From the Invisible Handshake to the Invisible Hand? How Import Competition Changes the Employment Relationship

Transcription

1 From the Invisible Handshake to the Invisible Hand? How Import Competition Changes the Employment Relationship Marianne Bertrand, University of Chiago, Center for Eonomi Poliy Researh, and National Bureau of Eonomi Researh Does import ompetition alter the extent to whih employers, after negotiating workers wages upon hire, subsequently shield those wages from external labor-market onditions? If inreased ompetition indues a swith away from these wage impliit agreements, then (1) the sensitivity of workers wages to the urrent unemployment rate should inrease as ompetition inreases and (2) the sensitivity of workers wages to the unemployment rate prevailing upon hire should derease. Using exhange-rate movements to generate exogenous variation in import ompetition, I find evidene supporting both of these preditions. I show that inreased finanial pressures on employers is one mehanism driving these effets. I am espeially grateful to Larry Katz, Caroline Minter-Hoxby, and Sendhil Mullainathan for many omments and helpful disussions. I have also benefited from the omments of George Baker, Olivier Blanhard, Ed Glaeser, John Horn, Mihael Kremer, Erzo Luttmer, Canie Prendergast, Andrei Shleifer, Jeff Wurgler, and seminar partiipants at the Harvard Labor Seminar. I thank John Horn and Greg Eastman for their help with the trade data. The researh has been supported by a Sloan Dotoral Dissertation Fellowship. Contat the author at marianne.bertrand@gsb.uhiago.edu. [ Journal of Labor Eonomis, 2004, vol. 22, no. 4] 2004 by The University of Chiago. All rights reserved X/2004/ $

2 724 Bertrand I. Introdution Complete employment ontrats are expensive to design and enfore. As a result, employers and workers often let their relationship be governed by informal agreements. They share an unwritten understanding about pay, hours, work quality, working onditions, job seurity, and other dimensions of employment. 1 To use Okun s (1981) haraterization, the market for labor is governed by an invisible handshake rather than leared by an invisible hand. Aording to the popular press, however, the past deade has witnessed a substantial transformation of the employment relationship in the United States. Multiple episodes of downsizing and an inreased reliane on temporary workers are ommonly ited as evidene of a deline of those informal arrangements between employers and workers. In partiular, it has been argued that inreased ompetitive and finanial pressures in U.S. produt markets, suh as those indued by trade openness, have turned the employment relationship into more of a spotmarket transation. In this artile, I develop an empirial framework to investigate this laim more systematially. 2 The unwritten understanding between employers and workers overs many dimensions. In this researh, I onentrate on impliit arrangements in the wage-setting proess. Under the arrangements I onsider, wages are negotiated when workers enter into a firm and are thereafter shielded from the external labor market. I fous on suh shielding agreements, or ohort effets, in the wage-setting proess for three main reasons. First, they are theoretially relevant. Under fairly general onditions, suh agreements optimally emerge as a way for risk-neutral firms to insure risk-averse workers against ylial flutuations. The resulting risk-sharing wages are sensitive to the external labor-market onditions at the time that the worker is hired (and the agreement negotiated) but do not respond to the urrent state of the labor market. Seond, previous ethnographi and ase studies have validated the relevane of these wage-setting arrangements. For instane, Doeringer and Piore (1971) desribe how ontats between internal and external labor markets in large manufaturing establishments are lim- 1 Union ontrats are an exeption. They are written and an be enfored in a ourt of law. However, they probably do not over all possible ontingenies and are, in that sense, inomplete. In this artile, I distinguish between union and nonunion workers. 2 This researh projet omplements a growing literature that aims to assess diretly evidene of a deline in internal labor markets. Researhers have previously looked at trends in labor-market outomes suh as job tenure (Swinnerton and Wial 1995; Diebold, Neumark, and Polsky 1997; Farber 1997a, 1997b), earnings instability (Gottshalk and Moffitt 1994), and wage dispersion between and within firms (Groshen and Levine 1998). Rather than looking solely at trends in labor-market outomes, I study the ausal link between suh labor-market outomes and hanges that independently ourred in other markets.

3 Import Competition and the Employment Relationship 725 ited to a subset of jobs, or ports of entry. Insiders wages are determined by a set of administrative rules and ustoms that are typially only weakly influened by hanges in the external labor market. Similarly, one of the main findings of Baker, Gibbs, and Holmstrom s (1994a, 1994b) ase study of 20 years of personnel data is the presene of ohort effets in wages. One workers are inside the firm, their wages seem to follow a ommon pattern that is very different from the pattern of wages for new entrants; muh of the wage differenes aross ohorts ome from variation in the starting wages. The third main reason for studying these wage-setting arrangements, besides their theoretial and pratial relevane, is that they an easily be examined empirially. I will regard a low elastiity of urrent wages to the urrent loal labor-market unemployment rate and a high elastiity of urrent wages to the unemployment rate at the time of job start as two symptoms of the use of shielding agreements in the employment relationship. I will ask whether and how inreased ompetitive pressures in the produt market affet both of these wage-unemployment elastiities. 3 Empirially, ompetitive pressures in the produt market an be orrelated with wage-setting praties without having a ausal effet. First, adverse labor ost shoks or hanges in wage-setting praties an themselves diretly affet the ompetitive position of a firm or industry. For example, employers who deide to buffer their workers wages from the vagaries of the external market may fae relatively higher osts during eonomi downturns, whih may weaken their ompetitive position at those times. Also, managers with weak inentives may be more likely both to protet their workers and also to be targeted by ompetitors. These examples justify the need to identify a soure of variation in produt market ompetition that is independent of adverse hanges in labor osts or other firm wage-setting praties. To that purpose, I fous on foreign ompetition as one form of produt market ompetition. I use soure- 3 It is worth stating at this point why my fous is not on the impat of produt market ompetition on wage levels or on job stability and tenure length. Let us onsider wage levels first. One might argue that a redution of wages following an inrease in import penetration is evidene of a hange in the wage-setting institutions that give workers above-market wages. However, a hange in wage levels an also be interpreted under onstant wage-setting rules. For example, import penetration redues the produt demand faed by domesti firms and shifts down the demand for labor. If the wage-setting urve is not perfetly elasti, the shift down of the labor demand urve will redue wages. Hene, import penetration an redue wages without any hange in wage setting. A similar argument justifies why I do not look at job stability and tenure length as dependent variables either. Any evidene that job stability and tenure length deline with ompetition (see Kletzer 1998), while onsistent with a deline in impliit ontrats, ould also result only from a negative labor demand shok.

4 726 Bertrand weighted exhange rate movements to generate exogenous variation in the level of import penetration by industry. Using merged labor, trade, and orporate data from the period , I find evidene onsistent with a weakening of informal wage shielding agreements in more ompetitive environments. I show that an inrease in import penetration inreases the elastiity of workers urrent wages to the urrent loal unemployment rate. I also show that an inrease in import penetration redues the elastiity of urrent wages to the unemployment rate that was prevailing at the time of job start. I verify the robustness of these results to various speifiation assumptions and sample seletion rules. I then study a speifi mehanism by whih import penetration may weaken the use of wage-shielding agreements. Wage-shielding agreements require that firms an ommit to insure their workers even at times where they ould lower their labor ost by paying a spot-market wage. A standard way to ahieve suh ommitment is through reputational onerns. Firms will not opportunistially renege on impliit ontratual agreements with their workers beause they may suffer higher labor osts in the future if they do so. I suggest that firms ability to ommit to impliit wageshielding ontrats dereases in more ompetitive environments. An inrease in import penetration redues firms earnings and inreases the risk of finanial distress. An inreased risk of finanial distress shortens orporate horizons and therefore weakens reputational onerns: the future ost of opportunistially reneging on impliit agreements goes down when firms fae shorter horizons. To test for the relevane of this mehanism, I introdue measures of orporate performane and apital struture as potential explanatory variables. I first show that delines in industry orporate returns indued by exhange rate movements inrease the sensitivity of wages to the urrent unemployment rate. Seond, I use industry leverage as a proxy for finanial distress. I show that inreases in import penetration redue the sheltering of workers wages more in highly leveraged industries. 4 The rest of the artile is organized as follows: in Setion II, I desribe in a simple illustrative model the finanial distress hannel by whih import penetration an alter the use of wage-shielding agreements. In Setion III, I present the empirial methodology and disuss the identifiation strategy. Setion IV reviews the different data soures. I present and in- 4 The leverage and orporate returns results ould also be reoniled with more behavioral models. First, workers may onsider it fair for firms that are experiening finanial pressure to pay the spot-market wages (Kahneman, Knetsh, and Thaler 1986). Seond, managers may have a preferene for honoring wagesetting arrangements but see their disretion to do so redued with higher levels of ompetition in the produt market (Shleifer and Summers 1988). I disuss these alternative models in Se. V.E.

5 Import Competition and the Employment Relationship 727 terpret the results in Setion V. I first study the ausal link between import penetration and the sheltering of workers wages (Se. V.B). I then investigate the mediating role of orporate attributes suh as finanial leverage and orporate earnings (Ses. V.C V.E). Setion VI summarizes and offers onluding remarks. II. An Illustrative Model In Setion II, I outline in a simple partial equilibrium model one mehanism by whih ompetitive pressures in general, and import ompetition in partiular, an hange the employment relationship. The model shows how greater import penetration, by reduing orporate earnings and inreasing the risk of finanial distress, weakens the use of impliit wageshielding agreements. The goal of the model is twofold. First, while merely illustrative, the model should larify and oordinate thoughts about various theoretial issues. Seond, and most important, the model provides a basi framework for the empirial analysis that follows. A. Setup Consider a firm having aess to a tehnology that requires one worker. Labor is the only input used in prodution. The firm is risk-neutral and has an infinite horizon. Firm s profit in period t is defined as p t(r t, w t), where Rt is revenue and wt is the wage paid to the worker. The firm s objetive at time t is to maximize the net present value of expeted future profits. Workers are risk-averse and live for one period. 5 Workers utility in period t is V(w t), with V 1 0 and V! 0. Workers reservation wage rt is subjet to random produtivity shoks and follows an AR(1) proess: rt p (1 r)r* rrt 1 ht, where r* is the long-run equilibrium labor pro- dutivity, 0! r! 1, and ht is equal to h 1 0 with probability.5 and to h with probability.5. When modeling the determination of the wage w t in the firm, I assume that expliit wage ontrats between the firm and the worker annot be implemented. Instead, the firm and the worker an adopt one of two polar wage-setting rules. First, the firm an pay the worker a spot-market wage s w t that makes the worker indifferent between working at the firm and working at his next best outside option. Assuming that the worker has s no bargaining power when negotiating with the firm, w t is simply equal to the worker s reservation wage in the external labor market: r t s wt p r t. (1) 5 The model ould be extended to allow workers to live for more than one period. The basi insight of the model, however, an already be aptured under the simplifying assumption of a one-period life.

6 728 Bertrand s s Thus, wt flutuates one for one with rt. A higher reservation wage rt inreases the spot-market wage w s. 6 t Alternatively, the firm and the worker may sign an impliit wage ontrat. The firm may offer a ontratual wage w t that protets the worker against random shoks to r t. More speifially, at the time of hire and before the shok h t to the reservation wage is realized, the firm and the worker an voluntarily agree to shift all labor-market risks to the riskneutral firm. 7 Workers expeted utility under the ontratual wage will be pushed down to their expeted utility under spot-market wage: s E t 1[V(w t)] p E t 1[V(w t )]. The optimal ontrat in this ase onsists in paying the risk-averse worker a onstant wage w t that is independent of the realization of. Suh a wage is impliitly given by r t V[(1 r)r* rrt 1 h] V[(1 r)r* rrt 1 h] V(w t ) p. (2) 2 Note that while wt is independent of the realization of rt, it does depend on the reservation wage at the time that the impliit ontrat is signed, r. Suh a ontratual wage w t 1 t mathes losely the ohort effets do- umented by Baker et al. (1994a, 1994b) as well as the idea of ports of entry disussed in Doeringer and Piore (1971). I assume that workers fae mobility osts that make them unwilling to swith jobs after having ontrated with a firm. 8 Firms, by ontrast, may s have an inentive to renege on the ontratual wage whenever wt 1 wt. I assume that the ontratual wage w t is enfored through reputational onerns. More speifially, I assume that if the firm reneges on w t, it loses reputation with future generations of workers and has to pay the spot-market wage s forever after. 9 w t 6 The results diretly generalize to a ase where the worker has stritly positive bargaining power, and the firm and the worker Nash-bargain over the spot-market s wage w t (Bertrand 1998). 7 Alternatively, I ould allow for a ontinuum of partial insurane ontrats and study how import ompetition affets the optimal level of insurane. For simpliity, I hoose to onsider only the polar opposite ases of no insurane or full insurane. 8 See Harris and Holmstrom (1982) and Beaudry and DiNardo (1991) for models where workers are ostlessly mobile between firms. I do test a version of the model in whih workers are not perfetly immobile and report these results in n. 35. The main onlusion of the artile is unaffeted under this alternative model. 9 The importane of the reputation mehanism in the enforement of impliit ontrats is a well-known point. So is its appliation to the employment relationship. Akerlof (1980, 1982) suggests that firms invest in their long-term reputation to generate produtive high morale among employees. Espinoza and Rhee (1989) develop a repeated game version of the monopoly union bargaining model. They show that if unions and firms have enough to gain from future interations, ooperation is possible, and the dynami version of the monopoly union model need not be as ineffiient as the stati version. Other well-known appliations inlude Bull (1987) and Kreps (1990).

7 Import Competition and the Employment Relationship 729 The firm is also exposed to revenue shoks every period. Corporate revenue Rt an be either low ( RL), with probability n, or high ( RH), with probability (1 n). One ould interpret a higher value of n as an inrease in import penetration in the firm s industry or as an appreiation of the dollar. More generally, a higher value of n will proxy for an inrease in ompetitive pressures in the firm s industry. I assume that shoks to revenue and shoks to the reservation wage r t are independent. I also assume, in this partial equilibrium setup, that hanges in n do not affet the reservation wage r t. These assumptions are valid if the firm is small ompared with the relevant external labor market. Clearly, if the firm is not small, a general equilibrium analysis that endogenizes the reservation wage r t would be more appropriate. Suh a general equilibrium analysis would be espeially warranted when trying to derive the aggregate effet of globalization on wage-setting praties. Instead, our goal in this theory setion and the empirial setions that follow is to analyze how, other things equal, a firm (or industry) that experienes an inrease in import penetration may hange its wage-setting praties. Finally, beause I want to highlight the role of finanial distress fators, I onsider the ase where the firm has an outstanding debt D in eah period, resulting from a previous investment. I assume that investors will liquidate the firm if an amount D is not repaid at the end of eah period t. 10 To make the problem interesting, I assume that the parameters of the model are suh that the debt D annot be repaid whenever Rt p RL: p (R, w ) 1 D 1 p (R, w ) t H t t L t The timing of every period t of the game is as follows: (1) a worker is hired by the firm and an impliit wage ontrat is or is not signed. (2) The reservation wage r t is realized. (3) The firm pays the worker: if an impliit wage ontrat is signed in step 1, the firm either pays w t or s s reneges on the ontrat and pays wt; otherwise, the firm pays wt. (4) Prodution takes plae. (5) Corporate revenue R t is realized. (6) The firm pays bak the debt D; the firm is liquidated if D annot be repaid. In the 10 Suh a risk of default derives naturally from an optimal debt ontrat under inomplete ontrat models. If managers an ostlessly divert the firm s ash flow beause the ash flow is not verifiable, investors must threaten to liquidate the firm in order to get repaid. See Hart and Moore (1989) and Bolton and Sharfstein (1996) for a general theoretial treatment. Chevalier and Sharfstein (1996) present an appliation of this theory to explain movements in markups over the business yle. 11 The positive eonomi profits ould result, e.g., from the presene of barriers to entry in the firm s industry. Alternatively, one ould assume that the firm s ability to repay the debt also depends on whether the firm is paying the spotmarket wage or the ontratual wage. An interesting impliation of suh an extension is that an inrease in n may in that ase also inrease the firm s ex ante preferene for wage flexibility (Bertrand 1998).

8 730 Bertrand s next subsetion, I derive the onditions under whih wt or wt will be the equilibrium wage-setting rule in the firm. B. Equilibrium Wage Setting The expeted ash flow stream for the firm at the beginning of eah period t is given by j t j t 1 t H t t j H j jpt 1 (1 n)[p (R, w ) D] d (1 n) {E [p (R,w )] D}, (3) where d is a disount fator, and E t represents expetation onditional on all the information available at the beginning of period t. In other words, when Rt p RL, the debt D annot be repaid, investors seize the remaining profits, and the firm is liquidated. When Rt p RH, the firm pays the debt D at the end of period t and stays in operation. Given expression (3), the firm will always prefer to offer the ontratual wage w t at the beginning of period t. Indeed, the expeted labor ost (ash flow stream) under w t is lower (higher) than under the spot-market s wage w t. This is beause a risk-averse worker is willing to reeive a lower wage on average in exhange for reeiving a sure wage. However, w t is not self-enforing. More speifially, if the firm ommits to pay w t at the beginning of period t, it will have an inentive to renege on this ommitment whenever a negative shok hits the reservation wage rt (i.e., whenever ht p h). The gain from reneging in period t is given by (1 n){wt [(1 r)r* rrt 1 h]} 1 0. (4) If a reneging firm is punished by having to pay a spot-market wage forever after, the ost of reneging in period t is given by j t j t s (1 n) d (1 n) [E t(wjfh t p h) E t(wjfh t p h)] 1 0. (5) jpt 1 The ontratual wage w t will be a self-enforing equilibrium as long as the ost of reneging equation (5) is larger than the gain from reneging equation (4). In other words, will be self-enforing as long as w t j j s d (1 n) [E t(wjfh t p h) E t(wjfh t p h)] jpt 1 1 wt [(1 r)r* rrt 1 h]. (6) As expression (6) learly indiates, a self-enforing equilibrium in w t beomes less and less likely as n inreases. As n inreases, the firm disounts the future at a higher rate and beomes less able redibly to ommit to pay the ontratual wage. As n inreases, workers beome less willing w t

9 Import Competition and the Employment Relationship 731 to aept impliit wage-setting agreements that are not redible. In other s words, as n inreases, wage setting moves away from wt and toward wt. Note that, under the assumptions of the model, an inrease in n does not affet the enforeability of w t in the absene of a risk of default. Indeed, if the firm does not fae a stritly positive probability of liquidation, for example when D p 0, the gain from reneging in period t is given by wt [(1 r)r* rrt 1 h] 1 0, (7) while the ost to reneging in period t is given by j t s d [E t(wjfh t p h) E t(wjfh t p h)] 1 0. (8) jpt 1 In the absene of a risk of finanial distress, neither the gain from nor the ost of reneging on impliit wage-setting ontrats depends on n. 12 C. Impliations for Empirial Work In the empirial work that follows, I propose to test the major impliations of the model. In pratie, I will use the state unemployment rate at time t as a proxy for the reservation wage r t, a higher unemployment rate implying a lower reservation wage. I will use the state unemployment rate at the time of job start as a proxy for the reservation wage r t 1. I will use industry-level measures of import penetration as a proxy for n, a higher level of import penetration implying that eah firm in the industry is more likely to experiene a negative revenue shok. Equations (1) and (2) diretly motivate how I empirially distinguish s spot-market wage setting ( wt) from ontratual wage setting ( wt). I will interpret a more negative orrelation between urrent wages and urrent unemployment and less negative orrelation between urrent wages and unemployment at the time of job start as evidene of a swith away from s wt and toward wt. Equations (1) and (2), in ombination with equation (6), motivate the tests I perform to study the impat of import penetration on the wagesetting proess: (i) as import penetration inreases, I expet the sensitivity of urrent wages to the urrent loal labor-market unemployment rate to inrease; (ii) as import penetration inreases, I expet the sensitivity of urrent wages to the unemployment rate at the time of job start to derease; and (iii) I expet both of these effets to be stronger in more- 12 However, even with perfet apital markets, an inrease in n ould affet the firm s disount rate by inreasing the probability of exiting the industry beause of negative eonomi profits. We have ruled out that ase here by assuming that p t(r H, w t) 1 p t(r L, w t) 1 0. See Bertrand (1998) for an extension of the model to the perfet apital market ase.

10 732 Bertrand leveraged firms. I now desribe in detail the empirial methodology I follow to test eah of these three preditions. III. Empirial Strategy Using repeated ross-setions of individual-level data, I first test predition i by estimating the following equation: log (w ) p au bimp (imp # u ) ijkt kt jt jt kt d e f gd w, (9) k t j ijkt ijkt where i, j, k, and t, respetively, index individual, industry, state, and time; wijkt is the wage rate for individual i; ukt is the unemployment rate in state k and time t; imp jt is a measure of import penetration in industry j at time t; dk is a vetor of state fixed effets; et is a vetor of time fixed effets; fj is a vetor of industry fixed effets; Di is a vetor of individual har- ateristis; and w is an error term. 13 ijkt Equation (9) is a slight modifiation of the standard wage urve (see Blanhflower and Oswald 1994; Card 1995; Blanhard and Katz 1997) where the negative relationship between urrent wage ( w ijkt ) and urrent unemployment rate ( u kt ) is allowed to depend on the level of import penetration. 14 The oeffiient of interest in equation (9) is. Under the model desribed above, I expet to be negative. A negative value for implies that a higher level of import penetration inreases (in absolute value) the urrent wage urrent unemployment elastiity. There are two points worth highlighting about the speifiation in equation (9). First, I rely on industry-level variation in import penetration. This may raise some onerns beause of the diffiulty of omparing ompetition levels between industries. The addition of industry fixed effets f j partly alleviates these onerns. Industry fixed effets allow me to look at within-industry variation in imp and imp # u. 15 jt jt kt Seond, I use individual-level data even though the soure of variation I exploit operates only at the industry-state-year level. The main advantage of a mirodata set is to mitigate the importane of potential omposition biases. Suppose that a demographi reshuffling of the workfore aom- 13 In pratie, I will define imp jt as the urrent level of import penetration in an industry but also as an average of urrent and future import penetration levels in order to apture more permanent hanges in import penetration, as implied by eq. (6). See Se. V.B for details. 14 Appendix A disusses in further detail identifying assumptions for this eonometri model. 15 Obviously, onerns ould still arise if there is between-industry variation in the wage-unemployment elastiity. I have onsidered alternative speifiations of eq. (9), where I allowed for interations of industry and also year fixed effets with. The results were qualitatively unaffeted. u kt

11 Import Competition and the Employment Relationship 733 panies hanges in the ompetitive environment and that the elastiity of wages to the unemployment rate differs aross demographi groups. One might then worry that the estimated oeffiient on imp jt # ukt reflets in part some of these ompositional hanges in the industry workfore. The vetor of individual harateristis D ijkt aptures observable hanges in the demographi mix of an industry. One major limitation of the ross-setional data is that they do not provide information on work history. 16 I annot test in the ross-setional data whether import penetration also affets the elastiity of the urrent wage to the unemployment rate at the time of job start (predition ii of the model). I therefore also use individual-level panel data to estimate the following equation: log (w ) p au bimp (imp # u ) hus ijkt kt jt jt kt ikt k(imp # us ) d e f g w, (10) jt ikt k t j i ijkt where us ikt is the state unemployment rate that was prevailing in the year that individual i started working for her present (period t) employer and g i is a vetor of individual fixed effets. The oeffiients of interest in equation (10) are and k. Under the model desribed above, I expet to be negative and k to be positive. A negative value for means that a higher level of import penetration inreases the urrent wage urrent unemployment elastiity. A positive value for k means that a higher level of import penetration dereases the urrent wage starting unemployment elastiity. 17 Another advantage of using panel data, besides the availability of a work history, is that it also allows me to ontrol for individual fixed 16 I use the Merged Outgoing Rotation Groups (MORGs) of the Current Population Survey (CPS) to onstrut the ross-setional sample. Advantages of the CPS data are a long time period and a large representative sample. The MORGs, however, do not ontain information about work history. The mobility supplements to various January CPSs do ontain information on how long workers have held their urrent jobs. However, these supplements are not available on a frequent basis. Only six mobility supplements are available between 1973 and If the true model is as desribed in Se. II, one might be onerned about possible biases in eq. (9), whih inludes neither the starting unemployment rate nor the starting unemployment rate interated with import penetration. More speifially, the estimates of oeffiients a and in eq. (9) will be biased if the urrent unemployment rate and the starting unemployment rate are orrelated. I used the panel data to assess the sign and magnitude of this orrelation. Controlling for state, year, and industry effets, I found a positive but statistially insignifiant orrelation between urrent and starting unemployment rate. This implies that, if anything, the estimate of a will be biased downward and the estimate of will be biased upward, assuming that the true model is as desribed in Se. II. The results in Se. V onfirm this onjeture.

12 734 Bertrand effets g i. Suh individual effets eliminate ompositional problems aused by unobserved harateristis that are onstant over time. The presene of individual fixed effets also lessens onerns of biases arising if individuals who start a new job in a reession differ along observable and unobservable dimensions. 18 Building on equations (9) and (10), one an easily test predition iii by omparing the estimated values of and k in high- and low-leverage industries. The model desribed above predits that and k should be larger in absolute value in more-leveraged industries. One ould, in priniple, diretly estimate equations (9) and (10) using ordinary least squares (OLS). However, one might suspet that import penetration measures are orrelated with some omponent of the error term w, in whih ase OLS will yield biased and inonsistent estimates of the parameters of the model. First, reverse ausation an ontaminate the OLS estimation as adverse shoks to labor osts or hanges in personnel praties may diretly affet the ompetitive position of a firm or industry. For example, firms that deide to buffer their workers wages from the vagaries of the external labor market may fae relatively higher prodution osts during reessions, whih may leave them more vulnerable to foreign ompetition. Seond, the OLS estimation may be subjet to an omitted variable bias. For example, managers with weak inentives may tend to protet their workers more from outside labor-market onditions and also be more likely targeted by foreign ompetitors. 19 These examples justify the need to identify a soure of variation in import penetration that is independent of all management hoie variables. I follow Revenga (1990, 1992) and instrument industry measures of import penetration with soure-weighted industry exhange rates. 20 The exhange rate variable is orrelated with industry-level measures of import penetration but is primarily determined by maroeonomi variables that, onditional on year dummies, an reasonably be regarded as exogenous to the behavior of a ertain industry in a ertain period. 21 Thus, it is a valid instrument for import penetration. 18 The individual fixed effets, however, do not deal with the possibility that job-mathing quality is proylial. It is diffiult, though, to explain why variations in import penetration would affet the proyliality of job-mathing quality. It is espeially diffiult to explain why urrent levels of import penetration would affet the quality of previously made mathes. 19 Note that both of these stories imply a positive orrelation between the level of import penetration and the use of impliit wage shielding ontrats. These stories thus work against finding a negative impat of import penetration on wage shielding in an OLS speifiation. 20 See Se. IV and app. B for a detailed desription of how the exhange rate variables are onstruted. 21 Controlling for year dummies is possible beause the exhange-rate variable varies between industry based on the omposition of imports by ountry of origin.

13 Import Competition and the Employment Relationship 735 In summary, the empirial strategy onsists in estimating equations (9) and (10) by two-stage least squares. In equation (9), two variables are instrumented: impjt and imp jt # ukt. The instruments are xrjt and xr jt # ukt, where xrjt is the soure-weighted exhange rate for industry j at time t. 22 In equation (10), three variables are instrumented: impjt, imp jt # ukt, and imp jt # usikt. The instrumental variables in that ase are xrjt, xr jt # u, and xr # us. kt jt ikt IV. Data This setion briefly reviews the prinipal data soures used in this artile. Appendix B ontains a more detailed desription of all the data soures. I use two different individual-level labor data sets: a Current Population Survey (CPS) extrat and a Panel Study of Inome Dynamis (PSID) extrat. The CPS extrat is the repeated ross-setion data set. It is a random subset of the entire set of manufaturing workers in the May CPSs and Merged Outgoing Rotation Groups (MORGs) of the CPS over the period I limit the sample to manufaturing workers sine import penetration data are available for only the manufaturing setor of the eonomy. I use the May CPSs over the period and the MORGs over the period Both the use of the May CPSs over the period and the absene of data for 1982 are justified by the need to extrat a ontrol for union status. Union status is not available in 1982 and is available in the May CPSs only prior to The PSID extrat is the panel data set. As I have already alluded to in Setion III, this seond data set has two main advantages over the CPS extrat. First, the PSID ontains information about the length of tenure with the urrent employer. 23 This information is ruial to the onstrution of a measure of loal labor-market onditions at the time of job start. Seond, the PSID extrat permits a superior ontrol for ompositional effets beause of its panel struture. Unfortunately, the PSID also has major drawbaks. First, the sample size is limited. It is espeially so in the ase of this study as I have to onentrate on workers in the manufaturing setor. Seond, the PSID does not ontain a detailed industry 22 In pratie, I often use both the urrent and lagged industry soure-weighted exhange rates as instruments. 23 Note that the wording of this question has substantially hanged over time. Before 1984, the question ould have been interpreted as duration of ontinuous employment rather than time together with the present employer. After 1984, the question expliitly asks for time together with the present employer (Polsky 1999). This impreise and hanging wording of the tenure question is likely to introdue measurement error and bias the estimated oeffiients on usikt and imp jt # usikt toward zero.

14 736 Bertrand ode prior to This limits the time span of the PSID extrat to the period Both the CPS and the PSID extrats are merged by industry and year with an import penetration variable and a real exhange rate index. The import penetration variable is onstruted from the National Bureau of Eonomi Researh (NBER) Trade Database. This database is desribed in detail in Feenstra (1996). Import penetration is defined as the ratio of imports over imports plus domesti prodution in a given industry and a given year. The real exhange rate index also varies at the industry-year level. It is defined as the weighted geometri average of the real exhange rates of the importing ountries. The weights for a given industry are the shares of eah foreign ountry s imports in the total imports of that industry in a base period ( ). I use a onordane between the 1972 standard industrial lassifiation (SIC) ode (in whih the trade data are expressed) and the ensus of population industry lassifiation ode to link the import penetration and exhange rate variables to the individuallevel labor data sets. The two labor data extrats are also merged by state and year with an unemployment rate measure. The unemployment rate measure is the log state unemployment rate for the total ivilian population in a given year. Finally, the two labor data extrats are merged by industry and year with a set of orporate variables onstruted from COMPUSTAT, whih reports historial finanial data for more than 7,500 orporations established in the United States. The data are drawn from shareholders, 10- K, and 10-Q reports and generally onsist of large ompanies with substantial publi ownership. While manufaturing firms in COMPUSTAT represent only a limited proportion of the total number of manufaturing firms in the United States, they represent a very large proportion of total manufaturing employment. For example, in 1985, manufaturing employment in COMPUSTAT totals about 16 million. 24 Aording to the Bureau of Labor Statistis s Employment and Earnings, 19 million workers were employed in manufaturing in the United States in The industry-level orporate variables I ompute are weighted industry averages of the equivalent firm-level variables. Eah firm is weighted in a given year by the share of its total assets in that year in the total assets of the industry. I use a onordane between the SIC-72 ode (in whih COMPUSTAT data are expressed) and the ensus of population industry lassifiation ode to merge the orporate variables with the CPS and PSID extrats. I ompute industry-level measures for aounting return, stok market return, and leverage. One merged with the import penetration, exhange rate, and orporate 24 Total manufaturing employment represented by COMPUSTAT firms is between 14 and 17 million workers in every available year.

15 Import Competition and the Employment Relationship 737 Table 1 Desriptive Statistis: CPS Individual Level Data All Male Nonunion Male Sample Mean SD Mean SD Mean SD Variable: Import penetration Female Age Years of experiene Years of eduation Union Wage Log wage Unemployment rate Log unemployment rate (Operating inome after depreiation)/(total assets) Common stok return (Total equity)/(total liabilities) in Net leverage in Sample size 130,567 87,615 60,110 Soure. Data are omposed of a random subsample of manufaturing workers over (soure: May Current Population Survey) and (soure: Merged Outgoing Rotation Groups, Current Population Survey). Note. Details of the onstrution and definition of variables are ontained in Se. IV and app. B. All orporate variables are industry means onstruted from COMPUSTAT. When required, variables are deflated by the onsumer prie index ( p 100). variables, the CPS extrat overs 62 different industries, while the PSID extrat overs 65 different industries. V. Empirial Results A. Desriptive Statistis Tables 1 and 2 present summary statistis for the main variables of interest in the CPS and PSID extrats, respetively. For eah of these two extrats, I also onsider two important subsamples. First, I onsider the subsamples of male workers. Impliit agreements typially require some relatively long-term attahment to the labor fore. Beause males have more stable labor-fore partiipation, one might suspet that they are more likely to be overed by the impliit wage-setting ontrats. This implies that they are also more likely to be affeted by hanges in import ompetition. Finding larger hanges in wage-unemployment elastiities for male workers would, in that sense, be a first hek of the validity of the results. Seond, I onsider the subsamples of nonunionized males. One might worry that the results in this artile apture hanges in expliit, rather than impliit, ontrating. Previous work has indeed doumented how hanges in the ompetitive environment of an industry may lead to a

16 738 Bertrand Table 2 Desriptive Statistis: PSID Individual Level Data All Male Nonunion Male Sample Mean SD Mean SD Mean SD Variable: Import penetration Female Age Years of experiene Months of tenure with present employer Years of eduation Union Wage Log wage Unemployment rate Log unemployment rate Log unemployment rate at job start (Operating inome after depreiation)/(total assets) Common stok return (Total equity)/(total liabilities) in Sample size 10,026 7,132 5,122 Soure. Data are omposed of manufaturing workers in the Panel Study of Inome Dynamis over Note. Details of the onstrution and definition of variables are ontained in Se. IV and app. B. All orporate variables are industry means onstruted from COMPUSTAT. When required, variables are deflated by the onsumer prie index ( p 100). breakdown of national union ontrats in that industry. For example, Belzer (1995) studies how the eonomi deregulation of the truking industry affeted the nature of union bargaining. He finds that bargaining between the Teamsters and general freight motor arriers beame more deentralized following deregulation. While wage levels in the truking industry used to be very uniform aross regions under national bargaining, deregulation inreased the sensitivity of wages to loal eonomi onditions. Hene, hanges in expliit ontrating ould explain why produt market ompetition inreases the elastiity of wages to state unemployment rates. I will show that results are not driven by hanges in the nature of union ontrats. 25 The CPS samples (table 1) appear to have a higher proportion of females and unionized workers than the PSID samples (table 2). Also, mean import penetration is higher in the PSID. Lower unionization and higher imports in the PSID both reflet the fat that the PSID extrat overs a 25 Note that, when I use the full sample or the sample of male workers, I always ontrol for union status. Thereby, I aount for the deline in unionization rate that might be orrelated with the inrease in produt market ompetition.

17 Import Competition and the Employment Relationship 739 more reent time period on average. Workers in the PSID samples are younger and, having eduation levels similar to the CPS workers, are less experiened. 26 In both the PSID and the CPS extrats, the full sample and two subsamples differ in expeted ways. Male workers earn higher wages, are somewhat more eduated (espeially if they are not unionized), and have a higher unionization rate. Nonunion males appear less experiened and slightly younger than the average male worker. Tenure with employer (only available in the PSID sample and reported in months) is higher for male workers, although it is about 4 months lower for nonunionized males than for the average worker in the full sample. B. Basi Results In this setion, I present evidene that, other things equal, wages in an industry beome more sensitive to the urrent unemployment rate and less sensitive to the unemployment rate at the time of job start when that industry faes an inrease in import penetration. I will first disuss the CPS results. The CPS data set is muh larger than the PSID data set and overs a longer time period. However, as I mentioned above, the CPS data set does not ontain the work history information required to ompute the time of job start. Hene, in the CPS data, I an only study hanges in the urrent wage urrent unemployment elastiity. In the seond part of this setion, I will remedy this weakness of the CPS by moving to the PSID data. 1. CPS Results Table 3 presents the effet of import penetration on the elastiity of wages to the urrent rate of unemployment. I onsider two different measures of foreign ompetition. First, I simply measure ompetition with the ontemporaneous level of import penetration. In light of the model presented in Setion II, one might argue against the use of suh a ontemporaneous measure. A more relevant measure should instead apture firms expetations about future levels of import penetration. Under the hypothesis that exhange rates follow a random walk (Meese and Rogoff 1983), the urrent level of import penetration may, however, be the best proxy for firms expetation about future level of import penetration, espeially in the IV model. Nevertheless, I will also use as an alternative 26 Part of this age gap is diretly indued by the onstrution of the PSID extrat. Beause I need state unemployment rate at the time of tenure start and beause state unemployment rates are only available sine 1970, I exlude from the PSID extrat observations for whih tenure with employer started prior to 1970.

18 740 Bertrand Table 3 Effet of Produt Market Competition on the Elastiity of Current Wage to Current Unemployment Current Import Penetration 3-Year Average Import Penetration Sample All (1) Male (2) Nonunion Male (3) All (4) Male (5) Nonunion Male (6) IV speifiation: Import penetration ( imp jt ) (.300) (.378) (.430) (.349) (.437) (.502) Log unemployment rate ( u kt ) (.008) (.009) (.010) (.008) (.009) (.010) imp jt # ukt (.093) (.114) (.142) (.085) (.105) (.133) 2 x -statisti, test that imp jt and imp jt # ukt are exogenous ( Prob 1 x -stat) (.001) (.004) (.013) (.001) (.003) (.012) Sample size 130,567 87,615 60, ,567 87,615 60,110 (7) (8) (9) (10) (11) (12) OLS speifiation: Import penetration ( imp jt ) (.056) (.075) (.089) (.057) (.078) (.092) Log unemployment rate ( u kt ) (.007) (.008) (.010) (.007) (.008) (.010) imp jt # ukt (.051) (.070) (.087) (.049) (.067) (.083) Adjusted R Sample size 130,567 87,615 60, ,567 87,615 60,110 Soure. Data are omposed of a random subsample of manufaturing workers over (soure: May Current Population Survey) and (soure: Merged Outgoing Rotation Groups, Current Population Survey). Note. Dependent variable: log( w ijkt ) (log deflated hourly wage); IV p instrumental variables; OLS p ordinary least squares. In ols. 1 3 and 7 9, imp jt measures urrent import penetration. In ols. 4 6 and 10 12, imp jt is the average of urrent, 1-year lead, and 2-year lead import penetration. Variables are defined in Se. IV and app. B. The dependent variable is the log deflated hourly wage. Both import penetration and the log unemployment rate have been demeaned. Covariates inluded in eah regression are 51 state dummies, 16 year dummies, and 62 industry dummies. Additional ovariates inluded in eah regression are eduation, a quarti in experiene, a union dummy, a nonwhite dummy, a female dummy, and interations for the four experiene terms with the female dummy. In the IV speifiations (ols. 1 6), both import penetration and the interation of import penetration with the log unemployment rate are instrumented. The instruments are urrent and 1-year lagged exhange rate indies and the interation of these indies with the log unemployment rate. Standard errors are in parentheses exept 2 that the numbers in parentheses below the x -statistis are p-values. They are orreted to allow for group effets within state-industry-year ells. measure the mean level of import penetration over a 3-year period: urrent period, 1-year, and 2-year leads. 27 The struture of table 3 is as follows: olumns 1 6 present the IV regressions. Columns 1 3 use urrent import penetration as the measure of foreign ompetition, while olumns 4 6 use the 3-year average import penetration. Columns 1 and 4 use the full sample, olumns 2 and 5 onentrate on male workers, and olumns 3 and 6 on nonunionized males. 27 For 1991, the mean import penetration is over a 2-year period: urrent period and 1-year lead. For 1992, I simply use urrent import penetration.

19 Import Competition and the Employment Relationship 741 Columns 7 12 are the equivalent OLS speifiations. The dependent variable in eah olumn is the log of deflated hourly wage. Eah regression inludes 16 year dummies, 51 state dummies, 62 industry fixed effets, and a set of demographi ontrols. The demographi ontrols are eduation, a quarti in experiene, a union dummy, a nonwhite dummy, a female dummy, and interations for the four experiene terms with the female dummy. In eah IV regression, two variables are instrumented: the import penetration measure and the interation of import penetration with the urrent state log unemployment rate. The instruments are urrent and 1-year lagged industry exhange rate indies as well as the interation of eah of these indies with the urrent state log unemployment rate. Standard errors are adjusted for group effets. For illustration, table C1 reports the two first-stage equations assoiated with regression 4 in table 3. The instruments are jointly very highly signifiant for eah of the first-stage regressions ( p!.001). I start by disussing the IV results. 28 First, onsistent with previous work, I find that a higher level of import penetration redues the level of wages in the IV speifiations. 29 More important, the results in table 3 show that import ompetition inreases the elastiity of wages to the urrent unemployment rate. The hange in elastiity is stronger for the subset of male workers, and stronger yet for nonunionized males. 30 However, none of these differenes is statistially signifiant. Note that the partial test of whether the instrumental variables are orrelated with the error term (not reported here) fails to rejet the null hypothesis of no orrelation at the.10 level in all ases. Consider table 3, olumn 1. At the mean level of import penetration 28 Beause of spae onstraints, I do not report the oeffiients on the demographi variables for all the regressions in table 3. Table D1 presents the full list of estimated oeffiients for table 3, regression 4. Wages are signifiantly higher for whites, males, unionized, more eduated, and more experiened workers. The returns to experiene are lower for females. 29 Several authors have already studied the impat of trade pressures on the level of wages. In a panel of manufaturing industries over the period , Revenga (1992) shows that hanges in import pries signifiantly redue wages. Using industry-state-year data, Goldberg and Tray (2001) find that import and export exhange rate movements have statistially signifiant effets on wages. They also find that the importane of these effets varies aross industries and regions. Interestingly, Campa and Goldberg (2001) find that the exhange rate effets on wages are stronger in lower profit margin industries. 30 In regressions not reported here, I have also split the sample based on whether the individual has at best ompleted high shool or has at least ompleted 1 year of eduation after high shool. I find that the hange in elastiity is stronger in the high eduation group. This is onsistent with the idea that eduated workers are more likely to be overed by impliit employment ontrats. This fat also mathes the popular pereption that the onditions of employment are espeially hanging for the white-ollar workfore.

20 742 Bertrand (about 10%), a one-standard-deviation inrease in the unemployment rate redues wages by about 1%. If import penetration inreases by 10 perentage points (a little more than one standard deviation), wages drop by about 1.6% for the same inrease in the unemployment rate. In olumn 3, a one-standard-deviation inrease in the unemployment rate redues nonunionized male workers wages by 2.5% instead of 1.5% when import penetration is 10 perentage points higher. A omparison of olumns 1 6 and 7 12 onfirms the importane of the instrumentation strategy. First, like Revenga (1990, 1992), I find no signifiant impat of trade on wage levels in the OLS speifiations. Seond, there is no apparent effet of foreign ompetition on the sensitivity of wages to the loal unemployment rate in the OLS model. The point estimates on the interation term imp jt # ukt are both eonomially and statistially insignifiant in the OLS speifiations. The Hausman test reported at the bottom of the IV panel in table 3 heks for a statistially signifiant differene between the OLS and IV estimates of the oeffiients on imp and. 31 jt # ukt impjt For all speifiations in table 3, the Hausman test rejets the null hypothesis that imp jt # ukt and impjt are exogenous. As I mentioned before, a potential explanation of the disrepany between the OLS and IV results is that industries that protet their workers are likely to engage at the same time in other personnel praties (suh as paying above market wages) that redue their ompetitiveness and weaken their protetion from foreign produers. In all the speifiations of table 3, I have restrited the returns to all the demographi variables to be onstant over the entire sample period. These restritions are not realisti. For example, it is well known that the returns to eduation have inreased over the period under study. Table 4 verifies the robustness of the results to these speifiation assumptions. I present only the IV results. In olumns 1 5, I allow the returns to eduation, unionization, rae, experiene, and sex, respetively, to vary over time. The oeffiient on imp # u is never affeted. 32 jt kt 31 The Hausman endogeneity test involves regressing the OLS residuals on the fitted values (from the first stage) of the variables that are instrumented as well as on all the variables in the OLS regression. In addition, N times the R 2 from this regression, where N equals the number of observations, asymptotially follows a x 2 distribution. 32 Another possible soure of onern with the eonometri speifiation is that the interation term imp jt # ukt might be proxying for the fat that log hourly wages wijkt are not linear in impjt or ukt. In regressions not reported here, I have studied the robustness of my results to allowing for a quadrati in u kt, a quadrati in imp jt, or both. The results are not sensitive to the addition of a quadrati in ukt. The point estimate on imp jt # ukt beomes even larger in absolute value when one ontrols for a quadrati in impjt or for both a quadrati in ukt and in impjt. The statistial signifiane of the estimate is, however, weakened in these last two ases ( p p.11).