WPS6419 Policy Research Working Paper 6419 Title Export Liberalization, Job Creation and the Skill Premium Evidence from the U.S.-Vietnam Bilateral Trade Agreement Emiko Fukase The World Bank Development Research Group Agriculture and Rural Development Team April 2013 Policy Research Working Paper 6419 Abstract This paper explores how the expansion of labor-intensive issue of endogeneity, the results confirm the existence of manufacturing exports resulting from the United States– a Stolper-Samuelson type effect. That is, those provinces Vietnam Bilateral Trade Agreement in 2001 translated more exposed to the increase in exports experienced into wages of skilled and unskilled workers and the relatively larger wage growth for unskilled workers and a skill premium in Vietnam through the channel of labor decline of (or a smaller increase in) the relative wages of demand. In order to isolate the impacts of trade shock skilled and unskilled workers. During the period 2000– from the effects of other market-oriented reforms, a 2004, the skill premium increased for Vietnam’s economy strategy of exploiting the regional variation in difference as a whole in the sample of panel individuals. Thus, the in exposure to trade is employed. Using the data on Stolper-Samuelson type effect appears to have mitigated panel individuals from the Vietnam Household Living but did not outweigh the impacts of other factors that Standards Surveys of 2002 and 2004, and addressing the contributed to the rise in the skill premium. This paper is a product of the Agriculture and Rural Development Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at efukase@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Export Liberalization, Job Creation and the Skill Premium: Evidence from the U.S.-Vietnam Bilateral Trade Agreement Emiko Fukase 1 The World Bank JEL Codes: F16, J23, J24, J31 Keywords — trade liberalization, skill premium, wage, Stolper–Samuelson theorem, Vietnam Sector Board: Poverty Reduction and Economic Management 1 Development Economics Research Group, World Bank. I would like to thank Wim Vijverberg for his advice and support throughout this research; Zadia Feliciano, the late Robert Lipsey and Jonathan Nelson for comments and suggestions; Chad Bown and Will Martin for guidance; and the General Statistics Office of Vietnam for providing the data. I also benefited greatly from comments by Ryan Edwards, Michael Grossman, David Jaeger, Yan Song, Thom Thurston and Merih Uctum. A shorter version of this paper has been published in World Development (2013a). I am grateful to the anonymous referees of World Development and the World Bank for their detailed and very helpful comments. Any remaining errors are mine. 1 1. INTRODUCTION Since Vietnam started its transition from a centrally planned to a market oriented economy under its doi moi (“renovation�) policy in 1986, Vietnam has been among the fastest growing economies with an average annual growth rate of 6.9 percent. 2 Vietnam’s trade, measured by its sum of imports and exports, grew even faster than its Gross Domestic Product (GDP), as the share of trade relative to GDP increased from 23.2 percent in 1986 to 112.5 percent in 2000, 139.0 percent in 2004 and 169.6 percent in 2007 (WDI, the World Bank). However, Vietnam’s industrial employment share initially remained unchanged at around 12 percent of total employment through the end of the 1990s (WDI), suggesting that expansion of trade does not necessarily lead to industrial job creation. In contrast, since the recent millennium, Vietnam’s industrial employment share in total employment grew substantially, rising from 12.4 percent in 2000 to 17.4 percent of total employment in 2004 (WDI). The U.S.-Vietnam Bilateral Trade Agreement (BTA) of 2001, which led to a dramatic expansion of labor-intensive manufacturing exports to the U.S., appears to have contributed to the surge. The standard Heckscher-Ohlin (H-O) theory predicts that, as developing countries are abundant in unskilled labor and scarce in skilled labor, freer trade would lead a developing country to specialize in a sector which uses its unskilled labor intensively, raising labor demand in the latter sector. Its companion theory, the Stolper-Samuelson (S-S) theorem (1941), suggests that the increase in the relative output prices of unskilled-labor-intensive goods relative to skilled-labor-intensive goods would translate into a rise in the relative wages of unskilled labor, narrowing the wage gap between skilled and unskilled workers. However, the validity of the Heckscher-Ohlin-Samuelson (H-O-S) theory has been challenged since, contrary to the prediction of the theory, many developing countries experienced an increase rather than a 2 The average for the period 1986-2011 (the World Development Indicators (WDI), the World Bank). 2 decrease in skill premium after episodes of trade liberalization (Goldberg & Pavcnik, 2007; Harrison, McLaren, & McMillan, 2010). Moreover, most of the empirical research finds little evidence that trade reforms induce labor reallocation across sectors toward unskilled-labor- intensive sectors in developing countries (Goldberg & Pavcnik, 2007). 3 In their extensive review on the distributional consequence of globalization, Goldberg and Pavcnik (2007) view that the H-O-S theorem is generally inconsistent with the empirical evidence and conclude that the direction of research on international trade tends to be shifting from the traditional focus on countries and industries to a new focus on firms and products. However, Goldberg and Pavcnik’s (2007) conclusions are mainly drawn from evidence on import liberalization, and little study has been devoted to how export liberalization resulting from policy changes by countries’ trading partner(s) would affect skill premium in developing countries. The impact of the BTA on Vietnam’s labor market provides an excellent opportunity to remedy this gap in the trade liberalization and wage inequality literature. First, the BTA presents an opportunity to examine, on the export side, how a tariff cut by a country’s trading partner influenced job opportunities and wages of workers with different skill levels. Second, the U.S. tariff cut on Vietnam’s exports was exogenous, 4 sudden and large. Before the BTA, Vietnam’s access to the U.S. market was quite limited since Vietnam faced the U.S. general tariff rate (at around 35 percent in simple average) which was much higher than the U.S. Most-Favored- Nation (MFN) tariff rate of around 4.9 percent (Fukase & Martin, 2000). Immediately after the 3 Examining 25 liberalization episodes, Wacziarg and Wallack (2004) find no systematic evidence that increased trade openness leads to increased labor shifts. A lack of labor reallocation across sectors after trade reforms is also reported for Argentina (Galiani & Sanguinetti, 2003), for Columbia (Attanasio, Goldberg, & Pavcnik, 2004), for Mexico (Feliciano, 2001), and for India (Kijima, 2006; Topalova, 2010). 4 The general tariff rates are for the most part the statutory rates that were applied to U.S. imports under the Tariff Act of 1930 (also known as the Smoot-Hawley Act). After the trade liberalization of the various General Agreement on Tariffs and Trade (GATT) rounds beginning in 1947, the United States has applied the MFN rate in the U.S. tariff schedule to almost all of its World Trade Organization (WTO) and non-WTO trading partners. However, the U.S. retained the general rates primarily against communist countries not participating in GATT (Fukase & Martin, 2000). 3 BTA came into effect on December 10, 2001, the U.S. granted MFN status to Vietnam lowering the tariff rates across the board. As a result, Vietnam’s exports to the U.S., in particular, those of labor-intensive manufacturing goods, expanded dramatically. Starting from a very low level, the U.S. absorbed 38.3 percent of Vietnam’s textiles exports, 56.9 percent of apparel, 16.6 percent of footwear/leather, and 26.2 percent of furniture and miscellaneous manufacturing exports by the year 2004 (the U.N. Comtrade System). Finally, unlike most of the cases in the literature, labor reallocation toward more labor-intensive manufacturing appears to have occurred in the aftermath of the BTA. Since doi moi, Vietnam has undertaken a number of reforms which introduced market forces in determining wages, including labor market reforms, privatization and rationalization of state owned enterprises (SOEs) and a variety of legal, regulatory and institutional reforms (Van Arkadie & Mallon, 2003). This paper undertakes a strategy of isolating the effects of the U.S. tariff cut on wages from the impacts of the domestic reforms, following recent literature to explore regional variation in exposure to trade in analyzing consequences of trade reforms (see, for instance, Castilho, Menéndez, & Sztulman, 2011; Chiquiar, 2008; Coello, 2009; Hanson, 2005; McCaig, 2011; Topalova, 2010; and Wei & Wu, 2001). A contribution of this paper is to model explicitly the change in labor demand induced by exports as a mechanism through which exports would influence wages. I construct an Export Index at the province level, taking account of Vietnam’s provincial industrial composition and its export- and labor-intensities. In order to overcome potential endogeneity, the U.S. tariff cut measure inspired by Topalova (2010) is used as an instrument. Then, using the panel individuals from the Vietnam Household Living Standards Surveys (VHLSS) who were interviewed both in 2002 and 2004, I evaluate how the 4 provincial variation in exposure to trade would have influenced the wage levels of skilled and unskilled workers and the skill premium in Vietnam. Section 2 demonstrates the trends of trade and industrial employment in Vietnam. Section 3 specifies and implements a series of regression models which relate changes in the Export Index to changes in wages of skilled and unskilled workers and in the skill premium. Section 4 concludes. 2. BACKGROUND (a) Trends of Vietnam’s trade and the U.S.-Vietnam Bilateral Trade Agreement The coming into effect of the U.S.-Vietnam BTA in December 2001 and Vietnam’s accession to the WTO in January 2007 contributed significantly to the expansion of Vietnam’s trade. Figure 1 plots the evolution of Vietnam’s exports to the United States. After the United States lifted its embargo in 1994, Vietnam’s exports to the United States grew steadily. Prior to the BTA, Vietnam’s exports to the U.S. were mainly concentrated in primary products such as coffee, shrimp and petroleum whose general tariff rates are zero or close to zero (Fukase and Martin, 2000). However, Vietnam faced almost prohibitive general tariff rates for many manufactured goods. 5 Immediately after the BTA came into force in December 2001, the United States extended normal trade relations and MFN status to Vietnam. As a result, the United States emerged as Vietnam’s top export destination in 2002, with Vietnam’s exports to the United States more than doubling from $1,066 million in 2001 to $2,453 million in 2002. They have continued to increase, reaching $11,903 million in 2008. During the same period, Vietnam’s exports to destinations other than the U.S. also grew rapidly with Vietnam’s total exports to the world rising from $14,483 million in 2000 to $62,685 5 See Fukase and Martin (2000, Table 2) for the differences between general and MFN tariff rates by commodity. For instance, the general tariff rates for apparel were as high as 68.9 percent compared to 13.4 percent for MFN rates. 5 million in 2008. Figure 2 plots the share of the United States in Vietnam’s total exports. Vietnam’s exports to the U.S., which accounted for 7.1 percent of Vietnam’s exports in 2001, jumped to 14.7 percent in 2002, and further increased to 19.6 percent in 2003; since then, the ratio has remained relatively steady at around 19 percent. Thus, it would be reasonable to assume that most of the immediate impacts of the U.S. tariff reduction took effect between 2001 and 2004. (b) Composition of trade Figures 3.1 through 3.3 show the composition of Vietnam’s “industrial� exports to the world (Figure 3.1), to the U.S. (Figure 3.2) and to the rest of the world (Figure 3.3) for the period 2000- 2007. 6 This paper focuses on the “industrial� sector which in turn is defined as mining (Categories 10-14 in the Vietnam Standard Industrial Classification (VSIC) which in turn is based on the International Standard Industrial Classification (ISIC)), and manufacturing (VSIC 15-41). 7 Figure 3.1 demonstrates the importance of export-oriented, labor-intensive 8 manufacturing, such as apparel/textiles, furniture, and footwear/leather, in Vietnam’s export values and growth. Figure 3.2 demonstrates that Vietnam’s exports to the United States are predominantly concentrated in labor-intensive sectors. For instance, starting from a negligible level, the U.S. absorbed 38.3 percent of Vietnam’s textiles exports (VSIC 17), 56.9 percent of apparel (VSIC 18), 9 16.6 percent of footwear/leather, and 26.2 percent of furniture and miscellaneous 6 See Appendix Table A.1 and A.2 for the values of exports to the world and to the U.S. respectively. 7 In my paper, I excluded the (unprocessed) agriculture/forestry/aquaculture sectors (VSIC 1-5) due to the difficulty in estimating the comparable impacts of exports on labor. For the impact of agricultural trade liberalization by Vietnam’s trading partners on cash crop production, see Coello (2009). 8 In this paper, manufacturing goods with relatively high “employment coefficients�, i.e., the number of workers required to produce one billion dong worth of output, are referred to “labor-intensive� goods. See Appendix Table A.4 for the employment coefficients at the VSIC two-digit level. 9 Vietnam’s exports to the United States of apparel and textiles expanded more than twenty-fold from $48 million in 2001 to $1,040 million in 2002, and then nearly doubled again to $2,020 million in 2003. However, the surge in 6 manufacturing (VSIC 36) exports by the year 2004. Figure 3.1 also reveals that the electronics, machinery and transport equipment sector has been one of the Vietnam’s fastest growing export sectors, perhaps stimulated by the process of Vietnam’s WTO accession. Figure 3.2 suggests that Vietnam’s expansion of this sector’s exports to the U.S. was relatively modest immediately after the BTA, but appears to be emerging at a later time, perhaps signaling Vietnam’s changing comparative advantage. 10 Figure 3.1 shows that food products have been an important class of Vietnamese export commodities throughout the period, reflecting Vietnam’s rich agricultural resources (Athukorala, 2009). However, the export growth of food products to the U.S. was inhibited in 2003 and 2004 by the U.S.’s imposition of antidumping duties against Vietnam’s frozen fillets and shrimp (Brambilla, Porto, & Tarozzi, 2009). Overall, mining (of which crude petroleum is Vietnam’s main export commodity) 11 remains one of Vietnam’s leading export sectors and its export values increased substantially starting in the mid-2000s, mainly due to an increase in world oil prices. Figures 4.1 through 4.3 demonstrate my estimates of the number of Vietnam’s workers engaged in producing exports to the world (Figure 4.1), to the U.S. (Figure 4.2) and to the rest of the world (Figure 4.3). The labor contained in exports is calculated using a similar methodology to the standard factor content analysis which in turn was used to estimate the employment effects of trade in Vietnam (see, for instance, Belser, 2000; Jenkins, 2004; Kien & Heo, 2009). The these categories of exports came to a halt in mid-2003 as the U.S. applied quantitative restrictions against them. Subsequent to Vietnam’s accession to the WTO in 2007, Vietnamese quotas were eliminated as the Agreement on Textiles and Clothing under the provisions of the Uruguay Round Agreement expired in 2005, abolishing the quotas against WTO-member exporters (Dimaranan, Duc, & Martin, 2005). Using an applied general equilibrium model, Dimaranan et al. (2005) show that the welfare gains to Vietnam coming from the abolition of the export quotas on textile and clothing would be substantial. 10 The U.S. emerged as Vietnam’s second largest export destination for electronics, machinery and transport equipment goods (after Japan) in 2006. This may reflect the start of operation of new Foreign Direct Investment (FDI) enterprises such as Inter Corporation and Taiwanese electronics contract manufacturers (e.g., Hon Hai Precision Industry Co., Foxconn) (Athukorala, 2009). 11 At the VSIC two-digit level, crude petroleum (VSIC 11) has been Vietnam’s largest single export commodity throughout the period. For instance, it accounted for about 92 percent of Vietnam’s mining exports and 21 percent of Vietnam’s total exports in 2004. 7 employment coefficients, i.e., the number of workers needed to produce one billion dong worth of goods, 12 were calculated using the employment and production data from the annual Enterprise Survey data and exports data from the UN Comtrade System at the two-digit VSIC level for the period 2000-2007. As the Enterprise Survey data do not include household enterprises, Vietnam’s goods are assumed to be exported through formal enterprises. 13 The visual comparison between Figures 3.1-3.3 and Figures 4.1-4.3 reveals that the employment effect of exports is disproportionately larger for labor-intensive industries due to the relatively larger employment creation per dollar of goods exported in such industries. In contrast, the employment effect of increases in the value of mining sector exports is relatively modest due to that sector’s low labor-intensity despite the importance of the mining sector in Vietnam’s exports in value terms. Since Vietnam’s exports to the U.S. consist mainly of labor-intensive goods, if exports are evaluated in terms of their labor content, the share of the U.S. in Vietnam’s “industrial� exports for the year 2004 rises to about 28 percent. Specifically for the period 2001- 2004, exports to the U.S. accounted for about 35 percent and 70 percent of changes in Vietnam’s total industrial exports in terms of export values and labor content respectively. In the empirical section, I will use these large changes to identify the impact of exports on wages. Appendix Table A.3 reports the evolution of Vietnam’s imports by the VSIC sub-categories for the period 2000-2008. Vietnam’s imports rose from $15,637 million in 2000 to $31,969 million in 2004, and, after Vietnam’s accession to the WTO in January 2007, increased at an accelerated pace to $80,714 million in 2008. Since United States goods enjoyed MFN tariff status in 12 The implicit assumption of this exercise is that export products are typical of their industries. However, Figures 4.1-4.3 may be understating the labor content of exports if the exports were relatively more labor-intensive than other non-exported products within the same industries. On the other hand, the figures may be overestimating the labor content of exports as exporting firms tend to be more productive and thus use less labor to produce a given value of output than non-exporters. 13 Some household enterprises sold directly to the international market. Out of 4,326 households who reported non- farm household activities in the VHLSS 2004, 50 households (about 1.2 percent) responded affirmatively to the question asking whether they “have sold goods/services on international market�. 8 Vietnam before the BTA, the share of the U.S. in Vietnam’s imports has remained relatively unchanged before and after the BTA at around 2.5 percent. 14 The composition of Vietnam’s imports is characterized by a high proportion of three VSIC sub-categories, namely, “basic manufacturing�, “chemical, rubber, and plastic products�, and “electronics, machinery, and transport equipment� which accounted for more than three quarters of Vietnam’s imports in 2004. If the latter products are capital and skilled-labor intensive, Vietnam’s trade pattern is consistent with the prediction of the H-O model. (c) Evolution of Vietnam’s industrial employment Vietnam’s “formal� 15 industrial employment recorded in the Enterprise Survey data (General Statistics Office (GSO)) grew substantially from 1.8 million workers in 2000 to 3.2 million in 2004 and further increased to 4.1 million in 2007. Figure 5 and Appendix Table A.5 show the actual number of workers employed by industrial enterprises for the period 2000-2007. It appears that employment growth is highly influenced by the expansion and the factor intensity of exports. The number of workers employed by export-oriented labor-intensive manufacturing, and in particular, those in the sectors whose exports to the U.S. expanded (namely apparel/textiles, footwear/leather and furniture and miscellaneous manufacturing) increased substantially. Employment growth in the electronics, machinery and transport equipment sector was also strong, perhaps helped by the rise in exports (e.g., of electronics goods). However, despite the high value of Vietnam’s crude oil exports, enterprise employment generation has been slowest in the mining sector due to the low labor intensity of crude oil production. At the two-digit VSIC 14 The rise of the U.S. share in Vietnam’s total imports to 4.5 percent in 2003 is attributable to Vietnam’s purchase of U.S. aircraft (Boeing 777s) (Parker, Riedel, & Quang, 2007). 15 Throughout the paper, jobs in “enterprises� as well as those in government are referred to as “formal� employment. The term “enterprise� is defined as “an economic unit that independently keeps business account and acquires its own legal status� under the relevant laws (GSO). “Informal� employment is defined to include wage workers who work in a variety of informal wage sectors as well as those who are self-employed. 9 level, no industry reveals a systematic employment decline during the same period. These patterns of employment are consistent with Kien and Heo’s (2009) study which investigates formally the impacts of trade liberalization on employment in Vietnam using a system generalized method of moments (GMM) model. 16 Figure 6 and Appendix Table A.6 show how Vietnam’s industrial employment spread across Vietnam’s two big cities and eight regions. Of these regions, Ho Chi Minh City (HCMC), the Southeast region (excluding HCMC), and the Red River Delta region (excluding Hanoi) are host to a majority of Vietnam’s industrial employment, accounting for about 62 percent of total industrial employment in 2004. Growth in industrial employment has been faster in the rest of the Southeast region than in HCMC and in the rest of the Red River Delta than in Hanoi, showing that industrial development has spread beyond Vietnam’s two big cities. In contrast, the North Central Coast, the Central Highlands, and the Northwest regions are clearly lagging behind in terms of industrial development. Vietnam’s establishment of normal trade relations with the United States appears to have induced a shift of its labor toward sectors that use the abundant factor more intensively. Using the Enterprise Survey data, Figure 7.1 shows the proportion of workers employed by “labor- intensive� industries in total industrial employment. 17 The ratio rose from 50.7 percent in 2001 to 53.6 percent in 2003 and then remained relatively unchanged, suggesting that Vietnam’s labor 16 Estimating a labor demand equation derived from the Cobb-Douglas production function for the period 1999- 2004, Kien and Heo (2009) find that, holding the output level constant, the rising export intensity (measured by export-output ratio) in an industry increased Vietnam’s derived labor demand. This implies that an increase in exports may create more job opportunities per unit of output because export-oriented goods are labor-intensive in Vietnam. In terms of imports, whereas intensified import penetration would likely have destroyed some jobs, it may also have stimulated labor demand positively when domestic production depends on the importation of raw materials and capital goods or imported inputs are assembled in Vietnam for exports (e.g., in the electronics industry). The coefficients of import penetration (measured by import-output ratio) in Kien and Heo’s regressions turn out to be positive but statistically insignificant suggesting that the increase in imports did not necessarily negatively impact Vietnam’s employment level. 17 By the “labor-intensive industries�, I refer the industries whose employment coefficients in 2004 are above median, which is 6.5 persons per billion dong of output. 10 appears to have shifted toward more labor-intensive industries in the aftermath of the BTA. Alternatively, Figure 7.2 demonstrates the change in the proportion of unskilled-labor-intensive industries in total industrial enterprise employment during the same period. 18 Figure 7.2 reveals a similar pattern, a shift of labor toward unskilled-labor-intensive industries paralleling the shift toward labor-intensive industries shown in Figure 7.1. However, this parallelism is expected since in Vietnam, labor-intensive industries are generally unskilled-labor-intensive industries. During the same period, “formal� industrial employment as a whole expanded substantially. Thus, besides labor reallocation within the industrial sector (if it occurred), there appear to have been two additional sources of labor supply for the expansion of labor-intensive manufacturing. First, the increase in industrial employment may reflect a shift of labor from agriculture to industry. While industrial employment grew rapidly in the recent years, agricultural employment declined from 65.3 percent in 2000 to 57.9 percent in 2004 (WDI). It is likely that this move was partly induced by trade openness (Dodzin & Vamvakidis, 2004; Fu & Balasubramanyam, 2005). 19 Second, the expansion of formal enterprise employment is likely to reflect a move of labor from the “informal� to the “formal� sector. The implementation of the Enterprise Law (2000) was a major step toward reducing numerous barriers to private business and has greatly encouraged the establishment of new enterprises (Ramstetter & Ngoc, 2007). It is also likely that this trend was partly accelerated through trade, as the largest expansion of industrial enterprise 18 Unfortunately, the Enterprise Survey data 2000-2007 do not break down different types of workers either in terms of education or occupation. However, the Industrial Survey (GSO, 2000) breaks down employees of 17 provinces into production workers, technicians and administrative workers for the year 1998. The “unskilled-labor intensive industries� here refer to the industries whose proportion of production workers in total labor force is above median (85.7 percent) in 1998. 19 Applying the Smith-Myint model of “vent for surplus� to China, Fu and Balasubramanyam (2005) find that the expansion of exports from labor-intensive manufacturing has accelerated the transfer of surplus labor from the agricultural to the export sector. Using a panel of 92 developing countries in the period 1960-2000, Dodzin and Vamvakidis (2004) find that an increase in openness to trade leads to an increase in the industrial value added share of production at the expense of the agricultural share suggesting that trade leads developing countries to industrialization. 11 employment occurred in export-oriented sectors. 20 As labor reallocation appears to occur across broad economic sectors, analyses which focus only on the formal industrial sector seem to be incomplete in capturing the whole impact of trade shock. The VHLSS 2002 and 2004 data used for analyzing wage movements in this study have an advantage relative to firm-level data as they allow the researcher to extend the analysis to all of Vietnam’s economic sectors. 3. THE IMPACTS OF THE BTA ON WAGES AND THE SKILL PREMIUM (a) Estimation strategy This section investigates empirically how the exogenous trade shock resulting from the BTA translated into changes in wages for skilled and unskilled workers. Whereas this paper is inspired by the basic insights from the H-O-S model, its analysis does not follow the strict version of the H-O-S theory. Instead, I relax some of its assumptions so that the conceptual framework is consistent with Vietnam’s “real world� data and my empirical design. First, whereas the Stolper- Samuelson theorem links changes in relative output prices to changes in relative wages, it is difficult to model this linkage directly since price data are not readily available. Thus, I take an indirect approach assuming that the impacts of changes in product prices induced by trade are “revealed� in changes in labor demand. Second, this paper extends the idea of the H-O-S theory to Vietnam’s provinces; i.e., it is assumed that Vietnam’s provinces have different specialization patterns given heterogeneity in endowments and difference in initial industrial development, and that the impacts of trade shocks translate differently at the province level. The model is designed to capture general equilibrium impacts of the trade shock within a province: it is assumed that labor is mobile across sectors and that factor prices tend to be equalized within province. In regard to labor mobility across 20 Discussing a similar situation, Anderson and Dimon (1999) report evidence that export oriented production in maquiladoras (across-border export processing plants) created formal sector job opportunities for single women in Mexico. 12 provinces, it is assumed that Vietnam’s workers are “sufficiently immobile� (Hanson, 2005, p.4) across provinces as province-specific labor demand shocks influence the wages of workers living in that province. 21 Overall, the empirical framework is similar to the model employed by Chiquiar (2008) who finds responses of the “Stolper-Samuelson type� effect following the North American Free Trade Area (NAFTA), by exploiting the regional variations in exposure to international markets. 22 Third, the S-S results may become less pronounced if the supply of unskilled labor is elastic. Winters (2002) presents a useful way of thinking about the impacts of trade on wages for unskilled workers by considering two polar forms of labor markets in developing countries. The first is that assumed by traditional trade theory in which factor supplies are exogenously fixed and wages are perfectly flexible (Winters, 2002, p.1348-1350). In this case, the Stolper- Samuelson Theorem, under particular conditions, generates the powerful result that an increase in the price of the unskilled labor-intensive good in production will increase the unskilled real wage and decrease that of skilled workers. The polar opposite view of labor markets for unskilled workers is one suggested in development theory that considers factor supplies to be infinitely elastic (Lewis, 1954; Winters, 2002, p.1350-1352). In the latter case, the formal sector can draw infinite amounts of labor from the informal sector or subsistence agriculture at a subsistence wage, such that the wage for unskilled workers is unaffected by the trade shock. 21 It appears that Vietnam’s labor is neither “perfectly� mobile nor “perfectly� immobile. The 1999 and 2009 census indicates that interprovincial migration accelerated in the past decade as the number of people aged five and older who had migrated across provinces within the census period increased from 2.0 million individuals (or 2.9 percent of population) in the 1999 census to 3.4 million (4.3 percent) in the 2009 census (GSO, p.21, 2011). The rest either moved intra provincially or stayed in the same location. The rise in interprovincial migration is not inconsistent with the theory since a large majority of people do not migrate and it appears to be precisely the disparity in earnings opportunities which would induce some individuals to migrate (Fukase, 2013b). 22 Chiquiar (2008) used one percent samples of male individuals taken from Mexico’s 1990 and 2000 population censuses and complemented them with site-specific data for globalization-related variables. Specifically, “globalization� is measured in Chiquiar’s study by the shares of state employment in agriculture and industry, the distance from the U.S., the share of FDI in the state GDP, the shares of maquiladora employment and imports, and migration rates. Controlling for personal characteristics in his cross section data, Chiquiar (2008) finds that Mexican states more exposed to globalization experienced a decrease in the skill premium relative to the other states. 13 Winters posits that neither of the polar extremes is likely to be precisely true and suggests the importance of determining the elasticity of labor supply in evaluating the supply response of unskilled workers. Thus, theoretical predictions with regard to the effects of trade on wages of unskilled workers are ambiguous, and the existence of S-S effects becomes an empirical question. (b) Data (i) The Vietnam Household Living Standards Surveys The VHLSS 2002 and 2004 were conducted by the GSO of Vietnam with the technical support of the World Bank and are generally recognized to be of high quality and are representative of all of Vietnam. The VHLSS 2002 and 2004 consist of 30,000 households and 9,188 households respectively, and about half of the households interviewed for the VHLSS 2004 came from the sample of the VHLSS 2002. The timeframe of the VHLSS 2002 and 2004, which reflect information for the years 2001/2002 and 2003/2004 respectively, overlaps reasonably with the period when Vietnam’s exports to the U.S. expanded in the aftermath of the BTA (see Figure 2). During the same period, protection appears to have been still in place for import-competing production. 23 Thus, the focus of the paper is to evaluate the impacts of trade on wages on the export side. Using the sampling weights available in the VHLSS 2004, out of 47.3 million people who had work in 2004 in Vietnam, about 14.6 million people (31.0 percent of total workers) held wage jobs of which 7.8 million (16.6 percent) and 6.8 million people (14.4 percent) worked in 23 For instance, the trade weighted Nominal Rate of Protection (NRP) for manufacturing was relatively high at 29.2 percent in 2003 (Athukorala, 2006, Table 3). Although the estimated Effective Rate of Protection (ERP) for manufacturing decreased from 95.9 percent in 2001 to 43.9 percent in 2003, the decline came from an increase in input tariff introduced to protect SOE engaged in intermediate production rather than from the decrease in tariff rates for final goods (Athukorala, 2006, p.174). 14 “formal� and “informal wage� sectors respectively. 24 The remaining 32.6 million workers (69.0 percent) were self-employed. In 2004, Vietnam’s workers on average had attained 7.7 years of education, and worked 34.8 hours per week, 25 and its wage workers earned .94 million dong per month. Formal sector workers worked longer hours (41.8 hours on average) than those who worked in the informal sectors (35.1 hours and 33.0 hours for those in the informal wage and self-employment sectors respectively). On average, workers in the formal sector were more educated with average total education of 11.1 years, whereas the corresponding figures were lower for the informal wage sector (6.6 years) and self-employed workers (7.1 years). In 2002, applying the sampling weight in the VHLSS 2002, only 29.6 percent of total workers had wage jobs: the increase in the proportion of wage workers in total workers is attributable to the rise in the proportion of workers employed in the formal sector from 14.2 percent in 2002 to 16.6 percent in 2004. One limitation of using the VHLSS is that self-employed individuals cannot be included as they have no wage information. Whereas it is possible to calculate income at the household level, it turns out to be difficult to attribute household income to individual household members. However, as the VHLSS includes wage information for informal wage workers, this enables the researcher to extend the sample coverage beyond the formal sector. The data are also limited in that they are likely to exclude a substantial number of migrants who hold a temporary or no 24 The classification of employment is based on the response to a question “for whom has [name] worked?� in the VHLSS 2004: those who responded to work in “foreign-invested sector�, “state�, “collective economic sector� and “private economic sector� are classified as “formal�; “informal wage� workers are assumed to be represented by those who responded to “work for other households�; “self-employed� includes those who answered “self- employment� and “working in his family owned company�. For 2002, based on the VHLSS 2002, those who responded to work for “government agencies, police, military�, “communist party, social organization�, “state owned enterprises�, “other state economic sector�, “collective economic sector�, “private capital economic sector�, “state capitalist economic sector� and “foreign shared enterprise� are referred to “formal�; “informal wage� and “self-employed� are represented those who responded to be �self-employed� and “work for other households� respectively. 25 The hours worked per week are computed by combining the hours worked for the first and second jobs. 15 registration status, since the sampling of the surveys is limited to persons who had permanent registration status. 26 In order to trace the wage changes during the period 2002 and 2004, I construct a sample of the panel individuals interviewed both in the VHLSS 2002 and 2004 who were 15-59 years of age in 2004 and had wage jobs as the most time consuming job in both years. 27 This resulted in a sample of 1,746 individuals. Relative to using the repeated cross section data, there exist advantages in analyzing the panel data component of the VHLSS. First, since the panel data model traces the changes in wages of the same individuals between 2002 and 2004, it controls for any impacts on wages resulting from time-invariant observable and unobservable personal characteristics. Second, the panel data approach adjusts for the “composition effect�, i.e., any impacts resulting from the difference in the composition of wage workers between two years. For instance, there is some evidence that the individuals who moved from a non-wage job in 2002 to a wage job in 2004 were less skilled on average than those who had a wage job in both years. 28 If the unobservable component of the selection process to wage employment and the export variable are correlated, the coefficient for the export variable using cross section data may be biased even controlling for observable personal characteristics. 26 For instance, the number of workers employed by foreign firms computed from the VHLSS 2004 (0.7 million) in 2004 is smaller than that taken from the Enterprise Survey data (1.0 million), most likely because the VHLSS data exclude migrants who hold a temporary or no registration status. According to the Vietnam Migration Survey 2004 data, only 10.2 percent of the migrants who work for foreign firms possess permanent registration status (Fukase, 2013b). 27 Since there exist some errors in the panel identifiers in the original dataset, I used the revised identifier codes provided by McCaig (2009) to match the individuals between 2002 and 2004. Then, I eliminated some observations whose information on gender, age and education are inconsistent. In order to mitigate further the measurement error, I dropped one percent of individuals who experienced the largest wage changes. However, the results are not sensitive to these adjustments. 28 The data on panel individuals reveal that those who had a wage job in both years tend to be more educated (with 9.4 years of education on average) than those who moved from a non-wage job in 2002 to a wage job in 2004 (with 7.3 years of education) (VHLSS 2002, 2004). During the same period, the BTA might have accelerated the rate at which less skilled non-wage workers became wage workers. If that were the case, the wage workers who lived in a province more exposed to exports might have consisted disproportionately of “new� wage workers relative to other provinces. 16 It is common in the literature to classify skilled and unskilled workers either by educational attainment or by occupation (white- vs. blue-collar). There is some difficulty in differentiating Vietnamese individuals into a skilled and unskilled worker dichotomy by educational attainment, however, since it is actually the intermediately educated who are most likely to be engaged in export-oriented production. 29 I therefore use the occupation codes available in the VHLSS which are similar to the International Standard Classification of Occupations (ISCO) by the International Labour Organization (ILO) to categorize individuals into skilled and unskilled workers. 30 The summary statistics of demographic and economic characteristics for skilled and unskilled workers are reported in Appendix Table A.7 along with all the other variables used in my regressions. For my sample of the panel individuals for the period 2002-2004, the average wage growth rate of 27 percent for skilled workers was larger than that for unskilled workers (18 percent), suggesting that the skill premium increased for Vietnam’s economy as a whole. (ii) Export index I first develop an index to represent the province-specific labor demand induced by exports as a mechanism through which trade affects wages. The increase in production for exports may increase wage levels in the location where export firms operate by raising the demand for labor in that location. In addition, it may influence wage levels indirectly, for instance, by increasing 29 Winters (2002) points out that it is not clear that the least skilled workers are the most intensively used factor in the production of tradable goods in developing countries. Thus, it might be the case that the wages of workers with primary education increase with trade liberalization, those of the least educated workers may be left behind (Winters, 2002). The information on whether or not individuals work for exports is not available in the VHLSS. Using production workers who worked for formal firms in export-oriented sectors (namely apparel and textiles, footwear and leather, furniture and miscellaneous manufacturing and electronics) in 2004 as a proxy for those working for export-oriented production, a large majority of these workers were intermediately educated, as 28 percent, 39 percent and 26 percent of them have upper-secondary, lower-secondary and primary education respectively. The workers in these fields who had tertiary education and those who had no school degree were relatively small at 2 percent and 5 percent respectively. 30 “Skilled� workers include leaders (11-19), top- and mid-level professionals (21-34), staff and other white-collar occupations (41-52), and skilled workers in agriculture, sylviculture, and aquaculture (61). “Unskilled� workers consist of production workers (71-83) and those generally classified as “unskilled workers� (91-93). The skill category is not assigned for those who served in “Army force� (00). Between parentheses are the occupation codes available in the VHLSS 2004. 17 the demand for inputs; by raising the demand for trade related services (e.g., transport); and through a multiplier effect if workers employed by exporting firms increase consumption in the area. The data on the number of workers engaged in exports are not available directly at each province level. I construct a province-specific export index (Exportpt), using the exports data at the national level as well as the employment and output data at the firm level. The Export Index is the sum of the workers over each industry j aggregated at the province level p, multiplied by the export intensity of each industry, relative to the total economically active population in each province p at time t. Specifically, Exportpt = ∑ j XS jt ⋅ Employment jpt (1) Employment pt where XSjt is the export intensity computed as the proportion of exports in gross output for industry j at time t at the national level; 31 Employmentjpt is the number of workers employed by industry j in province p at time t; and Employmentpt is the size of economically active population in province p at time t; j represents Vietnam’s industries at the two-digit VSIC level (VSIC 10- 41), p represents Vietnam’s 61 provinces, 32 and t = 2002, 2004 in my regressions. Thus, the impact of exports on labor demand is likely to be larger, the higher is the export-intensity in the composition of industries, and the larger is the share of workers employed by industries relative to total employment in province p. 31 Since the export intensity of each industry is not available at the province level, that at the national level is used. For some industries for some years, the export values from the Comtrade data exceed the production values from the Enterprise Survey data. As a potential source of this incompatibility, Jenkins (2004) points out that the gross output reported in the Enterprise Survey data may refer not to the value of the products produced, but to the processing cost. This is particularly the case of the garment and footwear industries, where exports are commonly contracted on a cut-make-trim (CMT) basis. Following Jenkins (2004), the reported figures for output in the garment and footwear industries are multiplied by 2.5 and 2.0 respectively. 32 Between 2002 and 2004, some provinces were subdivided creating three new provinces, namely Dien Bien, Dac Nong and Han Giang. For the purpose of consistency, the definition of provinces is based on the 2002 configurations. 18 Equation (1) can be re-written as: Exportpt = ∑ j XS jt ⋅ LaborCoeff jpt ⋅ Output jpt (1)' Employment pt Employment jpt where LaborCoeffjpt= Output jpt is the employment coefficient, i.e., the number of workers required to produce one unit of output. Thus, given industrial output, the impacts of exports on labor demand are likely to be larger, the higher is the labor-intensity of the industrial composition in province p. Overall, the Export Index is roughly interpreted as the proportion of workers affected by exports in each province. The employment and output data are taken from the annual Enterprise Survey data (GSO). The surveys cover all the registered “enterprises� in Vietnam but exclude household enterprises and agriculture and forestry cooperatives. In estimating the change in labor demand, the Enterprise Survey data have two advantages over the VHLSS 2002 and 2004. First, whereas the Enterprise Survey data are likely to include migrants, the calculation based on the VHLSS may underestimate the expansion of employment induced by exports, by excluding migrants who have no or temporary registration status. 33 Second, the pre-BTA comparable employment and production data, which are needed to construct an instrumental variable, are readily available in the Enterprise Survey data (for the year 2000 in my analysis). The data for the size of the economically active population aged 15 years and over by province were extracted from the GSO website, since the latter data are not available in either the VHLSS or the Enterprise Survey data for the year 2000. The trade data for Vietnam were extracted from the U.N. Comtrade system at the Harmonized System (HS) (1996) 6-digit level and aggregated at 33 When I use the VHLSS 2002 and 2004, the pattern of the change in industrial employment across sectors is less pronounced relative to the pattern resulting from use of the Enterprise Survey data. 19 the VSIC two-digit level using the correspondence file between HS 96 and ISIC (Rev. 3) which in turn was downloaded from the U.N. Statistics Division website. (c) Impact of a change in exports on wage growth (i) Model specification This section investigates the effects of exports on absolute wages for skilled and unskilled workers running regressions separately by skill levels. I specify a model which relates wage growth rates to changes in the province-specific Exports Index. ∆Ln Wip = β0 + β1 ∆Exportp + β2 ∆Xip + β3 Yip+ eip (2) where ∆Ln Wip is the monthly wage growth rate during 2002-2004 of an individual i who lives in province p, ∆Exportp is the change in Export Index for province p between 2002 and 2004, ∆Xip is a vector of differenced control variables, and Yip is a vector of other control variables which may affect wage growth rates. The monthly wage calculations are based on the annual wages earned and the hours worked by panel individuals during the twelve months prior to the survey.34 Monthly wages rather than hourly wages are employed in the model as they reflect both the compensation per unit of time and hours worked by individuals. Taking account of hours of work is potentially important since the trade-oriented work may involve long hours and the working hours appear to vary substantially in Vietnam across economic sectors. For instance, the panel individuals in the VHLSS reveal some evidence that those who moved from an informal to the formal sector increased their hours of work. 35 34 The wages include bonus/award, social allowances, and trip subsidy. The wages are converted to January 2002 constant prices using the deflators available in the VHLSS. 35 According to the panel individuals in the VHLSS, the workers who moved from the informal wage and self- employed sectors to the formal sector increased their hours worked by 8.3 percent and by 24.3 percent respectively. Unfortunately, the income changes for those who moved from self-employed to formal sector jobs are not reflected in the paper because no 2002 wage information was available for self-employed individuals. 20 An advantage of using a difference-difference specification relative to a model which relates the level of wage to the level of Export Index is that it controls for time invariant characteristics which are correlated both with wage levels and Export Index levels. For instance, there exist a number of variables which are likely to affect positively both wage levels and export levels such as better infrastructure, conducive business and investment climate, and favorable geographical conditions such as better access to seaports. Thus, one might find a spurious relationship between Export Index and wage levels in a level-level model. However, by differencing, all time- invariant province- and personal specific characteristics are controlled for. The first three columns in Table 1 report the Ordinary Least Squares (OLS) regression results of a model which relates the wage growth rate to the change in Export Index and other variables (Model A). Since there are both individual and province level variables in Equation (2), the standard errors are adjusted for within province correlation (clustering). Columns 1, 2 and 3 report the results for the full sample, for the subset of unskilled workers, and for the subset of skilled workers respectively. The distinction between skilled/unskilled categories is based on individual workers’ status in 2002. Between 2002 and 2004, some workers changed their skilled/unskilled status due to changes in their occupations. Thus, a variable reflecting their changes in skilled/unskilled status (= 1 if he/she moved from unskilled to skilled occupation; = 0 if his/her skilled/unskilled status remained unchanged; and = -1 if he/she moved from skilled to unskilled occupation) is included. Similarly, a variable to indicate the change in marital status (= 1 if his/her marital status changed from non-married to married; = 0 if marital status remained unchanged; and = -1 if his/her marital status changed from married to non-married) is added. Since education levels do not change much for adults and everybody is two years older in 2004, human capital variables such as education and potential experience are not included. In order to 21 capture potential trends of wage convergence or divergence, the lagged log of monthly wage is included (Topalova, 2010). For instance, because of wage convergence, individuals who enjoyed higher wage levels in 2002, who tended to reside in relatively high wage provinces, may have experienced slower wage growth. If such provinces were exposed to higher trade shock, without properly adjusting for this trend, β1 would be biased downward. For the full sample, the coefficient for the change in Export Index is found to be positive but statistically insignificant. The results of regressions run separately by the subsets of skilled and unskilled workers reveal differential impacts of exports by skill levels. The coefficient for the change in the Export Index for unskilled workers turns out to be positively significant at the five percent level implying that unskilled individuals who resided in a province which experienced more export expansion tend to experience higher wage growth rates. In contrast, the coefficient for the change in Export Index for skilled workers turns out to be insignificant. The coefficients for the lagged wages turn out to be negatively significant implying that Vietnam’s workers appear to have experienced wage convergence, i.e., the individuals who had higher wages in 2002 experienced on average lower growth in wages. 36 Whereas the differencing process controls for any time invariant characteristics that affect the level of wage, it does not rule out a possibility that these variables influence wage growth rates. Thus, I specify an alternative model which includes additional control variables which potentially affect wage growth (Model B). Although many personal characteristics variables do not change between 2002 and 2004, the level of these variables may influence the wage growth 36 Some caution needs to be exercised for the inclusion of this variable since the coefficients for the change in the Export Index become insignificant without the lagged log of wage. However, a model with the latter variable is a better model because the coefficient for the lagged log of wage turns out to be highly significant and the addition of the variable improves the overall performance of my regressions; since the initial log wage and the change in Export Index is positively correlated, the coefficient for the change in the Export Index would be biased downward without the initial log wage; and controlling for the initial level of the variable of interest is common both in macro growth literature and in micro literature to adjust for convergence or mean reversion. 22 rates, e.g., when the dynamics in a labor market favor certain kinds of workers. Thus, a series of personal characteristics, namely, gender, ethnicity, 37 educational attainment (measured by total education years) and potential experience, 38 are included. Changes in Export Index may be systematically correlated with province specific characteristics that affect wage growth (Topalova, 2010). For instance, if provinces which had more initial industrial base were enjoying higher wage growth even in the absence of exports and also experienced larger industrial exports expansion, the coefficient of the change in Export Index might capture a spurious relationship. In order to overcome this concern, I included the initial provincial employment composition at a more aggregate level, namely the proportion of workers in agriculture, construction, industry, government, and service sector in total workers constructed from the VHLSS 2002 (McCaig, 2011; Topalova, 2010). The growth path also may differ between rural and urban areas. Thus, Model B also includes a dummy variable for whether or not the individual lived in an urban area. Columns 4 through 6 in Table 1 report the results of Model B. Relative to the results obtained by Model A, the inclusion of the additional control variables led to an increase in R2. However, the coefficients for the changes in Export Index in Model B turn out to be similar to that obtained in Model A and the results appear to be robust to the inclusion of control variables. Some personal characteristics are found to have affected wage growth. For instance, the coefficients for education are found to be highly positively significant, highlighting the importance of educational attainment in wage growth. The positive impact of education on wage growth is also consistent with the rise in skill premium in Vietnam’s economy as a whole. 37 The minority is defined to represent all ethnic groups other than the Kinh majority and Hoa (the Chinese). 38 Potential experience is calculated as age minus six minus education years. 23 Columns 7 through 9 in Table 1 considers an alternative Export Index (U.S. Export Index (ExportUSpt)) which estimates the labor demand contained in Vietnam’s exports to the U.S. (Model C). ExportUSpt = ∑ USXS j jt ⋅ Employment jpt (3) Employment pt where USXSjt is the alternative export intensity computed as the proportion of exports to the U.S. in gross output for industry j at time t. By differencing, ∆ExportUSp is designed to capture the change in labor demand created by exports to the United States. The results of Model C reveal a pattern similar to that obtained by Model B. (ii) Instrumental Variables (IV) approach The change in Export Index (∆Exportp) in Equation (2) may be endogenous and the endogeneity of explanatory variable may cause a bias in the parameter estimates obtained by OLS. For instance, many empirical studies focusing on firm heterogeneity find that exporting firms are systematically different from non-exporting firms as they tend to be larger and more productive and tend to pay higher wages (e.g., Bernard, Jensen, Redding, & Schott, 2007). Firms that achieve a higher level of productivity and that have other advantages may tend to self-select into export markets since such firms are better able to overcome the sunk costs of entering international markets (Melitz, 2003). Productivity and wages may be correlated as productive firms may pay higher wages to attract better workers or firms may have invested in human capital among their workers and therefore pay higher wages to protect their investment. As greater productivity is associated with higher wages and with the likelihood of exporting, this may in turn cause a positive correlation between the change in Export Index and the error term. In this case, the coefficient estimate for the change in Export Index would be biased upward. Another potential source of endogeneity is that 24 wage growth may influence the change in exports, as a rise in wages may hamper export expansion. In the latter case, the change in Export Index and the error term would be negatively correlated, causing a downward bias in the coefficient estimate for the change in Export Index. Whatever the source of endogeneity may be, I address the potential endogeneity problem using the province-specific measure of the U.S. tariff reduction as an instrumental variable. The U.S. tariff cuts that each province faces are calculated using a methodology similar to the one developed by Topalova (2010) and applied to Vietnam by McCaig (2011) and Coello (2009). Topalova (2010) examines the 1991 Indian trade liberalization episode and finds that rural districts, in which production sectors more exposed to import liberalization were concentrated, experienced slower reduction in poverty in India. Following Topalova (2010), McCaig (2011) constructs a provincial measure of U.S. tariff reductions and finds that provinces that were more exposed to the U.S. tariff cuts in the aftermath of the BTA experienced faster decreases in poverty. My approach differs from the previous studies in terms of explicitly modeling the rise in labor demand induced by exports as a channel through which exports influence wage growth. 39 In order to overcome potential endogeneity problems, I use the U.S. tariff cut measure as an instrument. Furthermore, I extend the tariff cut measure developed by Topalova by including a 39 McCaig (2011) also examines the impacts of the BTA on provincial wages, but for different education groups. McCaig (2011) shows that the increase in the U.S. tariff cut measure is associated with an increase in the mean provincial hourly wage growth for workers with “at most a primary education� whereas the relationships are not statistically significant for the other education groups. His findings are consistent with a rise in labor demand for unskilled workers, but the study does not show or model a mechanism through which the tariff cut translated into the rise in wages for unskilled workers. Also, since the education groups of “no formal education� and “primary� are aggregated in McCaig (2011), one cannot know whether the U.S. tariff cut is associated with the rise in wages for the least educated workers (who are the most likely to be the poor). In terms of the data, whereas McCaig (2011) used cross section data aggregated at the province level, I used panel individual data. Finally, McCaig (2011) and most of previous studies used hourly wages in evaluating wage change, I used monthly wages which reflect both the compensation per unit of time and hours worked. 25 term representing the export intensity in 2000 since the pass-through from tariffs to prices in each sector is likely to be influenced by the sector’s export orientation. 40 Whereas Vietnam’s provinces all face the same tariffs abroad, the impacts of tariff cuts differ across provinces due to differences in initial industrial composition. The provincial measure of the impacts from the U.S. tariff cuts is the tariff drop rate aggregated at the province level using the structure of employment prior to the BTA (t = 2000). Specifically, TariffCutp = ∑ j XS j , 2000 ⋅ Employment jp , 2000 ⋅ ∆τ j (4) Employment p , 2000 where ∆ιj is the U.S. tariff drop at the two-digit VSIC level, which is taken from McCaig (2011, Figure 2); XSj,2000 is the export intensity which in turn is computed as the share of exports in gross output for industry j in 2000 at the national level; Employmentjp,2000 is the number of workers employed by industry j in province p in 2000; and Employmentp,2000 is the number of economically active population in province p in 2000. 41 The rationale of using the provincial measure of the U.S. tariff cut as an instrument is that the U.S. tariff reduction (∆ιj) is likely to influence the change in Exports Index, but the change in wages does not influence the U.S. tariff reduction. The U.S. tariff reduction is also unlikely to affect the change in wages other than via trade. In addition, the lagged level of industry mix is likely to affect the change in Export Index by influencing the impacts of the U.S. tariff cut at the province level, but the change in wages cannot affect the past level of industry composition. 40 The inclusion of the export intensity is consistent with Gonzaga, Filho, and Terra’s (2006) study which shows that the pass-through from tariffs to prices in each sector depends on the sector’s import penetration. For instance, the pass-through, measured by the share of trade relative to production, is obviously zero in sectors with no trade whereas the pass-through coefficient is one in sectors where all goods are traded. Investigating the impacts of trade liberalization implemented in Brazil from 1988 to 1995, Gonzaga et al. (2006) show that only after an adjustment of tariff changes for import penetration was the trade liberalization pattern with respect to skill intensity consistent with that of relative price changes. 41 As a robustness test, I construct the alternative tariff measure considering only those employed in the industrial sector so that the variation in the tariff measure is independent of the size of the industrial sector within a province (Topalova, 2010). The results remain essentially unchanged. 26 Figure 8.1 and Figure 8.2 plot the relationship between the U.S. tariff cut measure and the change in Export Index (for Model A and B) and the change in U.S. Export Index (for Model C) respectively. The figures reveal strong positive relationship between the U.S. tariff cut measure and the changes in the Export Indices. The correlation appears to be stronger when the change in U.S. Export Index is used, which in turn reflects the estimated change in employment contained in the exports to the United States. Finally, I repeat the same regressions in Table 1 using two-stage least square (2SLS) models. Table 2.1 demonstrates the results of the first stage regressions which relate the changes in Export Index (∆Exportp) and the U.S. Export Index (∆ExportUSp) on the measure of the U.S. tariff cut (TariffCutp) (excluded instrument (Bound, Jaeger, & Baker, 1995)) along with all the other exogenous variables included in the second stage. The coefficients for the tariff cut measure turn out to be highly positively significant confirming a strong positive relationship between the latter variable and the change in Export Indices. However, in Models A and B (except for the subset of unskilled workers in Model B), the first stage F statistics for excluded instrument are less than ten revealing a concern that 2SLS results may become unreliable (Staiger & Stock, 1997). In Model C, however, F statistics for excluded instrument turn out to be large, implying that the tariff cut measure is a strong instrument. Table 2.2 shows the results of the second-stage 2SLS regressions. The coefficients for the change in Export Index for unskilled workers estimated by the 2SLS remain positively significant in all the models confirming that the unskilled individuals residing in a province which experienced a larger increase in exports tended to benefit from a larger wage growth. The coefficients for the Export Indices for the subset of skilled workers are either insignificant or negatively significant. The last row in Table 2.2 reports the Wu-Hausman test statistics. Whereas 27 the test results reject the hypothesis that OLS estimates are consistent for some specifications, the results for the subset of unskilled workers in Model B and Model C imply that 2SLS is not required. Interpreting the results obtained by OLS for these models, a one unit increase in the change in the Export Index and in the U.S. Export Index is associated with about 2.8 percentage point and 4.1 percentage point higher wage growth for unskilled workers, respectively. Another way to deal with endogeneity is to relate the U.S. tariff cut measure to wage growth directly, which is a common approach in the previous literature (Castilho, Menéndez, & Sztulman, 2011; McCaig, 2011; Topalova, 2010). I repeat the same regressions replacing Export Indices by the U.S. tariff cut measure. The results reported in Appendix Table A.8 reveal a similar pattern. The coefficients for the U.S. tariff cut measure for a subset of unskilled workers turn out to be positively significant whereas those for a subset of skilled workers become either insignificant or negatively significant. I also re-run the same regressions using hourly wage growth instead of monthly wage growth as the dependent variable. The coefficients for the change in the Export Indices for the subset of unskilled workers remain positively significant for some models but the results are found to be sensitive to the inclusion of control variables (the regressions are not reported). (d) Impacts on the skill premium Many economists question the validity of the H-O-S theorem on empirical grounds. In particular, a number of Latin American countries drastically lowered their tariff and non-tariff barriers (NTBs) to trade pursuant to the GATT in the 1980s and 1990s. However, contrary to the prediction of the H-O-S theory, most Latin American countries experienced an increase rather than a decrease in the skill premium after episodes of trade liberalization (Goldberg & Pavcnik, 28 2007). This “puzzle� spurred the development of rich new theories which predict that trade may lead to a rise in wage inequality. 42 Some authors explore the “Second Puzzle� (Robertson, 2007) after Mexico’s entry into the NAFTA in 1994, i.e., why skill premium and other measures of inequality in Mexico began to fall or at least remained stable in the post-NAFTA period (Chiquiar, 2008; Goldberg & Pavcnik, 2007, Table 1, p.48; Robertson, 2004, 2007). A potential explanation is that Mexico’s further economic integration with the United States and Canada after NAFTA might have strengthened the H-O-S type effects. For instance, investigating the direct link between changes in relative product prices and changes in relative wages in the context of GATT and NAFTA, Robertson (2004) finds that the changes in the structure of wages in Mexico seem to be consistent with the prediction of the Stolper-Samuelson theorem. 43 Comparing the export patterns before and after NAFTA, Acosta and Montes-Rojas (2008) show that Mexico significantly changed its manufacturing trade patterns after NAFTA, becoming a major exporter of goods with low-skill content such as clothing and automotive supplies (Figure 2, p.765). In contrast to many Latin American countries, East Asian developing countries are generally viewed to have achieved both openness and relatively equitable development (e.g., World Bank, 1993). Some studies suggest that the experiences of East Asian Newly Industrializing Countries (NICs) are broadly consistent with the hypothesis that export-oriented industrialization reduces income inequality (Wood, 1999). For instance, based on time-series evidence, Wood (1999) 42 These theories include outsourcing, increase in capital flows and complementarity of capital and skilled labor, trade-induced Skill-Biased Technological Change (SBTC) and firm heterogeneity models (Goldberg & Pavcnik, 2007). For instance, firm heterogeneity models pioneered by Melitz (2003) tend to predict that an increase in export opportunities may increase demand for skilled labor. This is because the production of exporters may be more skill- intensive than production for non-exporters (Goldberg & Pavcnik, 2007); because trade may induce an upgrading of the “product quality� in exporting plants within industries (Verhoogen, 2008); and because reduction in trade costs may induce some firms to switch to more advanced technologies (Yeaple, 2005). 43 Whereas the relative prices of skill-intensive goods and the relative wages of skilled workers rose following Mexico’s entrance to the GATT (as Mexico protected less-skill-intensive industries prior to GATT), they diminished after NAFTA as Mexico further liberalized its trade with relatively skill-abundant Canada and the United States. 29 observes that, following a switch to export-oriented industrialization in Korea and Taiwan, China, in the 1960s and in Singapore in the 1970s, the wage gap between skilled and unskilled workers of these economies declined. However, there exist few studies based on detailed micro data for East Asian NICs, perhaps due to the lack of data (Goldberg & Pavcnik, 2007) and because of difficulty in formulating empirical designs given potential endogeneity of their trade policies. 44 For East Asian NICs, whereas the positive association between openness and equality is consistent with the H-O-S theorem, a question remains whether the equalizing development in these economies was a consequence of trade liberalization or an outcome of national policies such as broad based education. A number of studies report rising wage inequality in China since the introduction of market- oriented reforms in 1978 (e.g., Benjamin, Brandt, Giles, & Wang, 2005; Knight & Song, 2003). Many of them report that the Chinese wage structure was initially compressed under China’s centrally-planned system, but that the return on education rose sharply under the reforms. As China progressively liberalized its trade regimes during the same period, one challenge of empirical studies for China (as for other transition economies) is to separate the trade-related impacts on income from the effects of other market-oriented reforms. Exploiting regional variation, Wei and Wu (2001) find that Chinese cities (urban areas and adjacent rural counties) that experienced a greater increase in the export-to-GDP ratio tend to demonstrate a greater decline in urban-rural income inequality. As a potential explanation, Wei and Wu suggest that export-oriented activities by the Township and Village Enterprises (the TVEs), which in turn were permitted to industrialize under the policy of “Li Tu Bu Li Xiang� (“leaving-the-farm- work-but-not-the-farmland�), might have contributed to a rise in income in the rural areas. 44 With a combination of export subsidy and import protection, governments in East Asian countries intervened in trade and industrial policy in order to bring about “dynamic� comparative advantage. See Amsden (1989) for Korea and Wade (1990) for Taiwan, China. 30 In this sub-section, I investigate the impact of exposure to exports on skill premium introducing an interaction term between the skill status and the change in the Export Indices. Specifically, ∆Ln Wip = α0 + α1 Skillip + α2 Skillip·∆Exportp + α3 ∆Exportp + α4 ∆Xip + α5 Yip+ vip (5) where Skillip denotes a dummy variable to indicate individual i’s skill status evaluated by his/her occupation in 2002 (= 1 if he/she is a skilled worker; = 0 otherwise), Skillip·∆Exportp is an interaction term between the skill status and the change in the Export Index, and the other variables are those defined in Equation (2). Table 3 reports the results of the OLS regressions described above. In parallel to my regressions in Table 1 and in Table 2.2, regression (1) and regression (2) use the same control variables as those used in Model A and Model B respectively, and regression (3) re-estimates the same regressions using the change in the U.S. Export Index (∆ExportUSp) and the interaction term between the change in the U.S. Export Index and the skill status (Skillip·∆ExportUSp). In all three models, the coefficients for the interaction terms between the skill status and the Export Indices turn out to be negatively significant at the one percent level, suggesting that individuals who lived in a province more exposed to exports experienced a decrease (or a slower increase) in the skill premium. The coefficient estimates of the interaction terms imply that a one unit increase in the change in the Export Index and in the U.S. Export Index is associated with 2.3 percentage point and 4.5 percentage point lower increases in the growth rate of the skilled workers relative to unskilled workers respectively. 45 To deal with a potential endogeneity of the Export Indices, I repeat the regressions using the U.S. tariff cut measure instead of the change in the Export Indices. The results reported in Appendix Table A.9 suggest a robustness of the result 45 Evaluated at the sample means of the changes in the Export and US Export Indices, these figures correspond to 1.4 percentage point and 1.5 percentage point lower relative wage growth of skilled workers. 31 to the concern of endogeneity as the coefficients for the interaction term between skill status and the U.S. tariff cut measure turn out to be negatively significant at the one percent level. Finally, I repeat the same regressions measuring the wage growth with hourly wages rather than monthly wages and find no statistically significant relationship between the change in the exposure to exports and the change in skill premium (the regressions are not reported). The results in this paper confirm the existence of a Stolper-Samuelson type effect, i.e., those provinces more exposed to the increase in export opportunities experienced a decline of (or a smaller increase in) the skill premium relative to other provinces for the period 2002-2004. During the same period, the skill premium increased at the national level for my sample of panel individuals. 46 Thus, the S-S type effect appears to have mitigated, but did not outweigh the impacts of other factors which contributed to the rise in skill premium. These results are consistent with Wei and Wu’s (2001) study which finds a greater decline (or a smaller increase) in income inequality in the areas more exposed to exports, when China was experiencing a rise in skill premium at the national level. 4. CONCLUSION This paper explored how the expansion of labor-intensive manufacturing exports resulting from the U.S.-Vietnam BTA in 2001 translated into wages of skilled and unskilled workers in Vietnam through the channel of labor demand. The results demonstrate the existence of a “Stolper-Samuelson type� effect: those provinces which are more exposed to the increase in export opportunities experienced a larger wage growth for unskilled workers and a decline of (or smaller rate of increase in) the relative wage of skilled and unskilled workers relative to the other provinces. 46 The factors which may have contributed to the rise in skill premium include: the impacts of domestic reforms, SBTC, and other impacts of trade which might have occurred concurrently during the period studied. 32 Vietnam’s experience in the aftermath of the BTA highlights the importance of economic integration of a developing country with developed countries, in particular, in terms of creating job opportunities for its abundant unskilled workers and its positive impacts on their wages. The rise in wages for unskilled workers in the process of export-oriented industrialization is consistent with the East Asian experience, in particular, the historical shift of production locations to lower wage countries as wages increase; and the recent rise of wages for production workers in China. 47 The findings of this study also support the view that improving access to developed country markets for developing countries, in particular for unskilled-labor intensive goods, is essential for the latter countries to reap the benefits predicted by traditional trade theory. One limitation of this paper is that, whereas it focuses on identifying the S-S type effect exploiting regional variation, it does not analyze other factors which affected the wage structure. In fact, for the period 2002-2004, the skill premium increased at the aggregate level for the sample of panel individuals. Thus, the results imply that the S-S type effect mitigated, but did not outweigh the impacts of other factors which contributed to the rise in skill premium. It is also noted that the theory illustrated in this paper is not the pure version of the H-O-S whose simplified assumptions tend to be inconsistent with the complexity of the real world. For instance, the assumption of full employment would not strictly apply to Vietnam’s economy where a large majority of workers tend to be underemployed (Fukase, Table 2, 2013c) in agricultural and other informal sectors. Nonetheless, this paper suggests that the basic insights from the H-O-S theory remain valid as an important component of globalization and that, given 47 Several articles in The New York Times (e.g., February, 20, 2012) and The Economist (July 31, 2010) report the rising wages of production workers in coastal industrial areas in China. As labor costs have risen in China, some production of labor-intensive goods may be moving to lower-wage countries like Bangladesh, Vietnam and Cambodia (The New York Times, July 16, 2010; January 31, 2011). 33 their potential for economic development and poverty reduction, the Heckscher-Ohlin- Samuelson effects need to be addressed properly by policy makers in both developing and developed countries. REFERENCES Acosta, P., & Montes-Rojas, G. V. (2008). Trade Reform and Inequality: The Case of Mexico and Argentina in the 1990s. The World Economy, 31(6), 763-780. Amsden, A. H. (1989). South Korea and Late Industrialization. New York: Oxford University Press. Anderson, J. B., & Dimon, D. (1999). Formal Sector Job Growth and Women’s Labor Sector Participation: the Case of Mexico. The Quarterly Review of Economics and Finance, 39, 169- 191. Athukorala, P. (2006). Trade Policy Reforms and the Structure of Protection in Vietnam. The World Economy, 29(2), 161-87. Athukorala, P. (2009). Transition to a Market Economy and Export Performance in Vietnam. ASEAN Economic Bulletin, 26(1), 96-114. Attanasio, O., Goldberg, P. K., & Pavcnik, N. (2004). Trade Reforms and Wage Inequality in Colombia. Journal of Development Economics, 74, 331-366. Baum, C. F., Schaffer, M. E., & Stillman, S. (2007). Enhanced Routines for Instrumental Variables/GMM Estimation and Testing. The Stata Journal, 7(4), 465-506. Belser, P. (2000). Vietnam: On the Road to Labor-Intensive Growth? Policy Research Working Paper, 2389, the World Bank. Benjamin, D., Brandt, L., Giles, J., & Wang, S. (2008). Income Inequality during China’s Economic Transition. In L. Brandt, & T. Rawski (Eds.), China’s Great Economic Transformation, Chapter 18. Cambridge: Cambridge University Press. Bernard, A. B., Jensen, J. B., Redding, S. J., & Schott, P. K. (2007). Firms in International Trade. Journal of Economic Perspective, 21(3), 105-130. Bound, J., Jaeger, D. A., & Baker, R. M. (1995). Problems with Instrumental Variables Estimation When the Correlation between the Instruments and the Endogenous Explanatory Variable is Weak. Journal of the American Statistical Association, 90(430), 443-450. Brambilla, I., Porto, G., & Tarozzi, A. (2009). Adjusting to Trade-Policy Changes in Exports Markets: Evidence from U.S. Antidumping Duties on Vietnamese Catfish. Policy Research Working Paper, 4990, the World Bank. 34 Castilho, M., Menéndez, M., & Sztulman, A. (2011). Trade Liberalization, Inequality, and Poverty in Brazilian States. World Development, 40(4), 821-835. Chiquiar, D. (2008). Globalization, Regional Wage Differentials and the Stolper-Samuelson Theorem: Evidence from Mexico. Journal of International Economics, 74, 70-93. Coello, B. (2009). Exports Liberalization and Specialization in Cash Crop: Gains for Vietnamese Households? Economie Internationale, 118, 73-99. Dimaranan, B., Duc, L. T., & Martin, W. (2005). Potential Economic Impacts of Merchandise Trade Liberalization under Viet Nam’s Accession to the WTO. Purdue University. Dodzin, S., & Vamvakidis, A. (2004). Trade and Industrialization in Developing Economies. Journal of Development Economics, 75, 319-328. 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. Fu, X., & Balasubramanyam, V. N. (2005). Exports, Foreign Direct Investment and Employment: the Case of China. The World Economy, 28(4), 607-625. Fukase, E. (2013a). Export Liberalization, Job Creation and the Skill Premium: Evidence from the U.S.-Vietnam Bilateral Trade Agreement (BTA), World Development, 41(1), 317-337. Fukase, E. (2013b). Foreign Job Opportunities and Internal Migration in Vietnam. Policy Research Working Paper, forthcoming, the World Bank. Fukase, E. (2013c). Foreign Wage Premium, Gender and Education: Insights from Vietnam Household Surveys, forthcoming in The World Economy. Fukase, E., & Martin, W. (2000). The Effects of the United States Granting MFN Status to Vietnam. Weltwirtschaftliches Archiv, 136(3), 539-559. Galiani, S., & Sanguinetti, P. (2003). The Impact of Trade Liberalization on Wage Inequality: Evidence from Argentina. Journal of Development Economics, 72, 497-513. Goldberg, P. K., & Pavcnik, N. (2007). Distributional Effects of Globalization in Developing Countries. Journal of Economic Literature, 45(1), 39-82. Gonzaga, G., Filho, N. M., & Terra, C. (2006). Trade Liberalization and the Evolution of Skill Earnings Differentials in Brazil. Journal of International Economics, 68, 345-367. GSO. (2000). Analyzing the Results of the Industrial Survey of Vietnam 1999. Hanoi: Statistical Publishing House. 35 GSO. (2011). Migration and Urbanization in Vietnam: Patterns, Trends and Differentials. Hanson, G. (2005). Globalization, Labor Income, and Poverty in Mexico. NBER Working Paper, 11027. Harrison, A., McLaren, J., & McMillan, M. S. (2010). Recent Findings on Trade and Inequality. NBER Working Paper, 16425. Jenkins, R. (2004). Vietnam in the Global Economy: Trade, Employment and Poverty. Journal of International Development, 16, 13-28. Kien, T. N., & Heo, Y. (2009). Impacts of Trade Liberalization on Employment in Vietnam: A System Generalized Method of Moments Estimation. The Developing Economies, 47(1), 81-103. Kijima, Y. (2006). Why Did Wage Inequality Increase? Evidence from Urban India 1983-99. Journal of Development Economics, 81, 97-117. Knight, J., & Song, L. (2003). Increasing Urban Wage Inequality in China. Economics of Transition, 11(4), 597–619. Lewis, W. A. (1954). Economic Development with Unlimited Supplies of Labour. Manchester School of Economic and Social Studies, 22(2), 139-191. McCaig, B. (2009). The Reliability of Matches in the 2002-2004 Vietnam Household Living Standards Survey Panel. Discussion Paper, 622. The Australian National University, Centre for Economic Policy Research. McCaig, B. (2011). Exporting Out of Poverty: Provincial Poverty in Vietnam and U.S. Market Access. Journal of International Economics, 85(1), 102-113. Melitz, M. J. (2003). The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity. Econometrica, 71, 1696-1725. Parker, S., Riedel, J., & Quang, P. V. (2007). Assessment of the Five Year Impact of the U.S.- Vietnam Bilateral Trade Agreement on Vietnam’s Trade, Investment, and Economic Structure. A report by STAR-Vietnam project. Ramstetter, E. D., & Ngoc, P. M. (2007). Changes in Ownership and Producer Concentration after the Implementation of Vietnam’s Enterprise Law. Working Paper, 2007-06, Kitakyushu: International Centre for the Study of East Asian Development. Robertson, R. (2004). Relative Prices and Wage Inequality: Evidence from Mexico. Journal of International Economics, 64, 387-409. Robertson, R. (2007). Trade and Wages: Two Puzzles from Mexico. The World Economy, 30(9), 1378-1398. 36 Staiger, D., & Stock, J. H. (1997). Instrumental Variables Regression with Weak Instruments. Econometrica, 65(3), 557-586. Stolper, W., & Samuelson, P. (1941). Protection and Real Wages. Review of Economic Studies, 9, 58-73. Topalova, P. (2010). Factor Immobility and Regional Impacts of Trade Liberalization: Evidence on Poverty from India. American Economic Journal: Applied Economics, 2, 1–41. Van Arkadie, B., & Mallon, R. (2003). Viet Nam: a Transition Tiger? Canberra: Asia Pacific Press. Verhoogen, E. A. (2008). Trade, Quality Upgrading and Wage Inequality in the Mexican Manufacturing Sector. Quarterly Journal of Economics, 123(2), 489-530. Wacziarg, R., & Wallack, J. S. (2004). Trade Liberalization and Intersectoral Labor Movements. Journal of International Economics, 64, 411-439. Wade, R. (1990). Governing the Market: Economic Theory and the Role of Government in East Asian Industrialization. Princeton: Princeton University Press. Wei, S-J., & Wu, Y. (2001). Globalization and Inequality: Evidence from Within China. NBER Working Paper, 8611. Winters, L. A. (2002). Trade Liberalization and Poverty: What are the Links? The World Economy, 25(9), 1339-1367. Wood, A. (1999). Openness and Wage Inequality in Developing Countries: The Latin American Challenge to East Asian Conventional Wisdom. In Baldwin, R., Cohen, D., & Sapir, A. (Eds.). Market Integration, Regionalism and the Global Economy. Cambridge: Cambridge University Press, 153-188. World Bank. (1993). The East Asian Miracle: Economic Growth and Public Policy. Oxford and New York: Oxford University Press. Yeaple, S. R. (2005). A Simple Model of Firm Heterogeneity, International Trade and Wages. Journal of International Economics, 65, 1-20. 37 14000 12000 10000 manufactures (SITC 5-9) 8000 primary 6000 (SITC 0-4) 4000 Total 2000 0 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Figure 1. Evolution of Vietnam’s Exports to the United States ($ million). Source: the U.N. Comtrade System. 35.0 30.0 25.0 manufactures (SITC 5-9) 20.0 primary (SITC 0-4) 15.0 total 10.0 5.0 0.0 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Figure 2. Share of the United States in Vietnam’s Total Exports (%) Source: the U.N. Comtrade System. 38 = Source: Author’s calculation based on the U.N. Comtrade System. 39 Source: Author’s calculation based on the U.N. Comtrade System and the Enterprise Survey data. 40 1000 Apparel, textiles 900 Footw ear, leather 800 products Furniture, w ood, 700 manufact. n.e.s. Food products 600 Electronics, machinery, 500 transport equipment Metal, non-metallic 400 mineral products Basic manufacturing 300 Chemical, rubber, 200 plastic products Mining 100 0 2000 2001 2002 2003 2004 2005 2006 2007 Figure 5. Evolution of Vietnam’s Industrial Employment by the VSIC Sub-categories (1,000s). Source: the Enterprise Survey data (GSO). 1200 1000 South East 800 HCMC Red River Delta Mekong River Delta North East 600 Hanoi South Central Coast North Central Coast Central Highlands 400 North West 200 0 2000 2001 2002 2003 2004 2005 2006 2007 Figure 6. Vietnam’s Industrial Employment by Regions (1,000s). Source: the Enterprise Survey data (GSO). 41 55 54 53 52 51 50 49 2000 2001 2002 2003 2004 2005 2006 2007 Figure 7.1. Proportion of Workers Employed in Labor-intensive Industries in Total Industrial Employment (%). Note: The industries whose employment coefficients are 6.5 or above in 2004 are classified as labor- intensive industries (see Appendix Table A.4 for the employment coefficients). Source: the Enterprise Survey data (GSO). 55 54 53 52 51 50 49 2000 2001 2002 2003 2004 2005 2006 2007 Figure 7.2. Proportion of Workers Employed in Unskilled-labor-intensive Industries in Total Industrial Employment (%). Note: The industries whose proportion of production workers is 85.7 or above in 1998 in the Industrial Survey (GSO, 2000) are referred to unskilled-labor-intensive industries (see Appendix Table A.4 for the proportion of production workers). Source: GSO, 2000. 42 10 8 Changes in Export Index 4 6 2 0 0 1 2 3 4 5 U.S. Tariff Cut Measure Figure 8.1. Change in Export Index (∆Export) and U.S. Tariff Cut Measure 6 Changes in Export U.S. Index 2 0 4 0 1 2 3 4 5 U.S. Tariff Cut Measure Figure 8.2. Change in U.S. Export Index (∆ExportUS) and U.S .Tariff Cut Measure 43 Table 1. OLS Regression Results: Explaining Wage Growth Model A Model B Model C Total Unskilled Skilled Total Unskilled Skilled Total Unskilled Skilled ∆Export Index .012 .030** -.0007 .017** .028*** .0011 (.0098) (.013) (.0088) (.0079) (.011) (.0079) ∆Export US Index .019 .041** -.010 (.013) (.017) (.012) ∆Skilled/unskilled status -.054 .11 -.026 -.086 .055 -.071 -.085 .055 -.069 (.062) (.094) (.063) (.065) (.092) (.078) (.065) (.092) (.078) ∆Marital status -.12* -.15* -.014 -.10* -.14** .0025 -.10 -.14* .0016 (.072) (.079) (.093) (.063) (.069) (.091) (.063) (.069) (.092) Initial wage in log -.25*** -.36*** -.26*** -.40*** -.49*** -.33*** -.40*** -.49*** -.33*** (.020) (.029) (.034) (.029) (.033) (.043) (.030) (.033) (.043) Female -.097*** -.24*** .022 -.097*** -.24*** .021 (.024) (.034) (.036) (.024) (.035) (.036) Minority .0024 -.12 .099 -.00001 -.13 .099 (.087) (.12) (.072) (.088) (.12) (.072) Education .041*** .023*** .028*** .041*** .023*** .027*** (.0035) (.0045) (.0068) (.0035) (.0046) (.0067) Potential Experience .0015 -.00010 .00062 .0016 -.00006 .00058 (.0015) (.0016) (.0018) (.0015) (.0016) (.0018) Urban .044 .060 .0079 .043 .060 .0068 (.031) (.042) (.042) (.031) (.042) (.042) -.41* -.27 -.72*** -.35 -.19 -.72*** Agriculture share 2002a (.23) (.30) (.26) (.23) (.30) (.25) -.28 .072 -.71 -.099 .28 -.61 Industry share 2002a (.44) (.63) (.51) (.47) (.63) (.45) .14 1.51 -1.08 -.0051 1.33 -1.15 Construction share 2002a (.73) (.97) (1.07) (.71) (.97) (1.07) -1.27 -.20 -2.81** -1.30 -.24 -2.85** Government share 2002a (1.21) (1.39) (1.22) (1.23) (1.42) (1.23) Constant 3.50*** 4.91*** 3.85*** 5.45*** 6.53*** 5.12*** 5.39*** 6.44*** 5.10*** (.27) (.37) (.46) (.44) (.48) (.66) (.44) (.48) (.65) Observations b 1718 1051 647 1718 1051 647 1718 1051 647 R2 .115 .182 .146 .210 .274 .184 .209 .272 .184 Notes: The dependent variable is the log difference of monthly wage (∆Ln W in Equation (2)). The classification between skilled and unskilled workers is based on the status in 2002. *, **, *** indicate that the coefficients are significant at the 10, 5, and 1 percent level respectively. The standard errors in parentheses are based on heteroskedasticity-consistent estimates of the variance-covariance matrix and corrected for within province correlation (clustering). a The share of workers employed in the service sector is the omitted category. b The total sample includes those who worked for “Army force�. As the skill category is not assigned for the latter workers, the number of skilled and unskilled workers do not add up to total number. 44 Table 2.1. First Stage Regression Results Model A Model B Model C Dependent Variable: ∆Export Index Dependent Variable: ∆Export Index Dependent Variable: ∆Export US Index Total Unskilled Skilled Total Unskilled Skilled Total Unskilled Skilled U.S. tariff cut measure 1.21*** 1.39*** .94*** 1.39*** 1.60*** 1.02*** 1.03*** 1.07*** .93*** (.026) (.033) (.042) (.034) (.043) (.055) (.013) (.017) (.020) Observations 1718 1051 647 1718 1051 647 1718 1051 647 R2 .567 .648 .448 .708 .760 .636 .885 .893 .873 F test of excluded instrument 5.29 6.99 4.09 6.40 10.51 3.14 28.76 33.28 24.91 [p-value] [.025] [.011] [.048] [.015] [.0022] [.083] [.000] [.000] [.000] Notes: At the first stage, I regress the change in Export Indices on the measure of the U.S. tariff cut (excluded instrument) and on all the other exogenous variables included in the second stage. The regressions are estimated using STATA’s ivreg2 command with the robust and cluster options (Baum, Schaffer, & Stillman, 2007). The classification between skilled and unskilled workers is based on the status in 2002. *** indicate that the coefficients are significant at the one percent level. The standard errors in parentheses are based on heteroskedasticity-consistent estimates of the variance-covariance matrix and corrected for within province correlation (clustering). 45 Table 2.2. 2SLS Regression Results: Explaining Wage Growth Model A Model B Model C Total Unskilled Skilled Total Unskilled Skilled Total Unskilled Skilled ∆Export Index .024 .047* .0072 .0076 .028* -.027 (.021) (.027) (.017) (.011) (.016) (.023) ∆Export US Index .010 .042* -.029* (.017) (.023) (.017) ∆Skilled/unskilled status -.056 .11 .028 -.083 .055 -.060 -.083 .055 -.064 (.062) (.093) (.062) (.064) (.090) (.077) (.064) (.090) (.077) ∆Marital status -.13* -.16** -.017 -.10 -.14** .0069 -.099 -.14** -.00035 (.070) (.078) (.092) (.061) (.068) (.091) (.062) (.069) (.091) Initial wage in log -.25*** -.38*** -.27*** -.40*** -.49*** -.33*** -.40*** -.49*** -.33*** (.023) (.031) (.036) (.030) (.034) (.041) (.030) (.033) (.041) Female -.096*** -.24*** .018 -.096*** -.24*** .020 (.024) (.034) (.036) (.024) (.034) (.036) Minority .0024 -.12 .093 .0012 -.13 .10 (.087) (.12) (.073) (.087) (.12) (.071) Education .041*** .023*** .025*** .041*** .023*** .026*** (.0035) (.0044) (.0066) (.0035) (.0045) (.0066) Potential Experience .0015 -.00010 .00057 .0016 -.000063 .00052 (.0014) (.0016) (.0018) (.0014) (.0016) (.0017) Urban .044 .060 .0029 .043 .060 .0052 (.030) (.041) (.042) (.030) (.041) (.042) Agriculture share 2002a -.38 -.27 -.62** -.35 -.19 -.72*** (.24) (.30) (.25) (.23) (.30) (.25) Industry share 2002a -.074 .084 -.16 -.0059 .27 -.48 (.54) (.66) (.57) (.48) (.65) (.40) Construction share 2002a -.039 1.50 -1.46 -.090 1.34 -1.26 (.74) (1.00) (1.06) (.71) (.99) (1.07) Government share 2002a -1.34 -.20 -2.89** -1.34 -.24 -2.90** (1.22) (1.36) (1.21) (1.23) (1.39) (1.22) Constant 3.57*** 5.04*** 3.89*** 5.40*** 6.52*** 4.99*** 5.37*** 6.44*** 5.08*** (.30) (.39) (.47) (.46) (.50) (.62) (.44) (.48) (.64) Observationsb 1718 1051 647 1718 1051 647 1718 1051 647 R2 .114 .178 .145 .210 .274 .179 .209 .272 .184 Wu-Hausman F test 2.52 7.26 .59 .76 0.00 4.43 1.36 .000 3.59 [p-value] [.118] [.0092] [.447] [.386] [.966] [.040] [.248] [.966] [.063] Notes: The dependent variable is the log difference of monthly wage (∆Ln W in Equation (2)). The regressions are estimated using STATA’s ivreg2 command with the robust and cluster options (Baum et al., 2007). The classification between skilled and unskilled workers is based on the status in 2002. *, **, *** indicate that the coefficients are significant at the 10, 5, and 1 percent level respectively. The standard errors in parentheses are based on heteroskedasticity-consistent estimates of the variance-covariance matrix and corrected for within province correlation (clustering). a The share of workers employed in the service sector is the omitted category. b The total sample includes those who worked for “Army force�. As the skill category is not assigned for the latter workers, the number of skilled and unskilled workers do not add up to total number. 46 Table 3. Impact of Change in Exports on Skill Premium (1) (2) (3) Skill*∆Export Index -.023*** -.023*** (.0067) (.0080) ∆Export Index .027** .022** (.012) (.0089) Skill*∆Export US Index -.045*** (.014) ∆Export US Index .032** (.014) Skill status .29*** .21*** .22*** (.033) (.033) (.034) ∆Skilled/unskilled status .080 .023 .023 (.065) (.068) (.068) ∆Marital status -.12 -.12* -.12* (.070) (.064) (.064) Initial wage in log -.33*** -.41*** -.41*** (.022) (.029) (.029) Female -.12*** -.12*** (.026) (.026) Minority -.016 -.018 (.087) (.088) Education .026*** .026*** (.0037) (.0036) Potential Experience -.00023 -.00024 (.0013) (.0013) Urban .040 .039 (.030) (.030) Agriculture share 2002a -.36 -.33 (.22) (.22) Industry share 2002a -.053 .082 (.43) (.43) Construction share 2002a .53 .41 (.74) (.72) Government share 2002a -1.32 -1.35 (1.16) (1.17) Constant 4.44*** 5.63*** 5.58*** (.29) (.44) (.43) Observations 1698 1698 1698 R2 .173 .228 .228 Note: The dependent variable is the log difference of monthly wage (∆Ln W in Equation (5)). The standard errors in parentheses are based on heteroskedasticity-consistent estimates of the variance-covariance matrix and corrected for within province correlation (clustering). *, **, *** indicate that the coefficients are significant at the 10, 5, and 1 percent level respectively. The classification between skilled and unskilled workers is based on the status in 2002. a The share of workers employed in the service sector is the omitted category. 47 Appendix Table A.1. Vietnam’s Exports to the World by the VSIC Sub-categories ($ million) 2000 2001 2002 2003 2004 2005 2006 2007 2008 Industry Apparel, textiles (17, 18) 2104 2186 3045 3933 4890 5392 6547 8608 10163 Footwear, leather products (19) 1647 1829 2115 2562 3030 3437 4049 4623 5689 Furniture, wood, manufact. n.e.s. (20, 36, 37) 580 670 848 1115 1596 2117 2713 3574 4232 Food products (15, 16) 2463 2900 3066 2874 3620 4559 5352 6255 8634 Electronics, machinery, transport equipment (29-35) 1299 1366 1339 1832 2687 3274 4350 5892 7835 Metal, non-metallic mineral products (26, 28) 201 267 338 427 556 725 998 1279 1533 Basic manufacturing (21, 22, 23, 27, 40, 41) 360 351 300 349 537 739 1077 1505 3826 Chemical, rubber, plastic products (24, 25) 256 379 439 550 771 979 1414 1797 2547 Mining (10, 11, 13, 14) 3628 3309 3485 4086 6179 8172 9403 9714 12046 Agriculture, forestry, fishery (1-5) 1479 1407 1556 2255 2499 2915 3772 4972 5753 Others 464 365 174 166 119 139 151 343 429 Total 14483 15029 16706 20149 26485 32447 39826 48561 62685 Appendix Table A.2. Vietnam’s Exports to the U.S. by the VSIC Sub-categories ($ million) 2000 2001 2002 2003 2004 2005 2006 2007 2008 Industry Apparel, textiles (17, 18) 50 48 1040 2020 2550 2691 3148 4576 5241 Footwear, leather products (19) 92 117 236 352 502 702 893 1003 1208 Furniture, wood, manufact. n.e.s. (20, 36, 37) 11 22 81 173 381 672 927 1276 1448 Food products (15, 16) 304 488 678 617 560 588 682 771 801 Electronics, machinery, transport equipment (29-35) 5 8 33 88 117 240 484 740 983 Metal, non-metallic mineral products (26, 28) 9 11 30 36 62 82 117 146 186 Basic manufacturing (21, 22, 23, 27, 40, 41) 1 5 12 9 8 16 34 71 156 Chemical, rubber, plastic products (24, 25) 1 3 10 17 36 63 88 178 243 Mining (10, 11, 13, 14) 92 234 148 201 407 468 1030 798 1017 Agriculture, forestry, fishery (1-5) 151 129 182 422 403 404 446 542 617 Others 16 0 2 3 1 1 0 11 2 Total 733 1066 2453 3940 5027 5927 7850 10111 11903 Appendix Table A.3. Vietnam’s Imports from the World by the VSIC Sub-categories ($ million) 2000 2001 2002 2003 2004 2005 2006 2007 2008 Industry Apparel, textiles (17, 18) 1676 1620 2215 2582 3111 3509 3943 4998 5628 Footwear, leather products (19) 500 562 656 784 950 978 947 1107 1165 Furniture, wood, manufact. n.e.s. (20, 36, 37) 161 211 376 478 634 788 1027 1298 1382 Food products (15, 16) 629 795 922 1224 1487 1765 2065 3027 4347 Electronics, machinery, transport equipment (29-35) 4844 5025 5913 8102 9030 9621 11406 18545 23431 Metal, non-metallic mineral products (26, 28) 314 448 595 851 881 1065 1310 1921 2321 Basic manufacturing (21, 22, 23, 27, 40, 41) 3571 3739 4476 5848 8793 10966 14485 19216 26571 Chemical, rubber, plastic products (24, 25) 2879 3041 3610 4426 5717 6469 7720 10220 12597 Mining (10, 11, 13, 14) 142 130 183 166 280 300 330 415 541 Agriculture, forestry, fishery (1-5) 412 456 575 688 941 1152 1470 1785 2333 Others 508 192 224 109 145 147 188 233 398 Total 15637 16218 19746 25256 31969 36761 44891 62765 80714 Source: the U.N. Comtrade System. Note: Between parentheses are the VSIC codes at the two digit level. 48 Appendix Table A.4. Exports, Employment Coefficients and Proportion of Production Workers By VSIC 2-digit Level Share in Share of Total Employment production Exports Exports coefficient workers a VSIC categories Value 2004 2004 2004b 1998 ($ million) (%) (person/bn dong) (%) Crude petroleum (11) 5671 21.4 .1 88.9 Apparel (18) 3633 13.7 6.7 88.5 Food and beverages (15) 3492 13.2 2.8 81.7 Footwear and leather products (19) 3030 11.4 7.9 92.3 Furniture and manufacturing n.e.s. (36) 1276 4.8 9.1 88.7 Textiles (17) 1257 4.7 6.8 89.7 Electronics (31) 771 2.9 2.3 79.3 Office machinery (30) 655 2.5 .6 79.8 Radio, TV and communication machinery (32) 537 2.0 1.8 74.7 Rubber and plastics products (25) 454 1.7 3.5 81.0 Mining of hard coal lignite and peat (10) 354 1.3 6.5 86.9 Wood products (20) 320 1.2 10.4 85.7 Chemical products (24) 317 1.2 1.9 73.2 Non-metallic mineral products (26) 291 1.1 5.0 84.8 Transport equip. other than motor vehicle (35) 289 1.1 2.1 75.7 Metal products (28) 265 1.0 4.1 78.6 Machinery and equipment (29) 237 .9 4.1 74.3 Basic metals (27) 212 .8 1.4 85.6 Coke and refined petroleum products (23) 209 .8 .6 30.8 Tobacco products (16) 128 .5 1.1 85.3 Mining of metal ores (13) 120 .5 9.4 81.4 Motor vehicles (34) 102 .4 1.3 70.3 Medical, optical instruments, watches and clocks (33) 98 .4 4.8 78.1 Paper and paper products (21) 80 .3 4.2 81.7 Publishing and printing (22) 36 .1 4.0 72.2 Other mining and quarrying (14) 34 .1 11.8 92.7 Recycling (37) 0 .0 5.1 89.4 Electricity, gas, steam and hot water (40) 0 .0 2.4 54.3 Collection, purification and distribution of water (41) 0 .0 7.5 78.0 Exports other than industrial exports 2618 9.9 Total 26485 100.0 Sources: the Enterprise Survey Data 2004; Jenkins (2004); the UN Comtrade System. Notes: a Between parentheses are the VSIC codes at the two digit level. b Employment coefficient reflects the number of workers needed to produce one billion dong worth of goods. Following Jenkins (2004), output for apparel (18) and footwear and leather products (19) are multiplied by 2.5 and by 2 respectively, assuming that the gross output recorded in the Enterprise Survey data refers not to the value of the products but to the processing costs. Without this adjustment, the employment coefficients for apparel (18) and footwear and leather products (19) are 13.7 and 11.4 respectively. 49 Appendix Table A.5.Evolution of Industrial Employment by the VSIC Sub-sectors (1000s) 2000 2001 2002 2003 2004 2005 2006 2007 Apparel, textiles (17, 18) 355 394 509 602 666 700 789 901 Footwear, leather products (19) 297 332 397 472 518 551 582 615 Furniture, wood, manufact. n.e.s. (20, 36, 37) 129 157 203 257 329 376 435 483 Food products (15, 16) 280 311 358 392 425 442 454 478 Electronics, machinery, transport equipment (29-35) 153 183 221 253 287 319 370 440 Metal, non-metallic mineral products (26, 28) 179 206 248 287 332 350 378 418 Basic manufacturing (21, 22, 23, 27, 40, 41) 161 175 195 211 245 250 297 316 Chemical, rubber, plastic products (24, 25) 117 127 153 170 195 204 227 253 Mining (10, 11, 13, 14) 153 129 155 163 166 178 180 186 Total 1823 2014 2441 2807 3162 3370 3711 4091 Source: the Enterprise Survey data 2000-2007 (GSO). Appendix Table A.6. Evolution of Industrial Employment by Regions (1,000s) 2000 2001 2002 2003 2004 2005 2006 2007 HCMC 485 547 672 742 836 869 879 918 Southeasta 300 359 478 586 702 782 916 1041 Hanoi 163 178 210 235 248 267 275 291 Red River Deltab 248 255 305 381 427 466 545 637 Northeast 171 173 200 214 238 244 265 298 Northwest 7 8 10 14 14 15 17 19 North Central Coast 58 66 75 85 91 93 103 114 South Central Coast 122 137 160 183 207 215 234 253 Central Highlands 26 31 35 40 43 46 47 51 Mekong River Delta 127 139 170 199 224 249 290 332 Others 115 122 126 128 131 125 140 138 Total 1823 2014 2441 2807 3162 3370 3711 4091 Source: the Enterprise Survey data 2000-2007 (GSO). Notes: aThe Southeast region excludes HCMC. b The Red River Delta region excludes Hanoi. 50 Appendix Table A.7. Summary Statistics Total Unskilled Skilled Change Change Change 2002 2004 2002-2004 2002 2004 2002-2004 2002 2004 2002-2004 Personal Characteristics Observations 1746 1746 1065 1065 656 656 Female (%) .36 .36 .31 .31 .46 .46 (.48) (.48) (.46) (.46) (.50) (.50) Minority (%) .073 .073 .066 .066 .087 .087 (.26) (.26) (.25) (.25) (.28) (.28) Education (years) 9.25 9.38 7.21 7.30 12.41 12.61 (4.23) (4.26) (3.55) (3.59) (3.08) (3.03) Potential Experience 18.63 20.45 18.01 19.86 19.58 21.33 (years) (10.26) (10.32) (10.66) (10.75) (9.64) (9.66) Married (%) .68 .71 .025 .60 .63 .030 .80 .82 .015 (.47) (.45) (.20) (.49) (.48) (.21) (.40) (.38) (.18) Urban (%) .40 .40 .30 .30 .55 .55 (.49) (.49) (.46) (.46) (.50) (.50) Monthly wage 735.16 922.31 576.26 695.17 976.34 1260.19 (1,000 dong) (596.44) (715.31) (415.99) (504.16) (745.21) (848.34) Log monthly wage 13.27 13.49 .22 13.06 13.24 .18 13.59 13.86 .27 (.70) (.72) (.51) (.66) (.67) (.55) (.64) (.63) (.44) Province-specific Characteristics Observations 61 61 61 Export Index 2.35 2.93 .59 (4.76) (6.10) (1.44) Export US Index .42 .76 .34 (.82) (1.65) (.84) US tariff cut measure .45 (.87) Share of agriculture 2002 (%) .620 (.178) Share of construction 2002 (%) .041 (.022) Share of government 2002 (%) .053 (.018) Share of industry 2002 (%) .113 (.075) Share of service 2002 (%) .174 (.096) Sources: Author’s calculation based on the VHLSS 2002, VHLSS 2004 and the Enterprise Survey data. Note: Between parentheses are the standard deviations. 51 Appendix Table A.8. OLS Regression Results: Explaining Wage Growth with the U.S. Tariff Cut Measure Model A Model B and Model C Total Unskilled Skilled Total Unskilled Skilled U.S. tariff cut measure .029* .066*** .0068 .010 .045* -.027* (.015) (.017) (.014) (.018) (.024) (.014) ∆Skilled/unskilled status -.057 .096 .028 -.082 .058 -.067 (.062) (.093) (.063) (.065) (.091) (.078) ∆Marital status -.12* -.14* -.015 -.10 -.14* -.0022 (.072) (.077) (.094) (.063) (.069) (.093) Initial wage in log -.26*** -.38*** -.27*** -.40*** -.49*** -.33*** (.025) (.031) (.037) (.031) (.034) (.042) Female -.096*** -.24*** .021 (.024) (.034) (.036) Minority .00016 -.13 .10 (.089) (.12) (.071) Education .041*** .023*** .027*** (.0035) (.0046) (.0067) Potential Experience .0016 -.000014 .00046 (.0015) (.0016) (.0018) Urban .043 .059 .0076 (.031) (.042) (.042) -.34 -.13 -.77*** Agriculture share 2002a (.23) (.28) (.25) -.0041 .29 -.52 Industry share 2002a (.48) (.43) (.38) -.067 1.44 -1.31 Construction share 2002a (.73) (1.06) (1.14) -1.40 -.52 -2.77** Government share 2002a (1.24) (1.42) (1.20) Constant 3.60*** 5.07*** 3.90*** 5.37*** 6.43*** 5.10*** (.32) (.39) (.40) (.44) (.46) (.65) Observations b 1718 1051 647 1718 1051 647 2 R .118 .187 .146 .209 .272 .186 Notes: The dependent variable is the log difference of monthly wage (∆Ln W). The classification between skilled and unskilled workers is based on the status in 2002. *, **, *** indicate that the coefficients are significant at the 10, 5, and 1 percent level respectively. The standard errors in parentheses are based on heteroskedasticity-consistent estimates of the variance- covariance matrix and corrected for within province correlation (clustering). a The share of workers employed in the service sector is the omitted category. b The total sample includes those who worked for “Army force�. As the skill category is not assigned for the latter workers, the number of skilled and unskilled workers do not add up to total number. 52 Appendix Table A.9. Impact of Change in Exports on Skill Premium with the U.S. Tariff Cut Measure (1) (2) (3) Skill*U.S. tariff cut measure -.038*** -.047*** (.012) (.012) U.S. tariff cut measure .058*** .026 (.016) (.016) Skill status .30*** .23*** (.038) (.035) ∆Skilled/unskilled status .075 .022 (.063) (.068) ∆Marital status -.11 -.12* (.069) (.064) Initial wage in log -.34*** -.41*** (.027) (.030) Female -.12*** (.025) Minority -.019 (.088) Education .025*** (.0036) Potential Experience -.00031 (.0013) Urban .038 (.030) Agriculture share 2002a -.33 (.22) Industry share 2002a .16 (.43) Construction share 2002a .33 (.75) Government share 2002a -1.44 (1.17) Constant 4.57*** 5.57*** (.34) (.43) Observations 1698 1698 2 R .177 .228 Note: The dependent variable is the log difference of monthly wage (∆Ln W). The standard errors in parentheses are based on heteroskedasticity-consistent estimates of the variance-covariance matrix and corrected for within province correlation (clustering). *, **, *** indicate that the coefficients are significant at the 10, 5, and 1 percent level respectively. The classification between skilled and unskilled workers is based on the status in 2002. a The share of workers employed in the service sector is the omitted category. 53