The Decentralization of Wage Bargaining and Income Losses after Worker Displacement

The Decentralization of Wage Bargaining and Income Losses after Worker Displacement Abstract This paper uses administrative data to study the relationship between the decentralization of wage bargaining systems and the costs of worker displacement. Specifically, the paper exploits a major reform of the wage bargaining system in the Danish manufacturing sector, a reform that changed the wage-setting process from a highly centralized bargaining system at the national level to a decentralized system with a strong emphasis on firm-level wage bargaining. The results show that under the centralized wage bargaining system, displaced workers’ income losses were small, whereas under the decentralized wage bargaining system, these income losses increased substantially, particularly because displaced workers experienced worse wage growth under the decentralized system. The effect persists after controlling for a variety of macroeconomic indicators, and displaced workers’ income losses did not increase in sectors that were not affected by a comparable change in the wage bargaining system. 1. Introduction A great number of studies from many different countries show that job displacement results in large and long-lasting income losses (e.g., Jacobson, LaLonde, and Sullivan 1993; Bender, Dustmann, Margolis, and Meghir 2002; Kletzer and Fairlie 2003; Von Wachter, Weber Handwerker, and Hildreth 2009; Schmieder, von Wachter, and Bender 2010; Huttunen, Møen, and Salvanes 2011).1 However, the results are very diverse, and many of the studies have shown that the magnitude of displaced workers’ income losses varies strongly by the choice of empirical methods and data sources, by industry, and by business cycle conditions (e.g., Couch and Placzek 2010; Hijzen, Upward, and Wright 2010; Davis and von Wachter 2011). Nonetheless, researchers often find displaced workers’ income losses to be larger in the United States, where wage bargaining is decentralized, than in Europe, where many countries have centralized wage bargaining systems. Therefore, some researchers have argued that country-specific differences in displaced workers’ income losses may also reflect country-specific differences in institutions, such as wage bargaining systems (e.g., Burda and Mertens 2001; Bender et al. 2002). Yet no study has directly analyzed whether displaced workers experience larger income losses under decentralized wage bargaining systems than under centralized ones. Moreover, why displaced workers should experience larger income losses under decentralized wage bargaining remains unclear. The first contribution of this paper is to exploit a drastic reform of the wage bargaining system in the Danish manufacturing sector, to provide direct evidence for the relationship between the decentralization of wage bargaining and increasing income losses of displaced workers. The reform is ideal for studying this relationship, because it changed the wage bargaining from a centralized system to a decentralized one within a short period (e.g., Eriksson and Westergaard-Nielsen 2009). Moreover, I can rely on Statistics Denmark’s detailed register data, which allows me to study displaced workers’ income, wage, and unemployment patterns over long periods, both before and after the introduction of the decentralized wage bargaining system. The second contribution of this paper is to analyze the major channels through which the wage bargaining reform has affected the magnitude of displaced workers’ income losses. Centralized wage bargaining systems may impose rigid wage structures, such that displaced workers experience higher unemployment but lower wage losses under centralized wage bargaining systems, and experience higher wage losses but lower unemployment under decentralized ones. But such a simple model cannot explain why displaced workers’ income losses are larger under decentralized wage bargaining. Yet, a number of recent studies show that flexible wages resemble individual productivity more closely and induce greater variability in firm wage premiums (e.g., Card, Heining, and Kline 2013; Dahl, Le Maire, and Munch 2013). Thus, under decentralized wage bargaining, displaced workers’ losses of firm-specific human capital and firm wage premiums may result in larger income losses. The first contribution shows that the decentralization of the wage bargaining system is related to a substantial increase in displaced workers’ gross income losses including unemployment insurance (UI) and other social benefits. Before the reform, displaced workers’ income losses were about 1% of their average predisplacement income per year. Thereafter, displaced workers’ income losses ranged from 6% to 7% per year, and the losses appear very persistent over the long term. I provide a large number of robustness checks to ensure that neither other unrelated institutional changes nor changing macroeconomic conditions are driving the results. I show that the increase in displaced workers’ income losses persisted across changes in the business cycle. Moreover, displaced workers’ income losses did not increase in sectors that were affected by similar business cycle conditions and institutional changes but that were not exposed to a comparable change in their wage bargaining systems. The second contribution shows that the increase of displaced workers’ income losses under the decentralized wage bargaining system were largely related to higher losses in real wages, that is, wage losses increased from 0% to about 5%. In contrast, I do not find a comparable structural break in displaced workers’ unemployment patterns after the reform. Moreover, on average displaced workers reentered employment at wage levels very close to those of their predisplacement wages, both before and after the reform. These results are inconsistent with the common predictions of a simple neoclassical model, in which displaced workers trade off wage losses against unemployment. Instead, displaced workers experienced large income losses under the decentralized wage bargaining system, because they forwent wage growth, not because they initially reentered firms at lower wages. Under the centralized wages bargaining system, displaced workers appear to have benefited from wage drifts of general contracts and centralized negotiations, whereas firm-specific human capital and tenure became more important after the reform. In more detail, displaced workers who lost their jobs before the reform profited from considerable wage growth irrespective of the amount of tenure they accumulated after the job loss. In contrast, displaced workers who lost their jobs after the reform experienced much less wage growth after their job loss, but their returns to postdisplacement tenure more than doubled. Moreover, the variation in postdisplacement entry wages increased substantially after the reform, suggesting that firm-specific wage premiums and assortative matching may have become more important for displaced workers’ wage development under decentralized wage bargaining systems. However, the results also indicate that displaced workers with lower predisplacement wages refrain from reentering jobs with wages below the level of those in their previous job. In contrast, displaced workers with predisplacement wages above the median appear to have reentered employment more quickly and at relatively lower wages. As UI benefits in Denmark are relatively more generous for low- than for high-income workers, UI benefits may prevent the commonly expected trade-off between wage losses and unemployment—at least for displaced workers with lower wages. This paper contributes to the large literature on displaced workers’ income losses. Although many studies show that these income losses differ by worker characteristics, economic conditions, data sources, and countries,2 this paper is the first to provide direct evidence that labor market institutions such as wage bargaining systems determine the magnitude of displaced workers’ income losses. The paper also contributes to the literature showing that the decentralization of wage bargaining systems is related to rising skill premiums, rising returns to tenure and experience, and rising heterogeneity of firm-specific wage premiums (e.g., Card, Lemieux, and Riddell 2004; Dustmann, Ludsteck, and Schönberg 2009; Card et al. 2013; Dahl et al. 2013). This paper extends that literature by showing that displaced workers, who commonly lose firm-specific wage premiums and lack tenure, forgo substantially more wage growth under decentralized wage bargaining systems than under centralized ones. The remainder of this paper is structured as follows. Section 2 provides detailed background information on the Danish wage bargaining system and Denmark’s macroeconomic performance. Section 3 describes the estimation methods, data, and sample selection. Section 4 presents the main results, and Section 5 empirically analyzes the mechanisms to explain the results in Section 4. Section 6 provides a sensitivity analysis, and Section 7 concludes. 2. Institutions and Macroeconomic Conditions To provide the necessary background for interpreting the empirical results, this section describes the changes in the Danish wage bargaining system, the UI system, and macroeconomic performance. 2.1. Decentralization of the Wage Bargaining System Figure 1 shows the development of the Danish wage bargaining system, along with indicators of Denmark’s macroeconomic performance. The solid vertical lines in Figure 1 indicate the critical changes in wage bargaining in the manufacturing sector. Between 1956 and 1989, the Danish wage bargaining system was highly centralized and largely characterized by standard tariff wage contracts that established ranges for workers’ wages according to their occupation, education, and work experience. These contracts were not modified at the firm level, and wage floors and ceilings further limited the scope for firm-level wage bargaining. Moreover, a cost-of-living adjustment automatically tied workers’ wages to national inflation. Figure 1. View largeDownload slide Macroeconomic conditions and the wage bargaining system. The dashed line marks the national rate of registered unemployment. The solid line marks the share of standard tariff contracts in the DA/LO sector. The DA does not provide official data for standard tariff contracts for the time before 1989. However, Pedersen, Smith, and Stephensen (1998) report that 45% of all workers were covered by standard wage rate contracts before 1989. The vertical lines indicate changes in the wage bargaining system. Source: Statistics Denmark and Dansk Arbejdsgiverforening (2004) Arbejdsmakedrapport. Figure 1. View largeDownload slide Macroeconomic conditions and the wage bargaining system. The dashed line marks the national rate of registered unemployment. The solid line marks the share of standard tariff contracts in the DA/LO sector. The DA does not provide official data for standard tariff contracts for the time before 1989. However, Pedersen, Smith, and Stephensen (1998) report that 45% of all workers were covered by standard wage rate contracts before 1989. The vertical lines indicate changes in the wage bargaining system. Source: Statistics Denmark and Dansk Arbejdsgiverforening (2004) Arbejdsmakedrapport. Every two years, individual national unions and industry employer associations had the right to negotiate directly at the industry level. However, if an industry-level agreement was not reached, the Danish Federation of Trade Unions (LO) and the Danish Employer Federation (DA) were required to negotiate on behalf of their affiliates. If the LO and DA were also unable to reach an agreement, a state mediator had to mediate the dispute and make a proposal. The parliament typically enforced these proposals, even if the LO or DA rejected them (Wallerstein and Golden 1997). The cost-of-living adjustment was revoked in 1985, when the last national-level bargaining round took place (Dansk Arbejdsgiverforening (2004) Arbejdsmakedrapport). In 1987, employer organizations and unions negotiated at the industry level, agreeing to shift wage bargaining power away from the national level.3 Although the agreement was set for four years, the social partners renegotiated it after two years. Renegotiations in January 1989 marked a pivotal change in the manufacturing sector, with sector-specific negotiations between DI (a DA member organization that organizes firms in the manufacturing sector) and CO-metal (a bargaining conglomerate representing unions organizing workers employed in DI firms). These renegotiations shifted a substantial number of wage negotiations to the local level and substantially increased the scope of firm-level wage bargaining (Andersen and Svarer 2007). Although this decentralization process has affected all sectors of the Danish economy in recent decades, both unions and employer organizations in the manufacturing sector initiated this process in 1989. The metal industry’s skilled workers union, Dansk Metallarbejderforbund—traditionally the main proponent of a centralized wage bargaining system—was the first to accept strong decentralization in the late 1980s (Wallerstein and Golden 1997). Before 1989, standard wage rate contracts, for which wages were not modified at the firm level, consistently covered about 45% of the DA/LO area. Standard wage rate contracts are characterized by payment ranges based on workers’ education, occupation, and experience in the sector (Bingley and Westergaard-Nielsen 2003). Between 1989 and 1993, standard wage rate contracts became less widespread. In 1989, standard wage rate contracts covered only 34% of the DA/LO area. Between 1989 and 1993, this rate dropped to 16% and remained at this level until 2004 (see Figure 1). Between 1989 and 1993, wage arrangements leaving more room for firm-level wage bargaining became more popular. Specifically, the use of minimum wage and minimum pay systems, which represented wage floors for only very inexperienced workers, increased between 1989 and 1993 and covered nearly all remaining DA/LO firms and workers by 1993 (Andersen and Mailand 2005). Therefore, some researchers refer to the 1989–1993 period as one of coordinated decentralization within strict guidelines (Andersen 2003). Beginning in 1989, the social partners successively abandoned special guidelines such as ceilings on wage increases; instead, firm-level wage negotiations without any boundaries became more popular in the late 1990s and early 2000s, covering approximately 19% of DA/LO firms and workers by 2004.4 2.2. Unemployment Insurance In Denmark, UI is relatively generous, even in comparison to other European countries. In principle, the unemployment benefit compensates up to 90% of a worker’s previous wage. However, a relatively low cap of approximately 170,000 Danish kroners (DKKs) (e.g., 10,000 correspond to $1,160) restricts total compensation such that UI benefits are less generous for high-wage workers. In addition to common UI benefits, a specific “guarantee fund” compensates workers who forgo earnings if their employers are bankrupt or must lay them off for economic reasons. This fund guarantees ongoing salaries during the worker’s notice period, which commonly lasts up to three months.5 Most displaced workers who separate from closing firms are likely to receive payments from this fund. 2.3. Macroeconomic Performance Denmark experienced two economic downturns with high unemployment (see the dashed line in Figure 1). The first recession, in the early 1980s, was followed by a brief recovery period. In the late 1980s and the early 1990s, Denmark was hit by the “Nordic crisis” (i.e., a financial crisis in Denmark, Finland, Norway, and Sweden) (e.g., Kiander 2004). However, during the mid-1990s, Denmark experienced a phase of substantial economic growth and steadily decreasing unemployment (Kiander 2004). As the solid vertical lines in Figure 1 indicate, the major reform of the wage bargaining system took place at the beginning of a deep recession but remained flexible for the rest of the observation period, during which the economy recovered substantially. 3. Data For my investigation, I use the Integrated Database for Labor Market Research (IDA), a highly precise administrative data source that covers each worker and establishment in Denmark between 1980 and 2004 (e.g., Abowd and Kramarz 1999; Iversen, Malchow-Møller, and Sørensen 2010; Timmermans 2010). The Danish Bureau of Statistics collects administrative records from different register sources and matches the data, using a unique identification number for each individual and establishment. Every Danish resident is thus tracked and assigned to one—and only one—labor market state and work establishment on a specific date in November of each year. The data source contains a variety of worker and firm characteristics, including income, earnings, and firm size. 3.1. Identification of Worker Displacement As in most other register-based data sources, I cannot observe whether workers are displaced or leave their jobs for other reasons. Therefore, I follow the most common solutions to this problem by creating samples of workers separating from firms during mass layoffs and samples of comparable workers who do not separate from firms during mass layoffs (e.g., Jacobson et al. 1993; Couch and Placzek 2010; Davis and von Wachter 2011). In accordance with those previous studies, I define a mass layoff as a situation in which a firm’s workforce contracts by more than 30% from one observation year to the next. To distinguish true mass layoffs from regular employment volatility, I focus on firms with an employment of at least 50 employees before the mass layoff occurs. As in Jacobson et al. (1993), I incorporate closing firms with more than 50 employees. For many register data sources, researchers face the problem of misinterpreting simple changes in establishment identification numbers or transfers of large parts of the firm’s workforce to other buildings as mass layoffs. To avoid this confusion, the IDA provides prospective information on the status of each establishment (Timmermans 2010). Specifically, I have information for each successive year on whether an establishment (1) remains the same, (2) closes down, (3) merges with or is acquired by an existing establishment (i.e., either 30% or at least two workers move to the same other establishment), (4) closes down due to the transfer of its buildings to a new entering establishment, (5) moves to self-employed individuals without any employees, or (6) moves into a situation in which the firm has no employees. From this information, I apply the following additional restrictions to define a mass layoff: (1) Firms that experience a mass layoff but remain in business must remain the same during the mass layoff period and for at least three preceding observation periods (i.e., I avoid considering firms that transfer part of their workforce to other establishments); and (2) firms that experience mass layoffs but do not remain in business must go out of business when the separation takes place but must have remained the same for the three preceding observation periods.6 3.2. Displaced Workers Following the previous literature, my sample of displaced workers includes male manufacturing workers, 50 years or younger at the time of their displacement, who separate permanently from their establishments. I require displaced workers to have at least two years of tenure before their displacement occurs. During those two years, displaced workers must have been wage employed, that is, registered as wage employed for 365 days in each of the two predisplacement years. This restriction excludes seasonal workers and ensures that I do not mistakenly assign a share of the displacement loss to the year preceding the displacement. After the event of displacement, I apply no further restrictions, so the displaced workers may be employed, unemployed, or even out of the labor force after their job loss. 3.3. Nondisplaced Workers To keep the sample of nondisplaced workers most comparable to the sample of displaced workers, these nondisplaced workers consist of male manufacturing workers who are 50 years or younger when their respective counterfactual group of displaced workers was displaced. They work for firms with more than 50 employees. Throughout each respective observation period, nondisplaced workers were never laid off in response to a mass layoff. More specifically, the group of nondisplaced workers consists of both those who work in firms that never underwent a mass layoff event and those who work in firms that underwent a mass layoff event but were not laid off during it. I require that nondisplaced workers have had at least three years of tenure—that is, two years before their counterfactual group of displaced workers lose their jobs and an additional year because they do not switch jobs in the displacement year. Thereafter, I allow nondisplaced workers to switch jobs, but they must have had wage employment with a positive income for each subsequent year. This restriction ensures that nondisplaced workers have had a continuous employment pattern.7 The first column of Table C.1 in the appendix presents results with less restrictive sample selection criteria, but the key results of this paper remain very robust. 3.4. Variables The analysis relies on four main dependent variables. The first measures workers’ annual taxable gross income. This gross income measure captures a worker’s entire taxable income, including payments from UI or pension funds and other types of income. Notably, the measure covers income from the “guarantee fund”. The second variable measures workers’ annual labor earnings. I calculate this variable adding the workers’ November job earnings and the sum of their wages from all other jobs in a given year. This variable provides information on workers’ total displacement costs in the form of pure labor earnings, which do not include any form of unemployment or state-financed benefits. Because recent studies specifically note that zero earnings or income are an important part of workers’ displacement costs (e.g., Davis and von Wachter 2011), I specifically include displaced workers with zero income or earnings after displacement. Both income and labor earnings are deflated and measured in 2000 DKKs. The third dependent variable measures the workers’ hourly wage rate. Following the previous literature, I calculate this variable by dividing the sum of total labor income and mandatory pension fund payments by the total number of hours worked in a given year (e.g., Dahl et al. 2013). Statistics Denmark provides additional information on the quality of the measured wage rate. From this information, I set all wage observations of poor quality to “missing”. Moreover, to disentangle the wage effect from the quantity effect, which arises as a consequence of unemployment or the reduction in working time, I exclude observations with zero wages. The fourth variable measures a worker’s degree of unemployment in percent of a given year. A high replacement ratio and almost no experience rating for employers or workers implies that workers experience many short periods of unemployment. Moreover, employees have the right to claim supplementary UI benefits while on vacation, so that UI periods appear to cluster around summer and Christmas. To avoid capturing such unemployment periods not related to effective unemployment, I exclude the UI periods of workers who were registered as wage employed for the entire year.8 Finally, I have information about workers’ age, education, industry, and real—not potential—work experience since 1964. 4. Methods The main problem in estimating the cost of worker displacement is that researchers have no information on how displaced workers’ income would have developed in the absence of their displacement. A common solution to this problem is to create samples of displaced workers and compare their careers to a counterfactual group of comparable nondisplaced workers. This paper also follows the counterfactual approach by applying versions of the following regression model, which follows Jacobson et al. (1993) and represents best practices in the literature. Equation (1) is the regression equation:   \begin{equation} y_{it}=\alpha _{i}+\lambda _{t}+x_{it}\beta +\sum _{k}D_{it}^{k}\delta ^{k}+\varepsilon _{it}, \end{equation} (1)where i indicates a single worker, and t represents the calendar time in years. $$D_{it}^{k}$$ is a set of dummy variables equal to 1 in the kth year before or after the worker’s displacement, for example, $$D_{i1985}^{2}$$ equals 1 for displaced workers who lost their jobs in 1983. Thus δk represents the effect of displacement in the kth year before or after displacement. For nondisplaced workers, $$D_{it}^{k}$$ is always 0. The vector xit contains a set of control variables, which I restrict to five age categories ranging from under 25 to over 55 years of age. Following previous papers, such as Davis and von Wachter (2011), I incorporate only age as a control variable, because observable worker and job characteristics other than age are either time constant (e.g., predisplacement education) or endogenous to the displacement itself (e.g., tenure, work experience, or industry). The parameter λt represents a set of coefficients for each year in the observation period. The fixed effect ai summarizes all observable and unobservable time constant differences of workers i. I assume εit to have constant variance and to be uncorrelated with individual and time effects. The dependent variable yit represents the dependent variable of worker i at time t. yit can be the workers’ gross income (including UI benefits and other social benefits), their annual labor earnings (excluding all benefits), their hourly wages, or their incidence of unemployment in a given year. Standard errors are clustered at the individual level. All monetary outcomes are measured in absolute DKK values, and I include zero income and earnings in yit, so that I cannot measure the dependent variable in log values. However, I provide figures that express monetary displacement losses as a percentage of the workers’ base income, earnings, or wages throughout the paper. Therefore, I divide the δk by the average income, earnings, or wages of all displaced workers in the second year before the displacement event. The figures also contain confidence intervals, which I obtained using the delta method.9 Tables A.3–B.1 in the appendix present monetary displacement losses in absolute values. I estimate equation (1) on 19 separate equidistant and overlapping panels, one for each cohort of displaced workers between 1982 and 2000. Each subsample observes each displaced worker for two years before and three years after displacement. For example, the first sample includes all workers displaced in 1982, and I observe this group for six years (from 1980 to 1985); the second sample includes all workers displaced in 1983, and I observe this group for six years (from 1981 to 1986), and so forth. This approach allows me to present a very desegregated picture of the development of displaced workers’ income losses over time, and each subsample follows workers of the treatment and control group for the same amount of time before and after the job loss. Moreover, to provide ideal conditions for unbiased and comparable estimates of δks within each equidistant subsample, the number of observation is large in comparison to the number of time periods for each sample.10 However, analyzing the costs of worker displacement in separated samples does not allow me to test for significant differences in displaced workers’ income losses across the years. Moreover, I am unable to capture global trends in λt, which may finally account for the increase in displaced workers’ income losses. Thus I estimate regression equations of the following form on a pooled sample containing all individuals of all equidistant samples11:   \begin{equation} y_{it}=\alpha _{i}+\lambda _{t}+ x_{it}\beta + \mathit{ Reform}\cdot x_{it}\beta _{R} + \sum _{k}D_{it}^{k}\delta ^{k}+\sum _{k}\mathit{ Reform} \cdot D_{it}^{k}\delta _{R}^{k}+\varepsilon _{it}, \end{equation} (2)where $$ {Reform}$$ is a dummy being one if a worker was displaced after the reform of the decentralized wage bargaining system and zero otherwise. The $$\delta _{R}^{k}$$s directly measure the effect of the wage bargaining reform. As the $$ {Reform}$$ dummy is multicolinear to the time dummies λt and time constant for most displaced workers,12 I cannot include the $$ {Reform}$$ dummy separately, that is, λt and αi jointly pick up the isolated $$ {Reform}$$ effect. I emphasize here that the empirical approach in (2) relies on a combined sample that includes members of the treatment and control group from different periods into one regression. 5. Main Results This section presents the main results. Section 5.1 shows descriptive statistics for the sample composition of displaced and nondisplaced workers. Section 5.2 presents the main results. 5.1. Sample Composition and Baseline Characteristics Tables A.1 (pre-decentralization) and A.2 (post-decentralization) in the appendix present average statistics for the main analysis variables of all 19 subsamples, with t-statistics to test for significant differences between displaced and nondisplaced workers. The number of displaced workers (according to the definition in the previous section) ranges between 230 individual workers in sample “1984–1989” and 1,038 workers in sample “1986–1991”. The number of nondisplaced workers ranges from 52,699 in sample “1985–1990” to 61,135 in sample “1996–2001”. Baseline characteristics were not fully balanced between displaced and nondisplaced workers. Even two years before the displacement, displaced workers had somewhat lower income than their nondisplaced counterparts. Moreover, significant differences exist for age and education. These differences are very comparable to the differences of predisplacement characteristics in other studies (e.g., Jacobson et al. 1993; Korkeamäki and Kyyrä 2014).13 However, the decentralization of the wage bargaining system fell in the middle of an economic downturn that may have caused a major shift in economic activities across Danish manufacturing sectors (see Figure 1 of Section 2). Moreover, firms may tend to lay off more high-ability workers when wages are rigid but fewer of them when wages are flexible. As a result, the sample composition of displaced workers may have changed substantially after the introduction of the decentralized wage bargaining system in 1989. Figures 2 and 3 present descriptive evidence of the sample composition of displaced and nondisplaced workers by showing workers’ industry and education over time. Figure 2 describes the composition of workers’ education within the group of displaced workers (black dots) and nondisplaced workers (crosses). The first subfigure presents the fraction of low-educated workers with less than an apprenticeship degree. The second subfigure presents the fraction of medium-educated workers with an apprenticeship degree or comparable educational courses. The third subfigure presents the fraction of high-educated workers with a master craftsman degree or a university degree. The x-axis indicates the displacement year of the respective subsample, and the solid lines present the results of kernel-weighted local polynomial regressions. Figure 2. View largeDownload slide Composition of education over time. The figure presents descriptive statistics for the fraction of displaced and nondisplaced workers who have a low, medium, or high education. Low-educated workers have less than an apprenticeship degree, medium-educated workers have an apprenticeship degree, and high-educated workers have a master craftsman or a university degree. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Figure 2. View largeDownload slide Composition of education over time. The figure presents descriptive statistics for the fraction of displaced and nondisplaced workers who have a low, medium, or high education. Low-educated workers have less than an apprenticeship degree, medium-educated workers have an apprenticeship degree, and high-educated workers have a master craftsman or a university degree. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Figure 3. View largeDownload slide Composition of manufacturing industries over time. The figure presents descriptive statistics for the fraction of displaced and nondisplaced workers in four different industry categories before the displacement. The category “other mfr. industries” contains industrial workers in textiles and leather, chemicals and plastic products, other nonmetallic mineral products, furniture, and other small industries. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Figure 3. View largeDownload slide Composition of manufacturing industries over time. The figure presents descriptive statistics for the fraction of displaced and nondisplaced workers in four different industry categories before the displacement. The category “other mfr. industries” contains industrial workers in textiles and leather, chemicals and plastic products, other nonmetallic mineral products, furniture, and other small industries. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Even if significant differences between displaced and nondisplaced workers existed, Figure 2 does not indicate a systematic break in the educational composition of displaced and nondisplaced workers immediately after 1989. Both groups largely consist of medium-educated workers, followed by low- and high-educated workers. Figure 3 shows the composition of workers’ industries over time. For this purpose, I assigned all workers to four categories according to their predisplacement industry within the manufacturing sector. The first three categories correspond to those three manufacturing industries with the largest fractions of displaced workers in the sample: (1) food, beverages, and tobacco; (2) basic metals and fabric metal products; and (3) wood products and printing and publishing materials. The fourth category (4) represents all remaining manufacturing industries with fewer displaced workers, such as textiles and leather, chemicals and plastic products, other nonmetallic mineral products, and furniture. The y-axis of each subfigure indicates the fraction of workers within the respective industry category separately for the group of displaced workers (black dots) and nondisplaced workers (crosses). The relative industry composition of nondisplaced workers remains relatively stable over time, suggesting no major shifts in the manufacturing sector throughout the observation period. In contrast, the industry composition of displaced workers varies substantially over time. Moreover, the industry composition of displaced and nondisplaced workers differs significantly (see Tables A.1 and A.2). These results show that mass layoffs are not a smooth process occurring at constant rates within and across industries. However, the figure shows no evidence that the industry composition of displaced workers changed systematically immediately after the introduction of the decentralized wage bargaining system. 5.2. Decentralization and Displaced Workers’ Income Losses Figure 4 shows average real income developments (including UI and other social benefits) for displaced and nondisplaced workers for each sample. Each subfigure corresponds to a separate subsample of workers, indicated by the title of the subfigure. All subfigures follow displaced and nondisplaced workers from two years before their displacement occurred until three years after the displacement occurred (i.e., 0 on the respective x-axis denotes the displacement year). For example, the first subfigure (“1980–1985”) depicts average gross income developments for workers who were displaced in 1982, along with the income developments of their comparison group of nondisplaced workers. To keep subfigures comparable and to account for predisplacement differences between displaced and nondisplaced workers, I measure all income trajectories relative to the average income in the second year before the displacement occurred. Figure 4. View largeDownload slide Average gross income developments (measured in thousands). Each subfigure refers to a sample of displaced and nondisplaced workers for the assigned year. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Gross income is measured in 2000 DKKs and includes zeros. Displacements occurred between November in year (−1) and November in year (0). Average income is adjusted to a base income in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). Figure 4. View largeDownload slide Average gross income developments (measured in thousands). Each subfigure refers to a sample of displaced and nondisplaced workers for the assigned year. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Gross income is measured in 2000 DKKs and includes zeros. Displacements occurred between November in year (−1) and November in year (0). Average income is adjusted to a base income in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). Figure 4 shows that displaced workers who lost their jobs between 1982 and 1989 (i.e., workers in the “1980–1985” to the “1986–1991” samples) did not experience substantial income losses relative to their comparison groups of nondisplaced workers. However, after 1989—the year of the pivotal change in the wage bargaining system (sample “1987–1992”)—a clear structural break in the results appears. Displaced workers who lost their jobs after 1989 had worse income developments than nondisplaced workers, that is, displaced workers experienced stagnating or even negative income growth after their displacement. In other words, relative to nondisplaced workers, displaced workers who lost their jobs after 1989 experienced visible income losses. Tables A.3 and A.4 (which, because of their length, are in the appendix) present the corresponding estimation results of equation (1). Figure 5 illustrates the results of both tables by calculating the discounted present values (DPVs) of displaced workers’ income losses for each displacement from these tables. The DPVs express the workers’ income losses as the number of income years lost at the previous level of income for each single displacement year τ between 1982 and 2000.14 Figure 5. View largeDownload slide Discounted present values of income losses (DPV) and wage bargaining system. The figure presents average DPVs (interest rate 5%) of displaced workers’ income losses over the years. Gross income is measured in 2000 DKKs and includes zeros. Displacements occur between November in year (−1) and November in year (0). Average DPVs of displacement losses are expressed as the number of income years lost at the level of income two years before the displacement. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Figure 5. View largeDownload slide Discounted present values of income losses (DPV) and wage bargaining system. The figure presents average DPVs (interest rate 5%) of displaced workers’ income losses over the years. Gross income is measured in 2000 DKKs and includes zeros. Displacements occur between November in year (−1) and November in year (0). Average DPVs of displacement losses are expressed as the number of income years lost at the level of income two years before the displacement. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Figure 5 shows a substantial increase in the DPVs of displaced workers’ income losses after the wage bargaining system was decentralized. Before 1989, displaced workers lost between 0 and 0.05 years of their predisplacement income, that is, less than one month’s income over a period of four years. In contrast, displaced workers lost up to 0.20 years of their predisplacement income, which amounts to two and a half times their monthly income over a period of four years. This difference is substantial, although even after 1989, Danish income losses appear smaller than those in the United States. Thus the estimation results clearly support the descriptive results in Figure 4. The structural break in the findings for displaced workers’ income losses precisely corresponds to the structural break reported in earlier investigations of wage distributions conducted by Danish researchers. For example, as previously mentioned, Eriksson and Westergaard-Nielsen (2009) show that the dispersion of the wage structure suddenly increased after 1989, and Bingley and Westergaard-Nielsen (2003) show that returns to tenure more than doubled at the beginning of the decentralization period in 1989. Figure 6 presents estimations on a pooled sample that includes all individuals from all subsamples. The estimation approach follows equation (2) with interaction terms between an indicator for the reform and the displacement indicators $$D_{it}^{k}$$ and interaction terms between the reform indicator and all remaining control variables. Figure 6. View largeDownload slide Average real gross income losses (pooled sample). Estimation according to equation (2). Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Gross income is measured in 2000 DKKs and includes zeros. Average income is adjusted to a base income in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). Regressions include worker fixed effects, year fixed effects, and age categories. The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 6. View largeDownload slide Average real gross income losses (pooled sample). Estimation according to equation (2). Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Gross income is measured in 2000 DKKs and includes zeros. Average income is adjusted to a base income in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). Regressions include worker fixed effects, year fixed effects, and age categories. The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 6(a) presents the results as percentages of displaced workers’ base income for the standard specification without further macroeconomic controls. The absolute values of workers’ gross income losses appear in column (1) of Table B.1. Displaced workers who lost their jobs before the wage bargaining reform lost only between 290 DKK ($33) and 2060 DKK ($238) per year. In real terms, these income losses correspond to between 0.09% and 0.5% of their average predisplacement income (dashed line). At conventional significance levels, their income losses are not significantly different from zero. In contrast, displaced workers who lost their jobs after the reform lost between about 1,471 ($170) and 16,549 DKK ($1915) more per year. These differences are highly significant. Measured in relation to their predisplacement income, their income losses correspond to losses of between 6% and 7% of their predisplacement income (solid line). Of course, these estimations only present observable monetary displacement losses. In addition, unemployed workers may incur losses because they forgo their access to fringe benefits such as employer-provided pensions or on-the-job training. Therefore, total displacement losses may even be larger. However, these monetary loss findings are very similar to those for displaced workers’ income losses in other Nordic countries. For example, Huttunen et al. (2011) find income losses of about 5% for displaced workers in Norway. Moreover, as in other European studies, displacement losses do not peak during the displacement year but become significant only in the first year after the displaced workers lost their jobs. Previous studies such as Davis and von Wachter (2011) show that displaced workers’ income losses strongly relate to macroeconomic conditions such as national unemployment levels. Moreover, the decentralization of the Danish wage bargaining system occurred between 1989 and 1993, when the Danish economy experienced a substantial economic downturn. Therefore, Figure 6(b) divides the post-decentralization period after 1989 into two separate periods: one for the recession period between 1989 and 1993, and one for the expansion period between 1994 and 2000. Thus, instead of interacting the displacement dummies with a single dummy for the reform, I further split the reform dummy to represent the post-decentralization downturn period (“1989–1993”) and the post-decentralization recovery period (“1994–2000”) separately. Although income losses were on average slightly larger for workers who lost their jobs during the “1989–1993” Nordic crisis, the differences between the downturn and recovery period are negligible. Overall, the results allow me to draw two main conclusions. First, displaced workers’ income losses increased substantially after the reform of the decentralized wage bargaining system. Second, in general the “1989–1993” Nordic crisis appears not to account for those increases. Section 7 and Appendices A.1 and A.2 provide a large number of additional robustness checks showing that neither other unrelated institutional changes nor macroeconomic conditions are likely to explain the results. Moreover, a separate analysis in Appendix A.1 shows that displaced workers’ earnings losses also increased substantially after the reform. Yet in contrast to displaced workers’ income losses, earnings losses were more sensitive to changes in the business cycle, suggesting that generous UI benefits and large payments of the “guarantee fund” smooth income losses for displaced workers. Finally, I show that displaced workers’ income losses did not increase in other sectors that were affected by similar business cycle conditions but that were not exposed to a change in the wage bargaining system (Section 7.1). 6. Understanding the Relationship between the Decentralization of Wage Bargaining and Displaced Workers’ Income Losses The most common first intuition about the relationship between displaced workers’ income losses and the decentralization of wage bargaining may follow the idea of a very simple neoclassical model on the trade-off between wage losses and unemployment. Such a simple model describes (a) centralized wage bargaining systems by tariff contracts that impose downward rigid wage structures, which prevent displaced workers from obtaining jobs at their optimal reservation wages; and (b) decentralized wage bargaining systems by flexible wage structures that allow all efficient worker-firm matches to occur. This simple model implies that displaced workers on average experience higher unemployment but lower wage losses under centralized wage bargaining systems, and experience higher wage losses but lower unemployment under decentralized ones. However, such a simple model does not necessarily imply that displaced workers’ income losses should increase in response to the decentralization of the wage bargaining system.15 In contrast, the following three arguments suggest that displaced workers may experience larger income losses under decentralized wage bargaining systems. First, studies such as Card et al. (2013) and Dahl et al. (2013) have shown that flexible wages under decentralized wage bargaining resemble individual productivity more closely and induce a greater variability in firm wage premiums. Second, under decentralized wage bargaining, firms have more scope for designing incentive wage plans, such as seniority wage profiles and tournaments. Third, generous UI benefits and a more dispersed wage offer distribution may even provide incentives to prolong the job search under decentralized wage bargaining (see, e.g., Lazear 1986, 2012; Mortensen 1986). The following sections analyze displaced workers’ wage and unemployment patterns more closely to provide evidence for these arguments. 6.1. Hourly Wages and Unemployment Figure 7 decomposes displaced workers’ income losses into wage losses and unemployment. Figure 7. View largeDownload slide Average wage losses and unemployment (pooled sample). Estimation according to equation (1). Dep. vars.: (a) hourly wages and (b) degree of unemployment. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs, excluding zeros. Degree of unemployment measures the fraction of a year a worker received UI benefits. Average income is adjusted to a base income in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 7. View largeDownload slide Average wage losses and unemployment (pooled sample). Estimation according to equation (1). Dep. vars.: (a) hourly wages and (b) degree of unemployment. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs, excluding zeros. Degree of unemployment measures the fraction of a year a worker received UI benefits. Average income is adjusted to a base income in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 7(a) shows hourly wage losses for displaced workers who lost their jobs both before (dashed line) and after (solid lines) the decentralization of the wage bargaining system. I obtained the results by estimating equation (2) with the hourly wage rate as the dependent variable. In contrast to the previous estimations, for which I included zero income, I excluded observations with zero wages to capture the pure price mechanism in terms of displaced workers’ wage losses. For the post-decentralization period, the figure further distinguishes between workers who were displaced during the 1989–1993 recession and workers who were displaced during the recovery period after 1994 (solid line with crosses). Figure 7(a) shows that wage losses were small before the wage bargaining reform; indeed, those wage losses were virtually zero. In contrast, displaced workers who lost their jobs after the reform experienced significant wage losses of between 2% and 4%, including those who lost their jobs during the recovery period after 1994. Comparing the magnitude of these wage losses to the magnitude of income losses, which ranged between 2% and 8% (see Figure 6), suggests that a relatively large fraction of the increase in displaced workers’ income losses is related to an increase in actual wage losses. Figure 7(b) shows comparable results for displaced workers’ degree of unemployment in a given year. The results stem from a regression of equation (2), where the dependent variable measures the degree of unemployment in a given year. Before the reform, the degree of unemployment was on average between two and three percentage points larger for displaced workers than for nondisplaced workers (i.e., approximately between 7 and 10 days of a year). After the reform, the degree of unemployment increased for displaced workers who lost their jobs during the recession, that is, their average degree of unemployment was about three to nine percentage points larger than that of comparable nondisplaced workers. In contrast, the results for displaced workers who lost their jobs during the recovery period after 1994 are very similar to the results for displaced workers who lost their jobs before the reform of the wage bargaining regime. Table 1 analyzes the degree of displaced workers’ unemployment in more detail. The first row presents the fraction of displaced workers whose degree of cumulative unemployment never exceeded 20% of a given year during their first four years after their job loss. The second through fourth rows show the fraction of displaced workers whose degree of cumulative unemployment was at least in one year (2) between 20% and 50%, (3) between 50% and 80%, or (4) larger than 80%. Table 1. Largest degree of unemployment (displaced workers only).     Post-decentralization    Pre-decentralization  Recession  Recovery  Below 20%  0.869  0.739  0.841  Between 20% and 50%  0.074  0.122  0.094  Between 50% and 80%  0.034  0.066  0.041  Larger 80%  0.023  0.074  0.024  Observations  9241          Post-decentralization    Pre-decentralization  Recession  Recovery  Below 20%  0.869  0.739  0.841  Between 20% and 50%  0.074  0.122  0.094  Between 50% and 80%  0.034  0.066  0.041  Larger 80%  0.023  0.074  0.024  Observations  9241      Note: Degree of unemployment measures the fraction of a year a worker received UI benefits. The rows represent fractions of displaced workers whose degree of unemployment never exceeded (1) 20%, was at least once (2) between 20% and 50%, (3) between 50% and 80%, or (4) larger than 80%. Sample is restricted to all displaced workers with valid wage observations before and after their job loss. Displacements occur between November in year (−1) and November in year (0). View Large During the first four years after the job loss, the majority of displaced workers never received UI benefits for more than 20% in any of the four years. However, the statistics clearly show that displaced workers who lost their jobs during the recession between 1989 and 1994 were more likely to experience long-term unemployment than those who lost their jobs either before or after. Overall, the short-term results on unemployment indicate no structural break in unemployment (around the reform of the wage bargaining system) that would be consistent with displaced workers’ income or wage patterns. I find no evidence that overall unemployment consistently increased after the reform or that unemployment decreased, as a simple model on the trade-off between wage losses and unemployment would predict. Before and after the reform, the majority of displaced workers appear to have experienced short periods of unemployment. This result is in line with previous evidence such as Stevens (1997). However, displaced workers who lost their jobs during the recession were much more likely to experience long-term unemployment. If rigid wages under centralized wage bargaining systems would predominantly prevent displaced workers from finding jobs at their optimal reservation wages, the decentralization of the wage bargaining system should be visible in displaced workers’ short-term wage and employment patterns. In contrast, displaced workers may forgo wage losses in the long run if they end up on a different career path in response to the displacement. Therefore, Figures 8 and 9 present long-term analyses for displaced workers’ wage losses and their degree of unemployment, and follow workers for up to 14 years after their job loss. As the data source covers a finite period until 2003, I cannot observe displaced workers from each displacement cohort for the same amount of time after the job loss. Therefore, the coefficient estimates for the later postdisplacement periods are estimated from a different population than the coefficient estimates for the earlier postdisplacement periods. Thus for displaced workers who lost their jobs during the recovery period after 1994, the long-term results must be interpreted with some caution. Figure 8. View largeDownload slide Long-term wage losses (pooled sample). Estimation according to equation (1). Dep. vars.: hourly wages. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs, excluding zeros. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 8. View largeDownload slide Long-term wage losses (pooled sample). Estimation according to equation (1). Dep. vars.: hourly wages. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs, excluding zeros. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 9. View largeDownload slide Long-term unemployment (pooled sample). Estimation according to equation (1). Dep. vars.: degree of unemployment. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs, excluding zeros. Displacements happened between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 9. View largeDownload slide Long-term unemployment (pooled sample). Estimation according to equation (1). Dep. vars.: degree of unemployment. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs, excluding zeros. Displacements happened between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. The solid line in Figure 8 shows that displaced workers who lost their jobs before the reform did not experience wage losses under the direct influence of the rigid wage bargaining system but that they experienced small long-term wage losses in the long run (about 10 years after the job loss). However, displaced workers who lost their jobs after the introduction of the decentralized wage bargaining system experienced much larger short- and long-term wage losses—including those displaced workers who lost their jobs during the recovery period after 1994 (solid line with crosses in Figure 8(b)). Therefore, displaced workers who lost their jobs after the introduction of the decentralized wage bargaining system forwent wage growth in the short and the long run. In contrast, the results suggest that workers who were displaced before 1989 appear to have had the chance of stabilizing their career path before entering the post-decentralization period. Thus even if they also forwent wage growth after the implementation of the reform, their total wage losses were somewhat smaller. Figure 9 shows comparable long-term results for displaced workers’ degree of unemployment. The long-term unemployment pattern of those displaced workers who lost their jobs before the reform follows an inverted u-shape. More specifically, they have experienced the largest degree of unemployment between the 7th and 11th year after their job loss, that is, during the recession after the 1989 reform. Displaced workers who lost their jobs during the recession experienced even higher levels of unemployment—but only in the short run. In the later years after their job loss (when this group entered the period of economic recovery), their degree of unemployment decreased (solid line). Finally, although displaced workers who lost their jobs during the recovery period after 1994 experienced somewhat larger unemployment in the short run, their degree of unemployment decreased substantially thereafter (solid line with crosses). Overall, the results show that the degree of displaced workers’ unemployment is closely related to the general economic development in Denmark. However, unlike the results for wages and income, the results for unemployment do not show a clear structural break around the reform in 1989. Thus the results suggest that the sudden increase in displaced workers’ income losses is largely related to forgone wage growth, whereas generous UI benefits may have compensated displaced workers for their earnings losses from unemployment (see also Appendix A.1). 6.2. Forgone Wage Growth The movement away from the centralized wage bargaining system was also a movement away from rigid wage scales that determined workers’ wages based on observable characteristics such as age, education, or experience in the trade. Moreover, before the reform, unions with a preference for social equality were highly involved in the wage bargaining, so that low-productive workers may have received relatively higher wages (Freeman 1980; Card et al. 2004). As a result, even displaced workers who lose firm-specific human capital were more likely to have received wages similar to those of nondisplaced workers. In contrast, under decentralized wage bargaining, wages are more in accordance with individual productivity and local conditions. Moreover, if wage bargaining is decentralized, employers may have more scope for designing incentive wage plans, such as seniority wage profiles and tournaments, and steep wage profiles may also reflect employers’ learning about unobservable worker characteristics. Thus, after the reform, firm-specific human capital investments and careers in internal labor markets may have become more important in determining individual wage growth. In contrast, involuntary job losses and even short periods of unemployment may have resulted in large forgone wage growth. Table 2 analyzes in more detail the career-related wage growth of displaced workers after their job loss. The table presents simple regressions in the following form:   \begin{equation} \mathit{ lnw}_{it}^{d}=\alpha _{it}^{d}+ \delta _{1} f(\mathit{ time})_{it}^{d}+\delta _{2} {{\mathit{ Tenure}}}_{it}^{d}+\varepsilon _{it}^{d}, \end{equation} (3)where $$\mathit{ lnw}_{it}^{d}$$ is the displaced workers’ log hourly wages. $$f(\mathit{ time})_{it}^{d}$$ is a simple linear trend of the elapsed time since the displacement, and $${\mathit{ Tenure}}_{it}^{d}$$ measures their postdisplacement tenure. More specifically, $${{\mathit{ Tenure}}}_{it}^{d}$$ measures the cumulative time a displaced worker has worked for his current postdisplacement firm, that is, $${\mathit{ Tenure}}_{it}^{d}$$ does not measure postdisplacement experience. $$\alpha _{it}^{d}$$ is a worker fixed effect. The ds indicate that, to directly analyze displaced workers’ pre–post-displacement wage differences, this analysis contains only displaced workers. The first column of Table 2 shows the results for displaced workers who lost their jobs before the reform; the second column, for displaced workers who lost their jobs afterwards. Table 2. Returns to postdisplacement time and tenure (displaced workers only).   Pre-decentralization  Post-decentralization  Time since displacement ($$f(\mathit{ time})_{it}^{d}$$)  0.027***  −0.003    (0.002)  (0.003)  Tenure since displacement (Tenure$$_{it}^{d}$$)  0.002  0.012***    (0.002)  (0.003)  Observations:  14428  18921    Pre-decentralization  Post-decentralization  Time since displacement ($$f(\mathit{ time})_{it}^{d}$$)  0.027***  −0.003    (0.002)  (0.003)  Tenure since displacement (Tenure$$_{it}^{d}$$)  0.002  0.012***    (0.002)  (0.003)  Observations:  14428  18921  Notes: Author’s calculations with data from Statistics Denmark. Dep. var.: Log hourly wages. $$f(\mathit{ time})_{it}^{d}$$ is a linear time-trend measuring time since displacement. Hourly wages are measured in 2000 DKKs excluding zeros. Tenure$$_{it}^{d}$$ measures tenure since displacement. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Sample is restricted to all observations of displaced workers with valid wage observations. Column (1) shows the results for displaced workers who lost their jobs before the wage bargaining reform. Column (2) shows the results for displaced workers who lost their jobs after the reform. Displacements occur between November in year (−1) and November in year (0). Regressions include worker fixed effects, and age categories. Standard errors are in parentheses and clustered at the individual level. ***Significant at 1% level. View Large If displaced workers who lost their jobs before the wage bargaining reform benefited mainly from wage drifts of general contracts and centralized negotiations, we should expect a strong positive effect of the general time trend ($$f(\mathit{ time})_{it}^{d}$$), whereas the coefficient estimate of tenure should be small ($${\mathit{ Tenure}}_{it}^{d}$$). In contrast, if displaced workers who lost their jobs after the reform forwent wage growth, because they lost firm-specific human capital and career-related rents, tenure should become more important than the general time trend. The results of Table 3 indeed confirm this hypothesis. The first column shows that the coefficient estimate of the general time trend (δ1) is about three percentage points and highly significant, whereas the coefficient estimate on tenure (δ2) is only 0.3 percentage points. The second column shows a coefficient estimate of the general time trend that is essentially zero and insignificant, whereas the coefficient estimate of tenure is about 1 percentage point and highly significant. Table 3. Average pre–post-displacement wage differences (displaced workers only).   Pre-decentralization  Post-decentralization  All  0.029  −0.003  Low-wage workers  0.071  0.063  High-wage workers  −0.014  −0.070  Observations  3884  5189    Pre-decentralization  Post-decentralization  All  0.029  −0.003  Low-wage workers  0.071  0.063  High-wage workers  −0.014  −0.070  Observations  3884  5189  Notes: Author’s calculations with data from Statistics Denmark. Dep. var.: Log hourly wages. Hourly wages are measured in 2000 DKKs, excluding zeros. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Sample is restricted to all observations of displaced workers with valid wage observations. Column (1) shows the results for displaced workers who lost their jobs before the wage bargaining reform. Column (2) shows the results for displaced workers who lost their jobs after the reform. Displacements occur between November in year (−1) and November in year (0). View Large Even after the reform, returns to tenure were relatively small, a result consistent with the findings of the previous literature (e.g., Eriksson and Westergaard-Nielsen 2009). However, the results suggest that forgone returns to tenure may not be the only explanation for the increase of wage losses after the reform. Another potential explanation is that voluntary job shopping became more important in response to the more dispersed wage distribution after the reform. The next section analyzes this argument in more detail. 6.3. Wage Dispersion As previously mentioned, a number of studies find that the dispersion of wages increased after the 1989 wage bargaining reform in Denmark. Therefore, voluntary job shopping may have become a more important determinant for workers’ wage development. Thus if job displacements force workers to leave their path of lucrative voluntary job shopping, they may forgo more wage growth under decentralized wage bargaining systems. In addition, displaced workers may have incentives for prolonging their job search, because their option value of obtaining very high wages in the future may become relatively larger than their costs of unemployment. As a result, displaced workers may reject early job offers with low wages, because they now expect to find a job with higher wages in the future. This result is central to matching models such as Mortensen (1986), and pricing models such as Lazear (1986, 2012), all of which state that a more dispersed wage-offer distribution leads to higher search intensity and longer periods of unemployment. Providing sharp empirical tests for these ideas is difficult. However, analyzing the differences between displaced workers’ last hourly wages before the job loss and their first hourly wages after the job loss (hereafter, “pre–post-displacement wage differences”) can provide useful information.16 If displaced workers indeed respond to a more dispersed wage-offer distribution, realized pre–post-displacement wage differences should also become more dispersed after the reform. Moreover, displaced workers should be relatively unlikely to accept postdisplacement wages that lie substantially below the level of wages in their previous job. In contrast, a model in which centralized wages are simply a market friction that prevent displaced workers from finding jobs at their optimal reservation wages would predict a distribution shift to the left, simply because more displaced workers reenter employment with low wages. Table 3 presents simple descriptive statistics for pre–post-displacement wage differences. The first row of Table 3 presents average pre–post-displacement wage differences for all workers. Before the reform, average pre–post-displacement wage differences were about 3 percentage points. After the reform, pre–post-displacement wage differences were negative but close to zero. Thus, on average, displaced workers did not accept postdisplacement wages that were substantially below the level of their predisplacement wages. The second and third rows present average pre–post-displacement wage differences separately for displaced workers with low and high predisplacement wages, that is, a worker is considered as having a high predisplacement wage if it lies above the median of all predisplacement wages in that given year, and vice versa. The second row shows that displaced workers with low predisplacement wages reentered employment at wages that were on average always above the level of their predisplacement wages. In contrast, displaced workers with high predisplacement wages always reentered employment at wages below the level of their predisplacement wages. Moreover, their average pre–post-displacement wage differences changed from −0.014 to −0.070, suggesting that high-wage workers reentered employment after the reform at substantially lower wages. Figure 10 presents more detailed results by showing simple kernel density estimates for the distribution of log differences between pre- and postdisplacement wages, both before (solid line) and after (dashed line) the wage bargaining reform. Figure 10(a) presents the results for workers with low predisplacement wages, and Figure 10(b) presents the results for workers with high predisplacement wages. Both before and after the reform, pre–post-displacement wage differences were strongly centered at zero for both types of workers, suggesting that the median displaced worker did not accept postdisplacement wages that were substantially below the level of his predisplacement wages. Moreover, pre–post-displacement wage differences became more diverse for both types of workers. Interpreting these results in the light of the previous arguments suggests that displaced workers indeed responded to the increased dispersion of the wage-offer distribution. Thus the results are in line with the idea that job shopping became more important for them after the reform. Figure 10. View largeDownload slide Distribution of pre–post-displacement wage differences (pooled sample). Kernel density estimates Dep. vars.: Difference between log pre- and postdisplacement wages. (a) Low-wage worker and (b) high-wage worker. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. High-wage workers are workers who had a predisplacement wage that was larger than the median of all workers’ wages in the predisplacement year. Figure 10. View largeDownload slide Distribution of pre–post-displacement wage differences (pooled sample). Kernel density estimates Dep. vars.: Difference between log pre- and postdisplacement wages. (a) Low-wage worker and (b) high-wage worker. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. High-wage workers are workers who had a predisplacement wage that was larger than the median of all workers’ wages in the predisplacement year. Nonetheless, the distribution of displaced workers with high predisplacement wages exhibits a stronger shift to the left in response to the reform than the postreform distribution of displaced workers with low predisplacement wages. A potential reason is that low-income workers’ UI benefits are almost as large as their previous income, such that they have weak incentives to accept wages that are close to or below their predisplacement wages. In contrast, high-wage workers always received relatively lower UI benefits, such that they may have become more likely to trade off wage losses against unemployment after the reform. The next section analyzes this argument in more detail. 6.4. Unemployment Benefits and Reservation Wages If UI benefits are sufficiently generous, displaced workers’ outside options may offset the expected trade-off between wage losses and unemployment. Thus, because displaced workers remain unemployed until they find jobs that offer wages above their UI benefits, the decentralization of wage bargaining will not reduce unemployment very much. As previously mentioned, UI benefits constitute up to 90% of a worker’s previous wage, with a relatively low cap of approximately 170,000 DKKs for high-income workers. Thus, in particular, displaced workers with low predisplacement wages should always be unwilling to reenter employment for entry wages that are substantially below the wage level of their last employment before the job loss. In contrast, displaced workers with high predisplacement wages receive comparably less generous UI benefits and should, therefore, become more likely to accept wages below the level of their predisplacement wage and less likely to experience unemployment after the reform. Table 4 presents an analysis according to the following regression equation:   \begin{eqnarray} \Delta \text{log}w_{i,\mathit{ Pre-post}}^{d}&=&\alpha ^{d}+\lambda _{t}^{d}+x_{it}^{d}\beta + \gamma _{1} \mathit{ HW}_{it}^{d}\nonumber\\ &&+\,\,\delta \mathit{ HW}_{it}^{d}\cdot \mathit{ Reform}+\gamma _{2} \mathit{ Reform}+\varepsilon _{it}^{d}, \end{eqnarray} (4)where $$\Delta \text{log}w_{i,\mathit{ Pre-post}}^{d}$$ measures pre–post-displacement wage differences. Reform is a dummy for the wage bargaining reform, and $$\mathit{ HW}_{it}^{d}$$ indicates whether a displaced worker had a low or high predisplacement wage. Again, the d indicates that nondisplaced workers are excluded. Furthermore, the regressions include year dummies λt, a constant α,17 and age categories. δ is the coefficient of main interest and measures whether pre–post-displacement wage differences increased for high-wage workers after the wage bargaining reform. Thus if displaced workers with high predisplacement wages have reentered employment after the reform with relatively lower postdisplacement wages, the coefficient estimate of δ should be negative. Table 4. Trade-off between wages and unemployment (displaced workers only).     Degree of unemployment  Dep. vars.  Pre- and postdisp. wage diff.  Disp. year  1>Disp.  2>Disp.  3> Disp.  Post-dec.  0.025*  0.021**  −0.014  0.006  0.024**  ($$\mathit{ Reform}$$)  (0.013)  (0.007)  (0.010)  (0.009)  (0.009)  High wage  −0.064***  −0.008**  −0.017***  −0.017***  −0.015***  ($$\mathit{ HW}_{it}^{d}$$)  (0.006)  (0.003)  (0.004)  (0.004)  (0.004)  $$\mathit{ HW}_{it}^{d}\cdot \mathit{ Reform}$$  −0.053**  −0.020***  −0.008  −0.006  −0.005    (0.009)  (0.004)  (0.005)  (0.006)  (0.006)  Observations  9073  9073  9029  8988  8969      Degree of unemployment  Dep. vars.  Pre- and postdisp. wage diff.  Disp. year  1>Disp.  2>Disp.  3> Disp.  Post-dec.  0.025*  0.021**  −0.014  0.006  0.024**  ($$\mathit{ Reform}$$)  (0.013)  (0.007)  (0.010)  (0.009)  (0.009)  High wage  −0.064***  −0.008**  −0.017***  −0.017***  −0.015***  ($$\mathit{ HW}_{it}^{d}$$)  (0.006)  (0.003)  (0.004)  (0.004)  (0.004)  $$\mathit{ HW}_{it}^{d}\cdot \mathit{ Reform}$$  −0.053**  −0.020***  −0.008  −0.006  −0.005    (0.009)  (0.004)  (0.005)  (0.006)  (0.006)  Observations  9073  9073  9029  8988  8969  Notes: Author’s calculations with data from Statistics Denmark. Dep. var.: Column (1): Log pre- and postdisplacement wage differences. Columns (2)–(5): Degree of unemployment in respective year. Degree of unemployment measures the fraction of a year a worker received UI benefits. $$\mathit{ HW}_{it}^{d}$$ is a dummy variable indicating whether a displaced worker’s predisplacement wage was larger than the median of all workers’ wages in the predisplacement year. Reform is a dummy indicating whether the worker was displaced before or after the wage bargaining reform. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Sample is restricted to all displaced workers with valid wage observations before and after their job loss. Displacements occur between November in year (−1) and November in year (0). Regressions include year fixed effects and age categories. Standard errors are in parentheses and clustered at the individual level. *Significant at 10% level; **significant at 5% level; ***significant at 1% level. View Large The first column of Table 4 shows the results for pre- and postdisplacement wage differences. Columns (2)–(4) show results according to the same regression equation (4) but with a dependent variable measuring the degree of unemployment in the displacement year, as well as the first, second, and third years after displacement. As expected, the coefficient estimate of δ is negative and significant. Moreover, high wage workers also experienced relatively less unemployment after the reform (columns (2) through (5)). However, the unemployment effect is large and significant only for the displacement year and becomes much weaker and insignificant thereafter. These results are in line with the idea that generous UI benefits prevent displaced workers with lower wages from trading off wage losses against unemployment. 7. Placebo Tests and Alternative Explanations This section provides additional evidence improving the credibility of the main results. It also discusses a number of alternative scenarios that may explain why displaced workers experience larger income losses after the reform. 7.1. Sensitivity Analysis An ideal experiment for evaluating how the decentralization of wage bargaining systems influences the magnitude of displaced workers’ income losses would require a random assignment of centralized and decentralized wage bargaining systems to comparable sectors. However, changes in wage bargaining systems are commonly the result of complex political processes, and, unfortunately, policy makers do not tend to implement natural experiments. Therefore, no sectors unaffected by the wage bargaining reform are directly comparable to the manufacturing sector. However, this section provides two sensitivity checks to evaluate the credibility of the former results. The first sensitivity check analyzes displacement losses for a subset of low-skilled manufacturing workers likely to remain on the standard wage rate system—even after the reform. The second sensitivity check analyzes displacement losses in the Danish finance sector, a sector not exposed to a comparably drastic change of the wage bargaining system in the early 1990s. Many low-skilled workers are members of the General Workers Union, SID (Specialarbejderforbundet i Danmark), which strongly opposed the decentralization process in the early 1990s (Iversen 1996). Although the SID did not succeed in preventing the decentralization process, many low-skilled workers remained under the standard wage rate system after 1989. For example, Dahl et al. (2013) report that 59% of all workers covered by the standard wage rate system between 1992 and 2002 were low-skilled workers without an apprenticeship degree. Thus we should expect to find a weaker influence of the decentralization reform on the displacement losses of SID workers—if they remained in their predisplacement industry. Figure 11 shows estimates for a subset of all SID workers in my sample. To identify them, I rely on Statistics Denmark data that identifies the UI fund (Akasse) of each worker in a given year. Unfortunately, as the data are available only from 1985, I made the assumption that workers who had been SID members from 1985 had also been SID members between 1980 and 1984. Although making such an assumption may produce biased results, Danish workers seldom change their UI fund, so as to avoid losing their eligibility for benefits. For the purpose of this investigation, I restricted the sample to workers who did not change their manufacturing industry. Figure 11. View largeDownload slide Placebo analysis: SID workers. Estimation according to equation (1). Dep. vars.: (a) gross income, (b) hourly wages, and (c) degree of unemployment. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Gross income and hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs. Gross income includes zeros. Hourly wages exclude zeros. Degree of unemployment measures the fraction of a year a worker received UI benefits. Average income and wages are adjusted to base income and wages in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 11. View largeDownload slide Placebo analysis: SID workers. Estimation according to equation (1). Dep. vars.: (a) gross income, (b) hourly wages, and (c) degree of unemployment. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Gross income and hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs. Gross income includes zeros. Hourly wages exclude zeros. Degree of unemployment measures the fraction of a year a worker received UI benefits. Average income and wages are adjusted to base income and wages in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. The figure shows the results for income losses, wage losses, and unemployment. The first subfigure shows that SID workers experienced somewhat larger income losses than did the full sample before the reform. Displaced workers’ income losses increased slightly after the reform. However, the differences are not significant, and income losses did not increase as much as for the full sample. The second subfigure shows that SID workers experience almost no wage losses, either before or after the introduction of the reform. As the third subfigure suggests, income losses appear related to unemployment. If general macroeconomic conditions or unrelated institutional influences were to account for the increase in displaced workers’ income losses in the manufacturing sector, we should expect to also observe a similar pattern for displaced workers’ income losses in other sectors. Therefore, analyzing displaced workers’ income losses in other sectors can serve as another placebo test. In Denmark, the finance and banking sector is an ideal candidate for such a placebo analysis. Historically, the Danish finance sector was dominated by standard wage rate contracts negotiated between the financial service unions (Finansforbundet) and the employer associations of the financial sector (Finanssektorens Arbejdsgiverforening). The largest banks dominated these negotiations, and most smaller institutions adopted the agreements. However, wage bargaining in the banking sector is characterized by numerous subagreements for bank clerks and other employees, such as IT specialists and executive staff (e.g., Mayer, Andersen, and Muller 2001). Unlike in most other Danish sectors, the wage bargaining system in the finance sector was not changed in any comparable way during the 1990s. Only in 1997 did the social partners agree to implement a system that permitted greater use of performance-related pay at the company level. However, the system became effective only at the end of 1999, because managers in the banking sector were initially reluctant to implement the new pay system. Figure 12 presents the results for the finance sector. Before and after 1989, displaced workers’ income losses range between about 2% and 5%, and no structural break is visible. The same holds for wage losses and unemployment. Figure 12. View largeDownload slide Placebo analysis: Finance sector. Estimation according to equation (1). Dep. vars.: (a) gross income; (b) hourly wages; (c) degree of unemployment. Displaced workers were aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Gross income and hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs. Gross income includes zeros. Hourly wages exclude zeros. Degree of unemployment measures the fraction of a year a worker received UI benefits. Average income and wages are adjusted to base income and wages in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 12. View largeDownload slide Placebo analysis: Finance sector. Estimation according to equation (1). Dep. vars.: (a) gross income; (b) hourly wages; (c) degree of unemployment. Displaced workers were aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Gross income and hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs. Gross income includes zeros. Hourly wages exclude zeros. Degree of unemployment measures the fraction of a year a worker received UI benefits. Average income and wages are adjusted to base income and wages in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. If general macroeconomic conditions and unrelated institutional changes were to account for the increase in displaced workers’ income losses in the manufacturing sector, I should observe a similar pattern of displaced workers’ income losses in the finance sector. The results for the finance sector clearly contradict this expectation. Both of these sensitivity checks show that influences other than the decentralization of the wage bargaining system are unlikely to explain the increase in displaced workers’ income losses after 1989. 7.2. Alternative Policy Changes: UI Benefits In response to the continuously increasing unemployment at the end of the 1980s, Danish policy makers changed labor market institutions other than the wage bargaining system. One major reform changed the laws regulating UI benefits and active labor market policies. Before 1994, the eligibility period for UI benefits was effectively infinite. Despite a formal duration of seven years, workers could easily obtain a new benefit period by participating in a job activation program (Andersen 2003). In 1994 policy makers implemented a new system with two benefit periods. The first was a passive period of four years, to allow workers to focus on their job search without further requirements. The second was an activation period of three years, during which workers had to participate in active labor market programs in return for UI benefits. Policy makers further reduced the passive period to three years in 1996 and to two years in 1998. Previous evidence showed that unemployed workers indeed responded to the UI system reform by becoming more likely to reenter employment shortly before the compulsory participation in active labor market programs was to begin (Geerdsen 2006). Thus the question arises as to whether the change in the UI system can explain the increase of displaced workers’ income losses after 1989. For example, the newly implemented activation period may have led displaced workers to reenter employment faster but at lower wages. However, for the following four reasons, the reform of the UI system is unlikely to account for the increase in displaced workers’ income losses after 1989. First, although the UI reform was implemented only in 1994, displaced workers’ income losses had begun increasing earlier. Second, previous evidence suggests that most unemployed workers responded to the reform just about three months before the beginning of their activation period. But even after the UI reform, most displaced workers in my sample had the opportunity to benefit from a passive period of three or four years. Therefore, the change in the UI system cannot account for the effects I find within the first three or four years after the job loss. Third, most displaced workers appear to have reentered employment shortly after their job loss, with relatively few experiencing long-term unemployment. Therefore, the sample that I analyze in this paper does not include many long-term unemployed workers, who, according to Geerdsen (2006), are those who responded more strongly to the reform of the UI system. Fourth, the results in Section 7 showed that most displaced workers reentered employment at wages comparable to their predisplacement wages. This result is at odds with the idea that displaced workers reduced their unemployment at the cost of lower postdisplacement wages. 7.3. Alternative Policy Changes: Parental Leave In response to the economic crisis of the late 1980s, Danish policy makers implemented yet another labor market reform (Rostgaard and Christoffersen 1999). In 1992, they introduced a new parental leave provision that was available to all parents with children younger than eight. The aim was to lower unemployment by reducing the labor supply and by creating new opportunities for temporary employment. Although this reform initially gave parents benefits allowing them to leave work for 36 weeks, in 1993 policy makers extended this period to one year. However, as unemployment began to fall in 1994, policy makers reduced the child-care leave period to 13 weeks.18 Therefore, the reform of the child-care leave provision is unlikely to be the cause of the persistent increase in displaced workers’ income losses after 1989. Moreover, very few fathers took child-care leave. For example, in 1996 only 4% of fathers took advantage of any parental leave plan. As a result, a substantial influence of the child-care leave reform on the male-dominated labor market for manufacturing workers is unlikely. 8. Conclusion Many European countries have decentralized their wage bargaining systems over the past decades to foster economic growth and reduce unemployment. By showing the relationship between the decentralization of wage bargaining and displaced workers’ income losses, this paper provides important insights into the consequences of decentralized wage bargaining for the specific group involuntary job losers. More specifically, this paper uses administrative register data to analyze the relationship between the decentralization of wage bargaining systems and displaced workers’ income losses. It exploits a major reform of the Danish wage bargaining system, a reform that changed wage setting in the manufacturing sector from a highly centralized national system to a decentralized one with a substantial emphasis on firm-level wage bargaining. The results show small income losses under the centralized wage bargaining system, which was dominated by standard wage rate contracts that remunerated workers according to their education, occupation, and sector experience. After the decentralization of wage bargaining, a large percentage of workers had to negotiate their wages at the firm level, and displaced workers’ income losses increased substantially. Displaced workers who lost their jobs before the reform appear to have benefited by the wage drifts of general contracts and centralized negotiations. In contrast, displaced workers who lost their jobs after the reform appear to have forgone wage growth, because they failed to collect rents from tenure or voluntary job shopping. Moreover, both before and after the reform, displaced workers did not reenter employment for wages substantially below the level of their predisplacement wages, likely because of generous UI benefits. Appendix A: Additional Tables Table A.1. Baseline characteristics (pre-decentralization). Sample  1980–1985  1981–1986  1982–1987  Disp. year  1982  1983  1984  Gross income  292506  310624.2**  281861  286425.3  281504.2  270647.9*  Age  35.29  35.05  35.42  35.93*  35.62  34.83*  Experience  11.98  11.85  12.46  12.16*  12.98  11.48***  Education              Low  0.358  0.356  0.356  0.353  0.361  0.495***  Medium  0.527  0.512  0.527  0.535  0.528  0.360***  High  0.099  0.122*  0.1  0.092  0.096  0.123  Mfr. sector              Food and bev.  0.242  0.349***  0.242  0.117***  0.253  0.271  Metal products  0.46  0.306***  0.461  0.257***  0.449  0.457  Wood products  0.116  0***  0.113  0.245***  0.112  0.019***  Other mfr. indust  0.182  0.345***  0.184  0.382***  0.186  0.252**  Individuals  441  56870  781  59344  317  58421  Sample  1983–1988  1984–1989  1985–1990  Disp. year  1985  1986  1987  Gross income  285521.1  270212.3**  284364.2  277677.6  288474.6  259933.4***  Age  35.92  34.62***  36.06  34.77**  36.01  33.03***  Experience  13.53  12.28***  13.9  12.88**  14.08  11.64***  Education              Low  0.356  0.401**  0.353  0.357  0.343  0.390**  Medium  0.531  0.53  0.528  0.5  0.536  0.495**  High  0.098  0.046***  0.103  0.126  0.107  0.096  Mfr. sector              Food and bev.  0.251  0.585***  0.238  0.343***  0.224  0.188**  Metal products  0.456  0.290***  0.449  0.135***  0.452  0.590***  Wood products  0.107  0.0138***  0.109  0.287***  0.118  0.047***  Other mfr. indust.  0.186  0.111***  0.205  0.235  0.206  0.175**  Individuals  434  55005  230  52850  685  50350  Sample  1986–1991      Disp. year  1988      Gross income  291877.9  281774.6**          Age  35.83  34.95***          Experience  14.18  13.39***          Education              Low  0.338  0.390***          Medium  0.538  0.524          High  0.111  0.068***          Mfr. sector              Food and bev.  0.215  0.618***          Metal products  0.455  0.338***          Wood products  0.114  0.007***          Other mfr. indust.  0.215  0.037***          Individuals  1083  52699          Sample  1980–1985  1981–1986  1982–1987  Disp. year  1982  1983  1984  Gross income  292506  310624.2**  281861  286425.3  281504.2  270647.9*  Age  35.29  35.05  35.42  35.93*  35.62  34.83*  Experience  11.98  11.85  12.46  12.16*  12.98  11.48***  Education              Low  0.358  0.356  0.356  0.353  0.361  0.495***  Medium  0.527  0.512  0.527  0.535  0.528  0.360***  High  0.099  0.122*  0.1  0.092  0.096  0.123  Mfr. sector              Food and bev.  0.242  0.349***  0.242  0.117***  0.253  0.271  Metal products  0.46  0.306***  0.461  0.257***  0.449  0.457  Wood products  0.116  0***  0.113  0.245***  0.112  0.019***  Other mfr. indust  0.182  0.345***  0.184  0.382***  0.186  0.252**  Individuals  441  56870  781  59344  317  58421  Sample  1983–1988  1984–1989  1985–1990  Disp. year  1985  1986  1987  Gross income  285521.1  270212.3**  284364.2  277677.6  288474.6  259933.4***  Age  35.92  34.62***  36.06  34.77**  36.01  33.03***  Experience  13.53  12.28***  13.9  12.88**  14.08  11.64***  Education              Low  0.356  0.401**  0.353  0.357  0.343  0.390**  Medium  0.531  0.53  0.528  0.5  0.536  0.495**  High  0.098  0.046***  0.103  0.126  0.107  0.096  Mfr. sector              Food and bev.  0.251  0.585***  0.238  0.343***  0.224  0.188**  Metal products  0.456  0.290***  0.449  0.135***  0.452  0.590***  Wood products  0.107  0.0138***  0.109  0.287***  0.118  0.047***  Other mfr. indust.  0.186  0.111***  0.205  0.235  0.206  0.175**  Individuals  434  55005  230  52850  685  50350  Sample  1986–1991      Disp. year  1988      Gross income  291877.9  281774.6**          Age  35.83  34.95***          Experience  14.18  13.39***          Education              Low  0.338  0.390***          Medium  0.538  0.524          High  0.111  0.068***          Mfr. sector              Food and bev.  0.215  0.618***          Metal products  0.455  0.338***          Wood products  0.114  0.007***          Other mfr. indust.  0.215  0.037***          Individuals  1083  52699          Notes: Author’s calculations with data provided by Statistics Denmark. The table shows descriptive statistics for nondisplaced manufacturing workers two years before the displacement period of the respective sample. Labor earnings, wages, and gross income are measured in 2000 DKKs. Standard errors are in parentheses and clustered at the individual level. *Significant at 10% level; **significant at 5% level; ***significant at 1% level. View Large Table A.2. Baseline characteristics (post-decentralization). Sample  1987–1992  1988–1993  1989–1994  Disp. year  1989  1990  1991  Gross income  297192.3  293308.8  297500  281197.5**  302512.1  286267.8**  Age  35.66  34.51**  35.58  33.70***  35.65  34.39**  Experience  14.81  14.43  14.94  13.84***  15.22  15.14  Education              Low  0.311  0.306  0.308  0.417***  0.306  0.429***  Medium  0.559  0.545  0.563  0.479***  0.567  0.526  High  0.116  0.132  0.115  0.089*  0.114  0.036***  Mfr. sector              Food and bev.  0.175  0.306***  0.195  0.387***  0.202  0.309***  Metal products  0.473  0.524**  0.457  0.464  0.454  0.538**  Wood products  0.121  0.039***  0.116  0.015***  0.108  0.009***  Other mfr.  0.231  0.130***  0.232  0.134***  0.236  0.144***  Individuals  389  55281  501  57657  440  58053  Sample  1990–1995  1991–1996  1992–1997  Disp. year  1992  1993  1994  Gross income  297192.3  293308.8  297500  281197.5**  302512.1  286267.8**  Age  35.66  34.51**  35.58  33.70***  35.65  34.39**  Experience  14.81  14.43  14.94  13.84***  15.22  15.14  Education              Low  0.311  0.306  0.308  0.417***  0.306  0.429***  Medium  0.559  0.545  0.563  0.479***  0.567  0.526  High  0.116  0.132  0.115  0.089*  0.114  0.036***  Mfr. sector              Food and bev.  0.175  0.306***  0.195  0.387***  0.202  0.309***  Metal products  0.473  0.524**  0.457  0.464  0.454  0.538**  Wood products  0.121  0.039***  0.116  0.015***  0.108  0.009***  Other mfr. indust.  0.231  0.130***  0.232  0.134***  0.236  0.144***  Individuals  431  58174  470  59843  333  60136  Sample  1993–1998  1994–1999  1995–2000  Disp. year  1995  1996  1997  Gross income  299095.1  291924.5  303265  272220.5***  303552.8  317956.8**  Age  35.78  33.57***  35.81  34.37***  35.66  36.01  Experience  15.57  14.58***  15.84  15.6  15.95  16.45  Education              Low  0.306  0.448***  0.301  0.421***  0.3  0.246*  Medium  0.571  0.510**  0.574  0.521**  0.575  0.59  High  0.111  0.036***  0.113  0.0487***  0.113  0.149*  Mfr. sector              Food and bev.  0.199  0.568***  0.202  0.338***  0.199  0.343***  Metal products  0.463  0.336***  0.461  0.259***  0.465  0.295***  Wood products  0.109  0.03***  0.109  0.005***  0.108  0.134  Other mfr. indust.  0.229  0.066***  0.228  0.397***  0.229  0.228  Individuals  563  57899  390  58844  268  61225  Sample  1996–2001  1997–2002  1998–2003  Disp. year  1998  1999  2000  Gross income  303011.2  294882.2  306357.4  304647  312491.6  299407.7**  Age  35.68  35.69  35.82  35.28**  35.94  35.61  Experience  16.2  16.16  16.55  16.86  16.87  17.1  Education              Low  0.3  0.293  0.295  0.356***  0.29  0.325*  Medium  0.573  0.603  0.575  0.567  0.579  0.566  High  0.114  0.095  0.117  0.07***  0.118  0.089**  Mfr. sector              Food and bev.  0.198  0.451***  0.183  0.278***  0.182  0.486***  Metal products  0.456  0.158***  0.454  0.481  0.46  0.374***  Wood products  0.113  0.230***  0.111  0.041***  0.114  0.033***  Other mfr. indust.  0.233  0.161**  0.251  0.199**  0.244  0.107***  Individuals  348  61135  702  60084  551  58456  Sample  1987–1992  1988–1993  1989–1994  Disp. year  1989  1990  1991  Gross income  297192.3  293308.8  297500  281197.5**  302512.1  286267.8**  Age  35.66  34.51**  35.58  33.70***  35.65  34.39**  Experience  14.81  14.43  14.94  13.84***  15.22  15.14  Education              Low  0.311  0.306  0.308  0.417***  0.306  0.429***  Medium  0.559  0.545  0.563  0.479***  0.567  0.526  High  0.116  0.132  0.115  0.089*  0.114  0.036***  Mfr. sector              Food and bev.  0.175  0.306***  0.195  0.387***  0.202  0.309***  Metal products  0.473  0.524**  0.457  0.464  0.454  0.538**  Wood products  0.121  0.039***  0.116  0.015***  0.108  0.009***  Other mfr.  0.231  0.130***  0.232  0.134***  0.236  0.144***  Individuals  389  55281  501  57657  440  58053  Sample  1990–1995  1991–1996  1992–1997  Disp. year  1992  1993  1994  Gross income  297192.3  293308.8  297500  281197.5**  302512.1  286267.8**  Age  35.66  34.51**  35.58  33.70***  35.65  34.39**  Experience  14.81  14.43  14.94  13.84***  15.22  15.14  Education              Low  0.311  0.306  0.308  0.417***  0.306  0.429***  Medium  0.559  0.545  0.563  0.479***  0.567  0.526  High  0.116  0.132  0.115  0.089*  0.114  0.036***  Mfr. sector              Food and bev.  0.175  0.306***  0.195  0.387***  0.202  0.309***  Metal products  0.473  0.524**  0.457  0.464  0.454  0.538**  Wood products  0.121  0.039***  0.116  0.015***  0.108  0.009***  Other mfr. indust.  0.231  0.130***  0.232  0.134***  0.236  0.144***  Individuals  431  58174  470  59843  333  60136  Sample  1993–1998  1994–1999  1995–2000  Disp. year  1995  1996  1997  Gross income  299095.1  291924.5  303265  272220.5***  303552.8  317956.8**  Age  35.78  33.57***  35.81  34.37***  35.66  36.01  Experience  15.57  14.58***  15.84  15.6  15.95  16.45  Education              Low  0.306  0.448***  0.301  0.421***  0.3  0.246*  Medium  0.571  0.510**  0.574  0.521**  0.575  0.59  High  0.111  0.036***  0.113  0.0487***  0.113  0.149*  Mfr. sector              Food and bev.  0.199  0.568***  0.202  0.338***  0.199  0.343***  Metal products  0.463  0.336***  0.461  0.259***  0.465  0.295***  Wood products  0.109  0.03***  0.109  0.005***  0.108  0.134  Other mfr. indust.  0.229  0.066***  0.228  0.397***  0.229  0.228  Individuals  563  57899  390  58844  268  61225  Sample  1996–2001  1997–2002  1998–2003  Disp. year  1998  1999  2000  Gross income  303011.2  294882.2  306357.4  304647  312491.6  299407.7**  Age  35.68  35.69  35.82  35.28**  35.94  35.61  Experience  16.2  16.16  16.55  16.86  16.87  17.1  Education              Low  0.3  0.293  0.295  0.356***  0.29  0.325*  Medium  0.573  0.603  0.575  0.567  0.579  0.566  High  0.114  0.095  0.117  0.07***  0.118  0.089**  Mfr. sector              Food and bev.  0.198  0.451***  0.183  0.278***  0.182  0.486***  Metal products  0.456  0.158***  0.454  0.481  0.46  0.374***  Wood products  0.113  0.230***  0.111  0.041***  0.114  0.033***  Other mfr. indust.  0.233  0.161**  0.251  0.199**  0.244  0.107***  Individuals  348  61135  702  60084  551  58456  Notes: Author’s calculations with data from Statistics Denmark. The table shows descriptive statistics for displaced and nondisplaced manufacturing workers two years before the displacement period of the respective sample. Labor earnings, wages, and gross income are measured in 2000 DKKs. Standard errors are in parentheses and clustered at the individual level. *Significant at 10% level; **significant at 5% level; ***significant at 1% level. View Large Table A.3. Average real gross income losses (pre-decentralization). Disp. year  1982  1983  1984  1985  1986  $$D_{it}^{-1}$$  −5822.870  346.579  −5425.082**  7226.436***  −5203.899*     (7108.505)  (1265.987)  (2459.115)  (1933.702)  (2718.442)  Displacement  −10333.621  1234.724  −6467.426  1000.653  −5659.449*    (7958.089)  (1428.032)  (4456.945)  (2605.696)  (2918.881)  $$D_{it}^{1}$$  −6378.632**  −2566.888  −1838.633  −579.673  1234.947    (2838.265)   (1781.667)  (4424.249)  (5249.173)  (4119.185)  $$D_{it}^{2}$$  −1954.378  −1694.688  53.784  1738.79  −3478.647    (3398.521)   (2244.434)  (4473.693)  (5901.728)  (4035.749)  $$D_{it}^{2}$$  5968.243  −4328.676  −3760.320  −6344.822  −8386.598*     (4412.835)  (3007.733)   (4970.008)  (6550.623)  (5093.659)  Observations:  331127  345249  335713  316599  304206  Disp. year  1987  1988        $$D_{it}^{-1}$$  4592.791***  6345.078**          (1358.475)   (2036.713)        Displacement  3255.017  8478.351          (2114.587)   (5589.042)        $$D_{it}^{1}$$  −1034.001  −2892.284          (2673.976)  (2286.879)        $$D_{it}^{2}$$  1365.098  −152.887          (2828.742)  (4883.442)        $$D_{it}^{3}$$  6083.684  −2434.299          (4051.434)  (2870.725)        Observations:  294318  309727        Disp. year  1982  1983  1984  1985  1986  $$D_{it}^{-1}$$  −5822.870  346.579  −5425.082**  7226.436***  −5203.899*     (7108.505)  (1265.987)  (2459.115)  (1933.702)  (2718.442)  Displacement  −10333.621  1234.724  −6467.426  1000.653  −5659.449*    (7958.089)  (1428.032)  (4456.945)  (2605.696)  (2918.881)  $$D_{it}^{1}$$  −6378.632**  −2566.888  −1838.633  −579.673  1234.947    (2838.265)   (1781.667)  (4424.249)  (5249.173)  (4119.185)  $$D_{it}^{2}$$  −1954.378  −1694.688  53.784  1738.79  −3478.647    (3398.521)   (2244.434)  (4473.693)  (5901.728)  (4035.749)  $$D_{it}^{2}$$  5968.243  −4328.676  −3760.320  −6344.822  −8386.598*     (4412.835)  (3007.733)   (4970.008)  (6550.623)  (5093.659)  Observations:  331127  345249  335713  316599  304206  Disp. year  1987  1988        $$D_{it}^{-1}$$  4592.791***  6345.078**          (1358.475)   (2036.713)        Displacement  3255.017  8478.351          (2114.587)   (5589.042)        $$D_{it}^{1}$$  −1034.001  −2892.284          (2673.976)  (2286.879)        $$D_{it}^{2}$$  1365.098  −152.887          (2828.742)  (4883.442)        $$D_{it}^{3}$$  6083.684  −2434.299          (4051.434)  (2870.725)        Observations:  294318  309727        Notes: Author’s calculations with data from Statistics Denmark. Dep. var.: Annual gross income. Gross income is measured in 2000 DKKs and includes zeros. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Displacements occurred between November in year (−1) and November in year (0). Estimations are based on subsamples including displaced and nondisplaced workers. Each subsample follows both groups from two years before until three years after the displacement. Regressions include worker fixed effects, year fixed effects, and age categories. Standard errors are in parentheses and clustered at the individual level. *Significant at 10% level; **significant at5% level; ***significant at 1% level. View Large Table A.4. Average real gross income losses (postdecentralization). Disp. year  1989  1990  1991  1992  1993  $$D_{it}^{-1}$$  −4275.060**  −1814.476  35.11  −6449.536**  3089.666**    (1774.308)  (1567.708)  (2015.536)  (2079.534)  (1546.235)   Displacement  −7921.461**  −7230.973**  −4776.423  −10862.486***  −4999.634    (3245.806)  (3504.699)  (8104.956)  (3200.725)  (3071.603)   $$D_{it}^{1}$$  −21461.412***  −18033.971***  −13453.181**  −20182.018***  −9283.576**    (3616.340)  (5171.091)  (6422.683)  (4077.017)  (3736.323)  $$D_{it}^{2}$$  −20959.511***  −25164.434***  −7764.893  −17163.633***  −6469.091    (3973.129)  (3976.383)  (7992.864)  (4555.523)  (4058.477)  $$D_{it}^{3}$$  −28371.280***  −31038.667***  −22670.557**  −22614.100***  −11033.397**    (4447.770)  (4227.903)  (9537.456)  (4667.531)  (4178.573)  Observations  319360  333616  336281  338781  347311  Disp. year  1994  1995  1996  1997  1998  $$D_{it}^{-1}$$  −6788.029**  −1404.602  −6210.384***  10581.042***  9313.031***    (2460.529)  (1631.844)  (1499.809)  (2779.603)  (2581.047)   Displacement  −8922.475**  2452.191  −3682.919  38886.286***  −939.723    (3460.701)  (2869.926)  (3170.452)  (6410.583)  (5031.128)  $$D_{it}^{1}$$  −16853.147***  −13448.483***  −18191.982***  −11872.156**  −19319.945***    (5087.391)  (2986.011)  (3458.957)  (5082.981)  (4152.068)  $$D_{it}^{2}$$  −16730.114***  −19033.770***  −16315.782***  −577.569  −21456.393***    (4249.352)  (3479.684)  (3903.711)  (5063.351)  (3786.774)  $$D_{it}^{3}$$  −18627.864***  −18160.333***  −13593.930**  −2566.970  −17517.248***    (4560.996)  (4874.681)  (4380.230)  (5752.573)  (4187.611)  Observations:  347987  336725  340373  352623  352509  Disp. year  1999  2000        $$D_{it}^{-1}$$  6897.004**  7040.556***          (2340.955)  (1703.726)        Displacement  −1321.610  6448.736*          (3268.578)  (3384.888)        $$D_{it}^{1}$$  −27554.941***  −19806.119***          (3010.144)  (3422.487)        $$D_{it}^{2}$$  −19399.242***  −25884.981***          (3279.583)   (3570.380)        $$D_{it}^{3}$$  −18680.271***  −23407.572***          (3689.807)  (4082.181)        Observations:  348741  338037        Disp. year  1989  1990  1991  1992  1993  $$D_{it}^{-1}$$  −4275.060**  −1814.476  35.11  −6449.536**  3089.666**    (1774.308)  (1567.708)  (2015.536)  (2079.534)  (1546.235)   Displacement  −7921.461**  −7230.973**  −4776.423  −10862.486***  −4999.634    (3245.806)  (3504.699)  (8104.956)  (3200.725)  (3071.603)   $$D_{it}^{1}$$  −21461.412***  −18033.971***  −13453.181**  −20182.018***  −9283.576**    (3616.340)  (5171.091)  (6422.683)  (4077.017)  (3736.323)  $$D_{it}^{2}$$  −20959.511***  −25164.434***  −7764.893  −17163.633***  −6469.091    (3973.129)  (3976.383)  (7992.864)  (4555.523)  (4058.477)  $$D_{it}^{3}$$  −28371.280***  −31038.667***  −22670.557**  −22614.100***  −11033.397**    (4447.770)  (4227.903)  (9537.456)  (4667.531)  (4178.573)  Observations  319360  333616  336281  338781  347311  Disp. year  1994  1995  1996  1997  1998  $$D_{it}^{-1}$$  −6788.029**  −1404.602  −6210.384***  10581.042***  9313.031***    (2460.529)  (1631.844)  (1499.809)  (2779.603)  (2581.047)   Displacement  −8922.475**  2452.191  −3682.919  38886.286***  −939.723    (3460.701)  (2869.926)  (3170.452)  (6410.583)  (5031.128)  $$D_{it}^{1}$$  −16853.147***  −13448.483***  −18191.982***  −11872.156**  −19319.945***    (5087.391)  (2986.011)  (3458.957)  (5082.981)  (4152.068)  $$D_{it}^{2}$$  −16730.114***  −19033.770***  −16315.782***  −577.569  −21456.393***    (4249.352)  (3479.684)  (3903.711)  (5063.351)  (3786.774)  $$D_{it}^{3}$$  −18627.864***  −18160.333***  −13593.930**  −2566.970  −17517.248***    (4560.996)  (4874.681)  (4380.230)  (5752.573)  (4187.611)  Observations:  347987  336725  340373  352623  352509  Disp. year  1999  2000        $$D_{it}^{-1}$$  6897.004**  7040.556***          (2340.955)  (1703.726)        Displacement  −1321.610  6448.736*          (3268.578)  (3384.888)        $$D_{it}^{1}$$  −27554.941***  −19806.119***          (3010.144)  (3422.487)        $$D_{it}^{2}$$  −19399.242***  −25884.981***          (3279.583)   (3570.380)        $$D_{it}^{3}$$  −18680.271***  −23407.572***          (3689.807)  (4082.181)        Observations:  348741  338037        Notes: See Table A.3. View Large Appendix B: Macroeconomic Developments And Displaced Workers’ Earnings Losses Section 6 showed that macroeconomic conditions had only a weak influence on displaced workers’ short-term income losses. This outcome remains in contrast to the findings in the previous literature of a rather strong relationship between macroeconomic conditions and displaced losses. To infer more specifically to what extent macroeconomic changes may account for the increase of income losses after the reform, columns (2) and (3) of Table B.1 show further specifications of equation (2). These specifications include the national unemployment rate (II) and a dummy variable for the Nordic crisis between 1989 and 1993 (III). As both columns show, the results prove extremely robust to the inclusion of those macroeconomic indicators. Table B.1. Average real gross income losses (pooled sample). Specification  I  II  III  IV  V    Pre-decentralization period  $$D_{it}^{-1}$$  2901.831**  2903.621**  2903.640**  2901.660**  2903.663    (1085.318)  (1085.320)  (1085.321)  (1085.318)  (1085.321)  Displacement  2553.732  2557.736  2558.776  2553.191  2557.748    (1897.388)  (1897.391)  (1897.387)  (1897.388)  (1897.392)  $$D_{it}^{1}$$  −1190.575  −1185.547  −1190.123  −1190.251  −1183.038    (1167.817)  (1167.807)  (1167.817)  (1167.822)  (1167.801)  $$D_{it}^{2}$$  −290.036  −284.083  −293.713  −289.849  −286.725    (1731.657)  (1731.641)  (1731.665)  (1731.660)  (1731.648)  $$D_{it}^{3}$$  −2060.017  −2055.721  −2066.651  −2060.229  −2062.172    (1584.277)  (1584.269)  (1584.297)  (1584.278)  (1584.284)    Post-decentralization period  $$Reform \cdot D_{it}^{-1}$$  −696.665  −797.862  −649.109  −648.533  −453.310    (1255.200)  (1260.159)  (1257.714)  (1275.024)  (1317.415)  Reform ·  −1471.296  −1724.998  −1463.115  −1288.495  −814.662  Disp.  (2317.848)  (2347.678)  (2318.557)  (2362.736)  (2464.033)  $$Reform \cdot D_{it}^{1}$$  −14331.219***  −14747.380***  −14500.329***  −14175.279***  −13379.700***    (1816.255)  (1877.514)  (1808.330)  (1909.954)  (2082.575)  $$Reform \cdot D_{it}^{2}$$  −15220.223***  −15827.106***  −15571.643***  −15139.448***  −14365.461***    (2326.961)  (2416.846)  (2302.553)  (2393.104)  (2740.708)  $$Reform \cdot D_{it}^{3}$$  −16549.619***  −17326.460***  −17088.650***  −16354.378***  −15628.610***    (2427.847)  (2551.854)  (2394.630)  (2475.713)  (3102.946)  Observations:  2080471  2080471  2080471  2080471  2080471  Controls:            Unemployment  No  Yes  No  No  No  Nordic crisis  No  No  Yes  No  No  FDI flows  No  No  No  Yes  No  Ex-/imports  No  No  No  No  Yes  Specification  I  II  III  IV  V    Pre-decentralization period  $$D_{it}^{-1}$$  2901.831**  2903.621**  2903.640**  2901.660**  2903.663    (1085.318)  (1085.320)  (1085.321)  (1085.318)  (1085.321)  Displacement  2553.732  2557.736  2558.776  2553.191  2557.748    (1897.388)  (1897.391)  (1897.387)  (1897.388)  (1897.392)  $$D_{it}^{1}$$  −1190.575  −1185.547  −1190.123  −1190.251  −1183.038    (1167.817)  (1167.807)  (1167.817)  (1167.822)  (1167.801)  $$D_{it}^{2}$$  −290.036  −284.083  −293.713  −289.849  −286.725    (1731.657)  (1731.641)  (1731.665)  (1731.660)  (1731.648)  $$D_{it}^{3}$$  −2060.017  −2055.721  −2066.651  −2060.229  −2062.172    (1584.277)  (1584.269)  (1584.297)  (1584.278)  (1584.284)    Post-decentralization period  $$Reform \cdot D_{it}^{-1}$$  −696.665  −797.862  −649.109  −648.533  −453.310    (1255.200)  (1260.159)  (1257.714)  (1275.024)  (1317.415)  Reform ·  −1471.296  −1724.998  −1463.115  −1288.495  −814.662  Disp.  (2317.848)  (2347.678)  (2318.557)  (2362.736)  (2464.033)  $$Reform \cdot D_{it}^{1}$$  −14331.219***  −14747.380***  −14500.329***  −14175.279***  −13379.700***    (1816.255)  (1877.514)  (1808.330)  (1909.954)  (2082.575)  $$Reform \cdot D_{it}^{2}$$  −15220.223***  −15827.106***  −15571.643***  −15139.448***  −14365.461***    (2326.961)  (2416.846)  (2302.553)  (2393.104)  (2740.708)  $$Reform \cdot D_{it}^{3}$$  −16549.619***  −17326.460***  −17088.650***  −16354.378***  −15628.610***    (2427.847)  (2551.854)  (2394.630)  (2475.713)  (3102.946)  Observations:  2080471  2080471  2080471  2080471  2080471  Controls:            Unemployment  No  Yes  No  No  No  Nordic crisis  No  No  Yes  No  No  FDI flows  No  No  No  Yes  No  Ex-/imports  No  No  No  No  Yes  Notes: Author’s calculations with data from Statistics Denmark. Dep. var.: Annual gross income. Gross income is measured in 2000 DKKs and includes zeros. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Displacements occurred between November in year (−1) and November in year (0). Estimations are based on a pooled sample. Reform is a dummy indicating whether the worker was displaced before or after the wage bargaining reform. Exports, imports, and FDI inflows and outflows are measured in millions of US dollars. Regressions include worker fixed effects, year fixed effects, and age categories. Standard errors are in parentheses and clustered at the individual level. **Significant at 5% level; ***significant at 1% level. View Large However, national unemployment may be strongly related to the decentralization of the wage bargaining system, so that more exogenous indicators for measuring macroeconomic conditions may produce very different results. Therefore, specification IV controls for inflows and outflows of Foreign Direct Investments (FDI), and specification V accounts for exports and imports measured in millions of US dollars.19 Although both of these macroeconomic indicators appear to have a somewhat stronger effect on the results, the main outcome does not change. The main explanation for macroeconomic conditions having only little effect on displaced workers’ income losses may be that generous UI benefits and large payments of the “guarantee fund” smooth the income losses of displaced workers. I provide evidence for this argument in Figure B.1, which shows displaced workers’ real annual earnings losses (excluding UI benefits and all other kinds of social benefit payments). I calculate the labor earnings by summing workers’ November job earnings and the sum of their wages for all other jobs in a given year. If no labor income is reported for a worker in a given year, I count workers’ labor earnings as 0. Figure B.1. View largeDownload slide Annual earnings losses without UI benefits. Estimation according to equation (2). Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Annual earnings are measured in 2000 DKKs and includes zeros. Average earnings are adjusted to a base earning in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). Regressions include worker fixed effects, year fixed effects, and age categories. The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure B.1. View largeDownload slide Annual earnings losses without UI benefits. Estimation according to equation (2). Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Annual earnings are measured in 2000 DKKs and includes zeros. Average earnings are adjusted to a base earning in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). Regressions include worker fixed effects, year fixed effects, and age categories. The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure B.1 presents the average estimated earnings losses for the pre-decentralization period but divides the post-decentralization period (after 1989) into two separate periods: one for the recession period between 1989 and 1993, and one for the expansion period between 1994 and 2000. The figure presents the results as percentages of displaced workers’ predisplacement earnings losses. The figure shows that displaced workers’ real earnings losses are generally much larger than their real income losses, and—as in most the US studies—displaced workers’ earnings losses peak during the displacement year. Notably, the magnitudes of displaced workers’ earnings losses are similar to those in a Swedish study by Eliason and Storrie (2006), who also use labor earnings instead of total income. Most importantly, in contrast to the findings for income losses, displaced workers’ earnings losses respond much more strongly to macroeconomic conditions than do income losses and were significantly larger during the Nordic crisis than afterwards. Thus generous UI benefits indeed appear to smooth displaced workers’ income losses and explain why displaced workers’ income losses do not respond strongly to changes in the business cycle. Appendix C: Additional Sensitivity Analysis This section presents further three sensitivity analysis to prove the robustness of the results. First, a number of studies have shown that the selection of the groups of displaced and nondisplaced workers may have some influence on the magnitude of estimated income losses, and researchers rely on different selection criteria for different reasons. In this study I have decided to largely follow what the most prominent studies in the literature have done. As common in this literature, I have required that displaced workers separate from firms with more than 50 employees and that nondisplaced workers follow a relatively stable employment in firms with more than 50 employees (i.e., I require nondisplaced workers to have had at least one wage employment with a positive income for each subsequent year). However, other samples that allow for a larger amount of data can be constructed. Therefore, column (1) of Table C.1 provides a robustness check for which I reduced the firm size requirement to firms with only more than 30 employees. Moreover, I did not impose any other restriction on nondisplaced workers other than the restriction that they had at least three years of tenure. This sample selection increased the sample size by about 600,000 observations. As column (1) of Table C.1 shows, the results prove to be very similar to the results of the main specification. Table C.1. Alternative specifications (pooled sample). Specification  Sample cond.  First differences  Matching  $$D_{it}^{-1}$$  2227.190**  2244.689**  3516.580**    (942.617)  (1084.659)  (1115.474)  Displacement  1904.195  1261.441  3675.282*    (1646.380)  (1893.966)  (1941.637)  $$D_{it}^{1}$$  −1937.283*  −3078.235**  864.795    (1062.624)  (1161.792)  (1275.026)  $$D_{it}^{2}$$  −1744.543  −2714.031  2451.248    (1534.297)  (1721.671)  (1819.188)  $$D_{it}^{3}$$  −3108.941**  −5197.753**  1175.792    (1445.149)  (1587.526)  (1725.369)    Post-decentralization period  Reform$$\cdot D_{it}^{-1}$$  −1198.839  −913.765  −1629.755    (1092.562)  (1244.002)  (1313.097)  Reform · Disp.  −2488.040  −1830.713  −3072.338    (1991.826)  (2277.915)  (2413.820)  Reform$$\cdot D_{it}^{1}$$  −14207.696***  −14759.819***  −17863.161***    (1614.160)  (1758.585)  (1935.158)  Reform$$\cdot D_{it}^{2}$$  −14238.886***  −15921.435***  −19543.120***    (2046.070)  (2243.576)  (2440.424)  Reform$$\cdot D_{it}^{3}$$  −15740.132***  −17185.425***  −21110.164***    (2133.569)  (2308.938)  (2648.002)  Observations  2723280  1833907  159174  Specification  Sample cond.  First differences  Matching  $$D_{it}^{-1}$$  2227.190**  2244.689**  3516.580**    (942.617)  (1084.659)  (1115.474)  Displacement  1904.195  1261.441  3675.282*    (1646.380)  (1893.966)  (1941.637)  $$D_{it}^{1}$$  −1937.283*  −3078.235**  864.795    (1062.624)  (1161.792)  (1275.026)  $$D_{it}^{2}$$  −1744.543  −2714.031  2451.248    (1534.297)  (1721.671)  (1819.188)  $$D_{it}^{3}$$  −3108.941**  −5197.753**  1175.792    (1445.149)  (1587.526)  (1725.369)    Post-decentralization period  Reform$$\cdot D_{it}^{-1}$$  −1198.839  −913.765  −1629.755    (1092.562)  (1244.002)  (1313.097)  Reform · Disp.  −2488.040  −1830.713  −3072.338    (1991.826)  (2277.915)  (2413.820)  Reform$$\cdot D_{it}^{1}$$  −14207.696***  −14759.819***  −17863.161***    (1614.160)  (1758.585)  (1935.158)  Reform$$\cdot D_{it}^{2}$$  −14238.886***  −15921.435***  −19543.120***    (2046.070)  (2243.576)  (2440.424)  Reform$$\cdot D_{it}^{3}$$  −15740.132***  −17185.425***  −21110.164***    (2133.569)  (2308.938)  (2648.002)  Observations  2723280  1833907  159174  Notes: Author’s calculations with data from Statistics Denmark. Dep. var.: Annual gross income. Gross income is measured in 2000 DKKs and includes zeros. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Displacements occurred between November in year (−1) and November in year (0). Estimations are based on a pooled sample. Reform is a dummy indicating whether the worker was displaced before or after the wage bargaining reform. *Significant at 10% level; **significant at 5% level; ***significant at 1% level. View Large Second, if the income path of displaced workers had systematically differed from that of nondisplaced workers—even in the absence of the displacement—regression equations (1) and (2) may produce biased results. Moreover, if those worker-specific trends had changed over time, the previous results may have picked up such changes in unobserved dynamic heterogeneity. Wooldridge (2015) suggests testing for this possibility by performing both fixed effects and first difference estimators. Column (2) in Table C.1 shows that performing a first-difference regression leads to results similar to those in the standard fixed effects approach of equation (2). Third, Tables A.1 and A.2 show that baseline characteristics between displaced and nondisplaced workers were not fully balanced before the displacement event. To reduce the potential bias that may arise from systematic differences in unobserved characteristics of displaced and nondisplaced workers, previous studies suggested pairing displaced and nondisplaced workers according to their predisplacement characteristics before estimating equations such as (1) and (2) (Hijzen et al. 2010). This approach may be preferred to simple linear regression methods if the functional form assumptions of the latter are violated, and if members of the treatment or control group lie outside the support of the propensity distribution. Therefore, specification III of Table C.1 presents results of an estimation for which I used single-nearest-neighbor propensity score matching to pair displaced and nondisplaced workers. I estimate the propensity score from a set of characteristics in the second period before displacement. The characteristics include industry, education, experience, and age. Table C.1 shows that the results are very similar to those of the standard approach. In other words, systematic differences in predisplacement characteristics between displaced and nondisplaced workers appear not to account for the increase in displaced workers’ income losses. Notes The editor in charge of this paper was Claudio Michelacci. Acknowledgements I particularly thank Edward Lazear, Nils Westergaard-Nielsen, Thomas Siedler, Silke Anger, Uschi Backes-Gellner, Bob Hall, Paul Oyer, Kathryn Shaw, Hank Farber, Ulrich Kaiser, Jens Mohrenweiser, Malte Sandner, Nikolaj Harmon, Jens Iversen, and the participants of the Econ brown-bag seminar at the Stanford Graduate School of Business for their helpful comments and suggestions. I also thank Natalie Reid for language editing and Astrid Erismann for research assistance. This study is partially funded by the Swiss Federal Office for Professional Education and Technology through its Leading House on the Economics of Education, Firm Behavior, and Training Policies. During the work on this paper, I am grateful for financial support from the Swiss National Science Foundation. I thank the Cycles, Adjustment, and Policy research unit, CAP, Department of Economics and Business, Aarhus University, for support and for making the data available. Janssen is a Research Affiliate at IZA Footnotes 1 For earlier survey articles, see Fallick (1996) and Kletzer (1998). Other studies investigate how job displacement influences unemployment duration (e.g., Chan and Stevens 1999). Yet other studies show that worker displacement has substantial effects on health, mortality, fertility, and children’s labor market outcomes (e.g., Oreopoulos, Page, and Stevens 2008; Eliason and Storrie 2009; Sullivan and Von Wachter 2009; Stevens and Schaller 2011; Del Bono, Weber, and Winter-Ebmer 2012). 2 For example, the US studies find displaced workers’ earnings and wage losses of between 11% (e.g., Ruhm 1991; Farber 1993, 1997) and 40% per year (e.g., Ruhm 1991; Farber 1993, 1997; Stevens 1997; Couch 1998; Schoeni and Dardia 1997; Couch and Placzek 2010; Davis and von Wachter 2011). Most European studies find smaller income losses that range between 3% and 15% (e.g., Bender et al. 2002; Burda and Mertens 2001; Couch 2001; Eliason and Storrie 2006; Von Wachter and Bender 2006; Schmieder et al. 2010; Hijzen et al. 2010; Huttunen et al. 2011; Korkeamäki and Kyyrä 2014; Ichino et al. 2016). However, the evidence is more diverse for Europe. For example, Hijzen et al. (2010) find up to 35% for Britain, and Eliason and Storrie (2006) find earnings losses of up to 19% for Sweden. 3 The employer side pushed for this change because internationalization and technological change meant that wage contracts were not sufficiently flexible to accommodate local conditions. The employee side did not oppose this change, because it feared that labor demand might suffer if wages were not sufficiently flexible. Moreover, the employer side agreed to introduce mandatory labor market pensions, on-the-job training, and child-care leave. 4 Several studies show that the decentralization process affected wage setting in the manufacturing sector. For example, Bingley and Westergaard-Nielsen (2003) show that returns to tenure more than doubled at the beginning of the decentralization period in 1989, and Eriksson and Westergaard-Nielsen (2009) show that wage dispersion increased at the beginning of the late 1980s. Finally, Dahl et al. (2013) find that returns to skills are larger and that wage dispersion is greater under firm-level wage bargaining in Denmark. 5 In rare cases, guarantee fund payments can be extended for up to six months. 6 For example, Davis and von Wachter (2011) exclude all closing firms from their sample because they are unable to distinguish a firm closure from a simple change in a firm ID. As I do not face this problem in my data, I do not remove firms that go out of business. 7 Nondisplaced workers may become unemployed between job switches. 8 Some studies count a worker as unemployed only if he or she received UI benefits for more than 20% of the ongoing year. But given that displaced workers may indeed experience many short periods of real unemployment, such an approach may be too conservative for the purpose of this paper. 9 The markers in Figures 6–9, 11, 12, and B.1 express monetary displacement losses as a percentage of the workers’ base income. Displacement losses as function of the workers base income are essentially functions g() of the coefficient estimates δk. To calculate the confidence intervals for g(δk), I approximated the standard errors of g(δk) by $$\sqrt{(G^{\prime }VG)}$$, where G is a vector of the derivatives ∂g/∂δk and V is the estimated variance–covariance of the δks. This method is commonly referred to as the delta method. 10 If the number of time periods becomes large in comparison to the number of observations, fixed effect estimators may produce biased results. This may be especially problematic if the sample length is systematically longer for workers who are displaced after the reform. Wooldridge (2015) argues that both fixed effects and first difference estimators should produce similar results if the assumption of time constant unobserved heterogeneity holds. Therefore, Table C.1 in the appendix provides additional results from a first-difference approach. Both approaches produce very similar results in the short panels. 11 Although the separate samples overlap, the estimation sample for equation (2) contains only one observation per worker year. A very small number of workers experienced more than one displacement within a period of less than four years. For those few workers, I took into account only their first displacement. 12 Only very few workers in the sample experienced more than one displacement. However, experiencing a displacement both before and after the reform is a necessary condition for observing variation of the $$ {Reform}$$ dummy within a single worker’s observation period. 13 In contrast to the parallel trends assumption, balanced predisplacement characteristics are not an identification assumption of the difference-in-differences estimators in equations (1) and (2). However, more recently a number of studies have experimented with matching and reweighting techniques to more precisely estimate displaced workers’ income losses (e.g., Couch and Placzek 2010; Hijzen et al. 2010). Following these approaches, Table C.1 in the appendix presents additional results for which I matched the samples of nondisplaced workers so that their average baseline characteristics match those of displaced workers. As found in those earlier studies, matching and reweighting techniques do not substantially change the results. 14 I follow Davis and von Wachter (2011) and calculate the DPPVs as $$\mathit{ DPV}^{\mathit{ Loss}}_{\tau }=\sum _{k=1}^{3}\bar{\delta _{t}^{k}}({1}/{(1+r)^{t-1}})$$. I also assume a real interest rate of 5% and normalize the DPVs, using displaced workers’ mean income during the baseline period of each sample to account for changing income levels over time. However, I do not extrapolate the income losses for periods after the observation period as Davis and von Wachter (2011) do. 15 If rigid wages under centralized wage bargaining would primarily prevent displaced workers from reentering employment at their optimal reservation wages, displaced workers’ income losses should decrease, not increase, under decentralized wage bargaining systems. 16 The postdisplacement wage may stem from the first new job in the displacement year or from the first job that the worker found any time after that. However, more than 90% of all displaced workers found a job with a valid wage observation within less a year after their job loss. 17 I cannot include worker fixed effects in equation (4), because $$\text{log}w_{i,\mathit{ Pre-post}}^{d}$$ assigns one single observation to each worker. 18 With special permission from the employer, the period could be extended to 52 weeks. 19 The information on Foreign Direct Investments (FDI), exports, and imports stem from the homepage of the United Nations Conference on Trade and Development (http://unctad.org/en/Pages/Home.aspx). FDI inflows and outflows are measured in US dollars at current prices and current exchange rates. They comprise capital provided by a foreign direct investor to a FDI enterprise or capital received by a foreign direct investor from a FDI enterprise. FDI includes the following three components: equity capital, reinvested earnings, and intracompany loans. Imports and exports measure the total trade in goods and services in US dollars at current prices and current exchange rates. References Abowd John M., Kramarz Francis ( 1999). “ The Analysis of Labor Markets Using Matched Employer-Employee Data.” Handbook of Labor Economics , 3, 2629– 2710. Google Scholar CrossRef Search ADS   Andersen Torben M. ( 2003). “ Changes in Danish Labour Market Bargaining: The Prototypical Case of Organized Decentralization?” In National Regimes of Collective Bargaining in Transformation: Nordic Trends in Comparative Perspective , edited by Dølvik Jon E., Engelstad Fredrik. Makt-og Demokratiutredningen Norge  1918– 2003. Andersen Søren K., Mailand Mikkel ( 2005). “ The Danish Flexicurity Model.” Compiled for the Danish Ministry of Employment, Copenhagen. Andersen Torben M., Svarer Michael ( 2007). “ Flexicurity—Labour Market Performance in Denmark.” CESifo Economic Studies , 53, 389– 429. Google Scholar CrossRef Search ADS   Bender Stefan, Dustmann Christian, Margolis David, Meghir Costas ( 2002). “ Worker Displacement in France and Germany.” In Losing Work, Moving On: International Perspectives on Worker Displacement , edited by Kuhn Peter J., pp. 375– 470. WE Upjohn Institute. Google Scholar CrossRef Search ADS   Bingley Paul, Westergaard-Nielsen Niels ( 2003). “ Returns to Tenure, Firm-Specific Human Capital and Worker Heterogeneity.” International Journal of Manpower , 24, 774– 788. Google Scholar CrossRef Search ADS   Burda Michael C., Mertens Antje ( 2001). “ Estimating Wage Losses of Displaced Workers in Germany.” Labour Economics , 8, 15– 41. Google Scholar CrossRef Search ADS   Card David, Heining Jörg, Kline Patrick ( 2013). “ Workplace Heterogeneity and the Rise of West German Wage Inequality.” The Quarterly Journal of Economics , 128, 967– 1015. Google Scholar CrossRef Search ADS   Card David, Lemieux Thomas, Riddell Craig W. ( 2004). “ Unions and Wage Inequality.” Journal of Labor Research , 25, 519– 559. Google Scholar CrossRef Search ADS   Chan Sewin, Stevens Ann Huff ( 1999). “ Employment and Retirement Following a Late-career Job Loss.” American Economic Review , 89(2), 211– 216. Google Scholar CrossRef Search ADS   Couch Kenneth A. ( 1998). “ Late Life Job Displacement.” The Gerontologist , 38, 7– 17. Google Scholar CrossRef Search ADS PubMed  Couch Kenneth A. ( 2001). “ Earnings Losses and Unemployment of Displaced Workers in Germany.” Industrial and Labor Relations Review , 54, 559– 572. Google Scholar CrossRef Search ADS   Couch Kenneth A., Placzek Dana W. ( 2010). “ Earnings Losses of Displaced Workers Revisited.” American Economic Review , 100(1), 572– 589. Google Scholar CrossRef Search ADS   Dansk Arbejdsgiverforening ( 2004). Arbejdsmarkedsrapport 2004 Dansk Arbejdsgiverforening, Copenhagen. Dahl Christian M., Le Maire Daniel, Munch Jakob R. ( 2013). “ Wage Dispersion and Decentralization of Wage Bargaining.” Journal of Labor Economics , 31, 501– 533. Google Scholar CrossRef Search ADS   Davis Steven J., Wachter Till von ( 2011). “ Recessions and the Costs of Job Loss.” Brookings Papers on Economic Activity , 2, 1– 72. Google Scholar CrossRef Search ADS   Del Bono, Emilia, Weber Andrea, Winter-Ebmer Rudolf ( 2012). “ Clash of Career and Family: Fertility Decisions after Job Displacement.” Journal of the European Economic Association , 10, 659– 683. Google Scholar CrossRef Search ADS   Dustmann Christian, Ludsteck Johannes, Schönberg Uta ( 2009). “ Revisiting the German Wage Structure.” Quarterly Journal of Economics , 124, 843– 881. Google Scholar CrossRef Search ADS   Eliason Marcus, Storrie Donald ( 2006). “ Lasting or Latent Scars? Swedish Evidence on the Long-Term Effects of Job Displacement.” Journal of Labor Economics , 24, 831– 856. Google Scholar CrossRef Search ADS   Eliason Marcus, Storrie Donald ( 2009). “ Does Job Loss Shorten Life?” Journal of Human Resources , 44, 277– 302. Google Scholar CrossRef Search ADS   Eriksson Tor, Westergaard-Nielsen Niels ( 2009). “ Wage and Labor Mobility in Denmark, 1980-2000.” In The Structure of Wages: An International Comparison , edited by Lazear Edward P., Shaw Kathlyn L.. University of Chicago Press, pp. 101– 123. Fallick Bruce C. ( 1996). “ A Review of the Recent Empirical Literature on Displaced Workers.” Industrial and Labor Relations Review , 50, 5– 16. Google Scholar CrossRef Search ADS   Farber Henry S. ( 1997). “ The Changing Face of Job Loss in the United States, 1981-1995.” Brookings Papers on Economic Activity , 55– 142. Farber Henry S. ( 1993). “ The Incidence and Costs of Job Loss: 1982-91.” Brookings Papers on Economic Activity: Microeconomics , 1, 73– 132. Google Scholar CrossRef Search ADS   Freeman Richard B. ( 1980). “ The Exit-Voice Tradeoff in the Labor Market: Unionism, Job Tenure, Quits, and Separations.” The Quarterly Journal of Economics , 94, 643– 673. Google Scholar CrossRef Search ADS   Geerdsen Lars P. ( 2006). “ Is There a Threat Effect of Labour Market Programmes? A Study of ALMP in the Danish UI system.” The Economic Journal , 116, 738– 750. Google Scholar CrossRef Search ADS   Hijzen Alexander, Upward Richard, Wright Peter W. ( 2010). “ The Income Losses of Displaced Workers.” Journal of Human Resources , 45, 243– 269. Google Scholar CrossRef Search ADS   Huttunen Kristiina, Møen Jarle, Salvanes Kjell G. ( 2011). “ How Destructive is Creative Destruction? Effects of Job Loss on Job Mobility, Withdrawal, and Income.” Journal of the European Economic Association , 9, 840– 870. Google Scholar CrossRef Search ADS   Ichino Andrea, Schwerdt Guido, Winter-Ebmer Rudolf, Zweimüller Josef ( 2016). “ Too Old to Work, too Young to Retire?” The Journal of the Economics of Aging . Iversen Jens, Malchow-Møller Nikolaj, Sørensen Anders ( 2010). “ Returns to Schooling in Self-Employment.” Economics Letters , 109, 179– 182. Google Scholar CrossRef Search ADS   Iversen Torben ( 1996). “ Power, Flexibility, and the Breakdown of Centralized Wage Bargaining: Denmark and Sweden in Comparative Perspective.” Comparative Politics , 28, 399– 436. Google Scholar CrossRef Search ADS   Jacobson Louis S., LaLonde Robert J., Sullivan Daniel G. ( 1993). “ Earnings Losses of Displaced Workers.” American Economic Review , 83(4), 685– 709. Kiander J. ( 2004). “ Growth and Employment in Nordic Welfare States in the 1990s: A Tale of Crisis and Revival.” Government Institute for Economic Research VATT Discussion Papers No. 336, Helsinki, Finnland . Kletzer Lori G. ( 1998). “ Job Displacement.” Journal of Economic Perspectives , 12, 115– 136. Google Scholar CrossRef Search ADS   Kletzer Lori G., Fairlie Robert W. ( 2003). “ The Long-term Costs of Job Displacement for Young Adult Workers.” Industrial and Labor Relations Review , 56, 682– 698. Google Scholar CrossRef Search ADS   Korkeamäki Ossi, Kyyrä Tomi ( 2014). “ A Distributional Analysis of Earnings Losses of Displaced Workers in an Economic Depression and Recovery.” Oxford Bulletin of Economics and Statistics , 76, 565– 588. Google Scholar CrossRef Search ADS   Lazear Edward P. ( 1986). “ Retail Pricing and Clearance Sales.” American Economic Review , 76(1), 14– 32. Lazear Edward P. ( 2012). “ Why Should Inventories Rise when Demand Falls in Housing and other Markets?” The Singapore Economic Review , 57, 32. Google Scholar CrossRef Search ADS   Mayer Gerit, Andersen Torben, Muller Michael ( 2001). “ Employment Restructuring and Flexibility in Austrian and Danish Banking.” European Journal of Industrial Relations , 7, 71– 87. Google Scholar CrossRef Search ADS   Mortensen Dale T. ( 1986). “ Job Search and Labor Market Analysis.” Handbook of Labor Economics , 2, 849– 919. Google Scholar CrossRef Search ADS   Oreopoulos Philip, Page Marianne, Stevens Ann Huff ( 2008). “ The Intergenerational Effects of Worker Displacement.” Journal of Labor Economics , 26(3), 455– 483. Google Scholar CrossRef Search ADS   Pedersen Lars Haagen, Smith Nina, Stephensen Peter ( 1998). “ Wage Formation and Minimum Wage Contracts: Theory and Evidence from Danish Panel Data.” Economic Modeling Working Paper Series 5. Statistics Denmark, Copenhagen, Denmark . Rostgaard Tine M., Christoffersen Mogens N. ( 1999). “ Parental Leave in Denmark.” In Parental Leave: Progress or Pitfall? Research and Policy Issues in Europe , edited by Moss P., Deven F.. NIDI/CBGS Publications, pp. 25– 44. Ruhm Christopher J. ( 1991). “ Are Workers Permanently Scarred by Job Displacements?” American Economic Review , 81(1), 319– 324. Schmieder Johannes F., von Wachter Till, Bender Stefan ( 2010). “ The Long-Term Impact of Job Displacement in Germany During the 1982 Recession on Earnings, Income, and Employment.” Columbia University, Department of Economics Discussion Paper Series DP0910-07. Columbia University, New York, USA . Schoeni Robert F., Dardia Michael ( 1997). “ Earnings Losses of Displaced Workers in the 1990s.” Report, Northwestern University, University of Chicago Joint Center for Poverty Research . Stevens Ann Huff ( 1997). “ Persistent Effects of Job Displacement: The Importance of Multiple Job Losses.” Journal of Labor Economics , 15, 165– 188. Google Scholar CrossRef Search ADS   Stevens Ann Huff, Schaller Jessamyn ( 2011). “ Short-run Effects of Parental Job Loss on Children’s Academic Achievement.” Economics of Education Review , 30, 289– 299. Google Scholar CrossRef Search ADS   Sullivan Daniel, Wachter Till Von ( 2009). “ Job Displacement and Mortality: An Analysis using Administrative Data.” The Quarterly Journal of Economics , 124, 1265– 1306. Google Scholar CrossRef Search ADS   Timmermans Bram ( 2010). “ The Danish Integrated Database for Labor Market Research: Towards Demystification for the English Speaking Audience.” DRUID Working Paper No. 10-16. Aalborg University, Aalborg, Denmark . Von Wachter Till M., Handwerker Elizabeth Weber, Hildreth Andrew K. G. ( 2009). “ Estimating the ‘True’ Cost of Job Loss: Evidence Using Matched Data from California 1991-2000.” US Census Bureau Center for Economic Studies Paper No. CES-WP-09-14 . Google Scholar CrossRef Search ADS   Von Wachter Till, Bender Stefan ( 2006). “ In the Right Place at the Wrong Time: The Role of Firms and Luck in Young Workers’ Careers.” American Economic Review , 96, 1679– 1705. Google Scholar CrossRef Search ADS   Wallerstein Michael, Golden Miriam ( 1997). “ The Fragmentation of the Bargaining Society Wage Setting in the Nordic Countries, 1950 to 1992.” Comparative Political Studies , 30, 699– 731. Google Scholar CrossRef Search ADS   Wooldridge Jeffrey M ( 2015). Introductory Econometrics: A Modern approach . Nelson Education. © The Author 2017. Published by Oxford University Press on behalf of European Economic Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of the European Economic Association Oxford University Press

The Decentralization of Wage Bargaining and Income Losses after Worker Displacement

Loading next page...
 
/lp/ou_press/the-decentralization-of-wage-bargaining-and-income-losses-after-worker-0Km5aLlayo
Publisher
European Economic Association
Copyright
© The Author 2017. Published by Oxford University Press on behalf of European Economic Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
1542-4766
eISSN
1542-4774
D.O.I.
10.1093/jeea/jvx004
Publisher site
See Article on Publisher Site

Abstract

Abstract This paper uses administrative data to study the relationship between the decentralization of wage bargaining systems and the costs of worker displacement. Specifically, the paper exploits a major reform of the wage bargaining system in the Danish manufacturing sector, a reform that changed the wage-setting process from a highly centralized bargaining system at the national level to a decentralized system with a strong emphasis on firm-level wage bargaining. The results show that under the centralized wage bargaining system, displaced workers’ income losses were small, whereas under the decentralized wage bargaining system, these income losses increased substantially, particularly because displaced workers experienced worse wage growth under the decentralized system. The effect persists after controlling for a variety of macroeconomic indicators, and displaced workers’ income losses did not increase in sectors that were not affected by a comparable change in the wage bargaining system. 1. Introduction A great number of studies from many different countries show that job displacement results in large and long-lasting income losses (e.g., Jacobson, LaLonde, and Sullivan 1993; Bender, Dustmann, Margolis, and Meghir 2002; Kletzer and Fairlie 2003; Von Wachter, Weber Handwerker, and Hildreth 2009; Schmieder, von Wachter, and Bender 2010; Huttunen, Møen, and Salvanes 2011).1 However, the results are very diverse, and many of the studies have shown that the magnitude of displaced workers’ income losses varies strongly by the choice of empirical methods and data sources, by industry, and by business cycle conditions (e.g., Couch and Placzek 2010; Hijzen, Upward, and Wright 2010; Davis and von Wachter 2011). Nonetheless, researchers often find displaced workers’ income losses to be larger in the United States, where wage bargaining is decentralized, than in Europe, where many countries have centralized wage bargaining systems. Therefore, some researchers have argued that country-specific differences in displaced workers’ income losses may also reflect country-specific differences in institutions, such as wage bargaining systems (e.g., Burda and Mertens 2001; Bender et al. 2002). Yet no study has directly analyzed whether displaced workers experience larger income losses under decentralized wage bargaining systems than under centralized ones. Moreover, why displaced workers should experience larger income losses under decentralized wage bargaining remains unclear. The first contribution of this paper is to exploit a drastic reform of the wage bargaining system in the Danish manufacturing sector, to provide direct evidence for the relationship between the decentralization of wage bargaining and increasing income losses of displaced workers. The reform is ideal for studying this relationship, because it changed the wage bargaining from a centralized system to a decentralized one within a short period (e.g., Eriksson and Westergaard-Nielsen 2009). Moreover, I can rely on Statistics Denmark’s detailed register data, which allows me to study displaced workers’ income, wage, and unemployment patterns over long periods, both before and after the introduction of the decentralized wage bargaining system. The second contribution of this paper is to analyze the major channels through which the wage bargaining reform has affected the magnitude of displaced workers’ income losses. Centralized wage bargaining systems may impose rigid wage structures, such that displaced workers experience higher unemployment but lower wage losses under centralized wage bargaining systems, and experience higher wage losses but lower unemployment under decentralized ones. But such a simple model cannot explain why displaced workers’ income losses are larger under decentralized wage bargaining. Yet, a number of recent studies show that flexible wages resemble individual productivity more closely and induce greater variability in firm wage premiums (e.g., Card, Heining, and Kline 2013; Dahl, Le Maire, and Munch 2013). Thus, under decentralized wage bargaining, displaced workers’ losses of firm-specific human capital and firm wage premiums may result in larger income losses. The first contribution shows that the decentralization of the wage bargaining system is related to a substantial increase in displaced workers’ gross income losses including unemployment insurance (UI) and other social benefits. Before the reform, displaced workers’ income losses were about 1% of their average predisplacement income per year. Thereafter, displaced workers’ income losses ranged from 6% to 7% per year, and the losses appear very persistent over the long term. I provide a large number of robustness checks to ensure that neither other unrelated institutional changes nor changing macroeconomic conditions are driving the results. I show that the increase in displaced workers’ income losses persisted across changes in the business cycle. Moreover, displaced workers’ income losses did not increase in sectors that were affected by similar business cycle conditions and institutional changes but that were not exposed to a comparable change in their wage bargaining systems. The second contribution shows that the increase of displaced workers’ income losses under the decentralized wage bargaining system were largely related to higher losses in real wages, that is, wage losses increased from 0% to about 5%. In contrast, I do not find a comparable structural break in displaced workers’ unemployment patterns after the reform. Moreover, on average displaced workers reentered employment at wage levels very close to those of their predisplacement wages, both before and after the reform. These results are inconsistent with the common predictions of a simple neoclassical model, in which displaced workers trade off wage losses against unemployment. Instead, displaced workers experienced large income losses under the decentralized wage bargaining system, because they forwent wage growth, not because they initially reentered firms at lower wages. Under the centralized wages bargaining system, displaced workers appear to have benefited from wage drifts of general contracts and centralized negotiations, whereas firm-specific human capital and tenure became more important after the reform. In more detail, displaced workers who lost their jobs before the reform profited from considerable wage growth irrespective of the amount of tenure they accumulated after the job loss. In contrast, displaced workers who lost their jobs after the reform experienced much less wage growth after their job loss, but their returns to postdisplacement tenure more than doubled. Moreover, the variation in postdisplacement entry wages increased substantially after the reform, suggesting that firm-specific wage premiums and assortative matching may have become more important for displaced workers’ wage development under decentralized wage bargaining systems. However, the results also indicate that displaced workers with lower predisplacement wages refrain from reentering jobs with wages below the level of those in their previous job. In contrast, displaced workers with predisplacement wages above the median appear to have reentered employment more quickly and at relatively lower wages. As UI benefits in Denmark are relatively more generous for low- than for high-income workers, UI benefits may prevent the commonly expected trade-off between wage losses and unemployment—at least for displaced workers with lower wages. This paper contributes to the large literature on displaced workers’ income losses. Although many studies show that these income losses differ by worker characteristics, economic conditions, data sources, and countries,2 this paper is the first to provide direct evidence that labor market institutions such as wage bargaining systems determine the magnitude of displaced workers’ income losses. The paper also contributes to the literature showing that the decentralization of wage bargaining systems is related to rising skill premiums, rising returns to tenure and experience, and rising heterogeneity of firm-specific wage premiums (e.g., Card, Lemieux, and Riddell 2004; Dustmann, Ludsteck, and Schönberg 2009; Card et al. 2013; Dahl et al. 2013). This paper extends that literature by showing that displaced workers, who commonly lose firm-specific wage premiums and lack tenure, forgo substantially more wage growth under decentralized wage bargaining systems than under centralized ones. The remainder of this paper is structured as follows. Section 2 provides detailed background information on the Danish wage bargaining system and Denmark’s macroeconomic performance. Section 3 describes the estimation methods, data, and sample selection. Section 4 presents the main results, and Section 5 empirically analyzes the mechanisms to explain the results in Section 4. Section 6 provides a sensitivity analysis, and Section 7 concludes. 2. Institutions and Macroeconomic Conditions To provide the necessary background for interpreting the empirical results, this section describes the changes in the Danish wage bargaining system, the UI system, and macroeconomic performance. 2.1. Decentralization of the Wage Bargaining System Figure 1 shows the development of the Danish wage bargaining system, along with indicators of Denmark’s macroeconomic performance. The solid vertical lines in Figure 1 indicate the critical changes in wage bargaining in the manufacturing sector. Between 1956 and 1989, the Danish wage bargaining system was highly centralized and largely characterized by standard tariff wage contracts that established ranges for workers’ wages according to their occupation, education, and work experience. These contracts were not modified at the firm level, and wage floors and ceilings further limited the scope for firm-level wage bargaining. Moreover, a cost-of-living adjustment automatically tied workers’ wages to national inflation. Figure 1. View largeDownload slide Macroeconomic conditions and the wage bargaining system. The dashed line marks the national rate of registered unemployment. The solid line marks the share of standard tariff contracts in the DA/LO sector. The DA does not provide official data for standard tariff contracts for the time before 1989. However, Pedersen, Smith, and Stephensen (1998) report that 45% of all workers were covered by standard wage rate contracts before 1989. The vertical lines indicate changes in the wage bargaining system. Source: Statistics Denmark and Dansk Arbejdsgiverforening (2004) Arbejdsmakedrapport. Figure 1. View largeDownload slide Macroeconomic conditions and the wage bargaining system. The dashed line marks the national rate of registered unemployment. The solid line marks the share of standard tariff contracts in the DA/LO sector. The DA does not provide official data for standard tariff contracts for the time before 1989. However, Pedersen, Smith, and Stephensen (1998) report that 45% of all workers were covered by standard wage rate contracts before 1989. The vertical lines indicate changes in the wage bargaining system. Source: Statistics Denmark and Dansk Arbejdsgiverforening (2004) Arbejdsmakedrapport. Every two years, individual national unions and industry employer associations had the right to negotiate directly at the industry level. However, if an industry-level agreement was not reached, the Danish Federation of Trade Unions (LO) and the Danish Employer Federation (DA) were required to negotiate on behalf of their affiliates. If the LO and DA were also unable to reach an agreement, a state mediator had to mediate the dispute and make a proposal. The parliament typically enforced these proposals, even if the LO or DA rejected them (Wallerstein and Golden 1997). The cost-of-living adjustment was revoked in 1985, when the last national-level bargaining round took place (Dansk Arbejdsgiverforening (2004) Arbejdsmakedrapport). In 1987, employer organizations and unions negotiated at the industry level, agreeing to shift wage bargaining power away from the national level.3 Although the agreement was set for four years, the social partners renegotiated it after two years. Renegotiations in January 1989 marked a pivotal change in the manufacturing sector, with sector-specific negotiations between DI (a DA member organization that organizes firms in the manufacturing sector) and CO-metal (a bargaining conglomerate representing unions organizing workers employed in DI firms). These renegotiations shifted a substantial number of wage negotiations to the local level and substantially increased the scope of firm-level wage bargaining (Andersen and Svarer 2007). Although this decentralization process has affected all sectors of the Danish economy in recent decades, both unions and employer organizations in the manufacturing sector initiated this process in 1989. The metal industry’s skilled workers union, Dansk Metallarbejderforbund—traditionally the main proponent of a centralized wage bargaining system—was the first to accept strong decentralization in the late 1980s (Wallerstein and Golden 1997). Before 1989, standard wage rate contracts, for which wages were not modified at the firm level, consistently covered about 45% of the DA/LO area. Standard wage rate contracts are characterized by payment ranges based on workers’ education, occupation, and experience in the sector (Bingley and Westergaard-Nielsen 2003). Between 1989 and 1993, standard wage rate contracts became less widespread. In 1989, standard wage rate contracts covered only 34% of the DA/LO area. Between 1989 and 1993, this rate dropped to 16% and remained at this level until 2004 (see Figure 1). Between 1989 and 1993, wage arrangements leaving more room for firm-level wage bargaining became more popular. Specifically, the use of minimum wage and minimum pay systems, which represented wage floors for only very inexperienced workers, increased between 1989 and 1993 and covered nearly all remaining DA/LO firms and workers by 1993 (Andersen and Mailand 2005). Therefore, some researchers refer to the 1989–1993 period as one of coordinated decentralization within strict guidelines (Andersen 2003). Beginning in 1989, the social partners successively abandoned special guidelines such as ceilings on wage increases; instead, firm-level wage negotiations without any boundaries became more popular in the late 1990s and early 2000s, covering approximately 19% of DA/LO firms and workers by 2004.4 2.2. Unemployment Insurance In Denmark, UI is relatively generous, even in comparison to other European countries. In principle, the unemployment benefit compensates up to 90% of a worker’s previous wage. However, a relatively low cap of approximately 170,000 Danish kroners (DKKs) (e.g., 10,000 correspond to $1,160) restricts total compensation such that UI benefits are less generous for high-wage workers. In addition to common UI benefits, a specific “guarantee fund” compensates workers who forgo earnings if their employers are bankrupt or must lay them off for economic reasons. This fund guarantees ongoing salaries during the worker’s notice period, which commonly lasts up to three months.5 Most displaced workers who separate from closing firms are likely to receive payments from this fund. 2.3. Macroeconomic Performance Denmark experienced two economic downturns with high unemployment (see the dashed line in Figure 1). The first recession, in the early 1980s, was followed by a brief recovery period. In the late 1980s and the early 1990s, Denmark was hit by the “Nordic crisis” (i.e., a financial crisis in Denmark, Finland, Norway, and Sweden) (e.g., Kiander 2004). However, during the mid-1990s, Denmark experienced a phase of substantial economic growth and steadily decreasing unemployment (Kiander 2004). As the solid vertical lines in Figure 1 indicate, the major reform of the wage bargaining system took place at the beginning of a deep recession but remained flexible for the rest of the observation period, during which the economy recovered substantially. 3. Data For my investigation, I use the Integrated Database for Labor Market Research (IDA), a highly precise administrative data source that covers each worker and establishment in Denmark between 1980 and 2004 (e.g., Abowd and Kramarz 1999; Iversen, Malchow-Møller, and Sørensen 2010; Timmermans 2010). The Danish Bureau of Statistics collects administrative records from different register sources and matches the data, using a unique identification number for each individual and establishment. Every Danish resident is thus tracked and assigned to one—and only one—labor market state and work establishment on a specific date in November of each year. The data source contains a variety of worker and firm characteristics, including income, earnings, and firm size. 3.1. Identification of Worker Displacement As in most other register-based data sources, I cannot observe whether workers are displaced or leave their jobs for other reasons. Therefore, I follow the most common solutions to this problem by creating samples of workers separating from firms during mass layoffs and samples of comparable workers who do not separate from firms during mass layoffs (e.g., Jacobson et al. 1993; Couch and Placzek 2010; Davis and von Wachter 2011). In accordance with those previous studies, I define a mass layoff as a situation in which a firm’s workforce contracts by more than 30% from one observation year to the next. To distinguish true mass layoffs from regular employment volatility, I focus on firms with an employment of at least 50 employees before the mass layoff occurs. As in Jacobson et al. (1993), I incorporate closing firms with more than 50 employees. For many register data sources, researchers face the problem of misinterpreting simple changes in establishment identification numbers or transfers of large parts of the firm’s workforce to other buildings as mass layoffs. To avoid this confusion, the IDA provides prospective information on the status of each establishment (Timmermans 2010). Specifically, I have information for each successive year on whether an establishment (1) remains the same, (2) closes down, (3) merges with or is acquired by an existing establishment (i.e., either 30% or at least two workers move to the same other establishment), (4) closes down due to the transfer of its buildings to a new entering establishment, (5) moves to self-employed individuals without any employees, or (6) moves into a situation in which the firm has no employees. From this information, I apply the following additional restrictions to define a mass layoff: (1) Firms that experience a mass layoff but remain in business must remain the same during the mass layoff period and for at least three preceding observation periods (i.e., I avoid considering firms that transfer part of their workforce to other establishments); and (2) firms that experience mass layoffs but do not remain in business must go out of business when the separation takes place but must have remained the same for the three preceding observation periods.6 3.2. Displaced Workers Following the previous literature, my sample of displaced workers includes male manufacturing workers, 50 years or younger at the time of their displacement, who separate permanently from their establishments. I require displaced workers to have at least two years of tenure before their displacement occurs. During those two years, displaced workers must have been wage employed, that is, registered as wage employed for 365 days in each of the two predisplacement years. This restriction excludes seasonal workers and ensures that I do not mistakenly assign a share of the displacement loss to the year preceding the displacement. After the event of displacement, I apply no further restrictions, so the displaced workers may be employed, unemployed, or even out of the labor force after their job loss. 3.3. Nondisplaced Workers To keep the sample of nondisplaced workers most comparable to the sample of displaced workers, these nondisplaced workers consist of male manufacturing workers who are 50 years or younger when their respective counterfactual group of displaced workers was displaced. They work for firms with more than 50 employees. Throughout each respective observation period, nondisplaced workers were never laid off in response to a mass layoff. More specifically, the group of nondisplaced workers consists of both those who work in firms that never underwent a mass layoff event and those who work in firms that underwent a mass layoff event but were not laid off during it. I require that nondisplaced workers have had at least three years of tenure—that is, two years before their counterfactual group of displaced workers lose their jobs and an additional year because they do not switch jobs in the displacement year. Thereafter, I allow nondisplaced workers to switch jobs, but they must have had wage employment with a positive income for each subsequent year. This restriction ensures that nondisplaced workers have had a continuous employment pattern.7 The first column of Table C.1 in the appendix presents results with less restrictive sample selection criteria, but the key results of this paper remain very robust. 3.4. Variables The analysis relies on four main dependent variables. The first measures workers’ annual taxable gross income. This gross income measure captures a worker’s entire taxable income, including payments from UI or pension funds and other types of income. Notably, the measure covers income from the “guarantee fund”. The second variable measures workers’ annual labor earnings. I calculate this variable adding the workers’ November job earnings and the sum of their wages from all other jobs in a given year. This variable provides information on workers’ total displacement costs in the form of pure labor earnings, which do not include any form of unemployment or state-financed benefits. Because recent studies specifically note that zero earnings or income are an important part of workers’ displacement costs (e.g., Davis and von Wachter 2011), I specifically include displaced workers with zero income or earnings after displacement. Both income and labor earnings are deflated and measured in 2000 DKKs. The third dependent variable measures the workers’ hourly wage rate. Following the previous literature, I calculate this variable by dividing the sum of total labor income and mandatory pension fund payments by the total number of hours worked in a given year (e.g., Dahl et al. 2013). Statistics Denmark provides additional information on the quality of the measured wage rate. From this information, I set all wage observations of poor quality to “missing”. Moreover, to disentangle the wage effect from the quantity effect, which arises as a consequence of unemployment or the reduction in working time, I exclude observations with zero wages. The fourth variable measures a worker’s degree of unemployment in percent of a given year. A high replacement ratio and almost no experience rating for employers or workers implies that workers experience many short periods of unemployment. Moreover, employees have the right to claim supplementary UI benefits while on vacation, so that UI periods appear to cluster around summer and Christmas. To avoid capturing such unemployment periods not related to effective unemployment, I exclude the UI periods of workers who were registered as wage employed for the entire year.8 Finally, I have information about workers’ age, education, industry, and real—not potential—work experience since 1964. 4. Methods The main problem in estimating the cost of worker displacement is that researchers have no information on how displaced workers’ income would have developed in the absence of their displacement. A common solution to this problem is to create samples of displaced workers and compare their careers to a counterfactual group of comparable nondisplaced workers. This paper also follows the counterfactual approach by applying versions of the following regression model, which follows Jacobson et al. (1993) and represents best practices in the literature. Equation (1) is the regression equation:   \begin{equation} y_{it}=\alpha _{i}+\lambda _{t}+x_{it}\beta +\sum _{k}D_{it}^{k}\delta ^{k}+\varepsilon _{it}, \end{equation} (1)where i indicates a single worker, and t represents the calendar time in years. $$D_{it}^{k}$$ is a set of dummy variables equal to 1 in the kth year before or after the worker’s displacement, for example, $$D_{i1985}^{2}$$ equals 1 for displaced workers who lost their jobs in 1983. Thus δk represents the effect of displacement in the kth year before or after displacement. For nondisplaced workers, $$D_{it}^{k}$$ is always 0. The vector xit contains a set of control variables, which I restrict to five age categories ranging from under 25 to over 55 years of age. Following previous papers, such as Davis and von Wachter (2011), I incorporate only age as a control variable, because observable worker and job characteristics other than age are either time constant (e.g., predisplacement education) or endogenous to the displacement itself (e.g., tenure, work experience, or industry). The parameter λt represents a set of coefficients for each year in the observation period. The fixed effect ai summarizes all observable and unobservable time constant differences of workers i. I assume εit to have constant variance and to be uncorrelated with individual and time effects. The dependent variable yit represents the dependent variable of worker i at time t. yit can be the workers’ gross income (including UI benefits and other social benefits), their annual labor earnings (excluding all benefits), their hourly wages, or their incidence of unemployment in a given year. Standard errors are clustered at the individual level. All monetary outcomes are measured in absolute DKK values, and I include zero income and earnings in yit, so that I cannot measure the dependent variable in log values. However, I provide figures that express monetary displacement losses as a percentage of the workers’ base income, earnings, or wages throughout the paper. Therefore, I divide the δk by the average income, earnings, or wages of all displaced workers in the second year before the displacement event. The figures also contain confidence intervals, which I obtained using the delta method.9 Tables A.3–B.1 in the appendix present monetary displacement losses in absolute values. I estimate equation (1) on 19 separate equidistant and overlapping panels, one for each cohort of displaced workers between 1982 and 2000. Each subsample observes each displaced worker for two years before and three years after displacement. For example, the first sample includes all workers displaced in 1982, and I observe this group for six years (from 1980 to 1985); the second sample includes all workers displaced in 1983, and I observe this group for six years (from 1981 to 1986), and so forth. This approach allows me to present a very desegregated picture of the development of displaced workers’ income losses over time, and each subsample follows workers of the treatment and control group for the same amount of time before and after the job loss. Moreover, to provide ideal conditions for unbiased and comparable estimates of δks within each equidistant subsample, the number of observation is large in comparison to the number of time periods for each sample.10 However, analyzing the costs of worker displacement in separated samples does not allow me to test for significant differences in displaced workers’ income losses across the years. Moreover, I am unable to capture global trends in λt, which may finally account for the increase in displaced workers’ income losses. Thus I estimate regression equations of the following form on a pooled sample containing all individuals of all equidistant samples11:   \begin{equation} y_{it}=\alpha _{i}+\lambda _{t}+ x_{it}\beta + \mathit{ Reform}\cdot x_{it}\beta _{R} + \sum _{k}D_{it}^{k}\delta ^{k}+\sum _{k}\mathit{ Reform} \cdot D_{it}^{k}\delta _{R}^{k}+\varepsilon _{it}, \end{equation} (2)where $$ {Reform}$$ is a dummy being one if a worker was displaced after the reform of the decentralized wage bargaining system and zero otherwise. The $$\delta _{R}^{k}$$s directly measure the effect of the wage bargaining reform. As the $$ {Reform}$$ dummy is multicolinear to the time dummies λt and time constant for most displaced workers,12 I cannot include the $$ {Reform}$$ dummy separately, that is, λt and αi jointly pick up the isolated $$ {Reform}$$ effect. I emphasize here that the empirical approach in (2) relies on a combined sample that includes members of the treatment and control group from different periods into one regression. 5. Main Results This section presents the main results. Section 5.1 shows descriptive statistics for the sample composition of displaced and nondisplaced workers. Section 5.2 presents the main results. 5.1. Sample Composition and Baseline Characteristics Tables A.1 (pre-decentralization) and A.2 (post-decentralization) in the appendix present average statistics for the main analysis variables of all 19 subsamples, with t-statistics to test for significant differences between displaced and nondisplaced workers. The number of displaced workers (according to the definition in the previous section) ranges between 230 individual workers in sample “1984–1989” and 1,038 workers in sample “1986–1991”. The number of nondisplaced workers ranges from 52,699 in sample “1985–1990” to 61,135 in sample “1996–2001”. Baseline characteristics were not fully balanced between displaced and nondisplaced workers. Even two years before the displacement, displaced workers had somewhat lower income than their nondisplaced counterparts. Moreover, significant differences exist for age and education. These differences are very comparable to the differences of predisplacement characteristics in other studies (e.g., Jacobson et al. 1993; Korkeamäki and Kyyrä 2014).13 However, the decentralization of the wage bargaining system fell in the middle of an economic downturn that may have caused a major shift in economic activities across Danish manufacturing sectors (see Figure 1 of Section 2). Moreover, firms may tend to lay off more high-ability workers when wages are rigid but fewer of them when wages are flexible. As a result, the sample composition of displaced workers may have changed substantially after the introduction of the decentralized wage bargaining system in 1989. Figures 2 and 3 present descriptive evidence of the sample composition of displaced and nondisplaced workers by showing workers’ industry and education over time. Figure 2 describes the composition of workers’ education within the group of displaced workers (black dots) and nondisplaced workers (crosses). The first subfigure presents the fraction of low-educated workers with less than an apprenticeship degree. The second subfigure presents the fraction of medium-educated workers with an apprenticeship degree or comparable educational courses. The third subfigure presents the fraction of high-educated workers with a master craftsman degree or a university degree. The x-axis indicates the displacement year of the respective subsample, and the solid lines present the results of kernel-weighted local polynomial regressions. Figure 2. View largeDownload slide Composition of education over time. The figure presents descriptive statistics for the fraction of displaced and nondisplaced workers who have a low, medium, or high education. Low-educated workers have less than an apprenticeship degree, medium-educated workers have an apprenticeship degree, and high-educated workers have a master craftsman or a university degree. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Figure 2. View largeDownload slide Composition of education over time. The figure presents descriptive statistics for the fraction of displaced and nondisplaced workers who have a low, medium, or high education. Low-educated workers have less than an apprenticeship degree, medium-educated workers have an apprenticeship degree, and high-educated workers have a master craftsman or a university degree. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Figure 3. View largeDownload slide Composition of manufacturing industries over time. The figure presents descriptive statistics for the fraction of displaced and nondisplaced workers in four different industry categories before the displacement. The category “other mfr. industries” contains industrial workers in textiles and leather, chemicals and plastic products, other nonmetallic mineral products, furniture, and other small industries. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Figure 3. View largeDownload slide Composition of manufacturing industries over time. The figure presents descriptive statistics for the fraction of displaced and nondisplaced workers in four different industry categories before the displacement. The category “other mfr. industries” contains industrial workers in textiles and leather, chemicals and plastic products, other nonmetallic mineral products, furniture, and other small industries. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Even if significant differences between displaced and nondisplaced workers existed, Figure 2 does not indicate a systematic break in the educational composition of displaced and nondisplaced workers immediately after 1989. Both groups largely consist of medium-educated workers, followed by low- and high-educated workers. Figure 3 shows the composition of workers’ industries over time. For this purpose, I assigned all workers to four categories according to their predisplacement industry within the manufacturing sector. The first three categories correspond to those three manufacturing industries with the largest fractions of displaced workers in the sample: (1) food, beverages, and tobacco; (2) basic metals and fabric metal products; and (3) wood products and printing and publishing materials. The fourth category (4) represents all remaining manufacturing industries with fewer displaced workers, such as textiles and leather, chemicals and plastic products, other nonmetallic mineral products, and furniture. The y-axis of each subfigure indicates the fraction of workers within the respective industry category separately for the group of displaced workers (black dots) and nondisplaced workers (crosses). The relative industry composition of nondisplaced workers remains relatively stable over time, suggesting no major shifts in the manufacturing sector throughout the observation period. In contrast, the industry composition of displaced workers varies substantially over time. Moreover, the industry composition of displaced and nondisplaced workers differs significantly (see Tables A.1 and A.2). These results show that mass layoffs are not a smooth process occurring at constant rates within and across industries. However, the figure shows no evidence that the industry composition of displaced workers changed systematically immediately after the introduction of the decentralized wage bargaining system. 5.2. Decentralization and Displaced Workers’ Income Losses Figure 4 shows average real income developments (including UI and other social benefits) for displaced and nondisplaced workers for each sample. Each subfigure corresponds to a separate subsample of workers, indicated by the title of the subfigure. All subfigures follow displaced and nondisplaced workers from two years before their displacement occurred until three years after the displacement occurred (i.e., 0 on the respective x-axis denotes the displacement year). For example, the first subfigure (“1980–1985”) depicts average gross income developments for workers who were displaced in 1982, along with the income developments of their comparison group of nondisplaced workers. To keep subfigures comparable and to account for predisplacement differences between displaced and nondisplaced workers, I measure all income trajectories relative to the average income in the second year before the displacement occurred. Figure 4. View largeDownload slide Average gross income developments (measured in thousands). Each subfigure refers to a sample of displaced and nondisplaced workers for the assigned year. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Gross income is measured in 2000 DKKs and includes zeros. Displacements occurred between November in year (−1) and November in year (0). Average income is adjusted to a base income in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). Figure 4. View largeDownload slide Average gross income developments (measured in thousands). Each subfigure refers to a sample of displaced and nondisplaced workers for the assigned year. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Gross income is measured in 2000 DKKs and includes zeros. Displacements occurred between November in year (−1) and November in year (0). Average income is adjusted to a base income in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). Figure 4 shows that displaced workers who lost their jobs between 1982 and 1989 (i.e., workers in the “1980–1985” to the “1986–1991” samples) did not experience substantial income losses relative to their comparison groups of nondisplaced workers. However, after 1989—the year of the pivotal change in the wage bargaining system (sample “1987–1992”)—a clear structural break in the results appears. Displaced workers who lost their jobs after 1989 had worse income developments than nondisplaced workers, that is, displaced workers experienced stagnating or even negative income growth after their displacement. In other words, relative to nondisplaced workers, displaced workers who lost their jobs after 1989 experienced visible income losses. Tables A.3 and A.4 (which, because of their length, are in the appendix) present the corresponding estimation results of equation (1). Figure 5 illustrates the results of both tables by calculating the discounted present values (DPVs) of displaced workers’ income losses for each displacement from these tables. The DPVs express the workers’ income losses as the number of income years lost at the previous level of income for each single displacement year τ between 1982 and 2000.14 Figure 5. View largeDownload slide Discounted present values of income losses (DPV) and wage bargaining system. The figure presents average DPVs (interest rate 5%) of displaced workers’ income losses over the years. Gross income is measured in 2000 DKKs and includes zeros. Displacements occur between November in year (−1) and November in year (0). Average DPVs of displacement losses are expressed as the number of income years lost at the level of income two years before the displacement. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Figure 5. View largeDownload slide Discounted present values of income losses (DPV) and wage bargaining system. The figure presents average DPVs (interest rate 5%) of displaced workers’ income losses over the years. Gross income is measured in 2000 DKKs and includes zeros. Displacements occur between November in year (−1) and November in year (0). Average DPVs of displacement losses are expressed as the number of income years lost at the level of income two years before the displacement. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Figure 5 shows a substantial increase in the DPVs of displaced workers’ income losses after the wage bargaining system was decentralized. Before 1989, displaced workers lost between 0 and 0.05 years of their predisplacement income, that is, less than one month’s income over a period of four years. In contrast, displaced workers lost up to 0.20 years of their predisplacement income, which amounts to two and a half times their monthly income over a period of four years. This difference is substantial, although even after 1989, Danish income losses appear smaller than those in the United States. Thus the estimation results clearly support the descriptive results in Figure 4. The structural break in the findings for displaced workers’ income losses precisely corresponds to the structural break reported in earlier investigations of wage distributions conducted by Danish researchers. For example, as previously mentioned, Eriksson and Westergaard-Nielsen (2009) show that the dispersion of the wage structure suddenly increased after 1989, and Bingley and Westergaard-Nielsen (2003) show that returns to tenure more than doubled at the beginning of the decentralization period in 1989. Figure 6 presents estimations on a pooled sample that includes all individuals from all subsamples. The estimation approach follows equation (2) with interaction terms between an indicator for the reform and the displacement indicators $$D_{it}^{k}$$ and interaction terms between the reform indicator and all remaining control variables. Figure 6. View largeDownload slide Average real gross income losses (pooled sample). Estimation according to equation (2). Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Gross income is measured in 2000 DKKs and includes zeros. Average income is adjusted to a base income in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). Regressions include worker fixed effects, year fixed effects, and age categories. The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 6. View largeDownload slide Average real gross income losses (pooled sample). Estimation according to equation (2). Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Gross income is measured in 2000 DKKs and includes zeros. Average income is adjusted to a base income in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). Regressions include worker fixed effects, year fixed effects, and age categories. The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 6(a) presents the results as percentages of displaced workers’ base income for the standard specification without further macroeconomic controls. The absolute values of workers’ gross income losses appear in column (1) of Table B.1. Displaced workers who lost their jobs before the wage bargaining reform lost only between 290 DKK ($33) and 2060 DKK ($238) per year. In real terms, these income losses correspond to between 0.09% and 0.5% of their average predisplacement income (dashed line). At conventional significance levels, their income losses are not significantly different from zero. In contrast, displaced workers who lost their jobs after the reform lost between about 1,471 ($170) and 16,549 DKK ($1915) more per year. These differences are highly significant. Measured in relation to their predisplacement income, their income losses correspond to losses of between 6% and 7% of their predisplacement income (solid line). Of course, these estimations only present observable monetary displacement losses. In addition, unemployed workers may incur losses because they forgo their access to fringe benefits such as employer-provided pensions or on-the-job training. Therefore, total displacement losses may even be larger. However, these monetary loss findings are very similar to those for displaced workers’ income losses in other Nordic countries. For example, Huttunen et al. (2011) find income losses of about 5% for displaced workers in Norway. Moreover, as in other European studies, displacement losses do not peak during the displacement year but become significant only in the first year after the displaced workers lost their jobs. Previous studies such as Davis and von Wachter (2011) show that displaced workers’ income losses strongly relate to macroeconomic conditions such as national unemployment levels. Moreover, the decentralization of the Danish wage bargaining system occurred between 1989 and 1993, when the Danish economy experienced a substantial economic downturn. Therefore, Figure 6(b) divides the post-decentralization period after 1989 into two separate periods: one for the recession period between 1989 and 1993, and one for the expansion period between 1994 and 2000. Thus, instead of interacting the displacement dummies with a single dummy for the reform, I further split the reform dummy to represent the post-decentralization downturn period (“1989–1993”) and the post-decentralization recovery period (“1994–2000”) separately. Although income losses were on average slightly larger for workers who lost their jobs during the “1989–1993” Nordic crisis, the differences between the downturn and recovery period are negligible. Overall, the results allow me to draw two main conclusions. First, displaced workers’ income losses increased substantially after the reform of the decentralized wage bargaining system. Second, in general the “1989–1993” Nordic crisis appears not to account for those increases. Section 7 and Appendices A.1 and A.2 provide a large number of additional robustness checks showing that neither other unrelated institutional changes nor macroeconomic conditions are likely to explain the results. Moreover, a separate analysis in Appendix A.1 shows that displaced workers’ earnings losses also increased substantially after the reform. Yet in contrast to displaced workers’ income losses, earnings losses were more sensitive to changes in the business cycle, suggesting that generous UI benefits and large payments of the “guarantee fund” smooth income losses for displaced workers. Finally, I show that displaced workers’ income losses did not increase in other sectors that were affected by similar business cycle conditions but that were not exposed to a change in the wage bargaining system (Section 7.1). 6. Understanding the Relationship between the Decentralization of Wage Bargaining and Displaced Workers’ Income Losses The most common first intuition about the relationship between displaced workers’ income losses and the decentralization of wage bargaining may follow the idea of a very simple neoclassical model on the trade-off between wage losses and unemployment. Such a simple model describes (a) centralized wage bargaining systems by tariff contracts that impose downward rigid wage structures, which prevent displaced workers from obtaining jobs at their optimal reservation wages; and (b) decentralized wage bargaining systems by flexible wage structures that allow all efficient worker-firm matches to occur. This simple model implies that displaced workers on average experience higher unemployment but lower wage losses under centralized wage bargaining systems, and experience higher wage losses but lower unemployment under decentralized ones. However, such a simple model does not necessarily imply that displaced workers’ income losses should increase in response to the decentralization of the wage bargaining system.15 In contrast, the following three arguments suggest that displaced workers may experience larger income losses under decentralized wage bargaining systems. First, studies such as Card et al. (2013) and Dahl et al. (2013) have shown that flexible wages under decentralized wage bargaining resemble individual productivity more closely and induce a greater variability in firm wage premiums. Second, under decentralized wage bargaining, firms have more scope for designing incentive wage plans, such as seniority wage profiles and tournaments. Third, generous UI benefits and a more dispersed wage offer distribution may even provide incentives to prolong the job search under decentralized wage bargaining (see, e.g., Lazear 1986, 2012; Mortensen 1986). The following sections analyze displaced workers’ wage and unemployment patterns more closely to provide evidence for these arguments. 6.1. Hourly Wages and Unemployment Figure 7 decomposes displaced workers’ income losses into wage losses and unemployment. Figure 7. View largeDownload slide Average wage losses and unemployment (pooled sample). Estimation according to equation (1). Dep. vars.: (a) hourly wages and (b) degree of unemployment. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs, excluding zeros. Degree of unemployment measures the fraction of a year a worker received UI benefits. Average income is adjusted to a base income in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 7. View largeDownload slide Average wage losses and unemployment (pooled sample). Estimation according to equation (1). Dep. vars.: (a) hourly wages and (b) degree of unemployment. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs, excluding zeros. Degree of unemployment measures the fraction of a year a worker received UI benefits. Average income is adjusted to a base income in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 7(a) shows hourly wage losses for displaced workers who lost their jobs both before (dashed line) and after (solid lines) the decentralization of the wage bargaining system. I obtained the results by estimating equation (2) with the hourly wage rate as the dependent variable. In contrast to the previous estimations, for which I included zero income, I excluded observations with zero wages to capture the pure price mechanism in terms of displaced workers’ wage losses. For the post-decentralization period, the figure further distinguishes between workers who were displaced during the 1989–1993 recession and workers who were displaced during the recovery period after 1994 (solid line with crosses). Figure 7(a) shows that wage losses were small before the wage bargaining reform; indeed, those wage losses were virtually zero. In contrast, displaced workers who lost their jobs after the reform experienced significant wage losses of between 2% and 4%, including those who lost their jobs during the recovery period after 1994. Comparing the magnitude of these wage losses to the magnitude of income losses, which ranged between 2% and 8% (see Figure 6), suggests that a relatively large fraction of the increase in displaced workers’ income losses is related to an increase in actual wage losses. Figure 7(b) shows comparable results for displaced workers’ degree of unemployment in a given year. The results stem from a regression of equation (2), where the dependent variable measures the degree of unemployment in a given year. Before the reform, the degree of unemployment was on average between two and three percentage points larger for displaced workers than for nondisplaced workers (i.e., approximately between 7 and 10 days of a year). After the reform, the degree of unemployment increased for displaced workers who lost their jobs during the recession, that is, their average degree of unemployment was about three to nine percentage points larger than that of comparable nondisplaced workers. In contrast, the results for displaced workers who lost their jobs during the recovery period after 1994 are very similar to the results for displaced workers who lost their jobs before the reform of the wage bargaining regime. Table 1 analyzes the degree of displaced workers’ unemployment in more detail. The first row presents the fraction of displaced workers whose degree of cumulative unemployment never exceeded 20% of a given year during their first four years after their job loss. The second through fourth rows show the fraction of displaced workers whose degree of cumulative unemployment was at least in one year (2) between 20% and 50%, (3) between 50% and 80%, or (4) larger than 80%. Table 1. Largest degree of unemployment (displaced workers only).     Post-decentralization    Pre-decentralization  Recession  Recovery  Below 20%  0.869  0.739  0.841  Between 20% and 50%  0.074  0.122  0.094  Between 50% and 80%  0.034  0.066  0.041  Larger 80%  0.023  0.074  0.024  Observations  9241          Post-decentralization    Pre-decentralization  Recession  Recovery  Below 20%  0.869  0.739  0.841  Between 20% and 50%  0.074  0.122  0.094  Between 50% and 80%  0.034  0.066  0.041  Larger 80%  0.023  0.074  0.024  Observations  9241      Note: Degree of unemployment measures the fraction of a year a worker received UI benefits. The rows represent fractions of displaced workers whose degree of unemployment never exceeded (1) 20%, was at least once (2) between 20% and 50%, (3) between 50% and 80%, or (4) larger than 80%. Sample is restricted to all displaced workers with valid wage observations before and after their job loss. Displacements occur between November in year (−1) and November in year (0). View Large During the first four years after the job loss, the majority of displaced workers never received UI benefits for more than 20% in any of the four years. However, the statistics clearly show that displaced workers who lost their jobs during the recession between 1989 and 1994 were more likely to experience long-term unemployment than those who lost their jobs either before or after. Overall, the short-term results on unemployment indicate no structural break in unemployment (around the reform of the wage bargaining system) that would be consistent with displaced workers’ income or wage patterns. I find no evidence that overall unemployment consistently increased after the reform or that unemployment decreased, as a simple model on the trade-off between wage losses and unemployment would predict. Before and after the reform, the majority of displaced workers appear to have experienced short periods of unemployment. This result is in line with previous evidence such as Stevens (1997). However, displaced workers who lost their jobs during the recession were much more likely to experience long-term unemployment. If rigid wages under centralized wage bargaining systems would predominantly prevent displaced workers from finding jobs at their optimal reservation wages, the decentralization of the wage bargaining system should be visible in displaced workers’ short-term wage and employment patterns. In contrast, displaced workers may forgo wage losses in the long run if they end up on a different career path in response to the displacement. Therefore, Figures 8 and 9 present long-term analyses for displaced workers’ wage losses and their degree of unemployment, and follow workers for up to 14 years after their job loss. As the data source covers a finite period until 2003, I cannot observe displaced workers from each displacement cohort for the same amount of time after the job loss. Therefore, the coefficient estimates for the later postdisplacement periods are estimated from a different population than the coefficient estimates for the earlier postdisplacement periods. Thus for displaced workers who lost their jobs during the recovery period after 1994, the long-term results must be interpreted with some caution. Figure 8. View largeDownload slide Long-term wage losses (pooled sample). Estimation according to equation (1). Dep. vars.: hourly wages. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs, excluding zeros. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 8. View largeDownload slide Long-term wage losses (pooled sample). Estimation according to equation (1). Dep. vars.: hourly wages. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs, excluding zeros. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 9. View largeDownload slide Long-term unemployment (pooled sample). Estimation according to equation (1). Dep. vars.: degree of unemployment. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs, excluding zeros. Displacements happened between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 9. View largeDownload slide Long-term unemployment (pooled sample). Estimation according to equation (1). Dep. vars.: degree of unemployment. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs, excluding zeros. Displacements happened between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. The solid line in Figure 8 shows that displaced workers who lost their jobs before the reform did not experience wage losses under the direct influence of the rigid wage bargaining system but that they experienced small long-term wage losses in the long run (about 10 years after the job loss). However, displaced workers who lost their jobs after the introduction of the decentralized wage bargaining system experienced much larger short- and long-term wage losses—including those displaced workers who lost their jobs during the recovery period after 1994 (solid line with crosses in Figure 8(b)). Therefore, displaced workers who lost their jobs after the introduction of the decentralized wage bargaining system forwent wage growth in the short and the long run. In contrast, the results suggest that workers who were displaced before 1989 appear to have had the chance of stabilizing their career path before entering the post-decentralization period. Thus even if they also forwent wage growth after the implementation of the reform, their total wage losses were somewhat smaller. Figure 9 shows comparable long-term results for displaced workers’ degree of unemployment. The long-term unemployment pattern of those displaced workers who lost their jobs before the reform follows an inverted u-shape. More specifically, they have experienced the largest degree of unemployment between the 7th and 11th year after their job loss, that is, during the recession after the 1989 reform. Displaced workers who lost their jobs during the recession experienced even higher levels of unemployment—but only in the short run. In the later years after their job loss (when this group entered the period of economic recovery), their degree of unemployment decreased (solid line). Finally, although displaced workers who lost their jobs during the recovery period after 1994 experienced somewhat larger unemployment in the short run, their degree of unemployment decreased substantially thereafter (solid line with crosses). Overall, the results show that the degree of displaced workers’ unemployment is closely related to the general economic development in Denmark. However, unlike the results for wages and income, the results for unemployment do not show a clear structural break around the reform in 1989. Thus the results suggest that the sudden increase in displaced workers’ income losses is largely related to forgone wage growth, whereas generous UI benefits may have compensated displaced workers for their earnings losses from unemployment (see also Appendix A.1). 6.2. Forgone Wage Growth The movement away from the centralized wage bargaining system was also a movement away from rigid wage scales that determined workers’ wages based on observable characteristics such as age, education, or experience in the trade. Moreover, before the reform, unions with a preference for social equality were highly involved in the wage bargaining, so that low-productive workers may have received relatively higher wages (Freeman 1980; Card et al. 2004). As a result, even displaced workers who lose firm-specific human capital were more likely to have received wages similar to those of nondisplaced workers. In contrast, under decentralized wage bargaining, wages are more in accordance with individual productivity and local conditions. Moreover, if wage bargaining is decentralized, employers may have more scope for designing incentive wage plans, such as seniority wage profiles and tournaments, and steep wage profiles may also reflect employers’ learning about unobservable worker characteristics. Thus, after the reform, firm-specific human capital investments and careers in internal labor markets may have become more important in determining individual wage growth. In contrast, involuntary job losses and even short periods of unemployment may have resulted in large forgone wage growth. Table 2 analyzes in more detail the career-related wage growth of displaced workers after their job loss. The table presents simple regressions in the following form:   \begin{equation} \mathit{ lnw}_{it}^{d}=\alpha _{it}^{d}+ \delta _{1} f(\mathit{ time})_{it}^{d}+\delta _{2} {{\mathit{ Tenure}}}_{it}^{d}+\varepsilon _{it}^{d}, \end{equation} (3)where $$\mathit{ lnw}_{it}^{d}$$ is the displaced workers’ log hourly wages. $$f(\mathit{ time})_{it}^{d}$$ is a simple linear trend of the elapsed time since the displacement, and $${\mathit{ Tenure}}_{it}^{d}$$ measures their postdisplacement tenure. More specifically, $${{\mathit{ Tenure}}}_{it}^{d}$$ measures the cumulative time a displaced worker has worked for his current postdisplacement firm, that is, $${\mathit{ Tenure}}_{it}^{d}$$ does not measure postdisplacement experience. $$\alpha _{it}^{d}$$ is a worker fixed effect. The ds indicate that, to directly analyze displaced workers’ pre–post-displacement wage differences, this analysis contains only displaced workers. The first column of Table 2 shows the results for displaced workers who lost their jobs before the reform; the second column, for displaced workers who lost their jobs afterwards. Table 2. Returns to postdisplacement time and tenure (displaced workers only).   Pre-decentralization  Post-decentralization  Time since displacement ($$f(\mathit{ time})_{it}^{d}$$)  0.027***  −0.003    (0.002)  (0.003)  Tenure since displacement (Tenure$$_{it}^{d}$$)  0.002  0.012***    (0.002)  (0.003)  Observations:  14428  18921    Pre-decentralization  Post-decentralization  Time since displacement ($$f(\mathit{ time})_{it}^{d}$$)  0.027***  −0.003    (0.002)  (0.003)  Tenure since displacement (Tenure$$_{it}^{d}$$)  0.002  0.012***    (0.002)  (0.003)  Observations:  14428  18921  Notes: Author’s calculations with data from Statistics Denmark. Dep. var.: Log hourly wages. $$f(\mathit{ time})_{it}^{d}$$ is a linear time-trend measuring time since displacement. Hourly wages are measured in 2000 DKKs excluding zeros. Tenure$$_{it}^{d}$$ measures tenure since displacement. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Sample is restricted to all observations of displaced workers with valid wage observations. Column (1) shows the results for displaced workers who lost their jobs before the wage bargaining reform. Column (2) shows the results for displaced workers who lost their jobs after the reform. Displacements occur between November in year (−1) and November in year (0). Regressions include worker fixed effects, and age categories. Standard errors are in parentheses and clustered at the individual level. ***Significant at 1% level. View Large If displaced workers who lost their jobs before the wage bargaining reform benefited mainly from wage drifts of general contracts and centralized negotiations, we should expect a strong positive effect of the general time trend ($$f(\mathit{ time})_{it}^{d}$$), whereas the coefficient estimate of tenure should be small ($${\mathit{ Tenure}}_{it}^{d}$$). In contrast, if displaced workers who lost their jobs after the reform forwent wage growth, because they lost firm-specific human capital and career-related rents, tenure should become more important than the general time trend. The results of Table 3 indeed confirm this hypothesis. The first column shows that the coefficient estimate of the general time trend (δ1) is about three percentage points and highly significant, whereas the coefficient estimate on tenure (δ2) is only 0.3 percentage points. The second column shows a coefficient estimate of the general time trend that is essentially zero and insignificant, whereas the coefficient estimate of tenure is about 1 percentage point and highly significant. Table 3. Average pre–post-displacement wage differences (displaced workers only).   Pre-decentralization  Post-decentralization  All  0.029  −0.003  Low-wage workers  0.071  0.063  High-wage workers  −0.014  −0.070  Observations  3884  5189    Pre-decentralization  Post-decentralization  All  0.029  −0.003  Low-wage workers  0.071  0.063  High-wage workers  −0.014  −0.070  Observations  3884  5189  Notes: Author’s calculations with data from Statistics Denmark. Dep. var.: Log hourly wages. Hourly wages are measured in 2000 DKKs, excluding zeros. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Sample is restricted to all observations of displaced workers with valid wage observations. Column (1) shows the results for displaced workers who lost their jobs before the wage bargaining reform. Column (2) shows the results for displaced workers who lost their jobs after the reform. Displacements occur between November in year (−1) and November in year (0). View Large Even after the reform, returns to tenure were relatively small, a result consistent with the findings of the previous literature (e.g., Eriksson and Westergaard-Nielsen 2009). However, the results suggest that forgone returns to tenure may not be the only explanation for the increase of wage losses after the reform. Another potential explanation is that voluntary job shopping became more important in response to the more dispersed wage distribution after the reform. The next section analyzes this argument in more detail. 6.3. Wage Dispersion As previously mentioned, a number of studies find that the dispersion of wages increased after the 1989 wage bargaining reform in Denmark. Therefore, voluntary job shopping may have become a more important determinant for workers’ wage development. Thus if job displacements force workers to leave their path of lucrative voluntary job shopping, they may forgo more wage growth under decentralized wage bargaining systems. In addition, displaced workers may have incentives for prolonging their job search, because their option value of obtaining very high wages in the future may become relatively larger than their costs of unemployment. As a result, displaced workers may reject early job offers with low wages, because they now expect to find a job with higher wages in the future. This result is central to matching models such as Mortensen (1986), and pricing models such as Lazear (1986, 2012), all of which state that a more dispersed wage-offer distribution leads to higher search intensity and longer periods of unemployment. Providing sharp empirical tests for these ideas is difficult. However, analyzing the differences between displaced workers’ last hourly wages before the job loss and their first hourly wages after the job loss (hereafter, “pre–post-displacement wage differences”) can provide useful information.16 If displaced workers indeed respond to a more dispersed wage-offer distribution, realized pre–post-displacement wage differences should also become more dispersed after the reform. Moreover, displaced workers should be relatively unlikely to accept postdisplacement wages that lie substantially below the level of wages in their previous job. In contrast, a model in which centralized wages are simply a market friction that prevent displaced workers from finding jobs at their optimal reservation wages would predict a distribution shift to the left, simply because more displaced workers reenter employment with low wages. Table 3 presents simple descriptive statistics for pre–post-displacement wage differences. The first row of Table 3 presents average pre–post-displacement wage differences for all workers. Before the reform, average pre–post-displacement wage differences were about 3 percentage points. After the reform, pre–post-displacement wage differences were negative but close to zero. Thus, on average, displaced workers did not accept postdisplacement wages that were substantially below the level of their predisplacement wages. The second and third rows present average pre–post-displacement wage differences separately for displaced workers with low and high predisplacement wages, that is, a worker is considered as having a high predisplacement wage if it lies above the median of all predisplacement wages in that given year, and vice versa. The second row shows that displaced workers with low predisplacement wages reentered employment at wages that were on average always above the level of their predisplacement wages. In contrast, displaced workers with high predisplacement wages always reentered employment at wages below the level of their predisplacement wages. Moreover, their average pre–post-displacement wage differences changed from −0.014 to −0.070, suggesting that high-wage workers reentered employment after the reform at substantially lower wages. Figure 10 presents more detailed results by showing simple kernel density estimates for the distribution of log differences between pre- and postdisplacement wages, both before (solid line) and after (dashed line) the wage bargaining reform. Figure 10(a) presents the results for workers with low predisplacement wages, and Figure 10(b) presents the results for workers with high predisplacement wages. Both before and after the reform, pre–post-displacement wage differences were strongly centered at zero for both types of workers, suggesting that the median displaced worker did not accept postdisplacement wages that were substantially below the level of his predisplacement wages. Moreover, pre–post-displacement wage differences became more diverse for both types of workers. Interpreting these results in the light of the previous arguments suggests that displaced workers indeed responded to the increased dispersion of the wage-offer distribution. Thus the results are in line with the idea that job shopping became more important for them after the reform. Figure 10. View largeDownload slide Distribution of pre–post-displacement wage differences (pooled sample). Kernel density estimates Dep. vars.: Difference between log pre- and postdisplacement wages. (a) Low-wage worker and (b) high-wage worker. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. High-wage workers are workers who had a predisplacement wage that was larger than the median of all workers’ wages in the predisplacement year. Figure 10. View largeDownload slide Distribution of pre–post-displacement wage differences (pooled sample). Kernel density estimates Dep. vars.: Difference between log pre- and postdisplacement wages. (a) Low-wage worker and (b) high-wage worker. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. High-wage workers are workers who had a predisplacement wage that was larger than the median of all workers’ wages in the predisplacement year. Nonetheless, the distribution of displaced workers with high predisplacement wages exhibits a stronger shift to the left in response to the reform than the postreform distribution of displaced workers with low predisplacement wages. A potential reason is that low-income workers’ UI benefits are almost as large as their previous income, such that they have weak incentives to accept wages that are close to or below their predisplacement wages. In contrast, high-wage workers always received relatively lower UI benefits, such that they may have become more likely to trade off wage losses against unemployment after the reform. The next section analyzes this argument in more detail. 6.4. Unemployment Benefits and Reservation Wages If UI benefits are sufficiently generous, displaced workers’ outside options may offset the expected trade-off between wage losses and unemployment. Thus, because displaced workers remain unemployed until they find jobs that offer wages above their UI benefits, the decentralization of wage bargaining will not reduce unemployment very much. As previously mentioned, UI benefits constitute up to 90% of a worker’s previous wage, with a relatively low cap of approximately 170,000 DKKs for high-income workers. Thus, in particular, displaced workers with low predisplacement wages should always be unwilling to reenter employment for entry wages that are substantially below the wage level of their last employment before the job loss. In contrast, displaced workers with high predisplacement wages receive comparably less generous UI benefits and should, therefore, become more likely to accept wages below the level of their predisplacement wage and less likely to experience unemployment after the reform. Table 4 presents an analysis according to the following regression equation:   \begin{eqnarray} \Delta \text{log}w_{i,\mathit{ Pre-post}}^{d}&=&\alpha ^{d}+\lambda _{t}^{d}+x_{it}^{d}\beta + \gamma _{1} \mathit{ HW}_{it}^{d}\nonumber\\ &&+\,\,\delta \mathit{ HW}_{it}^{d}\cdot \mathit{ Reform}+\gamma _{2} \mathit{ Reform}+\varepsilon _{it}^{d}, \end{eqnarray} (4)where $$\Delta \text{log}w_{i,\mathit{ Pre-post}}^{d}$$ measures pre–post-displacement wage differences. Reform is a dummy for the wage bargaining reform, and $$\mathit{ HW}_{it}^{d}$$ indicates whether a displaced worker had a low or high predisplacement wage. Again, the d indicates that nondisplaced workers are excluded. Furthermore, the regressions include year dummies λt, a constant α,17 and age categories. δ is the coefficient of main interest and measures whether pre–post-displacement wage differences increased for high-wage workers after the wage bargaining reform. Thus if displaced workers with high predisplacement wages have reentered employment after the reform with relatively lower postdisplacement wages, the coefficient estimate of δ should be negative. Table 4. Trade-off between wages and unemployment (displaced workers only).     Degree of unemployment  Dep. vars.  Pre- and postdisp. wage diff.  Disp. year  1>Disp.  2>Disp.  3> Disp.  Post-dec.  0.025*  0.021**  −0.014  0.006  0.024**  ($$\mathit{ Reform}$$)  (0.013)  (0.007)  (0.010)  (0.009)  (0.009)  High wage  −0.064***  −0.008**  −0.017***  −0.017***  −0.015***  ($$\mathit{ HW}_{it}^{d}$$)  (0.006)  (0.003)  (0.004)  (0.004)  (0.004)  $$\mathit{ HW}_{it}^{d}\cdot \mathit{ Reform}$$  −0.053**  −0.020***  −0.008  −0.006  −0.005    (0.009)  (0.004)  (0.005)  (0.006)  (0.006)  Observations  9073  9073  9029  8988  8969      Degree of unemployment  Dep. vars.  Pre- and postdisp. wage diff.  Disp. year  1>Disp.  2>Disp.  3> Disp.  Post-dec.  0.025*  0.021**  −0.014  0.006  0.024**  ($$\mathit{ Reform}$$)  (0.013)  (0.007)  (0.010)  (0.009)  (0.009)  High wage  −0.064***  −0.008**  −0.017***  −0.017***  −0.015***  ($$\mathit{ HW}_{it}^{d}$$)  (0.006)  (0.003)  (0.004)  (0.004)  (0.004)  $$\mathit{ HW}_{it}^{d}\cdot \mathit{ Reform}$$  −0.053**  −0.020***  −0.008  −0.006  −0.005    (0.009)  (0.004)  (0.005)  (0.006)  (0.006)  Observations  9073  9073  9029  8988  8969  Notes: Author’s calculations with data from Statistics Denmark. Dep. var.: Column (1): Log pre- and postdisplacement wage differences. Columns (2)–(5): Degree of unemployment in respective year. Degree of unemployment measures the fraction of a year a worker received UI benefits. $$\mathit{ HW}_{it}^{d}$$ is a dummy variable indicating whether a displaced worker’s predisplacement wage was larger than the median of all workers’ wages in the predisplacement year. Reform is a dummy indicating whether the worker was displaced before or after the wage bargaining reform. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Sample is restricted to all displaced workers with valid wage observations before and after their job loss. Displacements occur between November in year (−1) and November in year (0). Regressions include year fixed effects and age categories. Standard errors are in parentheses and clustered at the individual level. *Significant at 10% level; **significant at 5% level; ***significant at 1% level. View Large The first column of Table 4 shows the results for pre- and postdisplacement wage differences. Columns (2)–(4) show results according to the same regression equation (4) but with a dependent variable measuring the degree of unemployment in the displacement year, as well as the first, second, and third years after displacement. As expected, the coefficient estimate of δ is negative and significant. Moreover, high wage workers also experienced relatively less unemployment after the reform (columns (2) through (5)). However, the unemployment effect is large and significant only for the displacement year and becomes much weaker and insignificant thereafter. These results are in line with the idea that generous UI benefits prevent displaced workers with lower wages from trading off wage losses against unemployment. 7. Placebo Tests and Alternative Explanations This section provides additional evidence improving the credibility of the main results. It also discusses a number of alternative scenarios that may explain why displaced workers experience larger income losses after the reform. 7.1. Sensitivity Analysis An ideal experiment for evaluating how the decentralization of wage bargaining systems influences the magnitude of displaced workers’ income losses would require a random assignment of centralized and decentralized wage bargaining systems to comparable sectors. However, changes in wage bargaining systems are commonly the result of complex political processes, and, unfortunately, policy makers do not tend to implement natural experiments. Therefore, no sectors unaffected by the wage bargaining reform are directly comparable to the manufacturing sector. However, this section provides two sensitivity checks to evaluate the credibility of the former results. The first sensitivity check analyzes displacement losses for a subset of low-skilled manufacturing workers likely to remain on the standard wage rate system—even after the reform. The second sensitivity check analyzes displacement losses in the Danish finance sector, a sector not exposed to a comparably drastic change of the wage bargaining system in the early 1990s. Many low-skilled workers are members of the General Workers Union, SID (Specialarbejderforbundet i Danmark), which strongly opposed the decentralization process in the early 1990s (Iversen 1996). Although the SID did not succeed in preventing the decentralization process, many low-skilled workers remained under the standard wage rate system after 1989. For example, Dahl et al. (2013) report that 59% of all workers covered by the standard wage rate system between 1992 and 2002 were low-skilled workers without an apprenticeship degree. Thus we should expect to find a weaker influence of the decentralization reform on the displacement losses of SID workers—if they remained in their predisplacement industry. Figure 11 shows estimates for a subset of all SID workers in my sample. To identify them, I rely on Statistics Denmark data that identifies the UI fund (Akasse) of each worker in a given year. Unfortunately, as the data are available only from 1985, I made the assumption that workers who had been SID members from 1985 had also been SID members between 1980 and 1984. Although making such an assumption may produce biased results, Danish workers seldom change their UI fund, so as to avoid losing their eligibility for benefits. For the purpose of this investigation, I restricted the sample to workers who did not change their manufacturing industry. Figure 11. View largeDownload slide Placebo analysis: SID workers. Estimation according to equation (1). Dep. vars.: (a) gross income, (b) hourly wages, and (c) degree of unemployment. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Gross income and hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs. Gross income includes zeros. Hourly wages exclude zeros. Degree of unemployment measures the fraction of a year a worker received UI benefits. Average income and wages are adjusted to base income and wages in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 11. View largeDownload slide Placebo analysis: SID workers. Estimation according to equation (1). Dep. vars.: (a) gross income, (b) hourly wages, and (c) degree of unemployment. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Gross income and hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs. Gross income includes zeros. Hourly wages exclude zeros. Degree of unemployment measures the fraction of a year a worker received UI benefits. Average income and wages are adjusted to base income and wages in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. The figure shows the results for income losses, wage losses, and unemployment. The first subfigure shows that SID workers experienced somewhat larger income losses than did the full sample before the reform. Displaced workers’ income losses increased slightly after the reform. However, the differences are not significant, and income losses did not increase as much as for the full sample. The second subfigure shows that SID workers experience almost no wage losses, either before or after the introduction of the reform. As the third subfigure suggests, income losses appear related to unemployment. If general macroeconomic conditions or unrelated institutional influences were to account for the increase in displaced workers’ income losses in the manufacturing sector, we should expect to also observe a similar pattern for displaced workers’ income losses in other sectors. Therefore, analyzing displaced workers’ income losses in other sectors can serve as another placebo test. In Denmark, the finance and banking sector is an ideal candidate for such a placebo analysis. Historically, the Danish finance sector was dominated by standard wage rate contracts negotiated between the financial service unions (Finansforbundet) and the employer associations of the financial sector (Finanssektorens Arbejdsgiverforening). The largest banks dominated these negotiations, and most smaller institutions adopted the agreements. However, wage bargaining in the banking sector is characterized by numerous subagreements for bank clerks and other employees, such as IT specialists and executive staff (e.g., Mayer, Andersen, and Muller 2001). Unlike in most other Danish sectors, the wage bargaining system in the finance sector was not changed in any comparable way during the 1990s. Only in 1997 did the social partners agree to implement a system that permitted greater use of performance-related pay at the company level. However, the system became effective only at the end of 1999, because managers in the banking sector were initially reluctant to implement the new pay system. Figure 12 presents the results for the finance sector. Before and after 1989, displaced workers’ income losses range between about 2% and 5%, and no structural break is visible. The same holds for wage losses and unemployment. Figure 12. View largeDownload slide Placebo analysis: Finance sector. Estimation according to equation (1). Dep. vars.: (a) gross income; (b) hourly wages; (c) degree of unemployment. Displaced workers were aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Gross income and hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs. Gross income includes zeros. Hourly wages exclude zeros. Degree of unemployment measures the fraction of a year a worker received UI benefits. Average income and wages are adjusted to base income and wages in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure 12. View largeDownload slide Placebo analysis: Finance sector. Estimation according to equation (1). Dep. vars.: (a) gross income; (b) hourly wages; (c) degree of unemployment. Displaced workers were aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of prior job tenure. Gross income and hourly wages of displaced and nondisplaced workers are measured in 2000 DKKs. Gross income includes zeros. Hourly wages exclude zeros. Degree of unemployment measures the fraction of a year a worker received UI benefits. Average income and wages are adjusted to base income and wages in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. If general macroeconomic conditions and unrelated institutional changes were to account for the increase in displaced workers’ income losses in the manufacturing sector, I should observe a similar pattern of displaced workers’ income losses in the finance sector. The results for the finance sector clearly contradict this expectation. Both of these sensitivity checks show that influences other than the decentralization of the wage bargaining system are unlikely to explain the increase in displaced workers’ income losses after 1989. 7.2. Alternative Policy Changes: UI Benefits In response to the continuously increasing unemployment at the end of the 1980s, Danish policy makers changed labor market institutions other than the wage bargaining system. One major reform changed the laws regulating UI benefits and active labor market policies. Before 1994, the eligibility period for UI benefits was effectively infinite. Despite a formal duration of seven years, workers could easily obtain a new benefit period by participating in a job activation program (Andersen 2003). In 1994 policy makers implemented a new system with two benefit periods. The first was a passive period of four years, to allow workers to focus on their job search without further requirements. The second was an activation period of three years, during which workers had to participate in active labor market programs in return for UI benefits. Policy makers further reduced the passive period to three years in 1996 and to two years in 1998. Previous evidence showed that unemployed workers indeed responded to the UI system reform by becoming more likely to reenter employment shortly before the compulsory participation in active labor market programs was to begin (Geerdsen 2006). Thus the question arises as to whether the change in the UI system can explain the increase of displaced workers’ income losses after 1989. For example, the newly implemented activation period may have led displaced workers to reenter employment faster but at lower wages. However, for the following four reasons, the reform of the UI system is unlikely to account for the increase in displaced workers’ income losses after 1989. First, although the UI reform was implemented only in 1994, displaced workers’ income losses had begun increasing earlier. Second, previous evidence suggests that most unemployed workers responded to the reform just about three months before the beginning of their activation period. But even after the UI reform, most displaced workers in my sample had the opportunity to benefit from a passive period of three or four years. Therefore, the change in the UI system cannot account for the effects I find within the first three or four years after the job loss. Third, most displaced workers appear to have reentered employment shortly after their job loss, with relatively few experiencing long-term unemployment. Therefore, the sample that I analyze in this paper does not include many long-term unemployed workers, who, according to Geerdsen (2006), are those who responded more strongly to the reform of the UI system. Fourth, the results in Section 7 showed that most displaced workers reentered employment at wages comparable to their predisplacement wages. This result is at odds with the idea that displaced workers reduced their unemployment at the cost of lower postdisplacement wages. 7.3. Alternative Policy Changes: Parental Leave In response to the economic crisis of the late 1980s, Danish policy makers implemented yet another labor market reform (Rostgaard and Christoffersen 1999). In 1992, they introduced a new parental leave provision that was available to all parents with children younger than eight. The aim was to lower unemployment by reducing the labor supply and by creating new opportunities for temporary employment. Although this reform initially gave parents benefits allowing them to leave work for 36 weeks, in 1993 policy makers extended this period to one year. However, as unemployment began to fall in 1994, policy makers reduced the child-care leave period to 13 weeks.18 Therefore, the reform of the child-care leave provision is unlikely to be the cause of the persistent increase in displaced workers’ income losses after 1989. Moreover, very few fathers took child-care leave. For example, in 1996 only 4% of fathers took advantage of any parental leave plan. As a result, a substantial influence of the child-care leave reform on the male-dominated labor market for manufacturing workers is unlikely. 8. Conclusion Many European countries have decentralized their wage bargaining systems over the past decades to foster economic growth and reduce unemployment. By showing the relationship between the decentralization of wage bargaining and displaced workers’ income losses, this paper provides important insights into the consequences of decentralized wage bargaining for the specific group involuntary job losers. More specifically, this paper uses administrative register data to analyze the relationship between the decentralization of wage bargaining systems and displaced workers’ income losses. It exploits a major reform of the Danish wage bargaining system, a reform that changed wage setting in the manufacturing sector from a highly centralized national system to a decentralized one with a substantial emphasis on firm-level wage bargaining. The results show small income losses under the centralized wage bargaining system, which was dominated by standard wage rate contracts that remunerated workers according to their education, occupation, and sector experience. After the decentralization of wage bargaining, a large percentage of workers had to negotiate their wages at the firm level, and displaced workers’ income losses increased substantially. Displaced workers who lost their jobs before the reform appear to have benefited by the wage drifts of general contracts and centralized negotiations. In contrast, displaced workers who lost their jobs after the reform appear to have forgone wage growth, because they failed to collect rents from tenure or voluntary job shopping. Moreover, both before and after the reform, displaced workers did not reenter employment for wages substantially below the level of their predisplacement wages, likely because of generous UI benefits. Appendix A: Additional Tables Table A.1. Baseline characteristics (pre-decentralization). Sample  1980–1985  1981–1986  1982–1987  Disp. year  1982  1983  1984  Gross income  292506  310624.2**  281861  286425.3  281504.2  270647.9*  Age  35.29  35.05  35.42  35.93*  35.62  34.83*  Experience  11.98  11.85  12.46  12.16*  12.98  11.48***  Education              Low  0.358  0.356  0.356  0.353  0.361  0.495***  Medium  0.527  0.512  0.527  0.535  0.528  0.360***  High  0.099  0.122*  0.1  0.092  0.096  0.123  Mfr. sector              Food and bev.  0.242  0.349***  0.242  0.117***  0.253  0.271  Metal products  0.46  0.306***  0.461  0.257***  0.449  0.457  Wood products  0.116  0***  0.113  0.245***  0.112  0.019***  Other mfr. indust  0.182  0.345***  0.184  0.382***  0.186  0.252**  Individuals  441  56870  781  59344  317  58421  Sample  1983–1988  1984–1989  1985–1990  Disp. year  1985  1986  1987  Gross income  285521.1  270212.3**  284364.2  277677.6  288474.6  259933.4***  Age  35.92  34.62***  36.06  34.77**  36.01  33.03***  Experience  13.53  12.28***  13.9  12.88**  14.08  11.64***  Education              Low  0.356  0.401**  0.353  0.357  0.343  0.390**  Medium  0.531  0.53  0.528  0.5  0.536  0.495**  High  0.098  0.046***  0.103  0.126  0.107  0.096  Mfr. sector              Food and bev.  0.251  0.585***  0.238  0.343***  0.224  0.188**  Metal products  0.456  0.290***  0.449  0.135***  0.452  0.590***  Wood products  0.107  0.0138***  0.109  0.287***  0.118  0.047***  Other mfr. indust.  0.186  0.111***  0.205  0.235  0.206  0.175**  Individuals  434  55005  230  52850  685  50350  Sample  1986–1991      Disp. year  1988      Gross income  291877.9  281774.6**          Age  35.83  34.95***          Experience  14.18  13.39***          Education              Low  0.338  0.390***          Medium  0.538  0.524          High  0.111  0.068***          Mfr. sector              Food and bev.  0.215  0.618***          Metal products  0.455  0.338***          Wood products  0.114  0.007***          Other mfr. indust.  0.215  0.037***          Individuals  1083  52699          Sample  1980–1985  1981–1986  1982–1987  Disp. year  1982  1983  1984  Gross income  292506  310624.2**  281861  286425.3  281504.2  270647.9*  Age  35.29  35.05  35.42  35.93*  35.62  34.83*  Experience  11.98  11.85  12.46  12.16*  12.98  11.48***  Education              Low  0.358  0.356  0.356  0.353  0.361  0.495***  Medium  0.527  0.512  0.527  0.535  0.528  0.360***  High  0.099  0.122*  0.1  0.092  0.096  0.123  Mfr. sector              Food and bev.  0.242  0.349***  0.242  0.117***  0.253  0.271  Metal products  0.46  0.306***  0.461  0.257***  0.449  0.457  Wood products  0.116  0***  0.113  0.245***  0.112  0.019***  Other mfr. indust  0.182  0.345***  0.184  0.382***  0.186  0.252**  Individuals  441  56870  781  59344  317  58421  Sample  1983–1988  1984–1989  1985–1990  Disp. year  1985  1986  1987  Gross income  285521.1  270212.3**  284364.2  277677.6  288474.6  259933.4***  Age  35.92  34.62***  36.06  34.77**  36.01  33.03***  Experience  13.53  12.28***  13.9  12.88**  14.08  11.64***  Education              Low  0.356  0.401**  0.353  0.357  0.343  0.390**  Medium  0.531  0.53  0.528  0.5  0.536  0.495**  High  0.098  0.046***  0.103  0.126  0.107  0.096  Mfr. sector              Food and bev.  0.251  0.585***  0.238  0.343***  0.224  0.188**  Metal products  0.456  0.290***  0.449  0.135***  0.452  0.590***  Wood products  0.107  0.0138***  0.109  0.287***  0.118  0.047***  Other mfr. indust.  0.186  0.111***  0.205  0.235  0.206  0.175**  Individuals  434  55005  230  52850  685  50350  Sample  1986–1991      Disp. year  1988      Gross income  291877.9  281774.6**          Age  35.83  34.95***          Experience  14.18  13.39***          Education              Low  0.338  0.390***          Medium  0.538  0.524          High  0.111  0.068***          Mfr. sector              Food and bev.  0.215  0.618***          Metal products  0.455  0.338***          Wood products  0.114  0.007***          Other mfr. indust.  0.215  0.037***          Individuals  1083  52699          Notes: Author’s calculations with data provided by Statistics Denmark. The table shows descriptive statistics for nondisplaced manufacturing workers two years before the displacement period of the respective sample. Labor earnings, wages, and gross income are measured in 2000 DKKs. Standard errors are in parentheses and clustered at the individual level. *Significant at 10% level; **significant at 5% level; ***significant at 1% level. View Large Table A.2. Baseline characteristics (post-decentralization). Sample  1987–1992  1988–1993  1989–1994  Disp. year  1989  1990  1991  Gross income  297192.3  293308.8  297500  281197.5**  302512.1  286267.8**  Age  35.66  34.51**  35.58  33.70***  35.65  34.39**  Experience  14.81  14.43  14.94  13.84***  15.22  15.14  Education              Low  0.311  0.306  0.308  0.417***  0.306  0.429***  Medium  0.559  0.545  0.563  0.479***  0.567  0.526  High  0.116  0.132  0.115  0.089*  0.114  0.036***  Mfr. sector              Food and bev.  0.175  0.306***  0.195  0.387***  0.202  0.309***  Metal products  0.473  0.524**  0.457  0.464  0.454  0.538**  Wood products  0.121  0.039***  0.116  0.015***  0.108  0.009***  Other mfr.  0.231  0.130***  0.232  0.134***  0.236  0.144***  Individuals  389  55281  501  57657  440  58053  Sample  1990–1995  1991–1996  1992–1997  Disp. year  1992  1993  1994  Gross income  297192.3  293308.8  297500  281197.5**  302512.1  286267.8**  Age  35.66  34.51**  35.58  33.70***  35.65  34.39**  Experience  14.81  14.43  14.94  13.84***  15.22  15.14  Education              Low  0.311  0.306  0.308  0.417***  0.306  0.429***  Medium  0.559  0.545  0.563  0.479***  0.567  0.526  High  0.116  0.132  0.115  0.089*  0.114  0.036***  Mfr. sector              Food and bev.  0.175  0.306***  0.195  0.387***  0.202  0.309***  Metal products  0.473  0.524**  0.457  0.464  0.454  0.538**  Wood products  0.121  0.039***  0.116  0.015***  0.108  0.009***  Other mfr. indust.  0.231  0.130***  0.232  0.134***  0.236  0.144***  Individuals  431  58174  470  59843  333  60136  Sample  1993–1998  1994–1999  1995–2000  Disp. year  1995  1996  1997  Gross income  299095.1  291924.5  303265  272220.5***  303552.8  317956.8**  Age  35.78  33.57***  35.81  34.37***  35.66  36.01  Experience  15.57  14.58***  15.84  15.6  15.95  16.45  Education              Low  0.306  0.448***  0.301  0.421***  0.3  0.246*  Medium  0.571  0.510**  0.574  0.521**  0.575  0.59  High  0.111  0.036***  0.113  0.0487***  0.113  0.149*  Mfr. sector              Food and bev.  0.199  0.568***  0.202  0.338***  0.199  0.343***  Metal products  0.463  0.336***  0.461  0.259***  0.465  0.295***  Wood products  0.109  0.03***  0.109  0.005***  0.108  0.134  Other mfr. indust.  0.229  0.066***  0.228  0.397***  0.229  0.228  Individuals  563  57899  390  58844  268  61225  Sample  1996–2001  1997–2002  1998–2003  Disp. year  1998  1999  2000  Gross income  303011.2  294882.2  306357.4  304647  312491.6  299407.7**  Age  35.68  35.69  35.82  35.28**  35.94  35.61  Experience  16.2  16.16  16.55  16.86  16.87  17.1  Education              Low  0.3  0.293  0.295  0.356***  0.29  0.325*  Medium  0.573  0.603  0.575  0.567  0.579  0.566  High  0.114  0.095  0.117  0.07***  0.118  0.089**  Mfr. sector              Food and bev.  0.198  0.451***  0.183  0.278***  0.182  0.486***  Metal products  0.456  0.158***  0.454  0.481  0.46  0.374***  Wood products  0.113  0.230***  0.111  0.041***  0.114  0.033***  Other mfr. indust.  0.233  0.161**  0.251  0.199**  0.244  0.107***  Individuals  348  61135  702  60084  551  58456  Sample  1987–1992  1988–1993  1989–1994  Disp. year  1989  1990  1991  Gross income  297192.3  293308.8  297500  281197.5**  302512.1  286267.8**  Age  35.66  34.51**  35.58  33.70***  35.65  34.39**  Experience  14.81  14.43  14.94  13.84***  15.22  15.14  Education              Low  0.311  0.306  0.308  0.417***  0.306  0.429***  Medium  0.559  0.545  0.563  0.479***  0.567  0.526  High  0.116  0.132  0.115  0.089*  0.114  0.036***  Mfr. sector              Food and bev.  0.175  0.306***  0.195  0.387***  0.202  0.309***  Metal products  0.473  0.524**  0.457  0.464  0.454  0.538**  Wood products  0.121  0.039***  0.116  0.015***  0.108  0.009***  Other mfr.  0.231  0.130***  0.232  0.134***  0.236  0.144***  Individuals  389  55281  501  57657  440  58053  Sample  1990–1995  1991–1996  1992–1997  Disp. year  1992  1993  1994  Gross income  297192.3  293308.8  297500  281197.5**  302512.1  286267.8**  Age  35.66  34.51**  35.58  33.70***  35.65  34.39**  Experience  14.81  14.43  14.94  13.84***  15.22  15.14  Education              Low  0.311  0.306  0.308  0.417***  0.306  0.429***  Medium  0.559  0.545  0.563  0.479***  0.567  0.526  High  0.116  0.132  0.115  0.089*  0.114  0.036***  Mfr. sector              Food and bev.  0.175  0.306***  0.195  0.387***  0.202  0.309***  Metal products  0.473  0.524**  0.457  0.464  0.454  0.538**  Wood products  0.121  0.039***  0.116  0.015***  0.108  0.009***  Other mfr. indust.  0.231  0.130***  0.232  0.134***  0.236  0.144***  Individuals  431  58174  470  59843  333  60136  Sample  1993–1998  1994–1999  1995–2000  Disp. year  1995  1996  1997  Gross income  299095.1  291924.5  303265  272220.5***  303552.8  317956.8**  Age  35.78  33.57***  35.81  34.37***  35.66  36.01  Experience  15.57  14.58***  15.84  15.6  15.95  16.45  Education              Low  0.306  0.448***  0.301  0.421***  0.3  0.246*  Medium  0.571  0.510**  0.574  0.521**  0.575  0.59  High  0.111  0.036***  0.113  0.0487***  0.113  0.149*  Mfr. sector              Food and bev.  0.199  0.568***  0.202  0.338***  0.199  0.343***  Metal products  0.463  0.336***  0.461  0.259***  0.465  0.295***  Wood products  0.109  0.03***  0.109  0.005***  0.108  0.134  Other mfr. indust.  0.229  0.066***  0.228  0.397***  0.229  0.228  Individuals  563  57899  390  58844  268  61225  Sample  1996–2001  1997–2002  1998–2003  Disp. year  1998  1999  2000  Gross income  303011.2  294882.2  306357.4  304647  312491.6  299407.7**  Age  35.68  35.69  35.82  35.28**  35.94  35.61  Experience  16.2  16.16  16.55  16.86  16.87  17.1  Education              Low  0.3  0.293  0.295  0.356***  0.29  0.325*  Medium  0.573  0.603  0.575  0.567  0.579  0.566  High  0.114  0.095  0.117  0.07***  0.118  0.089**  Mfr. sector              Food and bev.  0.198  0.451***  0.183  0.278***  0.182  0.486***  Metal products  0.456  0.158***  0.454  0.481  0.46  0.374***  Wood products  0.113  0.230***  0.111  0.041***  0.114  0.033***  Other mfr. indust.  0.233  0.161**  0.251  0.199**  0.244  0.107***  Individuals  348  61135  702  60084  551  58456  Notes: Author’s calculations with data from Statistics Denmark. The table shows descriptive statistics for displaced and nondisplaced manufacturing workers two years before the displacement period of the respective sample. Labor earnings, wages, and gross income are measured in 2000 DKKs. Standard errors are in parentheses and clustered at the individual level. *Significant at 10% level; **significant at 5% level; ***significant at 1% level. View Large Table A.3. Average real gross income losses (pre-decentralization). Disp. year  1982  1983  1984  1985  1986  $$D_{it}^{-1}$$  −5822.870  346.579  −5425.082**  7226.436***  −5203.899*     (7108.505)  (1265.987)  (2459.115)  (1933.702)  (2718.442)  Displacement  −10333.621  1234.724  −6467.426  1000.653  −5659.449*    (7958.089)  (1428.032)  (4456.945)  (2605.696)  (2918.881)  $$D_{it}^{1}$$  −6378.632**  −2566.888  −1838.633  −579.673  1234.947    (2838.265)   (1781.667)  (4424.249)  (5249.173)  (4119.185)  $$D_{it}^{2}$$  −1954.378  −1694.688  53.784  1738.79  −3478.647    (3398.521)   (2244.434)  (4473.693)  (5901.728)  (4035.749)  $$D_{it}^{2}$$  5968.243  −4328.676  −3760.320  −6344.822  −8386.598*     (4412.835)  (3007.733)   (4970.008)  (6550.623)  (5093.659)  Observations:  331127  345249  335713  316599  304206  Disp. year  1987  1988        $$D_{it}^{-1}$$  4592.791***  6345.078**          (1358.475)   (2036.713)        Displacement  3255.017  8478.351          (2114.587)   (5589.042)        $$D_{it}^{1}$$  −1034.001  −2892.284          (2673.976)  (2286.879)        $$D_{it}^{2}$$  1365.098  −152.887          (2828.742)  (4883.442)        $$D_{it}^{3}$$  6083.684  −2434.299          (4051.434)  (2870.725)        Observations:  294318  309727        Disp. year  1982  1983  1984  1985  1986  $$D_{it}^{-1}$$  −5822.870  346.579  −5425.082**  7226.436***  −5203.899*     (7108.505)  (1265.987)  (2459.115)  (1933.702)  (2718.442)  Displacement  −10333.621  1234.724  −6467.426  1000.653  −5659.449*    (7958.089)  (1428.032)  (4456.945)  (2605.696)  (2918.881)  $$D_{it}^{1}$$  −6378.632**  −2566.888  −1838.633  −579.673  1234.947    (2838.265)   (1781.667)  (4424.249)  (5249.173)  (4119.185)  $$D_{it}^{2}$$  −1954.378  −1694.688  53.784  1738.79  −3478.647    (3398.521)   (2244.434)  (4473.693)  (5901.728)  (4035.749)  $$D_{it}^{2}$$  5968.243  −4328.676  −3760.320  −6344.822  −8386.598*     (4412.835)  (3007.733)   (4970.008)  (6550.623)  (5093.659)  Observations:  331127  345249  335713  316599  304206  Disp. year  1987  1988        $$D_{it}^{-1}$$  4592.791***  6345.078**          (1358.475)   (2036.713)        Displacement  3255.017  8478.351          (2114.587)   (5589.042)        $$D_{it}^{1}$$  −1034.001  −2892.284          (2673.976)  (2286.879)        $$D_{it}^{2}$$  1365.098  −152.887          (2828.742)  (4883.442)        $$D_{it}^{3}$$  6083.684  −2434.299          (4051.434)  (2870.725)        Observations:  294318  309727        Notes: Author’s calculations with data from Statistics Denmark. Dep. var.: Annual gross income. Gross income is measured in 2000 DKKs and includes zeros. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Displacements occurred between November in year (−1) and November in year (0). Estimations are based on subsamples including displaced and nondisplaced workers. Each subsample follows both groups from two years before until three years after the displacement. Regressions include worker fixed effects, year fixed effects, and age categories. Standard errors are in parentheses and clustered at the individual level. *Significant at 10% level; **significant at5% level; ***significant at 1% level. View Large Table A.4. Average real gross income losses (postdecentralization). Disp. year  1989  1990  1991  1992  1993  $$D_{it}^{-1}$$  −4275.060**  −1814.476  35.11  −6449.536**  3089.666**    (1774.308)  (1567.708)  (2015.536)  (2079.534)  (1546.235)   Displacement  −7921.461**  −7230.973**  −4776.423  −10862.486***  −4999.634    (3245.806)  (3504.699)  (8104.956)  (3200.725)  (3071.603)   $$D_{it}^{1}$$  −21461.412***  −18033.971***  −13453.181**  −20182.018***  −9283.576**    (3616.340)  (5171.091)  (6422.683)  (4077.017)  (3736.323)  $$D_{it}^{2}$$  −20959.511***  −25164.434***  −7764.893  −17163.633***  −6469.091    (3973.129)  (3976.383)  (7992.864)  (4555.523)  (4058.477)  $$D_{it}^{3}$$  −28371.280***  −31038.667***  −22670.557**  −22614.100***  −11033.397**    (4447.770)  (4227.903)  (9537.456)  (4667.531)  (4178.573)  Observations  319360  333616  336281  338781  347311  Disp. year  1994  1995  1996  1997  1998  $$D_{it}^{-1}$$  −6788.029**  −1404.602  −6210.384***  10581.042***  9313.031***    (2460.529)  (1631.844)  (1499.809)  (2779.603)  (2581.047)   Displacement  −8922.475**  2452.191  −3682.919  38886.286***  −939.723    (3460.701)  (2869.926)  (3170.452)  (6410.583)  (5031.128)  $$D_{it}^{1}$$  −16853.147***  −13448.483***  −18191.982***  −11872.156**  −19319.945***    (5087.391)  (2986.011)  (3458.957)  (5082.981)  (4152.068)  $$D_{it}^{2}$$  −16730.114***  −19033.770***  −16315.782***  −577.569  −21456.393***    (4249.352)  (3479.684)  (3903.711)  (5063.351)  (3786.774)  $$D_{it}^{3}$$  −18627.864***  −18160.333***  −13593.930**  −2566.970  −17517.248***    (4560.996)  (4874.681)  (4380.230)  (5752.573)  (4187.611)  Observations:  347987  336725  340373  352623  352509  Disp. year  1999  2000        $$D_{it}^{-1}$$  6897.004**  7040.556***          (2340.955)  (1703.726)        Displacement  −1321.610  6448.736*          (3268.578)  (3384.888)        $$D_{it}^{1}$$  −27554.941***  −19806.119***          (3010.144)  (3422.487)        $$D_{it}^{2}$$  −19399.242***  −25884.981***          (3279.583)   (3570.380)        $$D_{it}^{3}$$  −18680.271***  −23407.572***          (3689.807)  (4082.181)        Observations:  348741  338037        Disp. year  1989  1990  1991  1992  1993  $$D_{it}^{-1}$$  −4275.060**  −1814.476  35.11  −6449.536**  3089.666**    (1774.308)  (1567.708)  (2015.536)  (2079.534)  (1546.235)   Displacement  −7921.461**  −7230.973**  −4776.423  −10862.486***  −4999.634    (3245.806)  (3504.699)  (8104.956)  (3200.725)  (3071.603)   $$D_{it}^{1}$$  −21461.412***  −18033.971***  −13453.181**  −20182.018***  −9283.576**    (3616.340)  (5171.091)  (6422.683)  (4077.017)  (3736.323)  $$D_{it}^{2}$$  −20959.511***  −25164.434***  −7764.893  −17163.633***  −6469.091    (3973.129)  (3976.383)  (7992.864)  (4555.523)  (4058.477)  $$D_{it}^{3}$$  −28371.280***  −31038.667***  −22670.557**  −22614.100***  −11033.397**    (4447.770)  (4227.903)  (9537.456)  (4667.531)  (4178.573)  Observations  319360  333616  336281  338781  347311  Disp. year  1994  1995  1996  1997  1998  $$D_{it}^{-1}$$  −6788.029**  −1404.602  −6210.384***  10581.042***  9313.031***    (2460.529)  (1631.844)  (1499.809)  (2779.603)  (2581.047)   Displacement  −8922.475**  2452.191  −3682.919  38886.286***  −939.723    (3460.701)  (2869.926)  (3170.452)  (6410.583)  (5031.128)  $$D_{it}^{1}$$  −16853.147***  −13448.483***  −18191.982***  −11872.156**  −19319.945***    (5087.391)  (2986.011)  (3458.957)  (5082.981)  (4152.068)  $$D_{it}^{2}$$  −16730.114***  −19033.770***  −16315.782***  −577.569  −21456.393***    (4249.352)  (3479.684)  (3903.711)  (5063.351)  (3786.774)  $$D_{it}^{3}$$  −18627.864***  −18160.333***  −13593.930**  −2566.970  −17517.248***    (4560.996)  (4874.681)  (4380.230)  (5752.573)  (4187.611)  Observations:  347987  336725  340373  352623  352509  Disp. year  1999  2000        $$D_{it}^{-1}$$  6897.004**  7040.556***          (2340.955)  (1703.726)        Displacement  −1321.610  6448.736*          (3268.578)  (3384.888)        $$D_{it}^{1}$$  −27554.941***  −19806.119***          (3010.144)  (3422.487)        $$D_{it}^{2}$$  −19399.242***  −25884.981***          (3279.583)   (3570.380)        $$D_{it}^{3}$$  −18680.271***  −23407.572***          (3689.807)  (4082.181)        Observations:  348741  338037        Notes: See Table A.3. View Large Appendix B: Macroeconomic Developments And Displaced Workers’ Earnings Losses Section 6 showed that macroeconomic conditions had only a weak influence on displaced workers’ short-term income losses. This outcome remains in contrast to the findings in the previous literature of a rather strong relationship between macroeconomic conditions and displaced losses. To infer more specifically to what extent macroeconomic changes may account for the increase of income losses after the reform, columns (2) and (3) of Table B.1 show further specifications of equation (2). These specifications include the national unemployment rate (II) and a dummy variable for the Nordic crisis between 1989 and 1993 (III). As both columns show, the results prove extremely robust to the inclusion of those macroeconomic indicators. Table B.1. Average real gross income losses (pooled sample). Specification  I  II  III  IV  V    Pre-decentralization period  $$D_{it}^{-1}$$  2901.831**  2903.621**  2903.640**  2901.660**  2903.663    (1085.318)  (1085.320)  (1085.321)  (1085.318)  (1085.321)  Displacement  2553.732  2557.736  2558.776  2553.191  2557.748    (1897.388)  (1897.391)  (1897.387)  (1897.388)  (1897.392)  $$D_{it}^{1}$$  −1190.575  −1185.547  −1190.123  −1190.251  −1183.038    (1167.817)  (1167.807)  (1167.817)  (1167.822)  (1167.801)  $$D_{it}^{2}$$  −290.036  −284.083  −293.713  −289.849  −286.725    (1731.657)  (1731.641)  (1731.665)  (1731.660)  (1731.648)  $$D_{it}^{3}$$  −2060.017  −2055.721  −2066.651  −2060.229  −2062.172    (1584.277)  (1584.269)  (1584.297)  (1584.278)  (1584.284)    Post-decentralization period  $$Reform \cdot D_{it}^{-1}$$  −696.665  −797.862  −649.109  −648.533  −453.310    (1255.200)  (1260.159)  (1257.714)  (1275.024)  (1317.415)  Reform ·  −1471.296  −1724.998  −1463.115  −1288.495  −814.662  Disp.  (2317.848)  (2347.678)  (2318.557)  (2362.736)  (2464.033)  $$Reform \cdot D_{it}^{1}$$  −14331.219***  −14747.380***  −14500.329***  −14175.279***  −13379.700***    (1816.255)  (1877.514)  (1808.330)  (1909.954)  (2082.575)  $$Reform \cdot D_{it}^{2}$$  −15220.223***  −15827.106***  −15571.643***  −15139.448***  −14365.461***    (2326.961)  (2416.846)  (2302.553)  (2393.104)  (2740.708)  $$Reform \cdot D_{it}^{3}$$  −16549.619***  −17326.460***  −17088.650***  −16354.378***  −15628.610***    (2427.847)  (2551.854)  (2394.630)  (2475.713)  (3102.946)  Observations:  2080471  2080471  2080471  2080471  2080471  Controls:            Unemployment  No  Yes  No  No  No  Nordic crisis  No  No  Yes  No  No  FDI flows  No  No  No  Yes  No  Ex-/imports  No  No  No  No  Yes  Specification  I  II  III  IV  V    Pre-decentralization period  $$D_{it}^{-1}$$  2901.831**  2903.621**  2903.640**  2901.660**  2903.663    (1085.318)  (1085.320)  (1085.321)  (1085.318)  (1085.321)  Displacement  2553.732  2557.736  2558.776  2553.191  2557.748    (1897.388)  (1897.391)  (1897.387)  (1897.388)  (1897.392)  $$D_{it}^{1}$$  −1190.575  −1185.547  −1190.123  −1190.251  −1183.038    (1167.817)  (1167.807)  (1167.817)  (1167.822)  (1167.801)  $$D_{it}^{2}$$  −290.036  −284.083  −293.713  −289.849  −286.725    (1731.657)  (1731.641)  (1731.665)  (1731.660)  (1731.648)  $$D_{it}^{3}$$  −2060.017  −2055.721  −2066.651  −2060.229  −2062.172    (1584.277)  (1584.269)  (1584.297)  (1584.278)  (1584.284)    Post-decentralization period  $$Reform \cdot D_{it}^{-1}$$  −696.665  −797.862  −649.109  −648.533  −453.310    (1255.200)  (1260.159)  (1257.714)  (1275.024)  (1317.415)  Reform ·  −1471.296  −1724.998  −1463.115  −1288.495  −814.662  Disp.  (2317.848)  (2347.678)  (2318.557)  (2362.736)  (2464.033)  $$Reform \cdot D_{it}^{1}$$  −14331.219***  −14747.380***  −14500.329***  −14175.279***  −13379.700***    (1816.255)  (1877.514)  (1808.330)  (1909.954)  (2082.575)  $$Reform \cdot D_{it}^{2}$$  −15220.223***  −15827.106***  −15571.643***  −15139.448***  −14365.461***    (2326.961)  (2416.846)  (2302.553)  (2393.104)  (2740.708)  $$Reform \cdot D_{it}^{3}$$  −16549.619***  −17326.460***  −17088.650***  −16354.378***  −15628.610***    (2427.847)  (2551.854)  (2394.630)  (2475.713)  (3102.946)  Observations:  2080471  2080471  2080471  2080471  2080471  Controls:            Unemployment  No  Yes  No  No  No  Nordic crisis  No  No  Yes  No  No  FDI flows  No  No  No  Yes  No  Ex-/imports  No  No  No  No  Yes  Notes: Author’s calculations with data from Statistics Denmark. Dep. var.: Annual gross income. Gross income is measured in 2000 DKKs and includes zeros. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Displacements occurred between November in year (−1) and November in year (0). Estimations are based on a pooled sample. Reform is a dummy indicating whether the worker was displaced before or after the wage bargaining reform. Exports, imports, and FDI inflows and outflows are measured in millions of US dollars. Regressions include worker fixed effects, year fixed effects, and age categories. Standard errors are in parentheses and clustered at the individual level. **Significant at 5% level; ***significant at 1% level. View Large However, national unemployment may be strongly related to the decentralization of the wage bargaining system, so that more exogenous indicators for measuring macroeconomic conditions may produce very different results. Therefore, specification IV controls for inflows and outflows of Foreign Direct Investments (FDI), and specification V accounts for exports and imports measured in millions of US dollars.19 Although both of these macroeconomic indicators appear to have a somewhat stronger effect on the results, the main outcome does not change. The main explanation for macroeconomic conditions having only little effect on displaced workers’ income losses may be that generous UI benefits and large payments of the “guarantee fund” smooth the income losses of displaced workers. I provide evidence for this argument in Figure B.1, which shows displaced workers’ real annual earnings losses (excluding UI benefits and all other kinds of social benefit payments). I calculate the labor earnings by summing workers’ November job earnings and the sum of their wages for all other jobs in a given year. If no labor income is reported for a worker in a given year, I count workers’ labor earnings as 0. Figure B.1. View largeDownload slide Annual earnings losses without UI benefits. Estimation according to equation (2). Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Annual earnings are measured in 2000 DKKs and includes zeros. Average earnings are adjusted to a base earning in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). Regressions include worker fixed effects, year fixed effects, and age categories. The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure B.1. View largeDownload slide Annual earnings losses without UI benefits. Estimation according to equation (2). Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Annual earnings are measured in 2000 DKKs and includes zeros. Average earnings are adjusted to a base earning in the second year before the displacement year. Displacements occurred between November in year (−1) and November in year (0). Regressions include worker fixed effects, year fixed effects, and age categories. The capped spikes indicate confidence bands at 5% level. Confidence bands are calculated with the delta method. Figure B.1 presents the average estimated earnings losses for the pre-decentralization period but divides the post-decentralization period (after 1989) into two separate periods: one for the recession period between 1989 and 1993, and one for the expansion period between 1994 and 2000. The figure presents the results as percentages of displaced workers’ predisplacement earnings losses. The figure shows that displaced workers’ real earnings losses are generally much larger than their real income losses, and—as in most the US studies—displaced workers’ earnings losses peak during the displacement year. Notably, the magnitudes of displaced workers’ earnings losses are similar to those in a Swedish study by Eliason and Storrie (2006), who also use labor earnings instead of total income. Most importantly, in contrast to the findings for income losses, displaced workers’ earnings losses respond much more strongly to macroeconomic conditions than do income losses and were significantly larger during the Nordic crisis than afterwards. Thus generous UI benefits indeed appear to smooth displaced workers’ income losses and explain why displaced workers’ income losses do not respond strongly to changes in the business cycle. Appendix C: Additional Sensitivity Analysis This section presents further three sensitivity analysis to prove the robustness of the results. First, a number of studies have shown that the selection of the groups of displaced and nondisplaced workers may have some influence on the magnitude of estimated income losses, and researchers rely on different selection criteria for different reasons. In this study I have decided to largely follow what the most prominent studies in the literature have done. As common in this literature, I have required that displaced workers separate from firms with more than 50 employees and that nondisplaced workers follow a relatively stable employment in firms with more than 50 employees (i.e., I require nondisplaced workers to have had at least one wage employment with a positive income for each subsequent year). However, other samples that allow for a larger amount of data can be constructed. Therefore, column (1) of Table C.1 provides a robustness check for which I reduced the firm size requirement to firms with only more than 30 employees. Moreover, I did not impose any other restriction on nondisplaced workers other than the restriction that they had at least three years of tenure. This sample selection increased the sample size by about 600,000 observations. As column (1) of Table C.1 shows, the results prove to be very similar to the results of the main specification. Table C.1. Alternative specifications (pooled sample). Specification  Sample cond.  First differences  Matching  $$D_{it}^{-1}$$  2227.190**  2244.689**  3516.580**    (942.617)  (1084.659)  (1115.474)  Displacement  1904.195  1261.441  3675.282*    (1646.380)  (1893.966)  (1941.637)  $$D_{it}^{1}$$  −1937.283*  −3078.235**  864.795    (1062.624)  (1161.792)  (1275.026)  $$D_{it}^{2}$$  −1744.543  −2714.031  2451.248    (1534.297)  (1721.671)  (1819.188)  $$D_{it}^{3}$$  −3108.941**  −5197.753**  1175.792    (1445.149)  (1587.526)  (1725.369)    Post-decentralization period  Reform$$\cdot D_{it}^{-1}$$  −1198.839  −913.765  −1629.755    (1092.562)  (1244.002)  (1313.097)  Reform · Disp.  −2488.040  −1830.713  −3072.338    (1991.826)  (2277.915)  (2413.820)  Reform$$\cdot D_{it}^{1}$$  −14207.696***  −14759.819***  −17863.161***    (1614.160)  (1758.585)  (1935.158)  Reform$$\cdot D_{it}^{2}$$  −14238.886***  −15921.435***  −19543.120***    (2046.070)  (2243.576)  (2440.424)  Reform$$\cdot D_{it}^{3}$$  −15740.132***  −17185.425***  −21110.164***    (2133.569)  (2308.938)  (2648.002)  Observations  2723280  1833907  159174  Specification  Sample cond.  First differences  Matching  $$D_{it}^{-1}$$  2227.190**  2244.689**  3516.580**    (942.617)  (1084.659)  (1115.474)  Displacement  1904.195  1261.441  3675.282*    (1646.380)  (1893.966)  (1941.637)  $$D_{it}^{1}$$  −1937.283*  −3078.235**  864.795    (1062.624)  (1161.792)  (1275.026)  $$D_{it}^{2}$$  −1744.543  −2714.031  2451.248    (1534.297)  (1721.671)  (1819.188)  $$D_{it}^{3}$$  −3108.941**  −5197.753**  1175.792    (1445.149)  (1587.526)  (1725.369)    Post-decentralization period  Reform$$\cdot D_{it}^{-1}$$  −1198.839  −913.765  −1629.755    (1092.562)  (1244.002)  (1313.097)  Reform · Disp.  −2488.040  −1830.713  −3072.338    (1991.826)  (2277.915)  (2413.820)  Reform$$\cdot D_{it}^{1}$$  −14207.696***  −14759.819***  −17863.161***    (1614.160)  (1758.585)  (1935.158)  Reform$$\cdot D_{it}^{2}$$  −14238.886***  −15921.435***  −19543.120***    (2046.070)  (2243.576)  (2440.424)  Reform$$\cdot D_{it}^{3}$$  −15740.132***  −17185.425***  −21110.164***    (2133.569)  (2308.938)  (2648.002)  Observations  2723280  1833907  159174  Notes: Author’s calculations with data from Statistics Denmark. Dep. var.: Annual gross income. Gross income is measured in 2000 DKKs and includes zeros. Displaced workers are aged 50 or younger, were separated from their main job during a mass layoff event, and have at least two years of previous job tenure. Displacements occurred between November in year (−1) and November in year (0). Estimations are based on a pooled sample. Reform is a dummy indicating whether the worker was displaced before or after the wage bargaining reform. *Significant at 10% level; **significant at 5% level; ***significant at 1% level. View Large Second, if the income path of displaced workers had systematically differed from that of nondisplaced workers—even in the absence of the displacement—regression equations (1) and (2) may produce biased results. Moreover, if those worker-specific trends had changed over time, the previous results may have picked up such changes in unobserved dynamic heterogeneity. Wooldridge (2015) suggests testing for this possibility by performing both fixed effects and first difference estimators. Column (2) in Table C.1 shows that performing a first-difference regression leads to results similar to those in the standard fixed effects approach of equation (2). Third, Tables A.1 and A.2 show that baseline characteristics between displaced and nondisplaced workers were not fully balanced before the displacement event. To reduce the potential bias that may arise from systematic differences in unobserved characteristics of displaced and nondisplaced workers, previous studies suggested pairing displaced and nondisplaced workers according to their predisplacement characteristics before estimating equations such as (1) and (2) (Hijzen et al. 2010). This approach may be preferred to simple linear regression methods if the functional form assumptions of the latter are violated, and if members of the treatment or control group lie outside the support of the propensity distribution. Therefore, specification III of Table C.1 presents results of an estimation for which I used single-nearest-neighbor propensity score matching to pair displaced and nondisplaced workers. I estimate the propensity score from a set of characteristics in the second period before displacement. The characteristics include industry, education, experience, and age. Table C.1 shows that the results are very similar to those of the standard approach. In other words, systematic differences in predisplacement characteristics between displaced and nondisplaced workers appear not to account for the increase in displaced workers’ income losses. Notes The editor in charge of this paper was Claudio Michelacci. Acknowledgements I particularly thank Edward Lazear, Nils Westergaard-Nielsen, Thomas Siedler, Silke Anger, Uschi Backes-Gellner, Bob Hall, Paul Oyer, Kathryn Shaw, Hank Farber, Ulrich Kaiser, Jens Mohrenweiser, Malte Sandner, Nikolaj Harmon, Jens Iversen, and the participants of the Econ brown-bag seminar at the Stanford Graduate School of Business for their helpful comments and suggestions. I also thank Natalie Reid for language editing and Astrid Erismann for research assistance. This study is partially funded by the Swiss Federal Office for Professional Education and Technology through its Leading House on the Economics of Education, Firm Behavior, and Training Policies. During the work on this paper, I am grateful for financial support from the Swiss National Science Foundation. I thank the Cycles, Adjustment, and Policy research unit, CAP, Department of Economics and Business, Aarhus University, for support and for making the data available. Janssen is a Research Affiliate at IZA Footnotes 1 For earlier survey articles, see Fallick (1996) and Kletzer (1998). Other studies investigate how job displacement influences unemployment duration (e.g., Chan and Stevens 1999). Yet other studies show that worker displacement has substantial effects on health, mortality, fertility, and children’s labor market outcomes (e.g., Oreopoulos, Page, and Stevens 2008; Eliason and Storrie 2009; Sullivan and Von Wachter 2009; Stevens and Schaller 2011; Del Bono, Weber, and Winter-Ebmer 2012). 2 For example, the US studies find displaced workers’ earnings and wage losses of between 11% (e.g., Ruhm 1991; Farber 1993, 1997) and 40% per year (e.g., Ruhm 1991; Farber 1993, 1997; Stevens 1997; Couch 1998; Schoeni and Dardia 1997; Couch and Placzek 2010; Davis and von Wachter 2011). Most European studies find smaller income losses that range between 3% and 15% (e.g., Bender et al. 2002; Burda and Mertens 2001; Couch 2001; Eliason and Storrie 2006; Von Wachter and Bender 2006; Schmieder et al. 2010; Hijzen et al. 2010; Huttunen et al. 2011; Korkeamäki and Kyyrä 2014; Ichino et al. 2016). However, the evidence is more diverse for Europe. For example, Hijzen et al. (2010) find up to 35% for Britain, and Eliason and Storrie (2006) find earnings losses of up to 19% for Sweden. 3 The employer side pushed for this change because internationalization and technological change meant that wage contracts were not sufficiently flexible to accommodate local conditions. The employee side did not oppose this change, because it feared that labor demand might suffer if wages were not sufficiently flexible. Moreover, the employer side agreed to introduce mandatory labor market pensions, on-the-job training, and child-care leave. 4 Several studies show that the decentralization process affected wage setting in the manufacturing sector. For example, Bingley and Westergaard-Nielsen (2003) show that returns to tenure more than doubled at the beginning of the decentralization period in 1989, and Eriksson and Westergaard-Nielsen (2009) show that wage dispersion increased at the beginning of the late 1980s. Finally, Dahl et al. (2013) find that returns to skills are larger and that wage dispersion is greater under firm-level wage bargaining in Denmark. 5 In rare cases, guarantee fund payments can be extended for up to six months. 6 For example, Davis and von Wachter (2011) exclude all closing firms from their sample because they are unable to distinguish a firm closure from a simple change in a firm ID. As I do not face this problem in my data, I do not remove firms that go out of business. 7 Nondisplaced workers may become unemployed between job switches. 8 Some studies count a worker as unemployed only if he or she received UI benefits for more than 20% of the ongoing year. But given that displaced workers may indeed experience many short periods of real unemployment, such an approach may be too conservative for the purpose of this paper. 9 The markers in Figures 6–9, 11, 12, and B.1 express monetary displacement losses as a percentage of the workers’ base income. Displacement losses as function of the workers base income are essentially functions g() of the coefficient estimates δk. To calculate the confidence intervals for g(δk), I approximated the standard errors of g(δk) by $$\sqrt{(G^{\prime }VG)}$$, where G is a vector of the derivatives ∂g/∂δk and V is the estimated variance–covariance of the δks. This method is commonly referred to as the delta method. 10 If the number of time periods becomes large in comparison to the number of observations, fixed effect estimators may produce biased results. This may be especially problematic if the sample length is systematically longer for workers who are displaced after the reform. Wooldridge (2015) argues that both fixed effects and first difference estimators should produce similar results if the assumption of time constant unobserved heterogeneity holds. Therefore, Table C.1 in the appendix provides additional results from a first-difference approach. Both approaches produce very similar results in the short panels. 11 Although the separate samples overlap, the estimation sample for equation (2) contains only one observation per worker year. A very small number of workers experienced more than one displacement within a period of less than four years. For those few workers, I took into account only their first displacement. 12 Only very few workers in the sample experienced more than one displacement. However, experiencing a displacement both before and after the reform is a necessary condition for observing variation of the $$ {Reform}$$ dummy within a single worker’s observation period. 13 In contrast to the parallel trends assumption, balanced predisplacement characteristics are not an identification assumption of the difference-in-differences estimators in equations (1) and (2). However, more recently a number of studies have experimented with matching and reweighting techniques to more precisely estimate displaced workers’ income losses (e.g., Couch and Placzek 2010; Hijzen et al. 2010). Following these approaches, Table C.1 in the appendix presents additional results for which I matched the samples of nondisplaced workers so that their average baseline characteristics match those of displaced workers. As found in those earlier studies, matching and reweighting techniques do not substantially change the results. 14 I follow Davis and von Wachter (2011) and calculate the DPPVs as $$\mathit{ DPV}^{\mathit{ Loss}}_{\tau }=\sum _{k=1}^{3}\bar{\delta _{t}^{k}}({1}/{(1+r)^{t-1}})$$. I also assume a real interest rate of 5% and normalize the DPVs, using displaced workers’ mean income during the baseline period of each sample to account for changing income levels over time. However, I do not extrapolate the income losses for periods after the observation period as Davis and von Wachter (2011) do. 15 If rigid wages under centralized wage bargaining would primarily prevent displaced workers from reentering employment at their optimal reservation wages, displaced workers’ income losses should decrease, not increase, under decentralized wage bargaining systems. 16 The postdisplacement wage may stem from the first new job in the displacement year or from the first job that the worker found any time after that. However, more than 90% of all displaced workers found a job with a valid wage observation within less a year after their job loss. 17 I cannot include worker fixed effects in equation (4), because $$\text{log}w_{i,\mathit{ Pre-post}}^{d}$$ assigns one single observation to each worker. 18 With special permission from the employer, the period could be extended to 52 weeks. 19 The information on Foreign Direct Investments (FDI), exports, and imports stem from the homepage of the United Nations Conference on Trade and Development (http://unctad.org/en/Pages/Home.aspx). FDI inflows and outflows are measured in US dollars at current prices and current exchange rates. They comprise capital provided by a foreign direct investor to a FDI enterprise or capital received by a foreign direct investor from a FDI enterprise. FDI includes the following three components: equity capital, reinvested earnings, and intracompany loans. Imports and exports measure the total trade in goods and services in US dollars at current prices and current exchange rates. References Abowd John M., Kramarz Francis ( 1999). “ The Analysis of Labor Markets Using Matched Employer-Employee Data.” Handbook of Labor Economics , 3, 2629– 2710. Google Scholar CrossRef Search ADS   Andersen Torben M. ( 2003). “ Changes in Danish Labour Market Bargaining: The Prototypical Case of Organized Decentralization?” In National Regimes of Collective Bargaining in Transformation: Nordic Trends in Comparative Perspective , edited by Dølvik Jon E., Engelstad Fredrik. Makt-og Demokratiutredningen Norge  1918– 2003. Andersen Søren K., Mailand Mikkel ( 2005). “ The Danish Flexicurity Model.” Compiled for the Danish Ministry of Employment, Copenhagen. Andersen Torben M., Svarer Michael ( 2007). “ Flexicurity—Labour Market Performance in Denmark.” CESifo Economic Studies , 53, 389– 429. Google Scholar CrossRef Search ADS   Bender Stefan, Dustmann Christian, Margolis David, Meghir Costas ( 2002). “ Worker Displacement in France and Germany.” In Losing Work, Moving On: International Perspectives on Worker Displacement , edited by Kuhn Peter J., pp. 375– 470. WE Upjohn Institute. Google Scholar CrossRef Search ADS   Bingley Paul, Westergaard-Nielsen Niels ( 2003). “ Returns to Tenure, Firm-Specific Human Capital and Worker Heterogeneity.” International Journal of Manpower , 24, 774– 788. Google Scholar CrossRef Search ADS   Burda Michael C., Mertens Antje ( 2001). “ Estimating Wage Losses of Displaced Workers in Germany.” Labour Economics , 8, 15– 41. Google Scholar CrossRef Search ADS   Card David, Heining Jörg, Kline Patrick ( 2013). “ Workplace Heterogeneity and the Rise of West German Wage Inequality.” The Quarterly Journal of Economics , 128, 967– 1015. Google Scholar CrossRef Search ADS   Card David, Lemieux Thomas, Riddell Craig W. ( 2004). “ Unions and Wage Inequality.” Journal of Labor Research , 25, 519– 559. Google Scholar CrossRef Search ADS   Chan Sewin, Stevens Ann Huff ( 1999). “ Employment and Retirement Following a Late-career Job Loss.” American Economic Review , 89(2), 211– 216. Google Scholar CrossRef Search ADS   Couch Kenneth A. ( 1998). “ Late Life Job Displacement.” The Gerontologist , 38, 7– 17. Google Scholar CrossRef Search ADS PubMed  Couch Kenneth A. ( 2001). “ Earnings Losses and Unemployment of Displaced Workers in Germany.” Industrial and Labor Relations Review , 54, 559– 572. Google Scholar CrossRef Search ADS   Couch Kenneth A., Placzek Dana W. ( 2010). “ Earnings Losses of Displaced Workers Revisited.” American Economic Review , 100(1), 572– 589. Google Scholar CrossRef Search ADS   Dansk Arbejdsgiverforening ( 2004). Arbejdsmarkedsrapport 2004 Dansk Arbejdsgiverforening, Copenhagen. Dahl Christian M., Le Maire Daniel, Munch Jakob R. ( 2013). “ Wage Dispersion and Decentralization of Wage Bargaining.” Journal of Labor Economics , 31, 501– 533. Google Scholar CrossRef Search ADS   Davis Steven J., Wachter Till von ( 2011). “ Recessions and the Costs of Job Loss.” Brookings Papers on Economic Activity , 2, 1– 72. Google Scholar CrossRef Search ADS   Del Bono, Emilia, Weber Andrea, Winter-Ebmer Rudolf ( 2012). “ Clash of Career and Family: Fertility Decisions after Job Displacement.” Journal of the European Economic Association , 10, 659– 683. Google Scholar CrossRef Search ADS   Dustmann Christian, Ludsteck Johannes, Schönberg Uta ( 2009). “ Revisiting the German Wage Structure.” Quarterly Journal of Economics , 124, 843– 881. Google Scholar CrossRef Search ADS   Eliason Marcus, Storrie Donald ( 2006). “ Lasting or Latent Scars? Swedish Evidence on the Long-Term Effects of Job Displacement.” Journal of Labor Economics , 24, 831– 856. Google Scholar CrossRef Search ADS   Eliason Marcus, Storrie Donald ( 2009). “ Does Job Loss Shorten Life?” Journal of Human Resources , 44, 277– 302. Google Scholar CrossRef Search ADS   Eriksson Tor, Westergaard-Nielsen Niels ( 2009). “ Wage and Labor Mobility in Denmark, 1980-2000.” In The Structure of Wages: An International Comparison , edited by Lazear Edward P., Shaw Kathlyn L.. University of Chicago Press, pp. 101– 123. Fallick Bruce C. ( 1996). “ A Review of the Recent Empirical Literature on Displaced Workers.” Industrial and Labor Relations Review , 50, 5– 16. Google Scholar CrossRef Search ADS   Farber Henry S. ( 1997). “ The Changing Face of Job Loss in the United States, 1981-1995.” Brookings Papers on Economic Activity , 55– 142. Farber Henry S. ( 1993). “ The Incidence and Costs of Job Loss: 1982-91.” Brookings Papers on Economic Activity: Microeconomics , 1, 73– 132. Google Scholar CrossRef Search ADS   Freeman Richard B. ( 1980). “ The Exit-Voice Tradeoff in the Labor Market: Unionism, Job Tenure, Quits, and Separations.” The Quarterly Journal of Economics , 94, 643– 673. Google Scholar CrossRef Search ADS   Geerdsen Lars P. ( 2006). “ Is There a Threat Effect of Labour Market Programmes? A Study of ALMP in the Danish UI system.” The Economic Journal , 116, 738– 750. Google Scholar CrossRef Search ADS   Hijzen Alexander, Upward Richard, Wright Peter W. ( 2010). “ The Income Losses of Displaced Workers.” Journal of Human Resources , 45, 243– 269. Google Scholar CrossRef Search ADS   Huttunen Kristiina, Møen Jarle, Salvanes Kjell G. ( 2011). “ How Destructive is Creative Destruction? Effects of Job Loss on Job Mobility, Withdrawal, and Income.” Journal of the European Economic Association , 9, 840– 870. Google Scholar CrossRef Search ADS   Ichino Andrea, Schwerdt Guido, Winter-Ebmer Rudolf, Zweimüller Josef ( 2016). “ Too Old to Work, too Young to Retire?” The Journal of the Economics of Aging . Iversen Jens, Malchow-Møller Nikolaj, Sørensen Anders ( 2010). “ Returns to Schooling in Self-Employment.” Economics Letters , 109, 179– 182. Google Scholar CrossRef Search ADS   Iversen Torben ( 1996). “ Power, Flexibility, and the Breakdown of Centralized Wage Bargaining: Denmark and Sweden in Comparative Perspective.” Comparative Politics , 28, 399– 436. Google Scholar CrossRef Search ADS   Jacobson Louis S., LaLonde Robert J., Sullivan Daniel G. ( 1993). “ Earnings Losses of Displaced Workers.” American Economic Review , 83(4), 685– 709. Kiander J. ( 2004). “ Growth and Employment in Nordic Welfare States in the 1990s: A Tale of Crisis and Revival.” Government Institute for Economic Research VATT Discussion Papers No. 336, Helsinki, Finnland . Kletzer Lori G. ( 1998). “ Job Displacement.” Journal of Economic Perspectives , 12, 115– 136. Google Scholar CrossRef Search ADS   Kletzer Lori G., Fairlie Robert W. ( 2003). “ The Long-term Costs of Job Displacement for Young Adult Workers.” Industrial and Labor Relations Review , 56, 682– 698. Google Scholar CrossRef Search ADS   Korkeamäki Ossi, Kyyrä Tomi ( 2014). “ A Distributional Analysis of Earnings Losses of Displaced Workers in an Economic Depression and Recovery.” Oxford Bulletin of Economics and Statistics , 76, 565– 588. Google Scholar CrossRef Search ADS   Lazear Edward P. ( 1986). “ Retail Pricing and Clearance Sales.” American Economic Review , 76(1), 14– 32. Lazear Edward P. ( 2012). “ Why Should Inventories Rise when Demand Falls in Housing and other Markets?” The Singapore Economic Review , 57, 32. Google Scholar CrossRef Search ADS   Mayer Gerit, Andersen Torben, Muller Michael ( 2001). “ Employment Restructuring and Flexibility in Austrian and Danish Banking.” European Journal of Industrial Relations , 7, 71– 87. Google Scholar CrossRef Search ADS   Mortensen Dale T. ( 1986). “ Job Search and Labor Market Analysis.” Handbook of Labor Economics , 2, 849– 919. Google Scholar CrossRef Search ADS   Oreopoulos Philip, Page Marianne, Stevens Ann Huff ( 2008). “ The Intergenerational Effects of Worker Displacement.” Journal of Labor Economics , 26(3), 455– 483. Google Scholar CrossRef Search ADS   Pedersen Lars Haagen, Smith Nina, Stephensen Peter ( 1998). “ Wage Formation and Minimum Wage Contracts: Th