Danmarks Nationalbank, "Job Retention during the Covid-19 Pandemic" (Working Paper No. 186, January 2022)
Summary
Danmarks Nationalbank Working Paper No. 186, dated 13 January 2022, by Pernille Valentin Borgensgaard, on the labor market effects of the Danish wage compensation for employees, which paid wage subsidies to private firms that furloughed workers instead of laying them off. Using monthly payroll records for employer-employee matches, the paper finds that the subsidy prevented at least 10,400 exits from employment. It also compensated workers whom firms were not planning to lay off. The effect was concentrated among workers with one year or less of tenure, while workers with 8+ years would less often have been laid off anyway. Furloughed workers' labor income fell significantly, and labor market mobility was basically unaffected.
Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used
Cite this document
“Danmarks Nationalbank, ‘Job Retention during the Covid-19 Pandemic’ (Working Paper No. 186, January 2022).”
Original at www.nationalbanken.dk. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/danmarks-nationalbank-job-retention-during-the-covid-19-pandemic-worki-a1e31ee7cf/
, SHA-256 c603c4d5edb09ea666970a61babaa80702353324494f37eacc8c262d3bea06ce.Full text
1 3 JANUARY 2 02 2 — NO. 186
Job Retention during the
Covid-19 Pandemic
Pernille Valentin The Working Papers of Danmarks Nationalbank describe research
Borgensgaard and development, often still ongoing, as a contribution to the
professional debate.
pvb@nationalbanken.dk
DANMARKS NATIONALBANK The viewpoints and conclusions stated are the responsibility of
the individual contributors, and do not necessarily reflect the views
of Danmarks Nationalbank.
WORK IN G PA PER — DANMARKS NATIONALBANK
13 JANUARY 2 02 2 — NO. 186
Job Retention during the Covid-19 Pandemic
Abstract Resumé
Following the outbreak of the covid-19 pandemic, Da covid-19 pandemien brød ud, valgte mange
countries worldwide adopted new or scaled up lande at introducere nye eller opskalere
existing job retention schemes. This paper eksisterende jobfastholdelsesordninger. Dette
examines the labor market effects of the Danish papir analyserer arbejdsmarkedseffekterne af den
wage compensation scheme for employees, which danske lønkompensationsordning for
offered wage subsidies to private firms that lønmodtagere, som tilbød løntilskud til private
furloughed workers instead of laying them off. virksomheder, der hjemsendte deres medarbejdere
Using payroll records at the monthly frequency for med løn i stedet for at afskedige dem. Ved brug af
Danish employer-employee matches, I find that the registerdata med månedlige lønoplysninger for
scheme prevented at least 10,400 exits from danske arbejdsgiver-arbejdstager matches finder
employment. However, it also compensated jeg, at ordningen forhindrede mindst 10.400
workers that firms were not planning to lay off. It afgange fra beskæftigelse. Ordningen
mainly prevented job losses for low-tenured kompenserede imidlertid også lønmodtagere, som
workers (≤ 1 year), while high-tenured workers (8+ virksomhederne ikke planlagde at afskedige. Den
years) would, to a lesser extent, have been laid off reddede primært jobs for nyansatte (≤ 1 år), mens
even in the absence of the scheme. Further, the lønmodtagere med mange års erfaring i
scheme led to a significant decline in furloughed virksomheden (8+ år) kun i mindre omfang ville
workers' labor income, which may reflect that firms være blevet afskediget, hvis ordningen ikke var
negotiated wage cuts with their employees before blevet indført. Ordningen førte endvidere til et fald i
applying for wage compensation. Finally, I find that de hjemsendte lønmodtageres arbejdsindkomst,
labor market mobility was basically unaffected by hvilket kan afspejle, at virksomhederne forhandlede
the scheme. lønnedgange med deres medarbejdere, før de
ansøgte om lønkompensation. Endelig finder jeg, at
mobiliteten på arbejdsmarkedet stort set var
upåvirket af ordningen.
Key words Acknowledgements
Economic activity and employment; public finances This paper is a revised version of my master's thesis
and fiscal policy. at the Department of Economics at the University of
Copenhagen. A special thanks to my supervisor
Claus Thustrup Kreiner and to colleagues at
JEL classification Danmarks Nationalbank, in particular Simon Juul
H12; J24; J38; J63 Hviid, Saman Darougheh, Henrik Yde Andersen,
Marcus Mølbak Ingholt, Martin Nygaard Jørgensen,
Peter Ejler Storgaard, and Thomas Harr, for
valuable comments and discussions. The author is
responsible for any remaining errors.
Job Retention during the Covid-19 Pandemic∗
Pernille Valentin Borgensgaard†
Abstract
Following the outbreak of the covid-19 pandemic, countries worldwide adopted
new or scaled up existing job retention schemes. This paper examines the labor
market effects of the Danish wage compensation scheme for employees, which of-
fered wage subsidies to private firms that furloughed workers instead of laying them
off. Using payroll records at the monthly frequency for Danish employer-employee
matches, I find that the scheme prevented at least 10,400 exits from employment.
However, it also compensated workers that firms were not planning to lay off. It
mainly prevented job losses for low-tenured workers (≤ 1 year), while high-tenured
workers (8+ years) would, to a lesser extent, have been laid off even in the absence
of the scheme. Further, the scheme led to a significant decline in furloughed workers’
labor income, which may reflect that firms negotiated wage cuts with their employees
before applying for wage compensation. Finally, I find that labor market mobility
was basically unaffected by the scheme.
Keywords: Economic activity and employment; public finances and fiscal policy.
JEL Classification: H12; J24; J38; J63.
∗
This paper is a revised version of my master’s thesis at the Department of Economics at the University
of Copenhagen. A special thanks to my supervisor Claus Thustrup Kreiner and to colleagues at Danmarks
Nationalbank, in particular Simon Juul Hviid, Saman Darougheh, Henrik Yde Andersen, Marcus Mølbak
Ingholt, Martin Nygaard Jørgensen, Peter Ejler Storgaard, and Thomas Harr, for valuable comments
and discussions. The viewpoints and conclusions stated are the responsibility of the author, and do not
necessarily reflect the views of Danmarks Nationalbank. The author is responsible for any remaining
errors.
†
Danmarks Nationalbank. E-mail: pvb@nationalbanken.dk.
1
1 Introduction
A significant amount of jobs were at risk when governments worldwide enforced lockdowns
of entire industries in an attempt to contain the covid-19 virus.1 To limit job losses,
many countries adopted new or scaled-up existing job retention schemes. The goal was to
preserve job matches by subsidizing wage costs for firms that furloughed workers rather
than laying them off. The concept of these policies is not new. Similar schemes prevented
job losses during the Great Recession.2 Yet, our knowledge about the role of job retention
schemes for labor markets remains limited.
This paper analyzes the labor market effects of a temporary wage compensation scheme
introduced in Denmark in March 2020. The scheme offered wage subsidies of up to 90
percent of salaries to firms that faced layoffs of at least 30 percent of their workforce, or
more than 50 employees, due to the pandemic but chose to furlough workers with pay
instead. For an evaluation of this scheme, there are at least four questions to be asked:
i) How many job losses did the scheme prevent? ii) Did it prevent loss of firm-specific
human capital? iii) How did it affect furloughed workers’ labor income? iv) How did
it affect labor market mobility? Using register data with monthly information on wage
compensation and labor market outcomes for Danish employer-employee matches and a
matched difference-in-differences estimator, the paper presents four key findings.
First, the preferred estimate indicates that the Danish wage compensation scheme
prevented at least 10,400 exits from employment from March to August 2020 (8,500 exits
at the lower bound of the 95% confidence interval). However, it also compensated workers
that firms were not planning to lay off. This is in line with results for the US Paycheck
Protection Program, which also intended to limit job losses during the pandemic but had
little effect on employment since aid-taking firms would not have laid off many workers
even in its absence (Chetty et al., 2020).
Second, this paper examines the role of job retention schemes for retaining firm-specific
1. The pandemic induced significant declines in hirings (Bess, Borgensgaard, and Iuel, 2020; Kahn,
Lange, and Wiczer, 2020; Hensvik, Le Barbanchon, and Rathelot, 2021) and large-scale job losses
(Coibion, Gorodnichenko, and Weber, 2020; Cajner et al., 2020; Bartik et al., 2020) that were partly
explained by health policies such as lockdowns (Gupta et al., 2020; Kong and Prinz, 2020; Juranek et al.,
2020).
2. Empirical studies find positive effects on employment of short-time work schemes during the Great
Recession (Felter, 2012; Hijzen and Martin, 2013), but they also targeted jobs that would have been
preserved anyway (Boeri and Bruecker, 2011).
2
human capital measured by workers’ ex ante firm tenure. I find that the Danish wage
compensation scheme mainly prevented job losses for low-tenured workers (≤ 1 year),
while furloughed workers with many years of experience in the firm (8+ years) would, to
a lesser extent, have been laid off even in the absence of the scheme. This may suggest
that firm-specific human capital would, to some extent, have been retained anyway. This
is arguably an interesting finding, as retaining firm-specific human capital is often part
of the motivation for adopting job retention schemes during recessions. It is believed to
shorten the recovery, as re-developing productive worker-firm matches is time-consuming
and costly (Oi, 1962; Barron, Bishop, and Dunkelberg, 1985; Blatter, Muehlemann, and
Schenker, 2012). More generally, the result implies that the benefits of compensating
jobs for high-tenured workers are limited. On the other hand, the scheme may have
supported future potential output growth by saving jobs for low-tenured workers. A recent
paper by Caggese, Cuñat, and Metzger (2019) shows that in the event of an economic
contraction like the covid-19 shock, financially constrained firms are induced to fire low-
tenured workers with high productivity growth prospects rather than less promising high-
tenured workers because of their lower firing costs.
Third, the results point to a fall in furloughed workers’ labor income in response to the
scheme. In the compensation period from March to June 2020, the scheme is estimated
to have reduced the average monthly labor income of furloughed workers by 2.1 percent,
corresponding to a total loss in earnings of almost DKK 2,400 (USD 370) per worker over
this period.3 This may suggest that firms and their employees use the option to negotiate
wage cuts before applying for wage compensation.
Fourth, and finally, this paper studies how job retention schemes affect labor market
mobility. Since furloughed workers only have limited incentives to apply for new jobs when
they are sent home with pay, the scheme may reduce the high degree of labor mobility
which is otherwise seen as one of the strengths of the Danish labor market model (Ander-
sen, Svarer, and Schröder, 2020). Labor mobility enables the allocation of workers from
less to more productive job matches. I find that furloughed workers’ probability of hav-
ing a job-to-job transition is basically unaffected by the wage compensation scheme. This
may reflect that job search costs were very high for both furloughed and non-compensated
workers during the lockdown of the Danish economy, as vacancies declined drastically.
3. While the average monthly labor income of furloughed workers is DKK 28,258 in this period, the
results indicate that it would have been approximately DKK 28,850 in the absence of the scheme.
3
A related study by Bennedsen et al. (2020) uses newly collected survey data on 10,642
firms to study the role of Danish relief packages for preventing job losses during the
pandemic. Comparing firms’ layoff decisions to their reported counterfactuals, they find
that the Danish relief packages, including cost aid, fiscal aid, and the wage compensation
scheme, saved 81,000 jobs. Their study is, however, not directly comparable to this paper,
as it estimates the job savings of several Danish relief packages and not only the wage
compensation scheme. It is also an earlier study, measuring the effect on firms’ intended
layoffs at the beginning of the pandemic rather than actual layoffs. It also employs a
different research design. When survey data is used, a concern is that firms misreport
counterfactual layoffs, leading to an upward bias in the estimated job savings. Aid-taking
firms that could otherwise afford to retain furloughed workers themselves have an incentive
to overstate counterfactual layoffs to be eligible for wage compensation, which they can
credibly do since the eligibility rule is unobserved.4
This paper estimates the effects of the wage compensation scheme using a matched
difference-in-differences estimator, comparing the growth rate of several labor market
outcomes from before to after the introduction of the scheme for two groups of workers:
those that were furloughed by wage compensated firms from March to June 2020 and
their non-compensated matches. The two groups are matched on the propensity score to
have similar pre-pandemic characteristics and a similar ex ante exposure to the covid-19
shock, exploiting the extraordinary degree of detail in the Danish registers. Exposure is
measured by workers’ ex ante industry and occupation, while the data does not allow for ex
ante firm-level measures. Firms operating within the same industries may not be equally
affected by the covid-19 shock, suggesting that the preferred estimate for the number
of averted exits from employment may be understated. Importantly, however, this issue
is not expected to be pronounced, as the results do not change if firms are additionally
matched on the change in sales of their workplaces from March 2019 to March 2020.
The empirical strategy provides a partial equilibrium analysis of the labor market
effects of the scheme, ignoring general equilibrium effects that may have influenced con-
sumer demand, and thereby employment at all firms. In the absence of the wage com-
pensation scheme, layoffs of furloughed workers would probably have led to a contraction
in consumer demand, an additional worsening of economic conditions, and another round
4. The eligibility rule is based on the number of layoffs a firm would perform in the absence of the
scheme.
4
of layoffs. This provides a further argument for interpreting the estimated number of
averted exits from employment as a lower bound for the true effect of the scheme.
The paper is organized as follows. Section 2 presents the wage compensation scheme
and a timeline of Danish covid-19 policies. Section 3 describes the data, and section 4
explains the empirical strategy. Section 5 presents the results, and section 6 concludes.
2 The Danish Wage Compensation Scheme
The wage compensation scheme for employees was introduced in March 2020, following
the lockdown of the Danish economy. The scheme became effective on March 9, 2020, and
was originally set to be phased out on June 8, 2020, but was later extended to August 29,
2020. By this time, approved applications for wage compensation amounted to almost
DKK 12.2 billion, supporting more than 273,000 jobs, cf. table 5 in appendix A.
The scheme offered temporary wage subsidies to private firms facing layoffs of at least
30 percent of their workforce, or more than 50 employees, due to the covid-19 pandemic.
For white- and blue-collar employees the compensation amounted to 75 and 90 percent
of salaries, respectively, but with a monthly cap of DKK 30,000 per full-time employee.5
Firms covered the remaining part of wages but were allowed to negotiate salary reductions
with their employees before they applied for compensation. If a salary reduction was not
agreed upon, employees were entitled to full pay during the entire compensation period
but were required to take some holidays.6
During the support period, firms were not allowed to dismiss employees for reasons
related to the pandemic and had to send compensated employees home on furlough.
Furloughed employees, except students, were not allowed to work for the compensated
firm. However, they might take on paid work in other firms during the furlough period if
their contract did not preclude this.
Box 1 in appendix B provides an overview of the timing of Danish covid-19 policies
surrounding the adoption and expiration of the scheme.
5. For firms that furloughed up to 25 employees, compensation was, in principle, computed based on
the highest average labor income of furloughed workers in the previous three or 12 months, including
add-ons to salaries. However, the salary could deviate from this rule if firms documented that it did not
reflect the agreed salary. For firms that furloughed more than 25 employees, the agreed salary for all
furloughed workers should be stated in the application, including pension contributions, foreseen shift
differentials, etc.
6. From March 9 to July 9, 2020, compensated employees had to take 1.67 holidays per furloughed
month. From July 9 to August 29, 2020, they had to take up to 15 holidays unless none were accrued.
5
3 Data
The empirical analysis is based on Danish administrative register data from Statistics
Denmark. To identify wage compensated jobs and furlough periods, I exploit a new
register with information on all approved applications for wage compensation.7 Labor
market outcomes are obtained from the register of wage earner employment, which con-
tains monthly information on wages, hours of work, occupation, and industry for Danish
employer-employee matches. This information is reported by firms to the Danish Tax
Agency and tax evasion on wage income is very small (Kleven et al., 2011). The register
uniquely identifies workers and workplaces over time by personal registration numbers
and firm identifiers, such that worker flows in and out of every workplace may be tracked.
It further enables me to link the data to various background characteristics of workers
and firms from other administrative registers. For an overview of the covariate set, see
table 6 in appendix C.
The baseline sample covers the period January 2018 to August 2020. It is restricted
to workers employed in private sector firms since public sector firms are not eligible for
wage compensation. The focus is on the cohort of workers who are 22-65 years old, have
monthly labor incomes of DKK 10,000-70,000, and who are employed in February 2020.
For individuals with more than one job, I consider only their main occupation, defined as
the job with the highest number of reported working hours.
The empirical design of this paper relies on comparing workers who are furloughed by
wage compensated firms from March to June 2020 (the treatment group) to similar workers
employed by non-compensated firms in February 2020 (the control group). The two groups
are matched on the propensity score to have similar pre-pandemic characteristics and a
similar ex ante exposure to the covid-19 shock. Among the 21 possible compensation
periods, the focus is only on the largest in terms of furloughed workers (March to June
2020) to ensure that different timings of treatment do not bias the results.8 For the
remainder of the paper, workers in the treatment and control groups are simply referred
7. A limitation to the data used is that it only contains information on approved and not actual
compensation periods. This means that it is not observed in the data if a firm leaves the scheme before
the approved compensation period ends either to lay off workers due to financial distress or to call back
workers due to recovered demand.
8. Table 5 in appendix A reports the number of furloughed workers in each of the 21 compensation
periods.
6
to as furloughed and non-compensated workers, respectively.
The treatment group does not include non-furloughed workers who are employed by
wage compensated firms, even though their jobs are protected by the scheme as well,
as these workers cannot credibly be identified.9 The main issue in this context is that
firms’ layoff and furlough decisions may not be perfectly related. Firms may decide to
furlough highly valued workers that are temporarily unproductive due to the lockdown
rather than less valued workers that would otherwise have been laid off. This indicates
that the estimated job savings of the scheme may be a lower bound for the true effect. To
ensure that those covered by the scheme do not incorrectly end up in the control group, I
discard observations for non-furloughed workers who are employed by wage compensated
firms in February 2020.
This leaves an unmatched sample of 743,509 workers, of which 51,632 were furloughed
from March to June 2020. After matching, there are 90,231 workers in the sample. Among
these, 51,632 are furloughed (treated), and 38,599 are non-compensated (controls). Tables
7-9 in appendix D present descriptive statistics for the full sample and the subsamples of
furloughed and non-compensated workers both before and after matching.
Before matching, furloughed workers are on average four years younger than non-
compensated workers, have one year less tenure, and earned DKK 5,434 less labor income
each month of employment in 2019. They also appear to be less educated than non-
compensated workers. There is an overweight of women, foreign citizens, and part-time
employees among furloughed workers compared to non-compensated workers. Further,
there are substantial differences in the distributions of workers across industries and oc-
cupations for the two groups. For industries, one example is accommodation and food
service, which employs 17 percent of furloughed workers but less than one percent of non-
compensated workers. For occupations, it appears that while 15 percent of furloughed
employees work with personal services, including waiters and hairdressers, only 2 percent
of non-compensated employees do. Hence, before matching, the groups of furloughed and
non-compensated workers differ in terms of potentially important confounding factors.
After matching, the two groups have very similar covariate distributions. The average
9. The wage compensation register does not contain information on these workers, and the group of non-
furloughed workers that receive salaries from wage compensated firms prior to the scheme also includes
workers who were laid off just before the firm applied for compensation. Observing only separations from
employment which are affected by resignation periods and not the time of layoff, I cannot distinguish
between these two groups of workers.
7
worker is just above 39 years old, has almost 4 years of tenure, and had an average monthly
labor income of DKK 29,622 when employed in 2019. Moreover, 55 percent are full-time
employed, 48 percent are men, and 10 percent have foreign citizenship. Most workers
do not have children or only children older than 15 years, and the vast majority have a
vocational education. The two most represented industries are wholesale and retail trade,
where 36 percent of workers are employed, and accommodation and food service, where
18 percent of workers are employed. The two most represented occupations are personal
services and sales, with worker shares of 14 and 17 percent, respectively.
4 Empirical Strategy
This section presents the empirical framework used to estimate the labor market effects
of the Danish wage compensation scheme. Workers have been allocated to treatment
and control groups based on who gets wage compensation rather than who is eligible for
it, because the eligibility rule is unobserved. It is based on the number of layoffs that
a firm would perform in the absence of the scheme. This gives rise to concerns about
selection into the scheme. A key goal of this paper is to identify a suitable control group
for furloughed workers among workers who are employed by non-compensated firms. For
this purpose, a matched difference-in-differences approach is proposed.10
The idea is to compare the growth rate of labor market outcomes from before to
after the introduction of the scheme for furloughed and non-compensated workers that
are matched on the propensity score to have similar pre-pandemic characteristics and a
similar ex ante exposure to the covid-19 shock. The key identifying assumption is that
no unobserved heterogeneity across furloughed and non-compensated workers affects both
participation in the wage compensation scheme and changes in labor market outcomes.
Unobserved heterogeneity may induce firms that are most severely affected by the pan-
demic to select into the scheme. This suggests that the estimated exits from employment
averted by the scheme may be a lower bound for the true effect.
The matched difference-in-differences approach is implemented in a two-step proce-
dure. First, the propensity score is estimated using a logit model that predicts the proba-
10. Other studies using this approach include Girma and Görg (2007), Aerts and Schmidt (2008),
Böckerman and Ilmakunnas (2009), Leth-Petersen (2010), Mu and Van de Walle (2011), Bentivogli and
Mirenda (2017), Espinosa, Desrieux, and Ferracci (2018), Moore, Grosskurth, and Themann (2019), and
Ronchetti and Terriau (2019, 2021).
8
bility of being furloughed by a wage compensated firm conditional on the covariate set. It
is used to match each furloughed worker with the non-compensated worker that minimizes
the difference in propensity scores.11 Matching is done with replacement so that the same
non-compensated worker can act as a match for more than one furloughed worker. While
one-to-one matching with replacement minimizes the risk of bias, it comes at the cost
of inefficiency, as many observations for the group of non-compensated workers are dis-
carded. With population-wide data on Danish employer-employee matches, this approach
does, however, yield reasonably precise estimates.
Second, using the matched sample of furloughed and non-compensated workers, I
estimate a difference-in-differences specification,
Aug-20
X Aug-20
X
yit = α + βDiT REAT + δt DtM ON T H + γt DiT REAT × DtM ON T H + κjm + it , (1)
t=Jan-18 t=Jan-18
where yit is the labor market outcome of interest: i) a dummy for having an exit from
employment, ii) log-labor income, or iii) a dummy for having a job-to-job transition
for worker i in month t ∈ {Jan-18,..., Aug-20}, where Feb-20 is omitted. DiT REAT is
an indicator variable equal to one if worker i is in the treatment group of furloughed
workers, and zero otherwise. DtM ON T H are month dummies. κjm is industry-calendar-
month fixed effects for the j = 18 industries and m = 12 calendar months, introduced
to absorb differences in seasonal patterns across industries. Non-compensated workers in
the control group are weighted according to the number of times they are matched.
The parameters of interest are γt . They measure the average change in the outcome
for furloughed workers relative to matched non-compensated workers in each month t.
Under the null hypothesis of parallel trends, γt equals zero in every pre-scheme month
t ∈ {Jan-18,..., Feb-20}, and identifies the effects of the wage compensation scheme in the
post-scheme period t ∈ {Mar-20,..., Aug-20}.
The validity of the parallel trends assumption relies crucially on the choice of covariates
for the matching procedure. The covariate set must include all variables that affect both
participation in the wage compensation scheme and changes in labor market outcomes.
Two sources of selection are particularly important to handle. First, firms select into the
scheme based on how severely they are affected by lockdown policies. This is accounted
for by including ex ante measures of industry and occupation in the covariate set. While
11. Rosenbaum and Rubin (1983) show that if potential outcomes are independent of treatment condi-
tional on covariates, then they are also independent of treatment conditional on the propensity score.
9
industry is an important predictor of taking up Danish relief packages (Bennedsen et
al., 2020), occupation captures the targeting of the covid-19 shock towards jobs that
require face-to-face interactions (Adams-Prassl et al., 2020; Garrote Sanchez et al., 2020;
Montenovo et al., 2020). Second, since the management in firms selects which employees to
furlough and lay off based on certain characteristics (Bess and Darougheh, 2021; Mattana,
Smeets, and Warzynski, 2020), a series of relevant background variables are included in
the covariate set. All covariates are measured in the most recently observed month prior
to the introduction of the scheme.12
A limitation to the covariate set used is that it does not control for firms’ use of other
covid-19 relief packages, including compensation for fixed costs and various measures of
fiscal aid such as deferral of tax obligations. However, Julin, Kuchler, and Otte (2021) find
no differences in debt developments between firms receiving and those not receiving wage
compensation. Moreover, the covariate set does not include any firm-level measures of
exposure to the covid-19 shock, as no ex ante measure exists. This further indicates that
the estimated exits from employment averted by the scheme may be a lower bound for
the true effect. In particular, for firms operating within the same industry, the likelihood
of receiving wage compensation is either unrelated to or increasing with the revenue drop
experienced in response to the pandemic, depending on the combination of relief packages
used (Bennedsen et al., 2020). A series of robustness checks will be performed in section
5.5 to examine the sensitivity of the results to the choice of covariates.
A further requirement for achieving identification is that the covariate distributions for
furloughed and non-compensated workers overlap (Heckman, Ichimura, and Todd, 1997).
In the next section, it is assessed that the common support condition is satisfied.
The empirical strategy provides a partial equilibrium analysis of the labor market
effects of the Danish wage compensation scheme, ignoring general equilibrium effects
that may have influenced consumer demand, and thereby employment at all firms. In
the absence of the scheme, layoffs of furloughed workers would probably have led to a
contraction in consumer demand, an additional worsening of economic conditions, and
another round of layoffs. This provides a further argument for interpreting the estimated
12. Table 6 in appendix C presents the covariate set, which includes measures of demographics (age,
gender, and children by age intervals in Jan-20, and citizenship in Jan-19), human capital (educational
attainment in Sep-19 and firm tenure in Feb-20), and labor market attachment (full-time, inflow to
employment, exit from employment, job-to-job transition, and several jobs in Feb-20, and average monthly
labor income in employed months of 2019).
10
number of exits from employment averted by the scheme as a lower bound for the true
effect.
4.1 Assessing the Quality of Matches
The propensity score is estimated using a logit model that predicts the probability of
being furloughed by a wage compensated firm conditional on the covariate set. Tables
10-13 in appendix E present the estimation results transformed to average marginal ef-
fects. It appears that both industry and occupation are important determinants of being
furloughed. Among industries, the furlough probability is highest for workers employed in
accommodation and food services. Among occupations, it is highest for personal service
workers. Employees with several jobs are more likely to be furloughed, while part-time
and full-time employees are equally likely to be furloughed. Moreover, the probability of
being furloughed is decreasing with average monthly labor income in employed months of
2019. Likewise, it is decreasing with age. Generally, men are less likely to be furloughed,
while foreign citizens are more likely to be furloughed. Having young children does not
appear to be an important determinant of whether an employee is furloughed. Finally,
the furlough probability is predominantly decreasing with both tenure and educational
attainment higher than primary school.
Figure 1: Kernel Density Estimates of Propensity Scores
(a) Before Matching (b) After Matching
Kernel Density Kernel Density
24 6
Treatment Group Treatment Group
Control Group Control Group
20 5
16 4
12 3
8 2
4 1
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Propensity Score Propensity Score
Note: Panels (1a) and (1b) show kernel density estimates of propensity scores for furloughed work-
ers (solid) and non-compensated workers (dashed) in the baseline sample before and after matching,
respectively. The kernel used is an Epanechnikov, and the bandwidth is 0.01.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
The propensity score is used to match furloughed workers to similar non-compensated
11
workers. It is crucial for the validity of the matching estimator that for every furloughed
worker, there exists a non-compensated worker with a similar probability of treatment.
Figure 1 depicts kernel density estimates of the propensity scores and shows that there is
common support for the two groups. Further, the distributions of propensity scores for
furloughed and non-compensated workers are almost identical after matching.
The goal of the matching procedure is to balance covariates across the treatment and
control groups. Looking at standardized biases, as suggested by Rosenbaum and Rubin
(1985), the two groups appear to have very similar characteristics after matching, cf.
figure 2 and figures 6-7 in appendix F. The mean standardized bias is reduced from 15.5
percent before matching to 1.1 percent after matching.
Figure 2: Standardized Biases Across Industries Before and After Matching
Accommodation and food service
18
Wholesale and retail trade
17 Unmatched
Arts, entertainment and recreation
16 Matched
Other service activities
15
Human health and social work
14
Education
13
Public administration, defense 12
etc.
Travel agents, cleaning etc.
11
Transportation
10
Mining and quarrying
9
Real estate
8
Water supply, sewerage etc.
7
Electricity, gas, steam etc.
6
Information and communication
5
Knowledge-based services
4
Financial and insurance
3
Manufacturing
2
Construction
1
0
-70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70
Standardized bias (%)
Note: The figure shows standardized percentage biases across industries before and after match-
ing. Standardized biases measure, for each covariate, the difference between the means of the
treatment and control groups as a percentage of the square root of the average variance across the
two groups (Rosenbaum and Rubin, 1985). A positive bias means that the value of the covariate
is higher in the treatment group than in the control group (and vice versa). Standardized biases
across occupations and the remaining covariates are presented in figures 6-7 in appendix F.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
12
5 Results
This section presents the main results. First, it is examined to what extent the wage
compensation scheme prevented job losses and retained firm-specific human capital. Af-
terwards, it is examined whether the scheme affected furloughed workers’ labor income
and labor market mobility.
5.1 Prevention of Job Losses
The first step in the empirical analysis is to examine to what extent the wage compensation
scheme prevented job losses for furloughed workers. As a measure of job loss, I use an
indicator for having an exit from employment, where an exit is defined as receiving wage
income in the current but not in the subsequent month.13
Figure 3a shows the monthly share of workers that have an exit from employment
(the exit probabilities) for the treatment and control groups. The black and gray dashed
vertical lines mark the introduction of the scheme (March 2020) and the end of the
compensation period (June 2020), respectively. Following the lockdown in March 2020,
there is a sharp increase in the exit share for the control group to 3.2 percent, implying
that many non-compensated workers lost their jobs. In the subsequent months, job losses
are less pronounced, albeit the monthly exit shares are still above the level seen prior
to the pandemic.14 Among the treated workers, the exit share remains slightly above
zero in March 2020, reflecting that furloughed workers’ jobs are being preserved. For the
remainder of the compensation period (until June), the exit share rises slowly.15
Figure 3b shows estimates for the average effects of the wage compensation scheme on
furloughed workers’ probability of having an exit from employment, with corresponding
95% confidence bands.16 In the months preceding the introduction of the scheme, almost
all estimates are insignificant, supporting the parallel trends assumption. After the intro-
duction of the scheme in March 2020, the exit probability for furloughed workers declines
13. This is a correlated but imperfect measure of job loss. For instance, a worker who is laid off may find
a new job before wage payments cease from the previous firm. This further suggests that the estimate for
the number of exits from employment averted by the scheme may be a lower bound for the true effect.
14. This either reflects that non-compensated firms enforce all their layoffs on impact or that fewer
layoffs are needed, as the gradual reopening of the Danish economy improves economic conditions.
15. This reflects that firms leave the scheme before their approved compensation period ends to lay off
workers and/or that workers are resigning to an increasing degree.
16. These estimates are also presented in panel (a) in table 14 in appendix H.
13
markedly relative to that for similar non-compensated workers in the control group. This
suggests that the scheme successfully saved jobs, which is consistent with the findings of
Bennedsen et al. (2020).
Figure 3: Monthly Effects of the Scheme on the Exit Probability
(a) Monthly Employment-exit Shares (b) Estimated Treatment Effects
Percent Percentage Points
6.0 2.0
Control
Treatment
5.0 1.0
4.0 0.0
3.0 -1.0
2.0 -2.0
1.0 -3.0
0.0 -4.0
Mar-18 Sep-18 Mar-19 Sep-19 Mar-20 Sep-20 Mar-18 Sep-18 Mar-19 Sep-19 Mar-20 Sep-20
Note: Panel (a) shows the monthly shares of workers that have an exit from employment for
the treatment and control groups. Panel (b) shows estimates for the average effects of the wage
compensation scheme on furloughed workers’ employment-exit probability. The gray lines that encircle
the estimates are 95% confidence bands based on clustered standard errors at the firm level. The black
and gray dashed vertical lines mark the introduction of the scheme and the end of the compensation
period, respectively.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
Table 1 panel (1) reports the preferred difference-in-differences estimate of the change
in the exit share of the treatment group from the four months preceding the introduction
of the scheme to the four months following it relative to the control group. It suggests that
the scheme reduces the average exit probability of furloughed workers by 3.8 percentage
points in the compensation period from March to June 2020. While the share of furloughed
workers that have an exit from employment during this period is 1.7 percent, it would
have been approximately 5.5 percent in the absence of the scheme.
Comparing this estimate to the total number of furloughed workers, regardless of
compensation periods, I find that the Danish wage compensation scheme prevented ap-
proximately 10,400 exits from employment during the first wave of the pandemic (8,500
exits at the lower bound of the 95% confidence interval). However, it also compensated
workers that firms were not planning to lay off.
The employed research design implies that the preferred estimate for the number of
exits averted by the scheme may be a lower bound for the true effect.
14
Table 1: Estimated Effect of the Scheme on the Exit Probability
—– Main —– ———————– Robustness ———————–
(1) (2) (3) (4)
DPOST × DTREAT -0.038*** -0.038*** -0.039*** -0.045***
(0.004) (0.004) (0.004) (0.004)
DPOST 0.044*** 0.044*** 0.047*** 0.055***
(0.003) (0.002) (0.002) (0.007)
DTREAT -0.001 -0.001 -0.001 -0.001**
(0.000) (0.000) (0.000) (0.001)
Industry-time FE Yes Yes Yes Yes
Occupation-time FE No Yes Yes No
Firm Covariates No No Yes No
Observations 203,969 203,969 186,192 176,482
Note: The table shows estimates for the average effect of the wage compensation scheme on furloughed
workers’ exit probability in the compensation period, comparing changes in exit shares of the treatment
and control groups from DP OST = 0 (Nov-19 to Feb-20) to DP OST = 1 (Mar-20 to Jun-20). Panel
(1) presents the preferred estimate from a model with industry-time fixed effects. Panels (2)-(4) each
represent a robustness test. Panel (2) includes occupation-time fixed effects. Panel (3) re-matches workers
on a covariate set extended with additional firm characteristics. Panel (4) excludes workers with more
than one job in February 2020. Parentheses report standard errors clustered at the firm level. * p<0.05,
** p<0.01, *** p<0.001.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
5.2 Retention of Firm-specific Human Capital
Another important aspect to explore is whether the scheme mainly prevented job losses
for workers with little or a fair amount of firm-specific human capital, as this is essential
for understanding its implications for both current and future productivity. As a proxy for
firm-specific human capital, I use months of tenure in the firm in February 2020.17 Table
2 shows preferred estimates for the average impact of the wage compensation scheme on
furloughed workers’ exit probability in the compensation period from March to June 2020
for four subgroups of workers with a) ≤ 1 year, b) ]1; 4] years, c) ]4; 8] years, and d) 8+
years of firm tenure, respectively.
The estimated job savings of the scheme appear to be declining with firm tenure.
For furloughed workers with up to one year of tenure, the results point to a scheme-
induced fall in the exit probability of 5.1 percentage points in the compensation period.
For workers with ]1; 4] years of tenure, the estimated fall is 3.5 percentage points. For
workers with ]4; 8] years of tenure, it is 1.8 percentage points, and for workers with more
than eight years of tenure, the response is insignificant.18 For all four subgroups, the exit
shares of furloughed and non-compensated workers appear to follow parallel trends in the
17. Firm tenure is positively correlated with firm-specific human capital, but there are limitations to
using this proxy, as workers may, for example, acquire firm-specific skills at different paces.
18. A caveat is that high-tenured workers may have longer resignation periods than low-tenured workers.
That is, they may be more likely to have an unobserved layoff because wage payments continue after the
sample ends.
15
pre-scheme period, cf. table 15 in appendix H.
Table 2: Heterogenous Effects on the Exit Probability Across Firm Tenure
(1) (2) (3) (4)
≤ 1 year ]1; 4] years ]4; 8] years > 8 years
DPOST × DTREAT -0.051*** -0.035*** -0.018*** -0.008
(0.005) (0.004) (0.004) (0.006)
DPOST 0.060*** 0.035*** 0.030*** 0.010***
(0.006) (0.003) (0.002) (0.002)
DTREAT -0.001 -0.001 0.000 -0.000
(0.001) (0.000) (0.000) (0.000)
Industry-time FE Yes Yes Yes Yes
Occupation-time FE No No No No
Firm Covariates No No No No
Observations 85,163 76,826 33,017 8,963
Note: The table presents estimates for the average effects of the wage compensation scheme on fur-
loughed workers’ exit probability in the compensation period from March to June 2020 for four subgroups
of workers with different levels of ex ante firm tenure. Each estimate compares the change in the exit
share of the treatment group from the four months preceding the introduction of the scheme to the four
months following it to the change of the control group. Panels (1)-(4) refer to workers with (1) ≤ 1 year,
(2) ]1; 4] years, (3) ]4; 8] years, and (4) > 8 years of tenure in February 2020, respectively. Parentheses
report standard errors clustered at the firm level. * p<0.05, ** p<0.01, *** p<0.001.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
These results indicate that the wage compensation scheme mainly prevented job losses
for low-tenured workers (≤ 1 year), while furloughed workers with many years of firm
tenure (8+ years) would, to a lesser extent, have been laid off even in the absence of the
scheme. This may suggest that firm-specific human capital would, to some extent, have
been retained anyway. More generally, the results indicate that the benefits of compen-
sating jobs for high-tenured workers are limited. On the other hand, the scheme may
have supported future potential output growth by saving jobs for low-tenured workers. A
recent paper by Caggese, Cuñat, and Metzger (2019) shows that in the event of an eco-
nomic contraction like the covid-19 shock, financially constrained firms are induced to fire
low-tenured workers with high productivity growth prospects rather than less promising
high-tenured workers because of their lower firing costs.
5.3 Intensive Margin Responses
The next step in the empirical analysis is to examine how furloughed workers’ labor
income was affected by the wage compensation scheme. A relatively narrow measure of
labor income is used, excluding mandatory pension contributions and fringe benefits.
Figure 4a shows the average monthly log-labor incomes for the treatment and control
groups. The solid black line is the outcome for the control group shifted downwards from
16
its actual position (the gray line) such that the share in February 2020 equals the one
for the treatment group. This graph allows for an easier comparison of pre-trends and
scheme effects. In the months preceding the introduction of the scheme, log-labor income
develops almost identically for the two groups, despite non-compensated workers earning
slightly more. Following the introduction of the scheme in March 2020, the average log-
labor income drops for the treatment group, while for the control group, it stays on the
pre-scheme path. Assuming that the two groups are equally affected by the pandemic,
this indicates that furloughed workers’ labor incomes fall in response to the scheme.19
Figure 4: Monthly Effects of the Scheme on Log-Labor Income
(a) Average Monthly Log-labor Income (b) Estimated Treatment Effects
Avg. log-labor income Percent
10.5 4.0
Control (shifted) Control (actual)
Treatment
10.4 2.0
10.3 0.0
10.2 -2.0
10.1 -4.0
10.0 -6.0
9.9 -8.0
Mar-18 Sep-18 Mar-19 Sep-19 Mar-20 Sep-20 Mar-18 Sep-18 Mar-19 Sep-19 Mar-20 Sep-20
Note: Panel (a) shows the average monthly log-labor income for the treatment and control groups.
The solid black line represents the control group shifted downwards from its actual position (the gray
line) such that the share in February 2020 equals the one for the treatment group. Panel (b) shows
estimates for the average effects of the wage compensation scheme on furloughed workers’ log-labor
income. The gray lines that encircle the estimates are 95% confidence bands based on clustered
standard errors at the firm level. The black and gray dashed vertical lines mark the introduction of
the scheme and the end of the compensation period, respectively.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
Figure 4b shows estimates for the average effects of the wage compensation scheme
on furloughed workers’ labor income, with corresponding 95% confidence bands.20 In the
pre-scheme period, all estimates but three are insignificant, supporting the parallel trends
assumption. From March 2020 and onward, the scheme is estimated to reduce furloughed
workers’ labor income. This may indicate that firms and their employees use the option
to negotiate wage cuts before applying for wage compensation.21
19. Furloughed workers’ labor income also appears to decline in most post-scheme months when com-
paring monthly labor income distributions for 2020 to those for the same months in the pre-scheme years
2018 and 2019, cf. figure 8 in appendix G.
20. These estimates are also presented in panel (a) in table 16 in appendix H.
21. The decline in furloughed workers’ labor income is not expected to be driven by a reduction in
17
Table 3 panel (1) reports the preferred difference-in-differences estimate of the change
in the average monthly log-labor income of the treatment group from the four months
preceding the introduction of the scheme to the four months following it relative to the
control group. It suggests that the scheme reduces the average monthly labor income
of furloughed workers by 2.1 percent in the compensation period from March to June
2020. While the average monthly labor income for furloughed workers is DKK 28,258 in
this period, it would have been approximately DKK 28,850 in the absence of the scheme.
This corresponds to a total loss in earnings of almost DKK 2,400 (USD 370) in the
compensation period.
Table 3: Estimated Effect of the Scheme on Log-Labor Income
—– Main —– —————————– Robustness —————————–
(1) (2) (3) (4) (5)
DPOST × DTREAT -0.021** -0.021*** -0.017** -0.018* -0.027***
(0.007) (0.006) (0.006) (0.007) (0.003)
DPOST -0.152*** -0.038 -0.100 -0.145*** 0.047*
(0.034) (0.054) (0.058) (0.041) (0.023)
DTREAT -0.015 -0.018 -0.015 -0.017 -0.021
(0.012) (0.010) (0.011) (0.013) (0.012)
Industry-time FE Yes Yes Yes Yes Yes
Occupation-time FE No Yes Yes No No
Firm Covariates No No Yes No No
Observations 203,969 203,969 186,192 176,482 183,512
Note: The table shows estimates for the average effect of the wage compensation scheme on furloughed
workers’ average monthly log-labor income in the compensation period, comparing changes in the average
monthly log-labor incomes of the treatment and control groups from DP OST = 0 (Nov-19 to Feb-20) to
DP OST = 1 (Mar-20 to Jun-20). Panel (1) presents the preferred estimate from a model with industry-time
fixed effects. Panels (2)-(5) each represent a robustness test. Panel (2) includes occupation-time fixed effects.
Panel (3) re-matches workers on a covariate set extended with additional firm characteristics. Panel (4) excludes
workers with more than one job in February 2020. Panel (5) re-matches workers on a restricted sample that
only includes workers who are employed during the compensation period. Parentheses report standard errors
clustered at the firm level. * p<0.05, ** p<0.01, *** p<0.001.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
5.4 Labor Market Mobility
The final step in the empirical analysis is to examine to what extent the wage compensa-
tion scheme affected labor market mobility. As a measure of labor market mobility, I use
an indicator for having a job-to-job transition, where a job-to-job transition is defined as
a change in firm identifiers of primary workplaces between the current and the previous
month.22
salary add-ons, as wage compensation accounted for all foreseen add-ons for planned work, such as
shift differentials. It even accounted for supplements for usual overtime work, unless the firm manually
corrected the salary in the application.
22. This is a highly correlated but imperfect measure of job-to-job transitions. For instance, if a work-
place gets a new CVR number (firm identifier) due to changes in ownership or company type, it will be
18
Figure 5a shows the monthly shares of workers that have a job-to-job transition for
the treatment and control groups. The solid black line is the outcome for the control
group shifted downwards from its actual position (the gray line) such that the share in
February 2020 equals the one for the treatment group. This illustration shows that before
the introduction of the scheme, the job-to-job transition shares develop fairly similarly for
the treatment and control groups. Following the lockdown of the economy in March 2020,
the job-to-job transition shares for both groups decline and stay below the pre-pandemic
level for the remainder of the observed period. This likely reflects that vacancies fell
markedly during this period (Bess, Borgensgaard, and Iuel, 2020). More interestingly,
from April 2020 and until the end of the compensation period, the job-to-job transition
share for the treatment group declines by relatively more than for the shifted control
group. This indicates that labor market mobility was slightly reduced by the scheme.
Figure 5: Monthly Effects of the Scheme on the Job-to-Job Transition Probability
(a) Monthly Job-to-Job Transition Shares (b) Estimated Treatment Effects
Percent Percentage Points
6.0 3.0
Control (shifted) Control (actual)
Treatment
5.0 2.0
4.0 1.0
3.0 0.0
2.0 -1.0
1.0 -2.0
0.0 -3.0
Mar-18 Sep-18 Mar-19 Sep-19 Mar-20 Sep-20 Mar-18 Sep-18 Mar-19 Sep-19 Mar-20 Sep-20
Note: Panel (a) shows the monthly shares of workers that have a job-to-job transition for the
treatment and control groups. The solid black line represents the control group shifted downwards
from its actual position (the gray line) such that the share in February 2020 equals the one for the
treatment group. Panel (b) shows estimates for the average effects of the wage compensation scheme
on furloughed workers’ job-to-job transition probability. The gray lines that encircle the estimates
are 95% confidence bands based on clustered standard errors at the firm level. The black and gray
dashed vertical lines mark the introduction of the scheme and the end of the compensation period,
respectively.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
Figure 5b presents the matched difference-in-differences estimates for the average ef-
fects of the wage compensation scheme on furloughed workers’ probability of having a
job-to-job transition, with corresponding 95% confidence bands.23 In the pre-scheme pe-
incorrectly counted as a job-to-job transition.
23. These estimates are also presented in panel (a) in table 17 in appendix H.
19
riod, almost all estimates are insignificant, supporting the parallel trends assumption.
In the compensation period, the scheme is estimated to have a small negative impact
on furloughed workers’ job-to-job transition probability, but only the estimate for April
is statistically significant. As a result, when estimating the aggregate response for the
compensation period, the estimate is very small and insignificant as well. In particular,
table 4 shows that the preferred estimate of the change in furloughed workers’ job-to-job
transition share from the four months preceding the introduction of the scheme to the
four months following it relative to the control group is −0.003. This suggests that labor
market mobility was basically unaffected by the wage compensation scheme.
This result likely reflects that job search costs were extremely high during the lock-
down of the Danish economy, as vacancies declined drastically. For instance, Bess, Bor-
gensgaard, and Iuel (2020) show that from late March to late April 2020, the number
of job postings in accommodation and food services was approximately 80 percent below
the historical average. As a result, neither furloughed nor non-compensated workers had
strong incentives to apply for new jobs.
Table 4: Estimated Effect of the Scheme on the Job-to-Job Transition Probability
—– Main —– ————————– Robustness ————————–
(1) (2) (3) (4) (5)
DPOST × DTREAT -0.003 -0.003 -0.010* -0.005 -0.007
(0.005) (0.005) (0.004) (0.004) (0.005)
DPOST 0.050 0.045 0.026 -0.059 0.054
(0.029) (0.033) (0.045) (0.044) (0.049)
DTREAT -0.013** -0.013** -0.010* -0.007 -0.008
(0.004) (0.004) (0.004) (0.004) (0.005)
Industry-time FE Yes Yes Yes Yes Yes
Occupation-time FE No Yes Yes No No
Firm Covariates No No Yes No No
Observations 203,969 203,969 186,192 176,482 183,512
Note: The table shows estimates for the average effect of the wage compensation scheme on furloughed
workers’ job-to-job probability in the compensation period, comparing changes in job-to-job transition shares
of the treatment and control groups from DP OST = 0 (Nov-19 to Feb-20) to DP OST = 1 (Mar-20 to Jun-20).
Panel (1) presents the preferred estimate from a model with industry-time fixed effects. Panels (2)-(5) each
represent a robustness test. Panel (2) includes occupation-time fixed effects. Panel (3) re-matches workers on
a covariate set extended with additional firm characteristics. Panel (4) excludes workers with more than one
job in February 2020. Panel (5) re-matches workers on a restricted sample that only includes workers who are
employed during the compensation period. Parentheses report standard errors clustered at the firm level. *
p<0.05, ** p<0.01, *** p<0.001.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
5.5 Robustness
To address possible caveats in the research design, panels (2)-(5) in tables 1 and 3-4 report
estimates associated with relevant robustness checks.
20
First, occupation-time fixed effects are included in the specification. These covariates
flexibly allow each of the 39 occupations to have its own time trend. Workers with
occupations that are often offered as seasonal positions, e.g., waiters, are more likely
to experience an increase in labor income during peak months than workers with other
occupations. They are also more likely to have an exit from employment or a job-to-
job transition when the temporary employment ends. Panel (2) shows that for all three
outcomes, the results are very similar to those of the preferred specification.
Second, firms may be more likely to apply for wage compensation if they experienced
negative growth before the pandemic (Nielsen and Hansen, 2020). To account for this,
I expand the covariate set with employment growth dummies, indicating whether the
workplace had negative, stable (up to 10 percent), positive (10-20 percent), or highly
positive (more than 20 percent) average employment growth in 2016-2019. For firms that
have not existed that long, there is a dummy indicating that it is a new firm. Another
concern is that firms operating within the same industries are not equally affected by
the pandemic. To account for this, the covariate set is extended with the most frequent
indicator of firm-level covid-19 exposure made possible by the Danish registers, which is
the change in sales from March 2019 to March 2020. Further, I include a measure of
average firm wages in February 2020, as firms with higher wages may be less likely to
select into the scheme due to the monthly wage compensation cap of DKK 30,000. Panel
(3) shows that re-matching workers on this extended covariate set does not change the
main results. For all three outcomes, the estimated effect of the scheme is close to that
of the baseline model.
Third, for workers with several jobs, exits from employment and job-to-job transitions
are potentially incorrectly measured because the sample is restricted to workers’ main
occupation. If a worker separates from one of several jobs, this will not be counted as
an exit from employment, but incorrectly as a job-to-job transition, if it is the main
job, or otherwise be unobserved, leading to bias in the estimated impact of the scheme.
Moreover, since workers are not matched based on industry and occupation of secondary
jobs, these may not be equally affected by the pandemic across treatment and control
groups. As a result, the estimated decline in total labor income may simply be explained
by furloughed workers being laid off from their side jobs to a larger extent than non-
compensated workers. To address this concern, I exclude workers with several jobs in
21
February 2020 and then re-match them. Panel (4) shows that this does not change the
estimated effect of the scheme significantly for any outcome.
Fourth, non-compensated workers in the control group who are laid off in the post-
scheme period could be different from those that are not. If those that are laid off
generally earn less, then the estimated effects on labor income would be negative even in
the absence of the scheme. Likewise, the estimated decline in the job-to-job transition
probability may simply be explained by fewer transitions of non-compensated workers
who are laid off compared to those that are not. To account for this, I re-match workers
on a restricted sample that only includes workers who are employed during the entire
compensation period from March to June 2020. Panel (5) in tables 3 and 4 shows that
for both labor income and the job-to-job transition probability, the estimated effect of the
scheme is numerically larger. However, neither estimate differs significantly from those of
the baseline model.
Fifth and finally, a concern is that the results do not carry over to furloughed workers
with other compensation periods than March to June 2020. However, it does not change
the main findings to define treatment as being furloughed in at least one post-scheme
month from March to August 2020 instead, cf. figure 9 in appendix I. I still find that the
scheme averted exits from employment, reduced labor income of furloughed workers, and
did not affect labor market mobility noticeably.
6 Conclusion
This paper has examined the labor market effects of a temporary wage compensation
scheme introduced in Denmark in March 2020 to limit job losses following the covid-19
pandemic. The scheme provided wage subsidies to private firms that faced extensive
layoffs due to the pandemic but chose to furlough workers with pay instead.
The effects are estimated using detailed register data and a matched difference-in-
differences estimator, comparing the growth rate of several labor market outcomes from
before to after the introduction of the scheme for furloughed workers and their non-
compensated matches. The two groups are matched on the propensity score to have
similar pre-pandemic characteristics and similar ex ante exposures to the covid-19 shock.
The findings suggest that during economic contractions, job retention schemes work as
22
intended by mitigating job losses. The Danish wage compensation scheme did, however,
also compensate workers that firms were not planning to lay off. It mainly prevented
job losses for low-tenured workers (≤ 1 year), while high-tenured workers (8+ years)
would, to a lesser extent, have been laid off even in the absence of the scheme. This
may suggest that firm-specific human capital would, to some extent, have been retained
anyway. Further, the results indicate that firms and their employees might have used
the option to negotiate wage cuts before applying for wage compensation and that labor
market mobility was basically unaffected by the scheme. The latter suggests that job
retention schemes do not prevent reallocation of workers from less to more productive job
matches in periods with few vacancies.
23
References
Adams-Prassl, Abi, Teodora Boneva, Marta Golin, and Christopher Rauh. 2020. “Inequality in
the impact of the coronavirus shock: Evidence from real time surveys.” Journal of Public
Economics, 189:104245.
Aerts, Kris, and Tobias Schmidt. 2008. “Two for the price of one?: Additionality effects of R&D
subsidies: A comparison between Flanders and Germany.” Research Policy, 37(5):806–822.
Andersen, Torben M., Michael Svarer, and Philipp Schröder. 2020. “Rapport fra den økonomiske
ekspertgruppe vedrørende udfasning af hjælpepakker.” Available at: https://fm.dk/media/
18033/rapport-fra-den-oekonomiske-ekspertgruppe-vedroerende-udfasning-af-hjaelpepakk
er.pdf (Accessed: May 3, 2021).
Barron, John M., John Bishop, and William C. Dunkelberg. 1985. “Employer search: The inter-
viewing and hiring of new employees.” The Review of Economics and Statistics, 67(1):43–
52.
Bartik, Alexander W., Marianne Bertrand, Zoë B. Cullen, Edward L. Glaeser, Michael Luca,
and Christopher T. Stanton. 2020. “How are small businesses adjusting to COVID-19?
Early evidence from a survey.” NBER Working Paper No. 26989.
Bennedsen, Morten, Birthe Larsen, Ian Schmutte, and Daniela Scur. 2020. “Preserving job
matches during the COVID-19 pandemic: firm-level evidence on the role of government
aid.” Covid Economics, 27.
Bentivogli, Chiara, and Litterio Mirenda. 2017. “Foreign ownership and performance: evidence
from Italian firms.” International Journal of the Economics of Business, 24(3):251–273.
Berlingske. 2020. “Detailhandel åbner – og så følger restauranter og skolernes ældste klasser trop.”
Berlingske. May 7, 2020. Available at: https://www.berlingske.dk/politik/detailhandel-
aabner-og-saa-foelger-restauranter-og-skolernes-aeldste (Accessed: May 24, 2021).
Bess, Mikkel, Pernille V. Borgensgaard, and Thea T. Iuel. 2020. “Arbejdsmarkedet er i bedring
efter historisk hårdt tilbageslag.” Danmarks Nationalbank, Economic Memo No. 13.
Bess, Mikkel, and Saman Darougheh. 2021. “Three lessons from the Danish wage compensation
scheme.” Danmarks Nationalbank, Economic Memo No. 5.
Bitsch, Alexander, and Daniel Skinbjerg. 2020. “Regeringen åbner grænsen til Tyskland, Island
og Norge.” FINANS. May 29, 2020. Available at: https://finans.dk/erhverv/ECE12177922/
regeringen-aabner-graensen-til-tyskland-island-og-norge/?ctxref =ext (Accessed: May 24,
2021).
Blatter, Marc, Samuel Muehlemann, and Samuel Schenker. 2012. “The costs of hiring skilled
workers.” European Economic Review, 56(1):20–35.
Bloch, Cathrine, Thue A. Holm, and Thomas S. Rohde. 2020. “Regeringen lukker nu restau-
ranter og butikker på stribe – og forbyder forsamlinger over ti personer: »Kampen mod
corona spidser til, og situationen bliver mere og mere alvorlig«.” Berlingske. March 17,
2020. Available at: https://www.berlingske.dk/politik/regeringen-lukker-nu-restauranter-
og-butikker-paa-stribe-og-forbyder-0 (Accessed: March 17, 2021).
24
Bloch, Cathrine, Pia G. Munksgaard, and Kasper Kildegaard. 2020. “Aftale på plads: Nye dele
af Danmark kan åbne.” Berlingske. May 20, 2020. Available at: https://www.berlingske.
dk/politik/aftale-paa-plads-nye-dele-af-danmark-kan-aabne (Accessed: May 24, 2021).
Böckerman, Petri, and Pekka Ilmakunnas. 2009. “Unemployment and self-assessed health: evi-
dence from panel data.” Health economics, 18(2):161–179.
Boeri, Tito, and Herbert Bruecker. 2011. “Short-time work benefits revisited: some lessons from
the Great Recession.” Economic Policy, 26(68):697–765.
Caggese, Andrea, Vicente Cuñat, and Daniel Metzger. 2019. “Firing the wrong workers: Financing
constraints and labor misallocation.” Journal of Financial Economics, 133(3):589–607.
Cajner, Tomaz, Leland Dod Crane, Ryan Decker, Adrian Hamins-Puertolas, and Christopher
Johann Kurz. 2020. “Tracking labor market developments during the COVID-19 pandemic:
A preliminary assessment.” FEDS Working Paper No. 2020-030.
Chetty, Raj, John N. Friedman, Nathaniel Hendren, Michael Stepner, and The Opportunity
Insights Team. 2020. “The economic impacts of COVID-19: Evidence from a new public
database built using private sector data.” NBER Working Paper No. 27431.
Coibion, Olivier, Yuriy Gorodnichenko, and Michael Weber. 2020. “Labor markets during the
COVID-19 crisis: A preliminary view.” NBER Working Paper No. 27017.
Dansk Erhverv (The Danish Chamber of Commerce). 2020. “Ny kompensationsordning for dækn-
ing af faste omkostninger.” Dansk Erhverv. June 17, 2020 Available at: https : / / www .
danskerhverv.dk/presse- og- nyheder/nyheder/2020/juni/ny- kompensationsordning- for-
dakning-af-faste-omkostninger/ (Accessed: May 24, 2021).
Erhvervsministeriet (Ministry of Industry, Business and Financial Affairs). 2020a. “COVID-19:
Nye retningslinjer skal hjælpe frisører og andre erhverv i gang igen.” Erhvervsministeriet.
April 19, 2020. Available at: https : / / em . dk / nyhedsarkiv / 2020 / april / covid - 19 - nye -
retningslinjer- skal- hjaelpe- frisoerer- og - andre- erhverv - i- gang - igen/ (Accessed: May 24,
2021).
Erhvervsministeriet (Ministry of Industry, Business and Financial Affairs). 2020b. “COVID-19:
Regeringen og alle Folketingets partier er enige om at justere og udvide hjælpepakker til
dansk økonomi.” Erhvervsministeriet. April 18, 2020. Available at: https://em.dk/nyheds
arkiv/2020/april/covid-19-regeringen-og-alle-folketingets-partier-er-enige-om-at-justere-
og-udvide-hjaelpepakker-til-dansk-oekonomi/ (Accessed: May 28, 2021).
Erhvervsministeriet (Ministry of Industry, Business and Financial Affairs). 2020c. “COVID-19:
Regeringen og alle Folketingets partier er enige om omfattende hjælpepakke til dansk
økonomi.” Erhvervsministeriet. March 19, 2020. Available at: https://em.dk/nyhedsarkiv/
2020/marts/regeringen-og-alle-folketingets-partier-er-enige-om-omfattende-hjaelpepakke-
til-dansk-oekonomi/ (Accessed: May 24, 2021).
Erhvervsministeriet (Ministry of Industry, Business and Financial Affairs). 2020d. “COVID-19:
Trepartsaftale skal hjælpe lønmodtagere.” Erhvervsministeriet. March 15, 2020. Available
at: https://em.dk/nyhedsarkiv/2020/marts/covid-19-trepartsaftale-skal-hjaelpe-loenmodt
agere/ (Accessed: May 24, 2021).
25
Espinosa, Romain, Claudine Desrieux, and Marc Ferracci. 2018. “Labor market and access to
justice.” International Review of Law and Economics, 54:1–16.
Felter, Megan. 2012. “Short-Time Compensation: Is Germany’s Success with Kurzarbeit an An-
swer to US Unemployment.” Boston College International and Comparative Law Review
35(2):481–509.
Garrote Sanchez, Daniel, Nicolas Gomez Parra, Caglar Ozden, and Bob Rijkers. 2020. “Which
Jobs Are Most Vulnerable to COVID-19? What an Analysis of the European Union Reveals.”
World Bank Research and Policy Briefs No. 148384.
Girma, Sourafel, and Holger Görg. 2007. “Evaluating the foreign ownership wage premium using a
difference-in-differences matching approach.” Journal of International Economics, 72(1):97–
112.
Gupta, Sumedha, Laura Montenovo, Thuy Dieu Nguyen, Felipe Lozano-Rojas, Ian M. Schmutte,
Kosali Ilayperuma Simon, Bruce A. Weinberg, and Coady Wing. 2020. “Effects of social
distancing policy on labor market outcomes.” NBER Working Paper No. 27280.
Hare, Henrik P. 2020. “Svømmehaller og fitnesscentre kan åbne fra på mandag.” FINANS. June
6, 2020. Available at: https://finans.dk/politik/ECE12196220/svoemmehaller-og-fitnessce
ntre-kan-aabne-fra-paa-mandag/?ctxref=ext (Accessed: May 24, 2021).
Heckman, James J., Hidehiko Ichimura, and Petra E. Todd. 1997. “Matching as an econometric
evaluation estimator: Evidence from evaluating a job training programme.” The review of
economic studies, 64(4):605–654.
Hensvik, Lena, Thomas Le Barbanchon, and Roland Rathelot. 2021. “Job search during the
COVID-19 crisis.” Journal of Public Economics, 194(C):104349.
Hijzen, Alexander, and Sebastien Martin. 2013. “The role of short-time work schemes during the
global financial crisis and early recovery: a cross-country analysis.” IZA Journal of Labor
Policy, 2(1):1–31.
Honoré, David R. 2020. “Danmark åbner grænserne: Fra 27. juni kan du rejse udenlands.”
Berlingske. June 18, 2020. Available at: https://www.berlingske.dk/samfund/danmark-
aabner-graenserne-fra-27-juni-kan-du-rejse-udenlands (Accessed: May 24, 2021).
IMF. 2020. Policy responses to COVID-19 - Policy Tracker. Available at: https://www.imf.org/
en/Topics/imf-and-covid19/Policy-Responses-to-COVID-19#D (Accessed: December 18,
2021).
Julin, Ida R., Andreas Kuchler, and Alexander M. Otte. 2021. “Firm financing and public support
measures during the pandemic.” Danmarks Nationalbank, Working Paper No. 184.
Juranek, Steffen, Jörg Paetzold, Hannes Winner, and Floris Zoutman. 2020. “Labor market effects
of COVID-19 in Sweden and its neighbors: Evidence from novel administrative data.” NHH
Dept. of Business and Management Science Discussion Paper No. 2020/8.
Kahn, Lisa B., Fabian Lange, and David G. Wiczer. 2020. “Labor Demand in the time of COVID-
19: Evidence from vacancy postings and UI claims.” NBER Working Paper No. 27061.
26
Kleven, Henrik Jacobsen, Martin B. Knudsen, Claus Thustrup Kreiner, Søren Pedersen, and
Emmanuel Saez. 2011. “Unwilling or unable to cheat? Evidence from a tax audit experiment
in Denmark.” Econometrica, 79(3):651–692.
Kong, Edward, and Daniel Prinz. 2020. “The impact of shutdown policies on unemployment
during a pandemic.” Covid Economics, 17:24–72.
Leth-Petersen, Søren. 2010. “Intertemporal consumption and credit constraints: Does total expen-
diture respond to an exogenous shock to credit?” American Economic Review, 100(3):1080–
1103.
Mattana, Elena, Valerie Smeets, and Frederic Warzynski. 2020. “Changing skill structure and
COVID-19.” Covid Economics, 45:1–30.
Montenovo, Laura, Xuan Jiang, Felipe Lozano Rojas, Ian M. Schmutte, Kosali I. Simon, Bruce
A. Weinberg, and Coady Wing. 2020. “Determinants of disparities in COVID-19 job losses.”
NBER Working Paper No. 27132.
Moore, Nils aus dem, Philipp Grosskurth, and Michael Themann. 2019. “Multinational corpo-
rations and the EU Emissions Trading System: The specter of asset erosion and creeping
deindustrialization.” Journal of Environmental Economics and Management, 94(C):1–26.
Mu, Ren, and Dominique Van de Walle. 2011. “Rural roads and local market development in
Vietnam.” The Journal of Development Studies, 47(5):709–734.
Nielsen, Peter Bøegh, and Kalle Emil Holst Hansen. 2020. “COVID-19: Hvilke virksomheder
har benyttet Lønkompensationsordningen? ” Available at: https://www.dst.dk/Site/Dst/
Udgivelser/nyt/GetAnalyse.aspx?cid=40811 (Accessed: April 22, 2021).
Oi, Walter Y. 1962. “Labor as a quasi-fixed factor.” Journal of political economy, 70(6):538–555.
Philipsen, Andrea. 2020. “Regeringen lukker de danske grænser.” Børsen. March 13, 2020. Avail-
able at: https : / / borsen . dk / nyheder / generelt / regeringen - lukker - de - danske - graenser
(Accessed: March 17, 2021).
Randeris, Mikkel. 2020. “Se kortet: Sådan åbner skoler og dagtilbud i din kommune.” Berlingske.
April 12, 2020. Available at: https : / / www . berlingske . dk / samfund / se - kortet - saadan -
aabner-skoler-og-dagtilbud-i-din-kommune (Accessed: May 24, 2021).
Retsinformation. 2020a. Bekendtgørelse om en midlertidig lønkompensationsordning til virksomheder
i økonomisk krise som følge af COVID-19. BEK No. 267 of March 25, 2020. Available at:
https://www.retsinformation.dk/eli/lta/2020/267 (Accessed: October 18, 2020).
Retsinformation. 2020b. Bekendtgørelse om en midlertidig lønkompensationsordning til virk-
somheder i økonomisk krise som følge af COVID-19. BEK No. 1116 of July 2, 2020. Available
at: https://www.retsinformation.dk/eli/lta/2020/1116 (Accessed: October 20, 2020).
Ronchetti, Jérôme, and Anthony Terriau. 2019. “Impact of unemployment on self-perceived
health.” The European Journal of Health Economics, 20(6):879–889.
Ronchetti, Jérôme, and Anthony Terriau. 2021. “Help me quit smoking but don’t make me
sick! The controversial effects of electronic cigarettes on tobacco smokers.” Social Science &
Medicine, 274:113770.
27
Rosenbaum, Paul R., and Donald B. Rubin. 1983. “The central role of the propensity score in
observational studies for causal effects.” Biometrika, 70(1):41–55.
Rosenbaum, Paul R., and Donald B. Rubin. 1985. “The bias due to incomplete matching.”
Biometrics, 41(1):103–116.
SKAT (The Danish Customs and Tax Administration). 2021. “eIndkomst: Indberetningsvejled-
ning. 8.1. Løntimer.” Available at: https://skat.dk/SKAT.aspx?oid=2045904 (Accessed:
May 20, 2021).
Stephensen, Emma K., and Thomas S. Hansen. 2020. “Danmark lukker ned: Her er regeringens
nye tiltag.” TV2. March 11, 2020. Available at: https://nyheder.tv2.dk/samfund/2020-03-
11-danmark-lukker-ned-her-er-regeringens-nye-tiltag (Accessed: March 17, 2021).
28
Appendices
Appendix A – Different Compensation Periods
Table 5: Furloughed Workers by Compensation Periods
—— Compensation period —— — Approved applications —
Start month End month Furloughed workers
Mar-20 Mar-20 2,667
Mar-20 Apr-20 12,434
Mar-20 May-20 10,745
Mar-20 Jun-20 133,481
Mar-20 Jul-20 28,026
Mar-20 Aug-20 17,270
Apr-20 Apr-20 3,841
Apr-20 May-20 4,694
Apr-20 Jun-20 27,516
Apr-20 Jul-20 10,030
Apr-20 Aug-20 4,971
May-20 May-20 1,075
May-20 Jun-20 5,368
May-20 Jul-20 3,907
May-20 Aug-20 2,045
Jun-20 Jun-20 1,003
Jun-20 Jul-20 2,288
Jun-20 Aug-20 1,110
Jul-20 Jul-20 75
Jul-20 Aug-20 713
Aug-20 Aug-20 12
Total 273,271
Note: The table shows the number of workers who were furloughed in each
of the 21 possible compensation periods. The bold line marks the largest group
of furloughed workers who were sent home from March 2020 to June 2020. It
is based on the raw wage compensation data from the register LONKOMP and
not the baseline sample used in the empirical analysis. However, observations
with missing worker identifiers are discarded.
Source: Own calculations based on Danish administrative register data from
Statistics Denmark.
29
Appendix B – Timeline of Danish Covid-19 Policies
Box 1. Timeline of Danish Covid-19 Policies in 2020
Feb 27: First confirmed case of covid-19 in Denmark.
Mar 11: Press conference with Danish Prime Minister Mette Frederiksen. Closure of
public schools, daycare and cultural institutions. Ban on indoor gatherings of
more than 100 participants.
Mar 14: Closure of Danish borders.
Mar 15: Announcement of tripartite agreement on temporary wage compensation for
private-sector employees.
Mar 17: Closure of hairdressers, malls, bars, gyms, and other businesses with close cus-
tomer contact. Cafes and restaurants close for guests but are allowed to deliver
takeaway orders. Ban on large gatherings tightened, now prohibiting both in-
door and outdoor gatherings of more than 10 people.
Mar 19: Announcement of agreement on temporary compensation for firms’ fixed costs.
Extension of entitlement to unemployment insurance and sickness benefits.
Mar 24: Danish Parliament passes the law on temporary wage compensation for employ-
ees in private firms.
Apr 15: Reopening of daycare institutions and schools from 0th to 5th grade.
Apr 18: Agreement on refund of paid VAT and payroll taxes as an interest-free loan.
Apr 20: Reopening of liberal professions, including hairdressers and masseurs.
May 11: Reopening of the retail sector, including shopping malls.
May 18: Reopening of cafes, restaurants, and schools for 6th to 10th grade.
May 27: Reopening of several cultural institutions, including theaters and cinemas.
Jun 08: Reopening of gyms and swimming pools.
Jun 15: Reopening of borders for tourists from Germany, Norway, and Iceland.
Jun 27: Reopening of borders for European countries that meet infection criteria.
Jul 08: Termination of the fixed costs compensation scheme for all firms but those that
are still subject to covid-19 restrictions.
Aug 29: Termination of the wage compensation scheme for employees.
Aug 31: Termination of the fixed costs compensation scheme for remaining firms.
Source: IMF (2020), Stephensen and Hansen (2020), Philipsen (2020), Bloch, Holm, and Rohde (2020), Retsinfor-
mation (2020a, 2020b), Bitsch and Skinbjerg (2020), Honoré (2020), Hare (2020), Bloch, Munksgaard, and Kilde-
gaard (2020), Randeris (2020), Berlingske (2020), Dansk Erhverv (The Danish Chamber of Commerce) (2020), and
Erhvervsministeriet (Ministry of Industry, Business and Financial Affairs) (2020a, 2020b, 2020c, 2020d).
30
Appendix C – Description of the Covariate Set
Table 6: Description of the Covariate Set
Covariate Description
Age Age in years on January 1, 2020.
Man An indicator that equals one if a worker is male, and zero otherwise.
Foreign An indicator that equals one if a worker has non-Danish or unobserved citizenship
on January 1, 2019, and zero otherwise.
Children indicators Four indicators that equal one if a worker’s youngest child is a) up to 5 years old,
b) 6-10 years old, c) 11-15 years old, or if a worker d) has no children or only
children older than 15 years on 1 January, 2020, respectively, and equal to zero
otherwise.
Educational indicators Indicators that equal one if a worker’s highest educational attainment is a) primary
school, b) high school, c) vocational education (e.g. carpenter, mechanic, nursing
home assistant), d) short higher education (e.g. 2-year marketing economist), e)
medium higher education (e.g. trained childcare worker), f) bachelor’s degree,
g) long higher education (e.g. master’s degree), and h) research (e.g. PhD),
respectively, in September 2020, and equal to zero otherwise. They act as measures
of general human capital.
Tenure The number of months a worker has been employed in a given firm in February
2020, censored above a threshold of 12 years and 2 months, as the data is only
observed from 2008. It acts as a measure of firm-specific human capital.
Full-time An indicator that equals one if an employee worked at least 160 hours (full time) in
February 2020, and zero otherwise. This classification follows SKAT’s reporting
instructions for the Danish Income Tax Register with a minor correction. The
instruction states that full-time work must always be reported as 160.33 hours per
month (SKAT (The Danish Customs and Tax Administration), 2021). The cor-
rection is made because there are many observations, where the reported number
of hours is 160, which is interpreted as incorrect reporting rather than part-time
work.
Inflow to employment An indicator that equals one if a worker has an inflow to employment in February
2020, and zero otherwise. An inflow is defined as receiving wages in the current
but not in the previous month.
Exit from employment An indicator that equals one if a worker has an exit from employment in February
2020, and zero otherwise. An exit is defined as receiving wages in the current but
not in the subsequent month.
Several jobs An indicator that equals one if a worker has several jobs in February 2020, and zero
otherwise. Several jobs are defined as receiving wages from at least two workplaces
with different CVR numbers.
Job-to-job transition An indicator that equals one if a worker has a job-to-job transition in February
2020, and zero otherwise. A job-to-job transition is defined as a change in firm
identifiers of workplaces between the current and the previous month.
Avg. monthly labor Average monthly labor income in 2019 conditional on employment. A relatively
income in 2019 narrow measure of labor income is used, excluding mandatory pension contribu-
tions and fringe benefits.
Industry indicators Indicators describing the industry of the workplace that employs the worker in
February 2020. I distinguish between 20 industries that are defined according to
the NACE classification. For a description of each industry, see Statistics Denmark.
Occupation indicators Indicators describing a worker’s occupation in February 2020. I distinguish between
38 occupations that are defined according to the DISCO-08 classification, but where
one of them ("other occupations") includes all occupational codes that are shared
by less than 150 non-compensated workers in February 2020. For a description of
the DISCO-08 classification, see Statistics Denmark.
Note: This table describes the baseline covariate set used to estimate the propensity score for being furloughed by a
wage compensated firm in section 4.1.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
31
Appendix D – Descriptive Statistics
Table 7: Descriptive Statistics Across Selected Covariates
—————– Unmatched —————– —————— Matched ——————
Full sample Treatment Control Full sample Treatment Control
Age (Years) 42.867 39.099 43.149 39.102 39.099 39.105
(11.839) (12.229) (11.761) (12.236) (12.229) (12.244)
Man 0.656 0.481 0.669 0.482 0.481 0.483
(0.475) (0.500) (0.471) (0.500) (0.500) (0.500)
Foreign 0.066 0.100 0.064 0.101 0.100 0.102
(0.249) (0.301) (0.244) (0.302) (0.301) (0.303)
Youngest child is ≤ 5 years 0.127 0.124 0.127 0.125 0.124 0.125
(0.332) (0.330) (0.333) (0.330) (0.330) (0.330)
Youngest child is 6-10 years 0.102 0.091 0.103 0.093 0.091 0.094
(0.303) (0.288) (0.304) (0.290) (0.288) (0.292)
Youngest child is 11-15 years 0.103 0.083 0.104 0.084 0.083 0.084
(0.303) (0.276) (0.305) (0.277) (0.276) (0.278)
High school 0.089 0.170 0.083 0.174 0.170 0.178
(0.284) (0.376) (0.275) (0.379) (0.376) (0.383)
Vocational training 0.434 0.435 0.434 0.430 0.435 0.426
(0.496) (0.496) (0.496) (0.495) (0.496) (0.495)
Short higher education 0.079 0.057 0.081 0.058 0.057 0.059
(0.270) (0.233) (0.273) (0.234) (0.233) (0.236)
Medium higher education 0.095 0.077 0.096 0.077 0.077 0.078
(0.293) (0.266) (0.295) (0.267) (0.266) (0.268)
Bachelor’s degree 0.025 0.022 0.025 0.023 0.022 0.023
(0.156) (0.148) (0.157) (0.149) (0.148) (0.151)
Long higher education 0.109 0.058 0.113 0.057 0.058 0.056
(0.312) (0.233) (0.317) (0.232) (0.233) (0.230)
Research education 0.008 0.002 0.008 0.002 0.002 0.002
(0.088) (0.048) (0.091) (0.049) (0.048) (0.049)
Full-time 0.611 0.551 0.616 0.553 0.551 0.554
(0.487) (0.497) (0.486) (0.497) (0.497) (0.497)
Inflow to employment 0.009 0.009 0.008 0.009 0.009 0.009
(0.092) (0.094) (0.092) (0.095) (0.094) (0.095)
Exit from employment 0.006 0.000 0.006 0.000 0.000 0.001
(0.075) (0.020) (0.078) (0.021) (0.020) (0.022)
Several jobs 0.066 0.138 0.061 0.147 0.138 0.155
(0.248) (0.345) (0.239) (0.354) (0.345) (0.362)
Job-to-job transition 0.019 0.029 0.018 0.031 0.029 0.032
(0.136) (0.168) (0.133) (0.172) (0.168) (0.177)
Tenure (Years) 4.850 3.886 4.922 3.813 3.886 3.741
(4.240) (3.859) (4.258) (3.818) (3.859) (3.776)
Avg. monthly labor income 34.701 29.644 35.078 29.622 29.644 29.600
in 2019 (DKK 1,000) (12.042) (10.822) (12.043) (10.877) (10.822) (10.932)
Observations 743,509 51,632 691,877 90,231 51,632 38,599
Note: The table shows sample means of selected covariates for the full sample and the subsamples of furloughed
(treated) and non-compensated (control) workers both before and after matching. The reference category for the three
children indicators is having no children or only children older than 15 years, and for the seven human capital indicators,
it is primary school. Parentheses report standard deviations.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
32
Table 8: Distribution of Workers Across Industries in February 2020
—————– Unmatched —————– —————– Matched —————–
Full sample Treatment Control Full sample Treatment Control
Mining and quarrying 0.003 0.001 0.003 0.001 0.001 0.001
(0.057) (0.024) (0.059) (0.026) (0.024) (0.027)
Manufacturing 0.220 0.086 0.230 0.085 0.086 0.085
(0.414) (0.280) (0.421) (0.279) (0.280) (0.279)
Electricity, gas, steam etc. 0.005 0.000 0.005 0.000 0.000 0.000
(0.070) (0.018) (0.072) (0.016) (0.018) (0.014)
Water supply, sewerage etc. 0.006 0.001 0.007 0.001 0.001 0.001
(0.079) (0.035) (0.081) (0.034) (0.035) (0.033)
Construction 0.137 0.024 0.145 0.023 0.024 0.023
(0.344) (0.152) (0.352) (0.151) (0.152) (0.150)
Wholesale and retail trade 0.183 0.356 0.170 0.360 0.356 0.365
(0.387) (0.479) (0.376) (0.480) (0.479) (0.481)
Transportation 0.058 0.049 0.059 0.048 0.049 0.047
(0.234) (0.215) (0.236) (0.214) (0.215) (0.213)
Accommodation and food service 0.018 0.173 0.006 0.177 0.173 0.181
(0.133) (0.378) (0.080) (0.382) (0.378) (0.385)
Information and communication 0.066 0.021 0.069 0.021 0.021 0.022
(0.248) (0.144) (0.254) (0.145) (0.144) (0.146)
Financial and insurance 0.076 0.006 0.081 0.006 0.006 0.006
(0.265) (0.076) (0.273) (0.075) (0.076) (0.074)
Real estate 0.023 0.015 0.024 0.015 0.015 0.016
(0.151) (0.121) (0.153) (0.123) (0.121) (0.124)
Knowledge-based services 0.084 0.029 0.088 0.029 0.029 0.029
(0.277) (0.169) (0.283) (0.168) (0.169) (0.167)
Travel agents, cleaning etc. 0.054 0.056 0.054 0.055 0.056 0.053
(0.225) (0.230) (0.225) (0.228) (0.230) (0.225)
Public administration, defense etc. 0.002 0.003 0.002 0.003 0.003 0.003
(0.046) (0.058) (0.045) (0.056) (0.058) (0.054)
Education 0.004 0.012 0.003 0.012 0.012 0.013
(0.061) (0.107) (0.056) (0.110) (0.107) (0.113)
Human health and social work 0.038 0.064 0.036 0.063 0.064 0.061
(0.192) (0.246) (0.187) (0.242) (0.246) (0.239)
Arts, entertainment and recreation 0.007 0.047 0.003 0.046 0.047 0.045
(0.081) (0.213) (0.059) (0.210) (0.213) (0.208)
Other service activities 0.010 0.055 0.007 0.052 0.055 0.048
(0.101) (0.229) (0.083) (0.221) (0.229) (0.214)
Observations 743,509 51,632 691,877 90,231 51,632 38,599
Note: The table shows sample means of the industry dummies that are included in the covariate set for the full sample
and the subsamples of furloughed (treated) and non-compensated (control) workers both before and after matching.
These are measured in February 2020, and the reference category is agriculture, forestry, and fishing. Parentheses report
standard deviations.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
33
Table 9: Distribution of Workers Across Occupations in February 2020
————— Unmatched ————— ————— Matched —————
Full sample Treatment Control Full sample Treatment Control
Top management and legislators 0.012 0.003 0.013 0.003 0.003 0.003
(0.110) (0.058) (0.113) (0.057) (0.058) (0.056)
Administrative managers etc. 0.019 0.013 0.019 0.012 0.013 0.012
(0.136) (0.112) (0.137) (0.110) (0.112) (0.109)
Production and services managers 0.014 0.005 0.014 0.005 0.005 0.005
(0.116) (0.070) (0.118) (0.069) (0.070) (0.068)
Hotel, restaurant, retail managers etc. 0.008 0.017 0.007 0.017 0.017 0.017
(0.087) (0.128) (0.084) (0.128) (0.128) (0.128)
Science and engineering professionals 0.046 0.011 0.048 0.011 0.011 0.011
(0.209) (0.106) (0.214) (0.105) (0.106) (0.104)
Health professionals 0.012 0.031 0.010 0.031 0.031 0.032
(0.108) (0.173) (0.102) (0.174) (0.173) (0.175)
Teaching professionals 0.019 0.018 0.019 0.018 0.018 0.018
(0.136) (0.134) (0.136) (0.134) (0.134) (0.135)
Business and administration professionals 0.063 0.017 0.066 0.018 0.017 0.018
(0.242) (0.131) (0.248) (0.131) (0.131) (0.132)
Information technology professionals etc. 0.045 0.007 0.047 0.007 0.007 0.006
(0.206) (0.084) (0.212) (0.081) (0.084) (0.079)
Legal, social and cultural professionals 0.017 0.007 0.018 0.008 0.007 0.009
(0.130) (0.084) (0.133) (0.088) (0.084) (0.093)
Science and engineering ass. professionals 0.045 0.018 0.047 0.018 0.018 0.017
(0.207) (0.132) (0.212) (0.131) (0.132) (0.131)
Health associate professionals 0.010 0.032 0.009 0.032 0.032 0.033
(0.101) (0.176) (0.092) (0.177) (0.176) (0.177)
Business and admin. ass. professionals 0.088 0.068 0.089 0.068 0.068 0.069
(0.283) (0.252) (0.285) (0.252) (0.252) (0.253)
Legal, social and cultural ass. professionals 0.007 0.033 0.005 0.031 0.033 0.029
(0.081) (0.179) (0.067) (0.173) (0.179) (0.167)
Information technicians etc. 0.012 0.004 0.013 0.005 0.004 0.005
(0.109) (0.067) (0.112) (0.068) (0.067) (0.070)
Secretaries and office clerks 0.045 0.041 0.045 0.042 0.041 0.043
(0.207) (0.199) (0.207) (0.201) (0.199) (0.204)
Customer services clerks 0.014 0.034 0.012 0.032 0.034 0.030
(0.116) (0.181) (0.109) (0.176) (0.181) (0.170)
Numerical and material recording clerks 0.041 0.025 0.042 0.025 0.025 0.026
(0.199) (0.156) (0.202) (0.158) (0.156) (0.159)
Other clerical support workers 0.010 0.013 0.010 0.013 0.013 0.014
(0.102) (0.113) (0.101) (0.115) (0.113) (0.116)
Personal service workers 0.026 0.149 0.017 0.139 0.149 0.130
(0.159) (0.356) (0.128) (0.346) (0.356) (0.336)
Sales workers 0.057 0.165 0.049 0.171 0.165 0.177
(0.233) (0.371) (0.217) (0.376) (0.371) (0.381)
Personal care workers 0.024 0.019 0.024 0.019 0.019 0.019
(0.153) (0.138) (0.154) (0.137) (0.138) (0.136)
Protective services workers 0.005 0.011 0.004 0.011 0.011 0.011
(0.068) (0.103) (0.065) (0.105) (0.103) (0.106)
Building workers and painters 0.067 0.014 0.071 0.015 0.014 0.015
(0.251) (0.119) (0.257) (0.121) (0.119) (0.122)
Metal and machinery workers 0.048 0.035 0.049 0.035 0.035 0.036
(0.213) (0.184) (0.215) (0.185) (0.184) (0.186)
Handicraft and printing workers 0.003 0.005 0.003 0.005 0.005 0.005
(0.058) (0.068) (0.057) (0.069) (0.068) (0.071)
Electrical and electronic workers 0.027 0.009 0.028 0.009 0.009 0.009
(0.162) (0.093) (0.166) (0.094) (0.093) (0.094)
Food processing workers etc. 0.013 0.009 0.013 0.009 0.009 0.009
(0.112) (0.094) (0.113) (0.095) (0.094) (0.097)
Stationary plant and machine operators 0.047 0.020 0.049 0.019 0.020 0.018
(0.211) (0.139) (0.215) (0.136) (0.139) (0.134)
Assemblers 0.010 0.006 0.011 0.006 0.006 0.006
(0.102) (0.080) (0.103) (0.079) (0.080) (0.078)
Drivers and mobile plant operators 0.047 0.026 0.048 0.026 0.026 0.026
(0.211) (0.160) (0.215) (0.160) (0.160) (0.159)
Cleaners and helpers 0.019 0.045 0.017 0.045 0.045 0.045
(0.136) (0.207) (0.129) (0.207) (0.207) (0.208)
Agricultural and forestry laborers etc. 0.001 0.000 0.001 0.000 0.000 0.000
(0.026) (0.022) (0.026) (0.021) (0.022) (0.020)
Mining and construction laborers etc. 0.064 0.045 0.066 0.045 0.045 0.046
(0.245) (0.207) (0.248) (0.208) (0.207) (0.209)
Food preparation assistants 0.006 0.032 0.004 0.037 0.032 0.041
(0.077) (0.177) (0.063) (0.188) (0.177) (0.198)
Refuse workers etc. 0.007 0.009 0.007 0.009 0.009 0.009
(0.086) (0.096) (0.085) (0.096) (0.096) (0.097)
Other occupations 0.000 0.000 0.000 0.000 0.000 0.000
(0.021) (0.017) (0.022) (0.019) (0.017) (0.021)
Observations 743,509 51,632 691,877 90,231 51,632 38,599
Note: The table shows sample means of the occupation dummies that are included in the covariate set for the
full sample and the subsamples of furloughed (treated) and non-compensated (control) workers both before and after
matching. These are measured in February 2020, and the reference category is agricultural, forestry, and fishery workers.
Parentheses report standard deviations.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
34
Appendix E – Results from the Propensity Score Estimation
Table 10: Estimated Effects of Covariates on the Furlough Probability
Logit estimates Average marginal effects
Age (Years) -0.010*** -0.001***
(0.001) (0.000)
Man -0.347*** -0.018***
(0.012) (0.001)
Foreign 0.203*** 0.011***
(0.020) (0.001)
Youngest child is ≤ 5 years 0.023 0.001
(0.017) (0.001)
Youngest child is 6-10 years -0.019 -0.001
(0.018) (0.001)
Youngest child is 11-15 years -0.052** -0.003**
(0.019) (0.001)
High school 0.169*** 0.009***
(0.020) (0.001)
Vocational training 0.094*** 0.005***
(0.015) (0.001)
Short higher education -0.252*** -0.012***
(0.026) (0.001)
Medium higher education -0.192*** -0.009***
(0.025) (0.001)
Bachelor’s degree -0.205*** -0.010***
(0.037) (0.002)
Long higher education -0.456*** -0.020***
(0.027) (0.001)
Research education -0.900*** -0.034***
(0.102) (0.003)
Full-time 0.013 0.001
(0.012) (0.001)
Inflow to employment -0.094 -0.005
(0.056) (0.003)
Exit from employment -3.529*** -0.066***
(0.230) (0.001)
Several jobs 0.659*** 0.039***
(0.017) (0.001)
Job-to-job transition -0.269*** -0.012***
(0.035) (0.001)
Tenure (Years) -0.022*** -0.001***
(0.001) (0.000)
Avg. monthly labor income in 2019 (DKK 1,000) -0.015*** -0.001***
(0.001) (0.000)
Constant -4.942***
(0.211)
Observations 743,509 743,509
Note: The table shows estimation results from a logistic regression of the baseline covariate set on the
treatment indicator. It both presents non-transformed logit estimates and average marginal effects. Average
marginal effects are evaluated for discrete changes of dummy variables from their base level and one unit
changes of continuous variables. Parentheses report standard errors. The reference category for the children
indicators is having no children or only children older than 15 years, and for the human capital indicators,
it is primary school. * p<0.05, ** p<0.01, *** p<0.001. Estimation results for the industry and occupation
dummies are presented in tables 11-13 in appendix E.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
35
Table 11: Estimated Effects of Industries on the Furlough Probability
Logit estimates Average marginal effects
Mining and quarrying 0.263 0.014
(0.230) (0.013)
Manufacturing 0.932*** 0.058***
(0.134) (0.010)
Electricity, gas, steam etc. -0.881** -0.033***
(0.277) (0.008)
Water supply, sewerage etc. 0.123 0.006
(0.185) (0.010)
Construction 0.175 0.009
(0.137) (0.008)
Wholesale and retail trade 2.220*** 0.160***
(0.133) (0.015)
Transportation 1.705*** 0.135***
(0.135) (0.015)
Accommodation and food service 4.159*** 0.557***
(0.134) (0.024)
Information and communication 0.904*** 0.059***
(0.137) (0.011)
Financial and insurance -0.654*** -0.027***
(0.145) (0.005)
Real estate 0.640*** 0.039***
(0.138) (0.010)
Knowledge-based services 0.841*** 0.053***
(0.135) (0.011)
Travel agents, cleaning etc. 1.480*** 0.111***
(0.134) (0.014)
Public administration, defense etc. 2.281*** 0.221***
(0.157) (0.024)
Education 3.303*** 0.393***
(0.142) (0.026)
Human health and social work 2.080*** 0.184***
(0.135) (0.018)
Arts, entertainment and recreation 4.155*** 0.551***
(0.137) (0.024)
Other service activities 3.359*** 0.403***
(0.135) (0.025)
Observations 743,509 743,509
Note: The table shows estimation results from a logistic regression of the baseline covariate set on the
treatment indicator. It both presents non-transformed logit estimates and average marginal effects of the
industry dummies. Average marginal effects are evaluated for a discrete change from its base level, which is
agriculture, forestry, and fishing. Parentheses report standard errors. * p<0.05, ** p<0.01, *** p<0.001.
Estimation results for the occupation dummies and the remaining covariates are presented in tables 10 and
12-13 in appendix E.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
36
Table 12: Estimated Effects of Occupations on the Furlough Probability (1/2)
Logit estimates Average marginal effects
Top management and legislators 0.944*** 0.063***
(0.183) (0.016)
Administrative managers etc. 1.763*** 0.148***
(0.171) (0.021)
Production and services managers 1.461*** 0.114***
(0.178) (0.020)
Hotel, restaurant, retail managers etc. 1.914*** 0.168***
(0.170) (0.022)
Science and engineering professionals 1.368*** 0.103***
(0.171) (0.018)
Health professionals 2.747*** 0.292***
(0.169) (0.028)
Teaching professionals 1.178*** 0.084***
(0.170) (0.016)
Business and administration professionals 1.497*** 0.116***
(0.170) (0.019)
Information technology professionals etc. 0.867*** 0.056***
(0.175) (0.014)
Legal, social and cultural professionals 0.754*** 0.048***
(0.175) (0.014)
Science and engineering ass. professionals 1.608*** 0.128***
(0.169) (0.019)
Health associate professionals 2.793*** 0.301***
(0.168) (0.028)
Business and admin. ass. professionals 1.816*** 0.146***
(0.167) (0.019)
Legal, social and cultural ass. professionals 3.048*** 0.348***
(0.169) (0.030)
Information technicians etc. 1.176*** 0.084***
(0.180) (0.017)
Secretaries and office clerks 1.753*** 0.143***
(0.167) (0.020)
Customer services clerks 2.668*** 0.280***
(0.168) (0.028)
Numerical and material recording clerks 1.375*** 0.102***
(0.168) (0.017)
Other clerical support workers 2.018*** 0.182***
(0.171) (0.023)
Observations 743,509 743,509
Note: The table shows estimation results from a logistic regression of the baseline covariate set on the
treatment indicator. It both presents non-transformed logit estimates and average marginal effects of the
occupation dummies. Average marginal effects are evaluated for a discrete change from its base level, which
is agricultural, forestry, and fishery workers. Parentheses report standard errors. * p<0.05, ** p<0.01, ***
p<0.001. Estimation results for the industry dummies and the remaining covariates are presented in tables
10-11 in appendix E.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
37
Table 13: Estimated Effects of Occupations on the Furlough Probability (2/2)
Logit estimates Average marginal effects
Personal service workers 3.083*** 0.353***
(0.167) (0.029)
Sales workers 2.325*** 0.212***
(0.166) (0.024)
Personal care workers 0.767*** 0.048***
(0.169) (0.013)
Protective services workers 2.528*** 0.259***
(0.173) (0.028)
Building workers and painters 1.323*** 0.098***
(0.171) (0.017)
Metal and machinery workers 1.806*** 0.149***
(0.167) (0.020)
Handicraft and printing workers 2.688*** 0.286***
(0.180) (0.030)
Electrical and electronic workers 1.708*** 0.142***
(0.173) (0.021)
Food processing workers etc. 1.537*** 0.122***
(0.173) (0.020)
Stationary plant and machine operators 1.453*** 0.111***
(0.169) (0.018)
Assemblers 1.903*** 0.167***
(0.175) (0.023)
Drivers and mobile plant operators 1.377*** 0.102***
(0.169) (0.017)
Cleaners and helpers 2.158*** 0.199***
(0.167) (0.024)
Agricultural and forestry laborers etc. 1.356*** 0.103***
(0.277) (0.029)
Mining and construction laborers etc. 1.694*** 0.135***
(0.167) (0.019)
Food preparation assistants 2.168*** 0.205***
(0.170) (0.025)
Refuse workers etc. 2.145*** 0.200***
(0.174) (0.025)
Other occupations 1.052** 0.073*
(0.327) (0.030)
Observations 743,509 743,509
Note: The table shows estimation results from a logistic regression of the baseline covariate set on the
treatment indicator. It both presents non-transformed logit estimates and average marginal effects of the
occupation dummies. Average marginal effects are evaluated for a discrete change from its base level, which
is agricultural, forestry, and fishery workers. Parentheses report standard errors. * p<0.05, ** p<0.01, ***
p<0.001. Estimation results for the industry dummies and the remaining covariates are presented in tables
10-11 in appendix E.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
38
Appendix F – Standardized Biases Before and After Matching
Figure 6: Standardized Biases Across Selected Covariates Before and After Matching
High School
20
Several jobs
19 Unmatched
Foreign
18 Matched
Job-to-job transition
17
Inflow to employment
16
Vocational training
15
Youngest child is up to 5 years
14
Bachelor's degree
13
Youngest child is 6-10 years
12
Medium higher education
11
Youngest child is 11-15 years
10
Research education
9
Short higher education
8
Exit from employment
7
Full-time
6
Long higher education
5
Tenure
4
Age
3
Man
2
Avg. labor income past 12 months
1
0
-70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70
Standardized bias (%)
Note: The figure shows standardized percentage biases across covariates other than industry
and occupation before and after matching. Standardized biases measure, for each covariate, the
difference between the means of the treatment and control groups as a percentage of the square root
of the average variance across the two groups (Rosenbaum and Rubin, 1985). A positive bias means
that the value of the particular covariate is higher in the treatment group than in the control group
(and vice versa). Standardized biases across industries and occupations are presented in figure 2 in
section 4.1 and figure 7 in appendix F, respectively.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
39
Figure 7: Standardized Biases Across Occupations Before and After Matching
Personal service workers
37
Sales workers
36 Unmatched
Food preparation assistants
35 Matched
Legal, social and cultural ass. professionals
34
Health associate professionals
33
Cleaners and helpers
32
Customer services clerks
31
Health professionals
30
Hotel, restaurant and retail managers etc.
29
Protective services workers
28
Other clerical support workers
27
Handicraft and printing workers
26
Refuse workers and other elementary workers
25
Teaching professionals
24
Other occupations
23
Agricultural and forestry laborers etc.
22
Secretaries and office clerks
21
Personal care workers
20
Food processing workers etc.
19
Assemblers
18
Administrative managers etc.
17
Metal and machinery workers
16
Business and admin. ass. professionals
15
Information technicians etc.
14
Mining and construction laborers etc.
13
Production and services managers
12
Numerical and material recording clerks
11
Legal, social and cultural professionals
10
Top management and legislators9
Drivers and mobile plant operators8
Electrical and electronic workers7
Stationary plant and machine operators6
Science and engineering ass. professionals5
Science and engineering professionals4
Business and administration professionals3
Information technology professionals etc.2
Building workers and painters1
0
-70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70
Standardized bias (%)
Note: The figure shows standardized percentage biases across occupations in the covariate set
before and after matching. Standardized biases measure, for each covariate, the difference between
the means of the treatment and control groups as a percentage of the square root of the average
variance across the two groups (Rosenbaum and Rubin, 1985). A positive bias means that the value
of the particular covariate is higher in the treatment group than in the control group (and vice
versa). Standardized biases across industries and the remaining covariates are presented in figure 2
in section 4.1 and figure 6 in appendix F, respectively.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
40
Appendix G – Labor Income Distributions for Furloughed Workers
Figure 8: Labor Income Distributions for Furloughed Workers
(a) February (b) March
(c) April (d) May
(e) June (f) July
(g) August
Note: The figures show monthly kernel density estimates of labor income distributions for
workers furloughed from March 2020 to June 2020 in the baseline sample for the years 2018-2020.
The kernel used is an Epanechnikov, and the bandwidth is 1.0. Panel (8a) shows the distributions
for the pre-treatment month February 2020, whereas panels (8b)-(8g) show the distributions for
the post-treatment months March to August 2020.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
41
Appendix H – Main Estimation Results
Table 14: Estimated Effects of the Scheme on the Exit Probability
—– Main —– ———————– Robustness ———————–
(1) (2) (3) (4)
DJan-19 × DTREAT -0.000 -0.000 -0.001 -0.000
(0.001) (0.001) (0.001) (0.001)
DFeb-19 × DTREAT -0.000 -0.000 -0.001 0.000
(0.001) (0.001) (0.001) (0.001)
DMar-19 × DTREAT 0.001 0.000 0.001 0.001
(0.001) (0.001) (0.001) (0.001)
DApr-19 × DTREAT -0.001 -0.001 -0.000 -0.001
(0.001) (0.001) (0.001) (0.001)
DMay-19 × DTREAT -0.000 -0.000 -0.001 0.001
(0.001) (0.001) (0.001) (0.001)
DJun-19 × DTREAT -0.000 -0.000 0.000 -0.000
(0.001) (0.001) (0.001) (0.001)
DJul-19 × DTREAT 0.000 0.000 -0.001 0.001
(0.001) (0.001) (0.001) (0.001)
DAug-19 × DTREAT -0.000 -0.000 -0.001 -0.000
(0.001) (0.001) (0.001) (0.001)
DSep-19 × DTREAT 0.000 0.000 -0.000 0.000
(0.001) (0.001) (0.001) (0.001)
DOct-19 × DTREAT -0.000 -0.000 -0.001 -0.001
(0.000) (0.000) (0.000) (0.001)
DNov-19 × DTREAT -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000)
DDec-19 × DTREAT -0.000 -0.000 -0.001 -0.001
(0.000) (0.000) (0.000) (0.000)
DJan-20 × DTREAT 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000)
DMar-20 × DTREAT -0.030*** -0.030*** -0.030*** -0.034***
(0.003) (0.003) (0.003) (0.003)
DApr-20 × DTREAT -0.010*** -0.010*** -0.011*** -0.012***
(0.001) (0.001) (0.001) (0.001)
DMay-20 × DTREAT -0.003*** -0.003*** -0.003*** -0.004***
(0.001) (0.001) (0.001) (0.001)
DJun-20 × DTREAT 0.000 0.000 0.000 0.000
(0.001) (0.001) (0.001) (0.001)
DJul-20 × DTREAT -0.000 -0.000 -0.000 -0.001
(0.001) (0.001) (0.001) (0.001)
DAug-20 × DTREAT 0.003*** 0.003*** 0.002* 0.004***
(0.001) (0.001) (0.001) (0.001)
Industry-month FE Yes Yes Yes Yes
Occupation-month FE No Yes Yes No
Firm Covariates No No Yes No
Observations 2,622,010 2,622,010 2,410,426 2,315,373
Note: The table shows estimates for the average effects of the wage compensation scheme on furloughed
workers’ probability of having an exit from employment in the post-scheme months, March 2020 to August
2020, and some pre-scheme months, January 2019 to January 2020. Panel (1) presents the main results from
the preferred matched difference-in-differences specification with industry-calendar-month fixed effects.
Panels (2)-(4) each represent a robustness test. Panel (2) includes occupation-calendar-month fixed effects.
Panel (3) re-matches workers on a covariate set extended with additional firm characteristics. Panel (4)
excludes workers with more than one job in February 2020. Parentheses report standard errors clustered
at the firm level. * p<0.05, ** p<0.01, *** p<0.001.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
42
Table 15: Heterogenous Effects on the Exit Probability Across Firm Tenure
(1) (2) (3) (4)
≤ 1 year ]1; 4] years ]4; 8] years > 8 years
DJan-19 × DTREAT -0.002 0.000 -0.000 0.002
(0.002) (0.001) (0.001) (0.001)
DFeb-19 × DTREAT -0.002 0.001** -0.001 0.002*
(0.002) (0.000) (0.001) (0.001)
DMar-19 × DTREAT -0.000 0.001 0.000 0.003*
(0.002) (0.001) (0.001) (0.001)
DApr-19 × DTREAT -0.002 0.001 -0.001 0.001
(0.002) (0.001) (0.001) (0.001)
DMay-19 × DTREAT -0.001 0.000 0.001 0.000
(0.002) (0.001) (0.001) (0.001)
DJun-19 × DTREAT -0.001 -0.000 0.001 0.001
(0.002) (0.001) (0.001) (0.001)
DJul-19 × DTREAT -0.000 -0.000 0.001* 0.002
(0.002) (0.001) (0.001) (0.001)
DAug-19 × DTREAT -0.001 0.000 -0.001 0.001
(0.001) (0.001) (0.001) (0.001)
DSep-19 × DTREAT 0.001 -0.001 0.000 0.002*
(0.001) (0.001) (0.001) (0.001)
DOct-19 × DTREAT -0.001 0.000 -0.000 -0.000
(0.001) (0.000) (0.001) (0.001)
DNov-19 × DTREAT -0.001 0.001 -0.000 0.000
(0.001) (0.000) (0.000) (0.000)
DDec-19 × DTREAT -0.001 -0.000 -0.000 0.001
(0.001) (0.000) (0.000) (0.000)
DJan-20 × DTREAT -0.000 0.000 -0.000 0.001
(0.000) (0.000) (0.000) (0.000)
DMar-20 × DTREAT -0.043*** -0.026*** -0.014*** -0.007
(0.004) (0.003) (0.003) (0.005)
DApr-20 × DTREAT -0.014*** -0.009*** -0.003* -0.001
(0.002) (0.002) (0.001) (0.001)
DMay-20 × DTREAT -0.005** -0.003* -0.003* 0.001
(0.002) (0.001) (0.001) (0.002)
DJun-20 × DTREAT 0.001 -0.000 0.000 -0.000
(0.002) (0.001) (0.001) (0.002)
DJul-20 × DTREAT 0.001 -0.002 0.000 0.002
(0.002) (0.002) (0.002) (0.002)
DAug-20 × DTREAT 0.003* 0.003* 0.003 0.006**
(0.001) (0.002) (0.002) (0.002)
Industry-month FE Yes Yes Yes Yes
Occupation-month FE No No No No
Firm Covariates No No No No
Observations 996,644 1,010,786 479,567 135,013
Note: The table shows the preferred matched difference-in-differences estimates for the average effects
of the wage compensation scheme on furloughed workers’ probability of having an exit from employment
across four subgroups of workers with different levels of ex ante firm tenure. Panels (1)-(4) refer to workers
with (1) up to 1 year, (2) ]1; 4] years, (3) ]4; 8] years, and (4) more than 8 years of tenure in February
2020, respectively. Parentheses report standard errors clustered at the firm level. * p<0.05, ** p<0.01,
*** p<0.001.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
43
Table 16: Estimated Effects of the Scheme on Log-Labor Income
—– Main —– ————————— Robustness —————————
(1) (2) (3) (4) (5)
DJan-19 × DTREAT -0.002 -0.003 -0.009 -0.005 0.001
(0.007) (0.006) (0.007) (0.006) (0.006)
DFeb-19 × DTREAT 0.007 0.006 0.002 0.009 0.014**
(0.005) (0.005) (0.004) (0.005) (0.005)
DMar-19 × DTREAT 0.001 -0.002 -0.006 -0.001 0.000
(0.004) (0.004) (0.004) (0.004) (0.005)
DApr-19 × DTREAT -0.003 -0.004 -0.007 -0.005 -0.005
(0.007) (0.007) (0.007) (0.007) (0.007)
DMay-19 × DTREAT -0.004 -0.005 -0.011* -0.007 -0.006
(0.005) (0.005) (0.005) (0.005) (0.005)
DJun-19 × DTREAT 0.005 0.005 0.002 0.006 0.011
(0.006) (0.006) (0.006) (0.006) (0.006)
DJul-19 × DTREAT -0.012* -0.012* -0.013* -0.015* -0.007
(0.006) (0.005) (0.005) (0.006) (0.006)
DAug-19 × DTREAT -0.012* -0.012* -0.012* -0.013* -0.007
(0.006) (0.005) (0.006) (0.006) (0.006)
DSep-19 × DTREAT -0.000 -0.001 -0.001 -0.001 0.000
(0.005) (0.004) (0.004) (0.004) (0.005)
DOct-19 × DTREAT -0.001 -0.001 -0.003 -0.000 0.001
(0.004) (0.004) (0.004) (0.004) (0.004)
DNov-19 × DTREAT -0.004 -0.003 -0.007 -0.006 -0.003
(0.005) (0.005) (0.005) (0.005) (0.005)
DDec-19 × DTREAT -0.012* -0.011* -0.013* -0.013* -0.013*
(0.005) (0.005) (0.005) (0.005) (0.005)
DJan-20 × DTREAT -0.006 -0.007 -0.008* -0.010** -0.007
(0.004) (0.004) (0.004) (0.004) (0.004)
DMar-20 × DTREAT -0.034*** -0.035*** -0.032*** -0.030*** -0.045***
(0.004) (0.004) (0.005) (0.004) (0.004)
DApr-20 × DTREAT -0.050*** -0.050*** -0.053*** -0.053*** -0.045***
(0.006) (0.006) (0.007) (0.007) (0.007)
DMay-20 × DTREAT -0.027*** -0.026*** -0.024*** -0.028*** -0.017***
(0.004) (0.004) (0.005) (0.005) (0.004)
DJun-20 × DTREAT -0.009 -0.008 -0.009 -0.013* -0.006
(0.005) (0.005) (0.005) (0.005) (0.005)
DJul-20 × DTREAT -0.020*** -0.019*** -0.015** -0.020*** -0.010*
(0.005) (0.005) (0.005) (0.005) (0.005)
DAug-20 × DTREAT -0.019*** -0.018*** -0.018*** -0.018*** -0.011*
(0.005) (0.005) (0.005) (0.005) (0.004)
Industry-month FE Yes Yes Yes Yes Yes
Occupation-month FE No Yes Yes No No
Firm Covariates No No Yes No No
Observations 2,622,010 2,622,010 2,410,426 2,315,373 2,454,997
Note: The table shows estimates for the average effects of the wage compensation scheme on furloughed
workers’ log-labor income in the post-scheme months, March 2020 to August 2020, and some pre-scheme months,
January 2019 to January 2020. Panel (1) presents the main results from the preferred matched difference-in-
differences specification with industry-calendar-month fixed effects. Panels (2)-(5) each represent a robustness
test. Panel (2) includes occupation-calendar-month fixed effects. Panel (3) re-matches workers on a covariate set
extended with additional firm characteristics. Panel (4) excludes workers with more than one job in February
2020. Panel (5) re-matches workers on a restricted sample that only includes workers who are employed during
the compensation period. Parentheses report standard errors clustered at the firm level. * p<0.05, ** p<0.01,
*** p<0.001.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
44
Table 17: Estimated Effects of the Scheme on the Job-to-Job Transition Probability
—– Main —– —————————– Robustness —————————–
(1) (2) (3) (4) (5)
DJan-19 × DTREAT 0.005 0.006* 0.001 0.004 0.004
(0.003) (0.003) (0.003) (0.002) (0.003)
DFeb-19 × DTREAT 0.005* 0.006* 0.001 0.003 0.005
(0.002) (0.002) (0.002) (0.002) (0.002)
DMar-19 × DTREAT 0.001 0.001 -0.002 0.001 -0.000
(0.002) (0.002) (0.002) (0.002) (0.003)
DApr-19 × DTREAT 0.002 0.002 -0.004 -0.001 -0.000
(0.003) (0.003) (0.003) (0.002) (0.003)
DMay-19 × DTREAT 0.001 0.001 -0.002 0.000 0.001
(0.003) (0.003) (0.003) (0.003) (0.003)
DJun-19 × DTREAT 0.002 0.002 -0.004 -0.001 -0.000
(0.002) (0.002) (0.002) (0.002) (0.003)
DJul-19 × DTREAT -0.004 -0.004 -0.001 -0.006 -0.004
(0.005) (0.005) (0.003) (0.005) (0.005)
DAug-19 × DTREAT 0.000 0.000 -0.003 -0.002 -0.002
(0.002) (0.002) (0.002) (0.002) (0.003)
DSep-19 × DTREAT -0.001 -0.001 -0.003 -0.004* 0.001
(0.002) (0.002) (0.002) (0.002) (0.002)
DOct-19 × DTREAT 0.002 0.002 0.000 0.001 -0.001
(0.002) (0.002) (0.002) (0.002) (0.003)
DNov-19 × DTREAT -0.002 -0.002 -0.003 -0.002 -0.001
(0.002) (0.002) (0.002) (0.002) (0.003)
DDec-19 × DTREAT -0.001 -0.001 -0.005* -0.002 -0.002
(0.002) (0.002) (0.002) (0.002) (0.002)
DJan-20 × DTREAT -0.001 -0.001 -0.005 -0.003 -0.002
(0.004) (0.004) (0.004) (0.003) (0.004)
DMar-20 × DTREAT -0.002 -0.002 -0.008*** 0.000 -0.002
(0.003) (0.003) (0.002) (0.002) (0.003)
DApr-20 × DTREAT -0.006** -0.006** -0.010*** -0.006*** -0.006**
(0.002) (0.002) (0.002) (0.001) (0.002)
DMay-20 × DTREAT -0.002 -0.002 -0.005* -0.006*** -0.001
(0.002) (0.002) (0.002) (0.002) (0.002)
DJun-20 × DTREAT -0.002 -0.002 -0.005** -0.004* -0.001
(0.002) (0.002) (0.002) (0.002) (0.002)
DJul-20 × DTREAT 0.000 0.000 -0.003 -0.000 0.002
(0.003) (0.003) (0.002) (0.002) (0.003)
DAug-20 × DTREAT 0.001 0.001 -0.002 0.002 0.002
(0.002) (0.002) (0.002) (0.002) (0.003)
Industry-month FE Yes Yes Yes Yes Yes
Occupation-month FE No Yes Yes No No
Firm Covariates No No Yes No No
Observations 2,622,010 2,622,010 2,410,426 2,315,373 2,454,997
Note: The table shows estimates for the average effects of the wage compensation scheme on furloughed workers’
probability of having a job-to-job transition in the post-scheme months, March 2020 to August 2020, and some
pre-scheme months, January 2019 to January 2020. Panel (1) presents the main results from the preferred matched
difference-in-differences specification with industry-calendar-month fixed effects. Panels (2)-(5) each represent a
robustness test. Panel (2) includes occupation-calendar-month fixed effects. Panel (3) re-matches workers on a
covariate set extended with additional firm characteristics. Panel (4) excludes workers with more than one job in
February 2020. Panel (5) re-matches workers on a restricted sample that only includes workers who are employed
during the compensation period. Parentheses report standard errors clustered at the firm level. * p<0.05, **
p<0.01, *** p<0.001.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
45
Appendix I – Estimation Results for All Furloughed Workers
Figure 9: Estimated Effects of the Scheme for All Furloughed Workers
(a) Employment-exit Probability (b) Log-labor Income
Percentage Points Percent
2.0 4.0
1.0 2.0
0.0 0.0
-1.0 -2.0
-2.0 -4.0
-3.0 -6.0
-4.0 -8.0
Mar-18 Sep-18 Mar-19 Sep-19 Mar-20 Sep-20 Mar-18 Sep-18 Mar-19 Sep-19 Mar-20 Sep-20
(c) Job-to-job Transition Probability
Percentage Points
3.0
2.0
1.0
0.0
-1.0
-2.0
-3.0
Mar-18 Sep-18 Mar-19 Sep-19 Mar-20 Sep-20
Note: The figure shows estimates for the average impact of the wage compensation scheme on
the three labor market outcomes from an alternative sample selection where all furloughed workers,
regardless of their compensation period, are included in the treatment group. Estimates are obtained
from a regression of the preferred matched difference-in-differences specification with industry-calendar-
month fixed effects. In panel (a), the outcome is the probability of having an exit from employment.
In panel (b), it is total log-labor income, and in panel (c), it is the probability of having a job-
to-job transition. The gray lines that encircle the estimates are 95% confidence bands based on
clustered standard errors at the firm level. The dashed vertical lines mark the introduction of the wage
compensation scheme.
Source: Own calculations based on Danish administrative register data from Statistics Denmark.
46
W ORKING PAPER — DANMARK S NATIONALBANK
13 JANUARY 2022 — NO. 186
As a rule, Working Papers are not translated, but are available in
the original language used by the contributor.
Danmarks Nationalbank's Working Paper are published in PDF format
at www.nationalbanken.dk. A free electronic subscription is also
available at the website. The subscriber receives an e-mail notification
whenever a new Working Paper is published.
Text may be copied from this publication provided that the source is
specifically stated. Changes to or misrepresentation of the content are
not permitted.
Please direct any enquiries directly to the contributors or to Danmarks
Nationalbank, Communications, Kommunikation@nationalbanken.dk.
File and source
- File
- u050_dnk_cc3p0jp5-job-retention-during-the-covid-19-pandemic-pdf.pdf
- Size
- 2,203,850 bytes
- SHA-256
- c603c4d5edb09ea666970a61babaa80702353324494f37eacc8c262d3bea06ce
- Original
- www.nationalbanken.dk