Human Capital and Earnings Inequality in Brazil, 1988­1998: Quintile Regression Evidence G. Reza Arabsheibani, Francisco Galrão Carneiro,* and Andrew Henley Abstract This paper undertakes an empirical examination of rates of return to human capital for men in Brazil through the period of macroeconomic stabilization and trade liberalization, using data from the PNAD household surveys. Simultaneous quintile equations are estimated to gain a picture of the impact of human capital on earnings across the hourly earnings distribution. We conclude that there is evidence for growing inequality in rates of return to education in Brazil. However we find evidence that education is no longer used as a screening device in the labor market, but rather rewarded for its innate association with higher productivity. Although increases in rates of return to education have been more pronounced at the top of the earnings distribution, this has not led to increased inequality. This is because levels of education and other labor market-rewarded endowments have increased and offset the rate of return effect. Keywords: Earnings, Human Capital, Inequality, Quintile Regression JEL Classification: J31, I20, C14 School of Management and Business, University of Wales Aberystwyth, UK. *The World Bank ­ Poverty Reduction and Economic Management (PREM). World Bank Policy Research Working Paper 3147, October 2003 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. 1 1. Introduction A far-reaching program of macroeconomic stabilization and rapid liberalization of trade in a developing economy are likely to have important and widespread consequences for the labor market. However there appears to be little evidence from a number of recent examples for the Stolpher-Samuelson theorem that trade liberalization in a developing economy will increase the relative demand for relatively abundant low-skilled labor, and so reduce income inequality.1 The Brazilian case has been particularly well-researched and there is an emerging conventional wisdom that, despite the upheavals of price and exchange rate stabilization and tariff reduction, earnings inequality in Brazil has remained stubbornly high between 1980 and the present (Dickerson et al., 2001; Green et al., 2001). One explanation for this is that steadily rising average levels of schooling have offset increased demand for skilled labor. At first glance this is a persuasive suggestion. In the sample of Brazilian male workers used in the present paper, average years of education for men have risen by almost one year between 1988 and 1998, and the rate of illiteracy has fallen from 16.6 to 11.6 per cent. Research has also established that rates of return to education in Brazil are low for primary schooling but very much higher for advanced levels of education. This may contribute significantly to Brazil's highly unequal distribution of earnings (Lam and Levinson, 1992). However while this may be true at the mean of the earnings distribution, little is known about rates of return across the range of this wide distribution. Furthermore, there is evidence that inequality peaked in the late 1980s and has been falling since, in contrast to what has happened elsewhere in the developing world and in particular in Latin America. If so this points to the possibility that important changes in the levels of and rates of return to education have taken place since the late 1980s. Consequently the reform experience may have been a very different one for workers in different positions in the income distribution. Protected product markets in the developing world prior to the early 1990s appear to have been associated with protected labor markets in Brazil in particular (Carneiro and Henley, 1998; Carneiro, 1998). Liberalization of the economy to overseas trade exposes domestic producers significantly to greater product market competition and this in turn serves to introduce much stronger forces of competition to the labor market. An important consequence of this process is that education ceases to serve as a device for rationing (screening) workers' access to economic rents. Rather education, if it reflects the acquisition of internationally marketable skills, begins to be associated with inherent productive potential. Whether the benefits of this process accrue more to those with lower or higher levels of education is a matter for empirical investigation. In common with other Latin American cases, such as Mexico, there is evidence that tariff protection was highest in sectors where employment of the least skilled was dominant (Mollick, 2002; Arbache and Corseuil, 2001). Overall there appears to have been little or no assessment of whether apparently high rates of return to formal education represent a genuine return to valuable skills acquired at school, or simply reflect the ability of the more educated to signal innately higher productivity to the labor market. This paper undertakes an empirical examination of these questions through the estimation of simultaneous quintile human capital equations. This is in order to gain a picture of the impact of human capital on earnings across different point of the distribution of earnings in Brazil. This exercise is performed a sample of male workers drawn from 1 See discussion and further references in Green et al., 2001. 2 household survey data from before, during and after the stabilization and liberalization program of the early 1990s. We conclude that there is evidence for growing inequality in rates of return to education in Brazil, but evidence that education is no longer used as a screening device in the labor market, but rather rewarded for its innate association with higher productivity. Although increases in rates of return to education have been more pronounced at the top of the earnings distribution, this has not led to increased inequality. This is because levels of education and other labor market-rewarded endowments have increased and offset the rate of return effect. The remainder of the paper is structured as follows. Section 2 discusses the empirical method used. Section 3 describes the data source used. Section 4 presents results and section 5 concludes. 2. Empirical Approach An important issue on the productivity of education concerns whether formal education acts as a screen, separating more able (and educated) individuals from the less able (and educated). The screening hypothesis (Arrow, 1973) observes that at the point of hiring workers' productivity is unknown to employers and argues therefore that employers use education as a proxy for latent productivity. In competitive sectors of the labor market productivity will matter and so returns to education will be higher. In non-competitive sectors of the labor market returns to subsequent education after hiring will be lower. It is therefore possible that the value of education as a screen may vary across the earnings distribution because of differing degrees of competition. In particular screening may be more important in the top of the distribution, where insider power may be more important. The empirical literature on screening distinguishes between the weak form and strong form of the hypothesis (Psacharopoulos, 1979; Arabsheibani and Rees, 1998). The weak form states that employers will pay a higher initial salary to recruits with higher levels of education, but is agnostic about the shape of the subsequent experience-earnings profile. The strong form states that employers will continue to pay high salaries even after observing working on the job, because education continues to enhance productivity as experience on the job rises. However the experience-earnings profiles of an educated worker will converge over time with that of a non-educated worker, as the original hiring "mistake" is gradually corrected. Psacharopoulos (1979) proposes what has become known as the P(sacharopoulos) test as a method of empirical investigation. Assume that log hourly earnings for individual i, yi , are determined according a Mincerian earnings function of the following form: yi = a0 + a1Si + a2Si + a3Ei + a4Ei + a5SiEi + bZi + ui 2 2 (1) where S is years of education, E is years of experience, Z are other socio-economic variables affecting earnings, aj and b are coefficients and u is a disturbance term. The inclusion of the interaction term between years of education and years of experience provides a straightforward test of convergent experience-earnings profiles under the strong screening hypothesis (Lee, 1980). If the hypothesis holds then a5 < 0, otherwise a5 0. Previous research has shown that modeling average earnings (i.e. OLS) fails to reveal that the effect of education on earnings is non-constant across the conditional wage 3 distribution (Buchinsky, 1994, 1998; Machado and Mata, 2001; Bauer and Haisken-DeNew, 2001; Hartog, Pereira and Vieira, 2001). This reinforces the need to investigate the screening hypothesis across the earnings distribution. An appropriate empirical strategy is to fit the earnings model across different points in the conditional sample distribution, using the quintile regression method. This was first introduced by Koenker and Bassett (1978). Assume yi , i = 1,...,n, is a sample of observations on log earnings, and that Xi is a K x 1 vector comprising the education, experience and other control characteristics contained on the right- hand side of equation (1). The quintile regression model can be expressed as: yi = Xi + ui, Quant (yi | Xi ) = Xi , (0,1) (2) where Quint(yi | Xi) denotes the quintile of log earnings conditional on the regressor vector. Following Koenker and Bassett (1978), the regression quintile can be defined as the solution to the problem: 1 1 n min yi - xi + ( 1- ) yi - xi = min ( ui ) (3) n i:yixi i:yi