Policy Research Working Paper 10536 Plant Closings and the Labor Market Outcomes of Displaced Workers Evidence from Mexico Francisco Arias Daniel Lederman Middle East and North Africa Region Office of the Chief Economist August 2023 Policy Research Working Paper 10536 Abstract This paper investigates the impacts of job displacement on week, respectively. Using the rotating panel of the survey, subsequent labor market outcomes, focusing on differenti- difference in differences coefficients are estimated, remov- ated effects by educational groups and gender. The findings ing time-invariant individual heterogeneity. Compared to show that job separations caused by plant closings result ordinary least squares, the difference in differences estimates in sizable and long-lasting wage reductions, with an aver- reduce the magnitude of the average impacts of plant clos- age decline of −7.5 percent over a nine-year period relative ing on wages, from −7.5 to −4.7 percent, and on working to workers who did not experience job losses. A stronger hours from 1.4 to 0.53 additional hours. These results sug- effect is estimated for highly educated workers than for gest that the ordinary least squares estimates are upwardly low educated workers, with initial effects being 18.4 and biased due to omitted individual worker heterogeneity. 9 percent wage drops, respectively. For working hours, the The paper discusses another potential remaining source of effect on low educated workers is double the effect on highly endogeneity concerning the quality of the match between educated workers, with 3.0 and 1.5 additional hours per employers and workers. This paper is a product of the Office of the Chief Economist, Middle East and North Africa Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at dlederman@worldbank.org and fariasvazquez@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Plant Closings and the Labor Market Outcomes of Displaced Workers: Evidence from Mexico Francisco Arias and Daniel Lederman1 Key words: Labor market, job displacement, wages, education, difference in difference. JEL codes: I26, J01, J30, J60 1 Arias is a Senior Economist with the World Bank’s Macroeconomics, Investment and Trade Global Practice, and Lederman is the Deputy Chief Economist for the Middle East and North Africa Region of the World Bank. We thank Diego Alvarez and Michelle Botello for their valuable research assistance. Lucila Venturi (Harvard, Center for International Development) provided invaluable research assistance during the early stages of this paper. We gratefully acknowledge comments received from Joana Silva on an earlier version of this paper. We also thank the anonymous referees who helped us improve this work with relevant observations and suggestions. All remaining errors are the responsibility of the authors. The opinions expressed in this paper do not necessarily reflect the opinions of the World Bank, its Board of Directors, or the countries which they represent. I. Introduction Labor reallocation and a flow of displaced workers reentering the workforce often accompany business cycles and economic shocks. The destruction and creation of jobs are fundamental to the efficient allocation of resources. Yet to the extent that mobility barriers and labor market distortions prevail in the labor market, displaced workers could experience a negative impact on their wages and other job characteristics such as their formality status. Once formerly displaced workers reenter the workforce, wages and other job characteristics might decline in relation to what they previously had. In these cases, an adequate response to labor policies requires information on the magnitude and length of workers´ losses. Additionally, the scarring effects of job displacement might be different for a different types of workers. Although earnings losses are the most common variable studied in the literature, the probability of working in the formal sector, hours of work, tenure, and the duration of unemployment are outcomes that deserve attention, especially in developing countries. In this paper, we look at how the Mexican labor market reabsorbs different types of displaced workers and provide empirical evidence of the magnitude and length of workers´ losses. We also explore the existence of heterogeneous effects by gender and educational categories. The results inform labor policies by identifying workers’ labor outcomes with the largest and longest-lasting impacts of job displacement. Most of the literature looking at the effects of job displacement on workers’ labor outcomes has focused on the US and developed countries. For instance, in one of the earliest studies on this topic, Ruhm (1991) finds long-lasting wage reductions for US displaced workers of around 14 percent compared to wages of their non-displaced counterparts. Stevens (1997), using a 2 longitudinal dataset, finds a drop in earnings of 30 percent at the time of displacement, which narrows to 10 percent after six years. In a seminal paper, using administrative data for workers in the state of Pennsylvania, Jacobson, LaLonde, and Sullivan (1993) look at the impact of large- scale layoffs on workers. They report earnings losses of more than 40 percent in the year of displacement, and six years after displacement earnings are yet 25 percent below their pre- displacement level. In a related paper for the state of Connecticut, Couch and Placzek (2010) find that earnings losses for workers displaced through mass layoff equal 32 percent at the time of job loss and six years later the reductions narrow to 14 percent. Davis and von Watcher (2011) calibrate a model of job search and unemployment fluctuations, and find that earning losses rise steeply with the unemployment rate at the time of displacement, concluding that tight labor market conditions at the time of displacement improve the medium- and long-term future earnings trajectory of displaced workers. Bonikowska and Morissette (2012) look at the Canadian labor market and find that better labor market conditions mitigate earnings losses for displaced workers. In periods of low unemployment earnings losses are 10 percent while in periods of high unemployment they reach 21 percent. For Swedeen, Eliason and Sotrrie (2006) use linked employer‐employee data and follow displaced workers over a pre-displacement period of 4 years and a post-displacement period of 11 years. They conclude that displaced workers experience both earnings losses and worsened labor‐market position not only during a transitory period of adjustment but also in the long run. For Germany, Toulemon and Weber-Baghdiguian, (2016) find earnings losses of 10 percent but the effect vanishes after four years. They also find that other indicators of job quality are negatively affected. For the United Kingdom, Hijzen, Upward and Wright (2010) track workers for up to nine years after displacement and implement difference in differences estimators using 3 propensity score matching methods. They report earnings losses during the first five years after displacement that range from 18 to 35 percent for workers whose firm closes, and from 14 to 25 percent for workers who exit a firm that suffers a mass layoff. The literature has poorly studied the impacts on labor market outcomes of job displacement in developing countries. Eslava, Haltiwanger, Kugler and Kugler (2010) study the Colombian labor market and look at how trade liberalization affected workers displaced due to plant closings. The authors document the job reallocation that followed trade liberalization in the country and analyze workers´ outcomes after they reentered the workforce. They find that displaced workers experience a reduction in wages and tenure (time in a job) and a lower probability of working in the formal sector within the first two years after their displacement. Yet they find that Colombian workers’ wages fully recover three years after a plant closing. For the case of Mexico, Kaplan, Martinez, and Robertson (2005) look at changes in formal wages following job displacement. They use administrative data from Social Security records to study the effect of job displacement on wages under different economic conditions. They find large wage losses when displacement occurs in periods of high unemployment and in regions with less economic activity. Their study is subject to two limitations. First, the Social Security records only collect information on formal employment and do not differentiate between informal employment, unemployment, and individuals who have left the labor force. Second, the Social Security records do not make it possible to identify the source of job separation, and as the authors acknowledge, “the cause of separation is important because workers who leave voluntarily are more likely to have more positive economic prospects after they leave their current firm. Including voluntary separation would, therefore, bias the estimated effects of displaced workers upwards.” (Kaplan et al., 2005). 4 In this paper, we investigate the impact of job displacement on labor market outcomes of Mexican workers, with special attention to differential effects on gender and educational groups. We use retrospective information on labor history of a cross-section of workers collected from the extended questionnaires of the Mexican Labor Force Survey (Encuesta Nacional de Ocupación y Empleo, ENOE). The dataset permits to identify six mutually exclusive categories of job displacement: 1) Plant closing; 2) Fired; 3) Quitting; 4) Closed own business; 5) Retired; and 6) Other reasons. We study wages, hours of work, formality status, and duration of unemployment that follow job displacement. After estimating the average impact of job displacement on these labor outcomes, we ask whether the impacts are transitory or permanent, and present results for educational groups and gender. We estimate a hazard model to look at the duration of unemployment spells of displaced workers. The analyses in this paper focus on workers displaced due to plant closings. We should be cautious, as the relevant coefficients might be subject to endogeneity or selection bias to the extent that displaced workers are more likely to have unobservable characteristics that make them less skillful than the benchmark category of non-displaced workers. Unlike Kaplan et al. (2005), we include informal workers in our analysis. The inclusion of informal workers is crucial, as they account for more than half of Mexico’s total employment. To test the robustness of the cross-section estimates, we estimate a difference-in-difference (DID) specification using the ENOE’s longitudinal structure. We find that displaced workers due to plant closings experience sizeable and long-lasting wage reductions, which average a reduction of -7.5 percent over a nine-year period. After an initial drop of 11 percent, the wage gap with respect to non-displaced workers narrows to 5 percent in the fourth year and equals 2 percent in the ninth year. We find no evidence of gender-differentiated 5 effects of plant closing, but considerable differential effects by educational groups. Plant closings have strong and permanent effects for high educated workers, wages of this group initially drop by 18.4 percent -in relation to the average wage of their non-displaced counterparts- narrowing to 11 percent in the third year and to 6 percent in the ninth year, while for low educated workers the initial drop is 9.0 percent, reduces to 5 percent in the second year and vanishes completely after the sixth year. Regarding other types of job displacement, we find that workers who quit have a permanent increase in wages that reach up to 10 percent in the ninth year, and workers who closed their own business have a transitory decline in wages that vanishes completely in the fourth year. There is no evidence of heterogeneous impacts for these two types of job separations. Displaced workers due to plant closings have a permanent increase in hours of work. The impact on low educated workers is double the magnitude for high educated workers, with 3.0 and 1.5 additional hours per week, respectively. There is no gender-differentiated effect, as both men and women experience similar increases in hours of work, which might be a coping mechanism to compensate for reduced wages. Interestingly, plant closings are associated with a permanent increase in the probability of working in the formal sector by 5.0 percentage points (pp.), however, there are marked differences by educational groups. For highly educated workers the probability initially drops 5.0 pp. relative to the probability of their non-displaced counterparts, and vanishes completely after the fifth year; for low educated workers the probability increases permanently by 9.0 pp. Finally, estimates of the hazard model show that after a plant closing event, unemployment spells –relative to their non-displaced counterpart– are longer for men than for women, and for high educated than for low educated workers. 6 The DID results show that displaced workers due to plant closings experience a reduction in wages of -4.7 percent, a moderate increase of 0.53 hours of work per week, and no impact on the probability of working in the formal sector. Compared to OLS estimates, DID reduces the magnitude of the impact of plant closings on wages, from -7.5 to -4.7 percent, and the effect on hours of work from 1.4 to 0.53 hours per week, while the DID coefficient on probability of working in the formal sector is tiny and thus not statistically different from zero. We interpret this change in the magnitude of the estimates as evidence that OLS might be inflating the impact of plant closings on labor outcomes. The rest of the paper is structured as follows: Section II describes the data. Section III describes the empirical specifications, outlines the robustness tests and discusses the results. Section IV presents the conclusions. II. Data The ENOE is a household survey that collects detailed information on labor force status, wages, employment, and demographic characteristics of the labor force. The survey has been collected quarterly since 2005 and is representative at the national and state levels. The survey in each quarter includes approximately 121,000 households. The survey is structured as a rotating panel; each household is interviewed for five consecutive quarters, and in each quarter, one-fifth of the sample is replaced. We restricted the sample to individuals between 15 and 65 years old. We use selected phases of the survey which implemented the extended questionnaire for the period 2005- 2019, which excludes the COVID-19 economic crisis. When the survey uses the extended questionnaire, it collects detailed information on individuals’ labor history. The survey instrument asks whether individuals have ever experienced a job 7 separation that left them temporarily unemployed. In addition, the survey asks the respondent to state the reasons for the job separation. We can identify six mutually exclusive types of job separation: 1) plant closings (PCE) 2; 2) quitting (Q); 3) discharges or firings (F) 3; 4) closing one’s own business (COB); 5) retirement; and 6) other types of displacement. However, the analysis in this version of the paper focuses only on four types of job separation, from type 1) to 4). The final sample includes 2,707,562 individuals. Table 1 presents summary statistics. We observe that 53.3 percent of the respondents reported they never experienced a job displacement, 8.4 percent experienced a plant closing, 5.5 percent were fired, 25.2 voluntarily quit, 2.7 percent closed their own business and 2.6 were retired. It is important to notice that the number of workers who never experienced a job termination—53.3 percent of the sample—might be biased upwards due to the structure of the question. The survey instrument considers a job to have been terminated only if the displacement event is followed by a period of unemployment; thus, workers who switch jobs without experiencing unemployment are classified in the “non-displacement” category. 2 Plant closing includes downsizing of staff. 3 Fired includes cases in which the person was not called again or when the contract ended. 8 Table 1: Summary Statistics Observations 2,707,562 Share of men 47.8% Average age 36.0 Share of married individuals 46.7% Average years of education 9.6 Size of locality (share of individuals) More than 100,000 inhabitants 60.4% Between 15,000-99,999 inhabitan 11.7% Between 2,500-14,999 inhabitant 12.0% Less than 2,500 inhabitants 15.9% Weekly hours 43.1 Hourly wage 32.5 Labor force composition Employed 61.8% Unemployed 2.7% Out of labor force 35.5% Employed status Employee 69.8% Employer 5.0% Self-employed 20.0% Unpaid worker 5.1% Type of displacement Plant Closing 8.4% Being Fired 5.5% Quitting 25.2% Closed Own Business 2.7% Retired 2.6% Other 2.2% Non displaced workers 53.3% Measures of informality status Formal Employment 47.5% With access to health 41.5% With a secondary job 4.2% Source: Own calculations based on ENOE 2005-2019 There are marked differences in current labor market outcomes between individuals who experience a plant closing and those who experience other types of job separation. We construct a transition matrix that relates the reason for job separation with the individual’s current labor market status. Through the matrix, we can identify both voluntary and involuntary separation. Involuntary separations are captured in the top two rows of table 2 —the “plant closing” and “fired” categories. The bottom two rows of table 2 captured voluntary separations—the “quit” and “closed own 9 business” categories. The current labor market statuses are captured in the “employed”, “unemployed” and “out of the labor force” columns of table 2. Among the four main sources of job separation, workers who experienced a plant closing are more likely to be currently employed. For example, for workers in the plant closing category, 72.0 percent are currently employed, compared to only 61.0 percent of those who were fired from their last job. Similarly, among plant closing workers, 9.6 percent are currently unemployed, compared to 14.6 percent of those who were fired. Table 2 also suggests that workers who were fired are more likely to leave the labor force, as 24.5 percent of the fired category are currently out of the labor force, compared to only 18.5 percent from the plant closing category. The differences in current labor status between the voluntary and involuntary separation categories are also significant. In general, workers in the voluntary separation categories (either quit or closed own business) are less likely to be currently employed in a salaried job and around half of them are currently out of the labor force, 51.6 and 49.8 percent for quit and closed own business, respectively. Table 2: Type of job displacement and current labor force status Current labor force status Employed Unemployed Out of labor force Plant 72.0 9.6 18.5 Closing Type of job Fired 60.9 14.6 24.5 displacement Quit 44.4 4.0 51.6 Closed own 44.8 5.4 49.8 business 10 We also construct a transition matrix that relates the type of displacement to current occupational categories. Workers who experienced a plant closing and workers who were fired are equally likely to have a salaried job once they reenter the workforce. Table 3 shows that approximately 80 percent of those in either the plant closing or fired categories once they are employed again continue being salaried employees, and approximately 15 percent move into self-employment activities. Workers who voluntarily quit have slightly different transitions to those who were involuntarily separated (in either plant closing or fired categories). For example, 73.9 percent of the workers in the quit category are currently salaried workers, and 18.3 percent move into self-employment activities. Table 3: Type of job displacement and current employment status Current employment status Employee Employer Self-employed Unpaid worker Plant 3.0 15.6 1.5 79.9 Closing Type of job Fired 80.7 2.6 14.5 2.3 displacement Quit 73.9 5.3 18.3 2.6 Closed own 58.5 8.5 29.9 3.2 business 11 III. Econometric approach In this section, we look at how job displacement influences current labor market outcomes. We compare different types of job displacement to the benchmark category of workers in the non- displacement group. The first specification captures the average effect of several types of job displacement. The second specification looks at the lagged effect of job displacement on current labor market outcomes. This specification permits us to identify temporary and permanent effects. When some effects disappear over time, workers recover labor outcomes similar to those of workers in the non-displacement group. We estimate a hazard model to study the role of job separation and other covariates on the duration of spells of unemployment. Finally, the last specification uses the ENOE rotating panel structure, which follows individuals for a one-year period to estimate a DID specification, which controls for a time-invariant individual heterogeneity term and provides robustness to the OLS estimates of temporary and permanent effects. First, we estimate the following specification: (1) Yi,t = δt + X i′,t β + ∑αdTDi,d + εi,t d 12 where Y represents the labor market outcome (log of real wages,4 hours of work, formality status or tenure). We include a time-fixed effect, dt, and the matrix X includes a rich set of covariates, including gender, marital status, education, age, age squared, state and industry fixed effects. The variable of interest is TD, which reflects the type of displacement, including plant closings, fired or discharged, quitting, etc. The omitted category is “non-displacement.” We explore the existence of temporary and permanent effects of job displacement on current labor market outcomes. We modify the above specification by adding lagged indicators to the type of displacement: D, s =10 Yi,t = δt + X i′,t β + ∑α d ,s TDi,d ,t − s + εi,t d , s =0 (2) For example, for workers who experience a plant closing event three years before the time of the survey interview, the coefficient aPC,3, captures the difference in current labor outcome, Yi,t, relative to workers in the non-displacement category. We model the duration of unemployment that follows job displacement with a parametric hazard model. Assuming a Weibull distribution for the hazard function, we estimate the following model, h(u) = αλα uα −1 (3) with the survival function, () = exp(−(λu)α), which represents the probability that the period of unemployment is a length that is at least U=u; in other words, S(u) = 1 − F(u) = Prob(U ≥ u) 4 We use the consumer price index to normalize nominal wages to 2005 prices. 13 with F(.) being the Weibull cumulative distribution function.5 We include the covariates in the hazard function by defining D ,s=10 δt + X i′,t β + ∑ α d ,sTDi ,d ,t−s +ε i ,t λi,t = e d ,s=0 (4) The shape parameter of the Weibull distribution, α, defines the time dependence of the model. The hazard rate is monotonically decreasing with < 0 and increasing with > 0. We present both the coefficients and the hazard-rate ratio of the model and complement the results with graphs of the survival functions for different groups of workers. A drawback in the specifications expressed in equations 1 and 2 is that they ignore the existence of individual heterogeneity, and unobservable worker characteristics might simultaneously be related to the probability of experiencing a plant closing and subsequent labor market outcomes. We check for the robustness of the OLS estimation of equations 1 by estimating a DID estimator and using the ENOE’s rotating panel structure that follows approximately 90,000 workers for one year. If we see a significant difference, it could be interpreted as evidence that OLS estimators are potentially confounding the displacement effects with those of unobservable characteristics of the individuals in our sample. Unfortunately, the DID approach might still not be free of omitted- variable bias, because the survey data do not provide information about the employer-worker matches. This might be problematic if the quality of the match (as opposed to the quality of the individual worker) is systematically correlated with both the probability of a plant closing and subsequent labor-market outcomes of the worker. 5 The Weibull density function is given by f (u) = αλuα −1e−λu . 14 A. Results In table 4 we present the average impact of job displacement on several labor outcomes. After reentering the workforce, displaced workers receive lower wages than non-displaced workers. PC- workers experience the largest wage reduction (7.5 percent), followed by F-workers (6.9 percent), Q-workers (2.3 percent) and COB-workers (5.3 percent). The impact on weekly hours of work is statistically significant for PC-workers (1.4 additional hours per week) and Q-workers (1.7 additional hours per week), and it is not significant for COB-workers.6 The third column shows that the impact on the worker’s probability to obtain a formal job varies with the type of job displacement. 7 PC-workers increase by 2.6 pp their probability of obtaining a formal job relative to non-displaced workers, and Q-workers increase their probability by 4.0 pp. In contrast, the probability drops for COB and F-workers, by 7.1 and 5.3 pp, respectively. The dependent variable in the fourth column is the number of years that workers have stayed in their current job, a proxy measure for job stability. Q-workers have the longest tenure in their current job (just 3.4 fewer years than the tenure of non-displaced workers) and COB and F-workers have the shortest tenures (approximately 6 fewer years than the tenure of non-displaced workers). The estimates in Table 4 show that among the four types of job displacement groups, plant closing workers experience the largest drop in wages and the longest increase in hours of work. F-workers also have a sizeable drop in wages and a reduction in the probability of finding a formal job. In sum, these two categories experience the largest decline in job quality, measured in terms of wages, hours of work and formality status. 6 For the rest of the paper PCE-workers refers to plant closing workers, Q-workers to quitting workers, F-workers to fired workers and COB to those who closed their own business. 7 We consider employment as informal if one of the following cases is true: 1) self-employed workers in the agriculture sector; 2) unpaid workers; and 3) workers without social security. 15 Table 4. Estimates of job displacement on labor market outcomes (1) (2) (3) (5) Log of hourly wage Hours of work Formality Tenure Plant Closing -0.0749*** 1.4492*** 0.0262*** -5.0338*** (0.006) (0.159) (0.004) (0.110) Fired -0.0687*** -0.1206 -0.0529*** -5.8829*** (0.009) (0.174) (0.005) (0.181) Quit -0.0229*** 1.7405*** 0.0401*** -3.3947*** (0.005) (0.149) (0.004) (0.097) Closed Own Busines -0.0532*** 0.3016 -0.0714*** -6.2715*** (0.005) (0.197) (0.007) (0.119) Observations 1,229,280 1,579,584 1,671,031 1,660,569 R-squared 0.3239 0.1302 0.2644 0.3766 Note: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.10. All regressions control for year of education, gender, marital status, age, age square, survey period, state and industry fixed effects. Standars errors are clustered at the state level. The ommited category is non-displaced workers. We now investigate whether the estimates in table 4 persist or tend to disappear over time. Figures 1–9 show lagged-coefficients (and confidence intervals) of different types of job displacement on real wages, weekly hours of work and formality status. The results show that a plant closing event has a long-lasting negative impact on real wages, as they decline by approximately 11 percent in the first two years after displacement, narrows to 5.5 percent after four years, and to 2 percent after nine years (figure 1). The increase in hours of work is permanent, PC-workers spend 1.5 additional hours per week than non-displaced workers (figure 2). Finally, plant closings have a short-term negative impact on the probability of working in the formal sector. Within the year of displacement PC-workers have a 2 pp. reduction in the probability of being formal, but a year after the probability increases by 2.7 pp. —above the probability of non-displaced workers— and stabilize to 5 pp. after nine years (figure 3). This finding is interesting for the discussion about remaining potential omitted-variable bias, because if the quality of the 16 match between employers and workers were an important determinant of worker performance after a separation due to a plant closing, then why should it not show up as a lasting negative impact on the probability of being formally employed? Figure 1. Plant closing coefficients on log hourly wages 0 -0.02 -0.04 -0.06 -0.08 -0.1 -0.12 -0.14 0 1 2 3 4 5 6 7 8 9 Lag years of plant closing Figure 2. Plant closing coefficients on weekly hours 3 of work 2.5 2 1.5 1 0.5 0 0 1 2 3 4 5 6 7 8 9 Lag years of plant closing Figure 3. Plant closing coefficients on formality 0.1 status 0.05 0 -0.05 0 1 2 3 4 5 6 7 8 9 Lag years of plant closing 17 Q-workers end up obtaining higher wages in the near term (figure 4). Although within the first year they experience a 1 percent reduction in wages, a year after they already have a 2 percent higher wage, and the gap increases further to 4 percent in the third year to reach 11.5 percent after the ninth year. This result confirms that, as expected from a voluntary separation, Q-workers end up obtaining higher wages than workers that experience a plant closing event. For Q-workers the impact on hours of work disappears gradually, however, during the first three years after quitting, they work between 1.7 and 2.5 additional hours than non-displaced workers (figure 5). Finally, the estimates show that Q-workers have permanent increase (in the range of 7 and 10 pp.) in the probability of working in the formal sector than the non-displacement group (figure 6). Figure 4. Quitting coefficients on log hourly wages 0.15 0.1 0.05 0 -0.05 0 1 2 3 4 5 6 7 8 9 Lag years of plant closing Figure 5. Quitting coefficients on weekly hours of 4 work 3 2 1 0 -1 0 1 2 3 4 5 6 7 8 9 18 Figure 6. Quitting coefficients on formality status 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 0 1 2 3 4 5 6 7 8 9 Lag years of plant closing COB-workers have an initial wage drop of 12 percent; however, the negative impact vanishes rapidly and by the second year the wage gap narrows to 5 percent and by the fourth year the estimates are no longer statistically different from zero (figure 7). COB-workers do not experience an impact on hours of work (figure 8), but they have a reduction in the probability of working in the formal sector. During the first five years after displacement, they have a probability in the range of 8 and 11 pp. lower than the non-displaced group of working in the formal sector, however, in the sixth year the estimates are no longer statistically different from zero (figure 9). Figure 7. Closed own business coefficients on log hourly wages 0.1 0.05 0 -0.05 -0.1 -0.15 -0.2 0 1 2 3 4 5 6 7 8 9 Lag years of plant closing 19 Figure 8. Closed own business coefficients on w2eekly hours of work 1 0 -1 -2 0 1 2 3 4 5 6 7 8 9 Lag years of plant closing Figure 9. Closed own business coefficients on 0.05 formality status 0 -0.05 -0.1 -0.15 0 1 2 3 4 5 6 7 8 9 Lag years of plant closing A1. Results by gender and educational categories In this section we explore the presence of differentiated effects of job displacement due to plant closing on labor outcomes by educational and gender categories. Descriptive statistics show significant labor outcome differences, for instance, throughout the analysis workers with more educational attainment have higher unemployment rates than workers with low education. Similarly, during the pre-crisis period —and since mid-2015— women have consistently higher unemployment rates than men (figures 10 and 11). 20 The results by educational categories show the existence of differentiated effects.8 Workers with high education (tertiary), in relation to workers with low education (primary and secondary), on average, have larger wage losses (12.1 percent vs 5.3 percent), a lower increase in hours of work (0.8 vs 1.5 hours), and a reduction in the probability of working in the formal sector (-0.5 vs 0.5 pp.). We further explore these results by estimating lagged coefficients of plant closing. For the 8 The interpretation of the coefficients is with respect of the benchmark of non-displaced workers in each educational category; for example, individuals with tertiary education who experienced a plant closing event have a wage reduction of 12.1 percent with respect of workers with similar education level in the non-displaced group. 21 low education group wages initially drop by 9.0 percent, the gap narrows to 5 percent in the second year and vanishes completely after the sixth year. For the high education group, the effect is considerably larger and permanent, with an initial drop of 18.4 percent, which narrows to 11.6 percent between the third and the sixth year and persists up to the ninth year with a wage gap of 6 percent (figure 11). The effect on hours of work is stronger for the low education group, with an estimate of 3 additional hours per week, an impact that persist over the entire horizon of analysis. For the high education group, the estimates point to 2 additional hours of work during first three years, with a gradual reduction over time that stabilizes at 1 additional hour at the end of the nine-year horizon (figure 12). Finally, there is a marked difference in the impact on the probability of working in the formal sector (figure 13). For the high education group, there is a transitory negative impact, with an initial drop of -10.0 pp., which narrows rapidly to -1.0 pp. between the third and sixth year, vanishing completely in the seventh year. The opposite occurs for the low education group, with a permanent increase in the probability that range between 7 and 10 pp. over the nine-year period. Table 5. Estimates of job displacement on labor market outcomes by education categories Log of hourly wage Hours of work Formality Primary and Primary and Primary and Tertiary Tertiary Tertiary Secondary Secondary Secondary Plant Closing -0.0536*** -0.1218*** 1.4798*** 0.8475*** 0.0461*** -0.0480*** (0.006) (0.006) (0.177) (0.125) (0.004) (0.005) Fired -0.0624*** -0.1229*** -0.1674 -0.0864 -0.0377*** -0.1099*** (0.010) (0.007) (0.203) (0.137) (0.005) (0.007) Quit -0.0120** -0.0412*** 1.9146*** 0.9811*** 0.0545*** -0.0081 (0.006) (0.006) (0.159) (0.127) (0.004) (0.005) Closed Own Business -0.0285*** -0.1266*** 0.0998 0.6714*** -0.0561*** -0.1265*** (0.005) (0.012) (0.220) (0.238) (0.007) (0.009) Observations 932,874 296,406 1,157,290 422,294 1,221,797 449,234 R-squared 0.2072 0.2782 0.1300 0.1470 0.2277 0.1218 Note: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.10. All regressions control for year of education, gender, marital status, age, age square, survey period, state and industry fixed effects. Standars errors are clustered at the state level. The omitted category is non-displaced workers. 22 Figure 11. Plant closing coefficients on log hourly wages: high and low educational grups 0.05 0 -0.05 -0.1 -0.15 Tertiary education -0.2 Primary and Secondary Education -0.25 0 1 2 3 4 5 6 7 8 9 Lag years of plant closing Figure 12. Plant closing coefficients on weekly hours of work: high and low educational grups 5 Tertiary education Primary & Sec education 4 3 2 1 0 0 1 2 3 4 5 6 7 8 9 Lag years of plant closing Figure 13. Plant closing coefficients on formality status: high and low educational groups 0.15 0.05 -0.05 Tertiary education Primary & Sec. education -0.15 0 1 2 3 4 5 6 7 8 9 Lag years of plant closing The results by gender show that the Mexican labor market reabsorbs displaced men and women similarly, however, point estimates suggest a slightly different effect of plant closing on wages, with a larger impact for males (a reduction of 8.0 percent) than for females (a reduction of 6.7 23 percent).9 We further explore this difference by estimating lagged coefficients of plant closing on real wages. The results in figure 14 confirm the existence of short-lived gender differences, which operate only in the first two years after the plant closing event has occurred, notwithstanding, after the second year the recovering of wages follow a similar path for men and women with no statistically significant differences in the magnitude of the coefficients. Figure 14. Plant closing coefficients on log hourly wages: male and female estimates 0.05 Male 0 Female -0.05 -0.1 -0.15 0 1 2 3 4 5 6 7 8 9 Lag years of plant closing To explore the duration of unemployment spells we estimate a hazard model by gender and educational categories. For this analysis the benchmark category consists of Q-workers.10 Table 6 presents both the coefficients and hazard rates ratios of the Weibull model. Men displaced due to plant closings, at each survival time, have a hazard rate equal to 83 percent of the hazard rate of male workers that quit, while for women the similar number is 93 percent. The results suggest that the unemployment spell that follow a plant closing event, relative to the spell of their counterparts that quit, is longer for men than for women (figures 15a and 15b). The coefficients for other types of displacement are not statistically significant. 9 In the Annex, tables A.1 and A.2 present average impacts by gender. 10 By construction this analysis compares unemployment spells among displaced workers, excluding non-displaced workers from the analysis. 24 There are moderately differentiated impacts by educational categories (table 7). High education group displaced due to plant closing has a hazard rate of 82 percent, approximately five percentage points less than the hazard rate of the low education group (87.6 percent). The survival functions of displaced workers due to plant closing for each educational group confirm that the low education group, at any time, has a higher probability of exiting unemployment (figures 16a and 16b). In sum, after a plant closing event, the unemployment spell for high educated workers is longer than the spells of less educated workers. Table 6. Weibull Hazard Model of Unemployment Spell Male sample Female sample Coefficients Hazard Ratio Coefficients Hazard Ratio Age -0.0043*** 0.9957*** 0.0023*** 1.0023*** (0.000) (0.000) (0.001) (0.001) Years Educ. -0.0228*** 0.9775*** -0.0074*** 0.9926*** (0.001) (0.001) (0.002) (0.002) Plant Closing -0.1832*** 0.8326*** -0.0745*** 0.9282*** (0.010) (0.009) (0.016) (0.015) Fired 0.0024 1.0024 -0.0262 0.9741 (0.010) (0.010) (0.018) (0.018) Closed Own Business 0.0213 1.0215 0.0341 1.0347 (0.019) (0.019) (0.028) (0.029) Alpha 1.26 1.26 Obs. 66,074 66,074 26,727 26,727 Note: All estimates control for marital status, state, survey period and industry fixed effects. The omi tted ca tegory i s workers tha t quit from the last job. Figure 15a Figure 15b 25 Table 7. Weibull Hazard Model of Unemployment Spell Low Education High Education Coefficients Hazard Ratio Coefficients Hazard Ratio Age -0.0021*** 0.9991** 0.0005 1.0006 (0.000) (0.000) (0.001) (0.001) Female -0.3025*** 0.7416*** -0.1506*** 0.8578*** (0.009) (0.007) (0.017) (0.014) Plant Closing -0.1308*** 0.8762*** -0.1972*** 0.8216*** (0.010) (0.009) (0.020) (0.016) Fired 0.0378*** 1.0457*** -0.1489*** 0.8630*** (0.010) (0.010) (0.022) (0.019) Closed Own Business 0.0381** 1.0401** -0.0193 0.9805 (0.018) (0.018) (0.036) (0.035) Alpha 1.27 1.26 Obs. 75,983 75,983 16,903 16,903 Note: All estimates control for marital status, state, survey period and industry fixed effects. The omi tted ca tegory i s workers tha t quit from the last job. Figure 16a Figure 16b B. Difference in differences estimation The DID results reported in table 8 show that displaced workers due to plant closing experience a reduction in wages of -4.7 percent, a moderate increase of 0.53 hours of work per week, and no impact on the probability of working in the formal sector. Compared to OLS estimates, DID reduces the magnitude of the impact of plant closing on wages, from a reduction of -7.5 to -4.7 percent, and the effect on hours of work from 1.4 to 0.53 hours per week, while the DID coefficient 26 on probability of working in the formal is not statistically different from zero. We interpret these changes in the magnitudes of the estimates as evidence that OLS estimates tend to exacerbate the impact of plant closing on labor outcomes. The DID estimation confirms the existence of heterogeneous effects of plant closing on labor outcomes. Highly educated workers have a reduction in wages of 9.3 percent, three times the decline for low educated workers (-2.7 percent). The increase in hours of work of high educated workers (0.57 hours) is slightly smaller than the increase of low educated workers (0.72 hours), and there is a marked difference of the effect of plant closing on the probability of working in the formal sector, the high education group have a reduction of 4.2 pp., while the low education groupan increase of 11.8 pp, in line with the results of figure 13. The DID estimation rejects the existence of gender-differentiated effects on wages, as both men and women have similar wage reductions, of 4.8 and 4.6 percent, respectively. However, we observe a slightly larger effect on hours of work for women (0.61 hours) than for men (0.48 hours), and a negligible impact on formality status. As with the OLS estimates, it is noteworthy that plant closings are unrelated to the subsequent formality status of workers, perhaps suggesting that remaining omitted variable biases are also negligible. Table 8. Difference in Difference estimates (1) (2) (3) Log of hourly wage Hours of work Formality All sample -0.0471*** 0.5311*** 0.0005 (0.006) (0.129) (0.004) Male -0.0485*** 0.4381** -0.0073** (0.006) (0.192) (0.003) Female -0.0461*** 0.6128*** 0.0114 (0.010) (0.219) (0.008) High Education -0.0933*** 0.5788** -0.0419*** (0.010) (0.281) (0.006) Low Education -0.0268*** 0.7233*** 0.0118*** (0.007) (0.188) (0.003) Note: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.10. All regressions control marital status, age, age square, survey period, state and industry fixed effects. Standars errors are clustered at the state level. The ommited category is non-displaced workers. 27 To further study potential biases in the DID estimates, Figure 17 presents the estimates of the industry fixed effects on the probability of occurrence of a plant closing event. Manufacturing is associated with a 14 percent higher probability of experiencing a plant closing, while agriculture is 36 percent less likely to experience this type of job separation. The results from figure 10 indicate the importance of controlling for workers’ industry of employment at the time of job separation. We re-estimate specification 2 with a subsample of workers for whom we can control for their industry of employment at the time of displacement and confirm that the results are robust to the inclusion of the industry of employment at the time of job displacement. 11 Furthermore, it is evident that plant closings are a phenomenon of agricultural sectors, particularly manufacturing. To the extent that higher quality workers and higher quality employer-employee matches are in the manufacturing sector where large enterprises operate rather than in agriculture, combined with the results regarding formality status of workers, these results lend further credence to the speculation that the DID estimates in particular are unlikely to be severely biased due to the omission of employer-employee fixed effects. Nonetheless, this remains a fertile area for future research. Figure 17: Industry fixed effects on probability of plant closing 0.50 0.35 0.20 0.05 -0.10 -0.25 -0.40 -0.55 11 The ENOE collects information on the industry of employment at the time of job displacement for only a subsample of workers— those who experienced a job termination within the last two years of the survey interview. 28 IV. Summary of Findings and Future Research We estimate the average impact of job displacement on labor market outcomes and ask whether these impacts are temporary or permanent. The results identify labor outcomes with the largest and the long-lasting impacts, confirming the existence of scarring effects of job displacement on workers´ future labor market outcomes. The empirical analysis focuses on heterogeneous effects on gender and educational groups. We find that the Mexican labor market reabsorbs men and women similarly, as both types of workers present similar recovering paths of wages, hours of work and probability of being formal after a plant closing event. We find marked differences in labor outcomes that follow plant closing events for high and low educated workers. Plant closing has a strong and permanent reduction on wages of high educated workers, while for low educated workers the impact on wages vanishes completely after the sixth year. Displaced workers due to plant closing have a permanent increase in the hours of work per week, with an impact on low educated workers that is double the magnitude of the impact on high educated workers. Additionally, plant closing permanently increases the probability of working in the formal sector for low educated workers. The estimates of the hazard model show that after a plant closing event, unemployment spells –relative to their non-displaced counterpart– are longer for men than for women, and longer for high than for low educated workers. An important result of this paper is the comparison of OLS and DID estimators. Using the rotating panel of the labor survey, we obtain the DID estimator and remove any time invariant fixed effect that could be biasing the OLS estimates of the cross-section analysis. The DID results show that displaced workers due to plant closing experience a reduction in wages of -4.7 percent, a moderate increase of 0.53 hours of work per week, and no impact on the probability of working in the formal 29 sector. Compared to OLS estimates, DID reduces the magnitude of the impact of plant closing on wages, from -7.5 to -4.7 percent, and the effect on hours of work from 1.4 to 0.53 hours per week, while the DID coefficient on probability of working in the formal is not statistically different from zero. We interpret this changes in the magnitude of the estimates as evidence that OLS estimates are upwardly biased due to the omission of unobserved individual worker characteristics. Finally, it is worth acknowledging that the DID estimates do not control for another potential source of omitted-variable bias: the quality of the match between employers and workers. That is, it is possible that the probability of a plant closing is systematically correlated with the quality of the match between the employer and its workers. Since the survey data, which nonetheless covers both informal and formal workers, does not offer information on the employer-worker relationships, we are unable to control for this potential source of omitted-variable bias. Future research should attempt to re-estimate the DID while controlling for employer-employee fixed effects, which requires administrative data, which would unfortunately leave out the informal labor market. This said, the fact that we did not find strong evidence of plant-closings being associated with higher probabilities of being informal after job separations could be interpreted as there being little remaining bias in our DID estimates since poor quality matches between employers and workers would presumably also be associated with informal employment of said workers. 30 References Bonikowska, A. and Morissette, R. (2012) “Earnings Losses of Displaced Workers with Stable Labour Market Attachment: Recent Evidence from Canada”, Statistics Canada, Analytical Studies Branch. URL: https://www150.statcan.gc.ca/n1/en/catalogue/11F0019M2012346 Couch, K. A. and Placzek, D. W. (2010) “Earnings Losses of Displaced Workers Revisited”, American Economic Review, 100 (1), pp. 572-589. DOI: 10.1257/aer.100.1.572 Davis, S. and Von Wachter, T.M. (2011) “Recessions and the Costs of Job Loss”, NBER, Working Paper 17638. Available at: https://www.nber.org/papers/w17638 Eliason, M. and Storrie, D. (2006) "Lasting or Latent Scars? Swedish Evidence on the Long-term Effects of Job Displacement", Journal of Labor Economics, 24(4), pp. 831–856. DOI: https://doi.org/10.1086/506487 Eslava, M., Haltiwanger, J., Kugler, A. and Kugler, M. (2010) “Trade liberalization and worker displacement: Evidence from trade reforms in Colombia”, Mimeo. URL: https://conference.iza.org/conference_files/worldb2011/kugler_a157.pdf Hijzen, A., Upward, R. and Wright, P. (2010) "The Income Losses of Displaced Workers", The Journal of Human Resources, 45(1), pp. 243–269. DOI: 10.3368/jhr.45.1.243 INEGI (2007) Cómo se hace la ENOE. Métodos y Procedimientos. Available at: https://en.www.inegi.org.mx/contenidos/productos/prod_serv/contenidos/espanol/bvinegi/produc tos/nueva_estruc/702825190613.pdf Jacobson, R., LaLonde, R. and Sullivan, D. (1993) “Earnings Losses of Displaced Workers”, American Economic Review, 83, pp. 685-710. URL: https://www.jstor.org/stable/2117574 31 Kaplan, D., Martinez, G. and Robertson, R. (2005) “What Happens to Wages After Displacement?”, Journal of the Latin American and Caribbean Economic Association, 5(2), pp. 197-234. URL: https://mpra.ub.uni-muenchen.de/3079/ Messina J. and Silva, J. (2018) Wage Inequality in Latin America. Understanding the past to prepare for the future. Washington, DC: The World Bank Group. Christopher, R. (1991) “Are Workers Permanently Scarred by Job Displacement?”, American Economic Review, 81, pp. 319-324. URL: https://www.jstor.org/stable/2006805 Stevens, A. H., (1997) “Persistent Effects of Job Displacement: The Importance of Multiple Job Losses”, Journal of Labor Economics, 13, pp. 165-188. DOI: https://doi.org/10.1086/209851 Toulemon, L. and Weber-Baghdiguian, L. (2016) “Long-term Impact of Job Displacement on Job Quality and Satisfaction: Evidence from Germany”, Paris School of Economics. Working Paper No. 2016-32. URL: https://halshs.archives-ouvertes.fr/halshs-01418183/ 32 Annex Table A1. Male sample: estimates of job displacement on labor market outcomes (1) (2) (3) Log of hourly wage Hours of work Formality Plant Closing -0.0802*** 0.7610*** 0.0101*** (0.005) (0.140) (0.004) Fired -0.0724*** -0.6139*** -0.0611*** (0.010) (0.182) (0.005) Quit -0.0209*** 1.7296*** 0.0421*** (0.006) (0.109) (0.005) Closed Own Busines -0.0527*** 0.3841** -0.0681*** (0.006) (0.176) (0.007) Retired -0.1364*** -3.6497*** -0.2221*** (0.013) (0.346) (0.012) Other -0.0889*** -0.9931*** -0.0920*** (0.006) (0.181) (0.005) Observations 753,619 966,028 1,018,629 R-squared 0.3197 0.1179 0.2818 Note: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.10. All regressions control for year of education, gender, marital status, age, age square, survey period, state and industry fixed effects. Standars errors are clustered at the state level. The ommited category is non-displaced workers. Table A2. Female sample: estimates of job displacement on labor market outcomes (1) (2) (3) Log of hourly wage Hours of work Formality Plant Closing -0.0667*** 1.7968*** 0.0308*** (0.007) (0.215) (0.006) Fired -0.0701*** -0.0953 -0.0551*** (0.007) (0.205) (0.005) Quit -0.0291*** 1.0244*** 0.0167*** (0.006) (0.185) (0.004) Closed Own Busines -0.0552*** -0.8703*** -0.1025*** (0.009) (0.280) (0.007) Retired -0.0830*** -4.2678*** -0.2740*** (0.023) (0.491) (0.012) Other -0.0855*** -1.0473*** -0.0793*** (0.013) (0.310) (0.007) Observations 475,661 613,556 652,402 R-squared 0.3383 0.0666 0.2813 Note: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.10. All regressions control for year of education, gender, marital status, age, age square, survey period, state and industry fixed effects. Standars errors are clustered at the state level. The ommited category is non-displaced workers. 33