of municipalities located in the Northeast of Brazil shows smaller controlled poverty rates, compared to the ones located in the Southeast. IV. Initial Conditions and Poverty Reduction According to Ravallion (2004), differences in initial conditions are probably the main determinants of pro-poor-growth. Therefore, we will investigate which economic variables, characterizing initial conditions of a spatial economic unit, also affect the change in poverty. Empirically, our strategy will resemble the economic growth 9

literature, since the independent variables will be set at values prevailing in the beginning of the period whereas the dependent variable will be the proportional change in some poverty index. As with the economic growth literature, this discussion will also bring to mind the idea of club convergence, which says that economies with initial different conditions, even if similar in their structural parameters, will have different steady states of per capita income. In this case, the question is if economies with initial different conditions will have not only different average steady states income values, but also different long run proportions of poor. Changes in poverty can be due to changes in average income, changes in the incomeelasticity of poverty and changes in inequality, according to Kraay (2001). We can interpret our results as testing if the chosen variables affect poverty though the first two channels. All groups of municipalities have shown a decrease in poverty during the 90 s, the reason why we found it easier to define our dependent variable as the absolute change in poverty from 1991 to 2000. Table 2 shows the results for this exercise, where initial conditions are included as independent variables. Column (1) shows the impact of the initial income and of initial inequality in the reduction of poverty. Only 8% of the differences in the rate of poverty reduction are related to these variables, which show, nevertheless, positive and significant coefficients, implying that poverty reduction was stronger in originally richer and unequal areas. The initial income level was not corrected by the Theil coefficient to keep the comparison to the usual economic growth regressions. Column (2) shows the results for the model when initial sectoral composition, urbanization and human capital endowments are included as independent variables. The model now explains 26% of the diversity in poverty reduction. Groups of municipalities with initial higher levels of human capital show higher poverty reduction, as the positive and significant coefficient of YEARS implies. Poverty reduction was also stronger in less urbanized areas, since the coefficient of URB was also negative and significant. On the other hand, stronger poverty reduction was not a 10

feature of areas with larger services sector, since the coefficient of SER is negative and significant. Relatively to agriculture, industrial areas did not perform differently in terms of poverty reduction, since the coefficient of IND is not significant. The same occurred in groups of municipalities with a higher initial share of their population working in the commercial sector In this specification, the coefficients of initial income and of initial inequality were not significant. Column (3) shows the results with the inclusion of the interaction terms. Richer regions were the ones with stronger poverty reduction, as implied by the positive and significant coefficient of initial per capita income. The following initial conditions show positive correlations with poverty reduction: years of schooling and the share of the labor force in the commercial sector, since the coefficient of these variables were positive and significant. On the other hand, some initial conditions had a negative impact on poverty reduction, as a large share of people working in services and higher rates of urbanization. The decline in poverty was also smaller in groups of municipalities with higher urbanization rates and with higher service sector due to a smaller income elasticity of poverty. Controlling for the interaction effects, less unequal areas are the ones which stronger gains in poverty reduction, since the coefficient of THEIL is negative and significant. The coefficient of COM turns to be positive and significant. Column (4) includes regional dummies, which are jointly significant. This final specification explains 55% of the differences in poverty reduction in our sample. The main results in columns (3) are robust to the inclusion of these dummies, except for the coefficient of the inequality variable, which is now not significant. There are still regional characteristics that our model is not capturing, which makes poverty reduction in groups of municipalities located in the SE (Southeast of Brazil) stronger than in the CO (Center-West), NE (Northeast) and N (North). Concluding, our results allow us to say that Brazilian groups of municipalities with a higher drop in their poverty rates during the 90 s were the ones better provided with 11

education, with a larger fraction of their population living in rural areas and working in agricultural or commercial production. The income elasticity of poverty was also higher the larger a fraction of the population living in rural areas and working in the services sector. Richer areas seem to be the ones with larger poverty reduction. V. Evidence of Poverty Traps in Brazilian Groups of Municipalities during the Nineties In this section, we will observe the shape of the distribution of poverty in 1991 and in 2000, and estimate the probability of persistence in this distribution. Following Quah (1997), we will also estimate the per capita income distribution in both periods and its transitions probabilities. Figure 1(a) shows the distribution of the poverty rates in 1991. It is possible to see that there are two modes in the distribution of the poverty rates, one around 50% and another with higher poverty rates, around 80%. Figure 1(b) shows the distribution of poverty rates in 2000, where it is shown that the distribution is shifting to the left, concentrating in lower values of poverty rates, and also displaying a decrease in the value of the modes. It is striking to see how clearly a significant number of groups of municipalities in Brazil was left behind in this general trend of diminishing poverty over the nineties. This observation motivates us to search for characteristics related with possible poverty traps in Brazil, and to set an agenda for testing candidate hypotheses to explain this feature. Figures 2 and 3 are useful to discuss the mobility of our groups of municipalities with respect to their poverty rates. Observing the mass of the conditional distribution for 2000 value over 1991 ones, it is possible to undertake a visual check on a broad result: Brazilian groups of municipalities tend to show smaller poverty rates in 2000 compared to 1991. 12

Although this widespread decline in poverty rates, there is persistence in the distribution in the sense that groups of municipalities with higher poverty rates in 1991 are the ones with higher poverty rates in 2000. Figure 4 shows the distribution of per capita income for each group of municipalities relative to the Brazilian average. It is shown that the distribution also has two modes, with the taller one around 0.008% of the Brazilian income and the richer units showing a peak around 0.02%. It is clear in Figure 4(b) that, in 2000, there was a larger concentration of regions below the second mode. The average values of the second mode is higher than in 1991, while the average value of the first mode is similar to 1991, implying that economic growth in the decade did not help to improve the living conditions of the poorest regions of Brazil. Finally, Figure 5 shows the kernel estimation of the Markov probabilities of transition for the average per capita income, while Figure 6 shows its contour plot. We can observe clearly that the distribution has two modes. Richer groups of municipalities, whose mode is almost on the diagonal, still show convergence, since the distribution cuts the diagonal from the left to the right. The poorest mode is located below the diagonal, implying that the poorest regions among the poor grew more than the others. The twin-peaked shape of the distribution does not seem to vanish. There is also no mobility between these two bases of attraction. The estimations above lead us to conclude that, during the 90 s, economic growth had different impacts depending on the initial income of the spatial unit. Middle-income groups grew more than richer ones, decreasing inequality in per capita income between these groups. For the poorer units in 1991, there was also a decrease in inequality within the group, but the distance between them and the others persisted. 13

Poverty has decreased among the units, but there is a significant group of regions showing a stronger decrease in the average value of poverty. The results of the previous sections informed us that poverty has decreased more in richer areas (controlling for other variables). We can conclude that although the majority of the regions have an increase in income in the period and a correspondent decrease in poverty, the poorest groups of municipalities in Brazil tend to continue poor. A large share of its population, consequently, will remain living below the poverty line. Conclusion Poverty alleviation was a widespread feature during the 90 s decade in Brazil. In 1991, groups of municipalities with higher education, higher urbanization rates, higher inequality and lower agricultural sector were the ones with higher income elasticity of poverty. A larger share of the service sector directly favored a decrease in poverty, either through associated higher economic growth or through associated lower inequality. Things have changed along the decade. Agricultural and rural units, in the beginning of the period, were exactly the ones that managed to decrease sharply their poverty rates. What has not changed was that more educated areas were not only the ones with lower poverty rates in 1991, but also more successful in alleviating poverty during the 90 s. It is also possible to say that the decrease of inequality during the 90 s was a major source of poverty change in this decade. There remains, nevertheless, a major concern for Brazilian policy makers: the very poor municipalities are not managing to escape poverty, either in the sense of persistent low average levels of per capita income, or directly with extremely high and persistent rates of poverty. 14

Tables Table 1 - Determinants of Poverty in Brazilian Groups of Municipalities - 1991 (Dependent Variable: LNP0) 1 2 3 4 5 6 LNY -0.926*** LNYCORR -1.959*** -1.700*** -1.593*** -1.039*** -1.124*** THEIL 1.995*** 2.437*** 2.395*** 2.073*** 1.545*** 1.592*** YEARS -0.064*** -0.028* 0.001-0.002 LE 0.003 URB -0.0004-0.001*** -0.0003 SER -0.035* -0.013** -0.012* IND -0.013** -0.003-0.004* COM -0.014-0.006-0.016 LNYIYEARS -0.034* -0.029* LNYURB -0.002-0.002* LNYSER -0.040** -0.039** LNYIND -0.045*** -0.041*** LNYCOM -0.050** -0.050** LNYTHEIL CO -0.015 NE -0.040** S 0.033** N 0.023** R 2 0.89 0.91 0.91 0.94 0.97 0.98 Source: Own Calculations from Raw Data Obtained at www.ipeadata.gov.br 15

Table 2 - Determinants of Poverty Change - 1991-2000 (Dependent Variable: Change in LNP0 (%)) 1 2 3 4 LNY 0.062*** -0.051 0.340*** 0.170* LNYCORR THEIL 0.461*** 0.193-0.217* 0.051 YEARS 0.090*** 0.105*** 0.087*** LE URB -0.004*** -0.006*** -0.006*** SER -0.051*** -0.29*** -0.024*** IND -0.005-0.002-0.0004 COM 0.080 0.084*** 0.052*** LNYIYEARS -0.007 0.028 LNYURB -0.005*** -0.006*** LNYSER -0.049* -0.064*** LNYIND -0.019* -0.019* LNYCOM -0.026-0.008 LNYTHEIL -0.063 0.131 CO -0.079*** NE -0.165*** S -0.017 N -0.274*** R 2 0,08 0.26 0.45 0.55 Source: Own Calculations from Raw Data Obtained at www.ipeadata.gov. 16

Figures Figure 1 Kernel Estimation for the Distribution of Poverty Rates Amongst Groups of Municipalities in Brazil, 1991 (a) and 2000 (b). (b) (a) Source: Own Calculations from Raw Data Obtained at www.ipeadata.gov.br 17

Figure 2 Stochastic Kernel Estimation for the Conditional Distribution of Poverty Rates Amongst Brazilian Groups of Municipalities in 2000 Values Over Their 1991 Values. 18

Source: Own Calculations from Raw Data Obtained at www.ipeadata.gov.br 19

Figure 3 Contour Plot of the Stochastic Kernel Estimation for the Conditional Distribution of Poverty Rates Amongst Brazilian Groups of Municipalities in 2000 Values Over Their 1991 Values. 20

Source: Own Calculations from Raw Data Obtained at www.ipeadata.gov.br Figure 4 Kernel Estimation for the Distribution of Relative Per Capita Income Amongst Groups of Municipalities in Brazil, 1991 (a) and 2000 (b). (b) (a) Source: Own Calculations from Raw Data Obtained at www.ipeadata.gov.br 21

Figure 5 Stochastic Kernel Estimation for the Conditional Distribution of Relative Per Capita Incomes Amongst Brazilian Groups of Municipalities in 2000 Values Over Their 1991 Values. 22

Source: Own Calculations from Raw Data Obtained at www.ipeadata.gov.br Figure 6 Contour Plot of the Stochastic Kernel Estimation for the Conditional Distribution of Relative Per Capita Income Amongst Brazilian Groups of Municipalities in 2000 Values Over Their 1991 Values. 23

Source: Own Calculations from Raw Data Obtained at www.ipeadata.gov.br 24

References Barreto, F., Marinho, E. & Soares, F (2003). Crescimento Econômico, Concentração de Renda e Redução da Pobreza nos Estados Brasileiros. Porto Seguro: Annals of the Annual Brazilian Economic Association Meeting Datt, G. & Ravallion, M. (1992) Growth and Redistribution Components of Changes in Poverty Measures: a decomposition with applications to Brazil and India in the 1980s, Journal of Development Economics, 38, 275-295. Hoffmann, R. (2004). Elasticidade da pobreza em relação à renda média e à desigualdade. João Pessoa: Annals of the Annual Brazilian Economic Association Meeting. Kakwani, N. & Pernia, E.M.(2000) What is Pro-Proor Growth, Asian Development Review, 18(1). Kraay, A.(2004). When is Growth Pro-Poor? Cross-Country Evidence, IMF Working Paper, WP/04/47. Leal Filho, R. S. (2004). Mensurando o impacto do crescimento econômico e da desigualdade sobre a evolução da pobreza nas Unidades da Federação do Brasil, IDHS/PUCMINAS, II Workshop Rumo aos objetivos do milênio no Brasil. Paes de Barros, R, Henriques, R. & Mendonça, R. (2000). Rio de Janeiro: IPEA, Desigualdade e Pobreza no Brasil. Quah, D. T. (1997). Empirics for Growth and Distribution: stratitification, polarization, an convergence clubs, Journal of Economic Growth, 2, 27-59.Ravallion, M. & Chen, S. (2003) Measuring pro-poor growth, Economic Letters, 78, p. 93-99. 25

Ravallion. M. & Datt, G. (1999) When is Growth Pro-Poor, The World Bank Policy Research, WP Series, 2263. Ravallion. M. (2001) Growth, inequality and poverty: looking beyond averages. World Development. Ravallion. M. (2004). Pro-Poor Growth: a primer, The World Bank Policy Research, WP Series, 3242. Tocheto, D. G. et al (2004). Crescimento Pró-Pobre no Brasil uma análise exploratória. João Pessoa: Annals of the Annual Brazilian Economic Association Meeting. Zhang, Y. & Wan, G. (2005). Why do Poverty Rates Differ From Region to Region?, WIDER Research Paper, 2005/56. 26