DP 1522

IAB-DISCUSSION PAPER
Articles on labour market issues
15|2022Beyond Lost Earnings: The Long-Term

Impact of Job Displacement on Workers’
Commuting Behavior
Yige Duan, Oskar Jost, Ramona Jost
ISSN 2195-2663
Beyond Lost Earnings: The Long-Term
Impact of Job Displacement on Workers’
Commuting Behavior
Yige Duan (University of British Columbia (UBC))

Oskar Jost (Institute for Employment Research (IAB))
Ramona Jost (Institute for Employment Research (IAB))
Mit der Reihe „IAB-Discussion Paper“ will das Forschungsinstitut der Bundesagentur für Ar-
beit den Dialog mit der externen Wissenschaft intensivieren. Durch die rasche Verbreitung
von Forschungsergebnissen über das Internet soll noch vor Drucklegung Kritik angeregt und
Qualität gesichert werden.
The “IAB-Discussion Paper” is published by the research institute of the German Federal Em-
ployment Agency in order to intensify the dialogue with the scientific community. The prompt
publication of the latest research results via the internet intends to stimulate criticism and to
ensure research quality at an early stage before printing.
IAB-Discussion Paper 15|2022 2

Contents
1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1 Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2 Data and Empirical Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Identification Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Coarsened Exact Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.1 Mechanism. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 The Value of Granularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.3 Additional Estimates and Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4 Does Commuting Mitigate the Wage Loss?. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5 A Job Search Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

5.1 Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.2 Dynamics of Surplus, Commuting, and Wage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.3 Empirical Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
A1 Additional Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
A2 Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
A2.1 Proof for Proposition 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
A2.2 Proof for Proposition 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
A3 Model Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
A3.1 Endogenous Job Search Effort . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
A3.2 Heterogeneous Job Security . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

Abstract
We study the long-term impact of job displacement on workers’ commuting behavior. Our
measures of commuting exploit geo-coordinates of workers’ places of residence and places
of work, from which we calculate the door-to-door commuting distance and commuting
time. Using German employee-employer matched data and an event study design, we
identify the causal effect of job loss on workers displaced during a mass layoff. Conditional
on finding a new job, workers’ commuting distance and commuting time rise sharply after
displacement and gradually decline in subsequent years. The recovery is due to employer
changes rather than migration, and a larger increase in commuting would mitigate the wage
loss due to job displacement. To rationalize our findings, we build an on-the-job search
model with heterogeneous firm productivity and commuting distances. Our model predicts
a joint recovery of wages and commuting despite a static tradeoff between the two
attributes.
Zusammenfassung
Wir analysieren den langfristigen Effekt von Entlassungen auf das Pendelverhalten von
Beschäftigten. Hierfür nutzen wir Geoinformationen zum Wohn- und Arbeitsort der
Beschäftigten, womit wir die Pendeldistanz und Pendelzeit bestimmen können. Mithilfe der
zur Verfügung stehenden Betriebsinformationen und einer Ereignisstudie können wir
zudem Massenentlassungen nutzen, um kausale Effekte auf Beschäftigte zu identifizieren.
Unsere Ergebnisse zeigen, dass Beschäftigte, die wieder in Arbeit finden, deutlich erhöhte
Pendeldistanzen und Pendelzeiten haben, welche in den Folgejahren abnehmen. Ferner
zeigen unsere Ergebnisse, dass Beschäftigte, die nach einer Entlassung länger pendeln
ihren Einkommensverlust reduzieren und die langfristige Erholung der Pendeldistanzen
und Pendelzeit auf den Wechsel von Firmen zurückgeht und nicht durch Umzüge getrieben
ist. Um unsere Ergebnisse analytisch zu diskutieren und zu begründen nutzen wir ein
suchtheoretisches Modell, welches eine gemeinsame Erholung des Einkommens und
Pendelns postuliert und damit eine statistische Betrachtung entscheidend erweitert.
JEL
J3, J6, R23, R4

Keywords
commuting, mobility, displacement, job search
Acknowledgements
We thank Stephan Brunow, Wolfgang Dauth, Nicole Fortin, David Green, Konstantin Körner,
Thomas Lemieux, Jordy Meekes, Duncan Roth, Raffaele Saggio, Kyungchul Song, and
seminar participants at the University of British Columbia for valuable comments and
support.

1 Introduction
Job displacement has profound consequences on individual workers. A large body of

literature has documented substantial and persistent earning losses suffered by displaced
workers (e.g., Jacobson/LaLonde/Sullivan, 1993b; Davis/von Wachter, 2011; Bertheau et al.,
2021). In particular, the authors attribute the slow recovery from job loss to job search
frictions, i.e., displaced workers take time to find a job that pays as much as before.
However, less attention has been paid to spatial frictions—the cost of commuting and
migration that erodes job opportunities outside displaced workers’ local labor markets
(Manning/Petrongolo, 2017; Marinescu/Rathelot, 2018). How do spatial frictions affect
workers’ recovery from job loss? How do workers respond to and trade off multiple job
search frictions? These questions are important for developing labor market policies to
alleviate the scarring effect of job loss (Bertheau et al., 2021).
In this paper, we investigate the job displacement effect on the commuting behavior of
German workers. Our paper makes three major contributions. First, using an event study
design, we document a joint recovery of wages and commuting costs. Symmetric to the
scarring effect on earnings, job displacement leads to a sharp increase—followed by a
gradual decline—in commuting by displaced workers. This symmetry contrasts with the
cross-sectional, substituting relationship between wages and commuting, and we use an
on-the-job search model to reconcile the static and dynamic patterns.1 Second, we combine
route planning algorithms with geo-coordinate data to precisely measure workers’
commuting distance and commuting time. This allows us to overcome bias in the
estimation using aggregate distance data, and to distinguish commuting from migration,
the latter of which was found to have no mitigating effect on the wage loss of displaced
workers (Fackler/Rippe, 2017; Huttunen/Møen/Salvanes, 2018). Third, we provide empirical
evidence that workers could alleviate the wage loss after job displacement by commuting
more. As such, policymakers could mitigate the impact of job loss by reducing spatial labor
market frictions and workers’ commuting costs.
To begin with, we use German social security data to estimate the effects of job
displacement on workers’ wages and commuting outcomes. Following the seminal event
study approach (Jacobson/LaLonde/Sullivan, 1993b), we compare workers displaced in
mass layoffs with a matched group of non-displaced workers exhibiting similar
pre-displacement characteristics. To rule out selection into mass layoffs, we control for
worker fixed effects and focus on full-time workers at displacing firms. After job
1
In cross-sectional data, wages and commuting costs tend to move in the same direction (Le Barban-
chon/Rathelot/Roulet, 2021; Dauth/Haller, 2020). On the contrary, we find an increase in wages accompa-
nied by a decrease in commuting costs in a dynamic setting.

displacement, the annual earnings of displaced workers fall by almost half in the short run
and gradually recover over subsequent years. Their daily wages exhibit smaller but similarly
persistent losses. However, the commuting distance and commuting time of displaced
workers both rise by roughly 20 percent after job displacement, and recover steadily
afterward. The recovery in commuting is driven by workers switching from distant to nearby
firms rather than migrating or relocating their homes.
Prior to our study, many researchers exploited regional-level data (e.g., the shortest
distance between two cities) to investigate workers’ commuting and migration decisions
(Eliasson/Lindgren/Westerlund, 2003; Fackler/Rippe, 2017; Meekes/Hassink, 2019, 2022). In
contract, our unique geo-coordinate data allows us to calculate, for each worker, the exact
commuting distance and commuting time along the most likely route between the place of
residence and workplace. We emphasize that individual-level commuting data is critical to
correctly identify the effects of job displacement. Using our estimates as a benchmark, we
show that measuring commutes at the regional level leads to notable bias. The bias either
overstates the true effect by omitting within-region commutes, or understates it by
misrepresenting commuting patterns across regions. According to our calculation,
measuring commutes by German municipalities could inflate the true effect by up to
one-fourth.
Next, we investigate whether workers can mitigate the wage losses by increasing
commuting or vice versa. Comparing the job loss effects by gender, we find that displaced
women’s wages drop more than men’s but their commuting increases to a lesser extent.
This implies women are more willing to accept a lower wage in order to avoid commuting
(Le Barbanchon/Rathelot/Roulet, 2021; Illing/Schmieder/Trenkle, 2021; Meekes/Hassink,
2022). At the individual level, we find workers with a greater estimated wage loss tend to
have a smaller increase in commuting. Therefore, we conclude that commuting attenuates
the wage losses of displaced workers.
The above results suggest that displaced workers experience declining commuting costs
and increasing wages in the years following their job loss. This contrasts with the static
tradeoff between wages and commuting, where workers compensate for longer commuting
distances by asking for higher wages. To reconcile the countervailing forces, we develop a
Burdett-Mortensen on-the-job search model (Burdett/Mortensen, 1998) with heterogeneous
firm productivity and commuting distances. Our model is able to capture both the static
and the dynamic relationship between wages and commuting.
In static settings, workers trade off the two attributes and are willing to sacrifice higher
wages for less commuting. Since both outcomes are associated with the joint match
surplus, we use the latter to establish a preference order among jobs, and claim that worker
prefer jobs at higher surplus levels. Using the first jobs taken on by displaced workers, we

graphically identify the set of jobs that create a positive surplus. Dynamically, however,
on-the-job search allows workers to gradually improve their job opportunities by
bargaining for higher wages or moving to more proximate firms. We prove that workers
accept a job offer only if it creates a higher match surplus than the current level, and the
expected productivity (commuting distance) is higher (lower) conditional on a higher
surplus. As is shown by the event study estimates, after the initial job separation, the
average productivity of new employers increases over time, and the average commuting
cost declines. More generally, our model is able to characterize the dynamics of the optimal
search effort and workers’ tradeoff between multiple job characteristics.
Our study delivers several insights for policymakers. First, it sheds light on the value of
transportation infrastructure in the labor market. Studies have found that better transport
facilities allow workers to access job opportunities more easily (Asher/Novosad, 2020;
Brooks/Donovan, 2020). We argue this will particularly benefit unemployed workers, who
incur more substantial commuting costs in job search. Our findings also point to the
benefits of commuting subsidies (Franklin, 2018; Moreno-Monroy/Posada, 2018), which
directly lowers the cost of commuting and has been adopted in several European countries.2
Temporary commuting subsidies would in particular help displaced workers recover from
job loss. Finally, the growing popularity of working from home (Barrero/Bloom/Davis,
2021), which substantially reduces the commuting cost of remote-job workers, would
attenuate the negative employment shocks caused by the Covid-19 pandemic.
1.1 Related Literature
Our paper contributes to a long literature on earning losses due to unemployment

(Jacobson/LaLonde/Sullivan, 1993b; Stevens, 1997; Couch/Placzek, 2010; Davis/von
Wachter, 2011; Flaaen/Shapiro/Sorkin, 2019; Schmieder/von Wachter/Heining, 2020;
Bertheau et al., 2021). Prior to our study, Eliasson/Lindgren/Westerlund (2003) and
Meekes/Hassink (2019) investigate how job loss impacts cross-regional commuting, while
Fackler/Rippe (2017) focus on the effect on migration. Our research differs from their work
in three ways. First, we examine commuters both within and across regions and distinguish
between commuting and migration. Since one-third of commutes take place within
administrative regions, focusing on cross-regional mobility is likely to overestimate the
displacement effect on commuting. Second, we demonstrate that the recovery of
commuting behavior is driven by workers changing firms rather than relocation. This
confirms Huttunen/Møen/Salvanes (2018) claim that migration by displaced workers is
mainly driven by non-economic factors. Third, commuting is not only costly, but also
2
For example, Austria, Denmark, Germany, and Sweden subsidize commuting via income tax deductions, and
France via tax-free reimbursements.

moderates the wage loss from job displacement. Therefore, our study depicts a more
complete picture of the consequences of job loss.
Next, our study is related to the theoretical literature on job search. To rationalize the
long-term behavior of displaced workers, we develop a job search model in the spirit of
Burdett/Mortensen (1998) and Postel-Vinay/Robin (2002a,b). Our model extends
Jacobson/LaLonde/Sullivan (1993a) and Le Barbanchon/Rathelot/Roulet (2021) to allow for
on-the-job search, and to study the long term impacts of job loss. Our model is also related
to Jarosch (2021), in which jobs differ in both productivity and security. However, we
demonstrate that the joint recovery of wages and commuting is driven by workers’ tradeoffs
rather than a correlation between the two attributes.
Finally, we contribute to the literature on spatial labor markets, specifically on how mobility
costs affect people’s location choices. Analyzing job application data in the U.K. and the
U.S., Manning/Petrongolo (2017) and Marinescu/Rathelot (2018) find that job seekers are
discouraged from applying for distant jobs. This suggests an aversion to commuting or
migration. Another strand of literature directly estimates the costs of relocation
(Kennan/Walker, 2011; Schmutz/Sidibé, 2019; Ransom, 2021) and the cost of commuting
(Van den Berg/Gorter, 1997; Mulalic/Van Ommeren/Pilegaard, 2014; Dauth/Haller, 2020;
Jost, 2020). We complement their work by showing that the willingness to commute varies
with workers’ employment status: Unemployed workers have the strongest incentive to
commute; but as workers recover from job loss, they exhibit a greater preference for jobs in
the proximity of their homes. Moreover, we exploit granular and precise commuting data to
distinguish between commuting and migration, and our findings reveal that they play
different roles in job search by displaced workers.
The rest of the paper is organized as follows. Section 2 discusses the data and empirical
methods. Section 3 presents the estimation results and robustness checks. Section 4
examines whether workers could mitigate the wage loss from job displacement by
increasing commuting. Section 5 introduces a job search model to rationalize the empirical
findings. Finally, Section 6 concludes.

2 Data and Empirical Methods
2.1 Data
Our study exploits two administrative data sets provided by the Institute for Employment
Research in Germany: The Integrated Employment Biographies (IEB) and the Establishment
History Panel (BHP). Our IEB data cover a representative sample of German workers subject
to social security, and contain information on each individual’s age, gender, educational
attainment, employment history, and receipt of unemployment benefits. However, the data
do not include civil servants and the self-employed. Meanwhile, the BHP provides
information on all establishments in Germany, e.g., their industry classification (2008
edition) and number of employees on June 30 each year.
A unique feature of our data is the additional information on the place of residence and
place of work in the form of geo-coordinates. Combining the granular location data with
road network data from OpenStreetMap.org (Huber/Rust, 2016; Dauth/Haller, 2020; Jost,
2020), we calculate the door-to-door commuting distance and commuting time by car for all
employed workers. We argue that the driving distance and driving time are representative
measures of commuting costs, as 68 percent of German commuters drive to work (Destatis,
2017). Using Google Maps data, we also validate that driving distances are highly correlated
with distances via other means of transportation. However, since the driving time is
calculated using estimated driving speeds in ideal road and traffic conditions, it likely
underestimates the true driving time, e.g., where traffic is heavy or parking takes time
(Dauth/Haller, 2020). To alleviate this bias, we estimate the effect of job displacement on
the within-individual, percentage change of commuting time, and show that the estimates
using driving distance and driving time are closely aligned.
As in Schmieder/von Wachter/Heining (2020) and Illing/Schmieder/Trenkle (2021), we

consider workers of both genders aged from 20 to 54. The sample period spans from 2000 to
2017, as the location information is only available from 2000 onwards. Following the data
preparation guide by Dauth/Eppelsheimer (2020), we transform the employment spells
from the IEB into a worker-year panel. For each year, we calculate the worker’s total
earnings, number of days in full-time, part-time, and mini jobs3 , the main employer as of
June 30, commuting time and distance to the main employer, and average daily wage paid
by the main employer. If a worker has multiple employers on June 30, we choose the one
with the highest daily wage as the main employer. In addition, we impute wages that are
3
Mini jobs are a form of marginal employment in Germany with a monthly income of 450 euros or less.
(Tazhitdinova, 2020; Illing/Schmieder/Trenkle, 2021)

top-coded due to the social security threshold for about 10 percent of workers
(Card/Heining/Kline, 2013) and missing education variables
(Fitzenberger/Osikominu/Völter, 2005). Finally, we exclude workers whose maximum
commuting distance exceeds 100 kilometers, as they are unlikely to commute on a daily
basis (Dauth/Haller, 2020).
On the establishment side, we consider mass layoffs as the source of involuntary job
displacement (Blien/Dauth/Roth, 2021; Burdett/Carrillo-Tudela/Coles, 2020;
Fackler/Mueller/Stegmaier, 2021; Jarosch, 2021). An establishment is said to have a mass
layoff in calendar year τ if (i) it has at least 20 workers on June 30 of year τ ; (ii) the number of
workers decreases by at least 30 percent from June 30 of year τ to June 30 of year τ + 1; and
(iii) the number of workers on June 30 of year τ is not higher than 130 percent of that on
June 30 of year τ − 1. After merging with the worker-level data, we identify 25,699 mass
layoff events. The average layoff establishment has 121 employees, 60 percent of whom are
laid off. Hereafter, we will use “firm” and “establishment” interchangeably.
To further distinguish mass layoffs from voluntary job-to-job transition, firm

reorganizations, and outsourcing (Goldschmidt/Schmieder, 2017), we follow the literature
and impose additional restrictions on the displaced workers. A worker is displaced in year τ
if he (i) leaves the displacing firm between years τ and τ + 1; (ii) receives unemployment
benefits for at least 30 days between June 30 of year τ and that of year τ + 1
(Burdett/Carrillo-Tudela/Coles, 2020; Jarosch, 2021); (iii) has worked full-time at the
displacing firm for at least three years before the mass layoff (Fackler/Mueller/Stegmaier,
2021). Furthermore, if a worker experiences multiple mass layoffs during our sample
period, only the first time is considered for analysis.
It is worth noting that the BHP only identifies mass layoffs between June 30 of two
consecutive years. As such, a worker could be displaced either in the second half of our
defined year of displacement or in the first half of the following year. Due to this
discrepancy, we expect mass layoffs to have a partial but immediate impact on workers’
annual total earnings. However, this does not affect the wage rate or commuting in the
same year, as they are measured on June 30, the earliest possible day of mass layoffs.
2.2 Identification Strategy
To identify the effect of mass layoffs on workers’ earnings, wages, and commuting behavior,
we exploit the seminal event study approach by Jacobson/LaLonde/Sullivan (1993b).
Specifically, let M Li indicate if worker i experiences a mass layoff and τ (i) be the mass

layoff year. The outcome of interest Yit , for worker i in year t, is determined by
X
11
Yit = [αk I{t = τ (i) + k} + βk I{t = τ (i) + k}M Li ] + Xit′ γ + ϕi + ψt + ϵit . (1)
k=−4,
k̸=−1
In (1), αk captures the outcome of non-displaced workers in the k -th year after (−k -th year
before) the mass layoff relative to the year immediately before (τ (i) − 1). For displaced
workers, this is captured by αk + βk . As such, βk represents the partial effect of job
displacement on Yit in the k -th year after (−k -th year before) the mass layoff. In addition,
Xit controls for time-varying worker characteristics, i.e., a cubic polynomial of age. Other
time-invariant characteristics, such as gender and education, are absorbed by the worker
fixed effect ϕi . In addition, ψt represents the calendar-year fixed effect and uit is the error
term.
We also estimate the average displacement effect over the entire post-displacement period.
The estimates are obtained by a difference-in-differences (DID) regression:
−2
X
Yit = [αk I{t = τ (i) + k} + βk I{t = τ (i) + k}M Li ]
k=−4 (2)
∗ ∗
+ α I{t ≥ τ (i)} + β I{t ≥ τ (i)}M Li + Xit′ γ + ϕi + ψt + ϵit ,
where β−4 to β−2 still capture the pretrend but β ∗ now represents a weighted average of βk
for all k ≥ 0, i.e., the average of the year-by-year displacement effects. Other components of
(2) remain the same as in (1).
To identify the causal effect of job displacement on workers’ outcomes, we rely on the
unconfoundedness assumption—that mass layoff incidences are not correlated with
workers’ characteristics relevant to their labor market outcomes. This assumption could be
violated if less productive workers or workers with unstable employment relationships face
higher risks of being displaced in a mass layoff, as it is more difficult for them to find new
jobs. As discussed earlier, we alleviate this concern by including worker fixed effects in the
regression, and focusing on workers who had worked at the displacing firm for at least three
years before the mass layoff. In Section 3.3, we perform additional robustness checks to
validate our estimates.
2.3 Coarsened Exact Matching
To identify the fully dynamic effects of job displacement, we also need a control group of
workers who are not displaced but have comparable characteristics to those of the

displaced workers (Krolikowski, 2018; Borusyak/Jaravel/Spiess, 2021). To this end, we
employ coarsened exact matching (CEM, Iacus/King/Porro, 2012) as described below.
To begin with, we select a pool of non-displaced workers eligible for CEM using the same
restrictions as for displaced workers. For each worker displaced in year τ , we identify
workers who were not displaced in that year, had been employed full-time and not changed
employers in the three years before τ (Krolikowski, 2018; Blien/Dauth/Roth, 2021;
Meekes/Hassink, 2019). Next, we match the displaced and non-displaced workers using
their characteristics in the year before displacement. The matching variables are selected
following Schmieder/von Wachter/Heining (2020). We use exact matching on gender,
educational attainment, one-digit industrial sector, and western versus eastern Germany.
We also use coarsened matching on age, firm tenure, firm size, and annual earnings. The
procedure yields a matched sample comprising 18,190 displaced workers and 273,142
non-displaced workers.
Table 1 reveals that displaced and non-displaced workers have balanced characteristics
after CEM. In both groups, roughly one-fourth of the matched workers are female, and the
average age prior to mass layoffs is 41 years old. The majority of those workers have
vocational training (77 percent) and live in western Germany (83 percent). At the time of the
mass layoff, an average worker has 15 years of total work experience and eight years of
experience with the current employer; moreover, the average worker earns 100 euros per
day and drives 15 kilometers (14 minutes) one-way between home and work. In
comparison, Stutzer/Frey (2007) and Giménez-Nadal/Molina/Velilla (2022) report the
average one-way driving time for all German workers to be 21-22 minutes. In addition,
Figure A1.1 in Appendix A1 reports the event study estimates using the regular (full-time or
part-time) employment indicator as the dependent variable. We find no pretrend in
workers’ employment status. Hence, the matching procedure yields balanced
pre-displacement characteristics.
Finally, we use the matched worker sample to construct the worker-year panel for analysis.
Each worker is tracked for a maximum of four years before and 11 years after displacement,
and is observed in the year when he is in regular employment on June 30. In total, the panel
comprises 3,213,415 yearly observations from the 291,332 matched workers. Table A1.1 in
Appendix A1 reports summary statistics for the worker-year panel.

3 Results
We estimate model (1) using ordinary least squares (OLS). Table 2 reports estimates of all βk
and Figure 1 plots the estimates with 95 percent confidence intervals. In Table 3, we also
present the DID estimates. All standard errors are clustered by workers.
First of all, we estimate the effect of job displacement on log daily wages and log annual
earnings. From Table 2 and Figure 1, (a)-(b), we draw two major conclusions. First, workers’
daily wage and annual earnings drop significantly upon job displacement. From the year
before displacement to the year afterwards, the average daily wage of displaced workers
falls by 19.6 percent, and the average annual earnings drop by 43.2 percent. The larger
effect on earnings is due to the reduced number of days worked by the displaced workers.
Second, both outcomes gradually recover over subsequent years. Ten years after
displacement, earnings and wages are roughly 17 percent lower than their
pre-displacement levels. In addition, the DID estimates in Table 3 reveal that a displaced
worker’s average earnings decrease by 20 percent and her wage by 25 percent over all
post-displacement years. These estimates are consistent with previous studies using the
same data (Burdett/Carrillo-Tudela/Coles, 2020; Jarosch, 2021).
Next, we estimate the effect of job displacement on workers’ commuting behavior. In Table
2 and Figure 1, (c)-(d), workers’ commuting distance and commuting time both increase
significantly upon job displacement and gradually decline in subsequent years. Relative to
the year before displacement, the average displaced worker commutes a 21.8 percent
longer distance and 18.7 percent longer time in the year after displacement (conditional on
having found a new regular job). Ten years after displacement, the average commuting
distance and commuting time have both recovered by one-half, but still exceed the
pre-displacement level by 9.9 and 8 percent, respectively. According to the DID estimates in
Table 3, the long-run increases in commuting distance and commuting time amount to 15.7
and 13.3 percent, respectively.
Using a simple back-of-the-envelope analysis, we calculate the pecuniary

value of the increased commuting for displaced workers. According to government
statistics (Stutzer/Frey, 2007; Destatis, 2018), the average hourly wage in Germany
is 20 euros and the average commuting time is 0.7 hours a day. As such, job displacement
increases the commuting time of an average German worker by 0.7 × 0.187 = 0.131 hours
per day or 0.131 × 250 = 32.75 hours in the first year after displacement. This is equivalent
to 32.75 × 20 = 655 euros of opportunity cost in annual earnings. Repeating this exercise for
all post-displacement years, we find the long-run opportunity cost of increased commuting
time to be 4,820 euros, or 3,955 euros in present value (assuming a discount factor of 0.95).

3.1 Mechanism
What explains the dynamics of commuting after job displacement? To reduce commuting
between home and workplace, workers can either switch to an employer closer to their
homes or move their homes closer to the employer. To determine which drives the recovery
of commuting, we estimate model (1) using the indicators of firm change and relocation as
the dependent variable, respectively. Firm changes are identified from changes in the firm
identifier, and relocation from changes in the place of residence (at the geo-coordinate
level). In the latter case, we also exploit the home location of unemployed workers. In
Figure 2 and Table 4, firm changes and relocation both increase in the first few years after
job displacement, but the increase in relocation is of a much smaller magnitude and does
not persist in subsequent years. Therefore, the long-term recovery of commuting is mostly
driven by firm changes. This validates Huttunen/Møen/Salvanes (2018) argument that
relocation of displaced workers is driven by non-economic factors and commuting is a more
effective response to job displacement than migration.
3.2 The Value of Granularity
A major advantage of our study is due to the granular commuting data. For each individual
worker, we know the driving distance between their exact place of residence and place of
work alone the most likely routes. In comparison, existing research on commuting often
identifies workers and firms at the geographic regional level and calculates the average
distance across regions (Eliasson/Lindgren/Westerlund, 2003; Fackler/Rippe, 2017;
Meekes/Hassink, 2019, 2022). In this subsection, we compare our estimates with those
using regional-level distances and illustrate the value of having granular commuting data.
We argue that measuring distances at the regional level is associated with three types of
errors. First, it omits all commutes within regions. This causes a censoring problem where
within-region commuters are assumed not to commute at all. Second, the distance
travelled by individual workers rarely coincide with the distance between two regional
centers, which is often used to define cross-region distances. Especially if cross-regional
commuters are more likely to live and work near regional borders, their commuting
distances will be overestimated at the regional level. Third, if the distance between two
regions is measured along the geodesic (the shortest path between two geo-coordinates), it
will underestimate the actual distance travelled along the roads.

To quantify the severity of (each of) the errors, we generate several commuting measures by
municipality, the lowest level of administrative areas in Germany.4 Then we compare the
estimates of model (1) using these measures as the outcome variable. The resulting
estimates are reported in Table 5 and visualized in Figure 3.
In Table 5, column (a), we force the log distance of within-municipality commutes to zero
and leave distances across municipalities unchanged. This is to shutdown commutes within
municipalities. Compared with Table 2, column (c), the results overestimate the effect of job
displacement on commuting by 16.8 to 32.3 percent. Averaged across all post-displacement
years, the bias amounts to roughly one-fourth of the true effect. The sign of the bias
suggests that, although displaced workers increase commuting relative to non-displaced
workers, many of the commutes still occur within municipalities.
Next, we calculate the driving distance between municipality centers using

OpenStreetMaps.org and substitute this for the driving distances along each individual’s
route. Hence, all workers commuting from municipality A to B are assumed to travel the
same distance. As shown In Table 5, column (b), the increase in commuting is still
overestimated, but to a lesser extent. Furthermore, we replace the driving distance between
municipalities by the geodesic between municipality centers. As geodesics ignore
geographic variations in the commuting routes across municipalities, the resulting
estimates are further driven down to an average of 7.5 percent, and a maximum of 14.4
percent each year (Table 5, column (c)).
To summarize, the use of aggregated commuting data yields significantly overestimated

effects of job loss. The overall bias consists of two countervailing biases: the omission of
within-municipality commutes leads to overestimation, and the aggregation of
cross-municipality commuting causes underestimation. It is worth noting that the largest
bias occurs in the first few years after job displacement, when the increase in commuting is
also the greatest. In Figure A1.2, we obtain robust results by performing the same
decomposition for commuting by districts Kreis, a more aggregated and commonly studies
level of regions in Germany (Schmidtlein/Seth/Vom Berge, 2020).
3.3 Additional Estimates and Robustness Checks
To conclude this section, we estimate the effect of job loss on other outcomes following
existing research, and present various robustness checks regarding our main results.
4
Germany has 11,014 municipalities in 2020. An average municipality has a population of 7,557.7 and an area
of 32.47 square kilometers.

Effect on Firm Wage Premiums. Related to the job displacement effect on wages,
Fackler/Mueller/Stegmaier (2021) argue that a major component of the wage losses is due
to workers moving to low-paying firms. Similarly, we estimate the impact of job loss on
firm-specific wage premiums, measured by Card/Heining/Kline (2013) estimates of firm
fixed effects. As shown in Figure A1.3 and Table A1.2, part (a), we find a negative and
persistent impact of job displacement on the firm fixed effects, and the effect magnitude
corresponds to half of the estimated wage loss.
Effect on Job Types. Next, job displacement could affect wages and commuting via
workers’ choice of flexible or regular jobs (Meekes/Hassink, 2022). In particular, displaced
workers take part-time or mini jobs as stepping stones back to full-time employment. In
Figure A1.3 and Table A1.2, parts (b)-(d), we estimate the effect of job loss on workers’ labor
supply in mini, part-time, and full-time jobs, measured by the number of days worked in
each job type. Immediately after displacement, workers first take up part-time and mini
jobs, in which their labor supply increases sharply. Four years after displacement, full-time
jobs start to catch up and labor supply in other types levels off or starts to decline. With
worker fixed effects, these effects are driven by within-worker adjustment of labor supply
rather than changes in worker composition.
Attrition. In contrast to the seminal event study design (Jacobson/LaLonde/Sullivan,

1993b), our estimation focuses on an unbalanced panel where workers are employed in
regular jobs. Thus, attrition could arise if the worker (i) is employed in a mini job or
temporarily unemployed; (ii) becomes a civil servant or self-employed; (iii) goes back to
school for vocational or advanced education; or (iv) goes abroad or leaves the workforce.
Referring to the unemployment registration data, we find the majority (70 percent) of
missing observations are accounted for by unemployment. To rule out attrition bias, we
estimate model (1) on a subsample of workers without any attrition. Specifically, we focus
on displaced workers who find another regular job in two years and remains in employment
until the seventh year after the mass layoff, and estimate their displacement effects up to
seven years. The estimates in Figure A1.4 are consistent with the main results.
Recalls. Our estimates are not driven by recalled workers, who account for 10 to 20
percent of all displaced workers (Leenders, 2021; Jost, 2022). For these workers, commuting
distance recovers to the pre-displacement level mechanically. In Figure A1.5, we exclude
workers recalled by the displacing firms and reestimate model (1). The estimated effects on
all outcomes are slightly larger but robust.
Alternative Sample Restriction. In Figure A1.6, we exclude worker-years in which the

main job on June 30 is part-time. There is little change in the estimates.

Alternative Estimation Method. Finally, we examine robustness to dynamic treatment
effects and treatment effect heterogeneity using the imputation method proposed by
Borusyak/Jaravel/Spiess (2021). As shown in Figure A1.7, the results are in line with the
original estimates using two-way fixed effects.
4 Does Commuting Mitigate the Wage

Loss?
The above estimates suggest that, on average, job displacement lowers the wage of
displaced workers and increases their commuting costs. However, the average
displacement effects might conceal sizeable heterogeneity across workers. On one hand,
workers could mitigate the wage losses by bearing higher commuting costs, leading to a
positive correlation between the two outcomes. On one hand, due to different labor market
conditions or job search effort, some workers may experience, relative to other workers, a
greater wage loss as well as a larger increase in commuting. In what follows, we find
empirical evidence for both mechanisms.
To investigate the mitigating effect, we modify model (2) and assume worker-specific
average displacement effects. Then we formulate a random coefficient model
−2
X
Yit = [αk I{t = τ (i) + k} + βk I{t = τ (i) + k}M Li ]
k=−4 (1)
+ αY∗ i I{t ≥ τ (i)} + βY∗ i I{t ≥ τ (i)}M Li + Xit′ γ + ϕi + ψt + ϵit , Y = W, C
where βW ∗ and β ∗ represent the job displacement effect on worker i’s wage and
i Ci
commuting distance, respectively. Then we calculate the correlation between βW ∗ and β ∗ .
i Ci
Following Verdier (2020), we estimate a fully saturated model with an interaction between
I{t ≥ τ (i)}M Li and all worker indicators. The resulting coefficients identify the
displacement effect on each displaced worker i. We find the correlation coefficient between
∗ and β̂ ∗ to be 0.084, which is positive and statistically significant.5 We interpret the
β̂W i Ci
correlation as the mitigating effect among displaced workers: They can attenuate their
wage losses from job displacement by accepting a larger increase in commuting distance.
5
These figures provide a lower bound of the true correlation. Since each worker is observed a maximum of
16 times, each individual displace effect is estimated with potentially large noise. Thus, the variances of
both β̂W i and β̂Ci will overstate the true variances of displacement effects (Kline/Saggio/Sølvsten, 2020).
∗ ∗
However, the covariance is consistently estimated if the estimation errors in β̂W i and β̂Ci from different
∗ ∗ ∗ ∗
equations are uncorrelated. In total, the correlation coefficient Cov(βW i , βCi )/[Var(βW i )Var(βCi )]1/2 is
underestimated.

Moreover, we apply the mitigating effect to compare the job displacement effects between
male and female workers. Existing studies show that after losing a job, women suffer from a
greater wage penalty, but increase commuting less than men (Illing/Schmieder/Trenkle,
2021; Le Barbanchon/Rathelot/Roulet, 2021; Meekes/Hassink, 2022). This is illustrated in
Figure 4, where women’s post-displacement earnings and wages are both lower than men’s
by almost 10 percentage points. Meanwhile, women’s commuting distance and commuting
time are lower than men’s, especially from the fifth year after displacement. Estimating the
random coefficient model separately, we find robust positive correlations of 0.079 for men
and 0.097 for women. As such, we argue that women’s commuting decisions are more
sensitive to the wage loss after job displacement.
In Appendix A1, we conduct more split-sample analysis. In Figures A1.8-A1.9, we estimate

model (1) for workers below and above the sample-medium age upon mass layoff (43 years
old) and residing in urban and rural areas, respectively. In Figures A1.10-A1.11, we compare
the effects on high-wage and low-wage workers and displacement from high-wage and
low-wage firms (Card/Heining/Kline, 2013). However, we do not find any significant
difference in the job displacement effects on commuting. As such, we cannot rule out the
other effect that wages and commuting are negatively correlated, i,e., workers who
experience a greater wage loss also increase commuting to a larger extent.
To summarize, we find two possible ways in which wages and commuting are correlated
after job loss. If displaced workers face different local labor market conditions, some will
generally find a job more easily than others, and experience a smaller wage loss associated
with less increase in commuting. In contrast, taking external job opportunities as given,
displaced workers will trade off wage losses with longer commuting. This gives rise to a
positive correlation between the two outcomes.
5 A Job Search Model
In this section, we develop a discrete-time job search model to rationalize the findings
above. Readers are referred to Appendix A2 for proof.
5.1 Setup
Consider a continuum of infinitely-lived workers and firms in a linear city R+ . All workers
are ex ante homogeneous and located at the origin. Following the empirical findings, we

assume they do not move home. In contrast, firms are ex ante heterogeneous in
productivity and locations. Thus, we denote a firm-type by θ = (y, r), where
y ∈ [y, ȳ] ⊂ R++ is the productivity and r is the distance between the firm and the origin. In
what follows, we assume y and r are independent, and their marginal distributions Fy and
Fr are twice differentiable with density functions fy and fr , respectively.
In every period of unemployment, a worker receives unemployment benefits z and a job

offer with probability λ0 . When employed at firm θ, he earns a flow wage w(θ, θ̂), incurs a
commuting cost c(r), and receives an job offer with probability λ1 . In particular, w(θ, θ̂)
depends on the current firm type θ and the worker’s best outside option θ̂; c(r) strictly
increases in r; and c(0) = 0. Throughout this section, we assume all workers receive job
offers from an exogenous distribution Fθ .6 All jobs dissolve with probability δ per period.
When a firm of type θ meets a worker with outside option θ̂, the value of a match is equal to
J(θ, θ̂) for the firm and W (θ, θ̂) for the worker. When unmatched or unemployed, the firm
gets zero value and the worker gets U ≡ W (θ, u). A job offer turns into a job match if and
only if the resulting joint surplus
S(θ) = max{0, W (θ, θ̂) − U + J(θ, θ̂)}, (1)
is strictly positive. Note that S(θ) does not depend on θ̂ because θ̂ only affects how the
surplus is split between the firm and the worker. No surplus is created by an unemployed
worker or a vacant firm, so we write S(u) = 0.
Wages are determined by Nash bargaining. With on-the-job search, bargaining takes the
form of sequential auctions (Cahuc/Postel-Vinay/Robin, 2006; Postel-Vinay/Robin, 2002a,b).
When an unemployed worker meets a firm of type θ, the worker bargains for an exogenous
share α ∈ [0, 1] of the joint surplus:
W (θ, u) − U = αS(θ). (2)
When a worker currently employed at firm θ1 with outside option θ̂ receives an offer from
firm θ2 , one of the following will happen. If S(θ2 ) > S(θ1 ), the worker will move to the new
firm and the new wage is determined by
W (θ2 , θ1 ) − U = S(θ1 ) + α(S(θ2 ) − S(θ1 )). (3)
If S(θ̂) < S(θ2 ) < S(θ1 ), the old firm retains the worker at a higher wage to match the better
6
As discussed in Section 2, the average number of workers displaced in a mass layoff is negligible relative to
the size of a local labor market. Therefore, mass layoffs are unlikely to have general equilibrium effects on
the job offer distribution or labor market tightness (Flemming, 2020).

outside option θ2 . The renegotiated wage satisfies
W (θ1 , θ2 ) − U = S(θ2 ) + α(S(θ1 ) − S(θ2 )). (4)
Finally, if S(θ2 ) ≤ S(θ̂), the worker remains at the old firm at the current wage. In what
follows, we use M1 (θ1 ) to represent the set of offers that incur a move to the new firm, and
M2 (θ1 , θ2 ) for those leading to wage renegotiation within the old firm. The corresponding
probability measures are p1 (θ1 ) and p2 (θ1 , θ2 ), respectively.
Now we pin down the value functions U , W , and J . They are respectively given by
( Z )
U =z+β λ0 W (x, u)dFθ (x) + (1 − λ0 p1 (u))U , (5)
M1 (u)
n
W (θ, θ̂) = w(θ, θ̂) − c(r) + β δU + (1 − δ)[1 − λ1 (p1 (θ) + p2 (θ, θ̂)]W (θ, θ̂)
"Z Z #)
(6)
+(1 − δ)λ1 W (x, θ)dFθ (x) + W (θ, x)dFθ (x) ,
M1 (θ) M2 (θ,θ̂)
and
( Z )
J(θ, θ̂) = y − w(θ, θ̂) + β̃ λ1 J(θ, x)dFθ (x) + [1 − λ1 (p1 (θ) + p2 (θ, θ̂)]J(θ, θ̂) , (7)
M2 (θ,θ̂)
where β̃ = β(1 − δ).
5.2 Dynamics of Surplus, Commuting, and Wage
To investigate the dynamics of wage and commuting costs after job displacement, we
associate both outcomes with the job match surplus. Plugging the value functions (5)-(7)
and wage bargaining rules (2)-(4) into (1), we obtain the following equation regarding the
surplus function:
˜ 1 Z Z
y − c(r) − z αβλ αβλ0
S(θ) = + [S(x) − S(θ)] dFθ (x) −
+
[S(x)]+ dFθ (x). (8)
1 − β̃ 1 − β̃ 1 − β̃
Given the exogenous parameters, the job characteristics θ = (y, r) uniquely pin down the
value of S(θ). As shown in Appendix A2, Lemma 2, the surplus level increases in the firm’s
productivity y but decreases in the commuting distance r. At the same surplus level,
workers’ commuting costs c(r) must be compensated one-to-one by the firm’s increased
productivity.

Motivated by this observation, we define isosurplus curves on the job space [y, ȳ] × R+ .
Figure 5 plots three isosurplus curves assuming the commuting cost is linear in distance. All
jobs (y, r) on the same curve yield the same surplus, and the curve on a higher position
represents a greater surplus level.
Proposition 1. Suppose a worker is exogenously displaced in period 0 and does not

experience another separation by period T . The worker’s total matched surplus S(θ) and their
surplus share W (θ, θ̂) − U are both non-decreasing in t for 0 ≤ t ≤ T .
Proposition 1 states that voluntary job transitions will never reduce a worker’s surplus level.
Next, we associate the recovery of surplus with the dynamics of productivity, commuting,
and wages.
Proposition 2. Let E[y |s] and E[r|s] be, respectively, the expected productivity and expected
distance of a firm that generates match surplus s. Also, define Fc (·) = Fr (c−1 (·)). If
(i) Fc is twice differentiable with density function fc ;
(ii) fc′ (x)/fc (x) and fy′ (x)/fy (x) are non-increasing in x;
then E[y|s] strictly increases in s and E[r|s] strictly decreases in s.
Therefore, workers with a higher job match surplus, in expectation, are matched to more
productive and more proximate firms. This is obvious when y and r are uniformly
distributed: In Figure 5, the conditional expectation is given by the midpoint of the
isosurplus curve. Proposition 2 extends this result to general distributions, provided the
density functions are not too volatile or skewed.
Combining the two propositions, we can characterize the recovery of displaced workers
over time. After a worker is displaced, the expected productivity of the new employers
increases over time, and the expected commuting distance decreases over time. This
process will continue until he is displaced again. As Jarosch (2021) points out, if α is high,
i.e., the worker gets a large share of the joint surplus, the expected wage will also increase
over time.7
In Appendix A3, we consider two extensions of the model: (i) endogenous search effort and
(ii) firm-specific separation rates. In the first extension, we show that displaced workers
search for jobs less intensively over time. However, this declining search effort is driven by
recovery of surplus, so it differs from the effect of exhausting unemployment benefits. In the
second case, workers move from less to more secure jobs as they recover from job loss. This
7
As Jarosch (2021) explains, when α is small, workers at more productive firms may accept a lower wage in
exchange for greater wage growth potential.

provides another explanation for our findings in Section 3.3—that workers view job security
as an independent job attribute and on-the-job search allows them to move from less to
more secure jobs.
5.3 Empirical Validation
According to Proposition 1, displaced workers will accept a job offer only if it yields a
positive surplus, i.e., lies above the zero-surplus curve. To empirically validate this claim,
we plot the distribution of first jobs accepted by displaced workers. Figure 6, panel (a)
depicts the productivity y and commuting distance r of first jobs accepted by a displaced
worker, where productivity is proxied by firm fixed effects (Card/Heining/Kline, 2013). As
distance increases, the distribution of firm productivity shifts up, and the likelihood of
observing a low productivity firms diminishes. In particular, most of the points lie above a
hypothetical, upward sloping curve as implied by the zero-surplus curve. This pattern
becomes more pronounced in panel (b), where we replace the firm fixed effects by log
wages. The implied lower bound represents an upward-sloping indifference curve between
wages and commuting. The slope of the indifference curve measures the worker’s
willingness to pay (WTP) for commuting.
The patterns in Figure 6 align our dynamic model with the static ones found in Dauth/Haller
(2020) and Le Barbanchon/Rathelot/Roulet (2021). Using the same data as ours,
Dauth/Haller (2020) estimate the WTP for commuting to be 0.06 for an average German
worker. Combining survey data with actual job acceptance in France, Le
Barbanchon/Rathelot/Roulet (2021) directly identify the slope of the indifference curve and
study gender gaps in the valuation for commuting. Our model differs from theirs by
considering on-the-job search and the shift in indifference curves themselves. With
on-the-job search, workers increase wages and reduce commuting at the same time,
yielding the opposite relationship between wages and commuting with respect to the static
case.
6 Conclusion
In this paper, we investigate the adjustment of commuting behavior by displaced workers.

Using an event study approach, we estimate the short-term and long-term impact of being
displaced during mass layoffs on the worker’s earnings, wages, and commuting costs.

Consistent with existing studies, we find a large and persistent loss in earnings and wages
after job displacement. In contrast, we show that displaced workers commute longer
distances after displacement, and their commuting patterns gradually recover to the
pre-displacement level in subsequent years. Further analysis reveals that the recovery of
commuting is driven by workers moving to proximate firms rather than migration or
relocation. Besides, we provide evidence that workers attenuate the wage loss after job
displacement by increasing commuting, or vice versa.
The declining commuting and increasing wage of post-displacement workers contrasts with
their static relationship, where longer commuting distances are compensated for by higher
wages. To reconcile the countervailing forces, we build a job search model with
heterogeneous firm productivity and commuting costs. We show that on-the-job search
plays an important role. With on-the-job search, workers can increase their job match
surplus by moving from less to more productive firms and moving from distant to proximate
firms. Whereas conditional on a fixed surplus level, they make a tradeoff between higher
wages and shorter commuting, leading to the opposite correlation. Using data on the first
jobs taken by displaced workers, we validate that workers rarely take up jobs with both low
pay and high commuting costs.
A large literature has documented the profound consequences of job loss on individual
workers. We contribute to the literature by demonstrating the multi-dimensional nature of
the impacts of job loss. Not only do displaced workers experience a negative shock on
wages and earnings, but they also face increased commuting costs to find new jobs. In
addition, displaced workers may suffer from poorer job match quality, job security, etc.
Hence, it would be interesting for future researchers to consider the effect of job loss on
other outcome dimensions and evaluate the welfare loss of displaced workers (e.g.,
Meekes/Hassink, 2022). Our multi-dimensional model in Appendix A3.2 provides a feasible
starting point.
For policymakers, our paper sheds light on the value of employment assistance programs.
Except for cash benefits and skill training, measures to reduce job search frictions would
also assist unemployed workers. Future research could examine whether commuting
subsidies, online job boards, or working-from-home options facilitate the return to
employment by displaced workers, and whether they accelerate the recovery from job loss.
Franklin (2018) and Gürtzgen et al. (2021) are two leading examples using experimental and
survey methods, respectively. It would also be valuable to study how job search frictions
impact the effectiveness of employment assistance programs.
Finally, our study highlights the importance of granular commuting data for studying
individuals’ responses to job loss. In other research areas, we see a similar value of
measuring commuting at the individual level. For example, studies of job search, labor

market frictions, monopsony, and social networks all emphasize individuals’ commuting
decisions. Hence, we also expect oppotunites for multidisciplinary research between
economics and geographical science.

References
Asher, Sam; Novosad, Paul (2020): Rural Roads and Local Economic Development. In:
American Economic Review, Vol. 110, No. 3, p. 797–823.
Barrero, Jose Maria; Bloom, Nicholas; Davis, Steven J. (2021): Why Working from Home Will
Stick. In: NBER Working Paper No. w28731.
Bertheau, Antoine; Acabbi, Edoardo; Barceló, Cristina; Gulyas, Andreas; Lombardi, Stefano;
Saggio, Raffaele (2021): The Unequal Cost of Job Loss across Countries. In: , working
Paper.
Blien, Uwe; Dauth, Wolfgang; Roth, Duncan H.W. (2021): Occupational Routine Intensity and
the Costs of Job Loss: Evidence from Mass Layoffs. In: Labour Economics, Vol. 68, p.
101 953.
Borusyak, Kirill; Jaravel, Xavier; Spiess, Jann (2021): Revisiting Event Study Designs: Robust
and Efficient Estimation. In: , working Paper.
Brooks, Wyatt J.; Donovan, Kevin (2020): Eliminating Uncertainty in Market Access: The
Impact of New Bridges in Rural Nicaragua. In: Econometrica, Vol. 88, No. 5, p.
1965–1997.
Burdett, Kenneth; Carrillo-Tudela, Carlos; Coles, Melvyn (2020): The Cost of Job Loss. In:
Review of Economic Studies, Vol. 87, No. 4, p. 1757–1798.
Burdett, Kenneth; Mortensen, Dale T. (1998): Wage Differentials, Employer Size, and
Unemployment. In: International Economic Review, p. 257–273.
Cahuc, Pierre; Postel-Vinay, Fabien; Robin, Jean-Marc (2006): Wage Bargaining with
On-the-Job Search: Theory and Evidence. In: Econometrica, Vol. 74, No. 2, p. 323–364.
Card, David; Heining, Jörg; Kline, Patrick (2013): Workplace Heterogeneity and the Rise of
West German Wage Inequality. In: Quarterly journal of economics, Vol. 128, No. 3, p.
967–1015.
Couch, Kenneth A.; Placzek, Dana W. (2010): Earnings Losses of Displaced Workers
Revisited. In: American Economic Review, Vol. 100, No. 1, p. 572–589.
Dauth, Wolfgang; Eppelsheimer, Johann (2020): Preparing the Sample of Integrated Labour
Market Biographies (SIAB) for Scientific Analysis. In: Journal for Labour Market
Research, Vol. 54, No. 1, p. 10.
Dauth, Wolfgang; Haller, Peter (2020): Is There Loss Aversion in the Trade-Off between
Wages and Commuting Distances? In: Regional Science and Urban Economics, Vol. 83,
p. 103 527.

Davis, Steven J.; von Wachter, Till (2011): Recessions and the Costs of Job Loss. In:
Brookings Papers on Economic Activity, Vol. 2011, No. 2, p. 1–72.
Destatis (2018): Average Hourly Wage of Persons Employed. In: , (accessed 2022-02-28)
https://www.destatis.de/EN/Themes/Labour/Labour-Market/Quality-Employment/Dimension2/2_5_Hour
Destatis (2017): Commuting in Germany: 68% Use a Car to Commute to Work. In: , (accessed
2022-02-28)
https://www.destatis.de/DE/ZahlenFakten/ImFokus/Arbeitsmarkt/PendlerArbeitsweg.html.
Eliasson, Kent; Lindgren, Urban; Westerlund, Olle (2003): Geographical Labour Mobility:
Migration or Commuting? In: Regional studies, Vol. 37, No. 8, p. 827–837.
Fackler, Daniel; Mueller, Steffen; Stegmaier, Jens (2021): Explaining Wage Losses after Job
Displacement: Employer Size and Lost Firm Wage Premiums. In: Journal of the
European Economic Association, Vol. 19, No. 5, p. 2695–2736.
Fackler, Daniel; Rippe, Lisa (2017): Losing Work, Moving Away? Regional Mobility after Job
Loss. In: Labour, Vol. 31, No. 4, p. 457–479.
Fitzenberger, Bernd; Osikominu, Aderonke; Völter, Robert (2005): Imputation Rules to

Improve the Education Variable in the IAB Employment Subsample. In: ZEW-Centre for
European Economic Research Discussion Paper, , No. 05-010.
Flaaen, Aaron; Shapiro, Matthew D.; Sorkin, Isaac (2019): Reconsidering the Consequences
of Worker Displacements: Firm versus Worker Perspective. In: American Economic
Journal: Macroeconomics, Vol. 11, No. 2, p. 193–227.
Flemming, Jean (2020): Costly Commuting and the Job Ladder. In: FEDS Working Paper No.
2020-025.
Franklin, Simon (2018): Location, Search Costs and Youth Unemployment: Experimental
Evidence from Transport Subsidies. In: Economic Journal, Vol. 128, No. 614, p.
2353–2379.
Giménez-Nadal, José Ignacio; Molina, José Alberto; Velilla, Jorge (2022): Trends in
Commuting Time of European Workers: A Cross-Country Analysis. In: Transport Policy,
Vol. 116, p. 327–342.
Goldschmidt, Deborah; Schmieder, Johannes F. (2017): The Rise of Domestic Outsourcing

and the Evolution of the German Wage Structure. In: Quarterly Journal of Economics,
Vol. 132, No. 3, p. 1165–1217.
Guglielminetti, Elisa; Lalive, Rafael; Ruh, Philippe; Wasmer, Etienne (2019): Home Sweet
Home? Job Search with Commuting and Unemployment Insurance. In: , working Paper.

Gürtzgen, Nicole; Diegmann, André; Pohlan, Laura; van den Berg, Gerard J. (2021): Do Digital
Information Technologies Help Unemployed Job Seekers Find a Job? Evidence from
the Broadband Internet Expansion in Germany. In: European Economic Review, Vol.
132, p. 103 657.
Huber, Stephan; Rust, Christoph (2016): Calculate Travel Time and Distance with
OpenStreetMap Data Using the Open Source Routing Machine (OSRM). In: Stata
Journal, Vol. 16, No. 2, p. 416–423.
Huttunen, Kristiina; Møen, Jarle; Salvanes, Kjell G. (2018): Job Loss and Regional Mobility.
In: Journal of Labor Economics, Vol. 36, No. 2, p. 479–509.
Iacus, Stefano M.; King, Gary; Porro, Giuseppe (2012): Causal Inference without Balance
Checking: Coarsened Exact Matching. In: Political Analysis, Vol. 20, No. 1, p. 1–24.
Illing, Hannah; Schmieder, Johannes; Trenkle, Simon (2021): The Gender Gap in Earnings
Losses after Job Displacement. In: , working Paper.
Jacobson, Louis S.; LaLonde, Robert J.; Sullivan, Daniel G. (1993a): The Costs of Worker
Dislocation. Kalamazoo, MI: WE Upjohn Institute for Employment Research.
Jacobson, Louis S.; LaLonde, Robert J.; Sullivan, Daniel G. (1993b): Earnings Losses of
Displaced Workers. In: American Economic Review, Vol. 83, No. 4, p. 685–709.
Jarosch, Gregor (2021): Searching for Job Security and the Consequences of Job Loss. In:
NBER Working Paper No. w28481.
Jost, Oskar (2022): See You Soon. Fixed-Term Contracts, Unemployment and Recalls in
Germany: A Linked Employer-Employee Analysis. In: Working Paper.
Jost, Ramona (2020): Persistence of Commuting Habits: Context Effects in Germany. In: IAB
Discussion Paper No. 14/2020.
Kennan, John; Walker, James R (2011): The Effect of Expected Income on Individual
Migration Decisions. In: Econometrica, Vol. 79, No. 1, p. 211–251.
Kline, Patrick; Saggio, Raffaele; Sølvsten, Mikkel (2020): Leave-Out Estimation of Variance
Components. In: Econometrica, Vol. 88, No. 5, p. 1859–1898.
Krolikowski, Pawel (2018): Choosing a Control Group for Displaced Workers. In: ILR Review,
Vol. 71, No. 5, p. 1232–1254.
Le Barbanchon, Thomas; Rathelot, Roland; Roulet, Alexandra (2021): Gender Differences in

Job Search: Trading Off Commute against Wage. In: Quarterly Journal of Economics,
Vol. 136, No. 1, p. 381–426.
Leenders, Frank (2021): Recall and the Scarring Effects of Job Displacement. In: , working
Paper.

Manning, Alan; Petrongolo, Barbara (2017): How Local Are Labor Markets? Evidence from a
Spatial Job Search Model. In: American Economic Review, Vol. 107, No. 10, p.
2877–2907.
Marinescu, Ioana; Rathelot, Roland (2018): Mismatch Unemployment and the Geography of
Job Search. In: American Economic Journal: Macroeconomics, Vol. 10, No. 3, p. 42–70.
Meekes, Jordy; Hassink, Wolter H.J. (2022): Gender Differences in Job Flexibility: Commutes
and Working Hours after Job Loss. In: Journal of Urban Economics, p. 103 425.
Meekes, Jordy; Hassink, Wolter H.J. (2019): The Role of the Housing Market in Workers’
Resilience to Job Displacement after Firm Bankruptcy. In: Journal of Urban Economics,
Vol. 109, p. 41–65.
Moreno-Monroy, Ana I.; Posada, Héctor M. (2018): The Effect of Commuting Costs and
Transport Subsidies on Informality Rates. In: Journal of Development Economics, Vol.
130, p. 99–112.
Mulalic, Ismir; Van Ommeren, Jos N.; Pilegaard, Ninette (2014): Wages and Commuting:
Quasi-Natural Experiments’ Evidence from Firms That Relocate. In: Economic Journal,
Vol. 124, No. 579, p. 1086–1105.
Postel-Vinay, Fabien; Robin, Jean-Marc (2002a): Equilibrium Wage Dispersion with Worker
and Employer Heterogeneity. In: Econometrica, Vol. 70, No. 6, p. 2295–2350.
Postel-Vinay, Fabien; Robin, Jean-Marc (2002b): The Distribution of Earnings in an

Equilibrium Search Model with State-Dependent Offers and Counteroffers. In:
International Economic Review, Vol. 43, No. 4, p. 989–1016.
Ransom, Tyler (2021): Labor Market Frictions and Moving Costs of the Employed and
Unemployed. In: Journal of Human Resources, p. 0219–10 013R2.
Schmidtlein, Lisa; Seth, Stefan; Vom Berge, Philipp (2020): Sample of Integrated Employer
Employee Data (SIEED) 1975–2018. In: .
Schmieder, Johannes; von Wachter, Till; Heining, Jörg (2020): The Costs of Job
Displacement over the Business Cycle and Its Sources: Evidence from Germany. In: ,
working Paper.
Schmutz, Benoît; Sidibé, Modibo (2019): Frictional Labour Mobility. In: Review of Economic
Studies, Vol. 86, No. 4, p. 1779–1826.
Stevens, Ann Huff (1997): Persistent Effects of Job Displacement: The Importance of
Multiple Job Losses. In: Journal of Labor Economics, Vol. 15, No. 1, p. 165–188.
Stutzer, Alois; Frey, Bruno S. (2007): Commuting and Life Satisfaction in Germany. In:
Informationen zur Raumentwicklung, Vol. 2, No. 3, p. 179–189.

Tazhitdinova, Alisa (2020): Do Only Tax Incentives Matter? Labor Supply and Demand
Responses to an Unusually Large and Salient Tax Break. In: Journal of Public
Economics, Vol. 184, p. 104 162.
Van den Berg, Gerard J.; Gorter, Cees (1997): Job Search and Commuting Time. In: Journal
of Business & Economic Statistics, Vol. 15, No. 2, p. 269–281.
Verdier, Valentin (2020): Average Treatment Effects for Stayers with Correlated Random
Coefficient Models of Panel Data. In: Journal of Applied Econometrics, Vol. 35, No. 7, p.
917–939.

Tables and Figures
Table 1: Coarsened Exact Matching: Results

Variable Unit Displaced Non-Displaced
Female Indicator 0.285 0.264
(0.452) (0.441)
Age Year 40.53 40.95
(8.082) (7.994)
High school or less Indicator 0.178 0.177
(0.383) (0.382)
Vocational training Indicator 0.765 0.771
(0.424) (0.420)
University or above Indicator 0.048 0.043
(0.213) (0.204)
Eastern Germany Indicator 0.177 0.170
(0.381) (0.376)
Work experience Year 14.77 15.52
(7.575) (7.714)
Firm tenure Year 8.336 8.924
(6.391) (6.718)
Firm size Count 224.9 218.3
(760.8) (784.0)
Daily wage Euro 99.78 101.7
(48.23) (49.91)
Commuting distance Kilometer 15.36 14.63
(15.93) (15.07)
Commuting time Minute 14.98 14.37
(12.95) (12.42)
Workers 18,190 273,142
Note: The table reports means and standard deviations (in parentheses) of displaced workers and matched
non-displaced workers in the year before mass layoff. The sample is obtained using exact matching on gender,
education qualification, one-digit industrial sector, and indicator of eastern Germany, and coarsened matching
on age, firm tenure, firm size, and average daily wage in 2015 real prices.
Source: IEB V14, 2000–2017.

Table 2: Effects of Job Displacement on Wages and Commuting: Event Study Estimates
(a) (b) (c) (d)
Daily wage Annual Commuting Commuting
(log) earnings (log) distance (log) time (log)
β−4 0.014*** 0.013*** -0.009 -0.006
(0.002) (0.002) (0.008) (0.006)
β−3 0.009*** 0.008*** -0.005 -0.002
(0.002) (0.002) (0.005) (0.004)
β−2 0.007*** 0.010*** -0.003 -0.002
(0.001) (0.001) (0.003) (0.003)
β0 -0.013*** -0.132*** -0.006** -0.005*
(0.002) (0.002) (0.003) (0.003)
β1 -0.196*** -0.432*** 0.218*** 0.187***
(0.004) (0.006) (0.016) (0.014)
β2 -0.230*** -0.334*** 0.206*** 0.177***
(0.004) (0.005) (0.014) (0.012)
β3 -0.230*** -0.281*** 0.176*** 0.152***
(0.004) (0.005) (0.014) (0.012)
β4 -0.217*** -0.248*** 0.172*** 0.147***
(0.004) (0.005) (0.015) (0.012)
β5 -0.204*** -0.228*** 0.158*** 0.135***
(0.004) (0.005) (0.015) (0.013)
β6 -0.191*** -0.209*** 0.145*** 0.122***
(0.005) (0.005) (0.016) (0.013)
β7 -0.185*** -0.198*** 0.135*** 0.113***
(0.005) (0.005) (0.016) (0.014)
β8 -0.181*** -0.188*** 0.122*** 0.100***
(0.005) (0.006) (0.017) (0.014)
β9 -0.172*** -0.183*** 0.110*** 0.091***
(0.005) (0.006) (0.018) (0.015)
β10 -0.164*** -0.174*** 0.099*** 0.080***
(0.006) (0.006) (0.019) (0.016)
β11 -0.156*** -0.160*** 0.088*** 0.073***
(0.006) (0.007) (0.020) (0.017)
Observations 3,213,415 3,213,415 3,213,415 3,213,415
Workers 291,332 291,332 291,332 291,332
R2 0.055 0.194 0.005 0.005
Notes: Each column represents estimates of model (1) with the dependent variable in the column title. Samples
comprise a yearly panel of workers with regular jobs on June 30 of the year. All estimates control for a polynomial
of age, worker fixed effects, calendar year fixed effects, and year relative to displacement fixed effects. Robust
standard errors in parentheses clustered by workers. (* p < 0.1; ** p < 0.05; *** p < 0.01)
Source: IEB V14, 2000–2017.

Table 3: Effects of Job Displacement on Wages and Commuting: DID Estimates
(a) (b) (c) (d)
Daily wage Annual Commuting Commuting
(log) earnings (log) distance (log) time (log)
β−4 0.014*** 0.015*** -0.010 -0.007
(0.002) (0.002) (0.008) (0.006)
β−3 0.009*** 0.008*** -0.005 -0.002
(0.002) (0.002) (0.005) (0.004)
β−2 0.008*** 0.010*** -0.003 -0.002
(0.001) (0.001) (0.003) (0.003)
β0 -0.013*** -0.132*** -0.006** -0.005*
(0.002) (0.002) (0.003) (0.003)
β∗ -0.200*** -0.250*** 0.157*** 0.133***
(0.004) (0.004) (0.013) (0.011)
Observations 3,213,415 3,213,415 3,213,415 3,213,415
Workers 291,332 291,332 291,332 291,332
R2 0.194 0.055 0.005 0.005
Source: IEB V14, 2000–2017.

Table 4: Effects of Job Displacement on Worker Mobility
(a) (b)
Relocation Firm change
β−4 0.010** 0.002**
(0.004) (0.001)
β−3 0.004 0.014***
(0.004) (0.003)
β−2 -0.001 0.007**
(0.004) (0.003)
β0 0.009** 0.000
(0.004) (0.000)
β1 0.049*** 0.852***
(0.006) (0.006)
β2 0.053*** 0.596***
(0.005) (0.006)
β3 0.023*** 0.298***
(0.004) (0.006)
β4 0.019*** 0.198***
(0.004) (0.005)
β5 0.021*** 0.145***
(0.005) (0.005)
β6 0.016*** 0.116***
(0.005) (0.005)
β7 0.014*** 0.095***
(0.005) (0.005)
β8 0.019*** 0.075***
(0.005) (0.005)
β9 0.011** 0.066***
(0.005) (0.006)
β10 0.014** 0.061***
(0.005) (0.006)
β11 0.006 0.049***
(0.005) (0.006)
Observations 3,213,415 3,213,415
Workers 291,332 291,332
R2 0.010 0.059
Notes: Each column presents estimates of model (1) with the dependent variable in the column title. Samples
Source: IEB V14, 2000–2017.

Table 5: Granular versus Coarse Commuting Distances
(a) (b) (c)
No commute within Single route btw. Geodesic btw.
municipality municipalities municipalities
β−4 -0.014 -0.014 -0.015
(0.010) (0.009) (0.010)
β−3 -0.003 -0.004 -0.005
(0.006) (0.006) (0.006)
β−2 -0.000 -0.001 -0.001
(0.004) (0.003) (0.004)
β0 -0.007** -0.006* -0.006*
(0.004) (0.003) (0.004)
β1 0.268*** 0.244*** 0.256***
(0.021) (0.020) (0.021)
β2 0.257*** 0.232*** 0.245***
(0.018) (0.017) (0.018)
β3 0.228*** 0.201*** 0.213***
(0.018) (0.017) (0.018)
β4 0.209*** 0.183*** 0.194***
(0.019) (0.017) (0.018)
β5 0.194*** 0.168*** 0.179***
(0.019) (0.018) (0.019)
β6 0.175*** 0.151*** 0.162***
(0.020) (0.019) (0.020)
β7 0.158*** 0.135*** 0.146***
(0.021) (0.019) (0.020)
β8 0.152*** 0.128*** 0.137***
(0.022) (0.020) (0.021)
β9 0.140*** 0.119*** 0.127***
(0.023) (0.021) (0.023)
β10 0.131*** 0.111*** 0.117***
(0.024) (0.023) (0.024)
β11 0.109*** 0.090*** 0.096***
(0.026) (0.024) (0.025)
Observations 3,213,415 3,204,807 3,204,807
Workers 291,332 290,747 290,747
R2 0.005 0.005 0.005
Notes: Each column presents estimates of model (1) using the following distance measures as the dependent
variable. Column (a): driving distance between geo-coordinates of an individual’s residence and workplace, but
zero log distance assumed for commutes within municipalities; Column (b): distance for commutes from one
municipality to another replaced by driving distance between municipality centers; Column (c): driving distance
between municipalities replaced by geodesic between municipality centers. Samples comprise a yearly panel
of workers with regular jobs on June 30 of the year. All estimates control for a polynomial of age, worker fixed
effects, calendar year fixed effects, and year relative to displacement fixed effects. Robust standard errors in
parentheses clustered by workers. (* p < 0.1; ** p < 0.05; *** p < 0.01)
Source: IEB V14, 2000–2017.

Figure 1: Effects of Job Displacement on Wages and Commuting
(a) Daily wage (log) (b) Annual earnings (log)
(c) Commuting distance (log) (d) Commuting time (log)
Notes: Each plot depicts estimates of model (1) with the dependent variable in the subtitle. Samples comprise a yearly panel of workers with regular jobs on June 30 of the year. The
shaded area indicates the displacement year, and each dot/bar represents the point estimate/95 percent CI of βk . All estimates control for a polynomial of age, worker fixed effects,
calendar year fixed effects, and year relative to displacement fixed effects. Robust standard errors clustered by workers.
Source: IEB V14, 2000–2017.
Figure 2: Effects of Job Displacement on Worker Mobility
Notes: The figure depicts estimates of model (1) with indicators of relocation and firm change as the dependent
variable. Samples comprise a yearly panel of workers with regular jobs on June 30 of the year. The shaded
area indicates the displacement year, and each dot/bar represents the point estimate/95 percent CI of βk . All
estimates control for a polynomial of age, worker fixed effects, calendar year fixed effects, and year relative to
displacement fixed effects. Robust standard errors clustered by workers.
Source: IEB V14, 2000–2017.

Figure 3: Individual- and Municipality-Level Commuting Distances
Notes: The figure depicts estimates of model (1) using various distance measures as the dependent variable.
The gray area represents the driving distance between geo-coordinates of an individual’s residence and work-
place (the baseline); the blue triangles assume zero log distance for commutes within municipalities; the red
circles further calculate the distance for all commutes from one municipality to another by the driving dis-
tance between municipality centers; the black stars further replace driving distances between municipalities
by geodesics between municipality centers. All estimates control for a polynomial of age, worker fixed effects,
calendar year fixed effects, and year relative to displacement fixed effects. Robust standard errors clustered by
workers.
Source: IEB V14, 2000–2017.

Figure 4: Effect Heterogeneity by Gender
Notes: Each plot depicts estimates of model (1) with the dependent variable in the subtitle. Samples comprise a yearly panel of workers with regular jobs on June 30 of the year. The
blue and red markers represent male and female workers, respectively. The shaded area indicates the displacement year, and each dot/bar represents the point estimate/95 percent
CI of βk . All estimates control for a polynomial of age, worker fixed effects, calendar year fixed effects, and year relative to displacement fixed effects. Robust standard errors clustered
by workers.
Source: IEB V14, 2000–2017.
Figure 5: Isosurplus Curves with c(r) = r
Notes: This figure depicts three isosurplus curves under the assumption c(r) = r. The support of productivity
y and distance r are given by [y, ȳ] = [0, 1] and R+ , respectively.
Source: Own depiction.

Figure 6: First Jobs after Displacement
(a) Productivity and Distance
(b) Wage and Distance
Notes: Sample comprises all first jobs accepted by displaced workers. Panel (a) plots Card/Heining/Kline’s
(2013) estimated firm fixed effects and commuting distances; Panel (b) plots average daily wages and commut-
ing distances (jobs with log daily wage above 5 are not shown).
Source: IEB V14, 2000–2017.

Appendix
A1 Additional Tables and Figures
Table A1.1: Summary Statistics

Variables Unit Mean S.D. Min. Max.
Female Indicator 0.103 0.304 0 1
Age Year 43.08 7.353 20 54
High school or less Indicator 0.097 0.296 0 1
Vocational training Indicator 0.883 0.322 0 1
University or above Indicator 0.019 0.137 0 1
Eastern Germany Indicator 0.074 0.261 0 1
Work experience Year 19.64 7.651 0.003 39
Firm tenure Year 13.06 8.290 0.003 39
Firm size Count 313.2 1,309 0 57,420
Full-time job Indicator 0.988 0.111 0 1
Daily wage Euro 111.9 42.55 0.011 1,369
Annual earnings Euro 36,662 11,619 3.650 311,510
Commuting distance Kilometer 13.66 13.62 0.003 100.0
Commuting time Minute 13.51 11.31 0.017 104.5
Firm change Indicator 0.054 0.225 0 1
Relocation Indicator 0.069 0.253 0 1
Observations 3,213,415
Workers 291,332
Note: The sample comprises a yearly panel of displaced workers and matched non-displaced workers using
CEM. Earnings and wages are in 2015 real prices.
Source: IEB V14, 2000–2017.

Table A1.2: Effects of Job Displacement on Other Outcomes
(a) (b) (c) (d)
Firm wage Days in Days in Days in
premiums full-time jobs part-time jobs mini jobs
β−4 -0.010*** 0.436 -2.136*** 0.231
(0.001) (0.317) (0.388) (0.905)
β−3 -0.005*** -0.820*** -1.787*** 4.948***
(0.001) (0.139) (0.152) (0.493)
β−2 -0.003*** -0.073 -0.04 0.773***
0.000 (0.052) (0.033) (0.230)
β0 0.006*** 0.858*** 0.428*** -34.427***
0.000 (0.089) (0.066) (0.490)
β1 -0.101*** 11.548*** 3.523*** -207.022***
(0.003) (0.422) (0.402) (1.133)
β2 -0.118*** 15.399*** 12.458*** -134.668***
(0.002) (0.589) (0.704) (1.478)
β3 -0.111*** 15.886*** 15.500*** -95.378***
(0.002) (0.667) (0.867) (1.510)
β4 -0.105*** 16.117*** 16.775*** -81.861***
(0.003) (0.740) (0.974) (1.578)
β5 -0.100*** 14.059*** 17.907*** -74.692***
(0.003) (0.749) (1.071) (1.623)
β6 -0.092*** 13.459*** 17.088*** -68.135***
(0.003) (0.783) (1.149) (1.684)
β7 -0.082*** 12.849*** 16.999*** -64.001***
(0.003) (0.813) (1.242) (1.741)
β8 -0.083*** 11.727*** 16.295*** -57.321***
(0.003) (0.857) (1.386) (1.850)
β9 -0.077*** 10.925*** 18.234*** -56.583***
(0.003) (0.902) (1.531) (1.984)
β10 -0.074*** 10.202*** 17.444*** -56.766***
(0.003) (0.954) (1.676) (2.124)
β11 -0.069*** 11.188*** 17.130*** -54.336***
(0.003) (1.035) (1.794) (2.247)
Observations 3,209,856 3,335,347 3,335,347 3,335,347
Workers 291,329 291,922 291,922 291,922
R2 0.918 0.012 0.043 0.097
comprise a yearly panel of workers with regular jobs on June 30 of the year (column (a)) or a complete yearly
panel of workers (columns (b)-(d)). All estimates control for a polynomial of age, worker fixed effects, calendar
year fixed effects, and year relative to displacement fixed effects. Robust standard errors in parentheses clus-
tered by workers. (* p < 0.1; ** p < 0.05; *** p < 0.01)
Source: IEB V14, 2000–2017.

Figure A1.1: Mass Layoffs and Regular Employment
Notes: The figure depicts estimates of model (1) with the probability of having regular employment as the de-
pendent variable. The sample comprises a complete yearly panel of workers. The shaded area indicates the
displacement year, and each dot/bar represents the point estimate/95 percent CI of βk . All estimates control
for a polynomial of age, worker fixed effects, calendar year fixed effects, and year relative to displacement fixed
effects. Robust standard errors clustered by workers.
Source: IEB V14, 2000–2017.

Figure A1.2: Individual- and District-Level Commuting Distances
Notes: The figure depicts estimates of model (1) using various distance measures as the dependent variable. The
gray area represents the driving distance between geo-coordinates of an individual’s residence and workplace
(the baseline); the blue triangles assume zero log distance for commutes within districts; the red circles further
calculate the distance for all commutes from one district to another by the driving distance between district
centers; the black stars further replace driving distances between districts by geodesics between district centers.
All estimates control for a polynomial of age, worker fixed effects, calendar year fixed effects, and year relative
to displacement fixed effects. Robust standard errors clustered by workers.
Source: IEB V14, 2000–2017.

IAB-Discussion Paper 15|2022
Figure A1.3: Effects of Job Displacement on Other Outcomes

(a) Firm wage premiums (b) Days in mini jobs
(c) Days in part-time jobs (d) Days in full-time jobs
Notes: Each plot depicts estimates of model (1) with the dependent variable in the subtitle. In panel (a), firm wage premiums are measured by Card/Heining/Kline’s (2013) estimated
firm fixed effects. Samples comprise a yearly panel of workers with regular jobs on June 30 (panel (a)) and a complete yearly panel or workers (panels (b)-(d)). The shaded area
indicates the displacement year, and each dot/bar represents the point estimate/95 percent CI of βk . All estimates control for a polynomial of age, worker fixed effects, calendar year
fixed effects, and year relative to displacement fixed effects. Robust standard errors clustered by workers.
Source: IEB V14, 2000–2017.
46
Figure A1.4: Robustness to Attrition

Notes: Each plot depicts estimates of model (1) with the dependent variable in the subtitle. Samples comprise a yearly panel of workers with regular jobs on June 30 from the second
to the seventh year after displacement. The shaded area indicates the displacement year, and each dot/bar represents the point estimate/95 percent CI of βk . All estimates control
for a polynomial of age, worker fixed effects, calendar year fixed effects, and year relative to displacement fixed effects. Robust standard errors clustered by workers.
Source: IEB V14, 2000–2017.
47
Figure A1.5: Robustness to Recalls

Notes: Each plot depicts estimates of model (1) with the dependent variable in the subtitle. Samples comprise a yearly panel of workers with regular jobs on June 30, excluding
displaced workers recalled by their pre-displacement firms. The shaded area indicates the displacement year, and each dot/bar represents the point estimate/95 percent CI of βk .
All estimates control for a polynomial of age, worker fixed effects, calendar year fixed effects, and year relative to displacement fixed effects. Robust standard errors clustered by
workers.
Source: IEB V14, 2000–2017.
48
Figure A1.6: Robustness to Excluding Part-Time Jobs

Notes: Each plot depicts estimates of model (1) with the dependent variable in the subtitle. Samples comprise a yearly panel of workers with regular jobs on June 30, limited to
full-time jobs. The shaded area indicates the displacement year, and each dot/bar represents the point estimate/95 percent CI of βk . All estimates control for a polynomial of age,
worker fixed effects, calendar year fixed effects, and year relative to displacement fixed effects. Robust standard errors clustered by workers.
Source: IEB V14, 2000–2017.
49
Figure A1.7: Robustness to Alternative Estimation Method

Notes: Each plot depicts estimates of model (1) with the dependent variable in the subtitle, using the imputation method of Borusyak/Jaravel/Spiess (2021). Samples comprise a yearly
panel of workers with regular jobs on June 30. The shaded area indicates the displacement year, and each dot/bar represents the point estimate/95 percent CI of βk . All estimates
control for a polynomial of age, worker fixed effects, calendar year fixed effects, and year relative to displacement fixed effects. Robust standard errors clustered by workers.
Source: IEB V14, 2000–2017.
50
Figure A1.8: Effect Heterogeneity by Age

Notes: Each plot depicts estimates of model (1) with the dependent variable in the subtitle. Samples comprise a yearly panel of workers with regular jobs on June 30. The black
and red markers represent workers under and over the age of 43 the displacement year, respectively. The shaded area indicates the displacement year, and each dot/bar represents
the point estimate/95 percent CI of βk . All estimates control for a polynomial of age, worker fixed effects, calendar year fixed effects, and year relative to displacement fixed effects.
Robust standard errors clustered by workers.
Source: IEB V14, 2000–2017.
51
Figure A1.9: Effect Heterogeneity by Urban and Rural Residence

Notes: Each plot depicts estimates of model (1) with the dependent variable in the subtitle. Samples comprise a yearly panel of workers with regular jobs on June 30. The black and
red markers represent workers living in urban and rural districts (Kreise) in the displacement year, respectively. The shaded area indicates the displacement year, and each dot/bar
represents the point estimate/95 percent CI of βk . All estimates control for a polynomial of age, worker fixed effects, calendar year fixed effects, and year relative to displacement
fixed effects. Robust standard errors clustered by workers.
Source: IEB V14, 2000–2017.
52
Figure A1.10: Effect Heterogeneity by Displacing Firms’ Fixed Effects

Notes: Each plot depicts estimates of model (1) with the dependent variable in the subtitle. Samples comprise a yearly panel of workers with regular jobs on June 30. The black and
red markers represent workers displaced from firms whose estimated fixed effects (Card/Heining/Kline, 2013) are below and above the sample median, respectively. The shaded
area indicates the displacement year, and each dot/bar represents the point estimate/95 percent CI of βk . All estimates control for a polynomial of age, worker fixed effects, calendar
year fixed effects, and year relative to displacement fixed effects. Robust standard errors clustered by workers.
Source: IEB V14, 2000–2017.
53
Figure A1.11: Effect Heterogeneity by Worker Fixed Effects

Notes: Each plot depicts estimates of model (1) with the dependent variable in the subtitle. Samples comprise a yearly panel of workers with regular jobs on June 30. The black
and red markers represent workers whose estimated fixed effects (Card/Heining/Kline, 2013) are below and above the sample median, respectively. The shaded area indicates the
displacement year, and each dot/bar represents the point estimate/95 percent CI of βk . All estimates control for a polynomial of age, worker fixed effects, calendar year fixed effects,
and year relative to displacement fixed effects. Robust standard errors clustered by workers.
Source: IEB V14, 2000–2017.
54
A2 Proof
A2.1 Proof for Proposition 1
When a worker is unemployed, his total surplus S(0) and surplus share W (0, 0) − U are
both zero and cannot decrease further. When he is employed at firm type θ with outside
option θ̂, in the next period he either moves to a new firm, renegotiates a higher wage at the
current firm, or stays at the current wage. In the first two cases, his share of surplus
increases according to either (3) and (4), and the total surplus either increases or stays the
same. In the last case, the total surplus and the share remain unchanged. The monotonicity
of W (θ, θ̂) − U follows immediately.
A2.2 Proof for Proposition 2
Since the joint surplus s increases over time, it suffices to prove that the expected
commuting distance declines, and the expected productivity increases in s.
Lemma 1. The surplus function S strictly increases in y and strictly decreases in r.
Proof. See Jarosch (2021).
Lemma 2. The surplus function S is continuous and linear in y and c(r) with opposite slopes.
Proof. Prove by contradiction. Suppose S is linear in y and c(r) with opposite slopes, such
that y1 − c(d1 ) = y2 − c(d2 ), but S(y1 , r1 ) =
̸ S(y2 , r2 ). Then (8) implies
ηλ1 (1 + α)
S(y1 , r1 ) − S(y2 , r2 ) =
1−η
Z (A2.1)

[S(x) − S(y1 , r1 )]+ − [S(x) − S(y2 , r2 )]+ dFθ (x),
which is not possible because the two sides have different signs. Continuity can be proved
in an analogous way.

As a result, we can express the surplus function as S(y, r) = a(y − c(r) + b) for some
constants a > 0 and b. Let Ls = {(y, r) : S(y, r) = s, r ≥ 0} denote the isosurplus curve
corresponding to the surplus level s. To circumvent the unknown function c, we define
L̃s = {(y, c) : S(y, c) = s, c ≥ 0}, the isomorphism of Ls in the space of (y, c). Along this
curve, we have c = y − s/a + b.
The expectation of y conditional on s is given by

Z
E[y|s] = ydF (y, c|s), (A2.2)
˜s
L
where F (y, c|s) is the conditional distribution of (y, c) given L̃s . By independence of y and
c(r), the density function is
s
fy (y)fc (c) f y (y)f c y − + b
f (y, c|s) = Z = Z y¯ a . (A2.3)
s √
dF (y, c|s) fy (y)fc y − + b 2dy
˜s
L ys a
where ys = s/a − b is the intercept of L̃s on the (y, c)-plane.
Plug (A2.3) into (A2.2),

Z y¯
s
yfy (y)fc y − + b dy
y a Y1
E[y|s] = Z sȳ ≡ . (A2.4)
s Y2
fy (y)fc y − + b dy
ys a
The derivatives of Y1 and Y2 with respect to s are, respectively,

Z y¯
1 s 1
Y1′ = − yfy (y)fc′ y − + b dy − ys fy (ys )fc (0) (A2.5)
a ys a a
Z y¯
1 s 1
Y2′ = − fy (y)fc′ y − + b dy − fy (ys )fc (0) (A2.6)
a ys a a
Note that
Z y¯
s
fy (ys )fc (0) (y − ys )fy (y)fc y − + b dy > 0, (A2.7)
ys a

Z ȳ
s
yfy (y)fc′ y − + b dy ′
a f (c)
= E y c s ,
ys
Z ȳ (A2.8)
s fc (c)
ys a
Z ȳ
s
fy (y)fc′ y − + b dy ′
a f (c)
= E c s ,
ys
Z ȳ (A2.9)
s fc (c)
ys a
and, because y and fc′ (c)/fc (c) are negatively correlated on L̃s ,
′ ′
fc′ (c) fc (c) f (c)
Cov y, s =E y s − E c s E [ y| s] < 0. (A2.10)
fc (c) fc (c) fc (c)
Combining (A2.7)-(A2.10), we can find
∂E[y|s] Y ′ Y2 − Y2′ Y1
= 1 > 0, (A2.11)
∂s Y22
Analogously, the expectation of r conditional on s is given by

Z
E[r|s] = c(c)dF (y, c|s), (A2.12)
L̃s
where c(·) = c−1 (·). As before, rewrite

Z c̄s
s
c(c)fy c + − b fc (c)dc
a R1
E[r|s] = 0Z c̄s ≡ . (A2.13)
s R2
fy c + − b fc (c)dc
0 a
where c̄s = (ȳ − s/a + b).
The derivatives of R1 and R2 with respect to s are, respectively,

Z c̄s
1 1 s
R1′ = − c(c̄s )fy (ȳ)fc (c̄s ) + c(c)fy′ c + − b fc (c)dc, (A2.14)
a a 0 a
Z c̄s
1 1 s
R2′ = − fy (ȳ)fc (c̄s ) + fy′ c + − b fc (c)dc, (A2.15)
a a 0 a

Since c(c) − c(c̄s ) < 0,
Z c¯s
s
fy (ȳ)fc (c̄s ) (c(c) − c(c̄s ))fy c + − b fc (c)dc < 0 (A2.16)
0 a
Following the previous steps,

Z c¯s
s
c(c)fy′ c + − b fc (c)dc
a fy′ (y)
0
Z c¯s = E c(c) s , (A2.17)
s fy (y)
0 a
Z c¯s
s
fy′ c + − b fc (c)dc ′
a fy (y)
Z c¯s
0
=E s . (A2.18)
s fy (y)
0 a

fy′ (y)
Cov c(c), s < 0. (A2.19)
fy (y)
Combining (A2.16)-(A2.19), we can find
∂E[r|s] R′ R2 − R2′ R1
= 1 < 0. (A2.20)
∂s R22

A3 Model Extensions
A3.1 Endogenous Job Search Effort
In this subsection, we extend our model to allow for endogenous search effort. Suppose the
worker pays a search cost h(R) to receive job offers from his neighboring locations [0, R],
where h(R) is increasing and convex in the radius R. The increasing cost of search with
respect to distance has been established in the empirical literature (Manning/Petrongolo,
2017; Marinescu/Rathelot, 2018). For traceability, we assume y and r are independent.
With the search cost, the value functions of unemployed and employed workers become
( " Z R Z ȳ
U = max z − h(R) + β λ0 W (y, r, y0 , 0)dFy (y)dFr (r)
R 0 c(r)−b (A3.1)
+(1 − λ0 p1 (u))U ]} ,
ˆ r̂) = max {w(y, r, y,

W (y, r, y, ˆ r̂) − c(r) − h(R) + βδU
R
Z R Z ȳ
+ β(1 − δ)λ1 W (y ′ , r ′ , y, r)dFy (y ′ )dFr (r′ )
S(y,r)
0 c(r ′ )+a
−b
Z RZ ′ S(y,r)
c(r )+ a −b
+ β(1 − δ)λ1 W (y, r, y ′ , r′ )dFy (y ′ )dFr (r′ )
S(ŷ,r̂)
0 c(r ′ )+ −b
Z
a
! )
R Z ȳ
+β(1 − δ) 1 − λ1 dFy (y ′ )dFr (r′ ) W (y, r, y,
ˆ r̂) .
S(ŷ,r̂)
0 c(r ′ )+ a
−b
(A3.2)
In either case, the optimal search radius R∗ satisfies

Z ȳ
h′ (R∗ ; u) = βλ0 W (y, R∗ , y0 , 0)dFy (y). (A3.3)
c(R∗ )−b
(Z
ȳ
h′ (R∗ ; y, r, y,
ˆ r̂) =β(1 − δ)λ1 W (y ′ , R∗ , y, r) − W (y, r, y,
ˆ r̂) dFy (y ′ )
S(y,r)
c(R∗ )+ a −b
Z )
c(R∗ )+
S(y,r)
a
−b
′ ∗ ′
+ W (y, r, y , R ) − W (y, r, y,
ˆ r̂) dFy (y ) .
S(ŷ,r̂)
c(R∗ )+ a
−b
(A3.4)

Note that h′ (R) is the increasing marginal cost of search, and the right-hand side (RHS) of
(A3.3) and (A3.4) are the present-value marginal payoffs from search. Since W (y, r, y,
ˆ r̂)
∗
decreases in r and r̂, the RHS diminishes in R, so R is unique.
A natural question that arises here is whether the optimal search radius decreases during
the recovery from displacement. To answer this question, we use (3)-(4) to rewrite the RHS
of (A3.4) as β(1 − δ)λ1 times
Z ȳ Z ȳ
α [S(y ′ , R∗ ) − S(y, r)]dFy (y ′ ) + (1 − α) [S(y, r) − S(ŷ, r̂)]dFy (y ′ )
S(y,r) S(ŷ,r̂)
c(R∗ )+ a −b c(R∗ )+ a −b
(A3.5)
Note that (A3.5) decreases in S(ŷ, r̂). Therefore, a greater outside option lowers the
marginal payoff from search, thereby reducing the optimal search radius.8
Proposition 3. The optimal search radius R∗

(i) decreases in S(ŷ, r̂);
(ii) decreases in S(y, r) if α is sufficiently large.
Proof. Examine the partial derivatives of (A3.5) with respect to S(ŷ, r̂) and S(y, r).
The first part of Proposition 3 suggests that, following a wage renegotiation with the current
employer, a worker will optimally reduce his search radius because S(ŷ, r̂) increases while
S(y, r) remains unchanged. However, the second part implies that if the worker moves to
another firm, the search radius unambiguously decreases only if the worker appropriates a
large share α of the additional surplus.
A3.2 Heterogeneous Job Security
In the second extension, we consider heterogeneous job separation rates. Assume δ is

firm-specific and distributed in the support (0, 1). With slight abuse of notation, we let
θ = (y, r, δ) include the separation rate as a third attribute of the firm. Rearranging (8) yields
y = c(r) + δG(S(θ)) + H(S(θ)). (A3.6)
8
Guglielminetti et al. (2019) suggest the exhaustion of unemployment benefits as a countervailing force driv-
ing up the search radius over time. However, this effect is muted once the worker finds a job after displace-
ment.

where
Z
G(s) = β s + λ1 (1 + α) [S(x) − s] dFθ (x) ,
+
(A3.7)
Z Z
H(s) = s − β s + λ1 (1 + α) [S(x) − s] dFθ (x) + λ0 [S(x)] dFθ (x) .
+ +
(A3.8)
As such, y , δ , and c(r) are linearly dependent given surplus level S(θ). It is easy to verify that
S strictly decreases in δ and both G and H increase in s.
Analogous to (8), (A3.6) defines an isosurplus surface on the three-dimensional job space
[y, ȳ] × R+ × (0, 1). As depicted in Figure A3.1, a higher surface indicates a higher surplus;
besides, the maximum job separation rate and maximum distance both decrease in the
surplus level. By a similar argument to Proposition 2, workers enjoy increased job security
as they recover from job loss.
The above results differ from Jarosch (2021) in that we assume firm productivity and
separation rate are independent. As such, their joint recovery is explicitly driven by workers’
tradeoffs rather than mechanical correlation between the two attributes. It is worth noting
that our framework can be generalized to incorporate multiple firm attributes.

Figure A3.1: Isosurplus Surfaces with c(r) = r
Notes: This figure depicts three isosurplus surfaces under the assumption c(r) = r. The support of productivity
y , distance r, and separation rate δ are given by [y, ȳ] = [0, 1], R+ , and (0, 1), respectively.
Source: Own depiction.

List of Figures
Figure 1: Effects of Job Displacement on Wages and Commuting . . . . . . . . . . . . . . . . . . . . . . . 36

Figure 2: Effects of Job Displacement on Worker Mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Figure 3: Individual- and Municipality-Level Commuting Distances. . . . . . . . . . . . . . . . . . . . . 38
Figure 4: Effect Heterogeneity by Gender . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Figure 5: Isosurplus Curves with c(r) = r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Figure 6: First Jobs after Displacement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Figure A1.1: Mass Layoffs and Regular Employment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Figure A1.2: Individual- and District-Level Commuting Distances. . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Figure A1.3: Effects of Job Displacement on Other Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Figure A1.4: Robustness to Attrition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Figure A1.5: Robustness to Recalls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Figure A1.6: Robustness to Excluding Part-Time Jobs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Figure A1.7: Robustness to Alternative Estimation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Figure A1.8: Effect Heterogeneity by Age . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Figure A1.9: Effect Heterogeneity by Urban and Rural Residence . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Figure A1.10: Effect Heterogeneity by Displacing Firms’ Fixed Effects . . . . . . . . . . . . . . . . . . . . . . . 53
Figure A1.11: Effect Heterogeneity by Worker Fixed Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Figure A3.1: Isosurplus Surfaces with c(r) = r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
List of Tables
Table 1: Coarsened Exact Matching: Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

Table 2: Effects of Job Displacement on Wages and Commuting: Event Study Estimates32
Table 3: Effects of Job Displacement on Wages and Commuting: DID Estimates . . . . . . . . 33
Table 4: Effects of Job Displacement on Worker Mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Table 5: Granular versus Coarse Commuting Distances. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Table A1.1: Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Table A1.2: Effects of Job Displacement on Other Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

Imprint
Publication Date
15 June 2022
Publisher
Institute for Employment Research
of the Federal Employment Agency
Regensburger Straße 104
90478 Nürnberg
Germany
All rights reserved

This publication is published under the following Creative Commons licence: Attribution -
ShareAlike 4.0 International (CC BY-SA 4.0)
https://creativecommons.org/licenses/by-sa/4.0/deed.de
Download
https://doku.iab.de/discussionpapers/2022/dp1522.pdf
All publications in the series “IAB-Discusssion Paper” can be downloaded from

https://www.iab.de/en/publikationen/discussionpaper.aspx
Website
www.iab.de/en
ISSN
2195-2663
DOI
10.48720/IAB.DP.2215
Corresponding author
Oskar Jost
Telefon +49 (0) 30 5555-99-5916
E-Mail [email protected]

DP 1522

Uploaded by

Copyright:

Available Formats

DP 1522

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

DP 1522

Uploaded by

Copyright:

Available Formats

IAB-DISCUSSION PAPER

Articles on labour market issues

15|2022Beyond Lost Earnings: The Long-Term

Yige Duan, Oskar Jost, Ramona Jost

Yige Duan (University of British Columbia (UBC))

IAB-Discussion Paper 15|2022 2

2 Data and Empirical Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

4 Does Commuting Mitigate the Wage Loss?. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

5 A Job Search Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

A1 Additional Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

IAB-Discussion Paper 15|2022 3

J3, J6, R23, R4

IAB-Discussion Paper 15|2022 4

commuting, mobility, displacement, job search

IAB-Discussion Paper 15|2022 5

Job displacement has profound consequences on individual workers. A large body of

IAB-Discussion Paper 15|2022 6

IAB-Discussion Paper 15|2022 7

1.1 Related Literature

Our paper contributes to a long literature on earning losses due to unemployment

IAB-Discussion Paper 15|2022 8

IAB-Discussion Paper 15|2022 9

As in Schmieder/von Wachter/Heining (2020) and Illing/Schmieder/Trenkle (2021), we

IAB-Discussion Paper 15|2022 10

To further distinguish mass layoffs from voluntary job-to-job transition, ﬁrm

2.2 Identiﬁcation Strategy

IAB-Discussion Paper 15|2022 11

2.3 Coarsened Exact Matching

IAB-Discussion Paper 15|2022 12

IAB-Discussion Paper 15|2022 13

Using a simple back-of-the-envelope analysis, we calculate the pecuniary

IAB-Discussion Paper 15|2022 14

3.2 The Value of Granularity

IAB-Discussion Paper 15|2022 15

Next, we calculate the driving distance between municipality centers using

To summarize, the use of aggregated commuting data yields signiﬁcantly overestimated

3.3 Additional Estimates and Robustness Checks

IAB-Discussion Paper 15|2022 16

Attrition. In contrast to the seminal event study design (Jacobson/LaLonde/Sullivan,

Alternative Sample Restriction. In Figure A1.6, we exclude worker-years in which the

IAB-Discussion Paper 15|2022 17

4 Does Commuting Mitigate the Wage

IAB-Discussion Paper 15|2022 18

In Appendix A1, we conduct more split-sample analysis. In Figures A1.8-A1.9, we estimate

5 A Job Search Model

IAB-Discussion Paper 15|2022 19

In every period of unemployment, a worker receives unemployment beneﬁts z and a job

S(θ) = max{0, W (θ, θ̂) − U + J(θ, θ̂)}, (1)

W (θ, u) − U = αS(θ). (2)

W (θ2 , θ1 ) − U = S(θ1 ) + α(S(θ2 ) − S(θ1 )). (3)

IAB-Discussion Paper 15|2022 20

W (θ1 , θ2 ) − U = S(θ2 ) + α(S(θ1 ) − S(θ2 )). (4)

where β̃ = β(1 − δ).

5.2 Dynamics of Surplus, Commuting, and Wage

IAB-Discussion Paper 15|2022 21

Proposition 1. Suppose a worker is exogenously displaced in period 0 and does not

IAB-Discussion Paper 15|2022 22

5.3 Empirical Validation

In this paper, we investigate the adjustment of commuting behavior by displaced workers.