Trade Liberalization, Productivity, Employment and Wages: A Difference-in- Difference Approach

Transcription

1 Trade Liberalization, Productivity, Employment and Wages: A Difference-in- Difference Approach Preliminary version August 2010 Abstract Adriana Peluffo* This paper analyzes the effects of increased competition resulting from the creation of the Southern Common Market (MERCOSUR) on productivity, employment and wages for the Uruguayan manufacturing sector at the plant level. We use impact evaluation techniques -which are not common in trade reform studies- namely matching and difference-in-differences estimation for the period The definition of an appropriate control group is a big challenge and it may be improved by the use of matching procedures. We define a set of tradable and non-tradable industries according to the level and stability of the import penetration and export orientation ratio of the industries. We further explore on the behavior of plants belonging to the export oriented and import competing sectors. One of the most robust findings is that increased trade liberalization seems to improve total factor productivity with a greater effect on plants belonging to the import competing sector. Furthermore, we find reductions in employment driven mainly by the decrease in blue collars, increases in wages and a reduction in the wage gap between white and blue collars as a result of increased trade exposure. Thus, the increase in productivity along with the unemployment of unskilled workers would indicate a room for training, labour and social policies in order to countervail the negative impact of trade liberalization on less qualified workers. JEL Classification: F13, O12, J2, J3 Keywords: trade policy, productivity, employment, wages. *Instituto de Economía, FCEA, Universidad de la República, Uruguay apeluffo@iecon.cceee.edu.uy, adriana.peluffo@gmail.com Acknowledgements: I am indebted to Prof. Germán Calfat and Prof. Ariel Barraud for their help and support. All remaining errors are on my own. 1

2 1. Introduction Recent models of international trade with firm heterogeneity (e.g. Melitz, 2003; Bernard et al, 2003; Bernard, Redding and Schott, 2007; Melitz and Ottaviano, 2008) predict than a movement from autarky to free trade leads to an increase in productivity and size led by expansion in high productivity firms, along with job creation and job destruction and increases in real wages. We analyze the effect of increased trade openness on firms productivity, employment and wages using matching and difference-in-difference techniques (MDID) which compare the effect of the increase in trade exposure on Uruguayan manufacturing plants performance before and after the creation of the Southern Common Market (MERCOSUR). The difference-in-difference approach has the advantage of removing the effects of common shocks providing so a more accurate analysis of the impact of trade openness. A contribution of this paper is the use of the difference-in-difference approach which is not common in trade empirical works 1 to analyse the impact of trade liberalization at the micro level for a small developing country -in particular the impact of MERCOSUR s creation-. To the best of our knowledge this is the first work to use this methodology for the MERCOSUR and in particular for Uruguay using micro level data to analyze the impact of increased openness on firms productivity, employment and wages. The impact of trade liberalization on economic performance and welfare gains is an important but controversial issue. Even though most of the cross-country empirical evidence finds that more open economies experience a faster growth, 2 some economists are sceptical to the robustness of this result (e.g. Rodriguez and Rodrik, 2000). Moreover, these studies do not identify the specific mechanisms by which trade openness may affect growth (Edwards, 1993; Hallak and Levinsohn, 2004). Traditional trade models predict that a movement towards free trade would increase efficiency by reallocating resources from comparative disadvantage industries to comparative advantage ones. These models assume representative firms, perfect competition and full employment. By assuming full employment the effect on labour markets is reduced to adjustment costs, though it is recognised that some workers can be better off and others worse off after trade liberalization. Gains in efficiency and long run growth are assumed to countervail the adverse effect on those who lose. In the neoclassical model a movement towards freer trade would lead to factor reallocation towards comparative 1 The difference-in-difference approach is commonly used in labour economics. As far as we know only 4 papers out of more than 95 use difference-in-difference methodology (Slaughter, 2001; Giavazzi and Tabellini, 2005; Girma, Greenaway and Kneller, 2008; Trefler, 2004) to analyse the effect of trade policies. 2 See for instance Dollar (1992), Sachs and Warner (1995), Edwards (1998), and Frankel and Romer (1999). For recent surveys see Lewer and Van den Berg (2003) and Lopez (2005). 2

3 advantage sectors, which are those intensive in relative abundant factors. In this model, the Stolper- Samuelson Theorem which links changes of price of production factors to changes in the prices of the goods produced states that opening to trade will raise the price of the abundant factor. Thus if developing economies are abundant in unskilled labour when opening to trade the relative wage of unskilled labour should rise, improving so income distribution. Up to the mid of the 90s trade models assume representative firms and usually perfect or monopolistic competition. Melitz (2003) was the first to elaborate a theoretical model introducing explicitly firm heterogeneity. He presents a model with one factor and one sector of production and constant markups, and shows that in the presence of firm s heterogeneity in productivity trade openness leads to significant within-industry reallocation effects from less to more productive firms. In Melitz s model exposure to trade generate within-industry reallocation effects which increases the average productivity and average size of firms while reduces its number. The new new models 3 of trade which incorporate firm heterogeneity, predict that trade liberalization could generate significant across and within-industry reallocation effects. In these models opening to trade and consequently increased trade exposure may not only generate the traditional resource reallocation effects from comparative disadvantage industries to comparative advantage ones, but also from less to more productive firms within industries. On the other hand, the new new trade models that introduce firm heterogeneity predict that trade reform will trigger job creation and job destruction in all sectors, as both net-exporting and net-importing sectors will be characterized by expanding high productivity firms while low-productivity firms shrink or close down. This implies that an important reshuffling of jobs takes place within sectors. Moreover, these new models predict an increase in real wages driven by the expansion of more productive firms. With respect to wage inequalities they have changed substantially over the past few decades, in developing as well as in developed countries. In some cases these changes are in line with the predictions of Heckscher-Ohlin theory: widening wages or unemployment gaps between skilled and unskilled workers in the North, and symmetrically narrowing gaps in parts of the South, particularly in East Asia in the 1960s and 1970s (Wood, 1994). In other cases the wage changes have diverged from these predictions. In particular in the South, wage inequalities rose in many countries in the 1980s and 1990s, most notably in middle-income countries (Robbins, 1996; Wood, 1998), but also in some low-income countries (UNCTAD, 1997). Several explanations have been put forth to explain these findings, some of them emphasise other forces than globalization (reform of labour markets, institutions or exogenous technical change), others suggesting alternative channels through which the effects of globalization might flow. 3 Recently, the examination of the new microeconomic evidence points out that exporting firms are more productive than non-exporting ones, and that increased exposure to international markets may increase productivity. This stylized fact gives raise to new models that incorporates firms heterogeneity. 3

4 Though the efficiency argument for trade liberalisation is usually accepted the main argument against trade reform in developing countries is that trade liberalization would increase income inequality and worsen the conditions of the poor. In particular concerns regarding higher unemployment among workers displaced by the contraction of import competing sectors, greater uncertainty and precariousness of job conditions, and the creation of new job opportunities only for the most qualified segments of the workforce. In this paper we examine the impact of increased exposure to trade on firms productivity, employment and wages. 4 We use a difference-in-difference methodology, also named treatment effect approach. We classify industries into two groups: tradable industries as those that received the treatment (which experienced an increased exposure to international markets) and non-tradable industries or control group (those that were not affected by trade liberalization). 5 We use matching propensity score methods (nearest neighbour and kernel techniques) and then difference-in differences estimates. Matching should improve the selection of the control group allowing a better insight on the effects of trade exposure on plants behaviour. Data come from two data sources. Data at the plant level was provided by the National Institute of Statistics (Instituto Nacional de Estadisticas del Uruguay, INE) for the period Data at the industry level on imports and gross product comes from UNIDO dataset (Nicita and Olarreaga, 2007). This last set of data was used to classify industries as tradable and non-tradable as explained below. This work structures as follows: after this introduction, we present briefly some theoretical issues on the links between trade liberalization, productivity, employment and wages. In the third section we present some empirical evidence. In the fourth section we present the empirical implementation and in the fifth the results and finally our preliminary conclusions. 2. Links between trade liberalization, productivity, employment and wages 2.1. Trade liberalization and productivity The allocative efficiency argument for free trade has been extensively debated by traditional trade theory in the context of perfectly competitive markets. These models predict that trade liberalization will induce the reallocation of resources across sectors from comparative disadvantage industries to comparative advantage ones. 4 In other related work we analysed the impact of trade on plants s mark-ups and sales. 5 This group is less affected by increased trade openness. 4

5 Since the late 1970s the new trade theory has shown that the gains from trade originating from specialisation according to comparative advantages are only part of the story, since in the presence of imperfectly competitive markets trade liberalisation can bring additional gains by reducing the dead weight losses created by domestic firm s market power. In particular it has been argued that trade liberalisation by increasing competition, forces firm s to lower cost-margins and hence move down their average cost curves, therefore raising firm size and scale efficiency. The New Trade Theory, based on models of monopolistic competition assumes the existence of a continuum of identical firms, each one producing a different variety of a product. Since firms have symmetric technologies, opening to trade induces all firms to export to all countries. Head and Ries (1999) use the Helpman and Krugman (1985) model of monopolistic competition to show that a tariff reduction leads to a decrease in the number of plants but has no effect on the output per plant. Thus, opening to trade would reduce the number of plants but would not affect the average size. Since the mid of the 90s several papers show that exporting firms are more productive than nonexporting ones. Further, they are bigger, pay higher wages and are more capital intensive. The studies by Bernard and Jensen (1995, 1999); Aw et al. (2000); Isgut (2001); Alvarez and Lopez (2005); Clerides et al. (1998) are some studies of this empirical literature. This new empirical evidence gives raise to theoretical models which introduce firm heterogeneity. These models show that in the presence of within industry firm heterogeneity, trade liberalisation causes more productive firms to expand at the expense of less efficient firms (which either shrink or die), thereby inducing additional efficiency gains also named rationalization effect. Melitz (2003) was the first widely cited work to introduce firm heterogeneity. In his model there is a continuum of firms, each one producing a different variety of a good. Preferences are of Dixit-Stiglitz type with a constant elasticity of substitution and fixed mark-ups. Labour is the only factor of production and productivity is randomly assigned to firms. Firms have to make an initial investment before they enter to the market and then they learn their productivity level and decide whether to produce or exit. Since not all firms can afford the entry cost to the export markets, only the more productive firms export. In addition, more firms, the ones with higher productivity draws, start producing the good. The expansion of exporters and the entry of new firms increase the demand for labour, which bids up the real wage and forces the least productive to exit. This reallocation of resources increases average productivity and average output per firm. Melitz and Ottaviano (2008) extend the Melitz s model by allowing for firms heterogeneity as well as endogenous mark-ups across firms. This model predicts that opening to trade reduces mark-ups, highlighting the potential pro-competitive effects of trade liberalization. In this model opening to trade 5

6 increases productivity and firms size while reduces its number. Further, aggregate productivity and average mark-ups respond to both the size of a market and the extent of its integration through trade (larger, more integrated markets exhibit higher productivity and lower mark-ups). Bernard, Redding and Schott (2007) introduce an additional industry and factor of production into Melitz s model with constant mark-ups. In this model comparative advantage plays an important role in determining the pattern of trade. This model can explain why some countries export more in certain industries, while at the same intra-industry trade exists. A movement to free trade increases firms productivity and size while decrease the number of firms. Average firm output increases more in comparative advantage industries while at the same time the largest decrease in the number of firms occurs in these industries. The selection on productivity drives the entry and exit process and it is more intense in comparative advantage industries because of the greater probability of export opportunities, therefore low productivity firms are less likely to survive trade openness in comparative advantage industries. Bernard et al. (2003) introduce a different type of trade model with heterogeneous firms. They assume stochastic firm productivity in the multi-country Ricardian model developed by Eaton and Kortum (2002). Domestic and foreign firms compete to produce the same variety, which generates an endogenous distribution of mark-ups. The model and the counterfactual experiment they develop predict that a decrease in trade barriers increases average productivity, reduce the number of firms but do not affect mark-ups. Aggregate productivity rises as employment shifts from low productivity plants driven out by import competition to high productivity plants turning towards export markets. The simulations also show that a 5 percent decrease in trade barriers decreases the number of firms by 3.3 percent while employment in the industry falls by 1.3 percent. Since employment falls less than the number of firms average employment per firm goes up. The degree of uncertainty with respect to technologies does not seem to affect the main predictions of these models. While Melitz (2003) assumes that technologies are randomly assigned to firms, Yeaple (2005) assumes that firms can choose between two technologies, a low-cost and a high-cost technology. These technologies use workers with different skills, which are completely observed by the firms. In Yeaple s model there are two industries: a differentiated manufactured good and a homogenous good. A reduction in trade costs increases average productivity and reduces the number of varieties produced in each country. Since each variety is produced by a single firm, the number of firm decreases, and sales concentration increases, increasing so firms size. Moreover, trade and investment liberalisation may promote technological progress and productivity growth in developing countries through several channels, such as technological progress embodied in 6

7 imported capital goods and intermediate inputs, technology transfers accompanying FDI and learning by exporting effects. As we stated above though the efficiency argument for trade liberalisation is usually accepted the main argument against trade reform in developing countries is that trade liberalisation would increase income inequality and worsen the conditions of the poor. In particular concerns regarding higher unemployment among workers displaced by the contraction of import competing sectors, greater uncertainty and precariousness of job conditions, and the creation of new job opportunities only for the most qualified segments of the workforce. We turn to this point in the section below Trade liberalization and its links with employment and wages Trade policies can have a significant impact on the level and structure of employment, on wages and wage differentials, and on labour market institutions and policies. Nevertheless labour and social policies also influence the outcomes of trade policies in terms of growth of output, employment and the distribution of income. Trade liberalization is associated with both job destruction and job creation. The net employment effect in the short run depends mainly on country specific factors such as the functioning of the labour market. In the long run, the efficiency gains due to trade liberalization are expected to generate positive employment effects, either in terms of quantity or quality of jobs or a combination of both. The theoretical literature provides insights into the process of job destruction and job creation following trade liberalization and illustrates how different country characteristics can affect temporary and permanent employment at the sectoral or country level (Lee and Vivarelli, 2006). The classical link between trade and income inequality is based on the Stolper-Samuelson Theorem developed in a model that assumed full employment. According to this theorem inequality is most likely to increase in industrialized countries as a consequence of trade with developing countries because the former are well endowed with skilled labour. While in developing countries is expected to observe a decline in inequality. This would happen because developing countries are typically well endowed with low skill labour relative to developed countries. With a move to free trade, developing countries will be more competitive in low skill intensive sectors which will expand. The increased demand for low skilled workers, who typically belong to the poorer segments of the population, will lead to an increase in their wages relative to the wages of skilled workers. 7

8 As we have mentioned above, traditional trade models assume full employment, though some workers may be better or worse off in the long run due to changes in wages. It is assumed that on average, individuals would be better off as a result of overall efficiency gains triggered by trade liberalization. However, many economies are not characterized by full employment. In this case trade liberalization would reduce demand for workers mainly in import competing sectors and unemployment would increase. Recent trade models point out that adjustment processes may not only be observed between sectors but also within sectors. The new-new trade models that introduce firm heterogeneity and fixed-market entry costs predict that trade reform will trigger job creation and job destruction in all sectors, as both net-exporting and net-importing sectors will be characterized by expanding high-productivity firms and low-productivity firms that shrink or close down. This implies that an important reshuffling of jobs takes place within sectors. On the other hand, on the question of the integration of countries with different level of development, some of the new geography models predict catastrophic agglomeration in developed countries and deindustrialization in less developed ones. Nevertheless, these effects would depend on the way integration is managed, the intensity and speed of the process, and the extent to which specific policies aimed at reducing the negative effects are implemented. 6 Policies such as those aimed at promoting technological spillovers would reduce the negative impacts on the less developed countries (Baldwin et al, 2003). With respect to wages, the theoretical literature predicts that trade liberalization raises average income levels, and some contributions to the theoretical growth literature suggest that trade also stimulates growth. A large number of multi-country case studies and econometric studies using cross-country datasets have tested the empirical validity of the trade-growth relationship but there is no full agreement among economists concerning the precise nature of this relationship. 7 The reasons for this disagreement may lay in differences in the quality of the data as well as on econometric issues of the studies. As we mention above the classical link between trade and income inequality is based on the Stolper- Samuelson Theorem developed in a model that assumed full employment. According to this theorem 6 Roughly speaking an increase in the mobility in goods and labor will foster agglomeration in the more developed countries while mobility of capital and knowledge will foster dispersion, i.e. convergence among economies. On the other hand, policies aimed at promoting the diffusion of technology spillovers and at reducing congestion would be first best policies (Baldwin et al, 2003). 7 See for instance Baldwin (2003), Rodríguez and Rodrik (2001), Dollar and Kraay (2004), Loayaza, Fajnzylber and Calderon (2005) and Wacziarg and Welch (2003). 8

9 inequality is most likely to increase in industrialized countries as a consequence of trade with developing countries because the former are well endowed with skilled labour. While in developing countries is expected to observe a decline in inequality. This would happen because developing countries are typically well endowed with low skill labour relative to developed countries. With a move to free trade, developing countries will be more competitive in low skill intensive sectors which will expand. The increased demand for low skilled workers, who typically belong to the poorer segments of the population, will lead to an increase in their wages relative to the wages of skilled workers. It is worth to note that the majority of trade in industrialized countries is intra-industry trade, i.e. trade with other industrial countries. Thus, the changes in relative demand for different factors of productions predicted by Stolper-Samuelson are not likely to hold. In this regard Manasse and Turrini (2001) analysed whether intra-industry trade have an impact on the demand for high-skilled and lowskilled labour and conclude that intra-industry trade can raise wage inequality within countries and within sectors. Duranton (1999) comes to a similar conclusion in a model that combines intra-industry trade with technological change. In his model trade and technological progress lead to increase wage inequality. 3. Evidence on the Effects of Trade Liberalisation on Firm Performance 3.1. Evidence on productivity gains Most studies that use plant or firm level data to analyse the impact of trade liberalisation on productivity show mixed results. Among these works are those by Tybout, de Melo and Corbo (1991) and Pavcnik (2002) for Chile, Tybout and Westbrook (1996) for Mexico, Harrison (1994) for Cote d Ivoire, Muendler (2002) for Brazil, Lopez-Cordova and Mesquita (2002) for Mexico and Brazil, Fernandes (2007) for Colombia and Krishna and Mitra (1997) and Topalova (2004) for India. Tybout, de Melo and Corbo (1991) find little evidence of productivity growth in manufacturing after trade reform in Chile. Tybout and Westbrook (1996) use plant-level data to study the efficiency gains induced by the Mexico trade liberalisation. They find modest increases in productivity due to output share reallocations among firms with different productivity levels, and even slighter from scale efficient effects. Most of the average increase in productivity comes from the catch-all residual effect. This implies that most of the estimated overall efficiency gains are indeed left unexplained. 9

10 Harrison (1994) working with a panel of firms for Cote d Ivoire finds significant productivity gain after trade liberalisation. Mulaga and Weiss (1996) working at the firm level for the period find that the reduction in protection increases total factor productivity in Malawi. Nevertheless, when productivity estimates were adjusted for changes in capacity utilisation the association between productivity growth and the fall in protection vanishes. Lopez-Cordova and Mesquita (2003) estimate firm level productivity for Brazil and Mexico and test its causal links with trade and foreign direct investment variables. The results suggest strong trade related gains, with import discipline emerging as the dominant effect. Topalova (2004) analyses the effects of India s trade reform and finds that the reduction in trade protectionism leads to higher levels and growth of firms productivity, with a stronger effect for private firms compared to state companies. All these studies were conducted on firm or plant level panel data. Muendler (2002) using data on Brazilian manufacturing finds that the shutdown probability of inefficient firms rises with competition from abroad, thus contributing positively to aggregate productivity. Krishna and Mitra (1997) working with a panel of firms for the period obtained some weak evidence of an increase in the rate of growth in productivity after trade reform. Pavcnik (2002) and Fernandes (2007) observe productivity gains for Chile and Colombia respectively. Pavcnik (2002) compares plants productivity growth in the export oriented and import competing sectors with that of firms in the non traded sector. Her results suggest a dramatic productivity growth differential in favour of plants exposed to international competition with respect to inward-oriented plants. They also suggest that, in the case of a unilateral trade liberalization (such as the one experienced by Chile) trade-induced productivity gains can be higher for import competing plants relative to export oriented plants. Fernandes (2007) analyze whether increased exposure to foreign competition generates productivity gains for Colombia manufacturing plants. She uses a two step approach as well as a production function equation that includes trade policy measures as a regressor (direct approach). She finds a strong positive impact of liberalization on plant productivity, even after controlling for plant and industry heterogeneity, real exchange rates and cyclical effects. 10

11 Pavcnik (2002) and Fernandes (2007) find that the reallocation of resources in favour of more productive firms is a critical determinant of productivity growth and that, consistent with Melitz s (2003) model, this effect can be largely due to trade liberalization. Trefler (2004) using a difference-in-difference approach finds evidence that the Canada-US Free Trade Agreement has led to substantial increases in plants labour productivity as well as contraction of plants in the import competing and an expansion of plants in the export oriented industries. Casacuberta et al. (2003) analysing the Uruguayan Manufacturing sector find increases in total factor productivity especially in sectors where tariff reductions were larger and unions were not present. The main methodological problems involved in estimating the productivity gains from trade reform are the identification of the trade effects, and simultaneity and self-selection bias in the estimation of productivity Evidence of the effect of trade liberalization on employment and wages Regarding to the impact of trade liberalization on the labour market the literature has flourished in the last decades. The economic literature has produced a large number of empirical studies analysing the effects of trade on labour market outcomes. Nevertheless, so far no clear message emerges from the literature. The only general conclusion that may be justified is that employment effects depend on a large number of country-specific factors. One shortcoming of the studies is that they fail to distinguish the different possible causes of employment changes. Labour market policies, macroeconomic policies, technological changes or movements along the business cycle are only a few examples of factors that may affect an economy s employment level. In this regard, the work by Gaston and Trefler (1997) on the Canada-US Free Trade Agreement, make a distinction between the employment effects of the trade agreement and those of a general recession affecting both trading partners in the same period. Gaston and Trefler (1997) find that tariffs cuts contributed to reduce employment during the years following the agreement but that they also contributed to important productivity increases leading to long run efficiency gains. However, after controlling for recession, it appears that the FTA accounted for only 9-14 per cent of the jobs lost over the period. Trefler (2001) analysing the Canada-US free trade agreement finds 8 The simultaneity/endogeneity bias arises since more productive plants are willing to hire more inputs so the error term is positively correlated with factor inputs. The self-selection bias is induced by plant closing down. 11

12 instead a bigger role for the tariff cuts in the employment declines. According to his estimates nearly 30 per cent of the observed employment losses in manufacturing were a result of the FTA tariff cuts. His work shows that the adjustment process took seven years and during this process many workers moved to high-end manufacturing jobs along with dramatic productivity growth. Both, aspects reflect important long run efficiency gains from trade. Trefler also finds increases in workers annual earning and these increases are significantly higher in those industries that cut tariff rates most. Rama (2003) points out that in many developing countries temporary increases in unemployment following trade reform have been observed, though he finds that unemployment rates do not appear to be systematically higher in more open economies. Milner and Wright (1998) analyzed labour market responses to trade liberalization in Mauritius. They show that manufacturing employment increased significantly in the period following the 1983 trade liberalization. Though employment increases in the long run exceeded those that occurred immediately after the reform, the short-run impacts on employment were significant and positive. Rama (1994), in contrast, finds a negative effect of trade liberalization on employment in his analysis of trade policy reform in Uruguay in the late 1970s and early 1980s. Further evidence on developing countries is given by Harrison and Revenga (1995). They find evidence of increases in manufacturing employment following trade liberalization periods in Costa Rica, Peru and Uruguay. Instead, in a number of transitional economies (Czechoslovakia, Poland and Romania), employment fell during the transition period. As the authors note, however, those countries were undergoing significant other reforms that went well beyond trade liberalization. There are some cross-country studies that provide insights into the income effects of trade reform for subgroups in the population. The study by Rama (2003) explicitly looks at the effects of trade reform on wages and finds that wages grow faster in economies that integrate with the rest of the world. The author finds that trade can have a negative impact on wages in the short run, but finds that it only takes a few years for this effect to change sign. Lopez (2004) distinguishes between the short and long run effect of trade policies. He finds that trade openness raises inequality and stimulates growth at the same time and refers to trade liberalization as a win-lose policy. Improvements in infrastructure and in education on the other hand reduce inequality and increase growth at the same time, so does inflation reduction. Most empirical works for Latin America suggest that trade liberalization has led to and increase in both income and wage inequality and a skill bias of labour demand (Slaughter, 2000; Wood, 1997; Robbins,1996; Attanasio et al. 2004; Feenstra and Hanson, 1997; Perry and Olarreaga, 2006; Barraud, 2008). Dollar and Kraay (2002) find that trade openness affects income distribution positively. A 12

13 similar result is obtained by Behrman, Birdsall and Szely (2003) for a set of Latin American countries. However, Sanchez-Paramo and Schady (2003) find the opposite result in six Latin American countries, where trade volumes would negatively affect inequality. Spilimbergo et al. (1999) also find that trade openness would be associated with higher inequality, whereas Edwards (1998) does not find any significant effect of trade on income distribution. This literature does not appear to allow for any general conclusion on the link between trade liberalization and income distribution and the impression arises that this link is country and situation specific. Barraud (2008) analysing the effect of trade liberalization on wages for Argentina using difference in differences and matching techniques finds that labour market and poverty indicators deteriorated in the liberalization period in Argentina. Galiani and Porto (2006) find a negative effect of tariff reforms on the wage levels in Argentina. 9 For the Uruguayan case Casacuberta and Vaillant (2002) find higher reductions in employment and wages in those industries with higher tariff reductions. Hence, so far empirical research into the link between trade liberalization and market labour outcomes has produced mixed results. While the evidence for Asia seems to confirm a reduction in inequality following trade liberalization in Latin America inequality seems to show an increase. Several explanations have been put forward for the differences in outcome between East Asia and Latin America. One line of argument focuses on the fact that Latin America opened its markets later than the East Asian economies (Wood, 1997). As a result, the entry of China and other large lowincome Asian countries into the world market for labour-intensive manufactures in the 1980s shifted the comparative advantage of middle-income Latin American countries into goods of medium skills intensity. Increased openness in middle income countries thus reduced the relative demand for unskilled workers by causing sectors of low-skill intensity to contract. This would explain why relative wages of unskilled workers decreased. Moreover, it has also been argued that most Latin American economies are abundant in natural resources rather than low-skilled labour. This would also explain why wage inequality did not decrease in Latin America. Another explanation for rising inequality in some developing countries is that liberalization introduces new skill intensive activities into developing countries. For example, the work by Feenstra and Hanson (1997) shows that there is evidence that liberalization in Mexico induced larger FDI inflows from the 9 Winters et al. (2004) and Hertel and Reimer (2004) surveyed the effects of trade on income levels. 13

14 United States. These FDI inflows reflect the shift of low-skill intensive activities from the US to Mexico. Nevertheless, Mexico s relative demand for skilled workers within industries in manufacturing rose along with FDI inflows into those industries, which led to increases in Mexican wage inequality. One possible explanation for this finding is that jobs which were low-skill intensive in the United States were relatively skill-intensive in a country like Mexico. The tariff schedule in place before trade liberalization may also affect the impact of trade on wage inequality. If protection was higher in the low-skill intensive sectors, then trade liberalization may actually lead to shrinkage of these sectors. As a consequence, wage inequality would increase. It has been suggested in the literature that this phenomenon has been observed in Mexico and Morocco. Attanasio et al. (2003) find evidence for a link between trade, skill-biased technological change and increases in wage inequality in line with the theoretical literature discussed before. They analysed the effects of the drastic tariff reductions in the 1980s and 1990s in Colombia on the wage distribution. They identify three main channels through which the wage distribution was affected: increasing returns to college education, changes in industry wages that hurt sectors with initially lower wages and a higher fraction of unskilled workers, and shifts of the labour force towards the informal economy that typically pays lower wages and offers no benefits. Their results suggest that trade policy affected each of the three channels. The increase in the skill premium was primarily driven by skilled-biased technological change. However, the authors suggest that this change may have been in part motivated by the increased foreign competition to which the trade reform exposed domestic producers. They also find that wage premiums decreased by more in sectors that experience larger tariff cuts. Finally, they find some evidence that the increased size of the informal economy was related to increased foreign competition as sectors with larger tariff cuts and higher trade exposure experienced a greater increase in informality, though this effect was concentrated in the years prior to the labour market reform. The authors conclude that increasing returns to education and changes in industry premiums and informality alone cannot fully explain the increase in wage inequality observed over the period, suggesting that overall the effect of the trade reforms on the wage distribution may have been small. As we have already mentioned we should keep in mind the difficulty of isolating the effects of trade from other policies implemented simultaneously with trade reform. In most studies, the identification of trade effects relies on the comparison before and after a policy change. As a consequence, this approach attributes changes originating from other sources to trade policy. Most studies use data covering only a short time period after the reform which implies that the results can be heavily affected by the cyclical behaviour of the economy. The difference-in-difference methodology should eliminate the effects of common shocks and provides more precise description of the impact of trade policy as we explain in Section 4. 14

15 4. Empirical implementation This paper use a difference-in-differences methodology which allows to study the impact of increased trade exposure (the treatment) on the liberalizing group (the treated) relative to firms in industries that did not increase their exposure to foreign competition (the control group). The effect of trade liberalization is the estimated difference-in-difference of the outcome variable (productivity, employment, wages and relative wages between white and blue collars) between the treated and the control groups. The difference-in-difference methodology is implemented matching plants with similar propensity scores and Difference-in-Differences estimation Matching firms with similar propensity scores and difference-in-difference estimation Our aim is to evaluate the impact of increased trade exposure on a set of outcome variables (Y), where Y represents productivity, employment and wages. Y is referred to as the outcome in the evaluation literature. 10 Let TL it { 0,1} be an indicator (dummy variable) of whether plant i was exposed to greater foreign competition after MERCOSUR s creation, i.e. after time period t=1991, and 1 Y i, t + s the outcome of the 0 treated at t+s, i.e. after the creation of the MERCOSUR. Also denote by Y i,t+s the outcome of plant i had it not experienced a greater trade exposure (control). The causal effect of trade openness for plant i 1 0 at period (t+s) is defined as: Y i, t + s Y i, t + s. 0 The fundamental problem of causal inference is that the quantity Y i, t + s is unobservable. Thus the analysis can be viewed as confronting a missing data problem. In common to most of the microeconomic evaluation literature (cf. Heckman et al., 1997) we define the average effect of trade openness as: { Y 1, 1} {, 1} {, 1}, Y TL i t = = E Y TL i t = E Y TL i = E i t + s i, t + s i, t + s i, t + s t Eq.(2) Causal inference relies on the construction of the counterfactual for the last term in equation (2) E { Y 0, 1}, TL i = i t + s t which is the outcome that plants would have experienced on average had they not been exposed to greater trade competition after the creation of the Southern Common Market. The counterfactual is estimated by the corresponding average value of plants that belong to non-tradable 10 Blundell and Costa Dias (2000) present a review of the microeconomic evaluation literature. 15

16 industries, and therefore almost unaffected due to increased trade exposure E { Y 0, 0}, TL i = i t + s t important issue in the construction of the counterfactual is the selection of a valid control group. We assume that all the difference in the outcome (Y) between firms affected by increased trade openness and the appropriately selected control group is captured by a vector of observables X it and the level of the outcome variable Y i,t-1, before MERCOSUR s creation. The basic idea of matching is to select from the group of plants belonging to the non-tradable industries (non-treated or control group) those plants in which the distribution of the variables affecting the outcome is as similar as possible to the distribution to the firms belonging to the tradable industries (treated group). Nevertheless, some assumptions have to hold for this to be a valid comparison group. One of these assumptions is conditional independence that states that the treatment status is random, conditional on some set of attributes, X it, and independent of the potential outcomes 1 0 ( Y it, Yit ). This assumption implies that given a set of observable characteristics the outcome of a carefully defined group of individuals unaffected by the policy can be used as a counterfactual of the outcome levels of the treated had them not be treated. The matching procedure consists in linking each treated individual with the same values of the X it. To solve for the difficulties that arises when X it is multidimensional, the results of Rosenbaum and Rubin (1983) show that if the conditional independence assumption holds, it will also hold conditional on a single index that captures the information from the X it in the so called propensity score, i.e. ( Y 1, 0 it Y it ) TL it P( X it ). Thus, we adopt the propensity score matching method of Rosenbaum and Rubin (1983). To this end, we first identify the probability of being a plant affected by increased trade openness (the propensity score ) for all plants, irrespective if they belong to tradable or non-tradable sectors by means of a logit or probit model. For instance for the probit model: P ( TLi, t = 1) = F( X it ) where F is the normal cumulative distribution for the probit model, or the logistic for the logit model and X stands for full set of control variables.. An Let P it denote the predicted probability of being affected by trade openness at time t for plant i (which is actually or potentially affected by increased trade exposure). A plant k belonging to the non-tradable industries, which is closest in terms of its propensity score to a plant belonging to the tradable industries, is then selected as a match for the former. More formally, at each point in time and for each firm exposed to increased foreign competition i, a non-tradable plant j is selected such that 11 P it P { } { Pit Pkt } = kt = k Lib i, t 0 min. 11 A firm unaffected by increased trade openness can be match to more than one firm that experienced the effect of increased trade exposure. 16

17 This type of matching procedure is preferable to randomly or arbitrarily choosing the comparison group because it is less likely to suffer from selection bias by picking plants with markedly different characteristics. There are several matching techniques, and in this work we use the nearest-neighbour matching method and we also try the kernel matching method to check the robustness of the results. 12 Once selected the comparison group, we adopt a difference-in-difference methodology 13 to isolate the role of increased trade exposure on plants dynamics. As Blundell and Costa Dias (2004) point out, a combination of matching and difference-in-difference is likely to improve the quality of non-experimental evaluation studies. The difference-in-difference approach is a two step procedure. Firstly, the difference between the average output variable before and after MERCOSUR s creation is estimated for firms belonging to the tradable sectors, conditional on a set of covariates. However, this difference can not be attributed only to increased trade exposure since after the creation of MERCOSUR the output variables might be affected by other macroeconomic factors, such as policies aimed to stabilization of the economy. To cater for this the difference obtained at the first stage is further differenced with respect to the before and after difference for the control group of non-tradable plants. The difference-in-difference estimator therefore removes effects of common shocks and provides a more accurate description of the impact of trade openness. According to the literature the independent variables to include in the logit/probit regression should be correlated to the outcome variable and to participation in the policy, but they should not be potentially changed by the policy itself. Thus, the choice of variables priorizes the use of time invariant variables which poses another challenge to the analysis. To tackle the issue we construct some categorical variables such as size, high value added defined as a dummy that takes the value of one if the plant has a value added higher than the median- high gross output, high capital intensity (capital labour ratio) and export status as we explained below Variables As outcome variables we consider total factor productivity (TFP), employment and wages at the plant level. TLit is the trade liberalization variable (the intervention on the treated). TLit takes the value of one after MERCOSUR creation and zero otherwise. It is constructed by interacting plants belonging to 12 The matching is performed using the command psmacht2 in Stata, version 9 as described in Sianesi (2001). Additionally, we tested the balancing properties with the command pscore. 13 See Meyer (1994) for a review of this approach. 17

18 the tradable industries ( Libit, where tradables=1 and non-tradables=0) with a time dummy that takes the value of one from 1992 onwards (a year after the creation of the MERCOSUR). 14 Total Factor Productivity estimates were provided by Casacuberta et al. (2004). TFP was estimated using the Levinshon and Petrin s methodology which allows more accurate estimates since this method reduces the simultaneity problem present in most works when estimating total factor productivity at the micro level. As is usual in the empirical literature the estimation of a production function uses the real value of output rather than physical units of output produced by the plant as a measure of output. A measure of productivity based on the real value of output might not reflect the ranking of plants on their productivity if plants charge different mark-ups. Discriminating between the true productivity and the specific mark-up is quite difficult in the productivity literature. In order to distinguish the true efficiency from the plant specific mark-ups we would need plant level price data which is not available for Uruguay. This caveat should be considered when interpreting the results. 15 We measure employment as the number of workers per plant and we also discriminate between white collar and blue collars. Wages are measure as total wages per plant, wages per employee, wages per white collar and wages per blue collar. We also analyse the effect of increased trade exposure on gross output per plant. Wages are deflated by a wage index while gross output is deflated by the wholesales price index. To construct the liberalization variable ( TLit ), we define the treated group as those establishments belonging to the tradable industries ( Libit ) after MERCOSUR s creation. Our control group is integrated by firms belonging to the non-tradable industries, which are likely to be less affected by trade openness. In so doing we follow a similar definition to the one adopted by Pavnick (2002) 16 who defined as tradable industries those with an import penetration ratio (IP) equal or greater than 0.20 and/or and exports to output (EXP_O) ratio greater than 0.20 and non-tradable otherwise. 17 We checked that these variables (IP and EXP_O) were greater than 0.20 for the whole period ( ) taking into account the level as well as the stability of the import penetration and export-output ratio. 14 The Asuncion Treaty, signed on the 26th March of 1991 is a regional integration agreement to create the Southern Common Market. It was signed by Argentina, Brazil, Paraguay and Uruguay. 15 Harrison is one of the few studies that explicitly models plants mark-ups, assuming that plants are Cournot competitors and allows the mark-ups to vary over time and across industries, but not across plants within and industry. This is the same as assuming that all firms within an industry have the same market share. If one does not have plant level price data we need to impose some assumptions on the joint distribution of productivity and mark-ups. Bernard, Eaton, Jensen and Kortum (2000) provide an example. They set up a model in which they show that on average, a more efficient plant charges a higher mark-up. 16 Pavnick (2002) takes an IP and EXP_O of 15 % and carries out a sensitivity analysis for 10 % and 20 % of these indexes finding that results are robust to these levels of IP and EXP_O. 17 Import penetration is defined as total imports in the industry over total output, while exports to output as total exports over total output at the industry level. To construct these variables we use data from Nicita and Olarreaga (2007). 18

19 We should note that this definition of the tradable and non-tradable groups is not free of criticism: on one hand it may be sensitive to the level of aggregation used. Moreover, usually the non-tradable industries except for the work of Pavcnik (2002) - are defined as the service sector (construction, communication, transport, and financial services). Lack of plant level data on services for the period analysed prevented us from checking the sensitivity of the result using services as a control group. Besides, Barraud and Calfat (2008) analysing the effect of trade liberalization on wages for Argentina find evidence of significant impacts of trade liberalization on several non-tradable sectors as well as an important shift of manufacturing workers to services, which would indicate that the service sector is also likely to be affected by liberalization. Furthermore, in the Uruguayan case services are liberalized and consumed mainly by not residents namely tourism, transportation and financial services- except for the public services provided to domestic residents by public companies (electricity, fuels and telecommunications). As we have mentioned above, we further classify tradable industries into different groups according to its trade orientation: export oriented industries, import competing industries and intra-trade industries, since trade liberalization may have a different impact on plants belonging to the tradable industries according to the trade orientation of the various sub-groups. We define export oriented industries as those industries (EXPO) with an export/output ratio equal or greater than 0.20 and an import penetration ratio lower than 0.20 during the whole period under study. The import competing industries (IMPC) are defined as those with an import penetration ratio equal or greater than 0.20 but an export to output ratio lower than 0.20 during the whole period. Intra-industry trade industries (IIT) are defined as those industries with an import penetration and export ratio equal or greater than 0.20 for the whole period. 18 In Appendix 1 we present the import penetration and export-output ratio for the 26 industries considered in this work. We exclude industries 353 and 354 due to missing data and also those industries that experienced important changes in its trade orientation after MERCOSUR s creation. In a first step we compare tradable versus non-tradable industries, while afterwards we take a closer look into the tradable group according to the trade orientation of the various groups of industries that composed it. We also examine if the impact of trade liberalization differs within the tradable industries depending on the trade orientation of the various sub-sectors that integrates it (EXPO and IMPC) The advantage of these openness measures is that they are specific to the manufacturing industries while cross-country comparisons use aggregate measures that avoid having a better insight on industry and plants dynamics. Nevertheless, as we note previously, even though the difference-in-difference methodology should eliminate common shocks, we should be cautious in interpreting the results since under the period analyzed Uruguay did not only liberalized trade but also pursue other set of 18 The average plant export propensity of the tradable group is 0.15, while for the export oriented this figure is 0.25 and 0.05 for import competing plants, according to this definition. 19

20 macroeconomic policies aimed to the stabilization of the economy. One of these policies was the exchange rate policy which was pegged to the dollar and domestic currency appreciated in order to control inflation. 19 Thus, we checked that the output variables analysed in the treated and control group have a common trend, otherwise results will we flawed unless we use inverse probability weighting techniques Data Sources We use two data sources. Data at the plant level comes from the National Institute of Statistics (INE) for the period Estimates of TFP were provided by Casacuberta et al. (2004). Data at the industry level on imports and gross product comes from UNIDO dataset (Nicita and Olarreaga, 2007). This last set of data was used to classify industries as tradable and non-tradable as explained above. The micro level data provided by the National Institute of Statistic, Uruguay (INE) contain information on gross product, value added, expenditures on intermediates and materials, energy, employment, wages, sales, exports, capital and age at the plant level for the period In 1988, the starting year of our sample, the Second National Economic Census was conducted. The rest of the data comes from annual surveys. The surveys report information from manufacturing plants with five or more employees. All the plants with more than 100 employees are compulsorily included in the sample. A random sampling process is conducted on plants with less than 100 employees and has to satisfy the requirement that the total employment of all the selected establishments must account at least for 60 % of the total employment of the sector according to the economic Census of These selection criteria biased the database towards big plants. Each year the INE revises the sample coverage, and if necessary, due to the closure of firms, includes new ones. Once a firm enters the survey, it is followed until its death. Therefore, when we have no more data for a particular establishment this is interpreted as a plant closure (exit or death). Gross output, value added, sales and exports were deflated by the wholesale index with base year Capital was deflated by specific industry price deflators for capital, as well as employment, wages and energy. The deflators were provided by the Department of Economics, School of Social Science, elaborated by the INE. In the period there was an important reduction in the number of plants, which along with missing observations left us with and important reduction in the number of observations. Actually in the period there is exit as well entry of new firms which is quite difficult to analyze due to the sampling methodology followed by the INE. The INE periodically includes new establishments, but these do not 19 In future work we will also try the methodologies of Inverse Probability Weighting which allows for variations in the trends between treated and control groups. 20