mar07_Article2 3/12/07 5:43 PM Page 39

Journal of Economic Literature Vol. XLV (March 2007), pp. 39–82

Distributional Effects of in Developing Countries

∗ PINELOPI KOUJIANOU GOLDBERG AND NINA PAVCNIK

The authors discuss recent empirical research on how globalization has affected income inequality in developing countries. They begin with a discussion of conceptu- al issues regarding the measurement of globalization and inequality. Next, they pres- ent empirical evidence on the evolution of globalization and inequality in several developing countries during the 1980s and 1990s. The authors then examine the chan- nels through which globalization may have affected inequality, discussing theory and evidence in parallel. They conclude with directions for future research.

1. Introduction that these are generally not better off, at least not relative to workers with higher skill or ne of the few uncontroversial insights of education levels. What explains this apparent trade theory is that changes in a coun- O paradox? Is the theory underlying the con- try’s exposure to international trade, and ventional wisdom too stylized to capture the world markets more generally, affect the dis- reality of the developing world? Or were tribution of resources within the country and there other forces at work that may have can generate substantial distributional con- overridden the effects of globalization? What flict. Hence, it comes as no surprise that the are the mechanisms through which globaliza- entry of many developing countries into the tion affected inequality? Did the experience world market in the last three decades coin- vary across countries and, if so, why? What cides with changes in various measures of are the general lessons we can draw from the inequality in these countries. What is more experience of the last three decades? It is surprising is that the distributional changes these and other related questions that this went in the opposite direction from the one article aims to address. suggested by conventional wisdom: while To this end, we present a large amount of globalization was expected to help the less evidence from several developing countries skilled who are presumed to be the locally regarding their exposure to globalization relatively abundant factor in developing and the parallel evolution of inequality. countries, there is overwhelming evidence While the evidence is subject to several measurement problems that we discuss ∗ Goldberg: . Pavcnik: Dartmouth. We extensively in this article, two trends emerge thank the editor, Roger Gordon, and two anonymous ref- clearly from the data analysis. First, the erees for many helpful comments. This research is sup- ported by funding from the National Science Foundation, exposure of developing countries to interna- Grant SES #0213459. tional markets as measured by the degree of

39 mar07_Article2 3/12/07 5:43 PM Page 40

40 Journal of Economic Literature, Vol. XLV (March 2007)

trade protection, the share of imports and/or and changes in inequality. By the mid- exports in gross domestic product (GDP), the 1990s, the economic landscape had however magnitude of capital flows—foreign direct changed, and factors other than trade liber- investment (FDI) in particular, and exchange alization, such as increased capital flows, rate fluctuations has increased substantially FDI, exposure to exchange rate fluctuations in recent years. Second, while inequality has that in turn affected exports, immigration, many different dimensions, all existing meas- etc., became increasingly more important ures for inequality in developing countries aspects of these countries’ integration in the seem to point to an increase in inequality, world market. Establishing a connection which in some cases (e.g., pre-NAFTA between these phenomena and inequality is Mexico, Argentina in the 1990s) is severe. more challenging compared to the case of We next investigate the question whether trade barrier reductions, but we discuss we can establish a causal link between the these aspects of globalization when related increase in inequality and globalization. We evidence is available. examine several mechanisms through which From a methodological point of view, we openness is presumed to have affected explore a variety of possible approaches to inequality and discuss related evidence. Our identify the impact of globalization on analysis here draws on several empirical inequality. A common theme across the studies of globalization and inequality in studies we draw upon is that they focus developing countries as well as existing sur- almost exclusively on the experience of par- veys of related topics (Ann E. Harrison and ticular developing countries within a rela- Gordon H. Hanson 1999; Adrian Wood tively short time span. While our survey has 1999; Pinelopi Koujianou Goldberg and a clear comparative aspect as we rely on evi- Nina Pavcnik 2004). We confine our discus- dence from a large set of countries, we sion to the experience of developing coun- abstain from relying on cross-country regres- tries in the last two to three decades. The sions to econometrically identify the effects primary reason for this focus is that meas- of trade policy changes or conducting com- ures of inequality are typically computed parisons of inequality measures over longer based on household survey data, and such time horizons. This focus is primary dictated data did not become available until the late by data constraints. Inconsistencies in the 1970s in many developing countries. In gen- measurement of inequality across countries, eral, the data have become more reliable changes in the household survey response over time, so that studies focusing on more rates over time as incomes rise, and frequent recent years tend to produce more credible changes in the design of household surveys results. The second reason we focus on the within the same country make inference last three decades is that, during that peri- based on cross-country evidence, or compar- od, many developing countries underwent isons of inequality measures over longer significant trade liberalization that substan- periods of time within a specific country, tially increased their exposure to interna- potentially less reliable compared to infer- tional markets. We argue that for many ence that relies on within-country evidence countries, most notably Latin American over shorter periods of time. To delineate countries in the 1980s and early 1990s and the scope of this study, we should also point India in the early 1990s, trade liberalization out that we focus our discussion on inequal- episodes represent a major part of their ity alone and not poverty, as the latter is dis- globalization. Furthermore, we argue that cussed extensively in a recent article in this the trade barrier reductions that occurred journal by L. Alan Winters, Neil McCulloch, during this period can be exploited to estab- and Andrew McKay (2004). Finally, we lish a causal link between trade openness abstract from effects of globalization on mar07_Article2 3/12/07 5:43 PM Page 41

Goldberg and Pavcnik: Distributional Effects of Golbalization 41

inequality that may have occurred through measurement of globalization and inequality the growth channel since the evidence on respectively. In section 3, we present empiri- the causal link between trade openness and cal evidence on the evolution of globalization growth has been controversial and inconclu- and inequality in developing countries and sive to date. However, this channel is poten- identify the main facts and trends that tially important; the perhaps most significant demand explanation. Section 4 discusses the benefit of globalization is presumed to be methodological challenges one faces in that it fosters economic growth, and growth attempts to causally link globalization to itself brings about distributional changes. inequality. Section 5, the core section of the Regarding our conclusions, we identify sev- paper, examines the channels through which eral channels that may explain why the recent globalization might affect inequality by pre- experience of developing countries did not senting theory and evidence in parallel. We conform to the “naive” thinking about global- start by focusing on the narrowest measure of ization. We argue that our understanding of inequality—the wage gap between skilled and the consequences of globalization for inequal- unskilled workers (or skill premium)—and ity has improved as the theoretical framework investigate the main globalization-related underlying the empirical work expanded to explanations for its documented increase. We include trade in intermediate products, inter- then progressively move to discuss the national flows of capital, trade-induced skilled impact of openness on broader concepts of biased technological change, short-run factor inequality. Section 6 concludes. immobility, and firm heterogeneity. We also find that the effect of globalization on 2. Conceptual Issues inequality depends on many factors, several 2.1 Measuring Globalization of which are country and time specific, including a country’s trade protection pattern Globalization is a broad concept casually prior to liberalization; the particular form of used to describe a variety of phenomena that liberalization and sectors it affected; the flex- reflect increased economic interdependence ibility of domestic markets in adjusting to of countries. Such phenomena include flows changes in the economic environment, in par- of goods and services across borders, reduc- ticular the degree of within-country labor and tions in policy and transport barriers to capital mobility; and the existence of other trade, international capital flows, multina- concurrent trends (e.g., skill-biased techno- tional activity, foreign direct investment, logical change) that may have interacted with outsourcing, increased exposure to exchange or even partially been induced by globaliza- rate volatility, and immigration. These move- tion. Given that different countries experi- ments of goods, services, capital, firms, and enced globalization in different ways and at people are believed to contribute to the different times, it is hardly surprising that the spread of technology, knowledge, culture, relevant mechanisms through which inequal- and information across borders. Research on ity was affected are case specific. From a pol- the effects of globalization in economics has icy point of view, this implies that attempts to concentrated on those aspects of globaliza- alleviate the potentially adverse distributional tion that are easier to capture empirically. effects of globalization in the short or medi- Accordingly, we confine our discussion on um run need to be grounded in a careful the more narrowly defined components of study of the nature of globalization and the globalization: trade liberalization, outsourc- individual circumstances in each country. ing, flows of capital across borders in the The remainder of this article is organized form of FDI, and exchange rate shocks. as follows. In section 2, we review some Even when one hones in on a narrow basic conceptual issues regarding the dimension of globalization, measurement mar07_Article2 3/12/07 5:43 PM Page 42

42 Journal of Economic Literature, Vol. XLV (March 2007)

challenges abound. The first hurdle is data course, tariffs are not the only policy instru- availability. Detailed information on trade ment through which governments in devel- barriers, outsourcing, or foreign direct oping countries regulate imports. Imports investment is often not readily available, into developing countries are also subject to especially when the analysis requires highly nontariff barriers to trade (NTBs) such as disaggregate data or longer periods of time import licenses and quotas. The information that span periods of policy liberalization. For on NTBs is often not available or not avail- example, in their recent survey of trade able at the same level of product/industry costs, James E. Anderson and Eric van aggregation as tariffs, especially for longer Wincoop (2004) note that data on trade pol- time periods surrounding trade liberaliza- icy barriers from UNCTAD’s TRAINS data tion episodes. Moreover, because many base that is systematically available for a NTBs are forms of protection that limit the large set of countries only covers years from quantity of imports allowed to enter a coun- 1989 onwards. In addition, in a given year, at try (rather than price-based measures), they most 17 percent of the included countries are more difficult to accurately measure. report both tariff and nontariff barriers to Researchers usually capture the extent of trade and trade flows. The lack of reporting NTBs at some level of industry aggregation is especially pronounced in developing by a nontariff barrier coverage ratio, which counties. Consequently, researchers have measures the share of products (or total often measured trade liberalization indirect- imports, or national production) in an indus- ly through more readily available data on try aggregate that is subject to NTBs. This trade volumes (i.e., exports and imports). measure however does not capture the true One problem with this approach is that trade restrictiveness of NTBs: for example, a cer- volumes are determined not only by (plausi- tain industry may have the same NTB cover- bly exogenous) changes in trade policy and age ratio in two different years, yet the NTB transportation barriers, but also by endoge- could be more or less restrictive in one of nous variables, some of which are in fact the the years because of different demand con- focus of interest in the globalization and ditions. As a result, measurement problems inequality debate (i.e., wages). As a result, are more severe in the case of NTBs and more recent studies have mainly relied on their comparability across countries, indus- national data sources to obtain trade policy tries, and time is more of an issue than in the information, as well as information on FDI case of tariffs. While the omission of reliable and outsourcing, spanning periods of policy NTB measures and their changes in empiri- reforms. cal studies is a potentially serious limitation, Trade liberalization episodes, and in par- a somewhat encouraging result is that corre- ticular reductions in tariff barriers, are per- lations between tariff rates and NTB cover- haps the most commonly studied age ratios (and their changes), whenever component of globalization. This focus is available, are positive, indicating that tariffs determined by practical considerations: tar- and NTBs have been used in recent years in iffs are relatively easier to measure than developing countries as complements and other forms of globalization. Because tariffs not substitutes.1 In terms of the interpreta- are usually imposed as ad valorem taxes on tion of empirical results, such correlations imported goods, they represent price based imply that the effect that is typically attrib- forms of trade protection. As such, they are uted to a tariff change represents an over- transparent, relatively easier to measure estimate of the pure tariff effect as it reflects consistently across industries and over time, and their magnitude reflects the true 1 See Goldberg and Pavcnik (2005), p. 89–90, for a dis- restrictiveness of the trade barrier. Of cussion of this correlation for the case of Colombia. mar07_Article2 3/12/07 5:43 PM Page 43

Goldberg and Pavcnik: Distributional Effects of Golbalization 43

the combined effect of the tariff and NTB 0.9 in years where both of these measures change. are available (Goldberg and Pavcnik 2005). Even if one limits the analysis to tariffs, Naturally, the focus on trade policy in measurement concerns remain. One of the studying the effects of “globalization” on most significant ones is aggregation. inequality is only useful to the extent that National governments set tariffs at a very trade policy is an important component of a disaggregate level as detailed tariff lines. country’s exposure to globalization. This was Researchers however typically need to the case in many of the countries that we dis- aggregate these tariffs to a higher level to cuss in this article, namely Latin American match the level of industry aggregation at countries such as Brazil, and Colombia, and which the outcome of interest, such as wages Mexico during the late 1980s/early 1990s or employment, is reported. This requires and India during the 1990s. In other settings, the use of concordances between tariff lines most notably Mexico after the implementa- and industries that are notoriously noisy, so tion of NAFTA, channels other than trade that aggregate industry tariffs are plausibly policy, for example, immigration, foreign measured with error. In addition, aggrega- direct investment, outsourcing, and the peso tion discards some potentially important crisis, have played a potentially more impor- variation in tariffs (or tariff changes) within tant role. Still, average tariff rates continue industry groups and thus precludes the to be high in many developing countries, researcher from examining some channels including some that have recently imple- through which individuals/firms adjust to mented trade reforms. India provides the trade liberalization within broadly defined most striking example. Although India industries. underwent a drastic trade liberalization A further concern is that industry tariffs reform starting in 1991, the average tariff in on final goods do not capture the true extent manufacturing was over 30 percent in 1999 to which an industry is affected by protec- (Petia Topalova 2004a). Thus, there remains tion (or liberalization) since they do not substantial scope for further tariff and NTB account for intermediate good linkages. One reductions and trade policy is likely to con- could, in principle, capture such linkages by tinue to be an important component of glob- constructing effective rates of protection, alization at least in some of the lower income which take into account not only the direct developing countries. protection granted to an industry through In addition to the role of trade reforms in nominal tariffs on final products but also the fostering trade in final goods, recent work by indirect one that results from tariffs on inter- Robert C. Feenstra and Hanson (1996, mediate inputs. Unfortunately, effective 1997, 2003) has emphasized the growing rates of protection are not readily available importance of trade in intermediate inputs. for many countries over periods that span This phenomenon is also referred to as “out- trade liberalization episodes. In addition, sourcing” or “production sharing.” Recent effective rates of protection present addi- trade liberalizations, coupled with the tional measurement/concordance problems removal of restrictions on capital flows and stemming from the use of information from technological change, have enabled firms to the input output tables required in their “outsource” some stages of production to construction. Fortunately, in cases where cost-minimizing locations abroad, either both nominal and effective measures of pro- through arm’s length imports of intermedi- tection are available, they tend to be highly ate inputs or by setting up their own produc- correlated. For example, the correlation tion facilities in a host country. A country can between industry effective rates of protec- be exposed to outsourcing as a purchaser of tion and industry tariffs in Colombia is above outsourcing activities (for example, firms in mar07_Article2 3/12/07 5:43 PM Page 44

44 Journal of Economic Literature, Vol. XLV (March 2007)

Hong Kong have been importing relatively capture the intensity of multinational activity labor-intensive intermediate products from by computing the share of foreign affiliates in China since the 1980s) or as a host of out- total industry employment or output to cap- sourcing activities (for example, Mexico’s ture horizontal linkages, or by additionally maquiladoras have been used to assemble applying input–output tables to this informa- intermediate products into final goods made tion to capture an industry’s exposure to FDI for U.S. markets since the early 1980s). through vertical linkages. One concern with In empirical work, one would ideally like to this measurement approach, raised recently rely on a measure of exposure to outsourcing by Wolfgang Keller and Stephen R. Yeaple that is related to plausibly exogenous changes (2003) in the context of the United States, is in trade and capital controls. From the receiv- that measures of an industry’s exposure to ing country’s perspective, this is subject to the FDI are highly sensitive to how the economic same data constraints we discussed in the con- activity of a foreign affiliate is allocated across text of effective rates of protection. the various industries in which it is active (for Consequently, the literature has mainly used example, main line of business versus other the share of imported inputs in total pur- lines). Another more general concern with chased intermediate inputs in an industry as a these measures of FDI is that the decision of measure of outsourcing (see Feenstra and a multinational to purchase an existing plant Hanson 2003; Chang-Tai Hsieh and Keong T. or to locate in a country/industry may Woo 2005). Because direct data on imported depend on unobserved wage and worker inputs by industry are often not available, the characteristics in a firm/industry/region, above outsourcing measure is constructed by which creates the potential for simultaneity combining information from input–output and selection bias. tables with information on total trade flows of Finally, the removal of capital controls com- final products. As a result, it is subject to the bined with a shift away from fixed and toward same endogeneity concerns as trade flows. more flexible exchange rate arrangements in Furthermore, this measure of outsourcing many developing countries has exposed these suffers from the same measurement problems countries to greater exchange rate volatility. we discussed earlier in the context of tariffs To the extent that these exchange rate regarding the concordances between trade changes are partially passed through onto data, industry data, and input–output tables. prices, the increased exposure to exchange Related to “global production sharing” is rate volatility impacts firms’ incentives to the presence of multinational firms and for- export (or import) and, hence, presents anoth- eign direct investment in developing coun- er channel through which globalization may tries. Their increased presence stems in part have affected inequality.2 The advantage of from the recent removal of controls on capital using exchange rate shocks as a measure of flows in these economies. The information on globalization is that they are easy to measure, affiliates of multinational companies in devel- plausibly exogenous—at least from a single oping countries is usually obtained from industry’s perspective, and large in magnitude. national surveys of firms, such as the Census of Manufacturers. In some countries, for 2 The evidence suggests that, for exchange rate shocks example Indonesia and Mexico, these surveys of the magnitude recently witnessed in several Latin provide information on the nationality of the American and Asian economies (for example, Mexico in 1994; Brazil in 1998; and Thailand, Korea, Indonesia, etc. capital sources so that one can identify in 1997), the price effects are larger than what is typically whether a particular firm is partly foreign- observed for more modest shocks. In these cases, the owned. These surveys are also used to create exchange rate shocks should be more accurately charac- terized as currency crises. The fallout from currency crises measures of the presence of multinationals in has potentially its own implications for inequality, but we an industry or region. Such measures usually do not explore these implications in this study. mar07_Article2 3/12/07 5:43 PM Page 45

Goldberg and Pavcnik: Distributional Effects of Golbalization 45

The disadvantage is that they represent aggre- countries do not consistently report expendi- gate shocks to an economy; they do not exhib- tures in their household surveys. The Living it any variation across industries or plants, so Standards Measurement Surveys project of that separating their effect from the one of the aims at changing this pattern, other concurrent macroeconomic shocks or so that research in future years may be able to policies can be challenging. take advantage of expenditure data to meas- ure inequality. To date however, most empiri- 2.2 Measuring Inequality cal studies had to contend themselves with The ideal measure of inequality would be employing income based measures of based on comparisons of individuals’ well- inequality, given that some measure of income being over their entire lifetime. The most is always included in household surveys. The appropriate variable for capturing lifetime most frequently used inequality indices (such well-being is arguably consumption (see as the Gini coefficient or the coefficient of Angus Deaton 1997 for a related discussion). variation) are based on the second moments Compared to income, consumption offers of the observed income distribution. three advantages. First, to the extent that The suitability of these indices for captur- consumers can intertemporally shift ing true changes in inequality, especially over resources through lending and borrowing, longer periods of time, has been questioned current consumption better captures life- recently for a variety of reasons. First, even time well-being. This argument may be less though most household surveys include some relevant for developing economies charac- measure of income, the coverage of income terized by severe capital market imperfec- sources and taxes tends to vary both across tions, yet the evidence suggests that, even in countries and, for a specific country, across these countries, some borrowing and lend- years; items such as in-kind gifts and govern- ing does take place (though this may occur in ment transfers, implicit rent from own hous- informal credit markets and at exceedingly ing, and capital income and profits tend to be high interest rates). Second, reporting prob- particularly problematic. To avoid these prob- lems are less pronounced for consumption lems, many studies have focused on a more than income. Specifically, it is well docu- narrow measure of inequality—wage inequal- mented that high-income households tend ity. A second set of problems is related to the to underreport their income (but not neces- fact that high-income households are known sarily their consumption), while most sur- to have higher nonresponse rates and under- veys collect data on pretax, and not after-tax report income, so that the income distribution income. Finally, many policies—trade poli- presented in household surveys is a truncated cies in particular—affect the relative prices version of the true one.4 Johan Mistiaen and of consumer goods so that they impact con- Martin Ravallion (2003) and Deaton (2003) sumers not only through income changes, have shown that, with nonresponse rates but also through changes in the purchasing increasing in income, it is possible that the power of their current incomes. Inequality variance of the truncated distribution is lower measures based on consumption data are by than the variance of the true distribution. In nature better suited to capture this effect. the context of inequality measurement, this Despite these advantages, consumption is counterintuitive result implies that indices rarely used as the basis for measuring inequal- ity in empirical studies of the effects of glob- 4 Miguel Szekely and Marianne Hilgert (1999), for alization.3 The reason is that many developing example, report that in many Latin American household surveys the top ten incomes reported in a given year are about the same as the salary of an average manager in the 3 Guido G. Porto (2006), Alessandro Nicita (2004), and country under consideration. This suggests that the truly Topalova (2004a) are notable exceptions. rich households are missing from the surveys. mar07_Article2 3/12/07 5:43 PM Page 46

46 Journal of Economic Literature, Vol. XLV (March 2007)

based on the second moments of the observed measures take into account consumption (truncated) income distribution may be mis- scale economies within the household and leading about changes in inequality; this is differences in the needs among individuals of especially the case if the comparisons involve different gender and age to construct scale- long periods of time during which income has and adult equivalent-adjusted versions of per substantially increased. On a similar note, capita income (see Deaton 1997). The prob- Abhijit Banerjee and Thomas Piketty (2004, lem with such adjustments is that the con- 2005) document that income data based on structed index of well-being will ultimately Indian tax returns (where underreporting is depend on the scale and adult-equivalency presumably less of an issue compared to parameters, which may be poorly known. household survey data) indicate that the “very Given the conceptual and measurement rich” in India, i.e., those who were in the top ambiguities involved in measuring inequali- 0.1 percent of the population, were getting ty, cross-country comparisons of inequality richer faster than anyone else in the 1990s. figures or investigations of long-term trends This group seems to be missing from the in developing countries appear problematic. Indian household survey (National Sample Studies of the effects of trade openness on Survey).5 Though tax return data provide a inequality have traditionally been narrower superior source of information for the pur- in focus, as the majority of them have ana- pose of documenting income inequality, they lyzed concrete trade liberalization episodes have not been used in studies of the causes or other policy changes in specific countries. behind changing trends in inequality since the Because most of these episodes unfolded confidential nature of the data prevents over the course of a few (2–3) years and the researchers from linking the income figures to related studies focus on one country at a other individual-specific variables of interest. time, many of the aforementioned measure- Another potential problem in inequality ment problems are less pronounced here. studies is that household surveys are often Furthermore, the increase in inequality doc- redesigned so that the wage or income data umented in many developing countries has are not comparable across years. Changes in been associated with an increase in the so- topcoding limits, for example, can affect the called skill premium, i.e., the wage gap range of top incomes reported in the surveys.6 between skilled and unskilled workers. In addition to these reporting problems, all Motivated by this finding, a substantial inequality studies face the conceptual issue of amount of related work has focused on an whether to focus on households or individu- even more narrow measure of inequality als. While the primary interest lies in the well- than the ones discussed above: the inequality being of individuals, people usually live in between skilled and unskilled workers. households and share resources. To take this The definition of skill varies depending on into account, many studies have focused on the kind of data employed. Studies that use some variant of per capita income. The sim- household survey or labor force survey data plest one is obtained by dividing household define skill based on the education of the income by family size; more sophisticated household head. Studies that exploit plant- or firm-level data typically differentiate 5 However, the authors point out that this group is too between production and nonproduction or small for its absence to explain important discrepancies in blue-collar and white-collar workers. This lat- the measurement of inequality and poverty based on NSS and the national accounts data. ter categorization is clearly unsatisfactory, 6 This was, for example, the case in Colombia, where especially since the skill composition of these a change in the topcoding procedures used in the groups is likely to vary over time. For many Encuesta Nacional de Hogares (National Household Survey) in 1994 affected the reported incomes of the countries however, plant-level data are more richer households. readily available over several years; moreover, mar07_Article2 3/12/07 5:43 PM Page 47

Goldberg and Pavcnik: Distributional Effects of Golbalization 47

they offer the advantage of providing infor- choice of time periods and countries is dic- mation about the sector of employment at a tated by the timing of trade reforms and data more disaggregate level compared to house- constraints. With few exceptions (Chile for hold surveys that in many developing coun- example), most developing countries did not tries report industry information only at the liberalize their trade regimes and did not two-digit level. Fortunately, cross-tabulations open their borders to foreign direct invest- of matched worker and employer surveys at ment until the 1980s. The countries dis- the plant level in the United States and the cussed in this section are representative in United Kingdom indicate a close relationship that sense since they all experienced drastic between the production/nonproduction sta- trade liberalization during the past two tus of workers and their educational level;7 decades. Furthermore, they all collect the nonproduction workers have more years of detailed micro data required to generate var- schooling and appear to be uniformly better ious measures of inequality that span the paid. Although there is no direct evidence on period before, during, and after policy this issue for developing countries, these cor- changes that increasingly exposed these relations are encouraging regarding the suit- countries to international markets. ability of plant-level data for analyzing the Consequently, these countries have served differential impact of globalization on work- as a testing ground for most empirical ers of different skill level. As with the income research investigating the channels through or wage based measures of inequality, com- which globalization may have affected parisons over short periods of time within a inequality. country are likely to be more credible than 3.1 Globalization cross-country comparisons or analyses of long time trends. Table 1 provides an overview of the glob- alization experience of the countries men- 3. Overview of the Evidence tioned above (changes in trade policy and Despite the difficulties associated with the other relevant measures of globalization) measurement of globalization and inequali- along with the reported changes in inequali- ty, research in the past fifteen years has tried ty measures. The same table also lists other to document their evolution by increasingly major reforms that took place during the relying on new and better data sources. In 1980s and 1990s in each of these countries. this section, we summarize the existing evi- Let us first focus on changes in globaliza- dence, focusing on the experience of a few tion measures, starting with trade liberaliza- representative countries (Mexico, Colombia, tion episodes. Table 1 indicates that, Argentina, Brazil, Chile, India, and Hong although some countries (i.e., Argentina and Kong) during the 1980s and 1990s.8 Our Colombia) experimented with short-lived trade reforms during the late 1970s, most 7 See Eli Berman, John Bound, and Stephen Machin (1997) and Machin, Annette Ryan, and John Van Reenen countries implemented unilateral trade (1996). reforms in the mid to late 1980s and early 8 One obvious omission is the set of Southeast Asian 1990s: Mexico 1985–87, Colombia 1985–91, countries (South Korea, Taiwan, and Singapore) that underwent trade reforms in the 1960s and 1970s. Argentina 1989–93, Brazil 1988–94, India Unfortunately, neither detailed data on tariffs nor micro 1991–94. Chile is an exception as it liberal- surveys are readily available for these countries. The exist- ized its trade regime early, from 1974 to ing evidence on these countries has been discussed in detail in Wood (1999). We also leave out China because empiri- 1979. cal work on the relationship between inequality and global- An important feature of the above reforms ization has just recently started to emerge for this country. was that they drastically reduced tariffs, Shang-Jin Wei and Yi Wu (2002) is to our knowledge the only study that examines the link between openness and which were high prior to liberalization and a inequality in China. crucial component of trade protection. The mar07_Article2 3/12/07 5:43 PM Page 48

48 Journal of Economic Literature, Vol. XLV (March 2007)

TABLE 1 GLOBALIZATION AND INEQUALITY IN SELECT DEVELOPING COUNTRIES

1970s 1980s 1990s MEXICO Globalization Unilateral trade NAFTA (1994) Measures liberalization 1985–87 Peso Crisis (WTO entry) Maquiladoras expansion Devaluation FDI Maquiladoras Immigration liberalization (1983) FDI liberalization (1989) Immigration Inequality Skill premium Increased Increased until mid-1990s Stable/declined after mid-1990s Increased between 2000–1990 Wage white collar/ Declined 1965– Increased Increased until mid-1990s Wage blue collar 80 Stable after mid-1990s 90-10 log N.A. Increased Increased up to 1996 wage differential Stable/decline after mid-1990s Gini of log wages Increased Increased up to mid-1990s Stable/decline after mid-1990s Income Inequality Declined Increased Stable/decline (Gini) Other Reforms Privatization Banking Crisis Labor Market Reform Deregulation COLOMBIA Globalization Partial Trade Reform Gradual trade liberalization Trade liberalization 1990–91 Measures starting 1979 starting 1985 Devaluation Inequality (urban) Skill Premium Slightly Declined Increased 90–10 log wage Slightly Declined Increased differential 1986–90 Gini of log wages Stable/ Slight Decline Increased Income Inequality Declined Stable/Increased Stable (Gini) Other Reforms Labor market reform 1990 Banking reform 1993

ARGENTINA Globalization Short Trade Reform Unilateral Trade Trade liberalization cont. Measures (1976–82) Liberalization (1989–93) Mercosur 1991 Appreciation Appreciation Inequality (urban) Skill Premium Decreased Increased Gini of log wages Increased Increased Income Inequality Increased Increased Increased Other Reforms Macroeconomic crisis Deregulation (1988–89) Privatization Privatization Financial liberalization Deregulation in early 1990s Financial Liberalization Convertibility Plan in the late 1980s mar07_Article2 3/12/07 5:43 PM Page 49

Goldberg and Pavcnik: Distributional Effects of Golbalization 49

TABLE 1 (continued)

1970s 1980s 1990s BRAZIL Globalization Partial unilateral trade Unilateral trade liberalization Measures liberalization (1988 onwards) (ends 1994) Mercosur 1991 Currency Crisis 1998 Inequality (national) Skill Premium N.A. Stable/Slight Increase Increased Mean log deviation N.A. Stable/Increased Stable of wage Gini of log wages Stable Stable/Small decline Income Inequality Stable Increased Stable/Small decline Other Reforms Labor market reform CHILE Globalization Trade Devaluation Measures Liberalization Inequality Skill Premium Increased Increased Declined early 1990s Overall increased 1990–2000 (national data) Wage white collar Increased /Wage blue collar Gini of log wages Increased Increased Decreased relative to late 1980s Stable during the 1990s Income Inequality Increased Increased Stable/Small increase late 1990s (national) Other Reforms Structural Reforms Devaluation Privatization Macroeconomic crisis Deregulation Tex Reform Labor Market Reform INDIA Globalization Limited Removal of Trade Liberalization 1991 Measures Import Licenses Unilateral FDI liberalization Inequality (urban) Skill Premium Relatively stable Increased 90-10 log wage Increased Increased more rapidly differential Income Inequality Increased Consumption Stable/Slight Increase Increased inequality Other Reforms Industrial delicensing Tax Reform Financial Reform HONG KONG Globalization Outsourcing to China Outsourcing to China Measures Inequality Skill Premium Slight decline Increased Increased (return to education) Wage non- Declined Increased Increased production/Wage production workers

(continued on next page) mar07_Article2 3/12/07 5:43 PM Page 50

50 Journal of Economic Literature, Vol. XLV (March 2007)

TABLE 1 (continued)

Notes: General sources on Latin American countries: Schady and Paramo-Sanchez (2003), Gasparini (2003), Robbins (1996). Sources for information on specific countries are: Argentina: Gasparini (2003); Brazil: Green, Dickerson, and Arbache (2001); Chile: Francisco H. G. Ferreira and Julie A. Litchfield (1999), Robbins (1996), Beyer, Rojas, and Vergara (1999), Pavcnik (2003); Mexico: Robertson (2004), Hanson (2004), Cragg and Eplebaum (1996); India: Kijima (2006), Topalova (2004a), Deaton and Dreze (2002); Colombia: Attanasio, Goldberg, and Pavcnik (2004); India: Kijima (2006), Topalova (2004a), Deaton and Dreze (2002); Hong Kong: Hsieh and Woo (2005).

high tariff rates reflect the lack of participa- Bernard, J. Bradford Jensen, Peter K. tion of most developing countries in the tar- Schott 2005). iff-reducing rounds of the GATT/WTO In addition to tariff reductions, the unilat- prior to their unilateral trade reforms: some eral trade reforms also reduced NTBs. developing countries were not GATT mem- Unfortunately, as discussed earlier, the infor- bers (for example, Mexico); others (such as mation on exact measures of NTBs is often Brazil, Colombia, and India) were GATT not available, especially for longer periods members on paper but did not have to surrounding trade liberalization episodes. reciprocate tariff concessions negotiated However, the available data on average NTB with the GATT until the Uruguay Round.9 coverage ratios in manufacturing industries Table 2 reports the average tariffs for the before and after the reforms (presented in manufacturing industries in the countries of columns 3 and 4 of table 2) suggest that table 1 in a year before and after the NTBs were high prior to trade reforms and reforms.10 The table illustrates that, prior to that liberalization drastically reduced their the reforms, tariff levels were high, ranging levels. For example, in Colombia the NTB from 117 percent in India to 23.5 percent in coverage ratio declined from 72.2 percent in Mexico. The comparison of average tariffs 1986 to 1.1 percent in 1992. In Mexico, the before and after the reforms suggests dras- share of manufacturing production subject tic tariff reductions: for example, 85 per- to import licenses dropped from 92 percent centage points in Chile, 73 percentage in 1985 to 23.2 percent in 1988. In India, the points in India, and 12.5 percentage points share of manufacturing imports covered by in Mexico. These tariff declines in develop- nontariff barriers dropped from 80 percent ing countries are in stark contrast to the low in 1990 to 17 percent in 1999 (Prachi Mishra tariff levels and rather minor tariff policy and Utsav Kumar 2005). Although we do not changes in the developed countries during have access to measures of NTBs in other this period. For example, in the Unites countries, NTBs were virtually eliminated in States—a country whose tariff policy resem- Chile (Rudiger Dornbusch and Sebastian bles the policy of most other developed Edwards 1994) and Brazil (Donald A. Hay economies—the average tariff was only 4.8 2001), while Argentina eliminated all import percent in 1982; tariffs declined on average licenses (Sebastian Galiani and Pablo by 0.6 percentage points to 4.2 percent Sanguinetti 2003). between 1982 and 1992 (Andrew B. Table 1 suggests that, subsequent to uni- lateral trade reforms, several countries also 9 Article XVIII of the GATT granted exemption from lowered their trade barriers vis a vis specific tariff concessions to developing countries. trading partners through regional trade 10 For each country, the actual year used to describe the period before and after the reforms is recorded below agreements. The most notable example is the country name in column 1. Mexico’s entry into a free trade agreement mar07_Article2 3/12/07 5:43 PM Page 51

Goldberg and Pavcnik: Distributional Effects of Golbalization 51

TABLE 2 GLOBALIZATION IN SELECTED DEVELOPING COUNTRIES

Trade Liberalization Trade Flows (% GDP) FDI inflows (% GDP) Average Tariff Average NTB Exports Imports Before After Before After 1980 2000 1980 2000 1980 2000 Argentina 45 12 n.a. declined 5.1 10.8 6.5 11.4 .88 4.09 Brazil 58.8 14.4 n.a. declined 9.1 10.9 11.3 12.1 .81 5.50 Chile 105 10 n.a. declined 22.8 31.8 27.0 30.8 .77 5.21 Colombia 50 13 72.2 1.1 16.2 21.9 15.6 20.4 .47 2.92 Hong Kong n.a. n.a. n.a. n.a. 89.9 150.0 90.8 145.3 n.a. n.a. India 117 39 82 17 6.1 14.0 9.7 16.6 .04 .51 Mexico 23.5 11 92 23.2 10.7 31.4 13.0 33.2 .96 2.31

Notes: The following years are specific dates refered above as before and after trade liberalization: Argentina (1988, 1994), Brazil (1987, 1994), Chile (1974, 1979), Colombia (1984, 1992), India (1991, 2000), Mexico (1985, 1988). Information on trade liberalization in selected countries is from Galiani and Sanguinetti (2003), Pavcnik et al. (2004), Dornbusch and Edwards (1994), Goldberg and Pavcnik (2005), Mishra and Kumar (2005), and Revenga (1996). World Development Indicators CD is the source of trade flow and FDI data.

with the United States and Canada in 1994 percent of its exports to the United States (NAFTA). Argentina and Brazil joined (Hanson 2004). Similarly, when China liber- Mercosur in 1991, along with Uruguay and alized its markets, many firms in Hong Kong Paraguay. These regional trade agreements shifted their relatively less-skilled-labor- likely induced changes in the geographic intensive activities to Chinese border composition of trade in these countries; how- regions, while specializing in higher-skill ever, the changes in trade policy implied by intensive activities, such as headquarter these agreements were substantially smaller services, at home. As a consequence, the than the declines in trade barriers observed share of intermediate inputs that were during the unilateral trade reforms. imported from China in Hong Kong’s total Furthermore, several countries (most intermediate inputs rose from less than 10 notably Mexico and Hong Kong) experi- percent in 1976 to almost 50 percent in 1996 enced increases in trade in intermediate (Hsieh and Woo 2005). A related develop- inputs associated with global production ment has been the growing presence of affil- sharing. For example, after the capital con- iates of multinational companies in trol liberalization in Mexico in the mid- developing countries during the 1980s and 1980s, many U.S. companies shifted 1990s following their capital market reforms. relatively low-skill intensive stages of pro- This is illustrated by the increased impor- duction to Mexico by setting up foreign tance of FDI inflows in the economies of assembly plants (maquiladoras). Intermediate developing countries. Table 2 reports FDI inputs were imported to Mexico, assem- inflows as a share of GDP in select countries bled in maquiladoras, and the final prod- and illustrates that, while the share of FDI in ucts exported to the United States. The total GDP was below 1 percent in 1980s in importance of maquiladoras for the these countries, it grew to about 3 percent in Mexico–U.S. trade was growing during the 2000 for Colombia and Mexico, to 4 percent 1980s and 1990s so that by 2000 maquilado- in Argentina, and 5 percent in Brazil. In ras accounted for 35 percent of Mexico’s India, however, it is still about 0.5 percent imports from the United States and for 48 of GDP. mar07_Article2 3/12/07 5:43 PM Page 52

52 Journal of Economic Literature, Vol. XLV (March 2007)

Finally, table 1 indicates that many devel- Epelbaum 1996). In other countries, the oping countries experienced large currency skill premium increased too, but by less: for fluctuations during the 1980s and 1990s. In example, the return to a university degree some instances, these exchange rate changes increased by 16 percent (relative to primary may have exposed the relevant countries to education) in Colombia between 1986 and international markets more than the trade 1998 (Orazio Attanasio, Goldberg, and reforms. Eric A. Verhoogen (2006), for Pavcnik 2004), by over 20 percent (relative example, argues that Mexico’s 1994 peso cri- to no complete education) in Argentina sis, during which the peso lost half of its orig- between 1992 and 1998 (Leonardo inal value, overshadowed the average tariff Gasparini 2004), by 13 percent in India (rel- changes from NAFTA. ative to primary education) between 1987 and 1999 (Yoko Kijima 2006), and by 10 per- 3.2 Inequality cent among men (relative to no complete The information on inequality is based on education) in Brazil (Gasparini 2003). Given empirical studies that have utilized micro that relatively large skill premium increases surveys of households or firms from the have been documented for several countries, country in question. The relevant sources it is unlikely that they are all a figment of the are cited in the notes to the table. Table 1 measurement problems discussed in section reports several measures of inequality: skill 2, although the exact magnitudes of the premium, wage inequality, income inequali- changes may be affected by these problems. ty, and consumption inequality. Note that A further pattern evident in table 1 is that because of data constraints, some of these the skill premium does not steadily increase measures, most frequently consumption throughout the two decades in all countries. inequality, are missing for many countries. Interestingly, the skill premium increases We begin by examining the evolution of seem to chronologically coincide with the the narrowest measure of inequality: the trade reforms in several countries. For wage gap between more and less skilled example, the skill premium grew steadily workers (the so-called skill premium). When during the 1980s and 1990s in Mexico,11 information on an individual’s education is which implemented a large trade reform in available, we use the returns to completed the mid-1980s and was continually exposed university degree as a measure of the skill to other forms of globalization, such as out- premium and report evidence based on a sourcing or FDI, for the next two decades. Mincerian regression; when data on the edu- On the other hand, skill premium increases cational attainment of workers are not avail- in Colombia, Brazil, Argentina, and India able, as is the case with plant surveys, we use were mainly confined to the 1990s; the latter the relative wage of white- to blue-collar countries implemented the bulk of their workers (or, alternatively, the relative wage trade reforms in the early 1990s. In Chile, of nonproduction to production workers), to where the reforms took place during the measure the skill premium. Several broad 1970s, the skill premium increased during patterns emerge. the 1970s and 1980s, declined in the early When we consider the 1980s and 1990s as a whole, all countries seem to have experi- 11 enced increases in the skill premium. The Most evidence on Mexico points to a rising skill pre- mium, at least until the mid-1990s. Gasparini (2003) and skill premium increases were largest in Hanson (2004) document skill premium increases over Mexico, where the return to university edu- the entire decade using nationally representative house- cation (relative to primary education) hold survey and population census data, respectively. However, Raymond Robertson (2004) argues that the skill increased by 68 percent between 1987 and premium declined (or remained relatively stable) after the 1993 (Michael Ian Cragg and Mario mid-1990s in urban areas. mar07_Article2 3/12/07 5:43 PM Page 53

Goldberg and Pavcnik: Distributional Effects of Golbalization 53

1990s (Donald J. Robbins 1996; Harald studies that decompose changes in wage Beyer, Patricio Rojas, and Rodrigo Vergara inequality into changes in the distribution of 1999), and then increased again between observable skills (such as education), 1990 and 2000 (Gasparini 2003). These changes in the prices of observable skills, time-series patterns have led many casual and changes in unobservables, which are observers to conclude that globalization was common in the literature on the evolution of the main source of growing inequality in inequality in the United States, rarely exist these countries. As we argue in the next sec- for developing countries. Kijima (2006) pro- tion, inference based on these before and vides an example of such decomposition. after comparisons can be misleading. She formally shows that most of the increase Finally, note that changes in the educa- in the postliberalization wage inequality in tion-based measure of the skill premium urban India can be attributed to increases in and the relative wage of white-collar to the prices for observable skills, and in partic- blue-collar workers tend to move in the ular to the return to tertiary education. same direction in countries and periods for However, the wage inequality increase of the which both measures are available. For 1980s (when returns to tertiary education example, in Mexico the average relative remained relatively stable), was largely due wage of nonproduction workers increased to changes in the quantity of observed skill. almost by a factor of 1.5 between 1987 and Similarly, Gasparini (2004) finds that wage 1995 (Robertson 2000). This parallel inequality increases during the 1990s in movement is reassuring for studies that Argentina can be to a large extent attributed rely on the white-collar/blue-collar distinc- to the rising skill premium, while changes in tion (or nonproduction/production worker the educational composition of the work- distinction) as a measure of skill. force importantly contributed to growing The observed changes in the skill premium wage inequality in the 1980s (when the skill are generally (but not always) reflected in premium actually slightly declined). changes in the wage inequality (usually Income-based measures of inequality measured by the Gini coefficient of log have been used less widely in the literature wages, or the 90–10 log-wage differential). on globalization and inequality. As men- As with the skill premium, wage inequality tioned earlier, this is partly due to the lack of increased in Mexico12 in the 1980s and early reliable survey data on nonwage sources of to mid-1990s, in Chile during the 1970s and income (especially in Latin American coun- 1980s, and in Colombia, Argentina, and India tries). Surveys that contain such information during the 1990s. Interestingly, increases in are more recent and often less frequently the skill premium are not mirrored in conducted than labor market surveys. The increases in wage inequality in Brazil, where limited information available in Latin the Gini coefficient remains remarkably sta- American countries (mainly drawn from ble during 1980s and 1990s (Carolina Gasparini 2003) suggests that income Sanchez-Parama and Norbert Schady 2003; inequality and wage inequality move in the Francis Green, Andy Dickerson and Jorge same direction, although changes in income Saba Arbache 2001; Gasparini 2003). Green, inequality are at times less pronounced than Dickerson, and Arbache (2001) attribute this changes in wage inequality or the skill pre- finding to the small share of university grad- mium (for example, in Mexico and Colombia uates in total population. Unfortunately, during the 1990s). Finally, a consumption- based measure of inequality is to our knowl- edge available over this period only for 12 Increases in wage inequality in Mexico during the 1990s are more pronounced in the first half of the decade, India, which has a nationally representative especially in urban areas. consumer expenditure survey that spans the mar07_Article2 3/12/07 5:43 PM Page 54

54 Journal of Economic Literature, Vol. XLV (March 2007)

1980s and 1990s.13 In urban areas, con- and 1990s contrasts with the experience of sumption inequality moves in the same several Southeast Asian countries (South direction as income and wage inequality; it is Korea, Taiwan, Singapore) that underwent relatively stable during the 1980s (a period trade reforms in the 1960s and 1970s. The prior to major liberalization) but increases latter observed a decline in inequality as they during the 1990s. Although this pattern can- opened up their economies to foreign mar- not be generalized to other countries, it is kets. We discuss the possible explanations reassuring that at least in the one case where proposed by Wood (1999) for the differ- both income and consumption inequality ences in these correlation patterns between measures are available, they both move in countries that globalized in the 1960s and the same direction. 1970s and countries that globalized in the In summary, the evolution of various 1980s and 1990s in section 5.1.1. measures of inequality suggests that most of Unfortunately, neither detailed data on tar- the developing countries experienced an iffs nor micro surveys are readily available increase in inequality during the past two for the early globalizers during the periods decades. More importantly, we find no evi- of reform to allow us to examine whether the dence that any measure of inequality declining inequality in these countries was decreased over this entire period when com- caused by globalization as opposed to being pared to earlier periods characterized by less just coincidental. globalization. As we note in the introduction In general, one needs to be careful draw- to this section, our discussion abstracts from ing conclusions regarding the link between several potentially important countries, most globalization and inequality simply based on notably China. Lee Branstetter and Nicholas before and after comparisons. Table 1 lists Lardy (2006) provide an excellent detailed other important reforms that took place dur- review of the process through which China ing periods of external liberalization in increasingly liberalized its trade and foreign selected countries. Perhaps the most striking direct investment policies during the 1980s feature of these reforms is the fact that not a and 1990s, culminating with the country’s single country implemented trade reforms entry into the WTO in 2001. During this or FDI liberalization in isolation from other period, income inequality in China has policy changes. For example, the most dras- increased (Wei and Wu 2002), so that tic trade policy liberalization in Colombia in China’s experience is consistent with the 1990–91 coincides with changes in labor positive correlation between inequality and market regulation that substantially exposure to globalization noted above. increased the labor market flexibility. The survey of the evidence confirms Mexico’s 1985 trade reform took place Wood (1999), who noted that inequality amidst privatization, labor market reform, increased in several middle-income Latin and deregulation. These concurrent policy American countries that liberalized their reforms combined with the simultaneous trade regimes during the 1980s and 1990s. It change of several globalization measures further suggests that this positive relation- make it particularly difficult for the researcher ship also holds in the cases of India, China, to disentangle the effects of trade liberaliza- and Hong Kong. As noted previously by tion (or other aspects of globalization) from Wood (1999), the experience of developing the effects of other policies. countries that globalized during the 1980s 4. Identification of Trade Policy Effects 13 There is however a large debate on whether these survey data allow for over time comparisons during the 1990s given the changes in the survey questionnaire. See The previous section documents that Deaton and Jean Dreze (2002) for an excellent discussion. many developing countries experienced an mar07_Article2 3/12/07 5:43 PM Page 55

Goldberg and Pavcnik: Distributional Effects of Golbalization 55

increase in inequality as they became more their consumption bundles and their educa- exposed to various dimensions of globaliza- tion endowments, they will be differentially tion. But establishing a causal link between impacted by price changes. For example, globalization and inequality by providing households in the left tail of the welfare dis- credible empirical evidence poses several tribution spend a higher share of their budg- challenges beyond the measurement issues et on basic items, such as food, and are less discussed in section 2. We highlight these educated than richer households. The issues for the case of trade policy below, model, combined with predictions about the however similar concerns exist with respect changes of traded good prices, estimates of to other aspects of globalization wage–price elasticities, and estimates of the Although there are several channels responsiveness of the nontraded good prices through which trade policy can affect to traded good prices, can be used to simu- inequality within a country (we discuss these late the effect of trade policy changes on the channels in detail in the next section), a distribution of household welfare (i.e., common theme in many of the mechanisms household expenditure per capita). discussed in the theoretical literature is that The main advantage of this approach is trade policy affects wage inequality by that it ultimately yields an answer to the changing the relative demand for skilled important question of how trade reform workers. The main empirical challenge is affects the welfare distribution within a how to isolate the effects of trade from other country in a general equilibrium setting that contemporaneous changes in the economic explicitly accounts for intermediate good environment that may have induced shifts in linkages and nontraded goods. However, the the relative demand and supply of skilled predictions of the model depend in a crucial labor. Governments in developing countries way on estimates of parameters that are typ- often implement trade reforms concurrent ically not known: the wage–price elasticities, with other economywide policy changes, the elasticity of nontraded good prices with ranging from labor market reform to indus- respect to traded good prices, and the trial delicensing, tax reforms, and privatiza- degree of pass-through from trade policy tion. Table 1 illustrates the prevalence of changes to product prices. These parameters reforms that may have had confounding are difficult to estimate consistently with effects on wage inequality. time-series data on wages and prices in a set- ting when many other policies change con- 4.1 A General Equilibrium Approach temporaneously with trade. Moreover, if Any study that attempts to address the labor and capital are assumed to be mobile ambitious question of “what is the overall within a country, as is often the case in long- effect of trade liberalization on inequality in run general equilibrium trade models, then a country” thus requires strong modeling the level of industry aggregation needed to and identification assumptions. Porto (2006) empirically implement the general equilibri- is an example of such a study. He examines um approach is very high, implying that the implications of the Argentinean trade there is not enough variation in the data to reform for the distribution of household wel- identify the relationship between trade poli- fare in a general equilibrium model of trade. cy and the variables of interest (prices, In his framework, trade policy influences wages, etc.). For example, in the household welfare by changing the relative Hecksher–Ohlin model, both skilled and prices of goods, which in turn affect labor unskilled labor are assumed to be perfectly income and consumption. Because house- mobile, so that—no matter what the cross- holds in different parts of the prereform wel- sectional pattern of trade protection or liber- fare distribution differ in the composition of alization is—the wages for skilled and mar07_Article2 3/12/07 5:43 PM Page 56

56 Journal of Economic Literature, Vol. XLV (March 2007)

unskilled wages should be equalized across premise is plausible in the short and medi- different sectors in an economy. If this were um run but questionable in the long run. At true, it would eliminate the prospect of any rate, it is important to note that failure of exploiting any cross-sectional variation in this premise to hold in practice does not trade barriers and wages in order to identify invalidate the approach; it simply implies the relationship between trade and wage that one would not find any differential trade changes. policy effects across industries/regions in this case, as wages are equalized across 4.2 Differential Exposure Approach industries/regions. However, studies that An alternative approach to identifying the have exploited industry or regional variation effects of trade liberalization on the wage or in developing countries do find effects, sug- income distribution is taken by several gesting that the assumption of constrained recent studies that have focused on cross- labor mobility is more appropriate in the sectional variation in changes in trade pro- context of developing countries. tection. Such studies examine whether, The main advantage of approaches that within a country, industries or regions that exploit differential time changes in trade were more exposed to trade liberalization protection across cross-sectional units is that experienced smaller or bigger changes in they require much weaker identification wage or income inequality than less-exposed assumptions than the general equilibrium industries/regions (Topalova 2004a; Hanson approach described above, so that the causal forthcoming; Goldberg and Pavcnik 2005; link between trade and inequality is perhaps Wei and Wu 2002). The empirical frame- more convincingly established. On the other work in this line of work usually exploits hand, such approaches can only identify household survey data that include informa- industry- or region-specific deviations from tion on individuals’ industry of occupation, aggregate trends that could, in principle, in wage, region of residence, and various part be due to trade policy. This limitation is demographic characteristics, such as age, discussed in Topalova (2004a), who exploits education, etc., to construct measures of the differential exposure of Indian districts average wages by industry (after controlling to trade liberalization to identify the effects for relevant worker, industry, and job charac- of trade on poverty. Her results indicate that teristics) or measures of wage or income districts that were more exposed to liberal- inequality by region. These measures are ization experienced a relative increase in then related to trade policy changes over the poverty (or, more accurately, a smaller span of a trade liberalization episode to iden- decrease in poverty). However, poverty tify the effect of trade barrier reduction on declines dramatically in India over this peri- inequality. To the extent that the tariff changes od. While her approach can plausibly identi- differ across industries/regions and are exoge- fy the role of trade in explaining nous (or can be instrumented for), the differ- district-specific deviations from this aggre- ential exposure of various industries/regions gate trend, it cannot identify the role of to tariff changes enable the researcher to sep- trade liberalization in explaining the trend arate the effects of trade liberalization from itself. the effects of concurrent policy changes. The An additional limitation of the aforemen- underlying premise of this line of work is that tioned studies is that their usual focus on labor is not perfectly mobile across industries nominal rather than effective tariff rates and/or regions (or at a minimum that certain implies that they ignore intermediate input skills are sector-specific and not easily trans- linkages, so that they are ultimately partial ferable across industries), so that wages are equilibrium in nature. This focus is, however, not equalized across sectors/regions. This not inherent in the nature of the identification mar07_Article2 3/12/07 5:43 PM Page 57

Goldberg and Pavcnik: Distributional Effects of Golbalization 57

approach but rather dictated by data con- negotiated with the WTO, rather than cater straints; effective rates of protection are to special lobby interests; as a result, indus- available for a few, isolated years at best and, tries with initially higher level of protection even then, they tend to be noisy. experienced greater tariff declines. Thus, Fortunately, for the few years for which trade liberalization did not simply lower tar- effective rates are available, the correlation iff levels but also changed the structure of between nominal and effective rates of pro- protection across industries. In fact, studies tection appears to be positive and high, so document that industries with larger pre- that the findings based on nominal tariff reform tariffs experienced larger tariff rates are likely to be robust to using effective changes in Colombia (Goldberg and Pavcnik tariff rates as a measure of protection. At any 2005), Brazil (Pavcnik et al. 2004) and India rate, this latter shortcoming could, in prin- (Topalova 2004a). This pattern suggests that ciple, be addressed with better data that industry lobbies may have had less influence would allow one to compute effective rates on the magnitude of the tariff changes during of protection. the reform period. With these methodological issues in mind, 4.3 The Endogeneity of Trade Policy we now examine the existing evidence on Another challenge facing the literature on various channels through which trade policy trade and inequality is that trade policy is the has affected inequality. outcome of a political process and thus endogenous. While there is a large theoreti- 5. The Relationship between Globalization cal and empirical literature on the determi- and Inequality nants of the protection structure across Globalization affects individuals through industries, empirical work on trade and three main channels: changes in their labor wages has only recently focused on the income, changes in relative prices and endogeneity of trade protection and liberal- hence consumption, and changes in house- ization. The concerns about the endogeneity hold production decisions. Consistent with of trade policy and political economy of pro- the income- or wage-based measurement of tection apply to all studies. For example, in inequality, most research to date has studies focusing on the overall effects of focused on the first channel. The first five trade reform, the political economy of pro- parts of this section are therefore devoted to tection might affect the assumptions on the summarizing the evidence related to the expected price changes subsequent to the effects through the labor income channel. reforms and the consistency of the estimates Since the increase in the skill premium has of wage and cross-price elasticities. been identified as one of the main con- Similarly, studies that exploit cross-industry tributing factors to rising wage inequality, or cross-regional changes in the pattern of we start by reviewing the main explanations protection have to answer the question, is it for the widely documented increase in the valid to treat such changes as exogenous? skill premium (part 5.1). Next we discuss Fortunately, the nature of the tariff reforms other ways through which globalization may in several developing countries, such as have impacted the income distribution: Colombia, Brazil, Mexico, and India, makes transitional unemployment (part 5.2); the usual concern about the endogeneity of changes in industry wages (part 5.3); uncer- trade policy in the context of these countries’ tainty (part 5.4); and potential effects on trade liberalization potentially less severe. labor market standards (part 5.5). The sixth Their governments’ goal in implementing part of this section focuses on the effects of trade reforms was to lower tariff levels across globalization on household production and industries to more uniform levels that were consumption decisions. mar07_Article2 3/12/07 5:43 PM Page 58

58 Journal of Economic Literature, Vol. XLV (March 2007)

5.1 Explanations for the Increase in the labor will specialize in the production of Skill Premium goods that are unskilled-labor intensive. The connection to the income distribution Whatever explanation for the widening is provided by the model’s companion theo- wage gap between skilled and unskilled rem, Stolper–Samuelson, that links changes workers in many developing countries one in product prices to changes in factor adopts, there seems to exist wide agreement returns. A trade-liberalization-induced that the skill premium increase was driven increase in the price of unskilled-labor- by an increase in the demand for skilled intensive products should, according to workers. The main evidence on this issue Stolper–Samuelson, increase the return to comes from studies that have documented the factor that is intensively in the produc- that wages and employment in various skill tion of these products, unskilled labor. In categories have moved in the same direction, contrast, the expected decrease in the price implying that demand shifts dominated of the skilled-labor intensive imported prod- (Robbins 1996; Sanchez-Paramo and Schady ucts should lead to a decline in the wage of 2003). The related arguments are similar to skilled labor. Based on this theorem and the the ones used in the context of the inequali- empirical evidence suggesting that develop- ty debate in developed countries but appear ing countries are richly endowed with even more convincing when applied to unskilled labor,15 one would expect the dis- developing countries, as many of them tributional changes induced by trade liberal- (Latin American countries in particular) did ization in developing countries to favor the not experience the same increase in the sup- unskilled workers. ply of educated workers as the U.S. and East The general equilibrium nature of the Asian economies (Attanasio and Szekely Heckscher–Ohlin model makes it extremely 2000; Sanchez-Paramo and Schady 2003). hard to bring it to the data. Given that the The causes of the increased demand for model’s predictions refer to economywide skilled workers have however been the sub- factor returns, one has only one observation ject of intense debate. per year to work with. In theory, one could 5.1.1 Stolper–Samuelson Effects try to identify Stolper–Samuelson effects by relating trade-policy-induced relative price The most direct link between trade open- changes to factor returns over time but, in ness and changes in the skill premium is pro- practice, this approach is fraught with prob- vided by the best known general equilibrium lems. Price data are often incomplete, model of International Trade, the while the changing mix of goods produced Heckscher–Ohlin model. This model has with different factor proportions within shaped thinking about the distributional statistically defined product categories effects of trade openness in the last makes price comparisons over time less decades, even though the theoretical and empirical shortcomings of the model are 14 widely recognized by now. In its simple 15 The most influential paper on this issue has been the 2 × 2 version, the model predicts that coun- study by Anne O. Krueger et al. (1981) that calculated the tries that are relatively rich in unskilled factor content of trade in manufactures for several devel- oping countries in the 1970s and showed that the export- ing sectors were less skill-intensive than the import-competing sectors. These results have been con- 14 On the theoretical side, the model rests on extreme- firmed in several other studies (see Wood 1999 for an ly restrictive assumptions such as perfect competition, overview). The only caveat is that most of these studies do perfect labor and capital mobility within a country and not cover the last two decades (they typically include data fixed technology. On the empirical side, there has been no up to the mid-1980s) and do not differentiate between support for the predictions of the model, at least not in its higher-skilled and lower-skilled trading partners for each strict version. developing country. mar07_Article2 3/12/07 5:43 PM Page 59

Goldberg and Pavcnik: Distributional Effects of Golbalization 59

informative than one would have hoped. India)), or to the existence of imperfect Furthermore, prices are determined product markets (so that firms respond by endogenously and may change for reasons lowering of profit margins (Mexico, unrelated to trade. For these reasons, a Morocco) and not through labor reallocation direct link between goods and factor prices across sectors). An alternative line of expla- as suggested by general equilibrium trade nation for the lack of labor reallocation is models has been empirically elusive. suggested by recent work by Kaivan Munshi Despite the lack of direct evidence on and Mark R. Rosenzweig (2005), who also Stolper–Samuelson effects, it seems safe to document very low spatial (and social) say that the widening wage gap between mobility in India.16 They argue that the skilled and unskilled workers in the last two social insurance provided by subcaste net- decades in many developing countries that works creates a disincentive to migrate or are presumed to be relatively abundant in out-marry out of fear of losing the services of unskilled labor seems inconsistent with the these networks. Interestingly, the increase in spirit of the Hecksher–Ohlin theory. In prin- inequality (possibly due to the Indian ciple, it is possible to reconcile the evidence reforms) is shown to lower mobility (which on wage inequality with the theory by con- was low to start with) even further, as sub- sidering various extensions of the original castes successfully coped with the conse- model. However, several other patterns doc- quences of rising inequality. In contrast, umented in developing countries seem Gene M. Grossman (1986) and Revenga inconsistent with Hecksher–Ohlin. (1992) find greater employment than wage First, a fundamental prediction of factor sensitivity to trade shocks for the United endowment based trade theories is that the States. These differences in the adjustment adjustment process to trade reforms would mechanisms are consistent with greater labor involve labor reallocations from sectors that mobility in the United States compared to experience price declines, and hence con- the developing economies. tract, toward sectors that experience relative A second piece of evidence that seems price increases and hence expand. However, inconsistent with Stolper–Samuelson effects most studies of trade liberalization in devel- is that empirical work on developing coun- oping countries find little evidence in sup- tries typically finds that the share of skilled port of such reallocation across sectors. The workers has increased substantially within lack of labor reallocation following trade most industries in the last two decades. reform has been documented by Ana L. Within-industry increases in the share of Revenga (1997), Hanson and Harrison skilled workers have been reported for (1999), and Zadia M. Feliciano (2001) for Argentina, Brazil, Mexico, Chile, and Mexico; by Attanasio, Goldberg, and Pavcnik Colombia (Robbins 1996; Sanchez-Paramo (2004) for Colombia; by Janet Currie and and Schady 2003; Attanasio, Goldberg, and Harrison (1997) for Morocco; by Topalova Pavcnik 2004); Hong Kong (Hsieh and Woo (2004a) for India; and by Romain Wacziarg 2005); and India (Kijima 2006). The higher and Jessica Seddon Wallack (2004) in a share of skilled workers in most industries cross-country study of trade liberalization has often been interpreted as evidence in where, however, trade liberalization is cap- tured only through a time dummy. These 16 Further evidence on this issue includes Topalova studies attribute the lack of labor realloca- (2004a), who documents little spatial mobility across dis- tion in response to trade reform to either tricts in India during the 1980s and 1990s, and Daniel rigid labor markets (so that the adjustment Chiquiar (2004), who finds little mobility of individual across Mexican regions in five-year intervals surrounding to trade liberalization occurs through rela- the Mexican trade reforms in the late 1980s and 1990s tive wage adjustments (Colombia, Mexico, (see table 5 in his paper). mar07_Article2 3/12/07 5:43 PM Page 60

60 Journal of Economic Literature, Vol. XLV (March 2007)

favor of skilled-biased technological expectations, it was the unskilled labor- change.17 intensive sectors that were protected the Given these patterns, several explanations most prior to trade reform. This protection for the increase in the skill premium have pattern has been reported for Mexico been suggested in the literature. The first (Hanson and Harrison 1999; Robertson one is to consider simple extensions of the 2000, 2004 for pre-NAFTA period), Morocco model that would reconcile the theory with (Currie and Harrison 1997), and Colombia the evidence. Specifically, the simple (Attanasio, Goldberg, and Pavcnik 2004). Stolper–Samuelson predictions may be The same studies document that it was in overturned if one introduces nontraded fact the unskilled-labor-intensive sectors that goods or additional factors in the model. were impacted the most by tariff cuts. Given Suppose, for example, that there are three this evidence, the increase in the skill premi- factors of production, skilled labor, um is exactly what Stolper–Samuelson unskilled labor, and land (or natural would predict: since trade liberalization was resources, or primary factors). Suppose fur- concentrated in unskilled-labor-intensive ther that some developing countries are rel- sectors, the economywide return to atively abundant in land (as is the case with unskilled labor should decrease—at least in many Latin American countries) and that the period immediately following the land is a complement to skilled labor. Then reforms. In fact, the only study that has to greater trade openness will favor land- our knowledge exploited price data, intensive goods in these countries. If pro- Robertson (2004), documents that relative duction of these goods requires a higher prices and relative wages in Mexico are ratio of skilled to unskilled workers, trade closely related along the lines suggested by openness will benefit skilled workers. A sim- the Stolper–Samuelson theorem (see figures ilar argument can be applied to the role of 4 and 5). This argument demonstrates the nontraded goods. Such arguments rest, advantages of exploiting the sectoral varia- however, on the rather implausible assump- tion in tariff changes as opposed to relying tion that land, or natural resources, require on time variation alone to identify the effects a higher ratio of skilled workers; to our of trade policy changes. Studies that simply knowledge, there has been no evidence sup- use “before–after comparisons” to uncover porting this claim. Moreover, even in that the effects of trade liberalization miss the case, one would expect labor reallocation to important fact that—unlike in textbooks of be the mechanism through which the International Trade—the comparison is not changes in the wage distribution are trans- between autarky and free trade, but rather mitted and, as noted above, the evidence on between protection and “less-protection” so such reallocation has been lacking for devel- that the pattern of protection across sectors oping countries. prior to liberalization is crucial in determining Another line of explanation for the the effects of trade reforms. increase in the skill premium focuses on the Still, this argument is not completely satis- pattern of protection prior to trade liberal- factory as it again implies sectoral labor real- ization in many developing countries and the location—a prediction that no empirical skill intensity of the sectors that were study of trade liberalization in a developing impacted the most by trade reforms. country has found strong support for. Several studies have noted that, contrary to Moreover, the initial pattern of protection that favored unskilled-labor-intensive sec- tors seems a puzzle by itself. Why did coun- 17 This interpretation is not uncontroversial. We dis- cuss it in more detail under “Skill-Biased Technological tries abundant in unskilled labor find it Change” in one of the following subsections. desirable to protect the low-skill-intensive mar07_Article2 3/12/07 5:43 PM Page 61

Goldberg and Pavcnik: Distributional Effects of Golbalization 61

sectors when the pattern of comparative of middle-income countries shifted to goods advantage would have suggested otherwise? of intermediate skill intensity. This shifting One possible answer is that the protection pattern of comparative advantage might patterns reflected political economy consid- explain why many middle-income countries erations that had little to do with compara- found it necessary to protect their low-skill tive advantage. Another is that high tariffs in intensive sectors from imports from low- low-skill-intensive industries, such as textiles income countries. It would also explain why or footwear, were left over from a time in greater openness in these countries would which these sectors were capital and high- not necessarily benefit low-skill workers, as skill intensive and when the protection of the trade barrier reductions in low-skill- these sectors would have been compatible intensive sectors (such as textiles) may lead to with the patterns of comparative advantage. an increase in the imports from China rather Hisahiro Naito (1999) offers an alternative than an increase of domestic production and explanation from a public finance perspec- exports. While this argument is a priori plau- tive: He argues that, contrary to the view of sible, there has been no direct empirical traditional trade theory, tariffs imposed on investigation of its implications to date. A unskilled-labor-intensive products can more disaggregate analysis of imports and Pareto improve welfare in a small open exports of middle-income countries that economy that uses a redistributive nonlinear would differentiate between “more skilled,” income system; this is because the tariffs high-income trading partners and “less change the unskilled/skilled wage ratio, skilled,” low-income trading partners might which in turn reduces the incentive problem shed light into this question in the future. of income redistribution. Since the unskilled A final explanation for the apparent ten- earn more, the tax burden of the skilled can sion between the increase in the skill premi- be reduced; the reduction of the tax burden um and theoretical predictions is that trade has a first order effect on welfare, while the affected the wage distribution through chan- distortion introduced by the tariff is only nels other than the ones suggested by simple second order. This argument applies even to Hecksher–Ohlin theory, or that there were developed countries with flexible income tax other forces at work (some of which may systems but much more so to developing have interacted with, or even been induced countries that have fewer alternative means by, trade openness). A common theme in of redistribution. A trade liberalization then subsequent research on alternative channels sacrifices this redistribution in the hope of through which trade affects inequality is to achieving enough growth to eventually focus on the mechanisms that lead to compensate the less skilled. increased relative demand for more edu- Another possibility is that the recent entry cated labor within industries (as opposed of China and other low-income developing to across industries). We consider these countries (India, Indonesia, Pakistan, etc.) in arguments next. the world markets shifted the existing pat- 5.1.2 The Role of Intermediate Goods and terns of comparative advantage in middle- Outsourcing income countries. This possibility is examined in detail in Wood (1999). Wood Most trade models assume that all trade postulates that, while in the 1960s and 1970s occurs in final goods: this assumption was middle-income countries had a comparative also implicit in the above discussion of the advantage in goods of low-skill intensity, in Hecksher–Ohlin mechanism. However, the 1980s and 1990s, when low-income recent work by Feenstra and Hanson (1996, developing countries started exporting to the 1997, 1999, 2003) has challenged this rest of the world, the comparative advantage assumption and emphasized the growing mar07_Article2 3/12/07 5:43 PM Page 62

62 Journal of Economic Literature, Vol. XLV (March 2007)

importance of trade in intermediate goods, (relatively unskilled-labor-intensive) produc- the so called “outsourcing” or “global pro- tion facilities from Hong Kong to China fol- duction sharing.” They argue that the rapid lowing China’s FDI liberalization in the late expansion in “global production sharing” 1970s. Hong Kong, in turn, specialized in over the past two decades can explain part of skill-intensive manufacturing activities and the observed increase in demand for skilled outsourcing-based services such as market- workers in both developed and developing ing and distribution. countries. The lack of empirical work on other devel- The basic framework in Feenstra and oping countries is partly explained by the Hanson relies on the premise that produc- predominant interest in the role of outsourc- tion of final goods can be split into interme- ing in developed rather than developing diate stages and that intermediate inputs countries. But it may also reflect the fact differ in their skill intensities. Consequently, there are few developing countries that have firms find it optimal to “outsource” some of received as large FDI flows as Mexico or the production stages to cost-minimizing that have outsourced as big a share of their locations abroad. Trade liberalization, cou- production as Hong Kong. However, it is pled with a removal of capital controls, unlikely that outsourcing affects wage opens new opportunities for firms to shift inequality only in Mexico and Hong Kong, the production of some of these intermedi- due to the proximity of these two countries ate goods from developed to developing to the United States and China, respectively. countries. While products shifted to devel- Evidence from Southeast Asia (Keith Head oping countries would be characterized as and John Ries 2002), Central America unskilled-labor-intensive from a developed (Andres Rodriguez-Clare 2001), and country’s perspective, they appear skilled- Eastern Europe (Dalia Marin 2006) suggests labor-intensive when compared with existing that many other developing or transition domestic production activities from the economies engage in production sharing developing country’s point of view. As a with developed economies. In fact, a recent result, “outsourcing” increases the average study by Andzelika Lorentowicz, Marin, and skill intensity of production in both the Alexander Raubold (2005) confirms the developed and developing economies, importance of outsourcing as an explanation inducing an increase in the skill premium in for wage inequality in a transition economy both places. setting. They find that outsourcing activities While descriptive statistics on trade flows of foreign multinationals in Poland are asso- suggest that outsourcing is a potentially ciated with a large increase in the relative important phenomenon in the developing demand for skilled workers in Poland. world (especially in Southeast Asia), the Examining the relationship between out- impact of outsourcing on wage inequality in sourcing and wage inequality in other devel- developing countries has so far been exam- oping countries remains a topic for future ined only for Mexico and Hong Kong. research. Feenstra and Hanson (1997) find strong 5.1.3 Increase in Capital Flows and support for the “global production sharing” Complementarity of Capital with hypothesis for Mexico, where many U.S. Skilled Labor firms export intermediate inputs to maquiladora plants that assemble the inputs A basic premise of the Stolper–Samuelson into final products. Similarly, Hsieh and Woo mechanism is that capital and labor, while (2005) document a large increase in the rel- perfectly mobile within a country, are ative demand for skilled workers in Hong immobile internationally. This premise is Kong after firms reallocated much of the clearly inconsistent with the recent adoption mar07_Article2 3/12/07 5:43 PM Page 63

Goldberg and Pavcnik: Distributional Effects of Golbalization 63

of outward-oriented policies in developing documented empirical phenomenon. How- countries that has been in many cases associ- ever, none of these explanations seems ated with substantial increases in interna- entirely convincing. Edward E. Leamer tional capital flows. If globalization leads to (1998), for example, argues that sector bias an increase in capital inflows into developing and not factor bias determines changes in countries and if the utilization of capital the wage distribution: skilled-biased tech- requires the use of a higher share of skilled nological change that is concentrated in labor, then the increase in capital flows will unskilled-intensive sectors benefits un- be associated with higher demand for skilled skilled workers in the general equilibrium, workers. This argument is put forward in while skilled-biased technological change Cragg and Epelbaum (1996) for pre-NAFTA concentrated in skilled-intensive industries Mexico and Jere R. Behrman, Nancy benefits skilled workers. This argument Birdsall, and Szekely (2000) for several Latin however requires that product prices do not American countries. Both studies focus on change, which is unlikely to be true during a the role of trade reforms in reducing the period of trade reforms. Moreover, the price of capital goods as the mechanism that (admittedly very scant) empirical evidence generates higher demand for both capital does not support this theory; Attanasio, goods and skilled labor. Similar arguments Goldberg, and Pavcnik (2004) do not find are developed in theories of endogenous any statistically robust evidence that skill- technological change (e.g., Daron Acemoglu biased technological change in Colombia was 2003), since new technology is often embod- concentrated in skilled-intensive industries; ied in capital good imports. Since these the- if anything, the (statistically insignificant) ories are most frequently used in the context point estimates of their regressions suggest of (endogenous) skill-biased technological that skilled-biased technological change was change, we discuss them in more detail in concentrated in low-skill sectors, which the next subsection. would have generated a decrease in the skill premium in the general equilibrium. 5.1.4 Skill-Biased Technological Change The past decade witnessed an intense and The main alternative explanation for the lively debate between those who favored increased demand for skilled labor has been the trade-openness-based explanations for skill-biased technological change. “Skill bias” the increase in the skill premium and those is inherently hard to measure and, because who considered skilled-biased technological most of the measures commonly employed change as the primary force behind the doc- in the literature are based on endogenous umented changes in the wage distribution outcome variables (e.g., the share of skilled worldwide. By now it has been recognized workers in a firm’s wage bill) rather than that the most credible explanations for the exogenous technology shocks, there exists no distributional changes witnessed in the past uncontroversial measure of skill-biased tech- few decades would most likely involve inter- nological change. Nevertheless, the repeat- actions of trade openness with skilled-biased ed finding of an increase in both the share of technological change. Along these lines, sev- skilled workers and their relative wage with- eral recent papers have postulated that, even in fairly narrowly defined industry categories though skilled-biased technological change in both developed and developing countries may have played a greater role in increasing has been interpreted as evidence for a the skill premium that particular trade policy worldwide skill bias in new technologies. changes, technological change was itself an As with the evidence on Stolper–Samuelson endogenous response to more “openness” so effects of trade, it is possible to come up that globalization was indirectly responsible with alternative explanations for this well for the increase in the skill premium. mar07_Article2 3/12/07 5:43 PM Page 64

64 Journal of Economic Literature, Vol. XLV (March 2007)

One of the earliest studies in this vein is demand for skilled workers by reducing the Wood (1995), who introduced the term prices of the relevant capital goods and, “defensive innovation” to describe firms’ hence, increasing their imports. From an response to trade openness. According to his empirical point of view, this model has two hypothesis, intensified competition from distinct implications: first, following a trade abroad may induce firms to engage in R&D liberalization episode in a developing coun- or take advantage of existing new technolo- try, total imports for office equipment and gies that they may have had little incentive advanced machinery from developed coun- to adopt prior to liberalization. This theory tries should increase; and second, the is developed further in Mathias Thoenig increase in the demand for skilled workers and Thierry Verdier (2003). While this should be more pronounced in sectors that argument seems more suitable to explain- import more foreign machinery. This sec- ing the increase in inequality in the devel- ond implication is investigated for the peri- oped world, it may be applicable to od surrounding the 1980s Mexican trade middle-income developing countries, such liberalization by Harrison and Hanson as Colombia or Brazil, that underwent sig- (1999), who find that within each Mexican nificant trade reforms in the 1980s and industry, firms that import machinery and 1990s. By that time, low-income developing materials are more likely to employ a high- countries (e.g., China) had entered the er share of white-collar workers than firms world markets, and the import competition that do not import these inputs. Pavcnik middle-income countries faced from the (2003), on the other hand, finds that the new entrants in their low-skill-intensive sec- increased relative plant demand for white- tors may have induced faster technological collar workers by Chilean plants in early change in these sectors. On the empirical 1980s cannot be attributed to the use of side, a common implication of these models imported materials and foreign technical is that in the short- and medium-run, skill- assistance to these plants once one controls biased technological change should be more for time-invariant plant characteristics. pronounced in the sectors that liberalized Marc-Andreas Muendler (2004) reports more. Attanasio, Goldberg, and Pavcnik that the use of imported intermediate prod- (2004) indeed document that, during ucts plays only a minor role for productivity 1984–98, the increase in demand for skilled improvements by Brazilian firms following workers in Colombia was largest in those the trade reform, while Ana M. Fernandes sectors that experienced the largest tariff (forthcoming) notes a positive association cuts. This provides some support for the the- between the use of imported intermediate ory that skilled-biased technological change products and productivity of domestic was itself an endogenous response to trade plants in Colombia. The evidence on the liberalization. role of machinery and office equipment A different mechanism through which imports in transmitting new technology and trade liberalization can induce (or acceler- creating demand for skilled workers is ate) skill-biased technological change is sug- therefore mixed. gested by Acemoglu (2003), who develops a An alternative mechanism through which model of endogenous technological change trade liberalization can affect technological and argues that, in the case of developing change, and thus indirectly inequality, is sug- countries, this technological change may gested by Philippe Aghion, Robin Burgess, take the form of increased imports of Stephen Redding, and Fabrizio Zilibotti machines, office equipment, and other cap- (2005). In their model, firms’ response to ital goods that are complementary to skilled trade liberalization depends on how close labor. Trade liberalization affects the they are to the technology frontier. Firms mar07_Article2 3/12/07 5:43 PM Page 65

Goldberg and Pavcnik: Distributional Effects of Golbalization 65

that are sufficiently close to the frontier can tries in recent years, this is a promising area survive or deter entry of (foreign) competi- for future research. tors by innovating; firms that are far from 5.1.5 Compositional Changes Within the frontier may not be able to fight external Industries: Exporting and “Quality” entry. Hence, the average effect of trade lib- Upgrading of Products, Plants, and eralization will depend on the fraction of Workers firms and sectors that are sufficiently close to the frontier to fight for their survival. In Recent literature has emphasized the addition, Aghion et al. emphasize the role of importance of firm heterogeneity in inter- domestic institutions, labor market restric- national trade (see James R. Tybout (2003) tions in particular, and their interactions for a survey). In particular, studies of the with technology adoption for the distribu- effects of trade reforms on productivity that tional effects of trade policy. In the empirical exploit plant- or firm-level data typically part of their paper, Aghion et al. look at the find major market share reallocations Indian trade liberalization of 1991 for sup- towards more efficient plants (often within port of their theory. Consistent with their the same industry) in the aftermath of liber- theoretical arguments, productivity and alization. This finding seems to contrast profits increased by more in industries that with the documented lack of labor realloca- were close to the Indian productivity fron- tion across industries in response to trade tier and in states that had more flexible labor shocks. One possible explanation for these market institutions. This differential impact seemingly conflicting findings is that the of trade liberalization across industries with documented reallocations are in reality sim- different proximity to the technology fron- ple “revenue-share” reallocations that could tier and states with different regulatory potentially result from changes in firms’ regimes had strong inequalizing effects. market power, rather than factor realloca- These conclusions find less support in tions. This is due to the fact that the plant Topalova (2004b), who documents, using level surveys that are typically employed to firm-level data and detailed information on measure productivity do not contain data on Indian industry tariffs from India, that tariff physical output or inputs, neither on plant- declines were associated with productivity specific prices, so that the above variables improvements in firms with both high and are measured in value terms while the price low productivity prior to the trade reform. indices that are used to deflate them are sec- Overall it seems fair to say that, even tor specific. Another possibility is that due though the premise that trade openness has to the factors discussed earlier on page 59, interacted with skill-biased technological labor market regulation in particular, labor change to increase the demand for skilled is in many developing countries less mobile labor seems both a priori plausible and the- than capital. Finally, it is also possible that oretically well founded, the empirical evi- there is in fact a lot of labor movement dence on the role of particular mechanisms across firms, often within the same industry, through which this increase occurred, is but this movement is not visible at the mixed and inconclusive. Clearly, more evi- aggregation level at which the industry of dence from other developing countries is employment is reported in household sur- needed before one can draw general conclu- veys. In fact, one recent study by John sions. There is also very little empirical work Haltiwanger, Adriana Kugler, Maurice linking skilled-biased technological change Kugler, Alejandro Micco, and Carmen in developing countries to the rise in exports. Pages (2004) finds substantial labor reallo- Given that exchange rate realignments have cation within sectors in several Latin affected exports in many developing coun- American countries. mar07_Article2 3/12/07 5:43 PM Page 66

66 Journal of Economic Literature, Vol. XLV (March 2007)

This latter possibility suggests a reorienta- firms within an industry differ is their tion of empirical analysis away from coun- exporting status. If production for export tries or industries, toward firms or plants, as markets is relatively more skill-intensive the relevant units of observation, a move- than production for developing countries’ ment that parallels recent developments in domestic markets, increased demand for international trade theory. The focus of tra- exports will increase the relative demand ditional trade theory and empirics on sectors for skilled workers within industries and or industries abstracts away from the sub- lead to a higher skill premium. Empirical stantial heterogeneity of products and firms evidence from the United States suggests that are included in statistically defined that exporting is indeed a skill-intensive aggregates. Products that fall into the same activity (see Bernard and Jensen 1997). two- or three-digit SIC category may be pro- Harrison and Hanson (1999) also find that duced with different factor proportions, exporters employ a higher share of white- while individual firms may vary both in collar workers than nonexporting plants in terms of their efficiency or “quality” and in Mexico. terms of the type of workers they employ. Production of higher quality products may Recent work has made this heterogeneity be one reason why exporting firms in devel- the main focus of the analysis by stressing oping countries may require relatively more the importance of compositional changes skilled labor than domestic firms. In addi- within industries in response to trade liberal- tion, “product quality” varies significantly ization, which may induce reallocation of across exporters from different countries. both capital and labor toward “better” firms. Schott (2004) provides strong evidence of The basic idea is that trade openness complete specialization by countries within induces a “quality” upgrading of firms, product categories, with the skill- and capi- where quality can mean either “firm produc- tal-abundant countries specializing in the tivity” or “product quality.” The quality production and export of higher unit value upgrading in response to trade openness can products, and unskilled-labor-abundant itself arise either because firms in import countries specializing in the production and competing sectors try to avoid competition export of low-unit value products. If one from cheaper countries by differentiating accepts his premise that unit values within themselves or because trade can shift very narrowly defined product categories resources from nonexporters to exporters reflect differences in product “quality,” then (see Marc J. Melitz 2003 for a related argu- the implication of Schott’s findings is that ment), and there is ample empirical evi- developed countries specialize in higher dence that exporters tend to be more quality products while developing countries “productive” than nonexporters. Despite the specialize in lower quality products within theoretical appeal and plausibility of these the same product category. While these find- arguments that emphasize firm and plant ings do not directly tell us how countries heterogeneity within an industry, the empir- adjust to trade liberalization, it seems plausi- ical evidence on how this channel affects ble to assume that, as middle-income devel- inequality is still scant and mostly indirect. oping countries become more open to trade, What is essential for establishing a con- they start upgrading their products like the nection between compositional changes more developed countries. If higher quality within an industry and the inequality debate products indeed require a higher share of is that “higher quality” firms have a higher skilled workers, then the shift toward higher demand for skill, so that quality upgrading quality products will benefit skilled workers. leads to an increase in the skill premium. Recent findings by Susan Chun Zhu (2005) For example, one dimension along which are consistent with this idea. She finds that mar07_Article2 3/12/07 5:43 PM Page 67

Goldberg and Pavcnik: Distributional Effects of Golbalization 67

wage inequality has increased by more in The increased demand for “skill”18 within countries and industries that (because of exporting plants could be met either by product cycles) shifted the within-industry attracting new, better-educated workers or composition of exports away from low-skill- by increasing the productivity of the existing intensive exports that were historically asso- workforce. In a subsequent paper, David S. ciated with less developed countries toward Kaplan and Verhoogen (2005) argue that it more sophisticated products that had been was the second mechanism that was at work: exported by richer countries in the past. the higher demand for skill in exporting A somewhat different mechanism involv- plants translated to higher efficiency wages ing upgrading is discussed in a recent in these plants, rather than changes in the paper by Verhoogen (2006). In Verhoogen’s proportions of white-collar and blue-collar study, trade openness leads to an upgrad- workers within each plant. Higher efficiency ing of the average product quality in wages could in turn reflect additional train- exporting plants, which in turn generates ing or effort by the white-collar workers demand for a better qualified workforce. employed in exporting plants. This finding The upgrading of the workforce can, how- implies that there was little labor realloca- ever, be satisfied by upgrading the existing tion across plants in the aftermath of the workforce in each plant, rather than hiring peso depreciation.19 The higher demand for of new, better qualified workers, so that skill was instead satisfied by increasing the trade openness does not necessarily gener- wage premia of the workers already ate labor reallocation across plants. Rather employed in exporting plants. than focusing on trade liberalization, The main challenge of this literature is to Verhoogen exploits a major exchange rate define “quality” in an operational way. As depreciation episode (the 1994 peso crisis Erkan Erdem and Tybout (2004) have in Mexico) to study the response of firms to pointed out, a separation of “firm produc- increased openness. The peso depreciation tivity” and what we typically mean by clearly benefited exporters. Instead of “product quality” is not possible given the focusing on the effects of an increase in available data sets. Moreover, the term aggregate exports on productivity or “quality” is itself elusive from an empirical demand for skilled workers, Verhoogen point of view, especially in the context of a considers the effects of the exchange rate horizontal differentiation model in which depreciation on firms of different produc- consumers value products differently. tivity. More productive firms produce high- er quality products and export; lower 18 We use the term “skill” here in the most general productivity firms produce lower quality sense of the word to include general human capital as products and sell in the domestic market reflected in a worker’s educational attainment: specific only. The basic hypothesis is that the human capital, motivation, and effort. Importantly, this interpretation does not match the white-/blue-collar increase in exports was associated with a worker dichotomy often used in the literature to differen- differential quality upgrading within tiate between skilled and unskilled workers. The mecha- Mexican manufacturing as higher-produc- nism discussed in Verhoogen (2006) demonstrates the limitations of this latter narrow definition in capturing the tivity, exporting plants shifted their within true quality of the workforce. plant product-mix toward higher quality 19 Kaplan and Verhoogen (2005) exploit matched varieties in order to appeal to U.S. con- employer–employee data from the Mexican social securi- ty agency, so that they can follow workers and their wages sumers. But this shift toward higher quali- over time. A potential caveat of their analysis is that the ty products required an upgrading of the data do not contain information on worker education so workforce. As a result, the peso deprecia- that one cannot be sure that within-plant changes of worker wages do not reflect changing returns to a partic- tion induced quality upgrading benefited ular worker characteristics, such a education, during that skilled workers. period. mar07_Article2 3/12/07 8:09 PM Page 68

68 Journal of Economic Literature, Vol. XLV (March 2007)

Schott (2004) tries to circumvent this prob- 5.1.6 Changing Returns to Skill-Intensive lem by implicitly assuming a vertical differ- Occupations entiation model so that higher unit values In some developing countries, the correspond to higher quality. Verhoogen increase in the skill premium has been (2006) uses a set of proxy variables (for linked to the increase in the returns to par- example, a plant’s total sales) or, alterna- ticular occupations that require a higher tively, a latent variable approach to capture level of education. Cragg and Epelbaum “product quality.” However, from the per- (1996) find strong support for this hypothe- spective of the inequality debate, it does sis in the case of pre-NAFTA Mexico, for not matter what definition of “quality” one which they document a rapid increase in the adopts. What matters is the proportion of occupational premia of professionals and skilled and unskilled workers that is administrators (including public administra- required to produce goods before and after tors).20 The authors attribute the increase in a trade liberalization or currency deprecia- these occupational premia to the rapid tion episode. If the demand for skill changes introduced in the economy by increases within firms, this is going to reforms that increased the demand for indi- induce an increase in the skill premium. viduals who could enact these reforms: man- Hence, rather than resorting to particular agers and professionals. The link to interpretations of product quality that may globalization is indirect: trade reforms be controversial, empirical work in this impacted these changing returns to occupa- area could directly examine how within- tion only to the extent that they were part of firm relative demand for skill is affected by the general reforms that generated demand trade liberalization and whether this effect for highly educated individuals. In related is different for firms with initially low ver- work, Kijama (2006) finds that increases in sus high skill-intensity (where “initial” the returns to tertiary degree were especial- refers to the skill-intensity observed prior ly pronounced for individuals in managerial, to the trade reform or exchange rate depre- professional, and technical job in urban ciation episode). India subsequent to 1991 reforms. Studies The second challenge facing this litera- on other countries have however found less ture is that, for the results to be relevant support for rapidly changing returns to for the inequality debate, it is important to skill-intensive occupations.21 have accurate measures of skill. Unfortunately, the information on worker 5.2 Transitional Unemployment and job characteristics provided in firm Perhaps the most commonly expressed level data sets is much more limited than concern regarding globalization in develop- what is usually provided in household sur- ing countries is that trade openness will lead veys, so that researchers have to resort to the familiar dichotomy between produc- tion and nonproduction, or white- and 20 These changing premia to skill-intensive occupa- blue-collar, workers. In the absence of tions account for a significant fraction of the estimated skill premium increase: controlling for occupation com- more detailed information, there is little presses the original estimate of the change in the premi- one can do in the short run. In the longer um of postsecondary to secondary education from 67 run, more information about the character- percent to 40 percent. Similarly, the increase in the pre- mium of postsecondary to primary education drops from istics of workers employed by different 70 percent to 42 percent once occupation is controlled for. firms (or plants) will be essential for estab- 21 In Colombia for example, Attanasio, Goldberg, and lishing a connection between firm hetero- Pavcnik (2004) document that occupational returns remained relatively stable over the 1986–98 period, with geneity and changes in the wage the exception of 1992, for which there was a short-lived distribution. spike in the returns to “managers and other professionals.” mar07_Article2 3/12/07 5:43 PM Page 69

Goldberg and Pavcnik: Distributional Effects of Golbalization 69

to transitional unemployment as the econo- in which industries they seek new employ- my adjusts to new conditions. To the extent ment; even when they do (as is the case in that this unemployment disproportionately the Colombian NHS for example), they affects the poor, it will have important con- report the industry at a very aggregate level sequences for income inequality. (one-digit ISIC). As a result, it is not possible Despite the prominence of this concern in to relate industry unemployment to more the public debate, there is remarkably little disaggregate tariff changes. Furthermore, theoretical or empirical work on its rele- empirical work in this area needs to deal vance. On the theory side, unemployment is with truncation issues, as workers who are absent in the mainstream models of interna- employed in any given survey interval can tional trade, which typically assume full only be assumed to be employed up to the employment. A notable exception is the end of the particular survey interval, and, work of J. Peter Neary (1978, 1982) that similarly, unemployed workers can be explores the consequences of factor speci- assumed to be unemployed only to the ficity in the short run. In Neary’s framework, extent that they have not found a new job it is possible that labor markets are at dise- before the end of that survey period. quilibrium in the short run as the economy An attempt to relate trade liberalization to adjusts to a terms-of-trade shock. This transitional unemployment was undertaken framework seems particularly relevant for by Attanasio, Goldberg, and Pavcnik (2004) developing countries that are often charac- in the context of the Colombian trade liber- terized by severe labor market rigidities (see alization. The authors examine whether the also David Tarr and Steven J. Matusz 1999 increase in the probability of being unem- for a discussion). ployed was greater for workers in the manu- On the empirical side, the lack of evidence facturing sector (where tariff cuts were the on the relationship between trade and tran- largest) than for workers with the same sitional unemployment is mainly due to the observable characteristics in nontraded- absence of appropriate data. Aggregate sta- good sectors (such as wholesale and retail tistics on total unemployment by year seem trade, restaurants, hotels, construction, etc.) to suggest that macroeconomic recessions in urban Colombia. They find that increases have a larger impact on unemployment than in the probability of unemployment before tariff reductions, but inferences based on and after tariff reductions were not larger in macroeconomic trends can be misleading, as manufacturing than in nontraded sectors. they do not indicate which industries and However, this evidence is based on a very which population groups are most affected, aggregate industry definition, while the what the causes of unemployment and information on unemployment is not direct- chances of reemployment are, and how long ly linked to changes in trade policy. the duration of unemployment spells are. Moreover, no attempt is made to link Such information is important for relating changes in probability of unemployment to unemployment to measures of well-being inequality. and inequality. The link between trade poli- 5.3 Industry Wages cy, unemployment, and inequality could be better identified by relating detailed indus- Among those who are and remain try tariff changes to changes in industry employed, our discussion so far has focused unemployment. The difficulty in pursuing exclusively on the impact of trade openness such an approach stems from the fact that on changes in the economywide skill premi- household surveys in developing countries um. We now turn our attention to other typically do not report in which industries ways in which globalization may have affect- the currently unemployed used to work and ed wage and income inequality. The first mar07_Article2 3/12/07 5:43 PM Page 70

70 Journal of Economic Literature, Vol. XLV (March 2007)

one is through changes in industry wage Moreover, industry wage premia may be premia. affected in cases where unions share in Industry wage premia are the part of industry rents through employment security worker wages that cannot be explained by guarantees rather than wages, and where observable worker characteristics, such as employment security is obtained through gender, age, education, experience, etc., but higher trade protection (Grossman 1984). can be attributed to workers’ industry affili- Finally, trade policy could affect industry ation. While studies have found that indus- wage premia via industry-level productivity try wage premia account for a significant changes. Several recent empirical studies portion of individual wage variation, there is have found that trade liberalization was asso- less agreement as to whether these premia ciated with productivity improvements in reflect compensating differentials, efficien- developing countries.23 If these improve- cy wages, industry rents, or returns to ments are passed on to workers in the form industry-specific skills. of higher wages, trade could increase wage There are several plausible channels premia in the sectors that experienced high- through which trade policy changes may er productivity gains due to their higher affect industry wage premia. In short- and exposure to trade liberalization. A related medium-run models of trade where workers argument is presented in the two studies by cannot easily move across sectors, tariff cuts Verhoogen (2006) and Kaplan and translate into proportional declines in the Verhoogen (2005) we examined earlier, wage premia of those industries that experi- although the (efficiency) wage increases in ence larger than average tariff declines.22 their mechanism are not generated by trade This possibility is particularly important in liberalization but rather a peso-crisis developing countries characterized by labor induced increase in exports destined for the market rigidities (James J. Heckman and U.S. market. Pages 2000). These rigidities may be irrele- The empirical evidence on the response vant in practice because of the existence of of industry wage premia to trade reforms is informal labor markets and the vast non- mixed: no association between tariff reduc- compliance with labor market regulation. tions and industry wage premia (Feliciano However, the lack of labor reallocation 2001 for Mexico; Pavcnik, Andreas Blom, across sectors in the aftermath of dramatic Goldberg, and Schady 2004 for Brazil), pos- tariff declines in several countries that we itive association (Goldberg and Pavcnik discussed earlier supports the premise of 2004 for Colombia), and negative associa- rigid labor markets. A further channel tion (Mishra and Kumar 2005 for urban through which trade may affect industry India).24 Feliciano (2001) reports a positive wages is suggested by models of imperfect association between declines in import competition and union bargaining. If prof- licenses and industry wage premia. The itable industries share part of their rents with workers because of union bargaining 23 power, tariff cuts in these industries may See Harrison (1994) for Cote d’Ivoire; Pravin Krishna and Devashish Mitra (1998), Aghion, Burgess, lead to lower wages as the industry rents Redding, and Zilibotti (2005), and Topalova (2004b) for stemming from protection disappear. India; Euysung Kim (2000) for Korea; Pavcnik for Chile (2002); Fernandes for Colombia (forthcoming); and Muendler (2004) and Hay (2001) for Brazil. 22 If such industries had lower wage premia in the pre- 24 Studies that rely on average firm or industry wages reform period, then such changes will further increase the rather than industry wage premia also report mixed wage dispersion, making those who received lower rela- results: Currie and Harrison (1997) find no association tive wages to start with even worse off. This turns out to between changes in industry wages and tariffs in be in fact the case with the trade reforms in Mexico in the Morocco; Revenga (1997) on the other hand reports a 1980s and Colombia in the 1990s. positive association for Mexico. mar07_Article2 3/12/07 5:43 PM Page 71

Goldberg and Pavcnik: Distributional Effects of Golbalization 71

heterogeneity of findings in the above stud- premia account for about 2 percent of ies is perhaps not surprising given the large explained variation in log hourly wages con- number of possible channels through which ditional on workers’ observable characteris- trade could affect industry wage premia. tics in this country. Thus, while changes in Kaplan and Verhoogen (2005) present evi- industry wages contribute to the increase in dence based on panel data that the wage wage inequality, it seems unlikely that the increases in plants with higher productivity change in industry wage premia is a first Verhoogen documented in his earlier work order effect. One potential explanation for are due to an increase in the efficiency the relatively small magnitude of industry wages of the workers employed in these wage responses and the simultaneous lack of plants, rather than higher wages of new labor reallocation across sectors is the exis- hires. tence of an informal sector in many develop- These findings have potentially important ing countries. This sector offers an implications for the effects of trade openness additional margin through which firms can on wage inequality. Interestingly, both in adjust to trade shocks. We investigate this Colombia and Mexico, studies that have doc- explanation in section 5.5. umented a decline in industry wages in 5.4 Uncertainty response to trade liberalization also find that the sectors that experienced the largest tariff A body of research has examined the idea cuts had the highest shares of less educated that globalization not only affects income workers and the lowest wages to start with levels but also exposes workers to increased (see, for example, Attanasio, Goldberg, and economic uncertainty through less secure Pavcnik 2004). As a result of trade liberaliza- employment and more volatile income. tion, the initially low wages declined even Conceptually, most empirical work in this lit- further increasing wage inequality. In this erature relies on a simple labor demand and sense, one could argue that less educated supply framework with a stochastic labor workers were hit twice: not only did the skill demand (see Kenneth Scheve and Matthew premium increase during that period but the Slaughter 2002 for an in depth review of this wages in industries that employed a propor- literature). In this setting, trade reform tionately higher share of unskilled workers might increase wage uncertainty in two declined relative to the average wage in the ways. economy. Kaplan and Verhoogen’s results go First, trade liberalization can lead to in the same direction: even though wages greater price volatility and productivity increase in absolute terms in the aftermath shocks (as in Dani Rodrik 1997, 1998), of the peso crisis, the wages of white-collar which in turn generates greater volatility in workers employed in high productivity wages and employment. Scheve and plants increase by more, thus contributing to Slaughter (2002) convincingly argue that an increase in wage inequality. empirical studies do not reach a consensus While these effects go in the direction of on whether trade liberalization increases increasing wage inequality, their magnitude price variation. This state of affairs is per- is estimated to be small, and so it is ques- haps not surprising. While trade liberaliza- tionable whether they are the primary force tion exposes domestic consumers and behind increases in wage inequality. In producers to the volatility of world prices, at Colombia, for example, the estimates sug- the same time the exposure to foreign mar- gest that the average tariff reduction in man- kets mitigates the effects of potentially large ufacturing sector of 37 percentage points domestic shocks on prices. Theoretical work would be associated with 4 percent decline by John McLaren and Andrew Newman in industry wage premium. Industry wage (2002) makes a similar point, suggesting that mar07_Article2 3/12/07 5:43 PM Page 72

72 Journal of Economic Literature, Vol. XLV (March 2007)

the relationship between globalization and demand elasticity in Turkey. Rana Hasan, risk is ambiguous. Mitra, and K. V. Ramaswamy (forthcoming) Second, Rodrik (1997) argues that trade find that labor demand becomes more elas- reforms may increase wage uncertainty not tic following the 1991 Indian trade reforms only through potentially greater demand and that more protected industries have volatility but also by increasing the (absolute lower labor demand elasticities. Their study value of) the own-price elasticity of labor is particularly interesting because it also demand. The reasoning is as follows. For a examines the differential effects of trade given vertical shift of the labor demand reform on labor demand elasticities in sec- curve (arising from productivity or product tors with differential tariffs located in states demand shocks), a more elastic labor with different labor market regulation. The demand implies greater variation in wages study finds that labor demand elasticities are and employment. In this case, greater open- greater in Indian states with more flexible ness increases the uncertainty faced by indi- labor laws and that trade reforms increased viduals only indirectly, not by exposing them labor demand elasticities by greater degree to greater demand volatility but by magnify- in states with more labor market flexibility. ing the effects that any given demand shock To our knowledge, Tom Krebs, Krishna, and will have on their wages and employment. William Maloney (2005) is the only study One way in which trade reform can that directly examines the link between increase the elasticity of labor demand is by trade reform and income variability. Using intensifying product market competition longitudinal income data on workers before (and thus increasing the elasticity of product and after the Mexican trade reforms in the demand from which the labor demand is 1980s and 1990s, they find that tariff derived). In fact, James Levinsohn (1993), declines are associated with increased Harrison (1994), and Currie and Harrison income uncertainty. (1997) find empirical support for increased To the extent that globalization increases product market competition following trade income uncertainty, risk averse individuals reforms. They show that domestic firms low- might be worse off even if trade reform does ered their markups following the trade not affect or increases their expected reforms in Turkey, Ivory Coast, and incomes. That said, the question still Morocco, respectively. Alternatively, trade remains open whether and how increased liberalization may make labor demand more uncertainty affects inequality. To the extent elastic by providing firms with increased that increases in uncertainty and/or risk access to substitutes for domestic labor such aversion vary across individuals of different as imported intermediate products. Our dis- education and/or ages, globalization induced cussion in section 5.1.2 emphasizes the rapid uncertainty could add to greater inequality expansion of trade in intermediate goods. across individuals. Yet, we are not aware of Unfortunately, the empirical work that any study that links liberalization-induced links trade reforms to wage uncertainty is increases in uncertainty to inequality. scarce, especially in the context of develop- A different but related point is that a more ing countries. Most studies examine the link uncertain product demand may induce firms between trade and wage uncertainty indi- to adapt hiring practices that increase a rectly by studying the relationship between firms’ flexibility to hire/fire workers in trade reform and labor demand elasticity. response to changing product demand. For The results of these studies are mixed. example, a firm that operates in a more vari- Krishna, Mitra, and Sajjid Chinoy (2001) able product market may find it beneficial to find no evidence that trade liberalization rely more heavily on informal or temporary increased (the absolute value of) labor labor to maintain flexibility. This, in turn, mar07_Article2 3/12/07 5:43 PM Page 73

Goldberg and Pavcnik: Distributional Effects of Golbalization 73

could lead to greater wage variability. We are work arrangements offered by this sector; to not aware of any empirical work that exam- the extent that this is true, the observed dif- ines how greater product demand uncertain- ferences in pay between formal and informal ty affects firms’ choice of workers and jobs may be partly driven by selection of contract types and, ultimately, inequality. individuals based on unobservable tastes or However this issue is partly related to a characteristics. broader line of work that examines the The usual argument that trade liberaliza- effects of globalization on the use of infor- tion will increase informality is that foreign mal labor and compliance with labor market competition forces firms to cut costs, which standards. We examine this issue next. they in part do by employing a higher pro- portion of informal workers. Goldberg and 5.5 Labor Market Standards Pavcnik (2003) present a model that formal- Many globalization opponents have izes this idea and show that, under certain argued that globalization may have adverse theoretical assumption, firms within an effects on inequality in the broader sense by industry may find it optimal to hire relative- inducing noncompliance of firms with labor ly more informal workers after a permanent market standards and by increasing the pro- decline in industry tariffs. To the extent that portion of workers in the informal sector of jobs in the informal sector are associated the economy. The informal sector is general- with relatively lower pay and worse working ly defined as the sector of the economy that conditions, the relative expansion of the does not comply with labor market regula- informal sector following a trade liberaliza- tions, such as minimum wage or minimum tion episode could contribute to growing working age laws, and it is associated in the inequality, especially since the informal sec- public’s mind with lower pay and worse tor tends to employ a higher proportion of working conditions. It accounts for a sizable less-educated workers. Alternatively, it is share of the labor market in developing possible that the expansion of the informal countries: for example, 50 to 60 percent of sector in the aftermath of trade liberalization the labor force in urban Colombia is reflects the entry of new firms into the employed in the informal sector during the market in response to new opportunities 1980s and 1990s. created by the reforms. Such firms are like- The claim that the informal sector offers ly to start small and informal, especially in worse working conditions is controversial. countries with rigid labor markets, and shift On one hand, several studies (Douglas into the formal sector only later if they are Marcouiller, Veronica Ruiz de Castilla, and successful. Christopher Woodruff 1997; Goldberg and Evidence on the link between informality Pavcnik 2003; Pavcnik et al. 2004) document and trade reform is scarce due to the lack of that workers with otherwise comparable data on informality and labor market regula- observable characteristics are paid lower tion compliance of firms. The few studies on wages in the informal sectors of Peru, El these issues yield mixed results. Currie and Salvador, Brazil, and Colombia; moreover, Harrison (1997) find that firms in Morocco workers employed in the informal sector are started hiring more temporary workers considerably less likely to receive nonwage (who are not entitled to benefits) in the benefits, and in household survey question- period following the trade reform. naires they express less satisfaction with Goldberg and Pavcnik (2003) focus on their working environment and job quality. Colombia and Brazil, two countries that On the other hand, some individuals may experienced expansions of their already choose to work in the informal sector large informal sectors in the years following because they value the greater flexibility in the trade reforms, and examine whether mar07_Article2 3/12/07 5:43 PM Page 74

74 Journal of Economic Literature, Vol. XLV (March 2007)

trade liberalization can explain the docu- plants subcontract to domestic establish- mented increase in informality. Interestingly, ments that may pay below the minimum most of the observed increase in the share of wage. Relying on information on wages of informal workers in the total labor force individual workers, Goldberg and Pavcnik occurred through within-industry increases, (forthcoming) find no association between rather than through shifts in employment the likelihood of industry compliance with across industries with different informality minimum wage laws and industry tariff intensity. However, the association between reductions in Colombia. within-industry tariff changes and probabili- Overall, existing studies provide little evi- ty of employment in the informal sector dence that trade liberalization or FDI con- varies across countries and time and seems tribute to growing inequality by expanding to be related more to the flexibility of the the size of the informal sector and inducing labor market than to trade policies. In Brazil noncompliance with minimum wage laws. (a country with a relatively flexible labor However, more work is needed in this area. market according to Heckman and Pages For example, the results in Goldberg and (2000)), industry tariff declines were not Pavcnik (2003) suggest that the relationship associated with changes in the probability of between trade reform and informality employment in the informal sector. In depends on the institutional setting in which Colombia (a country with more rigid labor trade reforms take place. To investigate this market institutions), industry tariff cuts were possibility further, one would ideally exploit associated with increased probability of the heterogeneity of labor market institu- informal employment, but only in the peri- tions over time and across administrative od prior to the implementation of labor areas within a country such as India, where market reforms that substantially increased labor market regulation varies across states. the flexibility of labor markets. In addition, a disadvantage of the within- Related work has examined firms’ compli- country analysis presented above is that it ance with minimum wage laws. This aspect cannot by its nature shed light on the empir- of the labor market regulation is particularly ical relevance of “race-to-the-bottom” argu- relevant in the globalization and inequality ments; that is, arguments that suggest that debate, because minimum wages affect pri- firms that have the option of relocating their marily workers at the bottom of wage distri- plants will choose, everything else being bution. Harrison and Jason Scorse (2004a, equal, the country with the lowest labor 2004b) study differences in compliance with standards. This induces competition among Indonesian minimum wage legislation across countries for footloose firms and leads to exporters, foreign-owned firms, and domes- further degradation of labor standards. tic firms using Indonesian surveys of manu- Bernard and Jensen (2003) and Bernard and facturers from the 1990s. In their study, they Fredrik Sjoholm (2003) present some evi- consider a firm to be compliant with the leg- dence for the United States and Indonesia islation if the average wage of production that suggests a higher propensity of multi- workers in the plant exceeds the minimum nationals to relocate; after accounting for the wage. They find that foreign owned plants fact that foreign affiliates are typically larger are actually more likely to have production and more productive than domestic firms, wages above the minimum wage. The use of foreign affiliates are more likely to shut down average wages as a measure of compliance than domestic firms in the host country. may also conceal individual instances of Determining whether these shut-down deci- wages below the legislated minimum wage. sions (and the set up of new operations else- Unfortunately, data constraints preclude where) are driven by labor market regulation them from examining whether foreign-owned remains a topic for future research. mar07_Article2 3/12/07 5:43 PM Page 75

Goldberg and Pavcnik: Distributional Effects of Golbalization 75

5.6 Household Production and Abstracting from household production and Consumption consumption may be defendable when one studies the consequences of manufacturing Our discussion so far has focused on the tariff declines on urban households in a mid- labor market effects of globalization. Two dle income country such as Mexico or additional channels through which globaliza- Colombia. However, it is substantially more tion may affect inequality are household pro- problematic to ignore these channels in poor duction and consumption. These channels economies such as India, Indonesia, or are particularly relevant in poorer develop- Vietnam, especially in rural areas and in ing countries, yet they have received little cases when trade liberalization affected the attention in the literature, perhaps because agricultural sector. the main focus of the globalization related The only study that has included house- research to date has been middle-income hold production in studying the relationship developing countries. between trade reforms and inequality is to Many individuals in poor countries are not our knowledge Topalova (2004a), who employed in the formal labor market for derives measures of inequality and poverty wages, but instead work in their household based on household expenditure data. As business or family farm and devote a sub- mentioned earlier, her results suggest that stantial amount of time to production of poverty declined less in districts that liberal- goods/services used for own consumption ized more, but the findings regarding inequal- (Rosenzweig 1988). For example, in ity are less clear-cut: the point estimates in Vietnam in 1993, about 19 percent of adults most of her specifications suggest that bigger age 20–64 report working for wages, while tariff cuts were also associated with bigger 90 percent of adults report working within increases in inequality within a district, their own household (Eric V. Edmonds and but these findings are never statistically Pavcnik 2006). Similarly, in Indonesia, less significant. than 30 percent (45 percent) of rural (urban) Other work has investigated how house- men and less than 12 percent (20 percent) of holds allocate their time between formal wage rural (urban) women worked in wage work markets and within household work, but has before the Indonesian crisis (James P. Smith not explicitly examined the relationship et al. 2002). In India, 46 percent of the labor between this allocation decision, globalization force (rural and urban) works for wages and inequality. The general lesson from this (Suresh D. Tendulkar 2003). work is that adjustment of household produc- The main reason for the limited amount of tion is an important way through which fami- empirical work on within household produc- lies in poor countries respond to economic tion and consumption is data constraints. shocks. Smith et al. (2002) and Elizabeth Specifically, many surveys focus only on the Frankenberg, Smith, and Duncan Thomas formal labor market and thus exclude the (2003) show that Indonesian families coped self-employed. To the extent that the self- with the 1998 crisis by increasing their within- employed are surveyed, measures of profits household production. Along the same lines, or net earnings associated with their busi- Edmonds and Pavcnik (2006) find that house- nesses are often missing or, to the extent that holds allocated time away from household they are available, they tend to be noisy. production toward wage work following the Moreover, because labor market surveys do rice market liberalization in Vietnam. not contain information on household Determining how these adjustments affect expenditures or consumption, the implicit inequality remains a topic for future research. value of products produced by households for Household consumption is equally impor- their own consumption cannot be captured. tant as a channel through which globalization mar07_Article2 3/12/07 5:43 PM Page 76

76 Journal of Economic Literature, Vol. XLV (March 2007)

may impact inequality. Most international the budget of households in the bottom tail trade models assume that individuals have of the welfare distribution. Second, his identical and homothetic preferences. In model also implies that changes in the prices these models, trade-policy-induced changes of traded goods lowered (through general in relative prices of goods change the con- equilibrium effects) the prices of nontraded sumption of individuals with different goods such as health, education, and leisure incomes in proportional terms; as a result, goods, which are consumed in greater pro- trade does not affect people’s relative posi- portion by the rich. Consequently, the con- tion in the welfare distribution through the sumption channel implied an increase in consumption channel. However, a large lit- inequality in the case of Argentina’s entry erature in development economics has into Mercosur. Although these findings are shown that poorer households devote a dis- subject to the same caveats discussed earlier proportionately large share of their house- in section 4, Porto’s study nicely illustrates hold expenditures to basic items such as the importance of the consumption channel. food. To the extent that household con- Furthermore, the pattern of predicted price sumption depends on the relative position changes serves as a reminder that it is impos- of households in the welfare distribution, sible to make general statements about the globalization-induced price changes may impact of trade liberalization on inequality, affect inequality through consumption. as the effects depend crucially on the Furthermore, the increased availability and specifics of the reform in question, in partic- lower prices of traded goods may shift ular the structure of tariff changes across demand away from nontradable services industries. (e.g., household services, such as housekeep- Interestingly, at the end of the study Porto ing, cooking, etc.) toward tradable goods concludes that the impact of the (washing machines, dryers, microwaves, Argentinean trade reforms on inequality via etc.), further depressing the earning the consumption channel was substantially prospects of the poor. smaller in magnitude than its impact The consumption channel has been through the labor income channel. Porto largely ignored in empirical work for the attributes the difference in the magnitude of reasons discussed in detail in section 2.2 of the two effects to the underlying assumption the paper. Porto (2006) is the only study of perfect factor mobility and the associated that explicitly considers how trade policy magnification theorem that states that affects the welfare distribution through changes in relative goods prices generate consumption.25 As we discussed in section more than proportional changes in factor 4, he examines the implications of the prices. Based on Porto’s results, it is tempt- Argentinean trade reform for the distribu- ing to conclude that the usual neglect of the tion of household welfare using a general consumption channel in the globalization equilibrium framework. Porto’s analysis and inequality debate may not be a first yields two interesting insights. First, his order concern. However, more work needs model implies that the structure of the to be done to establish whether his findings Mercosur-induced tariff cuts translated into generalize. increases in the prices of relatively low-skill- labor-intensive goods such as food and bev- 6. Conclusions erages. These goods have a larger share in

The substantial amount of evidence we 25 To the extent that consumption responses to trade reform differ across districts in India, this channel is also reviewed in this article suggests a contempo- captured by Topalova (2004a). raneous increase in globalization and mar07_Article2 3/12/07 5:43 PM Page 77

Goldberg and Pavcnik: Distributional Effects of Golbalization 77

inequality in most developing countries. point to constrained labor mobility as one However, establishing a causal link between plausible explanation for the lack of sectoral these two trends has proven more challeng- reallocation. Indeed, the strict labor market ing. Despite the ambiguities involved in regulation that many developing countries identifying the relationship between open- had in place prior to the recent reforms is a ness and distributional changes, it seems fair potential source of labor market rigidities. to say that the evidence has provided little The importance of these rigidities is likely to support for the conventional wisdom that diminish in the long run, especially since trade openness in developing countries many developing countries have by now sig- would favor the less fortunate (at least in nificantly liberalized their labor markets. relative terms). Still, from an empirical point of view, the dis- Our survey has identified several channels tinction between short- and long-run has that may explain why the recent experience always been elusive. We have surprisingly lit- of developing countries did not conform to tle knowledge as to how long it takes an the “naive” thinking about globalization. Our economy to adjust to external shocks, and understanding of the consequences of glob- what time frames we should use in practice alization for inequality has improved as the when we consider the short- versus long-run conceptual framework used in empirical effects of particular policies. work expanded to include trade in interme- The lack of sectoral reallocation could also diate products, international flows of capital, reflect that most of the adjustment to trade trade-induced skilled biased technological reform occurs within industries, but at a change, short-run factor immobility, and firm level of detail that cannot be detected in the heterogeneity. household or firm level surveys usually used Overall, there is little support for the in this line of work. Our survey highlights premise that adjustment to changing eco- several globalization-based explanations for nomic conditions would occur through labor the increased relative demand for more edu- reallocation from declining to growing sec- cated workers within industries. In some tors of the economy, at least at the aggregate cases, trade reforms liberalized in addition industry level usually considered in tradi- to goods flows, factor flows, most important- tional international trade models of compar- ly capital, that may have generated addition- ative advantage. A common finding of al demand for skilled workers. In other studies of the effects of trade reforms in instances, globalization affected not only developing countries is the lack (or small mag- trade in final goods, but also and foremost nitude) of sectoral labor reallocation (although trade in intermediate goods that from the it is possible that there is reallocation across developing country perspective were skill- firms within sectors that is not visible at the intensive. Even in those cases where liberal- relatively high level of aggregation used in ization was concentrated on final goods, the labor market surveys).26 In some instances, highest trade barrier reductions were often the data also suggest that the wage response concentrated—contrary to conventional to trade barrier reduction is more pro- wisdom—on low-skill sectors that had origi- nounced than the employment response. nally enjoyed a higher level of protection. While these findings are subject to many Technological change that favored skilled caveats—the high level of industry aggrega- workers may have interacted with trade tion being the perhaps most important reforms to further depress the demand for one—the cumulative evidence seems to low-skilled workers. Increased exposure to currency fluctuations boosted exports from 26 Recent evidence on constrained spatial mobility in developing countries in some cases and pro- developing countries is also in line with these findings. vided incentives to upgrade the product-mix mar07_Article2 3/12/07 5:43 PM Page 78

78 Journal of Economic Literature, Vol. XLV (March 2007)

of their domestic plants. These composition- As the nature of globalization keeps al changes may have fostered a quality changing, the channels through which the upgrading of plants that further contributed distribution of resources within countries is to the widening of the wage gap between affected changes too, and so does the skilled and unskilled. research agenda investigating the relation- Overall, it appears that the particular ship between globalization and inequality. In mechanisms through which globalization recent years, it has become increasingly affected inequality are country, time, and apparent that trade is more than the flow of case specific; that the effects of trade liberal- goods between countries as traditionally ization need to be examined in conjunction modeled in international trade theory. Trade with other concurrent policy reforms; and represents exchange between firms that are that implementation details of particular located in different countries. As traditional policies matter. This conclusion may seem cross-border restrictions are disappearing, disappointing, as it offers no simple predic- the focus of the analysis is shifting from the tions regarding the distributional impact of country to the firm, as the relevant unit of globalization and, hence, no straightforward observation. Accordingly, questions, such as recipe for remedial measures to alleviate what type of firms produce what goods and potentially adverse impacts. Yet, it is hardly for which markets, which firms export and surprising given the heterogeneity of coun- which ones produce for the domestic mar- tries, reforms, and overall globalization ket, what are the characteristics of workers experience within the developing world. employed by different types of firms, etc., Finally, we should emphasize that most of are becoming more prominent in the litera- the existing evidence refers to narrow meas- ture. Mechanisms that emphasize composi- ures of inequality, such as the skill premium tional effects of globalization, quality or wage inequality. Broader concepts of upgrading in response to intensified import inequality that focus on consumption and competition from lower-income countries or, general well-being have received substantial- alternatively, to higher export demand by ly less attention. The very scant evidence that more developed economies, and reallocation exists on these issues however seems to sug- of resources across firms or plants within a gest that the labor market effects of globaliza- sector, or even across products of different tion dominate its effects on consumption quality within a firm, seem more relevant to through relative price changes, so perhaps the developing countries these days. The main focus on wages alone is not as limiting as one challenge facing the empirical literature in would have thought. Along the same lines, we this area is that the heterogeneity of firms, know surprisingly little about one of the most plants, products, and workers emphasized in frequently voiced concerns regarding global- the theoretical arguments implies the need ization: its potential to generate transitional for highly disaggregate data. Such data are unemployment that might disproportionately typically available for plants and contain fair- affect less skilled individuals. It would be ly detailed information on many plant char- tempting to characterize these open ques- acteristics, including occasionally their tions as areas of future research, but the truth product lines. However, what is missing is that the same factors that have inhibited from such data sets is information on the research on these topics in the past (lack of characteristics of the workers employed by appropriate data being the primary one) are each plant/firm, which is the crucial step likely to do the same in the future. The most needed for establishing a connection to dis- pressing research priority in this regard is tributional questions. Hence, we do not the collection of additional data and the know for sure whether plants that are more improvement of existing collection methods. productive employ better educated workers; mar07_Article2 3/12/07 5:43 PM Page 79

Goldberg and Pavcnik: Distributional Effects of Golbalization 79

or whether the production of higher quality Journal of Development Economics, 59(1): 103–23. Branstetter, Lee, and Nicholas Lardy. 2006. “China’s products requires a more skilled labor force; Embrace of Globalization.” NBER Working Papers, or whether changes in the product mix or no. 12373. product quality are accompanied by changes Chiquiar, Daniel. 2004. “Globalization, Regional Wage Differentials, and the Stolper–Samuelson Theorem: in the characteristics and compensation of Evidence from Mexico.” Bank of Mexico Working the workforce. It is these kinds of questions Paper, no. 2004-06. that future research will need to address in Cragg, Michael Ian, and Mario Epelbaum. 1996. “Why Has Wage Dispersion Grown in Mexico? Is It the order to provide insight into how ongoing Incidence of Reforms or the Growing Demand for globalization will impact inequality in Skills?” Journal of Development Economics, 51(1): developing countries. 99–116. Currie, Janet, and Ann E. Harrison. 1997. “Sharing the REFERENCES Costs: The Impact of Trade Reform on Capital and Acemoglu, Daron. 2003. “Patterns of Skill Premia.” Labor in Morocco.” Journal of Labor Economics, Review of Economic Studies, 70(2): 199–230. 15(3): S44–71. Aghion, Philippe, Robin Burgess, Stephen Redding, Deaton, Angus. 1997. The Analysis of Household and Fabrizio Zilibotti. 2005. “Entry Liberalization Surveys: A Microeconometric Approach to and Inequality in Industrial Performance.” Journal of Development Policy. Baltimore and London: Johns the European Economic Association, 3(2–3): Hopkins University Press. 291–302. Deaton, Angus. 2003. “Measuring Poverty in a Anderson, James E., and Eric van Wincoop. 2004. Growing World (or Measuring Growth in a Poor “Trade Costs.” Journal of Economic Literature, World).” NBER Working Papers, no. 9822. 42(3): 691–751. Deaton, Angus and Jean Dreze. 2002. “Poverty and Attanasio, Orazio, Pinelopi Koujianou Goldberg, and Inequality in India: A Reexamination.” Economic Nina Pavcnik. 2004. “Trade Reforms and Wage and Political Weekly, 3729–48. Inequality in Colombia.” Journal of Development Dornbusch, Rudiger, and Sebastian Edwards. 1994. Economics, 74(2): 331–66. “Exchange Rate Policy and Trade Strategy.” In The Attanasio, Orazio, and Miguel Székely. 2000. Chilean Economy: Policy Lessons and Challenges, “Household Saving in East Asia and Latin America: ed. B. P. Bosworth, R. Dornbusch, and R. Laban. Inequality Demographics and All That.” In Annual Washington, D.C.: Brookings Institution, 81–104, World Bank Conference on Development Economics 112–15. 2000, ed. B. Pleskovic and N. Stern. Washington, Edmonds, Eric V., and Nina Pavcnik. 2006. “Trade D.C.: World Bank. Liberalization and the Allocation of Labor between Banerjee, Abhijit, and Thomas Piketty. 2004. “Are the Households and Markets in a Poor Country.” Journal Rich Growing Richer? Evidence from Tax Data.” In of International Economics, 69(2): 272–95. Data and Dogma: The Great Indian Poverty Debate, Erdem, Erkan, and James R. Tybout. 2004. “Trade ed. A. Deaton and V. Kozel. Houndmills: Macmillan Policy and Industrial Sector Responses in the Press. Developing World: Interpreting the Evidence.” In Banerjee, Abhijit, and Thomas Piketty. 2005. “Top Brookings Trade Forum 2003, ed. S. M. Collins and Indian Incomes, 1922–2000.” World Bank Economic D. Rodrik. Washington, D.C.: Brookings Institution Review, 19(1): 1–20. Press, 1–43. Behrman, Jere R., Nancy Birdsall, and Miguel Székely. Feenstra, Robert C., and Gordon H. Hanson. 1996. 2000. “Economic Reform and Wage Differentials in “Foreign Investment, Outsourcing, and Relative Latin America.” IADB Working Paper, no. 435. Wages.” In The Political Economy of Trade Policy: Berman, Eli, John Bound, and Stephen Machin. 1997. Papers in Honor of Jagdish Bhagwati, ed. R. C. “Implications of Skill-Biased Technological Change: Feenstra, G. M. Grossman, and D. A. Irwin. International Evidence.” NBER Working Papers, Cambridge and London: MIT Press, 89–127. no. 6166. Feenstra, Robert C., and Gordon H. Hanson. 1997. Bernard, Andrew B., and J. Bradford Jensen. 1997. “Foreign Direct Investment and Relative Wages: “Exporters, Skill Upgrading, and the Wage Gap.” Evidence from Mexico’s Maquiladoras.” Journal of Journal of International Economics, 42(1–2): 3–31. International Economics, 42(3–4): 371–93. Bernard, Andrew B., and J. Bradford Jensen. 2003. Feenstra, Robert C., and Gordon H. Hanson. 1999. “Firm Structure, Multinationals, and Manufacturing “The Impact of Outsourcing and High-Technology Plant Deaths.” Unpublished. Capital on Wages: Estimates for the United States, Bernard, Andrew B., J. Bradford Jensen, and Peter K. 1979–1990.” Quarterly Journal of Economics, Schott. 2005. “Trade Costs, Firms, and Productivity.” 114(3): 907–40. Unpublished. Feenstra, Robert C., and Gordon H. Hanson. 2003. Bernard, Andrew B., and Fredrik Sjoholm. 2003. “Global Production Sharing and Rising Inequality: A “Foreign Owners and Plant Survival.” NBER Survey of Trade and Wages.” In Handbook of Working Papers, no. 10039. International Trade, vol. 1, ed. E. K. Choi and J. Beyer, Harald, Patricio Rojas, and Rodrigo Vergara. Harrigan. Malden, Mass: Blackwell, 146–85. 1999. “Trade Liberalization and Wage Inequality.” Feliciano, Zadia M. 2001. “Workers and Trade mar07_Article2 3/12/07 5:43 PM Page 80

80 Journal of Economic Literature, Vol. XLV (March 2007)

Liberalization: The Impact of Trade Reforms in Americas: Ftaa and Beyond, ed. A. Estevadeordal et Mexico on Wages and Employment.” Industrial and al. Cambridge: Harvard University David Rockefeller Labor Relations Review, 55(1): 95–115. Center for Latin American Studies, 505–37. Fernandes, Ana M. Forthcoming. “Trade Policy, Trade Hanson, Gordon H. Forthcoming. “Globalization, Volumes and Plant-Level Productivity in Colombian Labor Income, and Poverty in Mexico.” In Manufacturing Industries.” Journal of International Globalizaton and Poverty, ed. A. Harrison. Chicago: Economics. Chicago University Press; Washington, D.C.: Ferreira, Francisco H. G., and Julie A. Litchfield. 1999. Brookings Institution Press. “Calm after the Storms: Income Distribution and Hanson, Gordon H., and Ann E. Harrison. 1999. Welfare in Chile, 1987–94.” World Bank Economic “Trade Liberalization and Wage Inequality in Review, 13(3): 509–38. Mexico.” Industrial and Labor Relations Review, Frankenberg, Elizabeth, James P. Smith, and Duncan 52(2): 271–88. Thomas. 2003. “Economic Shocks, Wealth, and Harrison, Ann E. 1994. “Productivity, Imperfect Welfare.” Journal of Human Resources, 38(2): Competition and Trade Reform: Theory and 280–321. Evidence.” Journal of International Economics, Galiani, Sebastian, and Pablo Sanguinetti. 2003. “The 36(1–2): 53–73. Impact of Trade Liberalization on Wage Inequality: Harrison, Ann E., and Gordon H. Hanson. 1999. “Who Evidence from Argentina.” Journal of Development Gains from Trade Reform? Some Remaining Economics, 72(2): 497–513. Puzzles.” Journal of Development Economics, 59(1): Gasparini, Leonardo. 2003. “Different Lives: Inequality 125–54. in Latin America and the Caribbean.” Unpublished. Harrison, Ann E., and Jason Scorse. 2004a. “Moving up Gasparini, Leonardo. 2004. “Argentina’s Distributional or Moving Out? Anti-Sweatshop Activists and Labor Failure.” Unpublished. Market Outcomes.” NBER Working Papers, no. Goldberg, Pinelopi Koujianou, and Nina Pavcnik. 10492. 2003. “The Response of the Informal Sector to Trade Harrison, Ann E., and Jason Scorse. 2004b. “The Liberalization.” Journal of Development Economics, Impact of Globalization on Compliance with Labor 72(2): 463–96. Standards: A Plant-Level Study.” In Brookings Trade Goldberg, Pinelopi Koujianou, and Nina Pavcnik. Forum 2003, ed. S. M. Collins and D. Rodrik. 2004. “Trade, Inequality, and Poverty: What Do We Washington, D.C.: Brookings Institution Press, Know? Evidence from Recent Trade Liberalization 45–96. Episodes in Developing Countries.” In Brookings Hasan, Rana, Devashish Mitra, and K. V. Ramaswamy. Trade Forum 2004, ed. S. M. Collins and C. Graham. Forthcoming. “Trade Reforms, Labor Regulations, Washington, D.C.: Brookings Institution Press, and Labor-Demand Elasticities: Empirical Evidence 223–69. from India.” Review of Economics and Statistics. Goldberg, Pinelopi Koujianou, and Nina Pavcnik. Hay, Donald A. 2001. “The Post-1990 Brazilian Trade 2005. “Trade, Wages, and the Political Economy of Liberalisation and the Performance of Large Trade Protection: Evidence from the Colombian Manufacturing Firms: Productivity, Market Share Trade Reforms.” Journal of International Economics, and Profits.” Economic Journal, 111(473): 620–41. 66(1): 75–105. Head, Keith, and John Ries. 2002. “Offshore Goldberg, Pinelopi Koujianou, and Nina Pavcnik. Production and Skill Upgrading by Japanese Forthcoming. “The Effects of the Colombian Trade Manufacturing Firms.” Journal of International Liberalization on Urban Poverty.” In Globalizaton Economics, 58(1): 81–105. and Poverty, ed. A. Harrison. Chicago: Chicago Heckman, James J., and Carmen Pages. 2000. “The University Press; Washington, D.C.: Brookings Cost of Job Security Regulation: Evidence from Institution Press. Latin American Labor Markets.” NBER Working Green, Francis, Andy Dickerson, and Jorge Saba Papers, no. 7773. Arbache. 2001. “A Picture of Wage Inequality and Hsieh, Chang-Tai, and Keong T. Woo. 2005. “The the Allocation of Labor through a Period of Trade Impact of Outsourcing to China on Hong Kong’s Liberalization: The Case of Brazil.” World Labor Market.” American Economic Review, 95(5): Development, 29(11): 1923–39. 1673–87. Grossman, Gene M. 1984. “International Competition Kaplan, David S., and Eric A. Verhoogen. 2005. and the Unionized Sector.” Canadian Journal of “Exporting and Individual Wage Premia: Evidence Economics, 17(3): 541–56. from Mexican Employer–Employee Data.” Unpub- Grossman, Gene M. 1986. “Imports as a Cause of lished. Injury: The Case of the U.S. Steel Industry.” Journal Keller, Wolfgang, and Stephen R. Yeaple. 2003. of International Economics, 20(3–4): 201–23. “Multinational Enterprises, International Trade, and Haltiwanger, John, Adriana Kugler, Maurice Kugler, Productivity Growth: Firm-Level Evidence from the Alejandro Micco, and Carmen Pages. 2004. “Effects United States.” NBER Working Papers, no. 9504. of Tariffs and Real Exchange Rates on Job Kijima, Yoko. 2006. “Why Did Wage Inequality Reallocation: Evidence from Latin America.” Increase? Evidence from Urban India 1983–99.” Journal of Policy Reform, 7(4): 191–208. Journal of Development Economics, 81(1): 97–117. Hanson, Gordon H. 2004. “What Has Happened to Kim, Euysung. 2000. “Trade Liberalization and Wages in Mexico since NAFTA? Implications for Productivity Growth in Korean Manufacturing Hemispheric Free Trade.” In Integrating the Industries: Price Protection, Market Power, and mar07_Article2 3/12/07 5:43 PM Page 81

Goldberg and Pavcnik: Distributional Effects of Golbalization 81

Scale Efficiency.” Journal of Development System and Its Implication for Production Economics, 62(1): 55–83. Efficiency.” Journal of Public Economics, 71(2): Krebs, Tom, Pravin Krishna, and William Maloney. 165–88. 2005. “Trade Policy, Income Risk, and Welfare.” Neary, J. Peter. 1978. “Short-Run Capital Specificity Unpublished. and the Pure Theory of International Trade.” Krishna, Pravin, and Devashish Mitra. 1998. “Trade Economic Journal, 88(351): 488–510. Liberalization, Market Discipline and Productivity Neary, J. Peter. 1982. “Intersectoral Capital Mobility, Growth: New Evidence from India.” Journal of Wage Stickiness, and the Case for Adjustment Development Economics, 56(2): 447–62. Assistance.” In Import Competition and Response, Krishna, Pravin, Devashish Mitra, and Sajjid Chinoy. ed. J. N. Bhagwati. Chicago: University of Chicago 2001. “Trade Liberalization and Labor Demand Press, 39–67. Elasticities: Evidence from Turkey.” Journal of Nicita, Alessandro. 2004. “Who Benefited from Trade International Economics, 55(2): 391–409. Liberalization in Mexico? Measuring the Effects on Krueger, Anne O., Hal B. Lary, Terry Monson, and Household Welfare.” World Bank Policy Research Narongchai Akrasanee, eds. 1981. Trade and Working Paper Series, no. 3265. Employment in Developing Countries. Vol. 1. Pavcnik, Nina. 2002. “Trade Liberalization, Exit, and Individual Studies. Chicago: University of Chicago Productivity Improvement: Evidence from Chilean Press. Plants.” Review of Economic Studies, 69(1): 245–76. Leamer, Edward E. 1998. “In Search of Pavcnik, Nina. 2003. “What Explains Skill Upgrading Stolper–Samuelson Effects on U.S. Wages.” In in Less Developed Countries?” Journal of Exports, Imports and the American Worker, ed. S. Development Economics, 71(2): 311–28. M. Collins. Washington, D.C.: Brookings Institution Pavcnik, Nina, Andreas Blom, Pinelopi Koujianou Press, 141–214. Goldberg, and Norbert Schady. 2004. “Trade Levinsohn, James. 1993. “Testing the Imports-as- Liberalization and Industry Wage Structure: Market-Discipline Hypothesis.” Journal of Inter- Evidence from Brazil.” World Bank Economic national Economics, 35(1–2): 1–22. Review, 18(3): 319–44. Lorentowicz, Andzelika, Dalia Marin, and Alexander Porto, Guido G. 2006. “Using Survey Data to Assess Raubold. 2005. “Is Human Capital Losing from the Distributional Effects of Trade Policy.” Journal Outsourcing? Evidence for Austria and Poland.” of International Economics, 70(1): 140–60. University of Munich Discussion Paper, no. 2005-22. Revenga, Ana L. 1997. “Employment and Wage Machin, Stephen, Annette Ryan, and John Van Effects of Trade Liberalization: The Case of Mexican Reenen. 1996. “Technology and Changes in Skill Manufacturing.” Journal of Labor Economics, 15(3): Structure: Evidence from an International Panel of S20–43. Industries.” IFS Working Papers, no. W96/06. Revenga, Ana L. 1992. “Exporting Jobs? The Impact of Marcouiller, Douglas, Veronica Ruiz de Castilla, and Import Competition on Employment and Wages in Christopher Woodruff. 1997. “Formal Measures of U.S. Manufacturing.” Quarterly Journal of the Informal-Sector Wage Gap in Mexico, El Economics, 107(1): 255–84. Salvador, and Peru.” Economic Development and Robbins, Donald J. 1996. “Evidence on Trade and Cultural Change, 45(2): 367–92. Wages in the Developing World.” OECD Technical Marin, Dalia. 2006. “A New International Division of Paper, no. 119. Labor in Eastern Europe: Outsourcing and Robertson, Raymond. 2000. “Trade Liberalisation and Offshoring to Eastern Europe.” Journal of the Wage Inequality: Lessons from the Mexican European Economic Association, 4(2–3): 612–22. Experience.” World Economy, 23(6): 827–49. McLaren, John, and Andrew Newman. 2002. Robertson, Raymond. 2004. “Relative Prices and Wage “Globalization and Insecurity.” Unpublished. Inequality: Evidence from Mexico.” Journal of Melitz, Marc J. 2003. “The Impact of Trade on Intra- International Economics, 64(2): 387–409. Industry Reallocations and Aggregate Industry Rodriguez-Clare, Andres. 2001. “Costa Rica’s Productivity.” Econometrica, 71(6): 1695–1725. Development Strategy Based on Human Capital and Mishra, Prachi, and Utsav Kumar. 2005. “Trade Technology: How It Got There, the Impact of Intel, Liberalization and Wage Inequality: Evidence from and Lessons for Other Countries.” Journal of Human India.” IMF Working Papers, no. 05/20. Development, 2(2): 311–24. Mistiaen, Johan, and Martin Ravallion. 2003. “Survey Rodrik, Dani. 1997. Has Globalization Gone Too Far? Compliance and the Distribution of Income.” World Washington, D.C.: Institute for International Bank Policy Research Working Paper Series, no. Economics. 2956. Rodrik, Dani. 1998. “Why Do More Open Economies Muendler, Marc-Andreas. 2004. “Trade, Technology, Have Bigger Governments?” Journal of Political and Productivity: A Study of Brazilian Economy, 106(5): 997–1032. Manufacturers, 1986–1998.” Unpublished. Rosenzweig, Mark R. 1988. “Labor Markets in Low- Munshi, Kaivan, and Mark R. Rosenzweig. 2005. Income Countries.” In Handbook of Development “Why Is Mobility in India so Low? Social Economics, vol. 1, ed. H. Chenery and T. N. Srinivasan. Insurance, Inequality, and Growth.” CID Working San Diego and Amsterdam: Elsevier Science. Paper, no. 121. Sanchez-Paramo, Carolina, and Norbert Schady. 2003. Naito, Hisahiro. 1999. “Re-examination of Uniform “Off and Running? Technology, Trade, and the Commodity Taxes under a Non-linear Income Tax Rising Demand for Skilled Workers in Latin mar07_Article2 3/12/07 5:43 PM Page 82

82 Journal of Economic Literature, Vol. XLV (March 2007)

America.” World Bank Policy Research Working Papers, no. 04/28. Paper Series, no. 3015. Tybout, James R. 2003. “Plant- and Firm-Level Scheve, Kenneth, and Matthew Slaughter. 2002. Evidence on “New” Trade Theories.” In Handbook “Economic Insecurity and the Globalization of of International Trade, vol. 1, ed. E. K. Choi and J. Production.” NBER Working Papers, no. 9339. Harrigan. Malden, Mass: Blackwell, 388–415. Schott, Peter K. 2004. “Across-Product versus within- Verhoogen, Eric A. 2006. “Trade, Quality Upgrading Product Specialization in International Trade.” and Wage Inequality in the Mexican Manufacturing Quarterly Journal of Economics, 119(2): 647–78. Sector.” Unpublished. Smith, James P., Duncan Thomas, Elizabeth Wacziarg, Romain, and Jessica Seddon Wallack. 2004. Frankenberg, Kathleen Beegle, and Graciela Teruel. “Trade Liberalization and Intersectoral Labor 2002. “Wages, Employment and Economic Shocks: Movements.” Journal of International Economics, Evidence from Indonesia.” Journal of Population 64(2): 411–39. Economics, 15(1): 161–93. Wei, Shang-Jin, and Yi Wu. 2002. “Globalization and Székely, Miguel, and Marianne Hilgert. 1999. “What’s Inequality without Differences in Data Definition, Behind the Inequality We Measure: An Investigation Legal System, and other Institutions.” Unpublished. Using Latin-American Data.” IADB Working Paper, Winters, L. Alan, Neil McCulloch, and Andrew McKay. no. 409. 2004. “Trade Liberalization and Poverty: The Tarr, David, and Steven J. Matusz. 1999. “Adjusting to Evidence So Far.” Journal of Economic Literature, Trade Policy Reform.” World Bank Policy Research 42(1): 72–115. Working Paper Series, no. 2142. Wood, Adrian. 1995. “How Trade Hurt Unskilled Tendulkar, Suresh D. 2003. “Organised Labour Market Workers.” Journal of Economic Perspectives, 9(3): in India Pre and Post Reform.” Unpublished. 57–80. Thoenig, Mathias, and Thierry Verdier. 2003. “A Wood, Adrian. 1999. “Openness and Wage Inequality in Theory of Defensive Skill-Biased Innovation and Developing Countries: The Latin American Challenge Globalization.” American Economic Review, 93(3): to East Asian Conventional Wisdom.” In Market 709–28. Integration, Regionalism and the Global Economy, ed. Topalova, Petia. 2004a. “Factor Immobility and R. Baldwin et al. Cambridge; New York and Regional Impacts of Trade Liberalization: Evidence Melbourne: Cambridge University Press, 153–81. on Poverty and Inequality from India.” Unpublished. Zhu, Susan Chun. 2005. “Can Product Cycles Explain Topalova, Petia. 2004b. “Trade Liberalization and Firm Skill Upgrading?” Journal of International Productivity: The Case of India.” IMF Working Economics, 66(1): 131–55.