77478 Trade Liberalization and Industry Wage Structure: Evidence from Brazil Nina Pavcnik, Andreas Blom, Pinelopi Goldberg, and Norbert Schady Industry affiliation provides an important channel through which trade liberalization can affect worker earnings and wage inequality between skilled and unskilled workers. This empirical study of the impact of the 1988–94 trade liberalization in Brazil on the industry wage structure suggests that although industry affiliation is an important component of worker earnings, the structure of industry wage premiums is relatively stable over time. There is no statistical association between changes in industry wage premiums and changes in trade policy or between industry-specific skill premiums to university graduates and trade policy. Thus trade liberalization in Brazil did not significantly contribute to increased wage inequality between skilled and unskilled workers through changes in industry wage premiums. The difference between these results and those obtained for other countries (such as Colombia and Mexico) provides fruitful ground for studying the conditions under which trade reforms do not have an adverse effect on industry wage differentials. Policymakers often promote trade liberalization and openness as a way to increase living standards and welfare in developing economies.1 From 1988 to 1994, Brazil, like many Latin American economies, followed these policy recommendations. The reforms not only reduced the average tariff level from about 60 percent in 1987 to 15 percent in 1998 but also changed the structure of protection across industries. These drastic tariff reductions were mirrored in increased import penetration in most sectors. Nina Pavcnik is assistant professor at Dartmouth College and faculty research fellow at the National Bureau of Economic Research; her e-mail address is nina.pavcnik@dartmouth.edu. Andreas Blom is education economist in the Latin American Region at the World Bank; his e-mail address is ablom@worldbank.org. Pinelopi Goldberg is professor at Yale University and research associate at the National Bureau of Economic Research; her e-mail address is penny.goldberg@yale.edu. Norbert Schady is senior economist in the Devel- opment Research Group at the World Bank; his e-mail address is nschady@worldbank.org. The authors thank Eric Edmonds, Carolina Sanchez-Paramo, seminar participants at Princeton and the 2002 Latin American and Caribbean Economic Association meetings, three anonymous referees, and two editors for thoughtful comments and suggestion. The authors are grateful to Marcello Olarreaga for providing Merco- sur trade data. 1. Although the theoretical relationship between free trade and welfare is ambiguous, careful empirical work based on cross-country data by Frankel and Romer (1999) confirms that countries with higher exposure to trade have higher living standards, as measured by per capita GDP. THE WORLD BANK ECONOMIC REVIEW, VOL. 18, NO. 3, Ó The International Bank for Reconstruction and Development / THE WORLD BANK 2004; all rights reserved. doi:10.1093/wber/lhh045 18:319–344 319 320 THE WORLD BANK ECONOMIC REVIEW, VOL. 18, NO. 3 Although empirical studies have documented that the trade reforms increased efficiency and growth (Hay 2001; Muendler 2002), the reforms might have also contributed to growing wage inequality. Several studies document growing returns to educated workers in Brazil that coincide with the timing of trade liberalization (Behrman and others 2000; Blom and others 2001; Green and others 2001; Sa ´nchez-Pa ´ramo and Schady 2003).2 Most of this literature con- centrates on the effects of trade on the returns to particular worker character- istics (such as skill) in the long run, when labor can move across sectors and industry affiliation does not matter. This article takes a different approach. It investigates the relationship between trade liberalization and industry wage premiums. Wage premiums represent the portion of worker wages that cannot be explained by character- istics of workers or firms but are attributed to a worker’s industry affiliation. Understanding this relationship is important for several reasons. First, indus- try affiliation is crucial in predicting the impact of trade reforms on workers’ wages in short- and medium-run models of trade and in models with imperfect competition and rent sharing. Studies that do not consider industry affiliation may thus miss an important channel of trade policy effects on wage distribution. These models seem a priori particularly relevant in Latin America, where labor market restrictions that can obstruct labor mobility across sectors are common (Heckman and Pages 2000) and where domestic industries are often shielded from foreign competition, giving rise to market power and industry rents. Second, the effect of trade policy on industry wage premiums has implications for wage inequality between skilled and unskilled workers. Because different industries employ different proportions of skilled and unskilled workers, changes in industry wage premiums translate into changes in the relative incomes of skilled and unskilled workers. If tariff reductions are proportionately larger in sectors employing unskilled workers, and if these sectors experience a decline in their relative wages as a result of trade liberalization, these unskilled workers will experience a decline in their relative incomes. This effect is distinct from the potential effect of trade liberalization on the economywide skill premium. Moreover, industry wage premiums might vary across workers with different levels of skill or education. For example, the more educated workers may be more or less mobile in the labor market, have accumulated more sector-specific human capital, or have more bargaining power over industry rents. If wage premiums differ across workers with different levels of education, and if trade liberalization increases industry-specific skill premiums, this could provide an additional channel through which reforms could affect wage inequality. Very 2. Rising skill premiums have been documented in Mexico and many other liberalizing Latin American economies (see Robbins 1996; Cragg and Epelbaum 1996; Harrison and Hanson 1999; Robertson 2000b; Behrman and others 2000; and Attanasio and others 2004). Pavcnik and others 321 few studies focus on the relationship between trade policy and industry wage premiums.3 Those that do have yielded mixed conclusions and except for Goldberg and Pavcnik (forthcoming) do not consider the implications of indus- try wage premiums for wage inequality between skilled and unskilled workers. This article empirically addresses the relationship between trade policy and industry wage premiums by combining detailed worker-level information from the Brazilian Monthly Employment Survey (PME) with industry-level data on tariffs, import penetration, and export exposure during 1987–98, which includes the Brazilian trade liberalization episode of 1988–94. The analysis finds no association between trade reforms and industry wage premiums. Although industry affiliation does play a role in determining workers’ earnings, accounting for 4–6 percent of the explained variation in log hourly wages, and although industry wage premiums vary widely across industries, the structure of industry wage differentials is very stable and is not affected by the changing structure of trade protection. Moreover, no statistical relationship was found between sector-specific skill premiums (measured by the return to a completed university education) and tariff reductions. Overall, the analysis concludes that trade reform in Brazil did not contribute to wage inequality between skilled and unskilled workers through differential changes in industry wage premiums or through increases in industry-specific skill premiums. I. THEORETICAL BACKGROUND Trade theory predicts how trade policy might affect industry wage premiums. In short- and medium-run models of trade, where labor is immobile across sectors and industries are perfectly competitive, workers’ wages depend on product prices and the marginal product of labor in an industry. The models predict a positive association between industry tariffs and wages, so that declines in industry tariffs lead to proportional declines in industry wages.4 These predic- tions are consistent with the popular belief that trade liberalization will make workers in previously protected sectors worse off. Models with imperfectly competitive product and labor markets provide addi- tional mechanisms through which industry tariffs affect industry wages. For example, in profitable industries unions might be able to bargain over industry rents and secure higher wages. Because trade liberalization likely lowers the profit margins of domestic firms that were previously sheltered from foreign competition 3. Revenga (1997), Gaston and Trefler (1994), Feliciano (2001), Robertson (2000a), Goldberg and Pavcnik (forthcoming), and Arbache and Menezes-Filho (2000) are examples of related work. Arbache and Menezes-Filho (2000) find significant evidence of rent sharing during trade liberalization in Brazilian manufacturing during 1989–95 when they instrument for value added with the effective tariffs. 4. In contrast, the long-run Hecksher-Ohlin model predicts that trade reform should affect only economywide returns to the factors of production but not industry-specific returns, because all factors of productions are mobile across uses. 322 THE WORLD BANK ECONOMIC REVIEW, VOL. 18, NO. 3 (Harrison 1994; Levinsohn 1993), lower tariffs are associated with lower industry wages. Grossman (1984) presents a model in which unions extract the rents associated with protection in the form of employment guarantees rather than wages. This channel implies a potentially negative association between tariffs and industry wages. Finally, trade liberalization might affect industry wages through trade-induced productivity improvements. Although trade theory does not yield clear-cut predic- tions on whether trade liberalization increases or decreases productivity (Rodrik 1991; Roberts and Tybout 1996; Melitz 2003), empirical work finds strong evidence that declines in tariffs are associated with productivity improvements (Harrison 1994; Krishna and Mitra 1998; Kim 2000; Pavcnik 2002; Fernandes 2001). Hay (2001) and Muendler (2002) estimate that the 1988–94 trade reforms had a significant impact on plant-level productivity in Brazil. As tariffs declined, firms had to become more productive to remain competitive. If the productivity enhancements were partially passed onto workers through higher industry wages, wages would increase in the industries with the largest tariff declines. Thus although industry affiliation provides an important channel through which trade policy can affect workers’ wages, these models do not yield unam- biguous predictions about the direction of the expected effect of trade liberal- ization on wages. The question is one that needs to be resolved empirically. II. METHODOLOGY AND DATA A two-stage estimation framework, familiar from the labor literature on indus- try wages, was used to empirically investigate the effect of trade exposure to wage premiums. In the first stage the log of worker i’s wages (wijt) was regressed on a vector of worker i’s characteristics (Hijt), such as education, age, age squared, gender, geographic location; an indicator for whether the person is self-employed; an indicator for whether the person works in the informal sector; and a set of industry indicators (Iijt) reflecting worker i’s industry affiliation: ð1Þ lnðwijt Þ = Hijt bHt þ Iijt à wpjt þ Eijt The coefficient on the industry dummy, the wage premium, captures the part of the variation in wages that cannot be explained by workers’ characteristics but can be explained by workers’ industry affiliation. Following Krueger and Sum- mers (1988), the estimated wage premiums are expressed as deviations from the employment-weighted average wage premium.5 This normalized wage premium can be interpreted as the proportional difference in wages for a worker in a given industry relative to an average worker in all industries with the same observable characteristics. The normalized wage differentials and their exact standard errors are calculated using the Haisken-DeNew and Schmidt (1997) 5. The sum of the employment-weighted normalized wage premiums is zero. Pavcnik and others 323 two-step restricted least squares procedure provided by the authors.6 The first- stage regressions are estimated separately for each year in the sample, as the subscript t in equation 1 indicates. The second stage pools the industry wage premiums (wpjt) over time and regresses them on trade-related industry char- acteristics in first differenced form: ð2Þ �wpjt ¼ �Tjt bT þ Dt bD þ ujt The primary variable included in Tjt, the vector of trade-related industry charac- teristics, is tariffs. Other controls in Tjt are also considered, such as lagged import penetration, lagged export to output share, and interactions of these variables with exchange rates. The vector Dt consists of a set of year indicators. Because the dependent variable in the second stage is estimated, equation 2 is estimated using weighted least squares, with the inverse of the standard error of the wage premium estimates from the first stage as weights. This procedure gives more weight to industries with smaller variance in industry premiums. To account for general forms of heteroscedasticity and serial correlation in the error term in equation 2, robust (Huber-White) standard errors were computed, clustered by industry. Labor Force Data The labor market data from the pme for 1987–1998 cover the six largest metropolitan areas in Brazil: Sa ˜ o Paulo, Rio de Janeiro, Porto Alegre, Belo Horizonte, Recife, and Salvador. These areas account for about 31.9 million of the country’s 79 million people in the economically active population. More- over, in 1997 the states in which the six metropolitan areas are located pro- duced 72 percent of Brazilian GNP.7 The findings are thus representative of the large and modern parts of the Brazilian labor market but do not necessarily carry over to the rural economy. Because the focus is on manufacturing, how- ever, this might not be problematic. Data were collected on workers affiliated with any of 18 manufacturing and 2 mining sectors and covered employees or self-employed workers ages 15–65 engaged in full-time work (defined as working more than 25 hours a week). The data were used to create several variables capturing worker demographic char- acteristics, such as wage, age, education, geographical location, employment in the informal sector, self-employment, and industry affiliation. The wage measure is the hourly wage (one-quarter of the monthly wage times the reported number of hours worked per week), deflated by the monthly national price index. All wages are expressed in September 1997 R$. The 6. Haisken-DeNew and Schmidt (1997) adjust the variance covariance matrix of the normalized industry indicators to yield an exact standard error for the normalized coefficients. 7. Brazilian gnp was 864,112 million reals (R$) and the six states (Sa˜ o Paulo, Rio de Janeiro, Rio Grande do Sul, Minais Gerais, Pernambuco, and Bahia) accounted for R$618,728 million, according to Brazilian Institute of Geography and Statistics accounts of gross regional products in current market prices. 324 THE WORLD BANK ECONOMIC REVIEW, VOL. 18, NO. 3 main education indicator is completed years of schooling, computed using an algorithm based on three survey questions on education.8 Workers are classified as those with no completed level of education, completed elementary education, completed lower secondary education, completed secondary education, and completed tertiary education.9 Formal and informal sector workers are distin- guished by whether they had a signed workcard (carteira assinada). Because a signed workcard legally entitles a worker to several rights and benefits, it can be used to identify whether a person works for a formal establishment that com- plies with labor market regulations. The variable ‘‘informal workers’’ takes a value of one if the worker is employed in the informal sector of the economy. Trade Exposure Data Until the 1980s Brazil pursued an import substitution policy to shield domestic firms from foreign competition. High tariffs and a large number of nontariff barriers provided high levels of protection to Brazilian firms and severely impeded access of foreign goods to the Brazilian market. Protection varied widely across industries, with tariffs ranging from more than 100 percent on clothing, the most protected sector, and 82–86 percent on textiles and rubber to almost 16 percent on oil (table 1). This suggests that Brazil strongly protected relatively unskilled, labor-intensive sectors, which conforms to a finding by Harrison and Hanson (1999) for Mexico and Goldberg and Pavcnik (forth- coming) for Colombia. From 1988 to 1994 Brazil underwent significant, if gradual, trade liberal- ization. In 1988 and 1989 the average tariff was reduced from about 60 to 39 percent. Kume (2000) and Hay (2001) argue that this initial reduction had no significant bearing on the exposure of domestic industries to increased foreign competition because substantial nontariff barriers remained, including import licenses, special import programs, and administrative barriers. These were eliminated in the second stage of the reforms that started in 1990 as the Collor government sought to improve productivity by exposing domestic firms to increased foreign competition.10 At the same time average tariffs were further reduced, from 34 percent in 1990 to 11 percent tariff in 1995. In 1995 the government partially reversed these trade reforms following real appreciation of the real that lowered the competitiveness of the manufacturing sector and widened the current account deficit. Nevertheless, the average tariff climbed only slightly between 1995 and 1998. In addition to the unilateral trade liberalization that took place from 1988 to 1994, Brazil joined Mercosur, a 8. The algorithm follows the standard conversion used elsewhere (see Lam and Schonie 1993; Barros and Ramos 1996). 9. Primary education in Brazil consists of four years of schooling. Secondary education (ensino medio) comprises two parts, 5–8 years of schooling and 9–11 years of schooling. Tertiary education includes 12–15 or more years of schooling. 10. Detailed information on nontariff barriers is not available. Pavcnik and others 325 T A B L E 1 . Industry Tariffs and Correlation of Industry Import Penetration and Tariffs for Brazil Tariff (%) Correlation with Import Penetrationa Industry 1986 1998 Current Tariff Lagged Tarriff Mineral extraction 20.5 6.4 À.88 À.69 Oil extraction 15.6 0.0 .73 .75 Nonmetalic mineral trasformation 63.7 13.7 À.66 À.73 Metalic products and steel 32.5 11.2 À.44 À.46 Machinery and equipment 47.0 17.2 À.80 À.83 Electrical and electronic equipment 59.8 18.8 À.91 À.91 Transportation vehicles 77.1 32.5 À.65 À.66 Wood and furniture 50.0 14.0 À.51 À.62 Paper, pulp, and cardboard 59.5 14.2 À.61 À.68 Rubber products 82.0 15.0 À.74 À.79 Chemicals 59.9 16.3 À.53 À.52 Petrochemicals 32.5 10.0 À.87 À.95 Pharmaceuticals 72.3 10.7 À.83 À.85 Plastics 36.6 18.1 À.74 À.82 Textiles 85.8 19.0 À.83 À.89 Clothing 102.7 22.8 À.71 À.79 Footwear 74.1 17.9 À.85 À.89 Tobacco 62.5 14.3 À.71 À.74 Foods 60.3 16.0 À.60 À.63 Beverages 80.5 19.0 À.69 À.78 a Import penetration refers to imports as a percentage of output plus net imports. Source: Authors’ calculations based on tariff data from Muendler (2002) (http://socrates. berkeley.edu/$muendler), which draws on Kume and others (2000). regional trading bloc also comprising Argentina, Paraguay, and Uruguay, in 1991. Although the focus here is the impact of the unilateral trade liberalization, Brazil’s tariffs on Mercosur imports and its trade within Mercosur are used to check the robustness of the findings. Brazil’s trade liberalization provides an excellent setting to study the relation- ship between wages and trade. From 1987 to 1998 the average tariff across 20 industrial sectors (table 1) declined from 58.8 percent to 15.4 percent (table 2).11 The reforms also changed the structure of protection across industries, as different industries experienced different rates of tariff changes, and tariff dispersion declined significantly. The changing structure of protection is reflected in the low year-to-year correlations of industry tariffs from 1987 to 1998. For example, the correlation coefficient between tariffs in 1987, a year preceding the trade reforms, 11. The original tariff data provide the tariff levels for 53 sectors at the level (nivel) 80 Brazilian industrial classification. So that the tariff information corresponds to the level of industry aggregation in the labor force data, the data were aggregated here to level 50, and some additional adjustments were made. The reported tariffs are simple averages of more disaggregated data. The tariff series was also constructed using level 80 import penetration as weights, which yielded similar aggregate means; the correlation coefficient between the two series was 0.98. Thus simple average tariffs were used for the study. 326 THE WORLD BANK ECONOMIC REVIEW, VOL. 18, NO. 3 T A B L E 2 . Trade Policy and Trade Exposure 1987–98 (percent) Tariffs Import Penetrationa Export Exposureb Year Mean SD Mean SD Mean SD 1987 58.8 22.8 5.7 8.6 9.7 11.2 1988 50.1 18.3 5.9 8.5 9.5 11.3 1989 39.1 16.4 6.1 8.4 9.4 11.5 1990 34.1 17.0 6.4 8.4 9.2 11.6 1991 25.2 13.3 7.6 8.6 10.9 12.4 1992 19.1 10.3 7.7 8.8 13.4 13.6 1993 14.4 7.2 8.0 8.4 13.0 13.2 1994 12.9 6.2 8.6 8.3 11.5 11.2 1995 10.9 5.7 9.8 8.1 11.0 10.8 1996 12.5 6.6 9.8 8.1 11.4 11.8 1997 12.8 7.0 10.6 8.3 11.7 12.2 1998 15.4 6.5 11.6 7.8 11.2 10.1 Note: The values cover 20 industries, except for 1998, which covers 18 for import penetration and export exposure. a Imports as a percentage of output plus net imports. b Exports as a percentage of output. Source: Muendler (2002). For tariffs Muendler draws on Kume and others (2000). and tariffs in 1989 is 0.81. The correlation coefficient between tariffs in 1987 and 1995, the year after the large reforms were completed, drops to 0.6. The vast variation in Brazilian tariffs across industries at a given time and across time provides an excellent setting to study the relationship between trade and wages. The shifts in Brazil’s trading environment are mirrored in the increased import penetration rate (imports as a percentage of output plus net imports) and export exposure (exports as a percentage of output).12 From 1987 to 1998 average import penetration increased from 5.7 percent to 11.6 percent and average export expo- sure from 9.7 percent to 11.2 percent. Whereas import penetration almost doubled, it continues to be low compared with a country such as Colombia, which liberalized during the same period. Colombia’s manufacturing import pene- tration rate was about 21 percent in 1984 and exceeded 30 percent after the 1990 tariff reductions (Goldberg and Pavcnik forthcoming). This difference could be attributed to the large size of Brazil relative to Colombia. Moreover, increases in import penetration rates in Brazil varied significantly across sectors. Industries with the largest gains are clothing, transport, textiles, machinery, electronics, and 12. Data on import penetration and export exposure were obtained from Muendler (2002) at online at http://socrates.berkeley.edu/$muendler. The data were adjusted so that the trade exposure information corresponds to the level of industry aggregation in the labor force data. The industry-level trade exposure measures used were weighted by the import penetration of the less disaggregated level 80 industry data. The correlation between the weighted import penetration series and the import penetration series based on simple averages is 0.99. Similarly, the correlation between the weighted export exposure series and the export exposure series based on simple averages is 0.99. Pavcnik and others 327 pharmaceuticals—also the industries that experienced large tariff declines. Corre- lation coefficients for import penetration and tariffs (and lagged tariffs) over time show, unsurprisingly, that imports and tariffs are negatively correlated (oil extrac- tion is an exception), ranging from À0.4 in steel to À0.9 in electrical and electronic equipment. The correlation increases in absolute value for lagged tariffs. III. INDUSTRY WAGE PREMIUMS AND TRADE POLICY: RESULTS Before exploring whether trade liberalization affected industry wage premiums, results are presented for the first-stage regressions of equation 1. First-Stage Results The first-stage results show, as in previous work, that several characteristics are associated with higher wages: age, being male, education, being self-employed, and working in the formal sector (table 3). The results also show that workers experience changes in returns to education over time. A noteworthy change is the decline in the wages of workers with secondary education relative to the wages of less skilled workers (no education or completed elementary) only and more skilled workers (complete tertiary education).13 Industry affiliation plays a material role in explaining the variation in log hourly earnings. For example, in 1987 worker characteristics and regional indicators alone account for 50 percent of the total variation in log hourly wages. The addition of industry indicators to the regression increases R2 to 0.52, which suggests that, conditional on other worker characteristics, industry indicators account for 4 percent of the explained variation in log hourly wages in 1987 (see table 3). In general, industry indicators account for 4–6 percent of the explained variation in log hourly wages between 1987 and 1998. Industry wage premiums vary widely across industries (table 4). The estimates for 1987, for example, range from 0.55 for the petrochemical industry to À0.20 for foods. A worker in 1987 with the same observable characteristics who switched from the textile industry, where the wage premium is À0.079, to the chemical industry, where the wage premium is 0.168, would experience a 25 percent increase in hourly wages. The standard deviations of the industry wage differentials reported at the bottom of table 4 summarize the overall variability of the industry wage premiums. The variation in industry wage differentials in a given year ranges from 13 percent to 16 percent, implying that changing industries has a large impact on worker earnings. The variation is largest in the period 1992 to 1994. 13. Some of this decline is presumably related to the increasing number of workers with a secondary education relative to the number with a primary or a university education (see Sa ´nchez-Pa ´ramo and Schady 2003). The increases in the returns to a university education are not confined to workers in manufacturing industries or to urban areas. Blom and others (2001) find similar patterns in the returns to education for workers in traded and nontraded industries. Green and others (2001) also document rising skill premiums using data from National Household Surveys that cover rural and urban areas. T A B L E 3 . First-Stage Regression Results for Worker Characteristics and Industry Indicators, 1987–98 Variable 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Age 0.067** 0.065** 0.064** 0.063** 0.059** 0.052** 0.054** 0.061** 0.056** 0.059** 0.059** 0.058** (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.003) Age squared À0.001** À0.001** À0.001** À0.001** À0.001** À0.001** À0.001** À0.001** À0.001** À0.001** À0.001** À0.001** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Female À0.452** À0.440** À0.462** À0.450** À0.424** À0.458** À0.430** À0.442** À0.438** À0.392** À0.384** À0.387** (0.010) (0.010) (0.011) (0.012) (0.011) (0.013) (0.013) (0.013) (0.012) (0.012) (0.012) (0.018) Elementary 0.268** 0.258** 0.252** 0.251** 0.227** 0.220** 0.219** 0.183** 0.190** 0.183** 0.202** 0.187** education (0.006) (0.006) (0.007) (0.008) (0.007) (0.008) (0.009) (0.009) (0.008) (0.008) (0.009) (0.013) Lower secondary 0.572** 0.551** 0.542** 0.523** 0.484** 0.452** 0.442** 0.430** 0.425** 0.421** 0.436** 0.421** education (0.008) (0.008) (0.009) (0.009) (0.009) (0.010) (0.010) (0.010) (0.010) (0.010) (0.010) (0.015) Upper secondary 1.079** 1.047** 1.051** 1.035** 0.951** 0.931** 0.922** 0.933** 0.906** 0.868** 0.867** 0.843** education (0.008) (0.008) (0.009) (0.010) (0.009) (0.010) (0.011) (0.011) (0.010) (0.010) (0.010) (0.015) Tertiary 1.823** 1.862** 1.880** 1.897** 1.831** 1.762** 1.795** 1.806** 1.778** 1.804** 1.766** 1.725** education (0.010) (0.011) (0.012) (0.013) (0.012) (0.014) (0.014) (0.015) (0.014) (0.014) (0.014) (0.021) Self-employed 0.091** 0.099** 0.119** 0.148** 0.097** 0.021 0.044** 0.078** 0.137** 0.072** 0.069** 0.074** 328 (0.016) (0.016) (0.018) (0.018) (0.016) (0.017) (0.017) (0.017) (0.016) (0.015) (0.015) (0.022) Informal À0.162** À0.238** À0.220** À0.162** À0.158** À0.265** À0.254** À0.205** À0.136** À0.124** À0.130** À0.165** (0.010) (0.010) (0.012) (0.012) (0.011) (0.012) (0.011) (0.011) (0.011) (0.010) (0.010) (0.015) R2 0.52 0.54 0.52 0.52 0.53 0.51 0.51 0.5 0.52 0.53 0.53 0.52 R2 without 0.50 0.52 0.50 0.50 0.51 0.48 0.48 0.47 0.50 0.51 0.50 0.49 industry indicators Variation .04 .04 .04 .04 .04 .06 .06 .06 .04 .04 .06 .06 attributed to industry indicators Number of 65,455 58,659 48,881 47,983 44,818 38,447 36,720 38,080 37,159 34,933 34,122 16,307 observations **Significant at the 5 percent level. Note: Numbers in parentheses are standard errors. All regressions include industry indicators and regional indicators. Source: Authors’ calculations based on data from Brazil’s PME. T A B L E 4 . Industry Wage Premiums, 1987–98 Industry 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Mineral extraction .238 .216 .115 .109 .142 .189 .166 .164 .037 .178 .269 .146 (.023) (.024) (.027) (.028) (.026) (.029) (.030) (.031) (.032) (.029) (.030) (.042) Oil extraction .092 .003 .036 .071 .102 .085 .089 .048 .014 .079 .094 .124 (.019) (.020) (.025) (.026) (.026) (.026) (.030) (.030) (.030) (.031) (.029) (.044) Nonmetalic mineral À.137 À.096 À.083 À.155 À.135 À.090 À.118 À.128 À.115 À.106 À.077 À.135 trasformation (.010) (.011) (.012) (.012) (.012) (.013) (.014) (.014) (.014) (.014) (.015) (.021) Metalic products and steel .021 .021 .027 .022 .012 .022 .001 .010 .016 À.010 À.009 .001 (.005) (.006) (.006) (.007) (.006) (.007) (.007) (.007) (.007) (.007) (.007) (.010) Machinery and equipment .129 .114 .083 .141 .110 .111 .093 .095 .103 .136 .091 .149 (.008) (.009) (.010) (.011) (.011) (.012) (.013) (.014) (.012) (.013) (.013) (.019) Electrical and electronic .051 .095 .105 .062 .085 .089 .104 .147 .088 .109 .079 .089 329 equipment (.009) (.010) (.011) (.011) (.011) (.014) (.015) (.015) (.015) (.015) (.015) (.022) Transportation vehicles .085 .133 .125 .098 .139 .227 .231 .215 .202 .198 .170 .183 (.007) (.007) (.008) (.009) (.009) (.010) (.009) (.010) (.009) (.010) (.010) (.014) Wood and furniture À.097 À.147 À.114 À.107 À.098 À.141 À.117 À.155 À.087 À.056 À.095 À.078 (.010) (.011) (.012) (.012) (.012) (.013) (.013) (.013) (.012) (.012) (.012) (.017) Paper, pulp, and cardboard À.031 À.048 À.019 .013 À.002 À.029 À.025 .029 .041 .030 .062 .070 (.009) (.010) (.011) (.010) (.010) (.011) (.012) (.012) (.011) (.011) (.011) (.016) Rubber products .057 .060 À.019 À.021 À.011 .002 .030 .062 .089 À.032 .019 .014 (.018) (.018) (.021) (.022) (.019) (.023) (.023) (.023) (.023) (.024) (.025) (.034) Chemicals .168 .172 .155 .200 .174 .178 .136 .168 .111 .088 .131 .085 (.010) (.010) (.011) (.012) (.012) (.014) (.014) (.015) (.015) (.015) (.016) (.025) (Continued) T A B L E 4 . Continued Industry 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Petrochemicals .550 .446 .426 .510 .396 .449 .440 .558 .468 .450 .468 .421 (.016) (.017) (.019) (.021) (.019) (.021) (.024) (.026) (.024) (.024) (.022) (.033) Pharmaceuticals .012 .015 .034 .053 .094 .018 .041 .046 .079 .089 .090 .162 (.016) (.017) (.020) (.020) (.019) (.022) (.021) (.022) (.022) (.020) (.020) (.030) Plastics À.081 À.071 À.082 À.070 À.025 À.086 À.057 À.051 À.092 À.098 À.091 À.101 (.014) (.015) (.016) (.016) (.016) (.018) (.019) (.019) (.017) (.017) (.017) (.025) Textiles À.079 À.095 À.037 À.060 À.077 À.089 À.065 À.124 À.117 À.073 À.080 À.120 (.011) (.011) (.013) (.014) (.013) (.015) (.016) (.016) (.016) (.018) (.019) (.029) Clothing À.141 À.177 À.133 À.155 À.144 À.196 À.180 À.210 À.146 À.145 À.178 À.159 (.013) (.013) (.015) (.015) (.015) (.017) (.016) (.016) (.015) (.016) (.016) (.024) 330 Footwear À.118 À.187 À.165 À.150 À.169 À.194 À.117 À.084 À.131 À.172 À.161 À.193 (.011) (.012) (.013) (.014) (.013) (.016) (.015) (.014) (.013) (.014) (.014) (.021) Tobacco .232 .332 .201 .116 .275 .395 .441 .288 .198 .047 .001 .277 (.041) (.042) (.048) (.051) (.053) (.055) (.058) (.056) (.056) (.065) (.064) (.100) Foods À.197 À.190 À.210 À.185 À.167 À.199 À.199 À.219 À.190 À.149 À.146 À.177 (.008) (.008) (.009) (.009) (.008) (.009) (.009) (.009) (.009) (.009) (.009) (.013) Beverages À.110 À.070 À.122 À.138 À.135 À.132 À.074 À.023 À.026 À.062 À.064 À.060 (.015) (.016) (.018) (.019) (.018) (.020) (.021) (.021) (.020) (.021) (.023) (.032) SD of industry premiums .135 .138 .128 .135 .127 .154 .143 .156 .133 .128 .131 .137 Note: Numbers in parentheses are SEs. Industry wage premiums and their standard errors are calculated using the Haisken-DeNew and Schmidt (1997) procedure and are expressed as deviations from the employment-weighted average wage premium. Source: Authors’ calculations based on data from Brazil’s PME. Pavcnik and others 331 Industry wage premiums tend to be highest in industries that employ a low share of unskilled workers (as measured by the share of workers without a completed university degree), such as the petrochemical industry, tobacco, and chemicals, and lowest in industries that employ a large share of unskilled workers, such food products, textiles, and clothing. The correlation of industry wage premiums with the share of unskilled workers in the industry in 1987 is always highly negative, and the correlation coefficient ranges from À0.89 in 1987 to À0.8 in 1998.14 Finally, the first-stage results suggest that the structure of Brazilian industry wages did not change substantially between 1987 and 1998 even though the structure of protection changed substantially. The year-to-year correlations in industry wage premiums are very high, with the correlation coefficient usually exceeding 0.9. This finding is surprising, given results from previous studies on trade liberalization episodes in Mexico (Robertson 2000a) and Colombia (Goldberg and Pavcnik forthcoming). Those studies found low year-to-year corre- lations of industry wages, suggesting that the trade reforms changed the structure of industry wages. The magnitude of the correlation in Brazil is in line with evidence for the United States, which shows very stable wage premiums across years (year-to-year correlations are always estimated at above 0.9; see Kreuger and Summers 1988 and Gaston and Trefler 1994). This resemblance could be attrib- uted to the fact that despite the large tariff reductions, most Brazilian industries continue to face relatively low import penetration rates, which is also the case for the United States. The stable structure of industry wage premiums suggests that changes in trade policy are unlikely to be associated with changes in industry wage premiums. This relationship is explored in more detail in the next section. Industry Wage Premiums and Tariffs Table 5 reports the results for wage premiums and tariffs in the regression framework described in the methodology section. Because the first-stage regres- sion controlled for worker characteristics, the relationship between industry wage premiums and tariffs does not simply reflect industry differences in worker composition that also affect the political economy of protection. Similarly, because the returns to all worker characteristics are allowed to differ from year to year in the first stage, the first-stage coefficients capture changes in the economywide returns to worker characteristics associated with changes in labor supply over time. All second-stage regressions are estimated in first differences and include year indicators. They thus account for unobserved time-invariant, 14. The positive correlation between industry wage premiums and the share of skilled workers in an industry may be related to the fact that in Brazilian unions tend to be concentrated in industries with the highest shares of skilled workers. Arbache (2001) writes that ‘‘unionization [in Brazil] is a clear char- acteristic of managers, skilled production workers, office workers and, in particular, professionals’’ and shows that unions are able to extract a large union wage premium—about 18 percent. Alternatively, this positive correlation could also reflect positive spillovers between skilled workers. T A B L E 5 . Regression Results for Industry Wage Premiums and Trade Exposure Variable 1 2 3 4 5 6 7 8 9 Tariff À.0686 À.0560 À.0483 À.0575 .0141 .0572 À.1594 À.1547 À.1461 (0.0599) (0.0543) (0.0506) (0.0558) (0.1159) (0.1229) (0.1130) (0.1099) (0.1118) Lagged import .1508 .1747 .2059 .1884 .1323 penetrationa (0.2134) (0.2403) (0.2305) (0.2048) (0.2136) Lagged export to 0.2642* 0.2627* 0.2618** 0.2900** 0.2605** exposureb (0.1298) (0.1291) (0.1174) (0.1317) (0.1173) Tariff*lagged import À.2276 penetration (0.6633) Lagged imports* À.1154 À.1252 exchange rate (0.0972) (0.1028) Lagged exports* À.0233 À.0250 exchange rate (0.0824) (0.0844) Mercosur tariff À.0856 À.1126 (0.1501) (0.1496) 332 Lagged Mercosur .0027 imports (0.0028) Lagged Mercosur À0.0026* exports (0.0015) Two-stage least squares No No No No No No Yes Yes Yes First differences Yes Yes Yes Yes Yes Yes Yes Yes Yes Year indicators Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of observations 240 240 240 240 240 198c 240 240 240 *Significant at the 10 percent level; **significant at the 5 percent level. Note: Numbers in parentheses are SEs. Reported SEs are robust and clustered by industry. In column 7 tariff changes are instrumented for by presample tariffs and the exchange rate interacted with presample tariffs. In columns 8 and 9 tariff changes are instrumented for by presample tariffs and coffee prices interacted with presample tariffs. a Imports as a percentage of output plus net imports. b Exports as a percentage of output. c The number of observations is lower because Mercosur exports and imports for nonmanufacturing industries are missing. Source: Authors’ calculations based on labor market data from Brazil’s PME and trade and trade policy data from Muendler (2002). Pavcnik and others 333 industry-specific variables (such as lobbying power) and macroeconomic shocks that could influence wages concurrently with tariffs. The results in table 5 suggest no relationship between tariffs and industry wage premiums. Although industry wage premiums are an important component of workers’ earnings, they do not seem to be associated with trade policy. Because Brazil’s tariff changes might overstate the extent of trade liberalization (because of the size of the economy and remaining nontariff barriers), the study next explores whether wage premiums are affected by alternative trade exposure measures. First, a specification is estimated that includes industry measures of lagged import penetration and lagged export exposure in addition to tariffs (see table 5, column 2).15 The results suggest that high export exposure is associated with higher industry wages. This result is intuitive because higher industry exports likely increase the demand for workers in that particular industry. However, there is no statistically significant effect of lagged import penetration on wage premiums. To capture the possibility that the effects of tariffs differ across sectors with different degree of import competition (as measured by import penetration), the interaction of tariffs with import penetration is added to the specification (column 3). The insignificant interaction coefficient suggests that import penetra- tion does not affect wage premiums differentially in industries with lower tariffs. Finally, exchange rate fluctuation might also affect wages. Although year effects capture exchange rate fluctuations over time, the effect of exchange rates might vary with the trade exposure of the sector. But when the exchange rate is interacted with lagged trade flows, none of the previous findings is affected (column 4). This study focuses on the relationship between unilateral trade liberalization and industry wage premiums, but Brazil’s trade with Mercosur members is also examined to check the robustness of the findings. In 1991 Mercosur members began to reduce tariffs on internal trade, and by 1995 most intra-Mercosur trade was duty-free (Chang and Winters 2002; Olarreaga and Soloaga 1998). Trade with Mercosur is controlled for in two ways. First, Brazil’s tariffs on Mercosur imports are included in the baseline specification in column 1 of table 5. As in Chang and Winters (2002), these tariffs were obtained by applying the negotiated tariff reductions to the most favored nation industry tariff rates.16 Note that the two tariff rates are strongly positively correlated, at 15. Because trade flows are likely endogenous (they depend on factor costs), the first lags of import and export measures are included in the estimation rather than their current values. To the extent that these variables are serially correlated, this approach might yield biased results, especially in industry fixed effects specifications with relatively small numbers of observations. Nevertheless, the inclusion of these lagged variables does not change the conclusions about the relationship between tariffs and industry wages. 16. The timeline of the negotiated reductions of the internal tariffs relative to the most favored nation rates was as follows: 47 percent after the ratification of the treaty, 54 percent by December 1991, 61 percent by June 1992, 68 percent by December 1992, 75 percent by June 1993, 82 percent by December 1993, 89 percent by June 1994, and 100 percent by December 1994. Although countries could exclude some products from internal free trade, Chang and Winters (2002) suggest that Brazil declared only 27 exceptions, so that most of its Mercosur trade was duty-free. 334 THE WORLD BANK ECONOMIC REVIEW, VOL. 18, NO. 3 0.95 during the entire sample period of 1987–98 (likely reflecting the fact that these tariffs were identical until 1991) and at about 0.57 in the Mercosur period, 1992–98. Two interesting findings emerge. First, the coefficient on the Mercosur tariff is negative and statistically insignificant (column 5). Second, even when Brazil’s tariff on Mercosur imports is included, there is no statistical association between most favored nation tariffs and industry wage premiums. In fact, the coefficient on the most favored nation tariff is even closer to zero than the coefficient reported in column 1. Because this specification would still not capture the potential effect of Mercosur membership on industry wage premiums through increased Brazilian exports to Mercosur partners, a second specification also controls for Brazil’s total exports and imports to Argentina and Uruguay.17 This specification, reported in column 6 of table 5, thus adds Brazil’s tariff on Mercosur imports and measures of total lagged exports and imports with Mercosur to the speci- fication in column 2. The only Mercosur-specific variable that is statistically significant is Brazil’s exports to Mercosur. Although higher export exposure in an industry continues to be associated with a higher industry wage premium, the negative coefficient on Mercosur exports suggests that an industry’s increased exports to Mercosur are associated with a lower industry wage premium con- ditional on total exports. Again, there is no statistical association between most favored nation tariffs and industry wage premiums even after controlling for Mercosur-specific trade. This analysis was replicated in unreported regressions using only data from 1991 onward, with similar results. This discussion of industry wage premiums has so far ignored the potential role of labor market institutions, such as minimum wages and union power. These factors are unlikely to affect the findings, however. First consider the minimum wage. It is set nationally and does not vary across industries. As a result, its effects are captured by the year effects in the second-stage regressions and by coefficients on education indicators in the first stage (in the case where the minimum wage is binding only for people with lower earnings). Moreover, any effects that changes in the minimum wage might have had on industry 17. This information is based on bilateral trade with Brazil reported by Argentina and Uruguay in Trade and Production Data complied by M. Olarreaga and A. Nicita available on the World Bank Web site (www.worldbank.org). This information was not available for Paraguay. This lack of data is unlikely to be a big problem because Chang and Winters (2002) suggest that Argentina is Brazil’s main trading partner within Mercosur. This is also confirmed in the sample data for this study, which show that average industry imports and exports between Brazil and Argentina are about five to six times larger than those between Brazil and Uruguay. Moreover, these bilateral data focus only on manufacturing industries (and not on mining as well, as is the case in the main data for this study). Finally, Mercosur-specific exports and imports could not be expressed as a share of total output or domestic consumption because industry-level information on output is lacking (the original data on total import penetration and export exposure from Muendler 2002 and sources cited therein do not report trade flows and output separately for detailed industry categories). Pavcnik and others 335 wages through compositional channels (for example, because some industries employ more unskilled workers than others) are already controlled for because the first-stage regressions control for industry composition in each year and allow the returns to various levels of education to change from year to year. Second, although the individual-level data do not provide information on union membership, preventing formal analysis, changes in unionization are unlikely to be driving the industry wage premium results. If changes in union strength vary by industry through time in the same way that changes in tariffs vary by industry, then changes in unionization could affect industry wages independently of tariff changes, potentially biasing the results.18 But to the extent that union power in each industry has not changed over time in Brazil, first differencing of data would capture the union effects. This may in fact be a realistic assumption. Arbache and Carneiro (2000) report the shares of union- ized workers in various manufacturing industries in 1992 and 1995. Their data show that the shares are relatively stable over time.19 Moreover, no study was found that suggests that changes in union power were industry specific and were correlated with (or led to) changes in tariffs. Finally, because the structure of protection changed in Brazil during the sample period, it could be argued that unobserved time-varying shocks, which may simultaneously affect tariff changes and sector-specific premiums, drive the results. Thus the analysis also accounts for the potential endogeneity of trade policy changes by instrumenting for changes in trade policy with presample tariffs and with presample tariffs interacted with the exchange rate. As in Goldberg and Pavcnik (forthcoming), the choice of instruments is guided by the institutional details of Brazilian trade liberalization. Kume (2000) suggests that at the macroeconomic level Brazil changed trade policy in response to exchange rate fluctuations. Moreover, as discussed earlier, some sectors experienced larger tariff reductions than others. Tariffs were widely dispersed across sectors prior to trade reforms. As a result of Brazil’s commit- ment to economywide liberalization, trade reform led to proportionately larger tariff reductions in sectors with historically higher tariff levels. Regression of the tariff decline from 1987 to 1998 on 1986 tariffs yields a coefficient of 0.8 on 1986 tariffs (t-statistic 16.77) and an R2 of 0.94. This discussion suggests that the 1986 industry tariff levels and their inter- action with exchange rates are highly correlated with industry tariff reductions and may provide good instruments for the tariff changes. Because coffee is a major Brazilian export and coffee prices likely affect the exchange rate, the 18. The situation in which lower tariffs reduce union power, leading to lower wage premiums, is not a concern because in this case unions simply provide a potential mechanism through which tariffs can affect wages. 19. The correlation between industry union membership in 1992 and in 1995 is 0.82. Arbache and Carneiro (2000) use data from pnad. These data are not available for 1991–94, during the trade liberal- ization of the early 1990s, and the surveys for 1989 and 1990 do not contain information on union status. 336 THE WORLD BANK ECONOMIC REVIEW, VOL. 18, NO. 3 interaction of coffee prices (rather than exchange rates) with presample tariffs was also tested as an instrument. The relationship between sector-specific skill premiums and tariffs was estimated in first differences using two-stage least squares. When instrumenting for tariff changes with presample tariffs and their interaction with the exchange rate (see table 5, column 7) and presample tariffs and their interactions with coffee prices (columns 8 and 9),20 the magnitude of the negative coefficient on tariffs becomes smaller in absolute value, but the coefficients are imprecisely estimated. Thus the results continue to show no statistical relationship between trade policy and industry wage premiums. Overall, there is no statistically significant evidence that Brazilian trade liberalization affected the industry wage structure and thus wage inequality between skilled and unskilled workers through their industry affiliation. This finding is consistent with the evidence from Mexico (Feliciano 2001), which shows no relationship between industry wages and tariffs, but is inconsistent with the evidence from Colombia (Goldberg and Pavcnik forthcoming) and Mexico (Revenga 1997), which shows that tariff reductions are associated with declines in industry wages.21 Industry Wage Premiums for University-Educated Workers Although the results show no relationship between trade exposure and industry wage premiums, trade policy could still account for part of the increase in the return to university-educated workers if tariff reductions are associated with increases in sector-specific skill premiums. Industry wage premiums could differ for workers with different levels of education for several reasons. For example, more educated workers might be more or less mobile in the labor market. Or workers with different amounts of education might differ in their accumulation of sector-specific skills or their ability to bargain over wages. Revenga (1997) finds in Mexico that the greater the proportion of unskilled workers in an industry, the lower the ability of workers in the industry to capture part of the industry rents. Finally, industry-specific skill premiums might reflect effi- ciency wages paid to skilled workers to prevent them from shirking if industries face different monitoring costs. Robbins and Minowa (1996), for example, find substantial variation in returns to schooling across industries for manufacturing workers in Sa ˜ o Paolo, Brazil, in 1977. They attribute these differences to efficiency wages that firms pay to skilled workers in capital-intensive industries to avoid shirking. To investigate the relationship between industry-specific skill premiums and trade policy, skill-specific industry wage premiums are computed by employing 20. The first-stage F-statistics in these two-stage least square regressions are F(12,207) = 30.5, F(12,207) = 25.5, and F(16,203) = 19.3, respectively. 21. The differences in results in Feliciano (2001) and Revenga (1997) could stem from differences in methodology and from the fact that Feliciano uses worker-level data similar to the data used here, whereas Revenga uses plant-level data. Pavcnik and others 337 a modified version of equation 1 that allows industry wage premiums to differ for skilled and unskilled workers: ð3Þ lnðwijt Þ ¼ Hijt bH þ Iijt à wpjt þ Iijt à Sijt à wpSjt þ Eijt The variable Sijt is an indicator for whether worker i in industry j is skilled at time t (has a university degree). The coefficient wpSjt represents the incremental wage premium skilled workers earn in industry j in addition to the base wage premium in industry j, wpjt, which is received by unskilled and skilled workers. The differential impact of trade policy on the industry wages of skilled and unskilled workers is investigated by relating these industry-specific returns to skill to trade policy measures in the second stage of the estimation along the lines discussed in the methodology section. The first-stage results suggest that industry-specific skill premiums are poten- tially important (table 6). As in the case of industry wage premiums, the reported coefficients and standard errors are computed using the Haisken-DeNew and Schmidt (1997) procedure, expressed as deviations from the employment-weighted average skill premium. This normalized industry-specific skill premium can be interpreted as the proportional differences in wages through the channel of an industry-specific skill premium for a university-educated worker in a given indus- try relative to an average university-educated worker in all industries with the same observable characteristics. Thus a negative industry-specific skill premium suggests that the industry has a lower industry-specific skill premium than the average economywide skill premium (and not that skilled workers in this industry earn less than unskilled workers in the industry). Although the inclusion of industry-specific skill premiums does not increase the explanatory power of the regression by much, the premiums vary widely across industries (see table 6).22 University-educated workers in the tobacco industry and in oil extraction have the largest skill premiums, whereas univer- sity-educated workers in paper and clothing have the smallest. For example, estimates for 1987 suggest that a university-educated worker who switches from the textile to the chemical industry would see an almost 14 percent increase (124À(À0.014)) in wages through the channel of industry-specific skill pre- miums. The standard deviation of the industry-specific skill premiums ranges between 12.2 and 19.8 percent over 1987–98. Are these changes in sector-specific skill premiums associated with changes in trade policy? The regression results show no statistical association between tariff changes and changes in industry-specific skill premiums (table 7, column 1). What about other trade exposure measures? The results show that the relationship between tariffs and sector-specific skill premiums is robust to the inclusion of other trade exposure measures (columns 2–4). Although there is no 22. The R2 in these regressions is basically identical to the R2 in regressions with industry fixed effects reported at the bottom of table 3 at two decimal points. T A B L E 6 . Industry-Specific Skill Premiums for University Graduates Industry 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Mineral .220 .242 .311 .389 .181 À.029 .088 .383 .370 .283 .153 .426 extraction (.058) (.057) (.068) (.074) (.084) (.074) (.074) (.099) (.089) (.079) (.082) (.121) Oil extraction .242 .639 .374 .429 .382 .275 .611 .719 .555 .412 .298 .246 (.068) (.068) (.087) (.081) (.074) (.094) (.108) (.124) (.109) (.105) (.110) (.142) Nonmetalic mineral .092 .259 .135 .218 .297 .248 .201 .358 .335 .459 .187 À.006 338 trasformation (.042) (.049) (.063) (.070) (.064) (.064) (.058) (.062) (.063) (.052) (.066) (.090) Metalic products .143 .055 .152 .133 .124 .110 .102 .136 .097 .147 .150 .048 and steel (.023) (.023) (.027) (.028) (.028) (.030) (.030) (.033) (.032) (.031) (.032) (.052) Machinery and À.032 À.088 À.117 À.172 À.067 À.119 À.035 À.211 À.166 À.108 À.004 À.055 equipment (.033) (.031) (.035) (.037) (.035) (.044) (.047) (.051) (.048) (.043) (.043) (.064) Electrical and À.001 .019 .016 À.043 À.110 À.001 À.061 .062 .046 .051 .021 .005 electronic equipment (.028) (.028) (.033) (.032) (.032) (.039) (.041) (.042) (.040) (.039) (.040) (.062) Transportation À.089 À.158 À.232 À.134 À.121 À.010 À.027 À.063 À.104 À.205 À.128 .029 vehicles (.030) (.030) (.035) (.033) (.034) (.039) (.035) (.042) (.039) (.040) (.037) (.053) Wood and À.145 À.009 .072 .077 À.096 À.525 À.662 À.353 À.367 À.641 À.439 À.061 furniture (.084) (.079) (.081) (.082) (.092) (.112) (.122) (.108) (.085) (.095) (.096) (.116) Paper, pulp, and À.322 À.197 À.086 À.223 À.287 À.147 À.104 À.081 À.136 À.149 À.212 À.086 cardboard (.032) (.035) (.037) (.039) (.036) (.038) (.039) (.041) (.036) (.037) (.036) (.050) Rubber À.036 À.278 À.182 À.010 .002 À.055 .269 À.067 .026 .047 À.095 .012 products (.069) (.072) (.083) (.084) (.078) (.098) (.104) (.115) (.119) (.099) (.098) (.173) Chemicals .124 .013 .028 À.024 .036 À.037 .006 À.094 À.038 .000 .089 .173 (.027) (.027) (.031) (.029) (.030) (.038) (.040) (.043) (.042) (.039) (.040) (.066) Petrochemicals À.063 À.086 À.113 À.139 À.056 .042 À.128 À.117 À.021 À.259 À.170 À.219 (.034) (.035) (.043) (.042) (.040) (.046) (.047) (.054) (.052) (.054) (.047) (.067) Pharmaceuticals À.120 À.110 À.181 .139 .140 À.153 À.082 À.221 .031 À.080 .041 .009 (.044) (.045) (.054) (.058) (.054) (.067) (.062) (.065) (.064) (.054) (.056) (.080) Plastics À.077 .033 À.122 .042 .110 À.065 À.007 .176 .029 .123 .194 À.111 (.060) (.067) (.072) (.074) (.079) (.097) (.087) (.094) (.073) (.076) (.067) (.095) Textiles À.014 .085 .196 .156 À.025 .063 À.048 À.030 À.169 .006 .030 .021 (.050) (.056) (.059) (.067) (.057) (.070) (.074) (.073) (.072) (.085) (.082) (.104) Clothing À.358 À.441 À.355 À.124 À.299 À.384 À.424 À.599 À.404 À.378 À.248 À.001 (.072) (.070) (.083) (.083) (.077) (.088) (.095) (.126) (.113) (.095) (.082) (.144) Footwear À.041 .079 À.352 À.174 À.397 .280 .159 .013 À.107 À.120 .115 .301 (.089) (.099) (.124) (.112) (.088) (.104) (.184) (.111) (.094) (.109) (.108) (.264) Tobacco .040 .630 .687 .424 .237 À.054 .164 À.060 À.157 À.465 À.223 .456 (.133) (.150) (.192) (.182) (.157) (.149) (.193) (.207) (.268) (.204) (.151) (.313) Foods .082 .173 .215 .262 .238 .121 .071 .024 .104 .140 .038 .050 (.037) (.038) (.043) (.044) (.041) (.047) (.047) (.049) (.050) (.046) (.042) (.059) Beverages .258 .255 .046 À.030 .321 .140 .330 .299 .370 .220 .413 À.204 339 (.073) (.073) (.087) (.084) (.073) (.079) (.104) (.090) (.094) (.092) (.085) (.117) SD .139 .163 .157 .156 .163 .122 .138 .175 .156 .198 .147 .098 Note: Numbers in parentheses are SEs. Industry wage premiums and their SEs are calculated using the Haisken-DeNew and Schmidt (1997) procedure. They are all expressed as deviations from the employment weighted average skill premium. Source: Authors’ calculations based on data from Brazil’s PME. T A B L E 7 . Regression Results for Industry-Specific Skill Premiums for University Graduates and Trade Exposure 1 2 3 4 5 6 7 8 9 Tariff À.1948 À.1334 À.2211 À.1296 À.3130 À.3059 À.0091 À.1667 À.1427 (0.1678) (0.1810) (0.2003) (0.1833) (0.3978) (0.4258) (0.1663) (0.1104) (0.1388) Lagged import penetrationa À.1567 À.3449 À.2708 À.2140 À.2819 (0.5131) (0.5667) (0.5346) (0.5627) (0.5456) Lagged export exposure 1.3691** 1.3740** 1.2357** 1.4292** 1.2342** (0.3443) (0.3509) (0.4161) (0.4809) (0.4137) Tariff*lagged import penetration 2.0329 (1.9014) Lagged imports*exchange rate .2845 .2834 (0.3351) (0.3383) Lagged exports*exchange rate .4136 .4134 (0.2934) (0.2928) Mercosur tariff .1221 .1751 (0.3258) (0.3550) 340 Lagged Mercosur imports .0019 (0.0054) Lagged Mercosur exports .0039 (0.0067) Two-stage least squares No No No No No No Yes Yes Yes First differences Yes Yes Yes Yes Yes Yes Yes Yes Yes Year indicators Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of observations 240 240 240 240 240 198 240 240 240 *Significant at the 10 percent level; **significant at the 5 percent level. Note: Numbers in parentheses are SEs. Reported SEs are robust and clustered by industry. In column 7 tariff changes are instrumented for by presample tariffs and the exchange rate interacted with presample tariffs. In columns 8 and 9 tariff changes are instrumented for by presample tariffs and coffee prices interacted with presample tariffs. a Imports as a percentage of output plus net imports. b Exports as a percentage of output. c The number of observations is lower because Mercosur exports and imports for nonmanufacturing industries are missing. Source: Authors’ calculations based on labor market data from Brazil’s PME and trade and trade policy data from Muendler (2002). Pavcnik and others 341 relationship between sector-specific skill premiums and import penetration, increases in export exposure within an industry are associated with increases in the skill premium in that industry. The findings are also robust to inclusion of Brazil’s tariff on imports from Mercosur countries (column 5) and of Brazil’s exports to and imports from Argentina and Uruguay (column 6). None of the Mercosur-specific trade measures is statistically significant, and their inclusion does not alter the findings on the relationship between skill-specific wage premiums and tariffs. Finally, instrumenting for tariff changes with presample tariffs and their interaction with the exchange rate (column 7) and presample tariffs and their interaction with coffee prices (columns 7 and 9) again shows a negative, but statistically insignificant relationship between tariff changes and changes in industry-specific skill premiums.23 In sum, the study finds no statistically significant evidence that tariff reduc- tions affected worker wages in Brazil through their industry affiliation or that tariff reductions contributed to wage inequality between skilled and unskilled workers through this channel. IV. CONCLUSION The analysis here was motivated in part by the current policy discussion on the benefits and costs of trade reforms. Many people have recently questioned whether the potential benefits of trade liberalization (increased efficiency and welfare) outweigh the potential costs (increased inequality, ‘‘race to the bottom’’ in wages). Several recent studies have proposed the use of labor market policies, such as minimum wages and government social protection programs, to offset the potential increase in inequality associated with trade liberalization (Rama 2001, 2003; Rama and Ravallion 2001). This study contributes to the policy debate in several ways. First, it is one of only a few studies that focus on trade policy variables (such as tariffs) rather than outcome variables (such as openness) in examining the implications of trade reforms for labor markets. Rodriguez and Rodrik (1999) point out the difficulties in assessing the impact of trade liberalization if trade reforms are measured using outcome variables such as openness, which reflect not only a country’s trade policy but also factors such as transport costs, technology, demand, and most important, changes in factor prices.24 The use of trade policy variables is thus an advantage. 23. The first-stage F-statistics in these two-stage least square regressions are F(12,207) = 26.8, F(12,207) = 22.3, and F(16,203) = 17.5, respectively. 24. One disadvantage of the tariff measures is that changes in tariffs may have little effect if large nontariff barriers remain. Although detailed information on nontariff barriers is not available, trade liberalization also significantly reduced nontariff barriers. Moreover, the findings of this study are essentially unaffected by the inclusion of measures of openness, such as (lagged) import penetration and the export exposure ratio that partially accounts for the effects of nontariff barriers. 342 THE WORLD BANK ECONOMIC REVIEW, VOL. 18, NO. 3 Second, opponents of globalization often claim that trade reforms make work- ers in previously protected sectors poorer and that trade liberalizations leads to a race to the bottom in wages. Some studies report results that are potentially consistent with this claim. For example, Goldberg and Pavcnik (forthcoming) and Revenga (1997) find that tariff reductions are associated with declines in industry wage premiums in Colombia and Mexico. Rama (2001) also finds some evidence of a negative association between openness and wages in the short run in a cross- country study. Rama (2001, 2003) has suggested that trade liberalization could be accompanied by increases in minimum wages to compensate potential losers. The evidence here from Brazil suggests that trade liberalization does not necessarily lead to lower industry wages through the channel of industry wage premiums in the short run. Obviously, trade liberalization could still lower wages through other channels, such as lower returns to education or experience, that are not the focus of this study. Exploring the differences in country characteristics or policies that determine how trade reform affects worker wages through various channels may thus be fruitful ground for future research. Finally, although no evidence was found that drastic tariff declines worsened inequality through changes in the structure of wage premiums, industry wage premiums were found to vary widely across Brazilian manufacturing sectors, accounting for 4–6 percent of the explained variation in log hourly wages. In addition, industry wage premiums are smallest in sectors with high shares of unskilled workers. This suggests that unskilled workers earn lower wages not only because of the growing economywide skill premium but also because they are disproportionately employed in industries with low wage premiums, a source of inequality that has been undetected in previous studies. This source of inequality, along with the rising skill premium, could be addressed through labor market policies, such as those promoted by Rama (2001) (changes in minimum wages and in social security programs), in addition to improved access to education. REFERENCES Arbache, J. S. 2001. ‘‘Unions and the Labor Market in Brazil.’’ University of Brasilia, Department of Economics. Arbache, J. S., and F. G. Carneiro. 2000. ‘‘Unions and Interindustry Wage Differentials.’’ World Devel- opment 27(10):1875–83. Arbache, J. S., and N. Menezes-Filho. 2000. ‘‘Rent-Sharing in Brazil: Using Trade Liberalization as a Natural Experiment.’’ University of Brasilia, Department of Economics. Attanasio, O., P. Goldberg, and N. Pavcnik. 2004. ‘‘Trade Reforms and Wage Inequality in Colombia.’’ Journal of Development Economics 74(2):331–66. Barros, R., and L. Ramos. 1996. ‘‘Temporal Evolution of the Relationship between Wages and Education of Brazilian Men.’’ In N. Birdsall and R. H. Sabot, eds., Opportunity Forgone: Education in Brazil. Washington, D.C.: Inter-American Development Bank. Behrman, J., N. Birdsall, and M. Szekely. 2000. ‘‘Economic Reform and Wage Differentials in Latin America.’’ IADB Working Paper. Inter-American Development Bank, Washington, D.C. Pavcnik and others 343 Blom, A., L. Holm-Nielsen, and D. Verner. 2001. ‘‘Education, Earnings, and Inequality in Brazil 1982– 1998: Implications for Education Policy.’’ World Bank, Washington, D.C. Chang, W., and L. A. Winters. 2002. ‘‘How Regional Blocs Affect Excluded Countries: The Price Effects of Mercosur.’’ American Economic Review 92(4):889–904. Cragg, M. I., and M. Epelbaum. 1996. ‘‘Why Has Wage Dispersion Grown in Mexico? Is It the Incidence of Reforms or the Growing Demand for Skills?’’ Journal of Development Economics 51(1):99–116. Feliciano, Z. 2001. ‘‘Workers and Trade Liberalization: The Impact of Trade Reforms in Mexico on Wages and Employment.’’ Industrial and Labor Relations Review 55(1):95–115. Fernandes, A. M. 2001. ‘‘Trade Policy, Trade Volumes and Plant-Level Productivity in Colombian Manufacturing Industries.’’ Yale University, Department of Economics, New Haven, Conn. Frankel, J., and Romer D. 1999. ‘‘Does Trade Cause Growth?’’ American Economic Review 89(3): 379–99. Gaston, N., and D. Trefler. 1994. ‘‘Protection, Trade, and Wages: Evidence from U.S. Manufacturing.’’ Industrial and Labor Relations Review 47(July):575–93. Goldberg, P., and N. Pavcnik. Forthcoming. ‘‘Trade, Wages, and the Political Economy of Trade Protection: Evidence from the Colombian Trade Reforms.’’ Journal of International Economics. Green, F., A. Dickerson, and J. S. Arbache. 2001. ‘‘A Picture of Wage Inequality and the Allocation of Labor through a Period of Trade Liberalization: The Case of Brazil.’’ World Development 29(11): 1923–39. Grossman, G. 1984. ‘‘International Competition and the Unionized Sector.’’ Canadian Journal of Eco- nomics 17(3):541–56. Haisken-DeNew, J. P., and C. M. Schmidt. 1997. ‘‘Inter-Industry and Inter-Region Wage Differentials: Mechanics and Interpretation.’’ Review of Economics and Statistics 79(3):516–21. Harrison, A. 1994. ‘‘Productivity, Imperfect Competition and Trade Reform: Theory and Evidence.’’ Journal of International Economics 36(1–2):53–73. Harrison A., and G. Hanson. 1999. ‘‘Who Gains from Trade Reform? Some Remaining Puzzles.’’ Journal of Development Economics 59(1):125–54. Hay, D. A. 2001. ‘‘The Post-1990 Brazilian Trade Liberalization and the Performance of Large Manu- facturing Firms: Productivity, Market Share, and Profits.’’ Economic Journal 111(473):620–41. Heckman, J., and C. Pages. 2000. The Cost of Job Security Regulation: Evidence from Latin American Labor Markets. nber Working Paper 7773. National Bureau for Economic Research, Cambridge, Mass. Kim, E. 2000. ‘‘Trade Liberalization and Productivity Growth in Korean Manufacturing Industries: Price Protection, Market Power and Scale Efficiency.’’ Journal of Development Economics 62(1):55–83. Krishna, P., and D. Mitra. 1998. ‘‘Trade Liberalization, Market Discipline and Productivity Growth: New Evidence from India.’’ Journal of Development Economics 56(2):447–62. Krueger, A. B., and L. H. Summers. 1988. ‘‘Efficiency Wages and the Inter-Industry Wage Structure.’’ Econometrica 56(2):259–93. ¸a Kume, H. 2000. ‘‘A politica brasileira de importac ¸a ˜ o no period 1987 descric ¸a ˜ o e avaliac ˜ o.’’ Instituto de Pesquisa Econo ˆ mica Aplicada, Rio de Janeiro. Kume, H., G. Piani, and C. F. Souza. 2000. ‘‘Instrumentos de Politica Comercial no Periodo 1987–1998.’’ Instituto de Pesquisa Econo ˆ mica Aplicada, Rio de Janeiro. Lam, D., and R. Schoeni. 1993. ‘‘The Effects of Family Background on Earnings and Returns to School- ing: Evidence from Brazil.’’ Journal of Political Economy 101(4):710–40. Levinsohn, J. 1993. ‘‘Testing the Imports-as-Market-Discipline Hypothesis.’’ Journal of International Economics 35(1–2):1–22. Melitz, M. 2003. ‘‘The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Produc- tivity.’’ Econometrica 71(6):1696–725. Muendler, M. A. 2002. ‘‘Trade, Technology, and Productivity: A Study of Brazilian Manufacturers, 1986–1998.’’ University of California, Berkeley, Department of Economics. 344 THE WORLD BANK ECONOMIC REVIEW, VOL. 18, NO. 3 Olarreaga, M., and I. Soloaga. 1998. ‘‘Endogenous Tariff Formation: The Case of MERCOSUR.’’ World Bank Economic Review 12(2):297–320. Pavcnik, N. 2002. ‘‘Trade Liberalization, Exit and Productivity Improvements: Evidence from Chilean Plants.’’ Review of Economic Studies 69(1):245–76. Rama, M. 2001. ‘‘Globalization, Inequality, and Labor Market Policies.’’ World Bank, Washington, D.C. ———. 2003. ‘‘Globalization and Workers in Developing Countries.’’ Policy Research Working Paper 2958. World Bank, Washington, D.C. Rama, M., and M. Ravallion. 2001. ‘‘Labor Market Regulation and Inequality: A Cross-Country Analysis.’’ World Bank, Washington, D.C. Revenga, A. 1997. ‘‘Employment and Wage Effects of Trade Liberalization: The Case of Mexican Manufacturing.’’ Journal of Labor Economics 15(3):S20–43. Robbins, D. J. 1996. ‘‘Evidence on Trade and Wages in the Developing World.’’ OECD Technical Paper 119. Organisation for Economic Co-operation and Development, Paris. Robbins, D., and M. Minowa. 1996. ‘‘Do Returns to Schooling Vary Across Industries?’’ In N. Birdsall and R. H. Sabot, eds., Opportunity Foregone: Education in Brazil. Washington, D.C.: Inter-American Development Bank. Roberts, M. J., and J. R. Tybout, eds. 1996. Industrial Evolution in Developing Countries. Oxford: Oxford University Press. Robertson, R. 2000a. ‘‘Inter-industry Wage Differentials across Time, Borders, and Trade Regimes: Evidence from the US and Mexico.’’ Macalester College, Department of Economics, St. Paul, Minn. ———. 2000b. ‘‘Trade Liberalization and Wage Inequality: Lessons from Mexico.’’ World Economy 23(6):827–49. Rodriguez, F., and D. Rodrik. 1999. Trade Policy and Economic Growth: A Skeptic’s Guide to the Cross-National Evidence. NBER Working Paper 7081. National Bureau of Economic Research, Cam- bridge, Mass. Rodrik, D. 1991. ‘‘Closing the Productivity Gap: Does Trade Liberalization Really Help?’’ In G. Helleiner, ed., Trade Policy, Industrialization and Development. Oxford: Clarendon Press. ´nchez-Pa Sa ´ramo, C., and N. Schady. 2003. ‘‘Off and Running? Technology, Trade, and the Rising Demand for Skilled Workers in Latin America.’’ Policy Research Working Paper 3015. World Bank, Washington, D.C.