International Economic Activities and Skilled Labor Demand: Evidence from Brazil and China

Transcription

1 International Economic Activities and Skilled Labor Demand: Evidence from Brazil and China Pablo Fajnzylber TheWorldBank Ana M. Fernandes TheWorldBank September 6, 2006 Abstract Using two new firm-level datasets, this paper investigates the impact of three international economic activities - the use of imported inputs, exports, and foreign direct investment - on skilled labor demand in Brazil and China. We find that Brazilian firms that engage in these activities exhibit a higher skilled labor demand than firms that do not. In contrast, Chinese firms that engage in these activities have a lower skilled labor demand than firms that do not. Thus, international economic activities act as a channel for skill-biased technology diffusion in Brazil but have an effect of specialization according to comparative advantage in unskilled labor-intensive goods in China. JEL Classification Numbers: F02, F16, J23, J31, O33. Keywords: technology diffusion, labor demand, trade and labor market interactions, foreign direct investment, exports. This paper has benefited from the suggestions of an anonymous referee and participants at the LACEA 2004, NEUDC 2004, and the Second Industrial Development Conference (Washington, 2004). The findings expressed in this paper are those of the authors and do not necessarily represent the views of the World Bank. Please address correspondence to Ana M. Fernandes, The World Bank, 1818 H Street, NW, Washington DC Phone: Fax: afernandes@worldbank.org.

2 1. Introduction Empirical evidence from industrial and developing countries shows that firms producing similar goods in the same location exhibit a large degree of heterogeneity in terms of size, investment, productivity, and wages. Interestingly, the literature also recognizes a systematic link between firm characteristics and the propensity to engage in international economic activities. For example, firms that export are larger and more productive than firms that do not, even within narrowly defined industries. 1 Our paper contributes to this literature by examining the heterogeneity in skilled labor demand of manufacturing firms in two developing countries: Brazil and China. We focus on the question of whether, within industries and regions, firms that are more integrated into world markets demand relatively more skilled labor. To proxy for a firm s degree of integration into world markets, we consider three international economic activities: the use of imported intermediate inputs, exports, and foreign direct investment (FDI). The use of imported intermediate inputs can have two effects on skilled labor demand in developing countries. First, if these inputs embody state-of-the-art technology not available domestically, then they act as a channel for technology diffusion (Grossman and Helpman, 1991; Keller, 2004; Lumenga-Neso et al., 2005). Moreover, by interacting with foreign suppliers, importers gain access to tacit forms of knowledge which are not transferable through market transactions. The technology and knowledge accessed through the use of imported inputs are likely to be skill-biased, as in Robbins (1996) skill-enhancing trade hypothesis. 2 Second, within an industry, firms that use imported inputs are likely to concentrate more heavily in stages of the production process or in goods in which the country has a comparative advantage. In a developing country with abundance of unskilled labor, firms located in export-processing zones tend to import skilled labor-intensive intermediate inputs and concentrate on unskilled 1 See, e.g., Bernard and Jensen (1999) and Fernandes and Isgut (2006). 2 Robbins (1996) argues that imports of skill-biased machinery can explain the increase in skilled labor demand in several developing countries over the last two decades. 1

3 labor-intensive assembly operations. Hence, the effect of the use of imported inputs on firm skilled labor demand depends on the importance of technology diffusion relative to specialization according to comparative advantage. Exports can also have two effects on skilled labor demand in developing countries. First, exporters may be pressured by foreign clients to meet higher quality standards than those prevailing in the domestic market and gain access to tacit or proprietary knowledge provided by foreign clients to help them meet those standards (Westphal, 2002). Thus, exports may act as a channel for technology diffusion. Second, within an industry, exporters who play to their strengths may specialize in stages of the production process or goods that use unskilled labor more intensively. Thus, whether exports result in higher skilled labor demand depends on the technology diffusion effect being stronger than the effect of specialization according to comparative advantage. FDI is expected to affect skilled labor demand positively since it allows for the diffusion of technology from multinational parents to subsidiaries. In fact, the presence of knowledge-based firm-specific assetsandthedifficulties associated with their market-mediated transfer (e.g., through technology licensing) explain the existence of multinationals (Markusen, 1995). Host country location advantages may also lead to FDI. In particular, lower costs of unskilled labor in developing countries may result in vertical FDI. However the extent of technology transfer to the host country is likely to be smaller for vertical FDI, which could imply a smaller impact of FDI on skilled labor demand in developing countries (Keller, 2004). Nevertheless, it is also possible that while the activities transferred through vertical FDI are unskilled labor-intensive in the multinational parent s country, the demand for skilled labor of subsidiaries is still higher than that of domestic firms in the host country (Feenstra and Hanson, 1996b). In sum, for firms in developing countries, international activities linked to trade and invest- 2

4 ment can (i) act as channels for the diffusion of skill-biased technology and (ii) be associated with the exploitation of comparative advantage. Hence, empirical evidence is crucial to determine the effect of these activities on skilled labor demand. This paper provides such evidence for manufacturing firms in Brazil and China. The study of Brazil and China is particularly interesting for four reasons. First, given their large size, both economies are crucial drivers of development in their regions. Second, during the last decade both countries experienced rapid trade liberalization and strong export dynamism and thus increased substantially their integration into the world economy. 3 Third, despite these similarities, per capita incomes are three times higher in Brazil than in China but average years of education are 30% higher in China. Moreover, despite the relatively smaller size of Brazil, the degree of trade openness of China is much higher. 4 Fourth, as a result of these differences in income, education, and trade openness, a higher share of FDI flows are aimed at supplying local markets in Brazil, while export-oriented FDI is much more common in China. Overall, Brazil and China are interesting case studies for understanding the links between globalization and labor demand, both in themselves, and as a reference for other developing countries. Our main findings are as follows. First, we find that the use of imported inputs and FDI have a positive effect on skilled labor demand in Brazil. These activities act as channels for the diffusion of skilled-biased technology and thus the skill-enhancing trade hypothesis is supported for Brazil. Second, we find a negative effect of exports on skilled labor demand of Brazilian firms that export more than half their output. However, the effect of exports is positive for the remaining exporters. Thus, for the former (latter) firms, the effect of specialization in unskilled labor-intensive production is larger (smaller) than the effect of exports in diffusing technology. Third, in contrast to the evidence for Brazil, the use of imported inputs and exports 3 Between 1999 and 2004, real exports grew at an average annual rate of 12.4% (24.7%) in Brazil (China), compared to a GDP growth rate of 2.6% (9.2%). 4 The sum of exports and imports represent about two thirds of GDP in China, but only 31% in Brazil. 3

5 are negatively associated with skilled labor demand in China. Hence, for Chinese firms engaged in these activities, the effect of specialization according to the country s comparative advantage in unskilled labor-intensive goods is larger than the effect of skill-biased technology diffusion. Fourth, FDI does not affect significantly skilled labor demand in China, suggesting that the effect of specialization in unskilled labor-intensive goods by foreign-owned firms counteracts the role of FDI in diffusing technology. Our main findings are robust to the use of different proxies for skilled labor demand and to the simultaneous inclusion in the baseline specification of measures of imports, exports, and FDI. The results are mostly unchanged when the impact of international economic activities is allowed to differ in industries with revealed comparative advantage, as well as when instrumental variables (IV) or Tobit estimation are used instead of OLS. Our paper relates to the literature that examines the reasons for the observed increase in skilled labor demand in developed countries in the 1990s. 5 Skill-biased technological change (SBTC) and increased trade with developing countries are the two major reasons proposed for the increase. Studies that estimate firm or industry skilled labor demand (known as wage bill studies) - whose methodology we also follow - provide strong evidence that SBTC is the driving force behind the increase in skilled labor demand (Berman et al., 1994; Doms et al., 1997; Autor et al., 1998; Machin and Van Reenen, 1998; Haskel and Heden, 1999). 6 Our findings for Brazil are in line with those for developed countries to the extent that the international activities that we focus on are channels for technology diffusion and thus for SBTC. Our findings for China contrast with those for developed countries. 7 5 Berman and Machin (2000) and Kranz (2006) show that the increase in skilled labor demand is expressed as an increase in relative employment of skilled workers, in relative wages of skilled workers, or both, depending on the country. 6 Wage bill studies estimate a direct link between investments in computerization or in research and development and skilled labor demand. Other studies decompose the observed increase in skilled labor demand into a between and a within-industry component and find that most of the increase occurs within industries and that skill upgrading occurs in similar industries - those that are technologically more advanced - across countries, which is indirect evidence of SBTC (e.g., Berman et al., 1998; Choi and Jeong, 2005). 7 However, our findingofnoeffectof FDI on skilled labor demand in China parallels that obtained by Bloningen and 4

6 More specifically, our paper contributes to the literature examining the effects of trade and technology on skilled labor demand in developing countries. Feenstra and Hanson (1997) show that the Mexican states and industries that received more FDI after trade and FDI regimes were liberalized exhibit a greater skilled labor demand. 8 Harrison and Hanson (1999) find that Mexican firms that export or are foreign-owned tend to employ a higher share of skilled workers, and so do firms that use more imported inputs and machinery. Pavcnik (2003) finds that the use of imported materials, patented technology, and foreign technical assistance are unrelated to skilled labor demand of Chilean plants in fixed effects and time-differenced regressions, but they have a positive effect on this demand in cross-sectional regressions. Our paper differs from this literature in three important respects. First, we consider two potential effects of international activities on skilled labor demand: (i) channels for technology diffusion and (ii) vehicles for specialization according to comparative advantage. Second, we include in our skilled labor demand equations controls for managerial ability and absorptive capacity which are crucial to obtain unbiased estimates of the effects of international activities. 9 Third, we provide evidence on the effect of international activities on skilled labor demand for two large cross-sections of firms in Brazil and China drawing on new survey data collected by the World Bank (Investment Climate Surveys). The use of panel data (possibly available in the Slaughter (2001) for the US. Also, our finding of a positive effect of exports on skilled labor demand for Brazil is in line with Bernard and Jensen (1997) who find that higher skilled labor demand within and across US manufacturing industries is largely due to employment movements towards exporters. 8 This finding is consistent with the model of Feenstra and Hanson (1996b) where a single final good is produced with a continuum of intermediate inputs traded between two countries, North and South, with relative abundance of skilled and unskilled labor, respectively. The North (South) specializes in inputs that are relatively intensive in skilled (unskilled) labor. A flow of capital from North to South leads to a shift of production activities to the South. Those activities are relatively intensive in unskilled labor by Northern standards, but they use more skilled labor than other inputs previously produced in the South. Thus, production outsourcing from a developed to a developing country results in an increase in relative wages of skilled workers in both countries. 9 First, better managers are more able to engage in these activities but may also demand more skilled labor. Pavcnik (2003) notes that unobserved managerial ability can explain the significant positive effects of foreign technology on skilled labor demand in her cross-sectional regressions, relative to the insignificant effects in plant fixed effects regressions. Second, firms with better absorptive capacity are likely to demand more skilled labor, as knowledge activities are skilled labor-intensive, and may also benefit more from pursuing international activities. Keller (1996) argues that the lack of a minimum local critical mass of absorptive capacity may explain why some outward-oriented countries have benefited less than others from the access to international technology flows. 5

7 near future) to explore this effect is an important avenue for future research. 10 For a developing country, traditional Hecksher-Ohlin trade theory predicts that activities linked to trade openness should be associated with the exploitation of comparative advantage in unskilled labor-intensive goods. Such an exploitation will increase the relative demand for unskilled labor and reduce the wage differential between skilled and unskilled labor. However, these predictions apply to industries, whereas our paper focuses on firms. As in Bernard et al. (2006a), we assume that a firm s input intensity proxies for the unobserved type of goods it produces within its industry: e.g., a firm demanding relatively more skilled labor is assumed to produce relatively more skilled labor-intensive goods within its industry. 11 Additionally, some features of recent theoretical models of international trade with heterogeneous firms are useful to motivate our analysis. First, Yeaple (2005) shows that international trade can lead to a reallocation of differently-skilled workers across firms and industries when firms choose the types of workers. Second, Bernard et al. (2006b) show that factor intensity and factor abundance, which traditionally determine between-industry resource reallocations due to trade, can also affect within-industry resource reallocations from less to more productive firms. The paper proceeds as follows. Section 2 presents the empirical methodology. Section 3 describes the data and Section 4 reports regression results. Section 5 discusses robustness checks and Section 6 offers concluding remarks. 2. Empirical Methodology This section presents the empirical methodology used to investigate the effect of international economic activities on skilled labor demand. We assume that capital is a quasi-fixed input and firms choose their skilled and unskilled labor inputs by minimizing a restricted vari- 10Liu et al. (2001) also use cross-sectional data to estimate the effect of advanced technology on skilled labor demand. 11This assumption is motivated by the work of Bernard and Jensen (1999). They find that (i) industry affiliation is not informative about the exporter status of a firm since factor intensities vary substantially within industries, and (ii) a large fraction of firms does not export even in comparative advantage export sectors. 6

8 able cost function subject to an output constraint. 12 For firm i, the minimum labor cost of producing value-added VA i given its capital stock K i, its internal absorptive capacity T 1 i,and its international activity T 2 i is given by: 13 VC i = f(wi S,wi U,K i, VA i,ti 1,Ti 2 ), (1) where w S i (wi U ) is the wage paid to skilled (unskilled) workers. We use a translog functional form for logarithmic variable costs as in Berman et al. (1994). The estimable equation for firm relative skilled labor demand is given by: 14 wi S L S i VC i w S = α + β 1 ln i Ki wi U + β 2 ln + β VA 3 ln VA i + β 4 Ti 1 + β 5 Ti 2 + ε i, (2) i where ε i is a stochastic error representing measurement or optimization error. The dependent variable in Equation (2) is the wage bill share of skilled workers. If β 4 > 0, thenthefirm s absorptive capacity is skill-biased. If β 5 > 0, then the international activity the firm engages in is skill-biased. Our coefficient of interest is β 5, but the cost function framework also controls for potential capital-skill complementarity (β 2 > 0) and for increasing returns to scale (β 3 > 0). Identifying the effect of relative wages on skilled labor demand in Equation (2) is difficult. The cross-sectional variation in relative wages observed is often due to variation in the unobserved quality of labor and thus is not exogenous. Following Berman et al. (1994), most researchers estimate variants of Equation (2) dropping the relative wage term, since instruments to correct for its endogeneity are not available. 15 We also drop the relative wage term, but we include industry and region dummy variables that control for differences in the excluded term and for other systematic differences in skilled labor demand across industries and regions. 12We measure output by value-added and the firm s variable costs by labor costs as in Berman et al. (1994). 13For brevity, Ti 1 and Ti 2 are described as single variables. However, Ti 1 is actually a vector of two dummy variables for firms doing R&D and for firms introducing new product lines. Also, Ti 2 is either a single international activity (Sections ) or a vector of three international activities (Section 4.4). 14Equation (2) is derived by (i) differentiating the cost function with respect to ln wi S, (ii) using Shephard s lemma, (iii) assuming symmetry of the effect of relative input prices on input demands, and (iv) assuming homogeneity of degree one in input prices. The derivation is available upon request. 15Autor et al. (1998), Machin and Van Reenen (1998), and Haskel and Heden (1999) argue that time dummy variables account for the variation in relative wages when using panel data. 7

9 To address the potential correlation between the error term in Equation (2) and the variables Ti 1 and Ti 2 due to unobserved managerial ability, we include in the regression a proxy for such ability, ME i.ourfinal estimable equation for firm i in industry j and region r is given by: w S ijrl S ijr VC ijr = α + β 2 ln( K ijr VA ijr )+β 3 ln VA ijr + β 4 T 1 ijr + β 5T 2 ijr + β 6ME ijr + I j + I r + ε ijr, (3) where I j and I r are industry and region fixed effects, respectively. 3. Data and Cross-Tabulation Results Our analysis draws on survey data recently collected by the World Bank based on a common questionnaire applied to random samples of firms in China and Brazil in 2001 and 2003, respectively. 16 In Brazil, stratified samples based on industry-state-size were randomly drawn for thirteen states and all the regressions estimated in the paper are weighted accordingly. 17 In China, the sample was randomly chosen from five cities and covers private firms and publiclyowned firms. 18 We restrict our analysis to private firms since publicly-owned firms tend to operate very differently: e.g., it is unlikely that publicly-owned firms satisfy the cost-minimization assumption of our framework. Table 1 shows the sample composition by industry (Panel A) and size category (Panel B). The apparel, auto-parts, and electronics industries are covered in both countries. Small firms represent about half of the Brazilian sample, whereas the majority of firms in the Chinese sample are large. These compositional differences do not influence our results since our regressions include industry dummy variables, and capital intensity and value-added, which control for firm size. The dependent variable in Equation (3) is the wage bill share of skilled workers. Following the literature, we also consider as a dependent variable the employment share of skilled workers. 16More information on the survey can be found at The survey covers manufacturing establishments, but we designate those as firms for brevity. 17Thesampleframewasthelistingoffirms provided by InstitutoBrasileiroGeografia eestatisticaand the states covered are: Amazonas, Bahia, Ceará, Goiás, Maranhão, Mato Grosso, Minas Gerais, Paraíba, Paraná, Rio de Janeiro, Rio Grande do Sul, Santa Catarina, and São Paulo. 18The sample frame was an electronic list of firms in Beijing, Chengdu, Guangzhou, Shanghai, and Tianjin. 8

10 Our surveys provide information on wages, employment, and the average number of years of education for five types of workers (management, professionals, skilled production workers, unskilled production workers and other nonproduction workers). We consider total compensation (i.e., wages and salaries, bonuses, and other benefits) as the measure of wages. We construct three measures of skilled workers. Our first measure referred to as nonproduction workers is the sum of management, professionals, and other nonproduction workers. Most previous studies use datasets for which the skilled/unskilled workers distinction can be proxied only by this nonproduction/production workers distinction. However, this is imperfect since some of the nonproduction workers are actually engaged in low skilled tasks. Our second measure referred to as workers in managerial, engineering, and technical occupations is the sum of management and professionals. This measure improves upon the nonproduction/production workers distinction since it excludes the other nonproduction workers in our surveys (e.g., janitors). Our third measure referred to as workers with some college education. For China, we use information on average educational levels for each type of workers. 19 For Brazil, we use direct information on the percentage of workers with some college education. 20 Panel A of Table 2 shows summary statistics for wage bill shares and employment shares of skilled workers. Brazilian firms have lower average shares of skilled workers than Chinese firms according to all skills measures, which is consistent with macro evidence in Barro and Lee (2000) that the average number of years of education in China (6.4) is higher than in Brazil (4.9). The learning and absorptive capacity of firms encompasses the ability to imitate existing technologies as well as to use outside knowledge as the basis for internal innovative activities. As Cohen and Levinthal (1989) argue, developing absorptive capacity is one of the motivations 19Skilled workers wage bill shares (employment shares) are defined as the ratio of the sum of wages (number of workers) of all types of workers which have an average number of years of education larger than the number of years needed to complete secondary school to the total wage bill (total number of employees). The implicit assumption made in constructing these measures is that for a given firm, all workers of a certain type have exactly the average level of education reported for that worker type by the firm in the survey. 20Wage bill shares for this education-based measure of skills are not available for Brazil. 9

11 for firms to engage in research and development expenditures (R&D). Accordingly, we use a dummy variable for R&D as a measure of the firm s absorptive capacity. We also use a dummy variable for the recent introduction of new product lines as another measure of the firm s absorptive capacity. Doing so captures the learning capacity accumulated by firms in the processes of imitation and technology adoption, even if strictly defined R&D activities are not performed. We use six measures of international economic activities: dummy variables for firms that use imported inputs, firms that export, and firms with FDI and the share of inputs that are imported, the share of sales that are exported, and the FDI share. 21 Panel B of Table 2 shows summary statistics for the absorptive capacity measures and the international activities. The estimation of Equation (3) also requires the use of firm-level data on value-added (sales minus materials costs), capital stocks measured by the book value of machinery and equipment, and a proxy for managerial quality given by three dummy variables for the level of education (secondary, college, or graduate) of the manager. 22 Finally, the industry dummy variables represent the industries in Table 1 and the region dummy variables represent 13 states in Brazil and five cities in China. 23 Panel C of Table 2 shows, for each country, the difference (i) between the average wage bill share of skilled workers of firms with absorptive capacity and of other firms, and (ii) between the average wage bill share of skilled workers of firms engaged in international activities and of other firms. In Brazil, there is a positive and significant correlation between the wage bill share of skilled workers and both measures of absorptive capacity and international activities. In China, absorptive capacity measures are also positively and significantly associated with wage 21For Brazil, we use data on direct exports (not channelled through a distributor) while for China we use data on total exports since the survey does not distinguish between direct and indirect exports. The findings for Brazil are similar when using total exports. 22To reduce the influence of extreme values, we exclude from the estimation the top and bottom 1% of observations for the ratio of capital to value-added in each industry. 23For brevity, we do not show in Tables 3-6 the estimated effects of the dummy variables for R&D, for new products, and for manager education, the capital to value-added ratio, value-added, and industry and region dummy variables. Those effects are discussed in Section 4.5 and are available upon request. 10

12 bill shares of skilled workers. Exports are negatively and significantly associated with wage bill shares of skilled workers, while imported inputs and FDI exhibit an insignificant correlation with those wage bill shares. 24 In sum, there are important differences in wage bill shares depending on the international activities of firms and on the country. Next, we explore these differences in a regression framework. 4. Regression Results In this section, we discuss the results from estimating Equation (3) separately for each country by ordinary least squares(ols)with standard errors corrected for heteroskedasticity (White correction). 25 There is a trade-off between T 2 ijr being a single international activity versus it being a vector of three international activities. Since the activities are correlated across firms, the regressions where T 2 ijr is a vector of activities may suffer from a multicollinearity problem. However, when focusing separately on each activity, the estimated effects may suffer from an omitted variables bias. Hence, we estimate specifications where T 2 ijr represents either (i) a single international activity (Table 3) or (ii) a vector of three international activities (Table 4) Imported Inputs Panels A and B of Table 3 shows the results from estimating Equation (3) including a dummy variable for firms that use imported inputs or the share of imported inputs. The use of imported inputs has a positive and significant effect on Brazilian firms wage bill shares and employment shares of skilled workers. The largest elasticity effect of the dummy is verified for the employment share of workers with college education (35%). 26 The elasticity effects are 24By performing decompositions of the overall variation in skilled labor demand as in Berman et al. (1994), we find that it is mostly due to variation across firms within industries, rather than to variation between industries in Brazil and China. These decompositions are available upon request. 25For Brazil, OLS estimation is weighted by survey weights. Using the test suggested by Deaton (1997, p. 72), we find that the use of unweighted OLS for Brazil is not appropriate. 26The elasticity effects discussed in Sections are calculated as the ratio of the coefficient on the relevant dummy variable to the average of the corresponding skilled labor share in the group of firms for which the dummy variable is zero. For example, the elasticity effect of the imported inputs dummy on wage bill shares of nonproduction workers is equal to the ratio of the coefficient on the imported inputs dummy to the average wage bill share of nonproduction workers for firms that do not use imported inputs. 11

13 smaller for employment shares than for wage bill shares. This finding suggests that the relative wages of skilled workers may be higher in firms that use imported inputs, thus reinforcing the effect of the latter on employment. Substantially different results are obtained for China where the elasticity effects of imported inputs on skilled labor demand are negative and significant and largest for employment shares of nonproduction workers (-14%). Chinese firms that use imported inputs concentrate in production activities that are more unskilled labor-intensive than their counterparts in the industry. The effects of imported inputs on wage bill shares of skilled workers are negative but insignificant as in Brazil, suggesting that while Chinese firms that use imported inputs employ relatively fewer skilled workers, they pay them higher wages than other firms Exports Panels C and D of Table 3 report the results when the export dummy or the share of exports are included in Equation (3). Panel C shows that in Brazil exporters demand relatively more skilled workers than nonexporters. The positive and significant elasticity effect of the export dummy is largest for workers in managerial, engineering, and technical occupations (24%). These findings provide some evidence of a role of exports in transmitting skill-biased technology. However, for employment shares the coefficients on the export dummy are insignificant. Thus, the effect on wage bill shares could result from exporters paying relatively higher wages to skilled workers. Panel D shows that firms with higher shares of exports have lower wage bill shares and employment shares of skilled workers. This result suggests that, within their industry, exporters specialize according to their comparative advantage in relatively unskilled labor-intensive goods. To investigate further these disparate results, we estimate Equation (3) including three dummy variables for firms that export (i) less than 10%, (ii) between 10% and 50%, or (iii) more than 50% of their output. The results show that firms that export less than 50% of their output 12

14 demand relatively more skilled workers, while firms exporting more than 50% of their output demand relatively less skilled workers. Thus, the results with the export dummy are driven by the former firms while the results with the share of exports are driven by the latter firms. The results for China show a negative and significant effect of the dummy and the share of exports on skilled labor demand. The strongest elasticity effect is found for employment shares of workers in managerial, engineering, and technical occupations (-29%). This evidence suggests that, within industries, Chinese exporters specialize according to their comparative advantage in unskilled labor-intensive goods. This effect more than compensates for the technology diffusion impact of exports. The elasticity effects of the export dummy are more negative for employment shares than for wage bill shares, suggesting that while exporters employ relatively fewer skilled workers, they pay them relatively higher wages FDI Panels E and F of Table 3 show the results when the FDI dummy or the FDI share are included in Equation (3). Foreign-owned firms exhibit a significantly higher skilled labor demand in Brazil. The elasticity effect of FDI is largest for wage bill shares of workers in managerial, engineering, and technical occupations (59%). Thus, FDI is a source of skill-biased technology diffusion in Brazil. The elasticity effects of FDI are larger for wage bill shares than for employment shares, suggesting that foreign-owned firms pay higher wages to skilled workers than domestic firms. The skilled labor demand of foreign-owned firms does not differ significantly from that of domestic firms in China. Hence, while FDI may play a technology diffusion role in China, that effect is offset by the specialization of foreign-owned firms in unskilled labor-intensive goods Multiple International Economic Activities Sections consider the effect of each international activity on skilled labor demand of 13

15 Brazilian and Chinese firms. Kraay et al. (2006) argue that these activities are highly correlated. Thus, focusing on a single activity may over or underestimate its true effects. Hence, we also estimate Equation (3) including simultaneously the three activities. Table 4 shows the results from regressions that include either dummy variables for the use of imported inputs, exports, and FDI, or shares of imported inputs, exports, and FDI. Overall, our results do not seem to be driven by the type of biases highlighted by Kraay et al. (2006). The estimates for Brazil show, as in Table 3, that the use of imported inputs, exports, and FDI contribute to higher skilled labor demand and so do higher shares of imported inputs and of FDI. Higher shares of exports are linked to lower skilled labor demand. The magnitude of the coefficients is smaller when the three activities are considered simultaneously. The estimates for China show, as in Table 3, that firms that export or import inputs demand less skilled labor than other firms. The coefficients on imported inputs are smaller in magnitude and less significant when the three activities are considered simultaneously. The coefficients on the FDI dummy and on FDI shares are larger in magnitude and significant for some skills measures, whereas they are insignificant in Table Other Effects on Skilled Labor Demand We now describe the estimated effectsofothervariables includedinequation(3),whichare similar regardless of the international activity considered. We find that firms doing R&D or introducing new product lines have higher skilled labor demand in Brazil and China, confirming the skill-biased nature of knowledge activities. The coefficients on the capital to value-added ratio are generally negative in Brazil and China indicating substitutability between capital and skilled labor. The sign of the value-added coefficient depends on the skills measure used. In Brazil, those coefficients are negative (positive), indicating decreasing (increasing) returns to scale for the employment share of workers in managerial, engineering, and technical occupations (wage bill share of nonproduction workers). In China, the coefficients are negative and significant 14

16 for three of the six dependent variables. The level of education of the firm s manager is an important determinant of skilled labor demand. Positive and generally significant coefficients are estimated on the dummy variables for managers with college education and managers with graduate education, relative to managers with lower educational levels. 5. Robustness To assess the robustness of our results, we consider different empirical specifications and estimation techniques Tobit Estimation and Disaggregated Industries The dependent variable in Equation (3) is the share of skilled workers in wages or employment and thus is restricted to vary between 0 and 1. When using OLS, the right-hand side of Equation (3) is unrestricted. If the variance of the error term in Equation (3) is large, then the estimated coefficients could be biased. To address this problem, we use Tobit estimation for Equation (3). The results are very close to those in Tables 3-6, suggesting that the restrictions on the dependent variable do not bias the estimates. To account for finer differences in skilled labor demand across industries, we estimate Equation (3) controlling for industry fixed effects at a more disaggregate level than that considered in Tables 3-6: 4-digit ISIC level (revision 3). The effects of imported inputs, exports, and FDI on skilled labor demand are qualitatively similar to those discussed in Section 4 for both countries. These findings show that international activities help explain the heterogeneity in skilled labor demand prevalent even across firms within very disaggregated industries Relative Wages and Relative Employment The results in Section 4 suggest that the use of imported inputs, exports, and FDI are associated with higher relative wages for skilled workers. Indeed, the elasticity effects of wage 27The results described in this section but not shown in the paper are available upon request. 15

17 bill shares of skilled workers with respect to these activities are generally larger than the corresponding elasticity effects of employment shares. To separate the effects of the use of imported inputs, exports, and FDI on wages and on employment of skilled workers, we estimate Equation (3) using relative wages or relative employment as the dependent variable. 28 Higher shares of imported inputs and exports are linked to higher relative wages of skilled workers in Brazil, but FDI shares do not affect those significantly. The effects of imported inputs and FDI on relative employment of skilled workers in Brazil are positive and significant, while the effect of exports is negative but insignificant. 29 The three international activities are associated with higher relative wages for skilled workers in China. However, the effects on relative employment of skilled workers are negative and significant for the shares of imported inputs and exports and insignificant for FDI shares. Qualitatively similar findings for Brazil and China are obtained using dummy variables for the use of imported inputs, exports, and FDI Industries with Comparative Advantage The effects of international activities on skilled labor demand may depend on whether the firm belongs to an industry with comparative advantage. We construct a dummy variable identifying industries with comparative advantage, based on revealed comparative advantage indexes for each 4-digit ISIC industry (revision 3) in Brazil and China. 30 A variant of Equation (3) is estimated where the activity T 2 ijr enters both separately and interacted with the dummy for comparative advantage industries. The results suggest that firms that use imported inputs demand relatively fewer (more) skilled workers in China (Brazil), regardless of whether they 28This approach follows Harrison and Hanson (1999) who estimate relative wage and relative employment regressions by OLS based on the cross-sectional variation in their panel of Mexican plants since there is very little time series variation in many of the [...] independent variables (p. 146). 29When estimating the effect of the three export dummy variables definedinsection4.2onrelativeemployment,wefind that only the coefficient on the dummy for firms exporting more than 50% of their output is negative. However, higher relative wages of skilled workers are found for exporters regardless of their export intensity. 30We use detailed trade flows for Brazil, China, and the world taken from World Integrated Trade Solution (WITS) to calculate the revealed comparative advantage indexes in the year prior to the survey as in Yeats (1998). Industries with values for those indexes larger than 1 have revealed comparative advantage. 16

18 belong to industries in which the country has a revealed comparative advantage. Exports tend to reduce skilled labor demand to a larger extent in industries with comparative advantage in Brazil and in China. Brazilian firms with FDI demand relatively more skilled labor across all industries, while Chinese firms with FDI demand relatively less skilled labor in industries with comparative advantage Instrumental Variables Our three international activities may be endogenous in Equation (3) due to the presence of reverse causality. For example, FDI may be directed toward firms that already exhibit a relatively higher share of skilled workers. Also, imports of intermediate inputs and exports are more likely to be pursued by firms whose workers are more educated. Our control for manager education in Equation (3) partly addresses this potential endogeneity bias. Moreover, the endogeneity concern is not too worrying for our main findings of opposite effects of international activities on skilled labor demand in Brazil versus China, unless the endogeneity bias is different for the two countries, which is unlikely. Nevertheless, we estimate Equation (3) using IV and show the results in Table Overall, the results confirm our findings in Section 3 from OLS estimation. Panel A of Table 5 shows that the effects of imported inputs estimated using OLS or IV are qualitatively similar for Brazil and China. Quantitatively, however, the effects are considerably larger using IV estimation. Panel B of Table 5 shows that for China the effects of the share of exports on skilled labor demand are qualitatively similar using OLS or IV estimation. However, for Brazil the IV estimates of the effect of exports are insignificant. Panel C of Table 5 shows that the effects of FDI on skilled labor demand from IV estimation are positive in Brazil and larger in magnitude than those from OLS estimation. For China, the IV 31The instruments used are mostly based on industry-region averages of some of the endogenous variables and are shown in Table 5. For Brazil, we also use industry-level tariffs, import penetration rates, and export shares taken from Muendler (2003a) and Muendler (2003b) as instruments. For China, we also use industry-region fractions of firms that have been restructured into shareholding as an instrument. 17

19 estimates of the effects of FDI are positive but significant only for two skills measures. For the three international activities and both countries, the specification tests support the validity of the instruments used. 6. Conclusion International activities linked to trade and investment can have two important effects on skilled labor demand. First, these activities act as channels for the diffusion of skill-biased technology. Second, these activities are also associated with the exploitation of the country s comparative advantage in the production of unskilled labor-intensive goods. In this paper, we investigate the relative importance of these two effects on the skilled labor demand of Brazilian and Chinese firms, considering three international activities: the use of imported intermediate inputs, exports, and FDI. Our main findings are as follows. First, we find that both the use of imported inputs and FDI have a positive effect on skilled labor demand in Brazil. These activities act as channels for the diffusion of skilled-biased technology. Second, we find a negative effect of exports on skilled labor demand of Brazilian firms that export more than half their output. However, the effect of exports is positive for the remaining exporters. Thus, for the former (latter) firms, the effect of specialization in unskilled labor-intensive production is larger (smaller) than the effect of exports in diffusing technology. Third, in contrast to the evidence for Brazil, the use of imported inputs and exports are negatively associated with skilled labor demand in China. Hence, for Chinese firms engaged in these activities, the effect of specialization according to the country s comparative advantage in unskilled labor-intensive goods is larger than the effect of skill-biased technology diffusion. Fourth, the impact of FDI on skilled labor demand in China is weak, indicating that the role of FDI as a source of skill-biased technology diffusion is offset by the specialization of firms with FDI in unskilled labor-intensive goods. 18

20 In sum, the effects of international activities on skilled labor demand depend on the country. Relative factor endowments may explain the differences across Brazil and China. Wood (1997) argues that the integration into world markets of countries with the highest abundance in unskilled labor (China) led countries with relatively smaller abundance in unskilled labor (Brazil) to shift their comparative advantage to goods of higher skill intensity. Hence, if importers and exporters play to their strengths, then imported inputs and exports are associated with a higher (lower) skilled labor demand in Brazil (China). Whether this argument extends to other developing countries is an interesting avenue for future research. As developing countries increase their integration into world markets and exploit their comparative advantage, the number of firms using imported inputs, exporting, or having FDI will likely increase. Our findings suggest that the impact of this integration on wage inequality between skilled and unskilled workers depends on the importance of the effects of technology diffusion and specialization according to comparative advantage. REFERENCES Autor, David, Katz, Lawrence, and Alan Krueger Computing Inequality: Have Computers Changed the Labor Market? Quarterly Journal of Economics 113: Barro, Robert and Jong-Wha Lee International Data on Education Attainment Updates and Implementation. NBER Working Paper Berman, Eli, Bound, John, and Zvi Griliches Changes in the Demand for Skilled Labor within U.S. Manufacturing: Evidence from the Annual Survey of Manufactures. Quarterly Journal of Economics 109: Berman, Eli, Bound, John, and Stephen Machin Implications of Skill-Biased Technological Change: International Evidence. Quarterly Journal of Economics 113: Berman, Eli and Stephen Machin Skill-Biased Technology Transfer Around the World. Oxford Review of Economic Policy 16: Bernard, Andrew and Bradford Jensen Exporters, Skill Upgrading, and the Wage Gap. Journal of International Economics 42: Bernard, Andrew and Bradford Jensen Exceptional Exporter Performance: Cause, Effect or Both? Journal of International Economics 47:

21 Bernard, Andrew, Jensen, Bradford, and Peter Schott. 2006a. Survival of the Best Fit: Exposure to Low-Wage Countries and the (Uneven) Growth of U.S. Manufacturing Plants. Journal of International Economics 68: Bernard, Andrew, Redding, Stephen, and Peter Schott. 2006b. Comparative Advantage and Heterogeneous Firms. Review of Economic Studies forthcoming. Bloningen, Bruce, and Matthew Slaughter Foreign-Affiliate Activity and US Skill Upgrading. Review of Economics and Statistics 83: Choi, Kang-Shik and Jinook Jeong Technological Change and Wage Premium in a Small Open Economy: The Case of Korea. Applied Economics 37: Cohen, Wesley, and Daniel Levinthal Innovation and Learning: The Two Faces of R&D. Economic Journal 99: Deaton, Angus, The Analysis of Household Surveys: A Microeconometric Approach to Development Policy. The Johns Hopkins University Press. Doms, Mark, Dunne, Timothy, and Kenneth Troske Workers, Wages, and Technology. Quarterly Journal of Economics 111: Fernandes, Ana M. and Alberto Isgut Learning-by-Exporting Effects: Are They for Real? World Bank mimeo. Feenstra, Robert, and Gordon Hanson. 1996a. Globalization, Outsourcing, and Wage Inequality. American Economic Review 86: Feenstra, Robert, and Gordon Hanson. 1996b. Foreign Investment, Outsourcing and Relative Wages. In Political Economy of Trade Policy: Essays in Honor of Jagdish Bhagwati, eds. Feenstra, Robert, Grossman, Gene, and Douglas Irwin. MIT Press. Cambridge. Feenstra, Robert, and Gordon Hanson Foreign Direct Investment and Relative Wages: Evidence from Mexico s Maquiladoras. Journal of International Economics 42: Grossman, Gene, and Elhanan Helpman Innovation and Growth in the World Economy. MIT Press Cambridge, MA. Harrison, Ann, and Gordon Hanson Who Gains from Trade Reform? Some Remaining Puzzles. Journal of Development Economics 59: Haskel, Jonathan, and Ylva Heden Computers and the Demand for Skilled Labour: Industry and Establishment-Level Panel Evidence for the UK. Economic Journal 109: Keller, Wolfgang, Absorptive Capacity: On the Creation and Acquisition of Technology in Development. Journal of Development Economics 49: Keller, Wolfgang International Technology Diffusion. Journal of Economic Literature, 42: