The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Federal Reserve Bank of Chicago Displacement, Asymmetric Information and Heterogeneous Human Capital Luojia Hu and Christopher Taber REVISED June 2, 2010 WP 2008-02 Displacement, Asymmetric Information and Heterogeneous Human Capital 1 Luojia Hu Federal Reserve Bank of Chicago Christopher Taber Department of Economics University of Wisconsin-Madison June 2, 2010 1 We thank Joe Altonji, Gadi Barlevy, Charlie Brown, Hank Farber, Jim Heckman, John Kennan, Bruce Meyer, Espen Moen, Ann Hu¤ Stevens and seminar participants at Iowa State University, University of Illinois, Northwestern University, Michigan State, NYU, UC-Berkeley, University of Chicago, Vanderbilt University, the Federal Reserve Bank of Chicago, and the 2006 SOLE meetings for helpful comments. All errors remain our own. An earlier version of the paper was circulated under the title “Layo¤s, Lemons, Race, and Gender.”Hu gratefully acknowledges …nancial support from the Searle Fund for Policy Research and the W.E. Upjohn Institute for Employment Research. Taber acknowledges support from NSF (SES-0617438). The opinions expressed here are those of the authors and not necessarily those of the Federal Reserve Bank of Chicago or the Federal Reserve System. Abstract Gibbons and Katz’s (1991) asymmetric information model of the labor market predicts wage losses following displacement should be larger for layo¤s than for plant closings. This was borne out in their empirical work. In this paper, we examine how the di¤erence in wage losses across plant closing and layo¤ varies with race and gender. We …nd that the basic prediction by Gibbons and Katz only holds for white males. We augment their asymmetric information model with heterogeneous human capital and show that this augmented model can match the data. 1 Introduction The role of asymmetric information in labor market outcomes has long been of interest to labor economists (e.g. Akerlof, 1976, Spence, 1973, Greenwald, 1986, and Laing, 1994). Empirical studies on this topic, however, have been scarce. In a seminal paper, Gibbons and Katz (1991; hereafter GK) construct a model of asymmetric information in the labor market. They use their model to argue that if …rms have discretion as to which workers to lay o¤, a layo¤ provides a signal to the outside market that a worker is of low quality. By contrast, virtually all workers lose their jobs when their plant closes so job loss from plant closing does not provide a negative signal. GK test for asymmetric information by looking at changes in wages for white collar workers.1 Since a layo¤ provides a negative signal about ability, one would expect wages to fall more following a layo¤ than for a plant closing. They con…rm this prediction in the data showing that wage penalties are substantially higher for layo¤s than for plant closings. In this paper, we take advantage of the fact that we have many more years of displaced workers data to expand on GK by looking at how the di¤erence in wage losses across plant closing and layo¤ varies with race and gender. Statistical discrimination against African Americans or women occurs when employers use race and gender as a predictor for productivity.2 If this is the case, then one would expect the information contained in a layo¤ to vary across racial and gender groups. Empirically, we …nd that the di¤erences between white males and the other groups are striking and complex. The basic prediction by GK actually only clearly holds for white males. Both black females and black males actually experience a much greater decline in earnings at plant closings than at layo¤s. For white females the losses at plant closing and layo¤ are very similar. These results arise from two reinforcing e¤ects. First, plant closings have substantially more negative e¤ects on minorities than on 1 They use white collar workers because they argue that blue-collar jobs are much more likely to be covered by collective bargaining agreements. In that case seniority is typically the main determinant of the layo¤ decisions so that a layo¤ will not necessarily convey negative information. One might expect seniority to be a more important factor for blue collar than white collar workers in nonunionized …rms as well. Abraham and Medo¤ (1984) have evidence suggestive of this in that seniority is more important for determining layo¤s for nonunionized hourly workers than for nonunionized salary workers. The information content may also di¤er for these di¤erent types of workers. For these reasons we follow GK and focus on white collar workers, but we present results for blue collar workers as well. 2 The theory of statistical discrimination was introduced by Phelps (1972) and Arrow (1973) and subsequently developed by, among others, Aigner and Cain (1977), Lundberg and Startz (1983), and Coate and Loury (1993). Empirical studies of statistical discrimination are still scarce. A notable exception is Altonji and Pierret (2001). Altonji and Blank (1999) presents a survey on this topic. 1 whites. Second, layo¤s seem to have more negative consequences for white men than for the other groups. Does this mean that one should discard the GK model? We think clearly not. However, the simple model is not su¢ cient to explain all of the data, so we augment it. We propose a new model that extends the asymmetric information model of GK by including heterogeneous human capital.3 In our model, di¤erent types of …rms hire di¤erent types of workers. Once an employee has worked for a …rm for one period, the current …rm knows his/her skill level, but outside …rms do not. We model layo¤s and plant closings as arising when shocks hit …rms. Severe shocks lead the …rms to cease operation (plant closings) while less severe shocks lead them to reduce the size of their workforce (layo¤s). On the one hand, plant closing may be more devastating than layo¤ because it may be associated with a larger negative shock to the human capital of a particular worker. On the other hand, layo¤s send a bad signal to the market and thus have additional negative consequences on the worker. If layo¤ is a substantially stronger signal for white males than for the other groups, this could lead the information hit for layo¤ to dominate for white males while the human capital aspect to dominate for the other groups. We provide some additional evidence that is suggestive that both asymmetric information and heterogeneous human capital are important. In support of both explanations we demonstrate that the racial and gender e¤ects are surprisingly robust to inclusion of region, industry, and occupation controls. We argue that this would seem unlikely if the explanation were simply that there is variation in the type of jobs performed by di¤erent demographic groups. To further look at asymmetric information, we make use of the Civil Rights Act of 1991 which induced employers to lay o¤ “protected”workers in mass layo¤s rather than …re them for cause. As a result, layo¤ should become a relatively more negative signal for blacks after 1991 than prior. Thus, if asymmetric information is important, one would expect the relative wage losses of blacks following layo¤s to increase after 1991 which is precisely what we …nd (although the standard errors are large). Our evidence on the importance of human capital heterogeneity arises from the distinction between two types of layo¤s. In the displaced worker survey, an individual can become laid o¤ either because of “position abolished”or “slack work.”In the spirit of our model, a “slack work” layo¤ can be thought of as arising from a shock to one’s …rm type (i.e. something 3 GK also informally make the point that if plant closing occurs in worse labor markets then we might see bigger drops in wages. This is related to our concept of heterogeneous human capital. 2 like an industry speci…c shock). By contrast, a “position abolished” layo¤ can be thought of as arising from a human capital type speci…c shock (i.e. something like an occupation speci…c shock). An individual can avoid the …rst order e¤ect of the former type of shock by switching sectors, but they can not avoid the …rst order e¤ect of the latter type shock. Thus if the shocks are of similar magnitude, we would expect a substantially larger fall in earnings from the second type of shock than for the …rst, and that is precisely what we see in the data. Earnings losses are much larger when layo¤ is associated with “position abolished” than when it is associated with “slack work.” Furthermore, we …nd that wages losses of plant closings fall in between these two types of layo¤s in terms of their magnitude. The distinction between di¤erent types of layo¤s are, to our best knowledge, new to the literature on displacement and we think they are interesting in their own right. Finally, we simulate our model and show that it can match the data. Due to sample size (and precision) considerations, we focus on the di¤erences across gender. We show in our model that if a layo¤ is a substantially stronger signal for men than for women, this could lead the information hit for layo¤ to dominate for men while the human capital aspect dominates for women. This allows us to reconcile the result. While we can not formally prove that one needs asymmetric information and heterogeneous human capital to match the moments, we think our model provides the most plausible story. A number of other studies have examined the comparison between layo¤s and plant closings using data other than the DWS. Using data from the PSID, Stevens (1997) also …nds that wage losses following layo¤s are larger than those following plant closings. However, she shows that this …nding can be explained by the larger wage reductions prior to displacement for plant closings than for layo¤s. Her analysis does not condition on race, gender, or blue/white collar so it is di¢ cult to directly compare to our study particularly given that it is not obvious how respondents answer the retrospective question about previous wages in the DWS. Using data from the NLSY, Krashinsky (2002) also …nds that workers displaced by layo¤s su¤er larger wage losses than those displaced by plant closings. However, he provides an alternative explanation attributing the e¤ect to di¤erences in …rm size of pre-displacement employers. He argues that small …rms are more likely to close down when facing adverse economic shocks, while larger …rms are more likely to reduce their workforce. Therefore laid-o¤ workers tend to lose any wage premium or rents they earned from working at large …rms. He shows that when …rm size is included in the wage loss regression, the di¤erence 3 between layo¤s and plant closings becomes statistically insigni…cant, although the con…dence intervals are wide enough to include substantial di¤erences as well. Song (2007) reexamines the GK study and argues that their …ndings can be partly attributed to di¤erential recall bias for layo¤s versus plant closings in the 1984 and 1986 DWS and, in later years, mostly by higher wage-tenure pro…le prior to displacement for layo¤s than for plant closings. We view all of these paper as potentially providing alternative explanations for the main GK …nding, however we do not view them as de…nitively establishing that GK’s explanation is incorrect. The main contribution of our paper is in looking at various interactions (among race, gender, white/blue collar and types of dismissals) which these other papers do not do. Therefore we proceed taking the GK explanation of the data, but it is important for the reader to keep these alternatives in mind. The remainder of the paper is organized as follows. Section 2 describes the data. Empirical results are reported in Section 3. We present the model in Section 4, discuss the di¤erence between slack work and position abolished in Section 5, and then use this information in the simulation in Section 6. Finally, Section 7 discusses the results and concludes. 2 Data We use data from the biennial Displaced Workers Surveys (DWSs) Supplement to the Current Population Survey (CPS) between 1984 and 2008. The DWSs were conducted as part of the January CPSs in 1984, 1986, 1988, 1990, 1992, 2002, 2004, 2006 and 2008 and the February CPSs in 1994, 1996, 1998 and 2000. Each of the supplements from 1984-1992 asks workers if they lost a job at any time in the previous 5-year period, and each supplement from 1994-2008 asks this question for the previous 3-year period.4 Displacement is de…ned as involuntary separation based on operating decisions of the employer such as plant closing, employer going out of business, layo¤ from which the worker was not recalled. Other events, including quits and being …red for cause, were not considered displacement. Thus, the supplement is designed to focus on the loss of jobs that results from business decisions of …rms unrelated to the performance of particular workers. If the response to the job loss question is positive, the respondent is then asked about the reason of job loss: 1) plant closing, 2) slack (insu¢ cient) work, 3) position or shift abolished, 4) seasonal jobs ended, 5) self-employment failed, and 4 The DWSs ask and collect information on at most one job loss for each individual. If the respondent lost more than one job in the reference period, she/he is asked about information only for the longest job lost. 4 6) other. The data have information on workers’demographics, tenure on pre-displacement job, occupation, industry and weekly earnings, weeks of joblessness after displacement and current weekly earnings.5 We restrict the sample to workers aged 20-64 who lost a job in the private sector in the preceding 3-year period due to plant closing, slack work or position or shift abolished, and are reemployed in the private sector at the survey date. We only focus on workers who made full time to full time job transitions (i.e. lost a full-time job and are re-employed on a fulltime job).6 We exclude workers who have re-employment weekly real earnings under $40. Earnings are de‡ated by the 1982-84=100 consumer price index (CPI). As in Gibbons and Katz, we distinguish between blue and white collar workers. The white collar sample consists of workers who were displaced from jobs as managers and administrators, professional and technical workers, clerical workers, and sales workers while the blue collar sample consists of workers who were displaced from jobs as craft and kindred workers, operatives, laborers, transport operatives, or service workers. We exclude workers in agriculture and construction industries. Descriptive statistics of the sample are reported in Table 1. We divide the data into sixteen di¤erent groups, classifying by gender, race, blue/white collar, and layo¤/plant closing. Sample means and standard deviations for all of the variables are displayed in the cells. In addition, for each group, we also report the t-statistics for testing the equality of sample means between layo¤s and plant closings. 3 Empirical Findings 3.1 Basic Results The main focus of our empirical work is on the wage losses associated with plant closings and layo¤s for various demographic groups. To a large extent, our main results can be seen from our summary statistics in the white collar section of Table 1. Note that we have a much longer history of data than Gibbons and Katz who only used 1984-1986. Since we can now extend the data until 2008, our sample size is large enough to condition on 5 In 1994 and later DWSs, individuals who report a job loss for the reasons other than the …rst three are not asked follow-up questions about the lost job. 6 We restrict to the sample to full time jobs (at least 35 hours per week) because before 1994 the DWSs only provided information on usual weekly earnings (and not hourly earnings) and the full/part time status of the worker’s old job. By limiting our sample to full time workers we attempt to control for hours of work on the old job. 5 speci…c demographic groups. The key variable is the change in the logarithm of the real wage which is shown in the third row. First, focusing on white males one can see that the main prediction of the Gibbons and Katz model holds up. White men lose approximately 6:5% of their wages at plant closings, but this rises to around 9:3% at layo¤s. This can be interpreted as evidence that asymmetric information is important. However, for the other three demographic groups the evidence is very di¤erent. In particular, for African American males and females the contrast is striking with substantially larger wage losses associated with plant closings than with layo¤s. For white women wage losses are very similar between plant closing and layo¤.7 In the rest of Table 1 we present results for blue collar workers, and like Gibbons and Katz, we …nd that wage losses are similar for plant closing as for layo¤. This result holds approximately for all four demographic groups. A key question is why the relative losses at plant closing and layo¤ vary so much across the demographic groups in Table 1. Is it because the losses at plant closing are larger, or is it that the losses at layo¤s are smaller? To add control variables and formally test for di¤erences, we set up the model in a regression framework. The main results are presented in Table 2. The key dependent variable is the change in log wages (i.e. log of post-displacement wage minus log of pre-displacement wage). We regress that variable on black and female dummy variables interacted with layo¤ and plant closing. Note that in columns (1)-(4) this speci…cation is not completely free in that we do not interact race with gender so that the gender e¤ect is constrained to be the same for the two di¤erent races.8 One can see that the results for white collar workers described above depend on di¤erences at both layo¤ and plant closing. In particular, blacks experience both smaller wage losses at layo¤ and larger losses at plant closing than do whites. However, the plant closing e¤ect seems to be the larger of the two and the layo¤ e¤ect is not statistically signi…cant at conventional levels. For women the story portrayed in Table 2 is quite di¤erent. We see only a small di¤erence at plant closing between men and women, particularly after including control variables. However, women experience smaller wage declines following a layo¤. In our simulation results below we will explore why this might be true. Estimating the interacted model in 7 In previous versions of our paper this result was di¤erent as we saw larger declines at plant closing than layo¤. When we added more recent waves of the DWS the point estimate has changed. However, this is not a large deviation from the previous result as the di¤erence changed from plant closing being slightly larger to plant closing being slightly smaller. One possibility is that over time the labor market treatment of men and women has become more similar. However, in our exploration of this idea we found the data to be too noisy to make strong predictions. 8 We do this to increase the precision of the results. 6 column (5) suggests that some additional interactions may exist but the basic results for white men and women remain. A particularly striking aspect of the results is the robustness of the results in Table 2 to inclusion of control variables. While parameters change some from column (1) to (2), all of the relevant coe¢ cients change very little between columns (2), (3), and (4). We view it as particularly surprising that region and one digit industry and occupation controls seem to make little di¤erence in the …nal result. This strongly suggests that the racial and gender patterns we document are not simply due to di¤erences in the sector of the economy or jobs in which workers were employed. One interesting question is whether these results could be explained by Stevens’(1997) …nding that the di¤erence between layo¤ and plant closing occurs because much of the wage losses accompanying plant closing occur prior to the displacement. For this to explain our …ndings, one would need that relative to the other groups, white males would experience larger losses at plant closing prior to displacement. This may seem unintuitive as one might expect white males to be more mobile and thus better able to avoid these losses. We do not have a panel of wages prior to displacement so we can not look at this directly. However, our best evidence is to look at wage di¤erences prior to displacement. One can see in Table 1 that for whites, the layo¤/plant closing gap in pre-displacement wages between men and women is virtually identical.9 Furthermore, we know from Table 2 that the main di¤erence across gender occurs at layo¤ not at plant closing. For blacks, in Table 1 one can see that black men goes in one direction (that is the pre-displacement gap is larger than for white males while for black women it goes in the other (i.e. smaller than for white males). In Table A1 in the appendix we rerun the speci…cations in Table 2 but use pre-displacement wages as the dependent variable. No clear evidence in support of this hypothesis arises. In the rest of Table 2 we present results for blue collar workers. The interactions are virtually all smaller in absolute value than those for white collar workers, and none of the interactions are statistically signi…cant at conventional levels. However, note that many of the signs are quite similar. Even though we can not reject that the coe¢ cients are zero, typically10 we can also not reject that they are di¤erent than the coe¢ cients for white collar workers.11 9 That is, for both white men and white women the di¤erence in predisplacement wages for layo¤ versus plant closing is roughly 0.05 as can be seen in the …rst row of Table 1. 10 The only exception to this in column (3) is the interaction between layo¤ and female. 11 We also rerun the speci…cations in Table 2 for blue collar workers using predisplacement wage as depen- 7 As mentioned in footnote 1, the reason we follow GK and focus our main analysis on white collar workers is because layo¤ decisions for blue collar workers are often determined by seniority. But if this is the case, within a plant’s blue collar work force, any high seniority blue collar worker who is laid o¤ must on average be particularly unproductive. This implies that we should see a large lemons e¤ect for this group.12 We verify this by rerunning the regressions of Table 2 but using a restricted sample of blue collar workers including only those who had tenure on the pre-displacement job for at least 3 years (which is about 44% of the full sample). In the speci…cation corresponding to column (9) of Table 2, the coe¢ cient on layo¤ is 0:041 with a standard error 0:020. So as predicted, there is a lemon’s e¤ect for white male high seniority blue collar workers who are laid o¤. All other coe¢ cients on the interactions remain statistically insigni…cant. 3.2 Employment Discrimination Legislation The GK model assumes that …rms maximize pro…ts and rationally decide whom to dismiss. It also assumes that the only way for an employer to dismiss low quality workers is through a layo¤. In reality, …rms can also dismiss workers by …ring them for cause. It is plausible that …rms can …re the lowest quality workers in the initial period, and when facing a shock, lay o¤ the next lowest quality workers in a later period. Non-economic factors, such as concerns about discrimination lawsuits, can lead employers to alter their methods of dismissal. For example, if workers are more likely to sue for wrongful termination when …red than when dismissed as a part of layo¤ (see for example, Donohue and Siegelman, 1993), then increases in the expected costs to …rms should induce substitution toward layo¤s and away from individual …rings (i.e. lowering cuto¤ in the initial screen for those who are more likely to sue). Oyer and Schaefer (2000) test the hypothesis by exploring the passage of the Civil Rights Act of 1991 (CRA91), which increases the expected costs to …rms of displacing “protected” employees (such as blacks and females). While previous federal employment discrimination legislation typically limited plainti¤ recovery to lost wages, CRA91 allows employees to sue for intentional gender and race discrimination up to $300,000 in punitive damages; furthermore, CRA91 allows employees to claim unlawful termination on the basis of race to sue dent variable. The results are reported in Table A1 in the appendix. Overall the basic pattern is similar to white collar. 12 We thank Charlie Brown for suggesting this idea to us. 8 for unlimited punitive damages. (See Oyer and Schaefer (2000) for more details of the law.) Using data from the 1987-1993 SIPP, they …nd that, relative to whites, rates of overall involuntary job loss (including both layo¤ and …ring) of black men were una¤ected by CRA91.13 However, while black men were signi…cantly more likely to be …red than white men in the pre-CRA91 period, this di¤erence disappeared in the post-CRA91 period. We use CRA91 to look at the implications of asymmetric information on blacks. We do not expect the same argument to work well for women for two reasons. First, the changes in the law a¤ect blacks to a larger extent than for women (and thus represent a larger increase in the expected costs of displacing the “protected”workers). While punitive damages (up to $300,000) apply to employment discrimination for both race and gender, the CRA91 explicitly extends the Civil Rights Act of 1866, which allows plainti¤ alleging racial discrimination to sue for unlimited damages, to cover both on-the-job activities and termination of employment. In other words, the CRA91 essentially removed all limits on damage awards in cases of racial discrimination in termination. So blacks could have the most to gain from the passage of the CRA91. Second, empirical evidence from Oyer and Schaefer (2000) suggests that …rms responded to the CRA1991 by lowering …ring rates for blacks more than whites, but the change in the di¤erence between men and women was statistically insigni…cant. For both reasons, we would expect the law to have a smaller wage e¤ect for women than for blacks. Following the logic of the GK model, a layo¤ (as opposed to a …ring) should be a more negative signal for black workers after 1991 than before. Since we are examining layo¤s rather than …rings, the GK model and the Oyer and Shaefer (2000) results imply that the lemon e¤ect for black workers should be larger after the CRA91 than before. Thus, we would expect wages to fall more dramatically at layo¤s for blacks relative to whites after 1991 than before.14 The DWS data contain information about the year in which workers lost their jobs, by 13 The data used in Oyer and Shaefer (2000) can not separately identify job losses due to plant closing from the other forms of layo¤s (selective downsizings such as abolished positions). 14 There might be other reasons for worrying about changes over time in general. It is widely believed that there has been an increase in the number of layo¤s, especially from white collar jobs in some large corporations, in the early to mid-1990s. Findings in Farber (1997 and 2003) lead support to this belief. He …nds that although the overall involuntary job loss rate did not change substantially from the 1980s to 1990s, there was a decade-long increase in the rate of job loss due to position abolished. If mass layo¤s occur increasingly frequently, then the event layo¤ might become less informative about individual worker’s productivity. Therefore we would expect the di¤erence in wage losses between layo¤s and plant closings to become smaller over time. 9 which we divide the sample into two sub-periods: 1981-1991 and 1992-2008. In Table 3 we repeat the speci…cation of Table 2 except that we interact all of the main coe¢ cients of interest with a dummy variable for post 1991. For white collar workers, the estimates tell a strong story that conforms with our prediction if signalling is important. Relative to whites, the wage hit for blacks associated with a layo¤ is substantially larger after 1991. To put it more literally, prior to 1991 whites had much larger wage declines at layo¤ than blacks, but that di¤erence essentially disappeared after the CRA91. Further evidence that this is not just sporadic comes from examining the other coe¢ cients. None substantially di¤er before and after the Civil Rights Act. We take this as evidence that asymmetric information plays an important role in the labor market. However, one should keep in mind that the con…dence interval for the key interaction found in the …rst row of Table 3 is very wide. It is signi…cant at the 5% level, but also is a very high point estimate. At the very least, we …nd these results highly suggestive that layo¤ appears to be a relatively more negative signal of quality for African American workers after the CRA91.15 3.3 Length of Unemployment Our results to this point have focused only on wages. However, an obvious selection problem arises since we focus only on workers who have been subsequently hired. We are also interested in the overall well being of these individuals which depends not only on the wage impact of displacement, but also the length of the subsequent unemployment spell. To examine this, we follow GK by using a Weibull proportional hazard model to analyze a sample of …rst spells of joblessness.16 The hazard can be speci…ed as t 1 Xi0 e 15 We also ran the regressions for the blue collar sample. The coe¢ cient is of the expected sign, but is smaller than for the white collar workers and statistically insigni…cant from both zero and the white collar coe¢ cient. 16 Each DWS has a question about weeks unemployed since job loss. In 1984 and 1986, it is total weeks of joblessness since displacement. In 1986, there was also information on the number of jobs held since displacement. These two variables allow us to determine the length of the initial spell of joblessness for those employed in their …rst job at the survey date. Since 1988, the question directly asks about weeks unemployed until found a job, i.e the initial spell length. (Due to a survey error, this variable was missing for most observations in 1994.) Workers who had not worked since displacement are always included in the sample just with censored length of the initial spell. We then construct a sample of …rst spells of joblessness from various years subject to the following additional restrictions: workers aged 20-64 who were displaced in previous three years from full-time, private sector jobs not in agriculture and construction industries and had weekly wage no less than $40. Note the subsequent duration analysis is applied to the sample using all years data between 1986 to 2008, but the results are robust when 1988 and 1994 data were removed. 10 where Xi is observable covariates, t is duration and ( ; ) are parameters. The nice aspect of the Weibull model is that the expected value of the log duration is linear so that if Ti represents the duration of unemployment for individual i; @E(log(Ti )) = @Xi : In Table 4 we report estimates of our model using a speci…cation analogous to Table 2. We report the results in terms of change in average log duration ( = ): For clarity, a positive number means that the average unemployment spell would be longer. The main results in Table 4 are similar to those found in Table 2. First, one can see that for white collar workers, plant closing is relatively worse than layo¤ for women and blacks in comparison to white males. We also see that for white males, layo¤ is associated with signi…cantly longer unemployment spells than plant closing. We also again see that plant closing has a much more negative impact on African Americans (and women) than on white males. The result for blue collar workers are quite similar in this case. Other results are somewhat di¤erent than for wage di¤erences in that we …nd that layo¤s are associated with longer unemployment spells for women and blacks than for white males. However, this should not be viewed as surprising. Our results in Table 2 were on wage di¤erences so that we implicitly allow for a …xed e¤ect. The length of unemployment is not analogous because there is nothing like a …xed e¤ect. Thus the comparisons of the level of unemployment by race and gender in Table 4 do not contradict the results in Table 2 which compare wage di¤erences by race and gender.17 It is straightforward to show in a search model (see for example Mortensen, 1987) that one would expect workers with higher wage options to experience shorter unemployment spells. Thus we do not view this result as at all surprising. Another result that does tell a somewhat di¤erent story than before is that in Table 2 we found that white males are the only white collar group for which layo¤ is substantially worse than plant closing. In terms of unemployment spells, the other three demographic groups seem to look similar to white men in the sense that unemployment spells are longer following a layo¤. One explanation for this result is that workers have advanced warning before a plant closing and may begin the search process at an earlier stage so that they are better prepared when it actually happens. 17 While including the wage at displacement is similar, one still …nds lower labor supply by race and gender conditional on wages. So while this might help, it does not completely account for the di¤erences. 11 Overall, we view these results as telling a story similar to those in Table 2. Relative to layo¤s, plant closings are associated with longer spells of unemployment for blacks and women than for white men. Note further that these results suggest that selection bias is not the main driving force behind the wage loss results. To see why, consider a simple reservation wage model in which workers accept a job when the o¤ered wage exceeds the reservation wage. When the reservation wage increases, one would expect the average re-employment wage to go up and the length of the unemployment spell to increase. However, this does not seem to be the driving force behind our estimates. We …nd that in cases in which reemployment wages fall, relative unemployment spells tend to lengthen. For example, the re-employment wage between layo¤ versus plant closing is relatively worse for men than women and the unemployment spell is relatively longer. This suggests that the results are not driven by di¤erent behavior in the reservation wages, but rather by changes in the demand for workers. In the example the relative demand for workers who lost their job from a layo¤ versus plant closing is worse for men than women. 3.4 Discussion To summarize our basic results, we …nd that among white collar workers, plant closings have substantially more negative e¤ects on minorities than on whites. By contrast we …nd that layo¤s seem to have more negative consequences for white men than the other groups. The only group for which wages clearly decline more at layo¤ than plant closing is white males. The question arises as to what models can explain these results. Perhaps the simplest explanation is that individuals from di¤erent demographic groups perform di¤erent types of jobs and thus have di¤erent displacement experiences. We …nd this di¢ cult to reconcile with Table 2 which shows that these e¤ects are remarkably robust to inclusion of industry and occupation controls. The cleanest evidence in favor of asymmetric information can be found in Table 3. As described above, if asymmetric information were important, one would expect the relative wage losses of blacks following layo¤s to increase after 1991 which is precisely what we …nd (although with large point estimates and standard errors). An intriguing aspect of our empirical results is that the negative consequences of plant closing are much worse for African American white collar workers. One explanation for this result is that some …rms discriminate against minorities more than others as in Becker 12 (1971). Minority workers should be more likely to match with nondiscriminatory …rms. In that case, the consequences of these nondiscriminatory plants closing is likely to have strong negative consequences for these workers. By contrast, the same argument would not hold for layo¤s. If discriminatory …rms hire minorities they may be more likely to be lay them o¤. If this is the case, one would not expect to see such an e¤ect in layo¤s.18 Formal development of this model would be relatively di¢ cult if one wants to avoid the unrealistic prediction of perfect segregation across …rms. Incorporating labor market frictions such as search frictions could be used to obtain more realistic predictions. For example one could augment a Burdett and Mortensen (1998) type model by allowing …rms have heterogeneity in tastes for workers by race. However, given the small sample size of blacks in our sample and the added complication of search frictions (in addition to asymmetric information and heterogeneous human capital), we do not address the race gap in the model below. Rather we dedicate the rest of this document to the gender gap. The basic result is that the wage loss is similar for men and women at plant closing, but larger for men at layo¤. In the next few sections we will show that an asymmetric model modi…ed to include heterogeneous human capital can reconcile these results. 4 Model We develop an equilibrium model with asymmetric information as well as heterogeneous worker and …rm types. We present the model in pieces. We …rst present an overview of the model and then discuss the agents and technology. Next we discuss the behavior of the …rms and workers and then describe the equilibrium. We …nish this section with a discussion of the speci…c parameterization we take of the model for the simulations. Derivation of the equilibrium conditions is presented in the appendix. 4.1 Overview of Model Our model is characterized by the following key features: Heterogeneous Human Capital: There are a …nite number (L) of worker types. The key feature of a worker type is that workers are perfectly substitutable with other workers 18 The fact that we don’t see much of a plant closing e¤ect for black workers in the blue collar data adds to the puzzle. Of course this can be consistent with the taste discrimination theory if there is much greater prejudice against minorities in white collar jobs than in blue collar jobs. 13 of the same type, but not perfectly substitutable with workers of a di¤erent type. Heterogeneity in Sectors: There are a …nite number (J) of sectors in the economy. Firms within each sector are identical, but …rms across di¤erent sectors have di¤erent production functions. In general …rms from all sectors hire all human capital types, but some sectors will hire certain types at higher rates than others. As an example, both law …rms and construction …rms hire both blue and white collar workers, but presumably law …rms hire relatively more white collar workers. Asymmetric Information: Within each worker type, workers are heterogenous in ability. A …rm can observe the ability of its current workers, but not the ability of other potential hires. Job Seperations: There are two distinct sources of job separation. The …rst, as in the model of GK, is that when the …rm learns the ability of its workers, it lays o¤ the lowest ability workers. The second source is that the …rm may be hit by a negative productivity shock and choose to lay o¤ some workers or to close its plants. Displacement as de…ned in the DWS occurs not from screening by ability, but rather by some shock to the …rm (plant closing, slack work, or position abolished). Thus the focus of this model is on workers who are displaced due to a productivity shock. If the shock is not severe enough to lead to a plant closing, the …rm chooses to lay o¤ the least productive workers so there will be a lemons e¤ect. At the same time, the nature of the technology shock will in‡uence the wage change since it leads not only to a job separation, but to a decline in the outside opportunity for the worker as well. Thus changes in earnings depend on both the nature and magnitude of shock that hits the economy. They a¤ect wages through equilibrium e¤ects as well as through an employer’s inference of worker quality from the layo¤. 4.2 Agents and Technology As mentioned in the previous section, we have J sectors and L di¤erent types of workers. We label the sectors by j for j = f1; :::; Jg and label labor types by ` for ` = f1; :::; Lg: We use Y` to denote the ability of an individual of type `. A key aspect of the model is asymmetric information. When …rms make o¤ers to workers, they do not know their ability. 14 However, after employing the worker for a period, they learn it. At that point the …rm may choose to lay o¤ unproductive workers. As in GK we need to give the incumbent …rm some comparative advantage in keeping the worker. We assume that during the …rst period that a worker works for a …rm, his productivity is Y` : For the second period and beyond, his productivity at that …rm is Y` where analysis. If > 1: This parameter plays an important role in the = 1; no workers would be retained in the equilibrium we examine (unless there is an upper bound to the support of productivity that has a positive probability of occurring). The reason is because the worker’s outside wage is essentially the expected productivity of all retained workers. With a smooth cdf, the worst retained worker must be lower than the productivity of the average retained worker, so one can not …nd an equilibrium. With > 1, workers have more value in the current …rm than in outside …rms so that some workers will be retained. In general, the larger is , the larger the fraction of workers who will be retained. We have modeled as if it is …rm speci…c human capital. Alternatively we could interpret 1) Y` as a training cost (or other type of hiring cost) that results in lower productivity ( during the …rst period.19 Let Ht`j be the aggregate human capital of type ` working in sector j at time t. We formally de…ne this in the appendix. It is basically the sum of the productivity across all workers of type ` who work in sector j at time t: Let Htj = (Ht1j ; :::; HtLj ) be the vector of inputs for a sector. The production function for sector j at time t takes the form Gtj (Htj ) : We assume further that each …rm within the sector is large enough so that the law of large numbers holds (so that average productivity is all that matters) and that G has constant returns to scale.20 There are a large number of …rms within each sector, free entry, and no search frictions, so the labor market for new hires will be competitive. 19 Gibbons and Katz (1991) argue similarly that their analogue could represent …rm speci…c human capital, mobility costs of the worker, hiring cost of the new employer, or a …ring cost from the old employer. 20 We model each sector as being composed of a large number of smaller …rms. With constant returns to scale we can focus on the aggregate human capital production function-and each of the smaller …rms will look identical. It is also easier to think about what an entering …rm would do. With increasing or decreasing returns this would be much more complicated. 15 4.3 Timing Two di¤erent aspects of the timing are important. The …rst is the de…nition of the di¤erent periods and the second is the timing within a period. We discuss these in order. Finite period versions of our model can have implications on wage changes that is driven by the fact that a worker is close to retirement. To avoid this problem, we allow for an in…nite number of periods although only the …rst few periods are of interest. We essentially follow a cohort of workers and …rms across time. The timing is de…ned as: Period 1: Workers are hired by …rms, but the …rms have only limited information about worker quality. Period 2: After learning about worker quality from the …rst period, during the second they only retain workers above a minimum threshold. Period 3: The economy is potentially hit by a shock that leads …rms to either close or lay o¤ workers. These workers are then rehired by other …rms. Period 4 and beyond: Nothing additional happens as workers continue to work for …rms. In terms of more speci…cs about the timing, the production function remains the same in periods 1 and 2. However, we assume that in period 3 the production process is potentially hit by a shock which may or may not a¤ect di¤erent sectors. After period 3, the parameters of Gtj remain …xed at those values forever. Thus the production function only changes between periods 2 and 3. In our model, …rms dismiss workers at the beginning of period 2 and then again at the beginning of period 3. It is important to point out that we view these as distinct phenomena. During period 1, …rms learn about the quality of a worker. They choose not to retain the worker because they have fallen below the screening value of the …rm. The event that leads to the separation in this case is the …rm learning about the quality of the worker. The retention decision at the beginning of period 3 is quite di¤erent. These workers have already made it beyond the initial screen, so the event that leads to workers being laid o¤ is an adverse shock to the …rm.21 Given the wording of the questions in the DWS, we assume that the data correspond to the latter type of dismissal rather than the former. 21 Of course it is still true that …rms typically do not lay o¤ all workers, but choose the ones of lower ability. Thus one sees a lemon e¤ect for both types of dismissals. 16 Since they are not of primary concern, workers who are below the cuto¤ value in period 2 leave this part of the economy permanently (which may seem reasonable given that we are focusing on the white collar sector). However, since our main goal is to focus on the displaced workers, we allow those who are not retained in period 3 to be rehired within this part of the economy. Note that in principle a …rm could …re a worker in period 4 that it hired in period 3 after learning about his ability. However, in practice this should happen only very rarely. The workers ability was high enough so that in period 2 the worker was retained, and now we are cutting o¤ the right hand tail so the wage premium to them is even smaller. Thus for computational reasons, we ignore this possibility and just assume that workers hired during period 3 will remain forever. Thus for t > 3 all that changes is that the productivity of the new hires improves and no worker mobility takes place.22 The timing within a period is the following: 1. The …rm observes the ability levels for all of its workers from the previous period. 2. It chooses whether to lay workers o¤ or to o¤er them a wage to continue. We do not allow the …rm to make an o¤er that it knows will be turned down. 3. Outside …rms make o¤ers to both the laid o¤ and retained workers. 4. Workers decide which o¤er to take. 5. Workers participate in the production process. Note that in part 2 we are not allowing …rms to make low ball o¤ers to workers they do not want, but are forcing them to lay them o¤. Since …rms are indi¤erent between laying o¤ a worker and retaining him with a wage that is lower than the market wage, there will be multiple equilibria in the model. GK describe this class of equilibria. We focus only on the equilibria in which …rms never pay a worker lower than their outside option.23 22 By assuming that all workers who are laid o¤ in period 2 disappear from the market, we have imposed that these workers can not be rehired either. 23 One could make some assumptions to guarantee that this condition holds. Alternatively, one could analyze all of the equilibria. However, our goal is to show that the model is consistent with the data, rather than to try to distinguish between equilibria. We strongly suspect that more than one equilibrium can reconcile the data. 17 4.4 Behavior of Workers and Firms Worker behavior is not particularly interesting in this model. Workers are risk neutral and simply make wage choices to maximize their present discounted value of income. We implicitly assume that workers do not know any more about their own ability than an outside …rm as we are not allowing outside …rms to construct contracts that might induce the worker to reveal their own ability. Firms are much more interesting as they have essentially three things to decide a) which workers to lay o¤ at the beginning of the period, b) the wage to o¤er the workers that are retained, and c) the wage to o¤er to outside workers. First consider the layo¤ decision. Firms strictly prefer higher ability workers to lower ability workers. Thus, for each human capital type in each sector there exists a cuto¤ value such that …rms only retain those workers whose productivity is above this cuto¤ value. Denote these cuto¤ values by yt`j so that workers are laid o¤ if Y` < yt`j . Next consider the o¤ers that …rms make to outside workers. We assume that outside …rms know the cuto¤ levels of other …rms in the economy, but do not know the level of productivity of individual workers. Thus, they make o¤ers based on the expected present value of revenue from the worker conditional on Y` < yt`j for workers who were laid o¤ and conditional on Y` yt`j for workers who were retained. In theory workers who are above the cuto¤ may be potentially poached by other …rms. While this will not happen in equilibrium, this potential “poaching”plays a key role in the analysis as it determines the outside wage. Finally we consider the retention o¤er. Given the timing of the model, the current …rm has no incentive to pay more than a tiny bit over the outside wage. Note that this means that the …rm is going to receive rents on many workers in these periods due to the training costs . However, since the hiring market is competitive, the expected value of these rents was essentially competed away in the form of higher wages given to the worker during the …rst period. 4.5 Equilibrium Equilibrium is characterized by the following four criteria: Outside wages are determined so that …rms earn zero pro…t on average for a worker. Inside wages (after the …rst period a worker has worked for a …rm) are chosen by the 18 …rm to make a worker indi¤erent between staying or leaving. Firms retain workers for whom it is pro…table to do so.24 Workers make employment choices to maximize their expected present discounted value of earnings. The details and derivation of the equilibrium conditions are provided in the appendix. 4.6 Parameterization in Simulation We use a CES production function Gtj (Htj ) = L X t`j Ht`j `=1 !1 with the number of sectors (J) equal to …ve. We assume that individuals of di¤erent genders are di¤erent worker types and allow for 10 worker types -…ve for each gender. A key aspect of the model is that di¤erent types of human capital are used di¤erently in di¤erent …rms. In particular we will assume that originally gender, 1`j 2 f1; 2; 3; 4; 5g: For each …rm and for each will take on each of these …ve numbers. Furthermore, the model is completely symmetric so for each labor type `; the share parameter takes on each of the 5 values for some sector.25 We simulate shocks to t`j to occur between periods two and three. We model three di¤erent types of shocks to this economy. 1. Modest proportional change in 2. Large change in t`j t`j by j (sector shock) by j causing sector to disappear (sector shock-plant closing) 3. Proportional change in t`j by ` (human capital type shock) The …rst type is a “sector speci…c shock”in which t`j falls for all values of ` in a given sector j. This can be thought of as a sector speci…c productivity shock, but could also be viewed as a demand shock to the sector. If the shock is large enough, the sector will 24 Given the assumptions we have made, one will only lay o¤ the worst workers because the outside market is identical for all workers. Thus there will always be a “lemon” e¤ect. 25 Making it symmetric substantially lowers the computational cost because of similarities in behavior across groups. Without this, the model would be much harder to estimate. 19 disappear which we view as analogous to a plant closing. We also consider a “human capital type shock”in which t`j falls for all values of j for a given human capital type `. We view this as “skill biased” technological shock that a¤ects the productivity of a particular skill type. An example of this type of shock is a technological discovery that is substitutable with type ` workers (such as the improvement of word processing software for typists). 5 Slack Work and Position Abolished A very nice thing about our data is that it contains something akin to the di¤erence between sector shocks and human capital shocks. To be in our layo¤ sample, an individual reported that the reason of job loss was either “slack work”or “position abolished.”We view these as mapping into our model well with the human capital type shock corresponding to position abolished while the …rm shock relates to slack work. Intuitively in the model, if the shocks are of similar magnitude, workers who are laid o¤ through a human capital type shock will likely experience a much larger wage loss. The reason is simply that if the shocks occur at the sector level, workers can move into a di¤erent sector that did not experience the shock. Due to the equilibrium e¤ect, their wage will still fall but by switching sectors they avoid the …rst order e¤ect of the shock. By contrast, workers laid o¤ because of a shock to their human capital can do no such thing. Their productivity, and analogously their wage, has fallen at all …rms.26 The model does not formally imply that the wage losses associated with a human capital shock will be larger than those from a sector shock. However, intuitively one might expect this to be the case and we consistently …nd it for reasonable parameter values. In particular, if the magnitudes of the shocks are similar, we consistently …nd larger wage losses for the human capital shock. In fact, in the simulation below, the sector shock is considerably larger than the human capital shock, but the wage losses are almost twice as large for the human capital shock. Thus our intuition from the model is that we would expect to see substantially larger wage losses for those that lose their jobs by position abolished as opposed to slack work. We …rst informally look at this issue in the data by presenting analyses analogous to Table 2, but distinguishing a layo¤ between “slack work” and “position abolished.” These 26 As previously mentioned in footnote 3, in their paper GK discuss the fact that a plant closing could be associated with poor local labor markets and thus be associated with worse outcomes. This is related to our model if one interprets sectors j as local labor markets. 20 results are shown in Table 5. Looking …rst at white collar workers, one can see that in every speci…cation, the point estimates indicate that every group experiences a larger fall in wages for position abolished than for slack work. At the bottom of the table we present the p-value of a joint test as to whether the coe¢ cients on slack work and position abolished are jointly the same. One can see that this hypothesis is strongly rejected in every speci…cation. Once again, one also sees the striking result that neither industry nor occupation is important in explaining these results. We believe these results in Table 5 are of interest in their own right as (to our knowledge) they have not been previously discussed in the literature on displaced worker e¤ects. Although we do not show them explicitly, we also looked at slack work versus position abolished using the Weibull proportional hazard model. We …nd that for white collar workers, position abolished leads to longer unemployment spells although the e¤ect is not statistically signi…cant. GK argue against using blue collar workers because their layo¤ decisions are determined in large part by collective bargaining decisions. This is an argument as to why one might not expect the lemons e¤ect to arise for blue collar workers, but has little to do with the distinction between slack work and position abolished. Our argument for slack work versus position abolished should hold for blue collar workers as well. We verify this in the blue collar part of Table 5. One can see that we …nd similar orders of magnitude in the results. The standard errors for blue collar workers are somewhat bigger leading to larger p-values.27 In fact the similarity between blue and white collar workers in Table 5 is remarkable. The simplest summary statistic is the di¤erence between the coe¢ cients on slack work and position abolished in columns (4) and (8). This is 0.040 in both cases (with a somewhat larger standard error for blue collar workers). Looking at the comparison with plant closing the results are again qualitatively similar. Workers experience larger wage losses at position abolished than either plant closing or slack work. The di¤erence between plant closing and 27 Note that we strongly reject the null hypothesis in column (4), but do so only at the 10% level for column (3). There are two major di¤erences between the two columns. First the null hypothesis is di¤erent since column (3) allows for more interactions. Second, the test is di¤erent since the test in column (4) is one- dimensional while the test in column (3) is a joint null about 3 sets of coe¢ cients. Given that we essentially have a one-sided alternative and that the results for the di¤erent groups basically all point in the same direction (with white women being weaker than the other groups), one might expect this test to be particularly weak. To see whether the smaller p-value comes from the more ‡exible null or the more powerful test, we reestimated the model with the same null as in column (3) but a one-dimensional test for the di¤erence between slack work and position abolished and we obtain a p-value of 0:0313 suggesting that it is the di¤erence in the test that leads to the di¤erence in p-values. 21 slack work is not statistically signi…cant for either group.28 Following the logic of our argument, if the magnitudes of the two types of shocks were similar, we would expect to see workers who are laid o¤ due to position abolished to be more likely to switch occupations than worker displaced for other reasons. To verify this, we looked at the probability of changing one digit occupation after displacement by job loss reason. The results, which are reported in Table 6, con…rm this prediction. For both blue and white collar workers, those who lost their jobs due to position abolished are more likely to switch occupations that those who lost their jobs due to plant closing or slack work. 6 Reconciling Model with Data We now turn toward reconciling the model with the data. With this in mind, we chose six moments in the data that we hope to match based on column (3) of Table 5. As mentioned above we focus on non-black workers only. By gender, we construct the change in wages at displacement for three di¤erent types of displacement: plant closing, position abolished, and slack work. Speci…cally we calculate the sample average for the other control variables in the model and give this sample average to all groups. The 6 moments we try to …t are presented in the data columns Table 7. One can see three key features of the data that we will show can be explained by the model. The di¤erence in wage loss by gender for plant closing is very small, and is in fact not statistically distinguishable. By contrast, for the two types of layo¤ we see substantially larger losses for men than for women.29 The second key feature of the data that we plan to match is the di¤erence between slack work and position abolished. We show that with heterogeneous human capital in the model it is straightforward to match this feature of the data. The third feature is that slack work is worse than plant closing for men, but this result goes the other way for women. We should point out that this …nal feature for men is not statistically signi…cant nor robust across columns in Table 5. We will discuss this issue more below in interpreting the results. In the simulation we assume that log(Y` ) is normal with standard deviation 28 g where g When we look at the duration model, we …nd the point estimate goes in the opposite direction, but is not statistically signi…cant. 29 While this can not be seen directly from either Table 5 or from Table 7, the gender di¤erence for position abolished is statistically signi…cant with a p-value of 0.033 and the gender di¤erences for position abolished and slack work are jointly signi…cant with a p-value of 0.06. 22 denotes gender. Allowing the information to vary by gender is important in …tting the data. We discuss below why 2 g might vary across gender. Thus the di¤erence in wage losses across groups is identical to what the gender dismissal reason would suggest. A key piece of this thought exercise is that we restrict the model in another important way by assuming that the gender productivity does not interact with sector at all because the results in Tables 2 and 5 suggest that industry and occupation di¤erences between men and women do not play a crucial role in explaining the results. For this reason we restrict the model so the “industry/occupation ”composition is the same for men and women. Formally, let ` = f1; :::; 5g denote men and ` = f6; :::; 10g denote women. We impose that for all states of the world, t`j = t(` 5)j for ` > 5: Thus shocks hit men and women in exactly the same way. So, for example, a human capital type shock that hits skill group ` = 3 for men will also hit group ` = 8 for women. This means in the end we have three data moments that di¤er by gender but in the simulation we only have one parameter that di¤ers by gender. Thus even though there are many parameters in the model, we are certainly not guaranteed to be able to …t the data. We assume that there are 16 states of the world in period 3. With probability 0.97 no shock is experienced. Then we put 0.2% probability on each of the other 15. These other 15 correspond to 5 of each type of the shock above. That is because we have 5 sectors, each sector has a 0.2% probability of getting hit by a modest proportional change in a 0.2% probability of getting hit by a large change in j` j` and also which causes it to shut down. For each gender, we also have 5 labor types and each has a 0.2% probability of getting hit by a proportional change in j` . The model has many di¤erent margins to complicate it. To ease the computational burden, we restrict turnover in a few ways. First, as mentioned above in Section 4, we restrict the amount of multiple layo¤s in two ways: a) workers that do not make the initial cut in period 2 disappear from this part of the labor market, and b) we do not allow for additional layo¤s after period 3. Additionally, for the sector speci…c shock we only allow for layo¤s in the sectors which experienced the shocks. In principal, since the outside option has changed, there could be some layo¤s for other …rms as well. However, this should be small, not of primary interest in this analysis, and incorporating it would make the model substantially more di¢ cult to solve. 23 This leaves us with essentially 4 types of parameters: the elasticity of substitution ( ), the training cost ( ), the standard deviation of the unknown component of ability ( g ), and the magnitudes of the shocks. Despite the simpli…cations described in the previous paragraph, we found the model to be very di¢ cult to solve. We need to solve for the …ring, hiring, and wage decisions for di¤erent sectors in di¤erent states of the world for di¤erent times. Thus this is a high dimensional problem and is non-smooth as every sector turns out to be at a corner solution either in terms of layo¤s or hiring or both at every point in time. We do not know which constraints will bind ex-ante. While ideally we would simultaneously solve for the parameters of the model to …t the data and solve for the equilibrium of the model, this did not work well. While we were able to …nd a set of parameters that …t our data perfectly, the equilibrium had some odd features. We would, of course, be happy to provide those details to any interested readers. The strange features primarily arose from the fact that we found = 1:011 for men but 1.16 for women and g = 0:005 for men but 0.0001 for women. This made the behavior of men and women look very di¤erent. We see no reason for to di¤er by gender and prefer results where we …x it to be the same by gender.30 Instead we chose a more adaptive approach to …nd solutions for the model. As a practical matter, we found that when a plant closes, the only parameter that matters substantially for the simulated results is the elasticity of substitution. Furthermore, in practice, the wage loss with plant closing in the simulations will be very close across gender-and the di¤erence in the data is not statistically signi…cant. Thus we do not try to match this di¤erence exactly, but rather choose a value of that matches approximately. This process led to the value = 0:15: We then searched over other parameters for a version of the model that was close to the data. While we have a number of di¤erent parameters, only g varies by gender. Thus again we should highlight that our goal is to show that the model can reconcile the data (which it is certainly not guaranteed to) and to give a sense of the parameter values that do that. The results of this simulation are presented in Table 7. These simulation results are very close to the four moments that we are trying to …t. The parameters that lead to this …t are shown in the bottom panel of Table 7. The standard deviation of unobserved ability is substantially higher for men than it is for women. In the simulation, this leads the “lemon” e¤ect to be larger for men than for women. It essentially embodies the idea that a layo¤ is 30 We never replicated the data exactly when a case does not exist. was …xed across gender, but this does not mean that such 24 a stronger signal for men than it is for women. It is important that the reader not take this parameter too literally. Another interpretation of the phenomena is that the decision to lay o¤ an individual is more complicated than in the model and involves other factors beyond just pure ability such as the value of home production which could change the threshold.31 If it were the case that the decision to layo¤ o¤ a man was based purely on his market ability, but the decision to lay o¤ a woman depended on both market ability and the value of non-market time, then the signal for a man would be stronger than the signal for a woman. We don’t view this as a fundamentally di¤erent model than the one we have written down, but rather view it as a potential reason why layo¤ is a more informative signal for men than it is for women (which is the essence of the higher value of A second feature of the results is that the value of g in the results).32 is quite large. Note that in the model, this does not lead to a higher measured return to tenure in a log wage regression because the …rm pays the outside wage. High values of can not be avoided when asymmetric information is important and retention rates are reasonably high. Since a …rm pays workers with identical observable characteristics the same-and the same as their outside market, they have a strong incentive to …re the workers with the worst unobservable attributes. If a majority of workers are retained it must be that these worst workers are relatively more productive for the current …rm than for the outside labor market. Thus it must be the case that some combination of hiring costs, training costs, or speci…c human capital are important. Any of these can be interpreted as : We simulate the e¤ects of three types of shocks. With plant closing, one sector disappears and all workers leave for another sector. For the sector shock, sector j is hit by a shock that leads to a 1.5% decline in the factor loading parameter. Formally this means that 3`j = n 0:985 2`j 2`j j=j otherwise : We see in Panel B of Table 7 that this leads 5.1% of men and 3.1% of women who were employed in this sector to be laid o¤. Note that this is a somewhat larger shock than it might initially appear to be. Part of the CES production function can be written as 31 This would involve a more complicated model in which home production was the relevant outside option for some. Individuals with higher value of home production could require more compensation and be laid o¤ earlier. 32 Note that one thing that can not be directly compared to Table 5 is the di¤erence in the relative layo¤ rates. In this table we present the layo¤ rates conditional on a shock occuring. Table 5 presents these results unconditionally. Thus by changing the relative frequency of the shocks we could …t this feature of the data. 25 H = 1= H : So multiplying by 0.985 is analogous to multiplying worker productivity by 0:9851= = 0:904: For the human capital type shock we consider a decline of 0.3% in the factor loading parameter. That is, if type ` 3`j 5 is hit by the shock this means that n 0:997 2`j ` = ` or ` = ` + 5 = : otherwise 2`j Since this is a negative shock to all sectors, the layo¤ probabilities are lower: 1.5% for men and 1.1% for women. This multiple is analogous to a 0:9971= = 0:980 multiple on human capital. Note that while the decline in productivity of the shocks is substantially smaller in this case, the wage penalty is much larger for position abolished than for slack work. This is a more general feature that comes out of the model rather than being an artifact of the particular normalization. As mentioned above, when a sector is hit by a shock, other sectors are not. Therefore a worker can move to a sector that was not hit. However, this is not possible for a human capital shock as the worker’s productivity has declined everywhere. We have shown in this section that our model is consistent with the data. Since the model is highly parameterized there might be a question of whether the fact that we can reconcile the data is particularly surprising or interesting. On this point we make three main comments. First, an important …nding in the empirical section is that occupation and industry play little role in explaining the gender di¤erence in plant closing and layo¤. Thus we do not allow this to reconcile the di¤erence. Second, heterogeneous human capital is crucial to explain the results. For both genders we see larger wage losses at position abolished than for slack work. The third and most nuanced is understanding the role that asymmetric information plays in the model. Given our parameterization, asymmetric information is important for two reasons. First, it is only g that can explain the di¤erence between men and women. Second, it is necessary to explain the fact that the loss for slack work is larger than the loss at plant closing for men. However, both of these might be regarded as weak support. We do not have a good theory as to why the standard deviation of unobserved ability varies by gender but other parameters do not. We could presumably allow all of the parameters to vary by gender and still be able to reconcile the results. The second reason is that in the model, the only way to explain the fact that men experience larger wage losses for slack work than for plant closings is because of the lemons e¤ect. However, as noted above, the di¤erence here is not statistically signi…cant nor robust across speci…cations. A better view of the data 26 is that the wage loss for slack work and plant closing is roughly the same. However, note that this is not what the model would predict in the absence of asymmetric information. Since a plant closing represents a much larger shock to the sector, it would lead to a larger wage loss. Thus the fact that slack work is similar to plant closing can be taken as evidence of asymmetric information. The strength of this evidence depends on one’s priors of how large this di¤erence might be. At the very least, we take the simulations as evidence that asymmetric information could be important and that the results for women should not be taken as evidence against this idea. 7 Conclusions In a seminal paper Gibbons and Katz (1991) develop and empirically test a model of asymmetric information in the labor market. They derive an implication of their model that if asymmetric information is important, one should expect a larger fall in earnings at layo¤ than at plant closing. Using the Displaced Worker Survey, they show this implication to be true for men. We revisit this question making use of the many more years of data that are available now. We test the hypothesis on four di¤erent demographic groups. The GK predictions stands for white men. However, for two of the four groups (black men and black women), we …nd the opposite of the Gibbons and Katz prediction; plant closings lead to more negative consequences than do layo¤s. For white women, the losses at plant closings and layo¤ are very similar. We show that these di¤erences occur for two reasons. First, white men experience larger earnings declines at layo¤ than the other groups. Second, black workers experience substantially larger decreases in their earnings at plant closing than do whites. We document four other aspects of the data. First, the basic results are remarkably robust to occupation and industry controls. Second, following Oyer and Schaefer (2000), we make use of the passage of the Civil Rights Act of 1991 (CRA91) to test an implication of the model. We show that black workers experience a relatively larger loss in earnings at layo¤s after 1991 than before which is consistent with asymmetric information. We think this is the strongest evidence in favor of asymmetric information. Third, we demonstrate similar patterns when we look at the length of unemployment spells following displacement. Fourth, we document for the …rst time in the literature that the two types of layo¤s reported in the DWS data have very di¤erent features in terms of earnings losses. In particular, we 27 …nd that losses in earnings are greater when layo¤ is associated with “position abolished” then when it is associated with “slack work.” We develop a model that incorporates heterogeneous human capital into an asymmetric information framework based on GK. The model includes di¤erent types of …rms and di¤erent types of workers. In the model, once a worker has worked for a …rm for a period the current …rm knows his/her skill level, but outside …rms do not. We model layo¤s and plant closings as resulting when shocks hit …rms in which the workers work. We then numerically simulate the model and show that one can …nd parameters of the model to make it consistent with the data. We simulate the model and show that it captures the key features of the data. In particular, asymmetric information plays the key role in explaining the gender gap in the model and this result is reconciled with the fact that in the data wage losses at both types of layo¤ (but not plant closing) di¤er substantially by gender. The basic idea is that for some reason, being laid o¤ is a relatively more important signal for a man than for a woman. The point estimates of the model suggest that the standard deviation of unobserved ability (to the outside …rm) is larger for men than for women. As we cautioned earlier, however, one does not necessarily want to take this literally. It could be standing in for some other feature of the data such as the layo¤ decision is more complicated for women than for men so that unobserved ability is a relatively more important factor. One interesting extension can be to extend the model to allow for layo¤ decisions to depend on both market ability and the value of non-market time, which in general can di¤er by gender, and then investigate the mechanisms directly. Another extension of the analysis would be to incorporate heterogeneity of taste discrimination across …rms in the model to explain the fact that blacks su¤er greater wage penalties from plant closings than whites. Both of these extensions are very interesting and important. More generally our model is overly simple in many dimensions. However, sorting out these alternatives require more data than we get in the DWS where we essentially just have the six numbers in Table 7. Other useful data sets are available and we leave this work for future research. Gibbons and Katz acknowledge at the end of their paper that “Unfortunately, the nature of asymmetric information seems to imply that direct empirical tests of its importance are not possible, so indirect tests of the kind presented here may be all that is possible.”We share 28 the sentiment and agree that our data are not rich enough to precisely distinguish between all potential explanations. However, we have provided additional evidence to bear on these issues and our results strongly suggest that heterogeneous human capital is important. While the evidence in support of asymmetric information is weaker, in our simulations it does play an important role. We hope that alternative data sources can be found which will shed more light on these important issues in the future research. 29 Appendix: Derivation of Equilibrium Conditions We start by de…ning the aggregate human capital inputs. Let f`j denote the fraction of workers of type ` who work in sector j at time 1 and `hj the fraction of `-type workers who worked for an h sector …rm in periods 1 and 2, were not retained by h in period 3, and were hired into sector j in state. In principle …rms can “poach” retained workers from other …rms and we should include notation to account for this. Since this does not happen in equilibrium we do not explicitly account for it. During the …rst period, Y` is not revealed so there will be no sorting among workers into sectors. Thus the aggregate human capital takes the value (A-1) H1`j = f`j E (Y` ) : During the second period some workers can be laid o¤. The ones who remain will be more productive by the factor so that H2`j = f`j E Y` 1 Y ` (A-2) y2`j where 1 ( ) is the indicator function. In the third period in addition some new workers can be hired from other sectors H3`j = f`j E Y` 1 Y ` y3`j + J X `hj E(Y` h=1 j y2`h < Y` (A-3) y3`h ): We allow …rms to both dismiss workers (if y3`j > y2`j ) and hire new workers (if `hj > 0): In our simulations we allow …rms to do both, but never found a case in which they do. Finally for all periods after period 3, Ht`j = f`j E Y` 1 Y ` y3`j + J X h=1 `hj E( Y` j y2`h < Y` y3`h ): (A-4) We de…ne some additional notation. Let ye`H be the conditional expectation of the pro- ductivity of a worker of type ` who has experienced labor market history H: During the second period, a worker who worked in sector j during period 1 and was retained will have history H = j and ye`j = E Y` j Y` y2`j : During period 3 and beyond, a worker who worked in sector j will have history H = jr where r is a dummy variable indicating whether the worker was retained. For example one history H may be that a worker started at …rm j and were retained throughout, then H = j1 and ye`j1 = E Y` j Y` 30 y3`j : Another potential history H is that an individual was retained by a …rm in sector j in period 2, but then laid o¤ by that …rm in period 3. In that case, H = j0 and ye`j0 = E Y` j y2`j < Y` It is easiest to solve this model by working backwards from period t y3`j : 4. Since nothing changes after period 4, all of these periods will look identical. Consider a …rm j trying to poach an ` type worker from another …rm. Then for any history H; the expected marginal value of workers who are hired in period 4 is @G4j (H4j ) X ye`H + @H4`j t=5 1 where t 4 ye`H @Gtj (Htj ) @G4j (H4j ) = ye`H @Ht`j @H4`j 1 + 1 is the discount rate since Gtj and Htj do not vary over time (after period 4). Wages will be constant across time and the wage will be determined by the best outside opportunity. This yields that the outside wage during period 4 for an individual of type ` with labor market history H is wt`H = for t max ye`H j=f1;:::;Jg @Gtj (H4j ) @Ht`j 1 + (A-5) 1 4: Employers must keep the workers indi¤erent between leaving and staying and the outside wage does not change after period 4, so employers will pay the constant outside wage wt`H in all periods after period 4. Now consider period 3 in which there will be some turnover. We …rst consider the market for workers of type ` from a …rm in sector h who are retained (Y` > y3`h ) : If a …rm in sector j considers hiring them, it will make expected pro…t @G3j (H3j ) ye`h1 @H3`j w3`h1 + @G4j (H4j ) ye`h1 @H4`j 1 w4`h1 where w3`h1 is the outside wage for such a retained worker. Since the market is competitive between …rms both within and between sectors, w3`h1 = max j=f1;:::;Jg @G3j (H3j ) ye`h1 + @H3`j 1 @G4j (H4j ) ye`h1 @H4`j w4`h1 : (A-6) In determining the cuto¤ y3`j , the …rm has to be indi¤erent between keeping and retaining a worker with Y` = y3`j . This yields that 8 < w3`j1 + y3`j = max @G3j (H3j ) : + @H3`j 1 w4`j1 1 @G4j (H4j ) @H4`j ; yj`2 Finally, consider a worker of type ` who was displaced (Y` be paid his best opportunity in the competitive market: 31 9 = ; : (A-7) y3`h ) from …rm h: He will w3`h0 = max j=f1;:::;Jg @G3j (H3j ) ye`h0 + @H3`j 1 @G4j (H4j ) ye`h0 @H4`j (A-8) w4`h0 where the expression on the right hand side comes from the zero pro…t constraint for each …rm. For workers who are above the cuto¤s during period 3, the …rms that retain them will receive rents. Since these rents will be bid away by …rms when hiring workers in earlier periods we will need to keep track of them. We de…ne the expected rent as a function of a worker’s ability y as: `j (y) = E 1 y y3`j @G3j (H3j ) y @H3`j w3`h1 + @G4j (H4j ) y @H4`j 1 w4`h1 : Next consider period 2. In this case workers who are not retained leave this labor market (and presumably go to the blue collar sector). We write the production function in sector j for periods 1 and 2 as Fj . Using notation analogous to above, we let w2`h be the outside wage for workers who are retained in sector h. The equilibrium outside wage is de…ned so that expected marginal pro…t is zero: w2`h = @G2j (H2j ) ye`h + E ( j=f1;:::;Jg @H2`j max (Y` ) j Y` `j (A-9) y2`h ) : A …rm chooses the cuto¤ value y2`h so that it is just indi¤erent about retaining the worker @G2j (H2j ) y2`h @H2`j w2`h + E ( `j (A-10) (y2`h )) = 0: Note from equations (A-9) and (A-10) one can see the importance of > 1: The …nal part of equation (A-9) involves future pro…ts for the average worker while the expression (A-10) involves future rents from the marginal worker which must be lower. One needs > 1 to …nd a nondegenerate value of y2`h to solve these equations. Finally during the …rst period, in equilibrium …rms will o¤er wages so that the marginal pro…t is zero w1`j = E (Y` ) @G1j (H1j ) @G2j (H2j ) + Pr(Y` > y2`h ) ye`j @H1`j @H2`j w2`j + E ( `j (Y` ) j Y` y2`h ) : (A-11) Workers choose …rms to maximize their expected present value of earnings. We assume that workers have no more information about their ability than do the …rms. We do not explicitly 32 give the form for the present value of earnings, but all of the components of the wages have been de…ned. The equilibrium of the model is characterized by equations (A-1)-(A-11). 33 References Abraham, Katharine G., and James L. Medo¤. 1984. Length of Service and Layo¤s in Union and Nonunion Work Groups. Industrial and Labor Relations Review 38: 87-97. Aigner, Dennis J., and Glen G. Cain. 1977. Statistical Discrimination in Labor Markets. Industrial and Labor Relations Review 30: 175-187. Akerlof, George A. 1976. The Economics of Caste and of the Rat Race and Other Woeful Tales. Quarterly Journal of Economics 90: 599-617. Altonji, Joseph G., and Rebecca M. Blank. 1999. Race and Gender in the Labor Market. In Handbook of Labor Economics, ed. Orley Ashenfelter and David E. Card. New York: Elsevier Science, North-Holland. Altonji, Joseph G., and Charles R. Pierret. 2001. Employer Learning and Statistical Discrimination. Quarterly Journal of Economics 116: 313-350. Arrows, Kenneth J. 1973. The Theory of Discrimination. In Discrimination in Labor Markets, ed. Orley Ashenfelter and Albert Rees. Princeton: Princeton University Press. Becker, Gary S. 1971. The Economics of Discrimination, 2nd Edition. Chicago: The University of Chicago Press. Burdett, Kenneth, and Dale T. Mortensen. 1998. Wage Di¤erentials, Employer Size, and Unemployment. International Economic Review 39: 257-273. Coate, Stephen and Glenn C. Loury. 1993. Will A¢ rmation-action Policies eliminate Negative Stereotypes? American Economic Review 83, 1220-1240. Farber, Henry S. 1997. Has the Rate of Job Loss Increased in the Nineties? Brookings Papers on Economic Activity: Microeconomics 55-128. Farber, Henry S. 2003. Job Loss in the United States, 1981-2001. Working Paper No. 471, Industrial Relations Section, Princeton University. Gibbons, Robert S., and Lawrence F. Katz. 1991. Layo¤s and Lemons. Journal of Labor Economics 9: 351-380. Greenwald, Bruce C. 1986. Adverse Selection in the Labor Market. Review of Economic Studies 53: 325-347. Krashinsky, Harry. 2002. Evidence on Adverse Selection and Establishment Size In the Labor Market. Industrial and Labor Relations Review 56: 84-96. Laing, Derek. 1994. Involuntary Layo¤s in a Model with Asymmetric Information Concerning Worker Ability. Review of Economic Studies 61: 375-392. Lundberg, Shelly and Richard Startz. 1983. Private Discrimination and Social Intervention in Competitive Labor Markets. American Economic Review 73: 340-347. Oyer, Paul and Scott Schaefer. 2000. Layo¤ and Litigation. Rand Journal of Economics 31: 345-358. Phelps, Edmund S. 1972. The Statistical Theory of Racism and Sexism. American Economic Review 62: 659-661. 34 Song, Younghwan. 2006. Recall Bias in the Displaced Workers Survey: Are Layo¤s Really Lemons? Labour Economics 14: 335-345. Spence, Michael. 1973. Job Market Signalling. Quarterly Journal of Economics 87: 355374. Stevens, Ann Hu¤. 1997. Persistent E¤ects of Job Displacement: The Importance of Multiple Job Losses. Journal of Labor Economics 15: 165-188. 35 -0.065 -0.093 (0.427) (0.463) -1.98 5.428 4.254 (6.872) (5.847) -6.48 39.223 39.427 (10.347) (10.321) 0.69 0.703 0.702 (0.457) (0.458) -0.10 0.032 0.021 (0.175) (0.144) -2.32 0.253 0.195 (0.435) (0.396) -4.95 0.299 0.296 (0.458) (0.457) -0.20 0.417 0.488 (0.493) (0.500) 5.03 Change in log real wage Tenure on previous job Age Married High School Dropout High School Graduate Some College College Graduate or above 0.260 0.439 (0.441) (0.498) 3.14 0.341 0.318 (0.476) (0.467) -0.41 0.341 0.217 (0.476) (0.413) -2.35 0.057 0.025 (0.233) (0.158) -1.34 0.593 0.503 (0.493) (0.502) -1.51 37.285 37.350 (10.180) (9.183) 0.06 5.711 3.865 (7.277) (5.534) -2.39 -0.112 -0.031 (0.445) (0.444) 1.39 5.651 5.827 (0.612) (0.538) 2.51 Plant Closing Layoff 5.773 5.895 (0.556) (0.572) 1.67 0.232 0.311 (0.422) (0.463) 6.09 0.349 0.365 (0.477) (0.482) 1.17 0.386 0.307 (0.487) (0.461) -5.76 0.033 0.017 (0.179) (0.130) -3.47 0.529 0.484 (0.499) (0.500) -3.15 37.500 38.115 (10.669) (10.717) 1.97 4.306 3.767 (5.673) (5.315) -3.36 -0.047 -0.053 (0.457) (0.426) -0.43 5.623 5.674 (0.510) (0.523) 3.30 Plant Closing Layoff 5.688 5.736 (0.560) (0.556) 2.82 White Collar Male Black Female White 0.153 0.226 (0.361) (0.419) 1.97 0.469 0.494 (0.500) (0.501) 0.53 0.340 0.239 (0.475) (0.427) -2.38 0.038 0.041 (0.192) (0.199) 0.16 0.373 0.370 (0.485) (0.484) -0.06 36.469 35.827 (10.110) (9.584) -0.69 5.420 3.650 (6.901) (5.252) -3.07 -0.102 -0.052 (0.591) (0.545) 0.86 5.474 5.520 (0.459) (0.485) 1.01 Plant Closing Layoff 5.577 5.584 (0.662) (0.597) 0.12 Female Black 0.046 0.060 (0.210) (0.238) 2.33 0.235 0.257 (0.424) (0.437) 1.87 0.522 0.518 (0.500) (0.500) -0.33 0.197 0.165 (0.398) (0.372) -3.04 0.700 0.649 (0.458) (0.477) -4.00 37.156 35.145 (10.969) (10.527) -6.93 5.746 3.429 (7.204) (5.332) -13.61 -0.088 -0.076 (0.456) (0.492) 0.91 5.650 5.615 (0.484) (0.493) -2.58 Plant Closing Layoff 5.734 5.690 (0.514) (0.524) -2.91 Male White 0.040 0.035 (0.197) (0.185) -0.29 0.225 0.270 (0.418) (0.445) 1.16 0.534 0.527 (0.500) (0.500) -0.15 0.201 0.168 (0.401) (0.375) -0.95 0.570 0.434 (0.496) (0.497) -3.09 36.498 34.695 (10.517) (10.173) -1.96 6.004 3.076 (7.772) (4.807) -5.04 -0.080 -0.063 (0.450) (0.398) 0.40 5.468 5.390 (0.495) (0.438) -1.84 Plant Closing Layoff 5.565 5.468 (0.509) (0.429) -2.13 Blue Collar Male Black 0.041 0.061 (0.197) (0.240) 1.96 0.188 0.207 (0.391) (0.405) 0.95 0.507 0.559 (0.500) (0.497) 2.15 0.264 0.174 (0.441) (0.379) -4.55 0.536 0.477 (0.499) (0.500) -2.45 38.781 36.754 (11.025) (11.103) -3.81 4.971 3.069 (6.354) (4.791) -6.93 -0.076 -0.090 (0.431) (0.413) -0.61 5.237 5.257 (0.433) (0.407) 0.94 Plant Closing Layoff 5.300 5.341 (0.449) (0.450) 1.84 Female White 0.000 0.008 (0.000) (0.087) 1.12 0.163 0.273 (0.370) (0.447) 2.32 0.548 0.515 (0.499) (0.502) -0.57 0.289 0.205 (0.455) (0.375) -1.67 0.331 0.303 (0.472) (0.497) -0.52 38.693 34.955 (10.315) (10.173) -3.23 7.024 3.600 (7.782) (4.807) -4.29 -0.108 -0.135 (0.393) (0.398) -0.54 5.104 5.120 (0.443) (0.438) 0.30 Plant Closing Layoff 5.234 5.259 (0.377) (0.429) 0.48 Female Black No. obs 2117 2882 123 157 2246 2488 209 243 2534 2981 249 256 913 818 166 132 SOURCE. --- Displaced Workers Survey 1984-2008. NOTE. --- Sample means are presented with standard deviations in parentheses. The t-statistics are for testing the null that the differences in sample means between the plant closing and layoff groups are zero. Sample selection: workers aged 20-64; lost job for 3 reasons: plant closing, position abolished or slack work; lost a job in previous 3 years; re-employed at survey date; full time to full time transition; private sector to private sector; re-employment weekly wage>=$40; not displaced from agriculture and construction. t-stat t-stat t-stat t-stat t-stat t-stat t-stat t-stat t-stat 6.012 6.033 (0.582) (0.590) 1.21 Log post-displacement real wage t-stat Log pre-displacement real wage Plant Closing Layoff 6.095 6.149 (0.580) (0.602) 2.94 Male White Table 1 Descriptive Statistics of Displaced Workers -0.027 (0.014) -0.052 (0.029) 0.017 (0.015) - - Layoff Plant Closing*Black Plant Closing*Female Layoff*Black*Female Plant Closing*Black*Female - Yr dummies, Yrs since disp, Region Pre-displacement tenure (1-3, 3-5, 5-10, 10+, omitted <1) Industry Occupation - - Y Y Y Y - - 0.004 (0.015) -0.052 (0.028) -0.038 (0.014) 0.030 (0.013) 0.008 (0.025) - Y Y Y Y Y - - 0.003 (0.015) -0.050 (0.028) -0.035 (0.014) 0.027 (0.013) 0.008 (0.025) Y Y Y Y Y Y - - 0.004 (0.015) -0.053 (0.028) -0.037 (0.014) 0.028 (0.013) 0.009 (0.025) Y Y Y Y Y Y -0.008 (0.058) -0.064 (0.051) 0.004 (0.015) -0.048 (0.045) -0.039 (0.014) 0.033 (0.014) 0.047 (0.040) - - - - - -0.087 (0.010) - - 0.007 (0.018) -0.006 (0.026) 0.013 (0.013) -0.021 (0.018) -0.006 (0.028) (6) - - Y Y Y Y - - 0.011 (0.018) -0.001 (0.026) -0.021 (0.013) -0.014 (0.018) -0.011 (0.028) (7) - Y Y Y Y Y - - -0.005 (0.018) -0.009 (0.026) -0.020 (0.013) -0.025 (0.018) -0.021 (0.027) (8) Blue Collar Y Y Y Y Y Y - - -0.005 (0.018) -0.010 (0.026) -0.019 (0.013) -0.026 (0.019) -0.022 (0.027) (9) Y Y Y Y Y Y 0.001 (0.053) -0.036 (0.057) -0.005 (0.020) -0.010 (0.033) -0.020 (0.014) -0.022 (0.020) -0.010 (0.033) (10) N 8750 8750 8747 8747 8747 6684 6684 6678 6678 6678 NOTE. ‐‐See Table 1 for sample restrictions. The cells present regression coefficients with standard errors in parentheses. The dependent variable in the regression is the difference in log wage between the post‐displacement job and pre‐displacement job. - Married, Age, Age2, Education -0.065 (0.011) 0.036 (0.013) Layoff*Female Constant 0.026 (0.025) Layoff*Black Table 2 Wage Losses after Layoffs and Plant Closings, by Race and Gender (1) (2) (3) (4) (5) White Collar Table 3 Effects of the CRA1991 on Relative Wage Changes (1) White Collar (2) (3) Blue Collar (4) Post91*Layoff*Black -0.130 (0.058) -0.117 (0.057) -0.032 (0.055) -0.025 (0.054) Post91*Layoff*Female -0.019 (0.027) -0.005 (0.027) -0.046 (0.037) -0.033 (0.036) Post91*Layoff 0.004 (0.029) 0.014 (0.028) 0.085 (0.027) 0.073 (0.026) Post91*Plant Closing*Black 0.013 (0.059) 0.001 (0.058) 0.008 (0.052) 0.004 (0.051) Post91*Plant Closing*Female -0.014 (0.029) -0.003 (0.029) -0.012 (0.036) -0.014 (0.035) Post91 0.027 (0.022) 0.000 (0.044) 0.025 (0.020) -0.001 (0.048) Layoff*Black 0.121 (0.050) 0.096 (0.049) 0.008 (0.041) -0.009 (0.041) Layoff*Female 0.047 (0.022) 0.031 (0.022) 0.004 (0.027) -0.009 (0.027) Layoff -0.032 (0.022) -0.046 (0.022) -0.034 (0.019) -0.057 (0.019) Plant Closing*Black -0.061 (0.046) -0.053 (0.045) -0.010 (0.035) -0.012 (0.035) Plant Closing*Female 0.024 (0.022) 0.005 (0.022) 0.013 (0.025) 0.002 (0.025) Constant -0.079 (0.016) Y -0.099 (0.013) Y Married, Age, Age2, Education - Y - Y Pre-displacement tenure (1-3, 3-5, 5-10, 10+, omitted <1) - Y - Y Yrs since disp., Yr dummies, Regions - Y - Y Industry, Occupation - Y - Y N 8750 8747 6684 6678 NOTE. ‐‐See Table 1 for sample restrictions. The cells present regression coefficients with standard errors in parentheses. The dependent variable in the regression is the difference in log wage between the post‐displacement job and pre‐displacement job. Table 4 Duration of Joblessness after Layoffs and Plant Closings, by Race and Gender (1) (2) (3) (4) White Collar (5) Blue Collar (6) Layoff*Black -0.028 (0.055) 0.035 (0.052) 0.050 (0.055) 0.176 (0.062) 0.113 (0.057) 0.104 (0.061) Layoff*Female 0.047 (0.030) 0.109 (0.029) 0.098 (0.031) 0.221 (0.045) 0.130 (0.041) 0.119 (0.044) Layoff 0.104 (0.034) 0.171 (0.032) 0.192 (0.034) -0.008 (0.035) 0.134 (0.032) 0.119 (0.034) Plant Closing*Black 0.129 (0.063) 0.134 (0.059) 0.118 (0.063) 0.227 (0.061) 0.166 (0.057) 0.148 (0.060) Plant Closing*Female 0.057 (0.035) 0.118 (0.034) 0.137 (0.037) 0.235 (0.044) 0.204 (0.041) 0.214 (0.044) Married, Age, Age2 - Y Y - Y Y Education - Y Y - Y Y Pre-displacement tenure (1-3, 3-5, 5-10, 10+, omitted <1) - Y Y - Y Y Yrs since disp., Yr dummies, Regions - Y Y - Y Y Industry, Occupation - Y Y - Y Y Log Pre-displacement wage - - 0.020 (0.024) - - 0.020 (0.032) 1.127 (0.009) 1.046 (0.008) 1.037 (0.009) 1.201 (0.011) 1.079 (0.010) 1.076 (0.010) Weibull scale parameter Percentage who lost job due to plant closing Percentage who lost job due to position abolished Percentage who lost job due to slack work 42.09 31.83 26.08 46.60 12.62 40.79 N 10277 10244 9095 7914 7887 7125 NOTE.‐‐The results are based on ML estimates of a Weibull proportional hazard model in which duration is the number of weeks on the first spell of joblessness since displacement. Cells present the marginal effects of covariates on the expected log durations. Standard errors are in parentheses. Sample selection: workers aged 20‐64; lost job for 3 reasons: plant closing, position abolished or slack work; lost a job in previous 3 years; displaced from full time jobs; displaced from private sector jobs; pre‐displacement weekly wage>=$40; not displaced from agriculture and construction. 0.0000 0.0056 - - Y Y Y Y 0.005 (0.015) -0.052 (0.028) -0.062 (0.016) -0.010 (0.017) 0.035 (0.022) 0.016 (0.024) 0.044 (0.044) 0.074 (0.046) 0.0162 Y Y Y Y Y Y 0.004 (0.015) -0.053 (0.028) -0.059 (0.016) -0.012 (0.017) 0.033 (0.022) 0.017 (0.024) 0.046 (0.044) 0.076 (0.046) -0.043 (0.028) ‐ 44.14 30.95 24.91 0.0020 Y Y Y Y Y Y 0.019 (0.010) -0.019 (0.019) -0.038 (0.011) 0.002 (0.012) 0.0240 - - - - - -0.087 (0.010) 0.007 (0.018) -0.006 (0.026) -0.035 (0.021) 0.027 (0.014) 0.025 (0.041) -0.021 (0.064) ‐ - 0.006 (0.041) ‐ 0.1167 - - Y Y Y Y 0.011 (0.018) -0.002 (0.026) -0.056 (0.021) -0.010 (0.014) 0.027 (0.041) -0.041 (0.027) -0.033 (0.063) -0.003 (0.040) (6) Blue Collar 0.0836 Y Y Y Y Y Y -0.005 (0.018) -0.010 (0.026) -0.058 (0.021) -0.007 (0.014) 0.024 (0.041) -0.034 (0.027) -0.035 (0.062) -0.006 (0.040) (7) 47.89 12.19 39.92 0.0305 Y Y Y Y Y Y -0.015 (0.013) -0.016 (0.019) -0.056 (0.018) -0.016 (0.012) - ‐ ‐ ‐ (8) N 8750 8750 8747 8747 6684 6684 6678 6678 NOTE. ‐‐See Table 1 for sample restrictions. The cells present regression coefficients with standard errors in parentheses. The dependent variable in the regression is the difference in log wage between the post‐displacement job and pre‐displacement job. Percentage who lost job due to plant closing Percentage who lost job due to position abolished Percentage who lost job due to slack work P -value for the F-test (null hypothesis: the coefficient on slack work and position abolished are jointly equal - Occupation -0.065 (0.011) Constant - 0.017 (0.015) Female Industry -0.052 (0.028) Black - -0.058 (0.016) Position Abolished Pre-displacement tenure (1-3, 3-5, 5-10, 10+, omitted <1) 0.010 (0.017) Slack Work - 0.028 (0.023) Female*Position Abolished Yr dummies, Yrs since disp, Region 0.013 (0.024) Female*Slack Work - 0.057 (0.045) Black*Position Abolished Married, Age, Age2, Education 0.097 (0.047) Black*Slack Work Table 5 Wage Losses after Displacement: Slack Work, Position Abolished and Plant Closing (1) (2) (3) (4) (5) White Collar Table 6 Propensity of Changing Occupation after Displacement (1) (2) White Collar (3) (4) Blue Collar Slack Work 0.017 (0.012) 0.028 (0.011) 0.066 (0.011) 0.048 (0.011) Position Abolished 0.038 (0.011) 0.063 (0.011) 0.081 (0.016) 0.089 (0.016) Constant 0.654 (0.007) Y 0.657 (0.008) Y Married, Age, Age , Education - Y - Y Yr dummies, Yrs since disp, Region - Y - Y Pre-displacement tenure (1-3, 3-5, 5-10, 10+, omitted <1) - Y - Y Industry - Y - Y Occupation - Y - Y 0.085 0.006 0.334 0.012 2 P -value for the F-test (null hypothesis: the coefficient on slack work and position abolished are jointly equal Mean of the Dependent Variable 0.670 0.670 0.693 0.694 N 10465 10421 8049 8006 NOTE.‐‐See Table 1 for sample restrictions. Cells present estimates from a linear probablity model in which the dependent variable equals one if a worker changed 1‐digit occupation after displacement. Robust standard errors are in parentheses. Table 7 Simulation Results for the Augumented Model A. Simulated Log Wage Differentials Male Female Data Simulation Data Simulation Slack Work Position Abolished Plant Closing ‐0.066 (0.013) ‐0.112 (0.012) ‐0.054 (0.011) ‐0.067 ‐0.112 ‐0.053 ‐0.045 (0.015) ‐0.075 (0.013) ‐0.049 (0.010) ‐0.046 ‐0.082 ‐0.053 B. Paramters of Simulated Model Male Female 0.15 1.49 0.06 0.02 0.776 0.999 ρ τ σ Initial Retention Probability Sector Shock Shock 0.985 Layoff Probability 0.051 0.031 Human Capital Type Shock Shock 0.997 Layoff Probability 0.015 0.011 NOTE.‐‐The six moments in the data are based on the regression in column 3 of Table 5. Standard errors are in parentheses. - Industry Occupation - - Y Y Y Y -0.264 (0.016) -0.139 (0.031) 0.042 (0.015) -0.298 (0.014) -0.149 (0.028) - Y Y Y Y Y -0.275 (0.016) -0.144 (0.031) 0.028 (0.015) -0.307 (0.014) -0.155 (0.028) Y Y Y Y Y Y -0.255 (0.016) -0.122 (0.030) 0.028 (0.015) -0.283 (0.014) -0.139 (0.027) - - - - - 5.730 (0.010) -0.421 (0.019) -0.130 (0.028) -0.043 (0.014) -0.334 (0.019) -0.177 (0.029) (5) - - Y Y Y Y -0.385 (0.018) -0.097 (0.026) 0.015 (0.013) -0.315 (0.018) -0.124 (0.027) - Y Y Y Y Y -0.339 (0.017) -0.080 (0.025) 0.013 (0.013) -0.277 (0.017) -0.106 (0.026) (6) (7) Blue Collar Y Y Y Y Y Y -0.318 (0.017) -0.078 (0.025) 0.007 (0.013) -0.257 (0.017) -0.100 (0.026) (8) N 9120 9120 9117 9117 7066 7066 7059 7059 NOTE. ‐‐See Table 1 for sample restrictions. Cells present regression coefficients with standard errors in parentheses. The dependent variable in the regression is the log pre‐displacement wage. - 6.088 (0.013) Constant Pre-displacement tenure (1-3, 3-5, 5-10, 10+, omitted <1) -0.394 (0.018) Plant Closing*Female - -0.194 (0.036) Plant Closing*Black Yr dummies, Yrs since disp, Region 0.058 (0.018) Layoff - -0.406 (0.016) Layoff*Female 2 Married, Age, Age , Education -0.192 (0.032) Layoff*Black Table A1 Wages Prior to Layoffs and Plant Closings, by Race and Gender (1) (2) (3) (4) White Collar Working Paper Series A series of research studies on regional economic issues relating to the Seventh Federal Reserve District, and on financial and economic topics. Firm-Specific Capital, Nominal Rigidities and the Business Cycle David Altig, Lawrence J. Christiano, Martin Eichenbaum and Jesper Linde WP-05-01 Do Returns to Schooling Differ by Race and Ethnicity? Lisa Barrow and Cecilia Elena Rouse WP-05-02 Derivatives and Systemic Risk: Netting, Collateral, and Closeout Robert R. Bliss and George G. Kaufman WP-05-03 Risk Overhang and Loan Portfolio Decisions Robert DeYoung, Anne Gron and Andrew Winton WP-05-04 Characterizations in a random record model with a non-identically distributed initial record Gadi Barlevy and H. N. Nagaraja WP-05-05 Price discovery in a market under stress: the U.S. Treasury market in fall 1998 Craig H. Furfine and Eli M. Remolona WP-05-06 Politics and Efficiency of Separating Capital and Ordinary Government Budgets Marco Bassetto with Thomas J. Sargent WP-05-07 Rigid Prices: Evidence from U.S. Scanner Data Jeffrey R. Campbell and Benjamin Eden WP-05-08 Entrepreneurship, Frictions, and Wealth Marco Cagetti and Mariacristina De Nardi WP-05-09 Wealth inequality: data and models Marco Cagetti and Mariacristina De Nardi WP-05-10 What Determines Bilateral Trade Flows? Marianne Baxter and Michael A. Kouparitsas WP-05-11 Intergenerational Economic Mobility in the U.S., 1940 to 2000 Daniel Aaronson and Bhashkar Mazumder WP-05-12 Differential Mortality, Uncertain Medical Expenses, and the Saving of Elderly Singles Mariacristina De Nardi, Eric French, and John Bailey Jones WP-05-13 Fixed Term Employment Contracts in an Equilibrium Search Model Fernando Alvarez and Marcelo Veracierto WP-05-14 1 Working Paper Series (continued) Causality, Causality, Causality: The View of Education Inputs and Outputs from Economics Lisa Barrow and Cecilia Elena Rouse WP-05-15 Competition in Large Markets Jeffrey R. Campbell WP-05-16 Why Do Firms Go Public? Evidence from the Banking Industry Richard J. Rosen, Scott B. Smart and Chad J. Zutter WP-05-17 Clustering of Auto Supplier Plants in the U.S.: GMM Spatial Logit for Large Samples Thomas Klier and Daniel P. McMillen WP-05-18 Why are Immigrants’ Incarceration Rates So Low? Evidence on Selective Immigration, Deterrence, and Deportation Kristin F. Butcher and Anne Morrison Piehl WP-05-19 Constructing the Chicago Fed Income Based Economic Index – Consumer Price Index: Inflation Experiences by Demographic Group: 1983-2005 Leslie McGranahan and Anna Paulson WP-05-20 Universal Access, Cost Recovery, and Payment Services Sujit Chakravorti, Jeffery W. Gunther, and Robert R. Moore WP-05-21 Supplier Switching and Outsourcing Yukako Ono and Victor Stango WP-05-22 Do Enclaves Matter in Immigrants’ Self-Employment Decision? Maude Toussaint-Comeau WP-05-23 The Changing Pattern of Wage Growth for Low Skilled Workers Eric French, Bhashkar Mazumder and Christopher Taber WP-05-24 U.S. Corporate and Bank Insolvency Regimes: An Economic Comparison and Evaluation Robert R. Bliss and George G. Kaufman WP-06-01 Redistribution, Taxes, and the Median Voter Marco Bassetto and Jess Benhabib WP-06-02 Identification of Search Models with Initial Condition Problems Gadi Barlevy and H. N. Nagaraja WP-06-03 Tax Riots Marco Bassetto and Christopher Phelan WP-06-04 The Tradeoff between Mortgage Prepayments and Tax-Deferred Retirement Savings Gene Amromin, Jennifer Huang,and Clemens Sialm WP-06-05 2 Working Paper Series (continued) Why are safeguards needed in a trade agreement? Meredith A. Crowley WP-06-06 Taxation, Entrepreneurship, and Wealth Marco Cagetti and Mariacristina De Nardi WP-06-07 A New Social Compact: How University Engagement Can Fuel Innovation Laura Melle, Larry Isaak, and Richard Mattoon WP-06-08 Mergers and Risk Craig H. Furfine and Richard J. Rosen WP-06-09 Two Flaws in Business Cycle Accounting Lawrence J. Christiano and Joshua M. Davis WP-06-10 Do Consumers Choose the Right Credit Contracts? Sumit Agarwal, Souphala Chomsisengphet, Chunlin Liu, and Nicholas S. Souleles WP-06-11 Chronicles of a Deflation Unforetold François R. Velde WP-06-12 Female Offenders Use of Social Welfare Programs Before and After Jail and Prison: Does Prison Cause Welfare Dependency? Kristin F. Butcher and Robert J. LaLonde Eat or Be Eaten: A Theory of Mergers and Firm Size Gary Gorton, Matthias Kahl, and Richard Rosen Do Bonds Span Volatility Risk in the U.S. Treasury Market? A Specification Test for Affine Term Structure Models Torben G. Andersen and Luca Benzoni WP-06-13 WP-06-14 WP-06-15 Transforming Payment Choices by Doubling Fees on the Illinois Tollway Gene Amromin, Carrie Jankowski, and Richard D. Porter WP-06-16 How Did the 2003 Dividend Tax Cut Affect Stock Prices? Gene Amromin, Paul Harrison, and Steven Sharpe WP-06-17 Will Writing and Bequest Motives: Early 20th Century Irish Evidence Leslie McGranahan WP-06-18 How Professional Forecasters View Shocks to GDP Spencer D. Krane WP-06-19 Evolving Agglomeration in the U.S. auto supplier industry Thomas Klier and Daniel P. McMillen WP-06-20 3 Working Paper Series (continued) Mortality, Mass-Layoffs, and Career Outcomes: An Analysis using Administrative Data Daniel Sullivan and Till von Wachter The Agreement on Subsidies and Countervailing Measures: Tying One’s Hand through the WTO. Meredith A. Crowley WP-06-21 WP-06-22 How Did Schooling Laws Improve Long-Term Health and Lower Mortality? Bhashkar Mazumder WP-06-23 Manufacturing Plants’ Use of Temporary Workers: An Analysis Using Census Micro Data Yukako Ono and Daniel Sullivan WP-06-24 What Can We Learn about Financial Access from U.S. Immigrants? Una Okonkwo Osili and Anna Paulson WP-06-25 Bank Imputed Interest Rates: Unbiased Estimates of Offered Rates? Evren Ors and Tara Rice WP-06-26 Welfare Implications of the Transition to High Household Debt Jeffrey R. Campbell and Zvi Hercowitz WP-06-27 Last-In First-Out Oligopoly Dynamics Jaap H. Abbring and Jeffrey R. Campbell WP-06-28 Oligopoly Dynamics with Barriers to Entry Jaap H. Abbring and Jeffrey R. Campbell WP-06-29 Risk Taking and the Quality of Informal Insurance: Gambling and Remittances in Thailand Douglas L. Miller and Anna L. Paulson WP-07-01 Fast Micro and Slow Macro: Can Aggregation Explain the Persistence of Inflation? Filippo Altissimo, Benoît Mojon, and Paolo Zaffaroni WP-07-02 Assessing a Decade of Interstate Bank Branching Christian Johnson and Tara Rice WP-07-03 Debit Card and Cash Usage: A Cross-Country Analysis Gene Amromin and Sujit Chakravorti WP-07-04 The Age of Reason: Financial Decisions Over the Lifecycle Sumit Agarwal, John C. Driscoll, Xavier Gabaix, and David Laibson WP-07-05 Information Acquisition in Financial Markets: a Correction Gadi Barlevy and Pietro Veronesi WP-07-06 Monetary Policy, Output Composition and the Great Moderation Benoît Mojon WP-07-07 4 Working Paper Series (continued) Estate Taxation, Entrepreneurship, and Wealth Marco Cagetti and Mariacristina De Nardi WP-07-08 Conflict of Interest and Certification in the U.S. IPO Market Luca Benzoni and Carola Schenone WP-07-09 The Reaction of Consumer Spending and Debt to Tax Rebates – Evidence from Consumer Credit Data Sumit Agarwal, Chunlin Liu, and Nicholas S. Souleles WP-07-10 Portfolio Choice over the Life-Cycle when the Stock and Labor Markets are Cointegrated Luca Benzoni, Pierre Collin-Dufresne, and Robert S. Goldstein WP-07-11 Nonparametric Analysis of Intergenerational Income Mobility with Application to the United States Debopam Bhattacharya and Bhashkar Mazumder WP-07-12 How the Credit Channel Works: Differentiating the Bank Lending Channel and the Balance Sheet Channel Lamont K. Black and Richard J. Rosen WP-07-13 Labor Market Transitions and Self-Employment Ellen R. Rissman WP-07-14 First-Time Home Buyers and Residential Investment Volatility Jonas D.M. Fisher and Martin Gervais WP-07-15 Establishments Dynamics and Matching Frictions in Classical Competitive Equilibrium Marcelo Veracierto WP-07-16 Technology’s Edge: The Educational Benefits of Computer-Aided Instruction Lisa Barrow, Lisa Markman, and Cecilia Elena Rouse WP-07-17 The Widow’s Offering: Inheritance, Family Structure, and the Charitable Gifts of Women Leslie McGranahan WP-07-18 Demand Volatility and the Lag between the Growth of Temporary and Permanent Employment Sainan Jin, Yukako Ono, and Qinghua Zhang WP-07-19 A Conversation with 590 Nascent Entrepreneurs Jeffrey R. Campbell and Mariacristina De Nardi WP-07-20 Cyclical Dumping and US Antidumping Protection: 1980-2001 Meredith A. Crowley WP-07-21 Prenatal Nutrition and Adult Outcomes: The Effect of Maternal Fasting During Ramadan Douglas Almond and Bhashkar Mazumder WP-07-22 5 Working Paper Series (continued) The Spending and Debt Responses to Minimum Wage Increases Daniel Aaronson, Sumit Agarwal, and Eric French WP-07-23 The Impact of Mexican Immigrants on U.S. Wage Structure Maude Toussaint-Comeau WP-07-24 A Leverage-based Model of Speculative Bubbles Gadi Barlevy WP-08-01 Displacement, Asymmetric Information and Heterogeneous Human Capital Luojia Hu and Christopher Taber WP-08-02 6