WPS6268 Policy Research Working Paper 6268 What Drives Short-Run Labor Market Volatility in Offshoring Industries? Evidence from Northern Mexico during 2007–2009 David S. Kaplan Daniel Lederman Raymond Robertson The World Bank Poverty Reduction and Economic Management Network International Trade Department November 2012 Policy Research Working Paper 6268 Abstract Recent research shows that employment in Mexico’s wages to trade shocks. Third, fluctuations in Mexico-U.S. offshoring maquiladora industries is twice as volatile as trade were associated with changes in the composition of employment in their U.S. industry counterparts. The employment, with the skill level of workers rising during analyses in this paper use data from Mexico’s social downturns and falling during upswings. This implies that security records and U.S. customs between the first the correlation between average wages and trade shocks is quarter of 2007 and the last quarter of 2009 to identify partly driven by labor-force compositional effects, which four channels through which economic shocks emanating may obscure individual-worker wage flexibility. Fourth, from the United States were amplified when transmitted trade shocks affecting related industries (industries into Mexico’s offshoring labor market of Northern linked by employment flows affect employment at least Mexico. First, employment and imports within industries as much as own-industry trade shocks, thus amplifying are complements, which is consistent with imports being employment volatility through the propagation of shocks used as inputs for the assembly of exportable goods across industries within Northern Mexico. Furthermore, within industries. That is, when imports fell during the data suggest that the observed fluctuations in U.S.- the crisis, employment in Mexico was reduced rather Mexico trade at the onset of the Great Recession in the than protected by the fall of imports. Second, contrary U.S. were not associated with pre-existing employment to other studies, employment is more responsive than trends in Northern Mexico. This paper is a product of the International Trade Department, Poverty Reduction and Economic Management Network. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank. org. The author may be contacted at dlederman@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team What Drives Short-Run Labor Market Volatility in Offshoring Industries? Evidence from Northern Mexico during 2007-2009 David S. Kaplan Daniel Lederman Raymond Robertson Key words: trade shocks, labor market dynamics, adjustment costs JEL codes: F16, J63, J31 Kaplan: Inter-American Development Bank, LMK and IZA, dkaplan@iadb.org. Lederman: World Bank, PRMTR, dlederman@worldbank.org. Robertson: Macalester College, robertson@macalester.edu. The authors gratefully acknowledge financial support from the World Bank executed Multi-Donor Trust Fund on Trade and from the World Bank’s Latin American and Caribbean Regional Studies project on Labor Markets and Business Cycles. We are also indebted to Judith Frias and Eduardo Alcaraz for help with the IMSS data. The authors received invaluable feedback on preliminary results presented at seminars in the World Bank’s Development Research Group, Trade and International Integration unit (DECTI), the Inter-American Development Bank’s Labor Markets analysis unit (LMK), the 2010 LACEA-TIGN meetings, and the 2011 Eastern Economic Association meetings. We particularly would like to acknowledge the insightful comments received from Tomas Castagnino, Julian Messina, Pravin Krishna, Marcelo Olarreaga, Carmen Pages, and Rodrigo Wagner. . 1. Introduction Offshoring is a significant and increasingly important feature of international trade (Feenstra and Hanson 1997, Feenstra 2010). Integration into global value chains, often through offshoring or outsourcing of manufacturing assembly tasks, is a key part of the globalization experience of many developing countries (Robertson et al. 2009). While the importance of offshoring is well established, the labor-market effects in developing countries where assembly operations take place have received less attention. Most papers studying labor markets and offshoring focus on wages, ranging from Feenstra and Hanson (1997) to Hummels et al. (2011). Bergin et al. (2009) highlight employment volatility as another important implication of outsourcing. Volatility is important for several reasons. First, comprehensive trade reforms have left countries much more integrated with the global economy. As a result, popular concerns about globalization have extended beyond the static implications of trade reforms and towards developing countries’ exposure to short-run international shocks (Jansen and von Uexkull 2010). Second, employment volatility is twice as high in Mexican maquiladora industries as in their U.S. counterparts. If this result holds across countries, the rise of outsourcing and offshoring would have significant implications for developing economies. Bergin et al. (2009) rule out higher overall economic volatility in Mexico relative to the U.S. as well as industry size as candidate explanations for the high responsiveness of offshoring employment to economic shocks. They do not, however, provide any explanations for this high volatility. The goal of this paper is to use the onset of the Great Recession in the United States as an exogenous short-run shock (we support the claim of exogeneity in section 2) to identify four channels through which shocks emanating in the U.S. affect employment in Mexico’s offshoring maquiladora industries of Northern Mexico. 2 First, in non-offshoring (traditional) trade, exports are assumed to be positively correlated with employment and imports are generally presumed to reduce employment by displacing import-competing firms (e.g. Kletzer 2001). It is noteworthy, however, that Amiti and Konings (2007) show that the effect of tariffs on firm productivity differs when imports are inputs rather than final goods, with reductions of input tariffs having a positive effect on productivity. We show that in an offshoring environment, namely Northern Mexico, employment and imports are complements. This result implies that a trade shock like the global financial crisis of 2008-09, which reduced both imports and exports, will have a larger effect on employment than in a traditional trade environment where reductions in imports could ameliorate the reduction of employment. We also show that the recovery in employment once trade flows return to normal will be dramatic because the effects of imports and exports reinforce each other, leading to overall higher volatility. Second, studies of the impact of trade reforms on wages and employment find that average industry wages are more responsive to trade shocks than employment (the Hoekman and Winters 2005). In Mexico’s offshoring region, however, we find that employment was more responsive to short-run trade shocks than the average wage of individual workers who remained employed between quarters. As will become apparent, the data is consistent with downward sticky nominal wages of individual workers. Our main point is simply that in Mexico’s offshoring environment, the relevant dimension for adjustment to shocks at the individual level was employment rather than wages, even during the rather severe downturn of 2008-09. The high responsiveness of offshoring employment may be related to adjustment costs. Labor-market adjustment costs feature prominently in a recent wave of research that examines the effects of trade liberalization on labor markets in the context of discrete choice models of 3 employment decisions (Artuc et al. 2010, Dix-Carneiro 2010) and in the presence of firm heterogeneity (Davidson et al. 2010, Helpman et al. 2010, Felbermayr et al. 2011). Robertson and Dutkowsky (2002) estimate demand-side labor-market adjustment costs (i.e., the ease with which firms can adjust the number of employees) for Mexico that appear to be an order of magnitude lower than in the United States. Given that Mexico’s overall employment volatility is not higher than the U.S., however, these lower adjustment costs alone cannot explain the excess volatility of Mexican offshoring employment. Instead, we show that Mexico’s relative employment of production workers is much higher in the maquiladoras than in the rest of Mexico. Given the finding that demand-side adjustment costs are lower for production workers than nonproduction workers in Mexico (Robertson and Dutkowsky 2002), it is possible that part of excess employment volatility is due to the relatively high proportion of employed production workers. We test an implication of this hypothesis by comparing changes in the skill composition of employment in response to shocks and find that, consistent with the adjustment cost hypothesis, the adverse trade shock increases the relative skill intensity within plants. To be more precise, we focus on the changes in the average worker fixed effect and find that it rose during the downturn and declined during upswings. Finally, we also evaluate whether linkages across industries in an offshoring environment reinforce the effects of a trade shock. For this purpose, we propose a new indicator of industry relatedness to analyze how shocks are spread through the labor market, which relies on the observed mobility of workers across industries over time. While there is a large literature on the transmission of shocks across countries (e.g., Classens et al. 2011), there are few studies of the diffusion of external trade shocks within countries. The importance of worker mobility in understanding the effects of trade liberalization is at the heart of recent papers, including 4 Ebenstein et al. (2009), but to our knowledge no other papers have applied this idea to study intra-national diffusion of trade shocks. Our analysis draws upon a unique matched worker-firm dataset that allows us to control for individual worker characteristics as they move between firms. In the presence of heterogeneous firms and assortative matching, matched worker-firm data are required to control for worker, firm, and match heterogeneity that has been shown to have significant effects on wages, employment, and inequality (Davidson et al. 2010, and Helpman et al. 2010). Our unique dataset of quarterly observations is also particularly well suited to capture leads and lags in response to the sudden reduction in U.S. demand after September 2008. It also allows us to examine the changes in the employment composition (the mix of skilled and less skilled workers) that is featured in this literature. The rest of the paper is organized as follows. Section 2 describes the trade shock and the general trends in wages and employment in Northern Mexico. Section 3 describes the labor market data. Section 4 discusses our strategies for identifying each of the four channels through which shocks emanating from the U.S. were transmitted to and amplified in Mexico’s offshoring industries. Section 5 presents the econometric results, and Section 6 concludes by summarizing the main findings and lays out suggestions for future research. 2. The Great Trade Collapse and U.S.-Mexican Trade The Great Trade Collapse has received a great deal of attention both in the academic literature and popular press. The collection of essays published in Baldwin (2009) suggest the “Great Trade Collapse� between the third quarter of 2008 and the second quarter of 2009 was primarily a demand-side shock induced in large part by European Union and U.S. firms and 5 consumers postponing purchases of consumer durables and investment goods. Eaton et al. (2009) estimate that changes in demand for manufactured goods accounted for about 70 percent of the global decline in international trade (relative to gross domestic product). These authors cite the World Trade Organization’s estimate that merchandise trade dropped by 23 percent in 2009 relative to the previous year—the largest drop in trade by a factor of four since World War II. The impact of the trade collapse was especially acute in Mexico. The decline in Mexican trade was highly correlated with the decline in U.S. GDP (Robertson 2009). Our data show that formal employment in the trade-intensive northern states fell more than 9 percent from September 2008 to March 2009. Real log wages of workers who did not lose their jobs (stayed with the same firm) fell on average by 0.1% from September 2008 to December 2008 and by 1.2% from December 2008 to March 2009. As exports and imports with the United States began to recover in the second quarter of 2009, employment and wages recovered as well. Since this shock originated outside of Mexico, and there have been few, if any, suggestions that factors within Mexico induced the crisis, we consider the shock to be reasonably exogenous from Mexico’s point of view. Figure 1 shows total Mexican imports and exports for the 2000-2010 period. The effect of the 2008 crisis on trade flows is clear. Between the local peak (around April 2008) and the trough (January 2009) Mexican trade fell about 43% in real terms. The drop erased nearly a decade of trade growth in a few months, but recovery was also relatively quick. The very close relationship between Mexican imports and exports is not a new phenomenon. The close relationship between imports and exports is characteristic of a vertically-integrated economy that is heavily engaged in processing activities (offshoring). 6 Mexico's significant engagement in processing has traditionally been tied to Mexico's close economic relationship with the United States. Over the past ten years, however, Mexico's trade has been diversifying away from the United States. Figure 2, for example, shows a declining trend in the U.S. share of both Mexico's imports and exports. While over 80% of Mexico's non-petroleum exports continue to go to the United States, non-petroleum imports from the United States fell from over 70% at the beginning of the decade to less than 50%. The crisis, however, does not seem to have affected the share of Mexico's trade with the United States other than possibly to stabilize the values. This is important for our analysis because the lack of inter- country substitution as a result of the crisis suggests that trade with the United States is a reasonable proxy for total Mexican trade. 1 The change in trade flows varied significantly across industries during the crisis. Using Harmonized System 6-digit industries, Figure 3 shows the distribution of the change in imports and exports between April 2008 and January 2009. The average change (difference in log values) in Mexican exports to the United States was about 42%. The standard deviation was nearly 1.14. The mean change in Mexican imports from the United States was smaller—about a 17% drop—but the standard deviation was larger - just over 1.18. We take advantage of the differences in the changes in trade across industries to analyze how the trade shock begins as an industry-specific shock and then ripples through the labor market. In even classical trade models, prices are fixed in small countries, which implies that demand shocks would affect quantities more than prices. We calculated the coefficient of variation across both unit values and quantities for all available products across the 24 months in 2008 and 2009 traded between the United States and Mexico, combining changes along the 1 In the aftermath of the crisis, the share of Mexican exports going to the U.S. did decline slightly after 2010. 7 extensive and intensive margins. 2 The median coefficient of variation for imports was 0.49 for quantities and 0.22 for unit values. The comparable values for exports were 0.59 and 0.26. Given that the shock to Mexico was primarily through demand, these numbers are consistent with a relatively elastic supply curve. In a small-country trade-in-tasks environment, it might not be surprising that U.S. firms would cut quantity first as a short-run response to the shock. The picture that emerges from these simple figures and statistics is that the labor market in Northern Mexico was subjected to a significant external trade shock that induced the kind of volatility identified by Bergin et al. (2009), creating an excellent opportunity to study to link between offshoring and volatility. We describe the data used to study that link in the next section. 3. Data The empirics require trade data and labor-market data, and we discuss each below. 3.1. Trade Data The bilateral monthly trade data used for the econometric models originate from U.S. customs records, specifically from the United States International Trade Commission’s data web interface. The monthly data were then summed over quarters. The industry classification system originating from the trade data (6-digit level of the Harmonized System (HS)) differs significantly from the industry classification system from the Mexican wage and employment data described below. We therefore constructed an industry classification concordance table to match the employment and wage data from the Mexican social security records with the U.S.- 2 Product-month observations with zero trade values or missing data were not used to compute the variation coefficients. 8 Mexico bilateral trade data. The resulting data set covers 105 tradable industries and one non- tradable industry, which covers all workers who were employed in industries that could not be matched to the trade data by HS industries. Examples of the 105 tradable industries include beer, sugar, prepared vegetables, plasters, cement, and industrial chemicals. An important feature of this level of aggregation is that it is probably not sufficient to exclude products that are used as inputs for the production of final goods within a given industry category. Consequently imports in each industry include imported inputs, which are likely to be complementary with labor. 3.2. Employment and Wage Data The wage and employment information come from Mexico’s confidential social security records maintained by the Instituto Mexicano del Seguro Social (IMSS) in Mexico City. The IMSS gathers data from all plants (establishments) on wages paid to each registered employee. We use these data to calculate total employment by industry, and we work with end-of-quarter data on employment and wages from 2007-2009. The frequency and end-of-period feature of the data have implications for interpretation and model specification, which are discussed further below. Overall, the data set covers between 3.6 and 4.2 million workers from the Mexican states that share a border with the United States. In order to produce a data set that did not violate the confidentiality of the information, IMSS staff provided data on worker-firm pairs. That is, we received data on the employment and wage history of a person while he or she remained at the same firm. If a person moves to a new firm, the new worker-firm pair is coded with an entirely new identifier. We therefore cannot follow a person over time when she changes firms. We also do not know if two people are working in the same firm. To compensate for our inability to follow workers when they leave 9 their firms, IMSS also provided us with industry-level calculations of transition probabilities, which we use to calculate the degree to which industries are related to each other. These transition probabilities were calculated using the entire country, not just data from the states that border the United States, and therefore provide a more robust indicator of inter-industry mobility. The resulting data can be used to illustrate the relationship between Mexico-U.S. trade and employment in Northern Mexico during 2007-2009, and the resulting trends provide strong motivations for the discussion of the empirical strategies. Figure 4 shows quarterly total employment in tradable industries (i.e., those industries for which we found a match with the trade data under the HS classification) and Mexican exports to the United States. A positive correlation appears very strong during the period 2007q3-2009q1. Figure 5 plots the same export series together with total employment in non-tradable industries. Surprisingly, the correlation between exports and employment in non-tradables appears to be even stronger, especially during non-crisis periods, although it is also strong during the crisis. We interpret this surprising finding as suggestive evidence that trade shocks affect employment decisions in industries that are not necessarily directly engaged in trade, which lends credence to the idea that inter-industry labor mobility or other inter-industry linkages can play an important role in determining the empirical correlation between labor-market outcomes and trade. During the crisis period, however, tradable-sector employment fell proportionately more than employment in non-tradable industries, as shown in Figure 6. Despite the important differences in magnitudes, changes in employment in tradable and non-tradable industries are also highly correlated. Figure 7a shows trends in real log wage changes for tradable and non- tradable industries for workers that remained with their employers for consecutive quarters (i.e., 10 the changes in wages control for firm-worker fixed effects). These two series appear even more correlated than the employment series. The correlation between traded and non-traded sectors (in both employment and wages) raises the possibility of two (not mutually exclusive) hypotheses: trade shocks can affect industries that are not directly exposed to trade (through input or output linkages) or the correlation is driven by common shocks (such as interest rate variations, remittances, or any other shocks that could have similar effects across all industries). We evaluate these hypotheses below by controlling for common shocks and differentiating between input and output linkages. As a way of showing the importance of controlling for firm-worker fixed effects, Figure 7b presents the changes in the average log wage using all workers. The fact that figures 7a and 7b appear so different indicates that compositional changes are also taking place. That is, the trade shocks we are studying in this paper are not only affecting employment levels, but also the types of workers who are employed in these industries. For example, the fact that wages in tradable sectors fell in the fourth quarter of 2008 for stayers even though average log wages rose suggests that the tradable sectors laid off the lower-wage workers between the third and fourth quarters of 2008. We continue to examine these compositional changes below. 4. Trade Shocks, Offshoring, and Employment Volatility: Four Possible Explanations In this section, we describe our strategy for evaluating four possible explanations of higher volatility Mexico’s offshoring region. The four possible explanations are employment- import complementary, import-wage and employment elasticities, employment composition changes and non-traded industry employment responsiveness. 11 4.1. Imports, Exports, and Employment: Sign and Dynamics Three key estimation issues shape our analysis of the effects of trade on employment in Mexico. First, there is some debate about whether prices or quantities are the relevant metric of trade. 3 We use quantities (imports and exports) in our analysis because we are dealing with essentially a macroeconomic shock rather than a change in tariffs that would change relative prices. Furthermore, in the context of a vertically-integrated small economy, quantities may be the more relevant adjustment margin. Nevertheless, we compare the relative changes of prices and quantities to affirm our choice. Second, quantities (imports and exports) are frequently used measures of globalization and in most of these studies, imports are generally considered to have adverse effects on wages and employment (e.g. Kletzer 2001). This assumption may not hold in the offshoring environment where imports are significantly comprised of intermediate inputs. We therefore remain open to the possibility of positive effects of imports. Third, since we are dealing with offshoring, it is important to consider assembly time in our model. Specifically, in the offshoring model, imports are primarily inputs that are assembled into final goods (or the next stage of production) and then exported. As a result, the estimated effect of imports and exports can be sensitive to the lag structure specified in the model. We therefore incorporate a general lag structure in our estimation. Our initial null hypothesis is that of Kletzer (2001): that imports reduce employment and therefore would mitigate employment volatility due to a drop in foreign demand. In the offshoring environment, the opposite is possible. The basic employment equation that we use to evaluate this hypothesis is simply formalized as: (1) 𝐸𝑖,𝑡 = 𝛽𝑚 ∙ 𝑚𝑖,𝑡 + 𝛽𝑥 ∙ 𝑥𝑖 ,𝑡 + 𝛼𝑖 + 𝛾𝑡 + 𝜀𝑖,𝑡 . 3 Richardson (1995), for example, makes this point in the context of the globalization and inequality debate. 12 The subscripts represent industries and time periods. E is the log of employment in industry i at the end of quarter t, m is the log of that industry’s imports from the United States, and x is the log of Mexican exports of that industry to the United States. The inclusion of industry fixed effects, 𝛼𝑖 , makes this model a typical within-industry model. The time-period specific fixed effect, 𝛾𝑡 , acquires particular importance in our application, because common effects across industries are especially important for identifying the effects of trade shocks in the context of a broader macroeconomic shock. The parameters 𝛽𝑚 and 𝛽𝑥 are the elasticities of interest. If imports are complementary inputs to labor, that is, if both are needed to produce a unit of a final good, then 𝛽𝑚 > 0 and 𝛽𝑥 > 0. The error term 𝜀𝑖,𝑡 is assumed to be the standard regression error that is uncorrelated with the independent variables. 4 We are also interested in allowing for assembly time. The resulting dynamic assembly time specification of the model can be written as: (2) 𝐸𝑖,𝑡 = 𝛽𝑚 ∙ 𝑚𝑖,𝑡 + 𝛽𝑥 ∙ 𝑥𝑖 ,𝑡 + 𝛽𝑚2 ∙ 𝑚𝑖 ,𝑡+1 + 𝛽𝑥2 ∙ 𝑥𝑖 ,𝑡+1 + 𝛼𝑖 + 𝛾𝑡 + 𝜀𝑖,𝑡 , in which the inclusion of exports one period ahead captures the time lag in the realization of exports. The inclusion of one-period-ahead imports toughens the test of the dynamic model, because evidence in favor of the model requires that 𝛽𝑥2 > 0 with 𝛽𝑚2 = 0. Also, lagged imports and exports are implicitly taken into account as the variables represent average trade flows during the quarter while the dependent variable is measured at the end of the quarter. In practice, both models are estimated in differences to take out the industry fixed effects. The final assembly time specification is thus: (3) ∆𝐸𝑖,𝑡 = 𝛽𝑚 ∙ ∆𝑚𝑖,𝑡 + 𝛽𝑥 ∙ ∆𝑥𝑖 ,𝑡 + 𝛽𝑚2 ∙ ∆𝑚𝑖,𝑡+1 + 𝛽𝑥2 ∙ ∆𝑥𝑖,𝑡+1 + 𝛾𝑡 + ∆𝜀𝑖,𝑡 . This equation is evaluated in Section 5.1. 4 We do, however, calculate the standard errors of our coefficients allowing for the possibility that the errors within an industry are arbitrarily correlated with each other over time (serial correlation) using the “cluster� option in Stata. 13 4.2. Wages within Industries One reason employment is volatile in offshoring industries is that the sector may take wages as given, either due to a relatively small size of the sector or due to a very elastic labor supply. Elastic labor supply implies that employment is more volatile than wages. Bergin et al. (2009) rule out industry size, leaving the possibility of elastic supply. The models used to estimate the wage effects of trade mimic the employment equations (1) - (3) presented above. As mentioned, however, it is important to control for firm-worker specific effects in the wage equation in order to be sure that the estimates are not affected by selectivity biases. Therefore the approach to estimating the wage effects follows two stages. The first strips out the match fixed effect using an underlying first-stage wage model: (4) 𝑤𝑓,𝑤,𝑖,𝑡 = 𝛾𝑖,𝑡 + 𝜃𝑤,𝑓 + 𝜀𝑓,𝑤,𝑖,𝑡 . The dependent variable is the log of wages of a worker w, in firm f, industry i, at time t. The first term on the right-hand side is the industry-time effect that we are interested in studying in the second stage. The second term is the worker-firm specific effect. By estimating (4) in differences, this second term disappears and the resulting model is: (5) ∆𝑤𝑓,𝑤,𝑖,𝑡 = 𝛾𝑖,𝑡 + ∆𝜀𝑓,𝑤,𝑖,𝑡 . In this case, 𝛾𝑖,𝑡 is the estimated average change in log wages for all workers that did not lose their jobs in industry i between t and t-1. Therefore the second stage estimation of the trade effects on wages within industries is: (6) 𝛾𝑖,𝑡 = 𝛽𝑚 ∙ ∆𝑚𝑖,𝑡 + 𝛽𝑥 ∙ ∆𝑥𝑖,𝑡 + 𝛽𝑚2 ∙ ∆𝑚𝑖,𝑡+1 + 𝛽𝑥2 ∙ ∆𝑥𝑖,𝑡+1 + 𝛾𝑡 + ∆𝜀𝑖,𝑡 . One concern about this model is that the dependent variable is an estimate, not a precise statistic. Consequently, we estimated (6) with Weighted Least Squares, with the weights for each 14 observation being the initial levels of employment. In addition, we used the inverse of the first- stage standard error of 𝛾𝑖,𝑡 as weights, and the results were virtually identical, because the standard errors of 𝛾𝑖,𝑡 are negatively correlated with the size of the industry. For the sake of brevity, the results reported below are limited to the WLS estimates with initial employment as 5 weights. 4.3. Labor Market Adjustment Costs: Changes in the Composition of Workers At least since Reder (1955), the labor literature has emphasized compositional changes in employment during business cycles. Reder (1955) argued that upturns (when labor is scarce) are associated with downgrading of the skill composition of workers and downturns (when employers have a larger pool to choose from) are associated with upgrading. 6 Gautier et al. (2002) found evidence in favor of these predictions in matched employer-employee data from the Netherlands. Since skilled workers in Mexico have been found to have higher adjustment costs (Robertson and Dutkowsky 2002), it seems possible that the financial crisis led to changes in the firm-level mix of skilled and less skilled workers. Using data from the Mexican Industrial Census, Bernard, Robertson, and Schott (2010) show both maquila/total employment ratios, state-wide non-production/production employment ratios, and maquila non-production/production employment ratios by Mexican states (and the Federal District) that have non-zero maquiladora employment. Using these data, Table 1 shows that the maquiladora industries have a much lower non-production/production worker employment ratio, as might be expected if one of the main motivating factors for offshoring in Mexico is to take advantage of lower labor costs. If adjustment costs are lower for production 5 The results with the standard errors as weights are available upon request. 6 On the labor force upgrading and business cycles, see also Hamermesh (1993). 15 workers, a trade shock in the offshoring industries would have much larger employment effects than in countries that have shifted employment towards skilled labor in response to offshoring (as in the United States). Furthermore, employment should shift towards non-production workers following an adverse trade shock. Although the empirical models of wage determination focus on the wage changes for stayers, it is useful to use a measure that gives us insight on the changing composition of workers as a result of trade shocks. More formally, we denote � 𝑠𝑡𝑎𝑦𝑒𝑟𝑠 ∑𝑤∈𝑠𝑡𝑎𝑦𝑒𝑟𝑠 𝑙𝑛�𝑤𝑤,𝑖,𝑡 � − 𝑙𝑛�𝑤𝑤,𝑖,𝑡−1 � ∑� 𝑤 𝑙𝑛�𝑤𝑤,𝑖,𝑡 � 𝑡 ∑� 𝑤 𝑡−1 𝑙𝑛�𝑤𝑤,𝑖,𝑡 � 𝑑𝑖𝑓𝑓𝑖 ,𝑡 = −� − � �𝑖,𝑠𝑡𝑎𝑦𝑒𝑟𝑠 �𝑖 ,𝑡 �𝑖,𝑡−1 as the difference between the change in average log wages for stayers and the change in average log wages for all workers in industry i in time t. What might this measure capture? Consider the following model of wage determination, 𝑙𝑛�𝑤𝑤,𝑠,𝑡 � = 𝛼𝑤 + 𝛾𝑖,𝑡 + 𝜀𝑤,𝑓,𝑡 , where 𝛼𝑤 is a fixed effect for each worker and 𝛾𝑖,𝑡 is a time varying industry or sector effect. Under this model of wage determination, the first term of the differential measure expressed above (the part corresponding to the wage changes of stayers) reduces to �𝛾𝑖,𝑡 − 𝛾𝑖,𝑡−1 � since the person fixed effects are differenced away. The second term would reduce to �𝑖,𝑡 − 𝛼 �𝛾𝑖,𝑡 − 𝛾𝑖,𝑡−1 � + �𝛼 �𝑖,𝑡−1 �, �𝑖 ,𝑡 is the average person effect in industry i in time t. This formulation implies that the where 𝛼 difference between the change in average log wages of stayers and the change in average log wages of all workers in industry i in time t could be expressed as �𝑖,𝑡−1 − 𝛼 𝑑𝑖𝑓𝑓𝑖,𝑡 = �𝛼 �𝑖,𝑡 �. 16 A positive value for 𝑑𝑖𝑓𝑓𝑖,𝑡 could therefore be interpreted as a downgrading of average skill or human capital in the industry while a negative value would be interpreted as an upgrading of skill or human capital in the industry. The econometric exercises, therefore, can ascertain whether employers tend to upgrade the quality of their workforce during downturns driven by trade shocks. Figure 8 presents the difference between the change in average log wages for stayers and the change in average log wages for all workers separately for tradable and non-tradable sectors. The figure shows that the variation of this measure was much larger in the tradable sector than in the non-tradable sector. In particular, figure 8 suggests that the tradable sector shed low-skilled workers in the fourth quarter of 2008 and hired them back in the third quarter of 2009. 4.4. Inter-industry “Related� Employment Labor mobility across industries may attenuate estimated wage effects (Ebenstein et al. 2009) and is costly (Artuc et al. 2007a and 2010, Felbermayr et al. 2011, Davidson et al. 2008, and Helpman 2010). Our new measure of relatedness specifically addresses labor mobility concerns. The empirical models of employment and wages can be augmented to address concerns about labor mobility. The additional explanatory variables need to satisfy two conditions. First, they need to capture the trade shocks affecting industries that employ similar workers, and second, they need to weigh these trade shocks in other industries by the extent to which workers move between industries. We propose the following “relatedness� indices for imports and exports that satisfy both conditions: ∑ 𝑙𝑗≠𝑖 ∙𝑚𝑗,𝑡 (7) 𝐼𝑖𝑚 ,𝑡 ≡ , and 𝐿𝑖 17 ∑ 𝑙𝑗≠𝑖 ∙𝑗 (8) 𝐼𝑖𝑥 ,𝑡 ≡ , 𝐿𝑖 where superscripts m and x denote the indices for imports and exports respectively. Denote l as the number of workers (in all of Mexico) employed in industry i at time t, but that were also employed in any other industry j during 2008-09. Intuitively, these indices are equal to the weighted average of “related� imports and exports in each time period, and the weights are the share of workers employed in industry i but who also were employed in any other industry j during 2008 and 2009. 7 Since confidentiality concerns preclude us from tracking workers over time when they change firms these industry-level “transition probabilities� were calculated separately by IMSS staff. It is also noteworthy that industries j can include non-tradable industries, in which case the value of imports and exports would be equal to zero (we set the log of exports or imports equal to zero in these cases). Hence the variance of these indices is lower than the variance of industry- specific trade flows. The variance of these indices is also lower because they are averages across numerous sectors. This feature of the indices is important for interpreting the economic magnitude of the coefficients. The augmented employment and wage equations in differences (within industries) in the assembly time specification that take into account the indirect effect of trade shocks on workers via their effects through related industries are: 𝑚 ∆𝐸𝑖,𝑡 = 𝛽𝑚 ∙ ∆𝑚𝑖,𝑡 + 𝛽𝑥 ∙ ∆𝑥𝑖,𝑡 + 𝛽𝑚2 ∙ ∆𝑚𝑖,𝑡+1 + 𝛽𝑥2 ∙ ∆𝑥𝑖,𝑡+1 + 𝛽𝐼𝑚 ∙ ∆𝐼𝑖,𝑡 + (9) 𝑥 𝑚 𝑥 𝛽𝐼𝑥 ∙ ∆𝐼𝑖,𝑡 + 𝛽𝐼𝑚2 ∙ ∆𝐼𝑖,𝑡+1 + 𝛽𝐼𝑥2 ∙ ∆𝐼𝑖,𝑡+1 + 𝛾𝑡 + ∆𝜀𝑖,𝑡 , and 𝑚 ∆𝑤𝑖,𝑡 = 𝛽𝑚 ∙ ∆𝑚𝑖,𝑡 + 𝛽𝑥 ∙ ∆𝑥𝑖,𝑡 + 𝛽𝑚2 ∙ ∆𝑚𝑖,𝑡+1 + 𝛽𝑥2 ∙ ∆𝑥𝑖 ,𝑡+1 + 𝛽𝐼𝑚 ∙ ∆𝐼𝑖,𝑡 + (10) 𝑥 𝑚 𝑥 𝛽𝐼𝑥 ∙ ∆𝐼𝑖,𝑡 + 𝛽𝐼𝑚2 ∙ ∆𝐼𝑖,𝑡+1 + 𝛽𝐼𝑥2 ∙ ∆𝐼𝑖,𝑡+1 + 𝛾𝑡 + ∆𝜀𝑖,𝑡 . 7 Econometric results using employment data from 1997-9 were strikingly similar to those reported below, thus suggesting that endogeneity in labor mobility across industries during the period under examination is not a serious problem. 18 The coefficients on the “related� trade flows can theoretically have positive or negative signs, depending on the economic nature of the mobility of labor across industries. On the one hand, related employment across industries can be due to worker characteristics, such as occupations, that are employed in different industries. If workers move between such horizontally-related industries, then positive shocks to related industries could be associated with declines in employment and increases in wages in a given industry, i.e., 𝛽𝐼𝑥 < 0 or 𝛽𝐼𝑚 < 0 (when imports complement employment) in equation (9) and 𝛽𝐼𝑥 > 0 or 𝛽𝐼𝑚 > 0 in the wage equation (10). On the other hand, inter-industry relatedness may be driven by local supply chains. If workers are more likely to move between vertically-related industries (i.e., industries characterized by input-output relationships) then the expected coefficients in the employment equation could be different from those discussed above, because positive shocks to related industries would imply positive shocks for the supply chain. Specifically, the expectation would be that 𝛽𝐼𝑥 > 0 or 𝛽𝐼𝑚 > 0 (again, when imports are complements of labor) in equation (9) and 𝛽𝐼𝑥 > 0 or 𝛽𝐼𝑚 > 0 in the wage equation (10). This estimation strategy errs on the side of caution. The inclusion of import and export variables both preceding the employment and wage observations (which correspond to the end of each period) and one period ahead as well as the inclusion of the related trade variables with the same leads and lags structure is a rather general specification of the model, and collinearity can mask the significance of the estimated coefficients. Therefore we present results from the models without leads, with leads, with “related� trade variables, and a final specification with only the explanatory variables that appear to be significant in the most general specification. Furthermore, we also discuss one model that tests for crisis-specific (during the two quarters of 2008q4 and 19 2009q1) coefficients of the latter model. Finally, due to concerns about collinearity, we also present results for the pure “time-to-assemble� specification, which includes imports and related imports together with only one lead of the export and related exports explanatory variables. 5. Results Before proceeding to the main results, we present some basic statistics for both the trade and labor market variables in Table 2. It is interesting to note that, on average, the quarterly changes in exports and imports were close to zero during this period. These figures, however, mask considerable heterogeneity across industries over time. Also, imports and exports are much more volatile than their associated relatedness indices, which will be important later when we interpret the coefficients from regressions using these variables. The presentation of the results follows the same structure of Section 4. The estimation results are discussed sequentially with the employment equations followed by the wage equations. We include the results for relatedness within the employment and wage sections. We then discuss change in composition and conclude with the continuous-treatment models of employment and wages that allow us to examine the robustness of our results. 5.1. Imports, Exports, and Employment Table 3 presents the results from the WLS estimations in which initial employment in 2007 in each industry is the weight. 8 The first column contains the basic results, and only imports from the United States appear significant and positive. The second specification is the assembly-time model that includes the one-period-ahead trade variables. Both export variables 8 Similar results were obtained when using the inverse of the standard errors from the first-stage wage equation. These are available upon request. 20 appear significant and positive. Model 3 is the dynamic model augmented with the related trade variables. The results are similar to those of the previous model, but related imports (preceding the employment observed at the end of the period) appear significant as well. The fourth specification excludes the explanatory variables that were insignificant in model 3, and it confirms that the related trade variables are significant. The final specification tests for crisis- specific coefficients and the results imply that the positive partial correlation between exports to the United States and employment was magnified during the crisis period. These results on employment suggest that the assembly time model is probably the correct specification; employment decisions tend to take place prior to and ahead of exports. Moreover, both own-industry and related-industry imports appear with positive coefficients, thus suggesting that they complement labor. This finding is consistent with the view that Northern Mexico’s industries are largely vertically-integrated maquila operations that rely on imported inputs. Given the level of aggregation of our industrial classification, which is similar to the aggregations used elsewhere in the literature, it is possible that numerous studies of the relationship between imports and labor (or even imports and firm productivity) tend to confound import-competing and imported-input effects. In Northern Mexico, the latter effect appears to dominate. 5.2. The relative responsiveness of wages and employment Table 4 contains the results for the wage equations, following the same sequence as the employment models. In column one, both trade variables appear significant and with positive coefficients. As we move across the table few other explanatory variables are significant, not even the interaction of the crisis dummy with own-industry exports. It is also worth noting that 21 the elasticities of wages with respect to exports are much smaller than the elasticities in the employment regressions discussed above. Hence we interpret these results as evidence that most of the inter-quarterly adjustment of labor markets in Northern Mexico occurred through employment quantities rather than wages. One possible interpretation of these results is that firms are small relative to the labor market and therefore take wages as given in the short run. These results are also consistent with an economy characterized by relatively low demand-side employment adjustment costs. 5.3. Employment composition In spite of the weak results concerning the wages of workers that stayed with a given employer, it is possible that employers changed the composition of their labor force. Table 5 presents the results related to the determinants of the relative wages of stayers relative to the average wage of all workers, which, as mentioned, can be interpreted as reflecting compositional changes in the employed labor force. Several results appear robust. Under column 1, both contemporaneous exports to and imports from the U.S. have positive and significant coefficients, which suggest that increases in exports and imports lead to the disproportional hiring of lower-skilled workers. Conversely, a trade collapse would lead to lower-skilled workers to disproportionately lose their jobs. The assembly-time model estimates under column 2 indicate that one-period-ahead exports are also positive and significant. These three variables remain positive and significant with only minor changes in the point estimates across specifications. Under the third column, the results suggest that related trade shocks are also significant with positive coefficients on both 22 related imports and exports. Overall, a clear picture emerges that lower-skilled workers are the primary beneficiaries of positive trade shocks and suffer the most from negative trade shocks. The fourth model’s results suggest that some coefficients were different during the crisis quarters. In particular, the coefficient of contemporaneous exports was significantly smaller, implying that the effect of exports on the composition of workers was essentially zero during the crisis period while it was about 0.03 otherwise. In contrast, the coefficients on one-period-ahead exports and related exports were both larger in magnitude during the crisis than during other quarters. Finally, it is noteworthy that the coefficients reported in Table 5 tend to be much larger than the coefficients reported in Table 4, which focused on the effects of trade shocks on the wages of workers that stayed with their employers. For example, comparing the coefficient on the only significant variable in Table 4, namely exports to the U.S., under columns 3 in both tables, the magnitude is more than three times larger in Table 5 than in Table 4 (0.024 versus 0.007). Consequently, the negative effects that exports have on average log wages that are associated with skill downgrading dominate the positive effects of exports on average wages associated with wage increases for stayers. To the extent that the previous literature has not adequately differentiated between short- run and long-run effects, our results could shed light on the reasons why the literature has had trouble finding the effects of trade on wages and has found mixed results for employment. Our results from Table 4 suggest that an increase in exports is associated with a small increase in wages for workers who remain with the same firm. Our results from Table 5, however, suggest that an increase in exports is associated with skill-downgrading, which has the effect of making industry wages appear to be negatively correlated with exports. Our results therefore highlight 23 the importance of distinguishing between short-run and long-run when examining the within- industry compositional effects of trade shocks found to be so important in recent papers. 5.4. Relatedness Above we argued that the fourth channel through which trade shocks are magnified in offshoring labor markets is through intra-national, inter-industry ripple effects. There is evidence of the diffusion of shocks through quantitative adjustments of Northern Mexico’s formal labor market in Tables 3 and 5, which focus, respectively, on changes in employment levels within industries and changes in the composition of the employed labor force. As with the other effects discussed thus far, the evidence of inter-industry ripple effects affecting the wages of stayers reported in table 4 is much weaker. Regarding the economic magnitude of the related-trade effects versus those of the own- industry effects, the elasticity of employment with respect to related-exports is substantially larger than that of own-industry exports (table 3). As mentioned in section 4, however, the variance of the related-trade variables is much lower than that of own-industry trade. In particular, the standard deviation (weighted by initial employment) of the change in log exports is 0.25, while the corresponding standard deviation for the related exports variable is only 0.04. Multiplying the coefficient on direct exports from column 4 of table 2 by its standard deviation yields a “standardized� coefficient of 0.008, implying that a one standard deviation increase in the change of log exports yields a 0.8% increase in employment. Undertaking the same exercise for the related export coefficient yields a “standardized� coefficient of 0.007, implying that a one standard deviation increase in the change of related log exports yields a 0.7% increase in employment. 24 The analogous “standardization� exercise for the import coefficients yields economic magnitudes that are somewhat lower than those for the export coefficients. Multiplying the coefficient on direct imports from column 4 of table 3 by its standard deviation (0.23) yields a “standardized� coefficient of 0.004, while multiplying the coefficient on related imports by its standard deviation (0.04) yields a “standardized� coefficient of 0.005. In table 5, as mentioned, related trade shocks are significant predictors of changes in the relative wage of stayers, which reflect compositional changes in the employed labor force within industries. Using the estimated elasticity in column 3 of table 5 (0.065), the results imply that a one standard deviation increase in related imports (0.04) is associated with a 0.26 percent increase in the relative wage growth of stayers. That is, given the definition of this variable – recall section 4.3 – this standardized coefficient implies that the average worker fixed effect of the employed labor force falls by 0.26 percent. In contrast, the standardized coefficient on direct imports implies that a one standard deviation increase in direct imports is associated with approximately a 0.14 percent increase in the relative wage of stayers, which implies an analogous reduction of 0.14 percent in the average employed-worker fixed effect. In other words, as far as the composition of the labor force is concerned, complementary indirect import shocks (via related industries) tend to have larger impact than direct import shocks (0.26 versus 0.14). 5.5. Robustness: Continuous Treatment Effects and Exogeneity of Trade Shocks One potential concern with our analysis is that exports and imports might not be truly exogenous. A related concern is that those industries that were most affected by the Great Trade Collapse were already experiencing different time trends prior to the collapse. In order to address these concerns, we adopt an empirical strategy that compares the labor markets of industries that 25 suffered comparatively large trade collapses with those that experienced comparatively mild collapses prior to the collapse, during the collapse, and after the collapse. In order to motivate models that focus on the large and negative export shock that occurred at the end of 2008, we present graphs on the effects of this shock on our three dependent variables. We estimated these models with various definitions of the crisis period, including October-November 2008, October-December 2008, October 2008-January 2009, and October 2008-Feberuary 2009. The results were very similar across these definitions and for the sake of brevity the corresponding results section presents graphs based on the decline from October-November 2008. We divided industries into three groups of approximately equal size in terms of their employment in the first quarter of 2007 that are grouped by the percent decline in exports between October 2008 and November 2008. We also divide industries into three groups using the same procedure for the shock to related exports. 9 Figure 9a shows the results for employment when dividing industries into groups based on the magnitude of the shock to direct exports. The first thing to notice from figure 9a is that the employment trends of the three groups of industries were quite similar prior to the trade shock at the end of 2008. As we mentioned earlier, the similarity of the pre-shock trends provides empirical support for our assertion that the differences across industries in the size of the shock were exogenous. Figure 9a also provides interesting information on the manner in which these exogenous trade shocks rippled through the labor market over time. Compared to industries that experienced only mild shocks, industries in the other two categories had larger reductions in employment in 9 Industries were categorized as having large export shocks if the decrease in log exports from October to November 2008 was greater than 0.23 log points. Industries were categorized as having medium export shocks if the decrease was greater than 0.18 log points but less than 0.23 log points. Industries were categorized as having small export shocks if the decrease was less than 0.18 log points. The analogous cutoff points for the groups defined by the shocks to related exports were 0.094 and 0.0654. 26 the fourth quarter of 2002 and the first quarter of 2003. There is some evidence that employment recovered towards the end of 2009. Figure 9b shows the analogous results for industries grouped by the magnitude of the shock to related exports and also reveals no evidence of differential trends prior to the shock. Compared to industries that experienced relatively mild related export shocks, industries in the other two categories experienced larger drops in log employment at the end of 2008 and beginning of 2009, with clear evidence of an employment recovery towards the end of 2009. 10 This clear evidence of recovery is likely due to the fact that industries that experienced comparatively larger negative shocks towards the end of 2008 also caught up (that is, experienced comparatively larger positive shocks) in 2009. These results are entirely consistent with the results presented in Table 3 (that is, with a positive correlation between trade flows and employment). An additional attractive feature about the evidence presented in Figure 9b is that one can see how trade shocks of different magnitudes across industries, along with the associated recoveries of different magnitudes across industries, were mirrored by similar employment effects over time. Figure 10a shows the results on wage changes of stayers dividing industries into groups based on the magnitude of the shock to direct exports. Once again, we see no evidence that the trends were different across these groups of industries prior to the trade collapse at the end of 2008. Compared to industries that experienced only mild shocks, industries in the other two categories had poorer wage performance in the beginning of 2009, with a recovery of wages occurring at the end of 2009. Figure 10b shows similar results for wage changes for stayers when grouping industries by the magnitude of the shock to related exports. 10 Although the cross-industry correlation between the shock to direct exports and the shock to exports in related industries is weakly positive (p-value of 0.53), subsequent tables confirm the results of these simple graphs in models that control for both variables at the same time. 27 Figure 11a shows the results on our measure of workforce composition (the difference between the change in average log wages for stayers and the change in average log wages for all workers), again dividing industries into groups based on the magnitude of the shock to direct exports. Again the pre-shock trends across industry groups appear quite similar. Compared to industries that experienced only mild shocks, industries in the other two categories shed low- skilled workers at the end of 2008 and beginning of 2009, but reversed this trend at the end of 2009. Figure 11b shows similar results when grouping industries by the magnitude of the shock to related exports. We now proceed to the econometric models that implement the continuous treatment design. Specifically, we estimated the following continuous-treatment effects models of the export and related-exports shocks on employment and wages: 𝑥 (11) ∆𝐸𝑖,𝑡 = 𝛽𝑥,𝑡 ∙ 𝛾𝑡 ∙ ∆𝑥𝑖,𝑡∈�𝑟𝑖𝑠𝑖𝑠 + 𝛽𝐼𝑥,𝑡 ∙ 𝛾𝑡 ∙ ∆𝐼𝑖,𝑡∈�𝑟𝑖𝑠𝑖𝑠 + ∆𝜀𝑖,𝑡 , 𝑥 (12) 𝑤𝑖,𝑡 = 𝛽𝑥,𝑡 ∙ 𝛾𝑡 ∙ ∆𝑥𝑖,𝑡∈�𝑟𝑖𝑠𝑖𝑠 + 𝛽𝐼𝑥,𝑡 ∙ 𝛾𝑡 ∙ ∆𝐼𝑖,𝑡∈�𝑟𝑖𝑠𝑖𝑠 + ∆𝜀𝑖,𝑡 , and 𝑥 (13) 𝑑𝑖𝑓𝑓𝑖,𝑡 = 𝛽𝑥,𝑡 ∙ 𝛾𝑡 ∙ ∆𝑥𝑖,𝑡∈�𝑟𝑖𝑠𝑖𝑠 + 𝛽𝐼𝑥,𝑡 ∙ 𝛾𝑡 ∙ ∆𝐼𝑖,𝑡∈�𝑟𝑖𝑠𝑖𝑠 + ∆𝜀𝑖,𝑡 . Equations 11-13 are estimated using the 99 industries for which both the trade and the labor market data exist for every observation in the sample, (1,089 observations). In a nutshell, these models provide estimates of crisis-specific deviations from common period-specific shocks across industries. The parameters of interest vary over time. If the export shocks observed during the crisis period did not affect ongoing trends across all industries then 𝛽𝑥,𝑡 = 0 and 𝛽𝐼𝑥,𝑡 = 0 across the whole sample period. If the crisis trade collapse induced inter- industry dispersion in employment and wages that deviated from common trends, however, then 𝛽𝑥,𝑡 > 0 or 𝛽𝐼𝑥,𝑡 > 0 during 𝑡 ∈ �𝑟𝑖𝑠𝑖𝑠. In other words, we should observe positive and significant coefficients in (11), (12), or (13) for the quarters of the crisis. If the crisis-induced 28 drop in exports caused a shedding of low-skilled workers we would expect positive coefficients at the end of 2008 and negative coefficients later in 2009 as exports recovered. The results from the estimations of models (11), (12), and (13) are summarized in figures 12, 13, and 14 respectively for the specifications that define the crisis shock as being from October 2008 to November 2008. We also present the underlying regression results in Table 6. 11 In Figure 12, industries less severely affected by export shocks from October 2008 to November 2008 (higher values for ∆𝑥𝑖,𝑡∈�𝑟𝑖𝑠𝑖𝑠 ) experienced better employment in the fourth quarter of 2008 (this result is significant at the 0.05 level). The fact that the coefficients for the direct export shock move around substantially both before and after the shock makes this coefficient, however, somewhat suspect. The results on the shock to related industries, in contrast, are much more convincing. There is strong evidence that when related industries received less severe export shocks, employment growth was much stronger in the first quarter of 2009 (significant at the 0.01 level). More severe shocks from October 2008 to November 2008 to related industries were associated with stronger recoveries in the fourth quarter of 2009 (significant at the 0.05 level). No other coefficients for the shock to related industries are significant at conventional levels. Turning now to the results for wages in Figure 13, we observe patterns similar to those we observed for employment. Industries with less severe export shocks from October 2008 to November 2008 (higher values of ∆𝑥𝑖,𝑡∈�𝑟𝑖𝑠𝑖𝑠 , that is, negative values that are smaller in absolute value) experienced higher wage growth in the first quarter of 2009 (this result is significant at the 0.05 level), but these gains were short lived as they experienced lower wage growth in the third quarter of 2009 (significant at the 0.01 level). Likewise, wages in industries that faced less severe 11 The analogous results using imports are similar to those using exports but with weaker statistical significance. 29 shocks to related industries grew faster in the first quarter of 2009 (significant at the 0.01 level), but, again, these wage gains were reversed in the third and fourth quarters of 2009 (both significant at the 0.10 level). It is also worth noting that all of the coefficients from the wage regressions from the continuous treatment models are substantially smaller in magnitude than in the corresponding employment regressions. These results provide further evidence that labor markets reacted to the trade shock mainly by reducing employment, although wages also fell. These results are consistent with either relatively low employment adjustment costs or a relatively elastic labor supply curve faced by "small" (wage-taking) firms in the short run. Finally, turning to the results on the difference between the change in average log wages for stayers and the change in average log wages for all workers, Figure 14 presents the results using 𝑑𝑖𝑓𝑓𝑖,𝑡 as the dependent variable. Sectors that were particularly hard hit by the direct export shock disproportionately shed low-skilled workers in the fourth quarter of 2008, that is, very quickly after the shock occurred. Industries that were particularly hard hit by shocks to related exports disproportionately shed low-skilled workers in the fourth quarter of 2008 and the first quarter of 2009, then brought them back at the end of 2009, consistent with relative adjustment costs (of unskilled versus skilled workers) playing a significant role in hiring and firing decisions. 6. Conclusions As developing countries become increasingly integrated into the world economy, understanding how international short-run shocks spread through domestic labor markets becomes increasingly important—especially in the context of vertical integration, labor market 30 adjustment costs, and dynamic adjustment. This paper exploits the exogenous variation in Mexico’s trade with the United States to study the employment and wage effects of trade shocks with unique data on formal labor markets from Northern Mexico. The data from social security records allow tracking individual workers across industries, which is critical for estimating the effects of trade on employment and wages whilst allowing for such effects to operate through labor mobility across industries. In addition, the data allow for a careful matching of the data on labor by industries to bilateral trade data from U.S. customs records. This combination of trade and employment data results in a quarterly dataset of employment and wages that permits the estimation of labor-market models with leads and lags around the time of the Great Trade Collapse. The econometric results revealed some interesting and novel patterns. First, imports appear to be complements to labor in Northern Mexico, which is consistent with outsourcing patterns whereby Northern Mexico is a processing stage in North American supply chains. We wonder whether the bulk of the empirical literature on trade and labor (and even the literature on trade and productivity) to some extent has confounded the import-competing and imported- inputs effects in models that utilize industrial classifications at medium levels of aggregation, which could partially explain the largely small estimated effects of trade that have been reported in the literature. Second, a significant portion of hiring decisions tends to occur prior to the realization of exports. Hiring and firing decisions seem to be more important than wage setting because most of the adjustment to trade shocks in Northern Mexico, including during the trade collapse at the end of 2008 and early 2009, seems to have taken place through adjustments in employment, 31 much more so than through wages. Wages were probably affected, especially during the crisis period, but the estimated elasticities were significantly smaller than those of employment. Third, we examined the employment-composition effects of the crisis. We found evidence suggesting that short-run positive shocks are associated with skill downgrading of the employed labor force, while downturns were associated with skill upgrading. Hence, while the wages of workers who did not lose their jobs during the crisis tended to be relatively (compared to employment) insensitive to the trade shock, the composition of the employed labor force changed significantly and to a larger extent than the wages of stayers. This result, which is consistent with a longstanding labor literature, suggests that short-run shocks can have significant effects not only on the level of employment within industries, but also on the composition of the employed labor force. This result therefore highlights the importance of considering timeframe in new dynamic trade models with search or labor-market adjustment costs. Fourth, the results of our measure of industry “relatedness� suggest that related-industry trade shocks appear to be significant determinants of employment, with the economic magnitude of these shocks being just as large as those from own-industry trade shocks. These results are consistent with the idea that Mexican industries are related through output, rather than input (labor) markets. We find that, controlling for general shocks, relatedness does matter and that the trade shocks spread to other industries proportional to the probability that workers move between industry pairs. The most likely explanation for this is that industries with supply relationships hire similar workers, and contrasts with the idea that workers released from a given industry create hiring opportunities for industries that hire similar workers. 32 7. References Amiti, Mary, and Jozef Konings. 2007. "Trade Liberalization, Intermediate Inputs, and Productivity: Evidence from Indonesia." American Economic Review, 97(5): 1611–1638. Artuc, Erhan; Shubham Chaudhuri; and John McLaren 2007 “Trade Shocks and Labor Adjustment: Theory.� NBER Working Paper 13463 http://www.nber.org/papers/w13463 Artuc, Erhan; Shubham Chaudhuri; and John McLaren 2010 “Trade Shocks and Labor Adjustment: A Structural Empirical Approach� American Economic Review, 100(3) (June): 1008-45. Baldwin, Richard (ed.) 2009 The Great Trade Collapse: Causes, Consequences and Prospects A VoxEU.org Ebook available at http://voxeu.org/index.php?q=node/4297. Bernard, Andrew; Raymond Robertson; and Peter K. Schott 2010 “Is Mexico a Lumpy Country?� Review of International Economics 18(5) (November): 937–950. Bergin, Paul, Robert Feenstra, and Gordon Hanson. 2009 “Offshoring and Volatility: Evidence from Mexico’s Maquiladora Industry.� American Economic Review 99: 1664-71. Claessens, Stijn; Hui Tong; and Shang-Jin Wei 2011 “From the Financial Crisis to the Real Economy: Using Firm-level Data to Identify Transmission Channels� NBER Working Paper No. 17360 National Bureau of Economic Research. Davidson, Carl; Steven Matusz, and Andrei Shevchenko. 2008 “Globalization and Firm-level Adustment with Imperfect Labor Markets� Journal of International Economics 75(2): 295-309. Davidson, Carl; Fredrik Heyman, Steven Matusz, Fredrik Sjoholm, and Susan Zhu, 2010 "Globalizaton and Imperfect Labor Market Sorting" University of Nottingham Research Paper 2010/30. Dix-Carneiro, Rafael 2010 “Trade Liberalization and Labor Market Dynamics�, mimeo, Princeton University. Eaton, Jonathan, Sam Kortum, Brent Neiman, and John Romalis. 2009 “Trade and the Global Recession.� Unpublished working paper. Ebenstein, Avraham, Ann Harrison, Margaret McMillan, and Shannon Phillips. 2009 “Estimating the Effect of Trade and Offshoring on American Workers Using the Current Population Surveys.� NBER Working Paper No. 15107, Cambridge, MA. Feenstra, Robert and Gordon H. Hanson.1997 “Foreign Direct Investment and Relative Wages: Evidence from Mexico’s Maquiladoras� Journal of International Economics 42: 371- 393. Feenstra, R.C., 2010. Offshoring in the Global Economy: Microeconomic Structure and Macroeconomic Implications, MIT Press, Cambridge, Massachusetts. Felbermayr, Gabriel, Julien Prat, Hans-Jörg Schmerer, 2011 “Globalization and Labor Market Outcomes: Wage Bargaining, Search Frictions, and Firm Heterogeneity� Journal of Economic Theory, 146(1) (January): 39-73. 33 Gautier, Peter; Gerald van den Berg; Jan van Ours; and Geert Ridder. 2002 “Worker Turnover at the Firm Level and Crowding Out of Lower Educated Workers.� European Economic Review 46: 523-38. Hamermesh, Daniel S. 1993 Labor Demand. Princeton, N.J.: Princeton University Press. Helpman, Elhanan. 2010 “Labor Market Frictions as a Source of Comparative Advantage, with Implications for Unemployment and Inequality� NBER Working Paper 15764. http://www.nber.org/papers/w15764. Helpman, Elhanan, Oleg Itskhoki, and Stephen Redding 2010 "Inequality and Unemployment in a Global Economy" Econometrica 78(4) (July): 1239-1283. Hoekman, Bernard and Alan Winters. 2005 “Trade and Employment: Stylized Facts and Research Findings.� World Bank Policy Research Working Paper #3676, Washington, DC. Hummels, David; Rasmus Jørgensen; Jakob R. Munch; and Chong Xiang (2011) “The Wage Effects of Offshoring: Evidence from Danish Matched Worker-Firm Data� NBER Working Paper No. 17496 October. Jansen, Marion and Erik von Uexkull. 2010 Trade and Employment in the Global Crisis International Labour Office (Geneva) and Academic Foundation (New Delhi). Kletzer, Lori (2001) Job Loss from Imports: Measuring the Costs Peterson Institute for International Economics, Washington D.C. Reder, Melvin. 1955 “The Theory of Occupational Wage Differentials.� American Economic Review 45(5): 833-50. Richardson, J. David. 1995 "Income Inequality and Trade: How to Think, What to Conclude" Journal of Economic Perspectives 9(3) (Summer): 33-55. Robertson, Raymond. 2009 " Mexico and the Great Trade Collapse" in Baldwin, Richard (ed.) The Great Trade Collapse: Causes, Consequences and Prospects A VoxEU.org Ebook available at http://voxeu.org/index.php?q=node/4297. Robertson, Raymond; Brown, Drusilla; Pierre, Gaëlle; Sanchez-Puerta, Laura (eds.). 2009 Globalization, Wages, and the Quality of Jobs Five Country Studies, The World Bank, Washington, D.C. Robertson, Raymond and Donald H. Dutkowsky, 2002 “Labor Adjustment Costs in a Destination Country: The Case of Mexico� Journal of Development Economics February, 67(1): 29-54. 34 Figure 1. Total Mexican Exports and Imports Billions 2009 U.S. Dollars Total Exports Billions $09 Total Imports Billions $09 27 24 Billions 2009 US$ 21 18 15 2001 2005 2010 Year Notes: Exports and Imports are in billions of U.S. dollars deflated by the U.S. Consumer Price Index for all urban consumers (series CUUR0000AA0) in which 2009=100. Totals include petroleum trade. Data are from http://dgcnesyp.inegi.gob.mx/cgi-win/bdieintsi.exe/Consultar#. 35 Figure 2. United States Share of Mexican Non-petroleum Exports and Imports US Share Mexican Imports US Share Mexican Exports .9 .8 .7 .6 .5 .4 2001 2005 2010 Year Notes: Series calculated by authors using data from http://dgcnesyp.inegi.gob.mx/cgi- win/bdieintsi.exe/Consultar#. 36 Figure 3a. Change in Mexican Exports to the U.S. by 6-digit HS Industry .167563 Fraction 0 -7 -5 -2 -1 0 1 3 5 7 Log Difference Notes: Mean (standard deviation) log difference is -0.419 (1.138). Difference is calculated as the difference in the log of U.S. imports from Mexico between April 2008 (peak) and January 2009 (trough). Difference shown only represents the intensive margin (HS6 categories that had positive trade values in both periods). Normal distribution is superimposed over the histogram. Figure 3b. Change in Mexican Imports from the U.S. by 6-digit HS Industry .167563 Fraction 0 -7 -5 -2 -1 0 1 3 5 7 Log Difference Notes: Mean (standard deviation) log difference is -0.166 (1.182). Difference is calculated as the difference in the log of U.S. exports to Mexico between April 2008 (peak) and January 2009 (trough). Difference shown only represents the intensive margin (HS6 categories that had positive trade values in both periods). Normal distribution is superimposed over the histogram. 37 Figure 4. Employment in Tradables and Exports to the United States, 2007-2009 Employment and Exports to the US (Border States: Tradable Sectors Only) 40000 45000 50000 55000 60000 1250 1300 1350 1400 1450 1500 Exports to US (1,000,000s) Total Empl (1,000s) 2007:1 2007:2 2007:3 2007:4 2008:1 2008:2 2008:3 2008:4 2009:1 2009:2 2009:3 2009:4 time Total Empl (1,000s) Exports to US (1,000,000s) Figure 5. Employment in Non-Tradables and Exports to the United States, 2007-2009 Employment and Exports to the US (Border States: Non-Tradable Sectors Only) 40000 45000 50000 55000 60000 2600 Exports to US (1,000,000s) Total Empl (1,000s) 2400 2007:1 2007:2 2007:3 2007:4 2008:1 2008:2 2008:3 2008:4 2009:1 2009:2 2009:3 2009:4 time Total Empl (1,000s) Exports to US (1,000,000s) 38 Figure 6. Net Percent Change in Employment in Tradables and Non-Tradables, 2007-2009 Net Percent Change in Employment (Border States: Tradables vs. Non-Tradables) 5 net % change in empl -5 -10 0 2007:2 2007:3 2007:4 2008:1 2008:2 2008:3 2008:4 2009:1 2009:2 2009:3 2009:4 time Tradables Non-Tradables 39 Figure 7a. Average Change in Log Real Wages for Stayers in Tradeables and Non- Tradables, 2007-2009 Average Change in Log Real Wages (workers who have not changed firms) .04 change in log wage 0 -.02.02 2007:2 2007:3 2007:4 2008:1 2008:2 2008:3 2008:4 2009:1 2009:2 2009:3 2009:4 qtr tradeable (stayers) non-tradeable (stayers) Figure 7b. Average Change in Log Real Wages in Tradeables and Non-Tradables, 2007- 2009 Average Change in Log Real Wages (all workers) .02 change in log wage -.01 0 -.02 .01 2007:2 2007:3 2007:4 2008:1 2008:2 2008:3 2008:4 2009:1 2009:2 2009:3 2009:4 qtr tradeable non-tradeable 40 Figure 8. Change in Average Log Wage for Stayers Minus the Change in Average Log Wage for All Workers, 2007-2009 Difference in Average Change in Log Real Wages (Stayers Minus All Workers) .04 differntial change in log wage -.02 0 .02 2007:2 2007:3 2007:4 2008:1 2008:2 2008:3 2008:4 2009:1 2009:2 2009:3 2009:4 qtr tradeable non-tradeable 41 Figure 9a. Changes in Log Employment for Tradable Industries Grouped by the Magnitude of the Direct Export Shock, 2007-2009 Average Change in Log Employment (industries grouped by size of shock to exports) .1 .05 0 -.05 -.1 2007:2 2007:3 2007:4 2008:1 2008:2 2008:3 2008:4 2009:1 2009:2 2009:3 2009:4 time Ave change in log empl in sectors with small shocks Ave change in log empl in sectors with medium shocks Ave change in log empl in sectors with large shocks Note: groups defined by % drop in ln(exports) from Oct - Nov 2008 Figure 9b. Changes in Log Employment for Tradable Industries Grouped by the Magnitude of the Related Export Shock, 2007-2009 Average Change in Log Employment (industries grouped by size of shock to exports in related industries) .05 0 -.05 -.1 2007:2 2007:3 2007:4 2008:1 2008:2 2008:3 2008:4 2009:1 2009:2 2009:3 2009:4 time Ave change in log empl in sectors with small related shocks Ave change in log empl in sectors with medium related shocks Ave change in log empl in sectors with large related shocks Note: groups defined by % drop in ln(exports) in related inds from Oct - Nov 2008 42 Figure 10a. Changes in Log Wages for Stayers in Tradable Industries Grouped by the Magnitude of the Direct Export Shock, 2007-2009 Average Change in Log Nominal Wage for Stayers (industries grouped by size of shock to exports) 0 .01 .02 .03 .04 .05 2007:2 2007:3 2007:4 2008:1 2008:2 2008:3 2008:4 2009:1 2009:2 2009:3 2009:4 time Ave change in log wage in sectors with small shocks Ave change in log wage in sectors with medium shocks Ave change in log wage in sectors with large shocks Note: groups defined by % drop in ln(exports) from Oct - Nov 2008 Figure 10b. Changes in Log Wages for Stayers in Tradable Industries Grouped by the Magnitude of the Related Export Shock, 2007-2009 Average Change in Log Nominal Wage for Stayers (industries grouped by size of shock to exports in related industries) .06 .04 .02 0 -.02 2007:2 2007:3 2007:4 2008:1 2008:2 2008:3 2008:4 2009:1 2009:2 2009:3 2009:4 time Ave change in log wage in sectors with small related shocks Ave change in log wage in sectors with medium related shocks Ave change in log wage in sectors with large related shocks Note: groups defined by % drop in ln(exports) in related inds from Oct - Nov 2008 43 Figure 11a. Differences between the Change in Average Log Wages for Stayers and the Change in Average Log Wages for All Workers in Tradable Industries Grouped by the Magnitude of the Direct Export Shock, 2007-2009 Differential Change in Log Nominal Wage (Stayers Minus All) .06 .04 .02 0 -.02 (industries grouped by size of shock to exports) 2007:2 2007:3 2007:4 2008:1 2008:2 2008:3 2008:4 2009:1 2009:2 2009:3 2009:4 time Diff log wage change (stayers minus all, small shocks) Diff log wage change (stayers minus all, medium shocks) Diff log wage change (stayers minus all, large shocks) Note: groups defined by % drop in ln(exports) from Oct - Nov 2008 Figure 11b. Differences between the Change in Average Log Wages for Stayers and the Change in Average Log Wages for All Workers in Tradable Industries Grouped by the Magnitude of the Related Export Shock, q12007-q42009 Differential Change in Log Nominal Wage (Stayers Minus All) (industries grouped by size of shock to exports in related industries) .06 .04 .02 0 -.02 2007:2 2007:3 2007:4 2008:1 2008:2 2008:3 2008:4 2009:1 2009:2 2009:3 2009:4 time Diff log wage change (stayers minus all, small related shocks) Diff log wage change (stayers minus all, medium related shocks) Diff log wage change (stayers minus all, large related shocks) Note: groups defined by % drop in ln(exports) in related inds from Oct - Nov 2008 44 Figure 12. Continuous Treatment Effects of Exports and Related Exports on Employment Coefficients on Direct Exports and Related Exports Dependent variable: change in log employment (Shock: October -November 2008) 0.08 1 0.06 0.8 0.04 0.6 0.02 0.4 0.2 0 0 -0.02 -0.2 -0.04 -0.4 -0.06 -0.6 2007:2 2007:3 2007:4 2008:1 2008:2 2008:3 2008:4 2009:1 2009:2 2009:3 2009:4 coef: direct export shock coef: related export shock Figure 13. Continuous Treatment Effects of Exports and Related Exports on Wages of Stayers Coefficients on Direct Exports and Related Exports Dependent variable: change in log wages of stayers (Shock: October -November 2008) 0.03 0.3 0.25 0.02 0.2 0.01 0.15 0.1 0 0.05 0 -0.01 -0.05 -0.02 -0.1 -0.15 -0.03 -0.2 2007:2 2007:3 2007:4 2008:1 2008:2 2008:3 2008:4 2009:1 2009:2 2009:3 2009:4 coef: direct export shock coef: related export shock 45 Figure 14. Continuous Treatment Effects of Exports and Related Exports on the Difference between the Change in Average Log Wage of Stayers and the Change in Average Log Wage of all workers Coefficients on Direct Exports and Related Exports Dep Variable: Diff Change in Log Wage (Stayers Minus All) (Shock: October -November 2008, Wage Regressions) 0.04 0.2 0.03 0.15 0.1 0.02 0.05 0.01 0 -0.05 0 -0.1 -0.01 -0.15 -0.2 -0.02 -0.25 -0.03 -0.3 2007:2 2007:3 2007:4 2008:1 2008:2 2008:3 2008:4 2009:1 2009:2 2009:3 2009:4 coef: direct export shock coef: related export shock 46 Table 1: Maquiladora Employment 1998 Employment Share N/P Employment Ratio State Maquila/Census Census Maquila Baja California Norte 0.868 0.153 0.078 Tamaulipas 0.769 0.239 0.086 Chihuahua 0.742 0.152 0.084 Sonora 0.644 0.212 0.065 Queretaro 0.552 0.422 0.083 Coahuila 0.485 0.217 0.056 Durango 0.340 0.17 0.052 Aguascalientes 0.286 0.261 0.041 Yucatan 0.227 0.266 0.055 Baja California Sur 0.226 0.319 0.031 Zacatecas 0.154 0.326 0.070 Nuevo Leon 0.142 0.285 0.090 Tlaxcala 0.103 0.243 0.068 Puebla 0.101 0.198 0.047 Jalisco 0.087 0.323 0.126 San Luis Potosi 0.073 0.308 0.027 Guerrero 0.060 0.282 0.022 Guanajuato 0.048 0.192 0.051 Morelos 0.023 0.348 0.092 Sinaloa 0.022 0.401 0.148 Mexico State 0.020 0.352 0.121 Hidalgo 0.008 0.186 0.069 Distrito Federal 0.004 0.506 0.108 National Average 0.242 0.293 0.073 Notes: Maquilas include services as well as manufacturing. In 1998, and over the 1990-2003 period, services average 4% of total maquila employment. The employment ratio is the non- production/production worker ratio. INEGI does not report data for all states, and we presume this reflects an insignificant number of maquiladoras and therefore enter "0" for these states. The Mexican states of Campeche, Colima, Michoacan, Nayarit, Oaxaca, Quintana Roo, Tabasco, and Veracruz have maquila/census employment shares of 0.000 and are not included in the table. The “Average� includes all 31 states and the Federal District. 47 Table 2: Descriptive Statistics Std. Obs. Mean Dev. Min. Max. Change in log exports 1,140 -0.007 0.250 -3.969 2.662 Change in log imports 1,144 0.003 0.227 -3.248 3.356 Relatedness index for imports 1,144 -0.001 0.042 -0.527 0.435 Relatedness index for exports 1,144 0.001 0.043 -0.545 0.349 Average change in log nominal wage 1,148 0.023 0.025 -0.349 0.433 Change in log employment 1,147 -0.013 0.066 -1.114 0.630 Note: All statistics are calculated using the industries of employment in the first quarter of 2007 as its weight. 48 Table 3. WLS Estimates of the Employment Equation (weight= employment by industry in 2007) Explanatory Variables: (1) (2) (3) (4) (5) (6) Dependent Variable: Change in log employment Imports from the U.S. 0.020* 0.014 0.014 0.017* 0.019* 0.017* (0.011) (0.009) (0.009) (0.009) (0.010) (0.009) Exports to the U.S. 0.021 0.034** 0.030* 0.025 (0.015) (0.016) (0.017) (0.017) One-period ahead: Imports from the U.S. -0.020* -0.020* (0.011) (0.011) One-period ahead: Exports to the U.S. 0.039*** 0.035** 0.032** 0.027* (0.015) (0.014) (0.014) (0.014) Related imports from the U.S. 0.138** 0.126* 0.039 0.145* (0.070) (0.070) (0.044) (0.080) Related exports to the U.S. 0.151 0.177* 0.015 (0.119) (0.103) (0.107) One-period ahead: Related imports from the U.S. -0.144 (0.109) One-period ahead: Related exports to the 0.180 0.172 U.S. (0.153) (0.124) Crisis dummy * Imports from the U.S. -0.021 (0.036) One-period ahead: Crisis dummy * Exports to the U.S. 0.001 (0.016) Crisis dummy * Related imports from the U.S. 0.417 (0.413) Crisis dummy * Related exports to the U.S. 0.451*** (0.159) Observations 1,132 1,027 1,025 1,027 1,027 1,027 R-squared 0.257 0.294 0.309 0.288 0.303 0.287 Robust standard errors in parentheses; errors are clustered by industry. *** p<0.01, ** p<0.05, * p<0.1 Time effects are included but not reported. Crisis dummy equals one during 2008q4 and 2009q1. 49 Table 4. WLS Estimates of the Wage Equation (weight= employment by industry in 2007) Explanatory Variables: (1) (2) (3) (4) (5) Dependent Variable: Change in log wages of stayers Imports from the U.S. 0.008* 0.006 0.006 0.007** (0.004) (0.004) (0.004) (0.003) Exports to the U.S. 0.010** 0.011*** 0.007* 0.007** (0.004) (0.004) (0.004) (0.003) One-period ahead: Imports from the U.S. -0.002 -0.003 (0.004) (0.003) One-period ahead: Exports to the U.S. 0.003 0.004 0.001 (0.003) (0.004) (0.004) Related imports from the U.S. 0.001 0.018 (0.030) (0.035) Related exports to the U.S. 0.140 (0.091) One-period ahead: Related imports from the U.S. 0.049* (0.027) One-period ahead: Related exports to the U.S. -0.002 0.003 (0.036) (0.009) Crisis * Exports to the U.S. 0.012 (0.009) Observations 1,133 1,028 1,025 1,133 1,027 R-squared 0.553 0.554 0.571 0.552 0.548 Robust standard errors in parentheses; errors are clustered by industry. *** p<0.01, ** p<0.05, * p<0.1 Time effects are included but not reported. Crisis dummy equals one during 2008q4 and 2009q1. 50 Table 5. WLS Estimates of Relative Wage Growth of Stayers (weight=employment by industry in 2007) Explanatory Variables: (1) (2) (3) (4) (5) Dependent Variable: Change in log wages of stayers minus change of the average Imports from the U.S. 0.008** 0.006* 0.006* 0.008** 0.008** (0.004) (0.003) (0.003) (0.004) (0.004) Exports to the U.S. 0.021*** 0.028*** 0.024*** 0.033*** (0.007) (0.007) (0.007) (0.007) One-period ahead: Imports from the U.S. 0.000 0.000 (0.004) (0.004) One-period ahead: Exports to the U.S. 0.016*** 0.015*** 0.015*** 0.009 (0.004) (0.004) (0.004) (0.005) Related imports from the U.S. 0.065** 0.045* 0.078* (0.033) (0.026) (0.043) Related exports to the U.S. 0.110*** 0.075* (0.040) (0.040) One-period ahead: Related imports from the U.S. -0.037 (0.041) One-period ahead: Related exports to the U.S. 0.041 0.044 (0.059) (0.048) Crisis * Exports to the U.S. -0.030*** (0.009) Crisis * Imports from the U.S. -0.004 (0.011) One-period ahead: Crisis * Exports to the U.S. 0.011** (0.005) Crisis * Related imports from the U.S. 0.058 (0.104) Crisis * Related exports to the U.S. 0.118* (0.071) Observations 1,132 1,027 1,025 1,025 R-squared 0.388 0.426 0.444 0.455 Robust standard errors in parentheses; errors are clustered by industry. *** p<0.01, ** p<0.05, * p<0.1 Time effects are included but not reported. Crisis dummy equals one during 2008q4 and 2009q1. 51 Table 6: WLS Estimates of Continuous Treatment Effects of Exports and Related Exports on Employment and Wages Change in log employment Change in log wage for stayers Differential wage change Export shock * 2007:2 -0.01 -0.01 -0.01 -0.01 -0.02** -0.02* (0.016) (0.016) (0.006) (0.007) (0.010) (0.010) Export shock * 2007:3 -0.02 -0.02 -0.02** -0.02* -0.02* -0.02 (0.017) (0.020) (0.008) (0.008) (0.010) (0.011) Export shock * 2007:4 0.00 0.01 -0.00 -0.00 0.01 0.01 (0.020) (0.028) (0.009) (0.010) (0.006) (0.008) Export shock * 2008:1 0.06 0.04 -0.00 0.01 0.00 0.00 (0.041) (0.038) (0.007) (0.007) (0.012) (0.012) Export shock * 2008:2 -0.03 -0.05* -0.01 -0.01* -0.01 -0.01 (0.029) (0.029) (0.005) (0.007) (0.013) (0.013) Export shock * 2008:3 0.03 0.01 -0.01 -0.01 -0.01 -0.01 (0.035) (0.029) (0.011) (0.011) (0.008) (0.008) Export shock * 2008:4 0.08*** 0.06** 0.02 0.02 0.04*** 0.03*** (0.021) (0.025) (0.014) (0.014) (0.010) (0.011) Export shock * 2009:1 0.10 0.04 0.04*** 0.03** 0.00 -0.01 (0.067) (0.057) (0.010) (0.012) (0.006) (0.010) Export shock * 2008:2 -0.05 -0.05 -0.01 -0.00 -0.01** -0.01 (0.045) (0.039) (0.005) (0.006) (0.005) (0.007) Export shock * 2009:3 0.03 0.04* -0.03*** -0.03*** -0.03 -0.02 (0.026) (0.026) (0.008) (0.009) (0.022) (0.019) Export shock * 2009:4 0.01 0.04 -0.01 0.00 0.00 0.02* (0.024) (0.025) (0.007) (0.007) (0.008) (0.009) Rel.exp. shock * 2007:2 -0.06 -0.04 -0.03 -0.02 -0.10* -0.06 (0.082) (0.082) (0.045) (0.048) (0.060) (0.048) Rel.exp. shock * 2007:3 0.00 0.05 -0.05 -0.01 -0.07 -0.03 (0.074) (0.080) (0.052) (0.054) (0.059) (0.054) Rel.exp. shock * 2007:4 -0.08 -0.10 -0.03 -0.03 -0.01 -0.03 (0.091) (0.133) (0.038) (0.039) (0.033) (0.041) Rel.exp. shock * 2008:1 0.29 0.20 -0.08** -0.09** 0.03 0.02 (0.203) (0.162) (0.037) (0.036) (0.058) (0.052) Rel.exp. shock * 2008:2 0.10 0.21 0.05 0.08* 0.01 0.03 (0.175) (0.172) (0.047) (0.044) (0.063) (0.055) Rel.exp. shock * 2008:3 0.30 0.28 -0.03 0.01 -0.04 -0.02 (0.218) (0.212) (0.066) (0.072) (0.049) (0.048) Rel.exp. shock * 2008:4 0.31* 0.17 0.03 -0.00 0.18** 0.12 (0.180) (0.177) (0.033) (0.030) (0.081) (0.070) Rel.exp. shock * 2009:1 1.00*** 0.92*** 0.29*** 0.24*** 0.14 0.16 (0.368) (0.337) (0.082) (0.078) (0.086) (0.106) Rel.exp. shock * 2008:2 -0.14 -0.04 -0.05 -0.04 -0.07** -0.05 (0.232) (0.179) (0.038) (0.043) (0.035) (0.039) Rel.exp. shock * 2009:3 -0.08 -0.18 -0.17** -0.11* -0.27* -0.23 (0.167) (0.172) (0.066) (0.066) (0.146) (0.144) Rel.exp. shock * 2009:4 -0.33** -0.42** -0.15** -0.15* -0.12* -0.16** (0.136) (0.160) (0.070) (0.082) (0.064) (0.080) Robust standard errors in parentheses with 1,089 observations. *** p<0.001, ** p<0.01, * p<0.05. Trade shock calculated from October 2008 to November 2008. 52