WPS3752 Spatial dynamics of labor markets in Brazil Kenneth M. Chomitz*, Daniel da Mata, Alexandre Ywata de Carvalho, João Carlos Magalhães * Development Research Group, World Bank Directorate of Regional and Urban Studies, Institute of Applied Economic Research (IPEA) World Bank Policy Research Working Paper 3752, October 2005 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 view of the World Bank, its Executive Directors, or the countries they represent. Policy Research Working Papers are available online at http://econ.worldbank.org. Acknowledgments This paper is part of a joint research program between the World Bank and the Instituto de Pesquisa Econômica Aplicada (IPEA), Brasília. We are grateful to Jonny Andersson, Piet Buys, George da Guia, Waldir Netto, and Edinaldo Tebaldi, for assistance, advice, data construction, and literature reviews. We have benefited from discussions with Uwe Deichmann, Somik Lall, Maria da Piedade Morais, Marcelo Piancastelli and Vernon Henderson. Abstract There was substantial spatial variation in labor market outcomes in Brazil over the 1990's. In 2000, about one-fifth of workers lived in apparently economically stagnant municipios where real wages declined but employment increased faster than the national population growth rate. More than one-third lived in apparently dynamic municipios experiencing both real wage growth and faster-than-average employment growth; these areas absorbed more than half of net employment growth over the period. To elucidate this spatial variation, we estimated spatial labor supply and demand equations describing wage and employment changes of Brazilian municípios. We used Conley's spatial GMM technique to allow for instrumental variable estimation in the presence of spatially autocorrelated errors. Chief findings include: a very strong influence of initial workforce educational levels on subsequent wage growth (controlling for possibly confounding variables such as remoteness and climate); evidence of positive spillover effects of own- municipio growth onto neighbors' wage and employment levels; an exodus from farming areas; relatively elastic response of wages to an increase in labor supply; and evidence of a local multiplier effect from government transfers. Resumo O mercado de trabalho brasileiro apresentou uma dinâmica espacial diversa durante a década de 1990. Em 2000, aproximadamente um quinto dos trabalhadores vivia em municípios aparentemente estagnados em termos econômicos, em que os salários reais caíam, mas em que o emprego crescia acima da taxa de crescimento populacional do Brasil. Por outro lado, mais de um terço dos trabalhadores vivia em municípios dinâmicos, com crescimento dos salários reais e crescimento do emprego acima do crescimento populacional brasileiro: essas áreas absorveram mais da metade do crescimento líquido do emprego durante o período. A fim de elucidar essa dinâmica, o presente artigo estimou um modelo espacial de demanda e oferta por trabalho no qual descreve as mudanças no nível de salários e empregos dos municípios. Foi utilizado o método GMM espacial desenvolvido por Conley (1999), que permite o uso de variáveis instrumentais na presença de autocorrelação espacial. Os principais resultados incluem: a influência muito forte do nível educacional inicial da força de trabalho na taxa de crescimento subseqüente dos salários (mesmo após controlar por diversas variáveis, tais como distância e clima); presença de efeitos de transbordamento positivos do crescimento do município sobre os níveis de salário e emprego de seus vizinhos; queda no emprego em atividades rurais; elasticidade na resposta dos salários a um aumento na oferta de trabalho; e presença de efeitos multiplicadores das transferências governamentais. Brazilian spatial labor dynamics Page 1 Introduction and motivation In Brazil, as in other large, heterogeneous countries, there is considerable policy concern about spatial patterns of growth and development. Brazil exhibits large and relatively persistent differentials in income and welfare measures between north and south, and between rural and urban areas. The semi-arid Northeast, in particular, has been an area of concern for at least a century, and still is home to the hemisphere's largest concentration of poor people. At the same time, there has been concern about environmental and social problems associated with rapid expansion of the largest metropolitan areas. For this reason, there is long-standing interest in finding regional development policies that would stimulate growth in lagging and rural regions. The underlying assumption is that targeting regional development policies on these poorer regions could simultaneously reduce aggregate poverty and regional inequality, and thereby reduce migration to large urban centers. Some of these policies include federal and state incentives for industrial location in lagging areas; provision of basic education; infrastructure development including dams and roads; support for agriculture, including familial agriculture, and community-driven development projects. More recently there has been an interest in a `territorial approach', which would promote integrated development in a secondary city and its rural hinterland. This approach emphasizes improvement of urban amenities and services as a means of unlocking local growth. The impacts of these approaches have not been rigorously examined. Indeed, many of the underlying assumptions are open to debate. Do we know how to boost the productivity of secondary cities? Does a thriving secondary city stimulate growth in its surroundings ­ or displace that growth? Are local investments reflected in higher local wages, or in higher employment? Are there persistent differences in returns to investment in large vs. small cities, or in high vs. low population density areas, in more or less agroclimatically favored areas? This paper addresses these issues by modeling wages and labor supply at the município1 level. In contrast to the more familiar Barro-style growth models, such as in Barro and Sala-i-Martin (1991), the proposed models focus on the role of policy and local social and environmental assets in shaping labor incomes and net migration across the landscape. Working in a spatial econometric framework, we allow for spatial spillovers and for spatial autocorrelation of unobserved variables. The plan of the paper is as follows. The second part of this paper reviews trends in regional growth in Brazil and reviews policies that have tried to reduce regional inequalities. The third section reviews the literature on income, wage, and employment growth at the subnational level. This leads to specification of a simple labor demand- supply model. Next, we describe the data and the econometric specification employed. The final sections present results, discussion, and conclusions. 1The município is roughly equivalent to a US county: the administrative level beneath the state. They are extremely heterogeneous in size, ranging in population from a few hundred to over 8 million. There are currently 5561 municípios in Brazil. Brazilian spatial labor dynamics Page 2 Issues in Brazilian growth Brazil is a country with large disparities in income between individuals and at all geographical scales. Much though not all of this inequality stems from inequalities between regions. In particular, the Northeast of the country has historically lagged behind the South and Southeast, where much of the Brazilian economy is concentrated. The income per capita ratio of the richest State and the poorest one was 8.9 in 1960 and 6.2 in 1996. (Azzoni et al., 2000). The 2000 figure was 7.7. This analysis can be extended also to social indicators such as the human development index (HDI). Out of the 10 cities with the worst HDI in Brazil, seven were in the Northeast in 1991 and eight in 2000.2 Within regions, there are substantial inequalities between states and among municípios. And even within municípios there are substantial inequalities, with many indigent people living in the municipíos with the highest mean income. Particular attention has been focused on longstanding regional inequalities, especially between the Northeast and the rest of the country. There is some evidence of convergence in incomes between states over the periods 1939-1985 (Azzoni 2001) and 1970-1985 (Ferreira and Diniz 1995). However, several papers find evidence that this convergence process stalled after 1985, with one or two groups of poor states tending towards a lower-income equilibrium than the richer part of Brazil. (Azzoni 2001; Ferreira 1998; Pontual e Porto Júnior 2000). Azzoni et. al. (2000) estimated traditional regressions of growth with microdata for 19 Brazilians states. They divided their dataset in cohorts and used a large list of variables as determinants of growth, including microdata from PNAD ­ National Household Survey. The results suggest that per capita incomes of the Brazilian states have already converged to their steady-state level and that they will remain unaltered due to huge educational and geographic differences between the states. What accounts for these disparities? The North and Northeast of the country lag far behind other regions in quantity and quality of education, and these differentials are strongly correlated with differences in labor income, as human capital theory would suggest (Fiess and Werner 2004). However, education and other observable individual characteristics do not fully explain interregional wage and income differentials. Azzoni e Santos (2002) compared the differences in salary in the 10 largest Brazilian metropolitan areas in 1992, 1995 and 1997. Those salary differences remained significant even after controlling for cost of living measures, the traits of the workers (education, age, sex, race and family position) and the traits of their jobs (occupational position, sector and experience). Fiess and Werner (2004) estimated mover/stayer models of wage and migration. They found that low-education Northeasterners could boost their wages by 80% through migration; the differential declined with higher levels of education. Evidently, judging from the evidence cited above, the substantial degree of interregional migration is not sufficient to induce convergence between the regions. Brazilian policy has been concerned with reducing interregional inequalities. Best known, perhaps, are the national programs of tax and fiscal incentives, including the Constitutional Funds, subsidized loans from development banks including BNDES and BNB, the Zona Franca of Manaus, and FINOR. Some programs were explicitly targeted on poor regions. For instance, Projeto Alvorada sought to provide education and other services to states (and later microregions) with below-median HDI. Individual states use fiscal incentives to influence intrastate location of industrial firms, but the scope and impact of these programs is not comprehensively known. Other programs 2The others were in the North Region. Brazilian spatial labor dynamics Page 3 were not explicitly regional in intent but may nonetheless have had differential regional impacts. Possibly the most important of these is transport. Brazil's transport infrastructure has improved markedly over the last 40 years (Castro 2003). Lall, Funderberg, and Yepes (2005) show that the productivity of manufacturing firms tends to increase with decreasing transport cost to São Paulo, the country's industrial hub. On the other hand, increased access to São Paulo or other metropolitan centers exposes local firms to increased competition from possibly more efficient firms, so the net effect on regional development is indeterminant. Another important policy affecting regional development was EMBRAPA's development of soybean varieties adapted to low latitudes. This innovation was responsible for the massive expansion of soybean cultivation in the Center West after 1970. There is widespread enthusiasm in Brasil and throughout Latin America for a new, more fine-grained approach to regional development, `territorial development.' This approach has many of the elements of the growth poles approach that was popular 30 or 40 years ago. Although there are differing interpretations of this concept, it typically focuses on spurring the development of secondary cities as means of stimulating growth in the surrounding area. For instance, the states of Ceará and Bahia have devoted considerable analysis and planning to articulating detailed visions of territorial development. Both have adopted territorial development strategies emphasizing the development of secondary or strategic cities as cornerstones of regional growth (Bar-el et al 2002; Governo do Estado da Bahia, 2003). Ceará's visions of regional development provide concrete examples of the territorial development approach (Secretaria de desenvolvimento Local e Regional 2004), emphasizing improvement of road and air transport, improvement of basic services including sanitation and communication, development of cultural and natural resources as the basis of a tourism industry, support services to agriculture, including sheep and goats, and expansion of irrigated fruiticulture. To sum up, interregional inequalities are a long-standing concern in Brazil. Principal strategies for combating these inequalities have been the use of explicit or tax subsidies to attract industrial firms; improvements in transport and urban amenities; and investments in human capital. These strategies are implicitly assumed to boost wages and thereby to reduce poverty and interregional inequality. However, quantitative estimates of these impacts are lacking. Literature review There has been an explosion in the production of subnational growth models. Most of these are in the tradition of Barro and Sala-i-Martin (1991), and are concerned with the detection of convergence in per capita income levels or growth rates across the subnational units. The models are sparsely parameterized; growth is represented primarily as a function of prior-period per capita income, possibly with some structural conditioning variables. This approach is not ideal for the present purpose because there is little scope for exploring the impact of policy interventions such as infrastructure investments. When policy-relevant variables are included as conditioning factors, they are constrained to affect growth rates rather than income or employment levels. More apt for our purposes are the models applied by Fan and co-authors to India and China (see Fan, Zhang, and Zhang (2002), Fan Hazell and Haque (2000), Fan and Brazilian spatial labor dynamics Page 4 Chan-Kang (2004)). These papers seek to measure the marginal impact of government investments on income and on poverty, with particular attention to the returns to investment in less-favored areas. In Fan, Zhang, and Zhang (2000), for instance, a provincial production function relates agricultural GDP per capita to land per worker, agricultural capital per worker, agroclimate, and infrastructure stock. However, land per worker and capital per worker are taken to be fixed and exogenous. There is no allowance for interprovincial labor mobility. While the assumption of immobile labor may be acceptable when considering large provinces, and for countries where there are strong legal or social barriers to migration, it is less apt for Brazilian municípios. Here, labor mobility may play an important role in understanding differential spatial patterns of development. As Pritchett (2004) points out, in the absence of barriers to migration, one would expect local market or technology shocks to be reflected in labor movements rather than wage changes. Hence local investments may be effective in alleviating poverty even if they do not result in a perceptible change in wages. It is possible, instead, that they attract labor from lower wage areas. In this case, regional interventions would be effective in alleviating poverty but not in reducing measured interregional inequalities. Araujo, de Janvry, and Sadoulet (2004) present a município-level, spatial model of employment growth in Mexico that is the closest precursor of the current paper. They relate growth in employment over 1990-2000 to 1990 values of employment, proximity to urban centers and other employment nodes, geographic characteristics, and wages, and use spatial econometric techniques. They find that rural employment growth in both services and manufacturing is inversely related to distance to the nearest urban center. Approach: the spatial dynamic of labor supply and demand We model subnational development through a spatially disaggregate model of labor market dynamics. The labor market approach has several advantages for our purposes over Barro-type growth models. First, our model offers potentially better insight into poverty alleviation strategies because it looks at wages rather than GDP/capita ­ an important consideration when wealth is unequally distributed. Second, while the growth models use regions as units of concern, looking only at mean income levels, the labor market approach looks also at employment. It is entirely possible for regional mean incomes to diverge even while incomes of individuals increase, if employment expands faster in higher-wage markets. Models focused only on mean income may entirely overlook such an outcome. Third, growth models typically employ a sparse set of policy-relevant variables. Here we examine the impact of a range of policy levers, including education, infrastructure, and transfers, while allowing for differential effects in regions with different agroclimatic conditions. Finally, our fine-scaled geographic approach facilitates an examination of local growth spillovers, allowing examination of the premises of territorial development policies. A spatial model of labor and demand These considerations motivate a labor supply/demand model, which examines the determinants of labor income and employment growth at the município level. It is important to recall that labor income, and indeed municipal household income, does not Brazilian spatial labor dynamics Page 5 necessarily track municipal GDP. For instance, some rural municípios may be dominated by capital-intensive farming systems, whose value-added accrues to absentee landlords. Nonetheless, a focus on labor income is justified by a policy concern with spatial aspects of poverty and welfare. Our results can be compared with those of growth regressions by taking wage growth as a proxy for productivity growth and employment as a proxy for município size. Let each município i have a production function f (Ka, L, Educ; A) for a composite output, where K is a vector representing industrial and agricultural capital, including land); L is the number of workers; Educ is a vector measure of the quantity and quality of worker education; A is a vector of productivity shifters, which include transport connectivity to markets, local governance quality, and agroclimate. These are typically time-invariant or change slowly over time. The município faces a price which is a function P(MP,GT) of local market potential or demand and of government transfers to individuals. The latter, which largely represent rural pensions, are locally important in some areas and may drive demand for local nontradeable services. MP is operationalized as an inverse-distance-weighted function of the total incomes of neighboring municipalities. A labor demand equation expresses the wage rate as the value of the marginal product: (1) w = P(MP,GT )(f L) Differencing over time, using a convenient ln-linear approximation, we have: lnw = X0 + 1ln L + 2ln MP + 3lnGT + 4ln K + 5Education where X is a set of determinants of the rate of growth of productivity, and Education is the quality-adjusted mean educational level of the work force. We assume that wages adjust rapidly to changes in capital, labor supply, and prices. We assume, in contrast, that capital and labor adjust relatively slowly to changes across the landscape, in relative wages and returns to capital. Thus we model the change labor supply from t-1 to t as a function of local wages, local amenities, and competing neighborhood wages in period t-1: lnL = lnL(lnwt ,lnEWFt ,lnMPt , AMENITIESt , AGROCLIMATE) -1 -1 -1 -1 (2) where EWFt-1 is the relative size of the cohort, in period t-1, which will enter the labor force by period t. We hypothesize that the labor force increases more rapidly when initial local wages are high, reflecting long-distance migration3. Holding constant wages within the município, employment is expected to grow less rapidly, the more rapid the change in local market potential, because local labor may be relatively easily attracted to 3Ideally the wage measure should be adjusted for spatial differentials in cost of living; unfortunately spatial price indexes are not available, so we use nominal wages, inflation-adjusted nationally over time. Brazilian spatial labor dynamics Page 6 neighboring dynamic areas. Agroclimate is postulated as an exogenous determinant of employment growth. This allows for the possibility that agricultural labor is shifting between agroclimatic zones ­ and perhaps, especially, out of less favorable zones ­ in response to changing agricultural market conditions. We model capital investment as ln K = ln K(MARKETACCESSt ,GOVERNANCE, AGROCLIMATE,lnMP) -1 (3) Better market access (measured by lower transport costs to São Paulo and to the nearest state capital) is associated with lower price levels and hence higher returns to capital. Better local government is expected to increase the município's attractiveness for investment. Increases in the market potential indicate location near a dynamic region; this potentially endogenous instrument must however be instrumented. Because we lack reliable data on capital, we substitute (3) into (1). While we have data on mean years of education of the work force, we lack information on employees' educational quality. Moreover, a dynamic município may experience changes in mean educational level due to compositional effects, depending on the relative educational level of in-migrants, out-migrants, and stayers. Hence we use initial educational level of workforce, and initial educational level of public school teachers (a proxy for local educational quality) in place of educational change in (1). We interpret the result as describing the demand curve while allowing for the endogenous response of capital investment to initial conditions. Data We examined patterns of employment and wage change over the period 1991- 2000. The basic spatial unit of analysis was the município. A complication, however, is that some municípios experience splits over the study period, the total number growing from 4491 to 5507. In complicated cases, two municípios are reorganized into three. To address this problem, we merged municípios into 4267 Minimum Comparable Areas (MCAs) ­ consisting of sets of one to three municípios whose borders were constant over the study period. Much of the data stem from the Brazilian Population Censuses of 1991 and 2000.4 The key variables of interest were employment: total number of people reporting any employment and wage: reported labor income divided by the number of workers. Self-employed workers and farmers are counted as workers, and their earnings are counted as labor income. Table 6 gives the sources and definitions of variables employed in this paper. Table 7 provides summary statistics for these variables. Some of them are mapped in the appendix. A unique feature of this analysis is the use of market potential and related variables. Market potential for a target MCA i is defined as the weighted sum of personal incomes of all other MCAs, where the weight is a decreasing exponential of distance to the target município: 4To be precise, they stem from the 12,5% Census sample of 1991 and 5% Census sample of 2000. Both samples represent município level data. Brazilian spatial labor dynamics Page 7 MPi = exp( -dij / 2 2) 2 ji where dij is the great-circle distance between the centroids of i and j in kilometers, and is the inflection point in kilometers, here set to 25. The effective radius of influence is about 50 kilometers. Analogous expressions were used to derive distance-weighted measures of population and of mean educational level, again excluding values from the target município. These variables were used as instruments for market potential. The results reported here are divided into three categories: Brazil, non- metropolitan Brazil and the latter without the North region. The difference between the first two categories is based on IPEA, IBGE and UNICAMP (2002). That work makes a comprehensive classification of Brazilian urban agglomerations. We used the municípios that belong to urban agglomeration only in the first category. Spatial patterns of labor dynamics Figure 1 and tables 1-5 illustrate the geographic diversity of Brazil's labor market dynamics over the period 1991-2000, a diversity we seek to explain. We classify MCAs into four categories, depending on whether wage growth was positive or negative, and whether employment growth was above or below the mean national population growth rate The most striking point evident from the maps is the general decline in real wages in the North and Northeast, and general increase elsewhere in the country. However, because population density varies tremendously across Brazil (see figure 2), maps can give a misleading impression of the number of people who fall into the labor dynamics categories. Tables 1-5, and population density maps (figure 2), provide a more accurate representation of population breakdowns. The main point of the tables is spatial divergence and heterogeneity in experience. Areas in which employment grew rapidly but real wages declined (E+W-, shown in yellow in figure 1) can be thought of as areas of economic stagnation. Here, labor demand failed to keep up with the growth of supply. This may well reflect natural increase; the mean total fertility rate for the E+W- areas was 3.78 in 1991, far above the other three categories, and the proportion of immigrants in the 2000 population was substantially lower than for the E+W+ category. By 2000, about 22% of employed workers lived in the stagnant E+W- areas. They account for a particularly large share of population in the North and Northeast (figure 3). A set of dynamic areas with both increasing wage growth and rapidly increasing employment (W+E+) absorbed more than half of the country's net increase in employment, ending up with 36% of Brazilian employment in 2000. Such an outcome may reflect a dynamic local economy, with a relative shift outward in the labor demand curve. Although the 1991 total fertility rate for these MCAs was a full point lower than that of the E+W- areas, the dynamic areas had a substantial higher proportion of recent immigrants in 2000; this suggests that these regions are growing via immigration, presumably drawn by economic opportunity. Most of the E+W+ population is urban. Among the regions, the Center West has by far the largest proportion of people living in E+W+ AMCs, nearly 70%. The Northeast has the lowest proportion, just 10%. Brazilian spatial labor dynamics Page 8 Tables 4 and 5 show also that about two-thirds of employment is in the metropolitan agglomerations. Over the nine year interval, employment grew a total of 15% in the nonmetropolitan areas vs 20% in the metropolitan areas. Dynamic areas (W+E+) account for a much larger share of employment in metropolitan areas than in nonmetropolitan areas. Stagnant (W-E+) areas are much more prevalent in the nonmetropolitan areas. Figure 4 shows an important correlate of wage growth that is associated with the north/south differential. It plots, in ln-ln form, the mean educational level of 1991 employees against wage growth over 1991-2000, by MCA. MCAs from the North and Northeast are shown in red; others are in blue5. The figure shows that wage growth is strongly and significantly associated with initial educational level. It shows also that a significant disjunction in these variables between north and south. Northern and northeastern MCAs have markedly lower educational levels, and markedly lower wage growth. Is this a causal relationship? Places with higher human capital may also have higher levels of social capital and may therefore be better able to attract further investment in human and physical capital, and to boost local productivity. On the other hand, causality may go in the opposite direction. Places with inherent social capital or other local advantages may experience both more rapid growth and greater investments in education. The labor supply and demand equations estimated below attempt to hold constant other characteristics that might be associated with educational levels. Econometric model We estimate independent, ln-linear approximations to the wage (labor demand) equation of (3) substituted into (1), and the labor supply equation (2). Endogenous variables are instrumented with time-lagged or space-lagged exogenous determinants. In the wage equation, lnL is instrumented with lagged demographic determinants of labor force growth; the exclusion of these variables from the wage equation identifies it. The labor supply equation is identified, more tenuously, via the exclusion of the market access variables and the government transfer variables6, and by the use of initial period wage rather than wage change. Table 6 lists the instruments for each endogenous variable. Finally, a key issue in estimation is the likelihood of spatial autocorrelation of unobserved variables. Failure to account for this autocorrelation could lead to inaccurate estimates of standard errors. To address this issue, we use the GMM technique of Conley (1999) (as did Araujo, de Janvry and Sadoulet 2004 and Baicker 2005). In contrast to the more standard spatial econometric approach using a weight matrix, Conley's nonparametric technique requires only that users locate data points in a real or virtual space. Users specify a cutoff distance beyond which spatial autocorrelation is thought to be unimportant. For our analysis we use the latitude and longitude of the MCA centroid to locate the data points. This technique is attractive because it is computationally feasible for large numbers of observations; allows for the use of 5Readers are again referred to the color pdf version of this paper. The figure shows that the northeastern and northern observations are clustered in the lower left of the diagram. 6We assume that the transfers go mostly to old-age pensioners and thus do not affect labor supply Brazilian spatial labor dynamics Page 9 instrumental variables; and is robust to misspecification of the degree of autocorrelation among neighboring observations. Carvalho, da Mata and Chomitz (2005) discuss and assess the technique in detail. Results We present the results for three samples. The model was intended to focus on nonmetropolitan Brazil, under the assumption that growth dynamics of these regions differ substantially from those of metropolitan areas. However, we test that assumption by running a regression also for the entire country. Finally, we run a model for nonmetropolitan regions outside the North. Many Northern MCAs have very small and volatile work forces, and Amazonian frontier regions may have very distinctive patterns of employment growth. The regression results are in tables 8-10. For each model we present the results for both 2SLS and spatial GMM specifications. As a sensitivity test, we ran the GMM regressions using different cutoffs for spatial autocorrelation: 0.5, 1.0 and 2.0 degrees.7 For the Brazil-wide demand regressions, there is reasonable agreement among the specifications on a number of variables. All four variants find that initial educational level is significantly related to wage growth, other things constant, and the estimated coefficient is stable at about 0.06. This implies that an increase of one year in the initial mean educational level of the workforce is associated with a 6% increase in mean wages over the nine year period. Higher initial teacher qualification does not have a similar effect. This may reflect collinearity with years of education combined with nonlinearity in the relationship. In all specifications, rainfall is statistically significant and quantitatively extremely important. A 500 mm increase in annual rainfall is associated with a nine-year increase of 30% or more in wages. It is possible that this variable is capturing other aspects of inter-regional variation. In all four specifications, delta ln employment is statistically significant. The estimated demand elasticity of wages with respect to labor increases in absolute magnitude with increasing cutoff distance (where we treat 2SLS as having a cutoff of 0). The elasticity is -0.49 in the 2SLS specification, increasing to -0.89 for a cutoff of 2.0. All four specifications show significant impact of growth of market potential. The elasticity of wage growth with respect to market potential increases with cutoff distance, increasing from 0.55 in 2SLS to 1.30 in the cutoff 2 case. Coefficients of this magnitude suggest either very substantial local spillovers, or failures of the instrumental variables to control for unmeasured growth effects common to the MCA and its neighbors. Other variables of interest are statistically significant in the 2SLS specification, but less so in the spatial GMMs. On theoretical and empirical grounds, we expected capital investment, and therefore wage growth, to decline with increasing transport costs to São Paulo and to the nearest state capital. However, in the 2SLS specification these variables, though statistically significant, were negligible in absolute magnitude. Significance and magnitude declined with increasing cutoffs. The change in government transfer payments was statistically significant in the 2SLS and the cutoff=0.5 specifications, with coefficients of about 0.3 and 0.2 respectively. This suggests marked 7Each unit in the cutoff measure is equivalent to a hundred kilometers. Brazilian spatial labor dynamics Page 10 local multiplier effects of such transfers. However, the variable declined in magnitude and significance at higher cutoffs. Finally a rough proxy for quality of municipal governance ­ an indicator of whether the municipal financial accounts had been computerized by 1999 -- had a modest effect under 2SLS, but not under the GMM specifications. We turn now to the countrywide labor supply equation. Here all four specifications show strong agreement on the coefficients and significance of number of variables, although standard errors increase with the cutoff levels. First we consider wage variables. Labor supply has an elasticity of about 0.18 with respect to the 1991 wage level, so that higher initial wages are associated with a mildly faster rate of employment growth. Although we expected the change in ln market potential to have a negative coefficient, it was robustly determined at about 0.23. This suggests that location in a booming region has spillovers on labor supply as well as demand. Consider next the impact of initial demographic and agroclimatic conditions. The 1991 ratio of workforce entrants to working age population was also robust and highly significant, as expected. Perhaps most striking, all the specifications found a very strong negative association between the proportion of employment in farming in 1991, and subsequent employment growth. This strongly suggests an exodus from farming regions. Lower temperature and higher precipitation are associated with faster employment growth. The first and third principal components of rainfall8, which capture variation in annual seasonality of rainfall, are also significant though difficult to interpret directly. Very small MCAs showed more rapid growth than very large ones, all else equal. Two demographic variables showed robust effects contrary to our expectations. We had expected the initial labor force participation rate to be negatively correlated with subsequent employment growth, on the argument that places which had already experienced growth in female labor force participation would have less scope for further employment expansion. But in fact labor force participation was strongly positively associated with employment growth. We expected also that places with high native proportion of population (i.e. born in the município of current residence) represented areas that had barriers to outmigration or were unattractive to in-migration, and hence predicted lower employment growth rates. However, the coefficient on this variable was positive, significant and robust. Finally, the regression includes a few variables intended to capture the impact of local amenities on attracting and retaining labor. Initial homicides per capita had a negative effect on employment growth, though the significance of this coefficient declined with higher cutoffs. A one standard deviation increase in homicides corresponded roughly to a 4% decrease in employment growth over the nine year period. The initial level of teacher qualifications had a negative effect on employment growth, contrary to expectations, but the effect was of negligible magnitude. The demand-side results for the nonmetropolitan sample (table 5) are quite similar to those for the nationwide sample, despite the huge disparity in MCA size. The coefficients on education and rainfall are of about the same magnitude as before. The labor elasticity coefficient is about -0.74 and significant in all the GMM specifications, slightly more negative than the 2SLS specification. The coefficient of the change in market potential is also relatively consistent across the GMM specifications, with an 8Calculated from the vector of the 12 monthly averages of precipitation. The second principal component is nearly collinear with total precipitation. Brazilian spatial labor dynamics Page 11 elasticity of about 0.6. In contrast to the nationwide sample, the coefficient on government transfers here stays robust across the GMM specifications, at about 0.16. The coefficients on transport cost are here more significant, though still negligible in magnitude. These results (including the effect of rainfall) are essentially unchanged when the North region is excluded. The supply-side results for the nonmetropolitan samples also are very similar to the nationwide results. The main difference is that the effect of the initial proportion in farming is reduced in magnitude, but it is still quite important quantitatively and statistically. In addition, the initial participation rate is no longer significant in the nonmetropolitan analyses. A final caution on our results relates to the J-test. For most of the GMM regressions, the test statistic is significant at the 5% level, which suggests potential problems with the instruments. It is not however significant at that level for the Brazilwide demand regression (cutoff of 2.0) and the nonmetropolitan supply regressions excluding the North (cutoffs of 1.0 and 2.0). Discussion We have found a number of relationships that appear to be robust to different choices of sample and different controls for spatial autocorrelation. Initial levels of workforce education are strongly related to subsequent wage growth, even after controlling for correlated variables such as remoteness and climate. Note that this is different from the well-known association between education and wage levels; it suggests an accelerator effect, where more-educated regions experience more rapid wage growth. And, the effect is large. In 1991, the mean educational level of workers was 3.9 years for the country as a whole and just 3.6 years for nonmetropolitan regions. The standard deviation for nonmetropolitan regions was 1.32 years. A change of this magnitude was associated with a 8% increase in wages over 1991-2000. If confirmed, this a sobering finding from the viewpoint of reducing inter-regional inequalities. It suggests that wages will continue to diverge between the more- educated south and the less-educated north. And because it takes 40 years to turn over the labor force, it will take a long time for lagging regions to reduce the educational gap with leading ones, even if the former were to make more rapid progress ­ itself a questionable assumption. Farming regions are losing employment, or growing more slowly, than other regions. In the nationwide sample, each 10 percentage point increase in the initial proportion of farmers among workers was associated with a 10 to 12 percent decrease in the nine- year rate of employment growth. Even when the sample is restricted to nonmetropolitan areas, this relationship holds, though it is less steep. Further investigation is needed to understand the degree to which this represents `pull' factors ­ attraction to growing areas, vs. `push' factors ­ e.g. displacement of smallholders by large farming enterprises. Low rainfall areas lagged other areas in wage and employment growth. These patterns were evident even controlling for remoteness and education, which might be correlated with rainfall. Wages respond relatively elastically to changes in labor supply. This suggests that in-migration could substantially reduce wages in an MCA with vigorous demand growth; that Brazilian spatial labor dynamics Page 12 population growth would depress wages in a stagnant MCA; and that outmigration could put upward pressure on wages for those left behind in declining MCAs. The supply-elasticity of wages is important to keep in mind when evaluating regional performance. According to our estimates, a dynamic region that boosts labor demand by 10% and accommodates labor force growth of 20% would experience unchanged wages, other things equal. Assessing the region only on the basis of wage growth would completely miss its success in providing new jobs, possibly to migrants who substantially increase their individual earnings. There appear to be positive spillover effects on wages and employment from income growth in nearby areas. Our estimates suggest that for nonmetropolitan areas a 10% increase in close neighboring regions' income is associated with a 7% increase in own wages and a 2% increase in employment. While we cannot rule out the possibility that this is due in part to a correlation with unobserved favorable factors common to the município and its environs, it suggests some support for the territorial development approach which advocates stimulating the growth of small cities in order to benefit nearby neighbors. Whether or not it is possible, in fact, to spur the growth of such cities is a different question. Government transfers ­ such as pensions--appear to stimulate local wage growth. More rapid growth in receipt of transfers is associated with more rapid wage growth. Since much of the growth in transfers over the 1990s is associated with the expansion of rural pensions, this finding supports the idea that such transfers result in favorable local multiplier effects, as the pensioners increase their demand for local goods such as services and some foodstuffs. This situation has been characterized as the `economy without production' (Maia Gomes, 2001). That epithet carries a negative connotation. A more positive view is that an increase in transfers to poor locales, with little natural or human assets, not only serves direct social needs but stimulates the local economy. Conclusions The paper began by mapping the heterogeneity of labor market outcomes across Brazil during the 1990s. To understand the determinants of that heterogeneity, we developed a spatial model of labor supply and demand, and applied an estimation strategy that takes into account endogeneity of the explanatory variables and spatial autocorrelation of unobserved determinants of labor force outcomes. Our results strongly emphasize the role of workforce education in determining growth prospects and the importance of local spillovers, and point to an exodus of labor from farming regions. This paper must be viewed as an initial step in a large research agenda. Areas for further investigation include more explicit treatment of the impacts of changes in agricultural prices and technologies, including the stimulus of growth of agricultural service cities; and incorporation of better data on capital stock, and on municipal governance, agroclimate, and infrastructure. Brazilian spatial labor dynamics Page 13 References ANSELIN, L. Spatial econometrics: methods and models. Dordrecht: Kluwer Academic, 1988. ARAUJO, C., A. de JANVRY, and E. SADOULET. 2004. Spatial patterns of non- agricultural rural employment growth in Mexico during the 90s. Department of Agricultural and Resource Economics, University of California, Berkeley. Processed. AZZONI, C. Economic Growth and regional income inequality in Brazil. The Annals of Regional Science, v. 35, p.133-152, 2001. AZZONI, C., MENEZES, T., MENEZES, N., NETO, R. Geografia e convergência da renda entre os estados brasileiros. In: HENRIQUES, R. (Org.). Desigualdade e pobreza no Brasil. Rio de Janeiro: IPEA, 2000, p.299-343. AZZONI, C., SANTOS, L. (2002). Education, cost of living and regional wage inequality in Brazil in the 90´s. Papers in Regional Science, v. 81, p. 157-175. BAICKER, C. (2005). The spillover effect of state spending. Journal of Public Economics, 89(2-3), 529-544. BAR-EL, Raphael et al. 2002. Reduzindo a Pobreza através do Desenvolvimento Econômico do Interior do Ceará. Fortaleza: Edições IPLANCE. BARRO, R. J. (1991) Economic growth in a cross section of countries. Quarterly Journal of Economics, v.106, n.2, p.407-43, 1991. BARRO, R.J., SALA-I-MARTIN, X. (1991) Convergence across states and regions, Brookings Papers on Economic Activity (1991) (1), pp. 107­182. CARVALHO, A., da MATA, D., CHOMITZ, K. Estimation of Multiequation Cross-Section Models in the Presence of Spatial Autocorrelation. 2005, in prep. CASTRO (2003). Logistic cost and Brazilian regional development. processed. FAN, S., CHAN-KANG, S. (2004). Returns to investment in less-favored areas in developing countries: a synthesis of evidence and implications for Africa. Food Policy 29, 431-444. FAN, S., HANZELL P., HAQUE, T. (2000). Targeting public investments by agroecological zone to achieve growth and poverty alleviation goals in rural India. Food Policy 25, 411-428. FAN S., L. ZHANG, X. ZHANG (2002) Growth, inequality and poverty in rural China : the role of public investments. International Food Policy Research Institute, report 125. FERREIRA, A. H. B. A distribuição interestadual e inter-regional da renda no Brasil: tendências recentes. 1995. 101f. Tese (Professor titular) - Departamento de Ciências Econômicas, Universidade Federal de Minas Gerais, Belo Horizonte, 1995. Brazilian spatial labor dynamics Page 14 FERREIRA, A. H. B., DINIZ, C. C. Convergência entre as rendas per capita estaduais no Brasil. Belo Horizonte: UFMG/CEDEPLAR, 1994. 27p. (Texto para Discussão, n.79) FERREIRA, P. C. G., ELLERY JR., R. G. Convergência entre a renda per capita dos estados brasileiros. Revista de Econometria, v.16, n.1, p.83-103, abr. 1996. FIESS, N.M. and D. WERNER. .2004. Migration and human capital in Brazil during the 1990s. World Bank Policy Working Paper no. 3093. GLAESER, E. L. et al. Growth in cities. Journal of Political Economy, v.100, n.6, p.1126- 1152, Dec.1992. GLAESER, E. L., SCHEINKMAN, J. A., SHLEIFER, A. Economic growth in a cross-section of cities. Cambridge, Mass.: NBER, 1995. 19p. (Working Papers, n.5013) GOVERNO DO ESTADO DA BAHIA (2003) O Futuro a Gente Faz: Plano Estratégico da Bahia. Salvador: SEPLAN. LALL, S., FUNDERBERG, R., YEPES, T. (2005). Location, concentration, and performance of economic activity in Brazil. World Bank Policy Research Working Paper 3268. LUCAS, Robert. On the mechanics of economic development. Journal of Monetary Economics, v.22, n.1, p.3-42, Jul. 1988. MAGALHÃES, A., HEWINGS, G., AZZONI, C. Spatial dependence and regional convergence in Brazil. Chicago: University of Illinois, 2000. 20p. (Working Papers REAL 00-T-11, Urbana) Disponível em: . MAIA GOMES, G. 2001. Velhas secas em novos sertões: continuidade e mudanças na economica do semi-arido e dos cerrados nordestinos. Brasilia: IPEA. PRITCHETT, L. (2004) Boom towns and ghost countries: geography, agglomeration, and population mobility. Center for Global Development, Working Paper no. 36. RIVKIN, S., HANUSHEK, E., KAIN, J. (2005) Teachers, schools and academic achievement, Econometrica, v.73, p.417-458. ROMER, P. M. Endogenous technological change. Journal of Political Economy, v.98, n.5 (part 2), p.71-102, Oct. 1990. ROMER, P. M. Increasing returns and long-run growth. Journal of Political Economy, v.94, n.5, p.1002-1037, Oct. 1986. SECRETARIA DE DESENVOLVIMENTO LOCAL E REGIONAL, n.d. (2004?). Planos de Desenvolvimento Regional. (various volumes: Baixo Jaguaribe, Maçico de Baturite, Vale deo Coreaú e Ibiapaba) Fortaleza: Governo do Estado do Ceará, Secretaria De Desenvolvimento Local e Regional. Brazilian spatial labor dynamics Page 15 SILVEIRA NETO, R. M. Crescimento e spillovers: a localização importa? Evidências para os estados brasileiros. Revista Econômica do Nordeste, Fortaleza, v.32, N.ESP., p.524- 545, 2001. Tables TABLE 1 - MCA Average Region Fertility Rate* Percentage of immigrants Population Growth - in population* geometric annual average** 1991 2000 1991 2000 1991 - 2000 North 4.42 3.33 24.41 22.99 2.87 Northeast 3.89 2.82 16.99 16.12 1.29 Southeast 2.46 2.16 19.21 19.08 1.61 South 2.64 2.30 22.13 21.32 1.42 Center-West 2.82 2.36 31.95 29.27 2.33 *Mean MCAs values weighted by population ** Calculated by the aggregate data TABLE 2 - MCA Average Percentage of immigrants in Population Growth - geometric Quadrant Fertilty Rate* population* annual average** 1991 2000 1991 2000 1991 - 2000 +, W+ 2.79 2.31 25.19 24.32 2.58 -, W+ 2.68 2.24 17.25 15.53 0.69 -, W- 3.26 2.54 15.47 14.56 0.51 +, W- 3.78 2.90 21.15 21.13 2.38 *Mean MCAs values weighted by population ** Calculated by the aggregate data TABLE3-MetropolitanAreas Employment Employment Employment Share Population Population growth Growth Frequency Quadrant 1991 2000 1991 2000 1991 2000 E+, W+ 14707336 19090991 4383655 40% 43% 37785537 47940239 10154702 332 E-, W+ 10144794 10924107 779313 27% 25% 25298240 27583837 2285597 154 E-, W- 6188863 6547091 358228 17% 15% 15430178 16591747 1161569 73 E-, W- 5978230 7957879 1979649 16% 18% 16504563 21302846 4798283 176 Total 37021214 44522068 7500854 100% 100% 95018518 113420669 18402151 735 Brazilian spatial labor dynamics Page 16 TABLE4 - Nonmetropolitan Areas Employment EmploymentEmployment Share Population Population growth Growth Frequency Quadrant 1991 2000 1991 2000 1991 2000 E+, W+ 3481048 4551220 1070172 19% 22% 9300552 10991493 1690941 701 E-, W+ 5949685 5950681 996 33% 28% 15123906 15312581 188675 1172 E-, W- 4005760 4008645 2885 22% 19% 11833148 11888764 55616 743 E-, W- 4837597 6599279 1761682 26% 31% 15549351 18187663 2638312 916 Total 18276081 21111825 2835744 100% 100% 51806957 56382501 4575544 3532 TABLE 5 - All Areas Employment Employment Share of Population Population growth employment Growth Frequency Quadrant 1991 2000 1991 2000 1991 2000 +, W+ 18188385 23642211 5453826 32% 35% 47086089 58931732 11845643 1033 -, W+ 16094479 16874788 780309 28% 25% 40422146 42896418 2474272 1326 -, W- 10194623 10555736 361113 19% 17% 27263326 28480511 1217185 816 +, W- 10815827 14557158 3741331 22% 23% 32053914 39490509 7436595 1092 Total 55295305 65631893 10336588 100% 100% 146825475 169801170 22975695 4267 Brazilian spatial labor dynamics Page 17 Table 6. Variables and Sources A. Variables used in the Wage Equation and their sources: Variable Source Instrument(s) Source Teacher qualification in 1991 IBGE - Census Exogenous Years of schooling in 1991 IBGE - Census Exogenous Total precipitation DECRG - IE Exogenous Government with Accountability IBGE - MUNIC Exogenous Employment rate in 1991 IBGE - Census Soil quality DECRG - IE Delta employment IBGE - Census First and third principal components of DECRG - IE monthly rainfall Transport cost to São Paulo NEMESIS Transport cost to São Paulo in 1968 NEMESIS Transport cost to nearest State Transport cost to nearest State Capital in NEMESIS NEMESIS Capital 1968 Dependency ratio in 1991 IBGE - Census Illiteracy rate in 1991 IBGE - Census Repetition rate in 1991 IBGE - Census Delta government transference IBGE - Census Proportion of elderly persons in 1991 IBGE - Census Population in 1991 IBGE - Census Proportion 5-15 over 15-55 yrs old in 1991 IBGE - Census Delta education market potential DECRG - IE Delta Market Potential DECRG - IE Urbanization rate in 1991 IBGE - Census Population density in 1991 IBGE - Census Brazilian spatial labor dynamics Page 18 B. Variables used in the Labor Equation and their sources: Variable Source Instrument(s) Source Wage in 1991 IBGE - Census Exogenous Proportion 5-15 over15-55 yrs old IBGE - Census Exogenous Proportion native in 1991 IBGE - Census Exogenous Delta market potential DECRG - IE Exogenous Teacher qualification in 1991 IBGE - Census Exogenous Homicides per capita in 1991 DATASUS Exogenous Proportion farmers in 1991 IBGE - Census Exogenous Bank Dummy Central Bank Exogenous Population in 1991 IBGE - Census Exogenous Mean Temperature DECRG - IE Exogenous Total Rainfall DECRG - IE Exogenous Rainfall - 1º Principal Component DECRG - IE Exogenous Rainfall - 3º Principal Component DECRG - IE Exogenous Illiteracy rate in 1991 IBGE - Census Employment rate in 1991 IBGE - Census Fertility rate in 1991 IBGE - Census Brazilian spatial labor dynamics Page 19 Table 7A. Summary Statistics, all Brazil Variable Obs Mean Std.Dev. Min Max Delta ln Wage (2000/1991) 4267 -0.0101 0.2935 -1.5242 0.8495 ln Teacher qualification in 1991 4267 1.3938 2.1088 -4.6052 4.2370 Years of Schooling in 1991 4267 3.9181 1.4556 0.4425 9.6572 Total rainfall 4266 1337.3190 425.4345 368.6636 3361.7090 Government Accountability 4267 0.9121 0.2832 0.0000 1.0000 Delta ln Employment (2000/1991) 4267 0.1422 0.2224 -0.6961 1.8755 lnTransport Cost to São Paulo in 1995 4266 7.0420 0.8469 2.3026 9.2603 ln Transport Cost to State Capital in 1995 4256 5.8180 0.7780 1.3863 8.6910 Delta ln Transferences (2000/1991) 4267 0.8083 0.3221 -1.1823 2.4240 Delta ln Market Potential (2000/1991) 4,258 0.4093 0.1375 -0.4691 2.0949 ln Population in 1991 4,267 9.5287 1.0888 6.6214 16.0821 Dependency Ratio in 1991 4267 0.4775 0.0466 0.3738 0.6205 ln Transport Cost to São Paulo in 1968 4266 7.6264 0.7874 3.2189 9.6385 ln Transport Cost to State Capital in 1968 4256 6.2548 0.7959 1.3863 8.6910 ln Proportion of Elderly in 1991 4267 -2.2502 0.2621 -3.9145 -1.5332 Illiteracy Rate in 1991 4267 35.5004 17.8266 2.4800 87.0500 Urbanization Rate in 1991 4267 53.6356 23.1683 2.1562 100.0000 Population Density in 1991 4267 93.6426 510.3321 0.0896 12199.7700 ln 5 to 15 over 15 to 55 years old Ratio 4267 -0.7491 0.2303 -1.6218 -0.0354 Repetition Rate in 1991 4267 4.2935 0.2071 2.3542 4.5756 ln Employment Rate in 1991 4267 -0.3760 0.1411 -1.2429 0.1601 Delta ln Education Market Potential (2000/1991) 4251 0.9737 0.1736 0.0535 2.7704 Percentage of Good Soils 4266 0.7904 0.3071 0.0000 1.0002 Rainfall - First Principal Component 4266 0.0000 2.2685 -4.2315 6.9861 Rainfall - Third Principal Component 4,266 0.0000 1.5749 -2.6324 9.2440 ln Wage in 1991 4267 8.0265 0.4963 6.6139 9.8121 Proportion of Natives in 1991 4267 0.6565 0.1790 0.0855 1.0134 Homicides per Capita in 1991 4265 0.0016 0.0011 0.0000 0.0090 Employment in Farming Activities in 1991 4,267 0.1777 0.0854 0.0000 0.5437 Bank Dummy (1991) 4,267 0.7492 0.4335 0.0000 1.0000 4267 9.5287 1.0888 6.6214 16.0821 Average Temperature 4266 22.3050 2.8451 13.8325 27.8109 Fertility Rate in 1991 4,267 3.6449 1.1885 1.7600 8.6800 Brazilian spatial labor dynamics Page 20 Table 7B, Summary statistics, Non-metropolitan Brazil Variable Obs Mean Std.Dev. Min Max Delta ln Wage (2000/1991) 3532 -0.0233 0.3097 -1.5242 0.8495 ln Teacher qualification in 1991 3532 1.1990 2.1519 -4.6052 4.2370 Years of Schooling in 1991 3532 3.6414 1.3197 0.4425 7.1707 Total rainfall 3,531 1,312.3340 436.0881 368.6636 3,361.7090 Government Accountability 3,532 0.9023 0.2969 0.0000 1.0000 Delta ln Employment (2000/1991) 3532 0.1231 0.2173 -0.6577 1.3689 lnTransport Cost to São Paulo in 1995 3531 7.1172 0.7733 4.1589 9.2603 ln Transport Cost to State Capital in 1995 3531 5.9358 0.6736 2.1972 8.6910 Delta ln Transferences (2000/1991) 3532 0.8316 0.3262 -1.1823 2.4240 Delta ln Market Potential (2000/1991) 3523 0.4053 0.1330 -0.4691 2.0949 ln Population in 1991 3,532 9.2629 0.8171 6.6214 12.9814 Dependency Ratio in 1991 3532 0.4840 0.0460 0.3738 0.6205 ln Transport Cost to São Paulo in 1968 3531 7.6962 0.7123 4.3694 9.6385 ln Transport Cost to State Capital in 1968 3531 6.3629 0.7037 3.1135 8.6910 ln Proportion of Elderly in 1991 3532 -2.2260 0.2566 -3.9145 -1.5332 Illiteracy Rate in 1991 3532 38.0138 17.5883 2.4800 87.0500 Urbanization Rate in 1991 3532 48.7978 20.6294 2.7639 100.0000 Population Density in 1991 3532 31.1089 33.0202 0.0896 522.2028 ln 5 to 15 over 15 to 55 years old Ratio 3,532 -0.7264 0.2329 -1.6218 -0.0354 Repetition Rate in 1991 3,532 4.2707 0.2125 2.3542 4.5756 ln Employment Rate in 1991 3532 -0.3722 0.1462 -1.2429 0.1601 Delta ln Education Market Potential (2000/1991) 3516 0.9745 0.1732 0.0535 2.7704 Percentage of Good Soils 3531 0.7816 0.3179 0.0000 1.0002 Rainfall - First Principal Component 3531 0.0033 2.2334 -4.2315 6.8953 Rainfall - Third Principal Component 3531 0.0262 1.6087 -2.4997 9.2440 ln Wage in 1991 3532 7.9349 0.4619 6.6139 9.3168 Proportion of Natives in 1991 3532 0.6778 0.1760 0.0855 1.0134 Homicides per Capita in 1991 3530 0.0015 0.0010 0.0000 0.0090 Employment in Farming Activities in 1991 3532 0.1975 0.0743 0.0000 0.5437 Bank Dummy (1991) 3532 0.7087 0.4544 0.0000 1.0000 3532 9.2629 0.8171 6.6214 12.9814 Average Temperature 3531 22.4151 2.8196 13.8325 27.7363 Fertility Rate in 1991 3532 3.7759 1.2169 2.0100 8.6800 Brazilian spatial labor dynamics Page 21 Table 7C. Summary statistics, Non-metropolitan Brazil without North Region Variable Obs Mean Std.Dev. Min Max Delta ln Wage (2000/1991) 3300 0.0000 0.2901 -1.3942 0.8495 ln Teacher qualification in 1991 3300 1.3156 2.1187 -4.6052 4.2370 Years of Schooling in 1991 3300 3.6858 1.3320 0.4746 7.1707 Total rainfall 3299 1251.7990 367.9959 368.6636 2532.2020 Government Accountability 3300 0.9124 0.2827 0.0000 1.0000 Delta ln Employment (2000/1991) 3300 0.1103 0.2035 -0.6577 1.2464 lnTransport Cost to São Paulo in 1995 3299 7.0422 0.7355 4.1589 8.3051 ln Transport Cost to State Capital in 1995 3299 5.8691 0.6163 2.1972 7.7619 Delta ln Transferences (2000/1991) 3300 0.8448 0.3073 -0.2986 2.3336 Delta ln Market Potential (2000/1991) 3299 0.4101 0.1179 -0.2664 1.5442 ln Population in 1991 3300 9.2337 0.8054 6.6214 11.9770 Dependency Ratio in 1991 3300 0.4807 0.0447 0.3738 0.6205 ln Transport Cost to São Paulo in 1968 3299 7.6219 0.6676 4.3694 8.7367 ln Transport Cost to State Capital in 1968 3299 6.2990 0.6601 3.1135 7.9077 ln Proportion of Elderly in 1991 3300 -2.1936 0.2204 -3.6893 -1.5332 Illiteracy Rate in 1991 3300 37.7150 17.8150 2.4800 81.4600 Urbanization Rate in 1991 3300 49.2604 20.7435 2.7639 100.0000 Population Density in 1991 3300 32.7412 33.3495 0.2065 522.2028 ln 5 to 15 over 15 to 55 years old Ratio 3300 -0.7464 0.2243 -1.6218 -0.1870 Repetition Rate in 1991 3300 4.2817 0.1976 2.6575 4.5756 ln Employment Rate in 1991 3300 -0.3625 0.1406 -1.2429 0.1601 Delta ln Education Market Potential (2000/1991) 3297 0.9707 0.1575 0.1650 1.6027 Percentage of Good Soils 3299 0.7881 0.3192 0.0000 1.0002 Rainfall - First Principal Component 3299 -0.0130 2.1967 -4.2315 6.8953 Rainfall - Third Principal Component 3299 -0.1705 1.3696 -2.4997 5.4633 ln Wage in 1991 3300 7.9233 0.4684 6.6139 9.3168 Proportion of Natives in 1991 3300 0.6765 0.1694 0.0855 1.0134 Homicides per Capita in 1991 3299 0.0015 0.0010 0.0000 0.0090 Employment in Farming Activities in 1991 3300 0.1995 0.0752 0.0000 0.5437 Bank Dummy (1991) 3300 0.7197 0.4492 0.0000 1.0000 Average Temperature 3299 22.1518 2.7223 13.8325 27.7363 Fertility Rate in 1991 3300 3.6513 1.1051 2.0100 7.9600 Brazilian spatial labor dynamics Page 22 Table 8: Regression results, Brazilwide Spatial GMM Spatial GMM Dependent Variable: 2SLS Est. 2SLS SE Est. SE Delta ln Wage (Cutoff = 0.5) (Cutoff = 0.5) Intercept -.23969856 .06883838 -.56879473 .30864844 ln Teacher qualification in 1991 .00548926 .00230455 .00416922 .0057475 Years of Schooling in 1991 .06204049 .00435942 .05989964 .01154468 Total Rainfall .00005344 9.645e-06 .00006334 .00002661 Government Accountability .02233476 .01293292 .01689856 .03458774 Delta ln Employment -.49393135 .05272667 -.60942137 .21566178 ln Transport Cost SP in 1995 -.03205953 .00590571 -.02147947 .02295913 ln Transport Cost Capital in 1995 -.04812287 .00601824 -.01909574 .02022743 Delta ln Transferences .29663134 .03118072 .20129193 .11094594 Delta ln Market Potential .54819572 .09031346 .98877352 .29459548 crit. fn. test of overid. restrictions ---- ---- 24.106339 24.106339 Spatial GMM Spatial GMM Spatial GMM Spatial GMM Dependent Variable: Est. SE Est. SE Delta ln Wage (Cutoff = 1) (Cutoff = 1) (Cutoff = 2) (Cutoff = 2) Intercept -.71932918 .39269048 -.89149666 .46470872 ln Teacher qualification in 1991 .00313908 .00588544 .00077822 .00636871 Years of Schooling in 1991 .06201083 .01286978 .06663528 .01474433 Total Rainfall .0000674 .00002795 .00007651 .00002947 Government Accountability .02179158 .03504303 .03300684 .035756 Delta ln Employment -.74141878 .24696933 -.89418358 .27639595 ln Transport Cost SP in 1995 -.01160188 .02949758 .00306388 .03553532 ln Transport Cost Capital in 1995 -.00840394 .02391369 .00218674 .02606355 Delta ln Transferences .13809767 .12475475 .05152986 .13939323 Delta ln Market Potential 1.1562178 .34490082 1.3035762 .36538869 crit. fn. test of overid. restrictions 19.804367 19.804367 14.762519 14.762519 Brazilian spatial labor dynamics Page 23 Spatial GMM Spatial GMM Dependent Variable: 2SLS Est. 2SLS SE Est. SE Delta ln Employment (Cutoff = 0.5) (Cutoff = 0.5) Intercept -.61966401 .1396479 -.59779578 .19288857 ln Wage in 1991 .18148355 .01471158 .1805902 .02061008 ln 5 to 15 over 15 to 55 yrs Ratio .32737947 .03751615 .3284415 .05359506 Proportion of Natives in 1991 .26274058 .04047279 .28160825 .06166641 Delta ln Market Potential .23244707 .02541535 .22884513 .04132408 ln Teacher qualification in 1991 -.01008945 .0023123 -.01034963 .00348222 Homicides in 1991 -23.377317 4.2814245 -26.064766 11.646685 Employment in Farming in 1991 -1.0925149 .16331391 -1.160606 .25061706 Bank Dummy -.00883696 .00937983 -.00637937 .01263533 ln Population in 1991 -.0271189 .00513829 -.02845369 .00796092 Mean Temperature -.00216998 .00263983 -.00152014 .00396835 Total Rainfall .00002401 .00001217 .00002763 .00001916 Rainfall - 1º Princ. Component .00693925 .00183254 .00660851 .00291956 Rainfall - 3º Princ. Component .01082346 .00425524 .0101329 .00634718 Employment rate in 1991 .50765127 .14583623 .55593884 .22256766 crit. fn. test of overid. restrictions ---- ---- 4.3194265 4.3194265 Spatial GMM Spatial GMM Spatial GMM Spatial GMM Dependent Variable: Est. SE Est. SE Delta ln Employment (Cutoff = 1) (Cutoff = 1) (Cutoff = 2) (Cutoff = 2) Intercept -.58264148 .21309273 -.55973566 .23194173 ln Wage in 1991 .17833635 .02288821 .17472738 .02582378 ln 5 to 15 over 15 to 55 yrs Ratio .32701012 .05974778 .32103962 .06498975 Proportion of Natives in 1991 .28581622 .07064752 .2838648 .08167097 Delta ln Market Potential .22956501 .04807927 .22942847 .05732953 ln Teacher qualification in 1991 -.01069225 .00410323 -.01055971 .00482079 Homicides in 1991 -29.090509 16.986828 -34.329254 26.672443 Employment in Farming in 1991 -1.1970719 .28437662 -1.2222365 .31455466 Bank Dummy -.00550414 .01287836 -.00412273 .01383933 ln Population in 1991 -.028508 .00898918 -.02745768 .01065978 Mean Temperature -.00111862 .00469047 -.00119964 .00556795 Total Rainfall .00003371 .00002314 .00003857 .00002848 Rainfall - 1º Princ. Component .00608015 .00340453 .0056972 .00382822 Rainfall - 3º Princ. Component .00928193 .00771074 .00883827 .00899555 Employment rate in 1991 .5768252 .24739395 .58084285 .26483339 crit. fn. test of overid. restrictions 3.4865079 3.4865079 2.7619863 2.7619863 Brazilian spatial labor dynamics Page 24 Table 9: Regression results, Non-metropolitan Brazil Spatial GMM Spatial GMM Dependent Variable: Est. 2SLS Est. 2SLS SE SE Delta ln Wage (Cutoff = 0.5) (Cutoff = 0.5) Intercept .09597151 .10607452 -.07212585 .15562966 ln Teacher qualification in 1991 .00297498 .00255803 -.00046191 .00367207 Years of Schooling in 1991 .06753124 .00544935 .06342638 .00746079 Total Rainfall .00005995 .00001096 .00007959 .00001616 Government Accountability .01715587 .01423606 .00466584 .02099178 Delta ln Employment -.58726752 .07915421 -.72860634 .13080938 ln Transport Cost SP in 1995 -.05128457 .00834921 -.04954901 .01126003 ln Transport Cost Capital in 1995 -.06169838 .00863113 -.04358863 .0138009 Delta ln Transferences .24978392 .04062804 .16556984 .06593844 Delta ln Market Potential .33007705 .12288672 .67970574 .17956081 crit. fn. test of overid. restrictions ---- ---- 51.198551 51.198551 Spatial GMM Spatial GMM Spatial GMM Spatial GMM Dependent Variable: Est. SE Est. SE Delta ln Wage (Cutoff = 1) (Cutoff = 1 (Cutoff = 2) (Cutoff = 2) Intercept -.04565513 .17729723 -.0494362 .21368676 ln Teacher qualification in 1991 -.00177951 .00383909 -.00199813 .00412284 Years of Schooling in 1991 .06576312 .00830697 .06912613 .00973923 Total Rainfall .00008265 .0000173 .00007853 .0000177 Government Accountability .00246969 .02073262 .00032706 .01966279 Delta ln Employment -.73485869 .14010873 -.74297255 .15258759 ln Transport Cost SP in 1995 -.05022805 .01231244 -.04537549 .01413209 ln Transport Cost Capital in 1995 -.04676843 .01599272 -.04803535 .01860846 Delta ln Transferences .16973922 .07140593 .16184129 .08098298 Delta ln Market Potential .64660937 .19555643 .60036469 .22895648 crit. fn. test of overid. restrictions 35.984318 35.984318 22.100483 22.100483 Brazilian spatial labor dynamics Page 25 Spatial GMM Spatial GMM Dependent Variable: 2SLS Est. 2SLS SE Est. SE Delta ln Employment (Cutoff = 0.5) (Cutoff = 0.5) Intercept -1.0038159 .14846142 -.97731119 .21782738 ln Wage in 1991 .19207975 .01553089 .190244 .02302395 ln 5 to 15 over 15 to 55 yrs Ratio .25067678 .03711903 .24860755 .0545722 Proportion of Natives in 1991 .292077 .04174566 .31104589 .06760336 Delta ln Market Potential .2035026 .02778313 .19855203 .04571589 ln Teacher Qualification in 1991 -.00699189 .00239773 -.00710868 .00376014 Homicides in 1991 -9.7408304 4.5089288 -13.952994 12.869347 Employment in Farming in 1991 -.55168569 .17153628 -.62525272 .27117524 Bank Dummy -.0029771 .00978021 .00085738 .01357927 ln Population in 1991 -.02653262 .00623792 -.02801059 .009894 Mean Temperature -.00246354 .00267179 -.00177658 .00416755 Total Rainfall 3.313e-06 .00001244 7.357e-06 .00002025 Rainfall - 1º Princ. Component .00715906 .00194123 .00712434 .0032946 Rainfall - 3º Princ. Component .01638063 .0044276 .01515498 .00688489 Employment rate in 1991 .17582282 .14548762 .21597567 .23001887 crit. fn. test of overid. restrictions ---- ---- 6.5507803 6.5507803 Spatial GMM Spatial GMM Spatial GMM Spatial GMM Dependent Variable: Est. SE Est. SE Delta ln Employment (Cutoff = 1) (Cutoff = 1 (Cutoff = 2) (Cutoff = 2) Intercept -.94565964 .24317868 -.8953951 .27096564 ln Wage in 1991 .1860382 .02520692 .17956887 .02802359 ln 5 to 15 over 15 to 55 yrs Ratio .2456395 .06085758 .2379445 .06769622 Proportion of Natives in 1991 .31289083 .07724686 .30829117 .08909906 Delta ln Market Potential .19815785 .05295998 .19621347 .06306989 ln Teacher Qualification in 1991 -.00740787 .00443293 -.00709898 .00504883 Homicides in 1991 -18.404268 18.498964 -26.86664 28.508017 Employment in Farming in 1991 -.6742841 .31386852 -.72451919 .35913906 Bank Dummy .00197269 .0137091 .00300001 .01441328 ln Population in 1991 -.02758414 .01119957 -.02459399 .01372423 Mean Temperature -.00145903 .00495593 -.00197288 .00596594 Total Rainfall .00001497 .00002414 .00002119 .00002897 Rainfall - 1º Princ. Component .00662112 .00376597 .00617786 .00420715 Rainfall - 3º Princ. Component .01389925 .00844409 .01335848 .009971 Employment rate in 1991 .23755644 .25803682 .25136152 .28428775 crit. fn. test of overid. restrictions 5.2505856 5.2505856 4.0939112 4.0939112 Brazilian spatial labor dynamics Page 26 Table 10: Non-Metropolitan Brazil excluding North Region Spatial GMM Spatial GMM Dependent Variable: 2SLS Est. 2SLS SE Est. SE Delta ln Wage (Cutoff = 0.5) (Cutoff = 0.5) Intercept -.15950009 .10277873 -.12074952 .17974661 ln Teacher Qualification in 1991 .0052535 .00254925 .00236128 .00410509 Years of Schooling in 1991 .06246437 .00536942 .06088094 .0082013 Total Rainfall .00007979 .00001297 .00009786 .00002296 Government Accountability -.00584819 .01443974 -.00617702 .02409457 Delta ln Employment -.53408098 .07787127 -.63237196 .14710456 ln Transport Cost SP in 1995 -.0404624 .00821161 -.04469861 .0127028 ln Transport Cost Capital in 1995 -.04374225 .00881383 -.0422301 .01622356 Delta ln Transferences .22328359 .04042839 .18064082 .07351591 Delta ln Market Potential .58404297 .11290579 .62033021 .18144919 crit. fn. test of overid. restrictions ---- ---- 44.336521 44.336521 Spatial GMM Spatial GMM Spatial GMM Spatial GMM Dependent Variable: Est. SE Est. SE Delta ln Wage (Cutoff = 1) (Cutoff = 1 (Cutoff = 2) (Cutoff = 2) Intercept -.04462965 .19892652 .00716942 .22615147 ln Teacher Qualification in 1991 .00093057 .00426831 .00001553 .00453575 Years of Schooling in 1991 .06330124 .00894142 .06656591 .01025243 Total Rainfall .00010275 .00002402 .00010328 .00002395 Government Accountability -.01646986 .02421216 -.02936971 .02324613 Delta ln Employment -.68066132 .16307582 -.76383001 .18872485 ln Transport Cost SP in 1995 -.04688983 .01363969 -.042707 .01531044 ln Transport Cost Capital in 1995 -.04913757 .01864702 -.05431722 .02134752 Delta ln Transferences .17948617 .0803842 .15744109 .09175749 Delta ln Market Potential .58104956 .19482417 .52911681 .22650729 crit. fn. test of overid. restrictions 33.939201 33.939201 21.138334 21.138334 Brazilian spatial labor dynamics Page 27 Spatial GMM Spatial GMM Dependent Variable: 2SLS.Est. 2SLS.SE Est. SE Delta ln Employment (Cutoff = 0.5) (Cutoff = 0.5) Intercept -.95503801 .15016687 -.9338026 .21124536 ln Wage in 1991 .19092101 .0156507 .18901078 .02148466 ln 5 to 15 over 15 to 55 yrs Ratio .1983383 .03620152 .19401434 .05055868 Proportion of Natives in 1991 .28972116 .04454679 .3009288 .06903289 Delta ln Market Potential .30840242 .03091629 .30340591 .05130368 ln Teacher Qualification in 1991 -.00807114 .00235408 -.00812438 .00351447 Homicides in 1991 -6.9250641 4.325995 -8.436305 11.13018 Employment in Farming in 1991 -.46696147 .16523983 -.50415081 .25446269 Bank Dummy .00072828 .00950343 .00397452 .01257344 ln Population in 1991 -.02531787 .00636568 -.02616172 .00975625 Mean Temperature -.00678719 .00265857 -.00660142 .00409379 Total Rainfall -.00005046 .00001317 -.00004882 .00002093 Rainfall - 1º Princ. Component .00971466 .00196289 .00973935 .00324289 Rainfall - 3º Princ. Component .02245587 .00465159 .02228627 .00712442 Employment rate in 1991 .15742175 .14161712 .17703837 .2183038 crit. fn. test of overid. restrictions ---- ---- 3.819309 3.819309 Spatial GMM Spatial GMM Spatial GMM Spatial GMM Dependent Variable: Est. SE Est. SE Delta ln Employment (Cutoff = 1) (Cutoff = 1 (Cutoff = 2) (Cutoff = 2) Intercept -.91925794 .23347463 -.8959102 .25081733 ln Wage in 1991 .18708293 .02332669 .183212 .0248753 ln 5 to 15 over 15 to 55 yrs Ratio .19600923 .05612046 .19536939 .06081697 Proportion of Natives in 1991 .30085957 .07981306 .29273868 .09316006 Delta ln Market Potential .30237946 .05999995 .30237827 .06949838 ln Teacher Qualification in 1991 -.008132 .00415266 -.0078909 .0047722 Homicides in 1991 -9.6619034 15.947038 -12.576632 24.664917 Employment in Farming in 1991 -.52499872 .29569033 -.53298384 .3420592 Bank Dummy .00465381 .01249154 .00546396 .01298614 ln Population in 1991 -.02584312 .01101769 -.02346004 .01342835 Mean Temperature -.006529 .00497929 -.00703587 .00608096 Total Rainfall -.00004341 .0000258 -.00003918 .00003218 Rainfall - 1º Princ. Component .00924989 .00368997 .00853621 .00395734 Rainfall - 3º Princ. Component .02213984 .00880441 .02222459 .01032099 Employment rate in 1991 .18801478 .2473453 .18705417 .27703223 crit. fn. test of overid. restrictions 3.0782778 3.0782778 2.4566422 2.4566422 Brazilian spatial labor dynamics Page 28 Labor Market Dynamics E+, W+ E-, W+ E-, W- E+, W- N W E S 900 0 900 1800 Miles Figure 1 Labor dynamics, Brazil Categorization of MCA by its labor dynamics Real wage change E-W+ E+W+ Employment change relative to population growth E-W- E+W- Brazilian spatial labor dynamics Page 29 PopulationDensity 2000 0.131- 3.816 3.816- 12.225 12.225 - 85.984 85.984 - 12878.968 N W E 1000 0 1000 2000 Miles S Figure 2 Population density, 2000 Brazilian spatial labor dynamics Page 30 Population distribution by region and labor market outcome 80% 70% 60% 50% 40% 30% 20% 10% 0% N NE SE S CW E+, W+ E-, W+ E-, W- E+, W- Figure 3 Population distribution by region and labor market outcome Brazilian spatial labor dynamics Page 31 Figure 4 Worker education vs. wage growth Red ­ North and Northeast Regions. Blue ­ Southeast, South and Center-West Regions. Brazilian spatial labor dynamics Page 32 Employment Growth. Brazil. 1991/2000. Variação no emprego - 91/00 -8508.163 - 0 0 - 7000 7000 - 21000 21000 - 215865.8 N W E 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 Kilometers S Map 1 Brazilian spatial labor dynamics Page 33 Salary Growth Salary-1.524 --0.5 Growth- 91/00 -0.5 - 0 00.3- 0.3 - 0.85 N W E 1000 0 1000 2000 Miles S Map 2 Salary. Brazil. 2000. Salário - 00 325.614 - 2500 2500 - 5000 5000 - 7500 7500 - 16701.68 N W E 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 Kilometers S Map 3 Brazilian spatial labor dynamics Page 34 Years of study growth - 1991/2000 Years of study growth -91/00 -3.739- -0.5 -0.5- 0 0- 0.3 0.3 - 1.245 N W E 1000 0 1000 2000 Miles S Map 4 Years of Study. Brazil. 2000. Anos de estudo - 00 1 - 4 468 -- 68 - 10.493 N W E 1000 0 1000 2000 Miles S Map 5 Brazilian spatial labor dynamics Page 35 Teacher Qualification Growth. Brazil. 1991/2000. Crescimento da qualificação dos professores - 91/00 -7.044 - 0 01.5- 1.5 - 3 3 - 8.056 N W E 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 Kilometers S Map 6 Percentage of income from government transferences. Brazil. 2000 Percentage of income from government transferences - 2000 2 - 13 14 - 18 19 - 23 24 - 35 N W E 1000 0 1000 2000 Miles S Map 7 Brazilian spatial labor dynamics Page 36 Changes in transport costs to capital. 1995/1980 Changes in transport cost to capital -95/80 -0.916 - -0.5 -0.5 - -0.259 -0.259 - -0.098 -0.098 - 0 N W E 1000 0 1000 2000 Miles S Map 8 Change in transport cost to SP. 1995/1980 Changes in transport cost toSP- 95/80 -0.916- -0.331 -0.331- -0.197 -0.197- -0.087 -0.087- 0 N W E 1000 0 1000 2000 Miles S Map 9 Brazilian spatial labor dynamics Page 37 Transport Cost to São Paulo. Brazil. 1995. Custodetransportepara São Paulo- 1995 10- 500 500- 1500 1500 -2500 2500 -5000 5000 -10512 N W E 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 Kilometers S Transport Cost to Capital. Brazil. 1995. Custo0300- de-transporte àcapital - 1995 300600 600- 1000 1000 - 3000 3000 - 5949 N W E 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 Kilometers S Map 10 Brazilian spatial labor dynamics Page 38 Government accountability in computers. 1999 Governmentaccountability in computers 01 N W E 1000 0 1000 2000 Miles S Map 11 Market Potential. Brazil. 2000. Market0100,000 - 2000 Potential 1,000,000- -50,000,000 - 100,0001,000,000 50,000,000-994,320,704 N W E 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 Kilometers S Map 12