Corporate Financial Frictions and Employee Mental Health

Abstract

This article argues that corporate financial frictions can have an adverse effect on employee mental health, an important determinant of employee productivity. To identify the causal effects of financial frictions, we exploit variation in firms’ need to refinance their long-term debt in 2008, a period when refinancing became more difficult due to the credit crunch. Using administrative microdata, we find that antidepressant use grows significantly more among employees of firms in higher need of debt refinancing. Much of this effect occurs at employees keeping their jobs, pointing to decreased perceptions of job security as a transmission channel.

Full text

JOURNAL OF FINANCIAL AND QUANTITATIVE ANALYSIS
© The Author(s), 2023. Published by Cambridge University Press on behalf of the Michael G. Foster
School of Business, University of Washington. This is an Open Access article, distributed under the terms
of the Creative Commons Attribution-NonCommercial-NoDerivatives licence (https://creativecommons.org/
licenses/by-nc-nd/4.0), which permits non-commercial re-use, distribution, and reproduction in any medium,
provided that no alterations are made and the original article is properly cited. The written permission of
Cambridge University Press must be obtained prior to any commercial use and/or adaptation of the article.
doi:10.1017/S0022109023000595
Corporate Financial Frictions and Employee
Mental Health
Dániel Kárpáti
Tilburg University Department of Finance
d.karpati@uvt.nl
Luc Renneboog
Tilburg University Department of Finance
luc.renneboog@uvt.nl (corresponding author)
Abstract
This article argues that corporate financial frictions can have an adverse effect on employee
mental health, an important determinant of employee productivity. To identify the causal
effects of financial frictions, we exploit variation in firms’need to refinance their long-term
debt in 2008, a period when refinancing became more difficult due to the credit crunch. Using
administrative microdata, we find that antidepressant use grows significantly more among
employees of firms in higher need of debt refinancing. Much of this effect occurs at
employees keeping their jobs, pointing to decreased perceptions of job security as a trans-
mission channel.
I. Introduction
A growing literature documents that financial constraints amplify the adverse
effects of economic shocks on firms’human capital. Giroud and Mueller (2017)
provide evidence that high-leverage firms decreased their employment more during
the Great Recession in response to local demand shocks. The authors argue that
leverage may impair firms’ability to retain temporarily unnecessarily employees
(labor hoarding), a practice that firms may otherwise find optimal in order to
preserve human capital and avoid hiring/rehiring costs. Caggese, Cuñat, and
Metzger (2019) find that financial constraints prompt firms experiencing economic
distress to implement suboptimal dismissal policies, firing short-tenured workers
with high future expected productivity. Baghai, Silva, Thell, and Vig (2016) doc-
ument that firms lose workers with the highest cognitive and noncognitive skills due
to financial distress as they approach bankruptcy, whereas Brown and Matsa (2016)
show that financial distress can discourage talented job applicants.
Results are based on calculations by Dániel Kárpáti of Tilburg University using nonpublic
microdata from Statistics Netherlands. The research for this publication was partly funded by the
Open Data Infrastructur e for Social Science and Economic Innovations (OD ISSEI) in the Netherlands
(www.odissei-data.nl).
1
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

In this article, we document a novel cost of financial constraints on firms’
human capital: we provide evidence that financial constraints can exacerbate the
adverse effects of economic shocks on employee mental health. Employees’mental
well-being should be a primary concern for any firm, given its role in employee
productivity, absenteeism, and employee turnover (Duijts, Kant, Swaen, van den
Brandt, and Zeegers (2007), Burton, Schultz, Chen, and Edington (2008), and
Bubonya, Cobb-Clark, and Wooden (2017)). We argue that financial constraints
contribute to a greater probability of job loss and that decreased job security may
trigger mental health problems also for employees who manage to keep their jobs.
To study the effects of financial constraints on employee mental health, we
must overcome two empirical challenges. First, we need to establish a quantitative
measure of mental health. To do so, we exploit rich administrative data from the
Netherlands, in particular a population-wide medicine use register, which records
annual binary indicators of medicine use grouped by 4-digit ATC (Anatomical
Therapeutic Chemical) codes. As a measure of mental health, we focus on the use
of antidepressants (ATC: N06A), drugs that are predominantly prescribed to treat
serious mental illnesses, such as depressive disorders, anxiety disorders, or bipolar
disorders (Gardarsdottir, Heerdink, van Dijk, and Egberts (2007), Simon, Stewart,
Beck, Ahmedani, Coleman, Whitebird, Lynch, Owen-Smith, Waitzfelder, Soumerai,
and Hunkeler (2014)). Although antidepressant use does not cover the complete
spectrum of mental health problems, especially milder conditions, general prac-
titioners in the Netherlands frequently employ antidepressants as the first line of
treatment for mental health complaints.
1
Furthermore, as the medically unjusti-
fied use of antidepressants is reported to be low (Piek, van der Meer, Hoogendijk,
Penninx, and Nolen (2011)), patients prescribed these medicines indeed suffer
from mental problems.
The second empirical challenge is how to disentangle the effects of financial
constraints from the effects of economic distress that make these constraints bind.
As the papers cited in the introductory paragraph also highlight, the adverse effects
of financial constraints on human capital are the most pronounced in bad economic
times. Yet, during economic distress, variables that could serve to measure a firm’s
financial health (such as profitability or firm leverage) are also likely correlated with
the firm’s sensitivity to the economic shock, the firm’s labor demand, and ultimately
the mental health of its employees.
Therefore, instead of focusing on contemporaneous measures of financial
health, we identify a balance sheet vulnerability that made firms more likely to
be financially constrained during a subsequent economic shock. In particular, we
exploit the unforeseen credit supply shock presented by the Global Financial Crisis
and employ an empirical strategy motivated by Almeida, Campello, Laranjeira, and
Weisbenner (2011).
2
Weconsider the long-term debt maturity structure of 352 large
Dutch companies that employed over 330,000 people on Jan. 1, 2008, and identify
firms as financially constrained if they had to refinance a large part (minimally 25%
1
In 2010, 30% of patients with any psychological diagnoses were prescribed antidepressants
(Nuijen, Emmen, Smit, Stirbu-Wagner, Veerbeek, and Verhaak (2012)).
2
A similar methodology was applied in several recent papers (e.g., Carvalho (2015), Benmelech,
Frydman, and Papanikolaou (2019), and Duval, Hong, and Timmer (2020)).
2 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

in our baseline model) of their long-term debt outstanding in 2008 (we call these
firms the high-repayment or treated firms).
The underlying idea of this identification strategy is that firms that had to repay
a larger share of their outstanding long-term debt in 2008 faced refinancing diffi-
culties due to the credit crunch. We offer two pieces of evidence in support. First,
bank lending is the main source of external financing for Dutch firms (Kalara and
Zhang (2018)), and the Netherlands experienced a strong negative bank credit
supply shock in 2007–2008 (Duchi and Elbourne (2016)). As Figure 1 reveals,
almost all Dutch banks tightened their lending standards (for large firms) in each
quarter starting from end-2007, which contributed to the slowdown of business
lending observed from mid-2008 (DNB (2009), van der Veer and Hoeberichts
(2016)). In the last quarter of 2008, the net borrowing of Dutch firms turned
significantly negative for the first time in many years (Figure 2). Second, there is
direct survey evidence indicating that firms experienced a negative credit supply
shock: in 2009:Q1, 21% of the Dutch companies reported the unavailability of bank
lending as the most important crisis-related problem they faced (56% among those
firms that reported any problems).
3
As the maturity profile of long-term debt is the cumulative outcome of
hard-to-reverse decisions made several years prior to 2008, it is unlikely that the
FIGURE 1
Credit Standards of Dutch Banks on Loans to Large Enterprises
Figure 1 shows the net percentages of banks tightening and easing their credit standards (overall) in the preceding quarter,
weighted by loans outstanding. Source: ECB (SDW item BLS.Q.NL.ALL.O.E.Z.B3.ST.S.BFNET).
–80
–60
–40
–20
0
20
40
60
80
100
120
2005:Q1
2005:Q4
2006:Q3
2007:Q2
2008:Q1
2008:Q4
2009:Q3
2010:Q2
2011:Q1
2011:Q4
2012:Q3
2013:Q2
2014:Q1
2014:Q4
2015:Q3
3
COEN Business Survey Netherlands 2009:Q1, administered by Statistics Netherlands. The sample
consists of establishments with more than 5 employees; the average sample size of the COEN surveys
is approximately 6,000 establishments. The crisis-related questions were first added in 2009:Q1. The
question of interest asks about the most important effect of the economic downturn that the respondent
experiences (problems acquiring credit, problems attracting equity, losses on deposits, value loss of
investments, increased debtor risk, or problems saving surplus funds), 62% mentions that none of these
effects are important (i.e., no important effects or important effects are unlisted). A total of 21% mentions
problems acquiring credit (22% for establishments with over 100 employees) and 11% mentions
increased debtor risk.
Kárpáti and Renneboog 3
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

2008 repayment share is correlated with the sensitivity of the firm to the
economic downturn or other unobservable factors. This is particularly true because
in our regression models, we control for time-invariant employee unobservables
(employee fixed effects) and we allow for different flexible time trends for firms
with distinct pre-crisis characteristics (controls × year fixed effects). The included
control variables (industry, firm size, cash ratio, long-term debt to assets, and cash
flow) aim to pick up any systematic differences in firms’long-term debt maturity
structure that might have also affected the firm’s economic perspectives and per-
sonnel policies, and thus its employees’mental health, during the crisis.
The results from the regression models suggest a significant and persistent
effect of the credit supply shock on employee mental health. People employed on
Jan. 1, 2008, by firms with at least 25% of their long-term debt maturing in 2008
faced a 0.44 pp (percentage points) higher average probability of antidepressant use
in the period of 2008 to 2012, which is an economically significant 9% increase
with respect to the 5% unconditional prevalence. The 9% increase in the probability
of antidepressant use is comparable in magnitude to the 7.5% rise in antidepressant
prescription volume due to a 20% decline in US housing prices between July 2006
and Feb. 2009 as estimated by Lin, Ketcham, Rosenquist, and Simon (2013).
These results are qualitatively robust to variations in control variables, restrict-
ing or broadening the sample of firms, altering the 25% refinancing cut-off, and
using pre-regression matching to remove any imbalances between employees of
treated and control firms. Wealso perform placebo tests to verify that our results are
not driven by the excess sensitivity of treated firms to the economic downturn in
2008–2009 (i.e., macroeconomic effects unrelated to the credit supply shock) and
that the relation between financial constraints and mental health does not apply in
firms where financial constraints are not expected to be binding because of internal
capital markets.
FIGURE 2
New Loans Minus Retired Bank Loans of Dutch Nonfinancial Companies
Figure 2 shows the retired bank loans minus new loans of Dutch nonfinancial companies (EURm). Source: Statistics
Netherlands –Quarterly sectoral accounts (CBS –Kwartaalsectorrekeningen).
2005:Q1
2005:Q2
2005:Q3
2005:Q4
2006:Q1
2006:Q2
2006:Q3
2006:Q4
2007:Q1
2007:Q2
2007:Q3
2007:Q4
2008:Q1
2008:Q2
2008:Q3
2008:Q4
2009:Q1
2009:Q2
2009:Q3
2009:Q4
2010:Q1
–15,000
–10,000
–5,000
0
5,000
10,000
15,000
20,000
4 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

The estimated 0.44 pp increase in antidepressant use is a weighted average
treatment effect on employees who left their job during the sample period (leavers)
and on those who stayed in their jobs (stayers). Based on the literature, we argue
that a main transmission channel from refinancing difficulties to employee mental
health is job loss for leavers and decreased job security for stayers. Financial
constraints may negatively affect firms’labor demand (Benmelech, Bergman,
and Seru (2011), Chodorow-Reich (2014), Giroud and Mueller (2017), Huber
(2018), and Popov and Rocholl (2018)), and the ensuing job loss can have an
adverse effect on employees’mental health (Browning and Heinesen (2012),
Ganster and Rosen (2013), and Schaller and Stevens (2015)). However, decreased
job security can damage employee mental health even in the absence of actual job
loss (Witte (1999), Burgard, Brand, and House (2009), Reichert and Tauchmann
(2011), and Kim and Von Dem Knesebeck (2015)). Green (2011) concludes that for
an employee of average employability the mental health effect of extreme job
insecurity is similar to the effect of unemployment.
We do indeed find that employees of high-repayment firms had a 6.2 pp higher
probability of job separation in the period of 2008 to 2010. Although this estimate
does not distinguish between involuntary job loss and voluntary job separation, we
provide two supplementary analyses that show that employees of high-repayment
firms likely faced greater job insecurity: These employees were more likely i) to be
dismissed with a permit from the Dutch Employee Insurance Agency, and ii) to
experience job separation followed by a gap in employment. For the period prior to
2008, we show that employee turnover was similar in high-repayment firms and in
control firms, and that treated and control employees in our sample experienced
similar trends in labor force attachment.
Can a greater propensity of job loss in treated firms completely explain the
increase in antidepressant use? We argue that this is not the case and that stayers also
suffered from deteriorating mental health. First, in a back-of-the-envelope calcu-
lation, we multiply the job loss estimates (with an upper bound of 6.2 pp) with the
effects of job loss on depression/anxiety reported by Schaller and Stevens (2015)
(1.6 pp). From this calculation, it is clear that the 0.44 pp overall increase in
antidepressant use may be rather high to be explained by greater job loss alone.
Second, we restrict our sample to employees who kept their jobs at least till the end
of the year in which we measure antidepressant use. In this subsample, we still find
that the probability of antidepressant use in treated firms was 0.28 pp higher in the
period of 2008 to 2012.
4
Finally, we study treatment heterogeneity among stayers to test whether job
insecurity is indeed a driver for greater antidepressant use for these employees.
Based on the economics and psychology literatures, we identify five personal/
household characteristics that are expected to increase the mental health burden
4
The group of employees who keep their job is a selected sample and selection is possibly endog-
enous to changes in mental health outcomes. For example, employees who stayed with financially
constrained firms might be in general more resilient to job insecurity. These employees might have
reacted more mildly to increasing job insecurity due to the economic downturn even in the absence of
financial constraints, introducing a downward bias in our estimates of the financial constraints’effects on
these employees’mental health. As we lack any good instruments for job separation, we cannot claim a
causal interpretation for our results on the subsample of employees who keep their job.
Kárpáti and Renneboog 5
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

of job insecurity: older age, being male, living without a partner, having children in
the household, and having a salary that constitutes a large share of total household
income. When we interact our treatment indicator with these moderator character-
istics, we find statistically significantly larger treatment effects for employees
without a partner, for those with children in their household, and for employees
whose salary constitutes a large share of their total household income. Treatment
effects appear to be larger for employees who are at least 45 years old, but the
difference is not statistically significant at any conventional level, while male and
female employees appear to be similarly affected by corporate financial stress. We
also find that the relation between corporate financial frictions and antidepressant
use is higher for employees with medium tenure, and for employees whose hourly
wage is in the top quartile of hourly wages in their firms (a proxy for managerial
employees). This latter group of managerial employees is potentially more aware
of the financial and economic difficulties of their firms and may be involved in
resolving them, both of which can generate additional mental stress. Finally,
antidepressant use increases more for treated employees who retain their jobs but
work in business units with a higher increase in job separations. Reasons could be
that the departure of colleagues induces an enhanced perception of job insecurity,
tensions consequential to the reorganization of the work, as well an increased
burden of work pressure. Furthermore, the severing of collegial or friendship ties
may reduce work satisfaction. Taken together, these results provide support to our
hypothesis that greater job insecurity is driving increased antidepressant use among
employees who do not lose their jobs.
This article relates to three strands of literature in finance and economics. First
of all, as cited in the introductory paragraph, a growing literature in finance studies
the effects of financial constraints on firms’human capital. We combine firm-level
financial data with rich employee-level data on antidepressant use to document a
novel cost of financial constraints, their detrimental effect on employee mental
health. We show that the mental health toll of financial constraints is not restricted
to dismissed employees but it is also substantial for employees who stay with the
firm. As argued above, the mental health of employees, particularly of those not
dismissed, should be a prime concern of firms due to mental illnesses’burden on
employee productivity.
5
Another strand of literature related to our work studies the health effects of
financial and economic crises. Several papers in this field report a negative corre-
lation between unemployment rates and mental health status (e.g., Charles and
DeCicca (2008), Tefft (2011), and Bradford and Lastrapes (2014)). We also study
how employment relations contributed to the mental health of employees during
a crisis period, but contrary to the previous literature, we use employer–employee
matched data to disentangle the mental health effects of the financial crisis (credit
supply shock) from the effects of the ensuing economic crisis (the Great Recession).
5
Another related strand of the literature studies the interaction of firm financial events, such as
takeovers and buyouts, and employee health. Bach, Baghai, Boss, and Silva (2021) document that
takeovers increase the incidence of stress, anxiety, depression, psychiatric medication usage, and even
suicide among affected employees. Garcia-Gomez, Maug, and Obernberger (2020) investigate the
health effects of private equity buyouts. Although the authors find no evidence that buyouts worsen
employees’health, they report that health characteristics strongly predict job loss after buyouts.
6 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

Furthermore, we show that crisis periods may have an adverse mental health
effect even on employees who manage to keep their jobs but who may suffer from
decreased perceptions of job security.
Finally, this article also relates to the literature on the health effects of job
displacement. Findings of this literature generally indicate a negative causal rela-
tion between job loss and mental health (Browning and Heinesen (2012), Schaller
and Stevens (2015)), although not unequivocally (Salm (2009)). The key difference
between these papers and our work is that while the job displacement literature’s
main interest is the effect of job loss per se, we focus on the effects of a firm-level
financial shock that may be propagated by job loss, among other channels. We argue
that it is not possible to infer the mental health effects of the financial constraints that
we study directly from the job displacement literature, most importantly because
the majority of employees in financially constrained firms do not lose their jobs,
yet they may suffer from workplace stress and increased job insecurity.
The rest of this article is organized as follows: Section II describes the data and
the institutional setting, and presents our empirical strategy. Section III documents
the baseline results and financial constraints’effects on employees’antidepressant
use. Section IV studies a transmission channel, increased job insecurity, and pre-
sents evidence that the increase in antidepressant use is not restricted to employees
losing their jobs. Section V presents robustness and placebo tests. Section VI
concludes.
II. Data, Institutional Setting, and Empirical Specification
We use administrative data from the Netherlands. Our data set combines
medicine use and employment data at the individual level with financial data at the
corporate (employer) level. All administrative data are provided by Statistics
Netherlands (SN), and separate databases are linked using unique (pseudony-
mized) identifiers at the individual or firm level. Appendix A provides details on
the databases used and Appendix B provides variable definitions.
A. Firm-Level Financial Data
Under the data framework of Statistics Netherlands, the definition of a firm is
hierarchical, whereby the enterprise group stands on top of the hierarchy and is
considered the center of financial decision-making. All corporate financial data are
provided at the (consolidated) enterprise group level. An enterprise group consists
of one or more business units, which are characterized by independent production
decisions and the ability to offer their products to external parties, and comprise
one or more legal entities (e.g., BVs –private limited liability companies) over
which the enterprise group has majority control. An enterprise group in our sample
consists of on average 5.5 business units, although 115 of the 352 enterprise
groups only have a single business unit. Hereafter we use firm and enterprise
group interchangeably.
Firm-level financial data is from the Annual Statistics of Finances of Large
Enterprises (SFLE, in Dutch: Statistiek Financiën van Grote Ondernemingen,
SGFO), which contains information on the consolidated balance sheets and income
Kárpáti and Renneboog 7
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

statements of the largest Dutch enterprise groups. In 2007, all enterprise groups
with at least EUR 23 million in total assets were surveyed, amounting to a sample of
1,204 firms. The scope of consolidation is the Netherlands; foreign subsidiaries
of Dutch internationals and Dutch subsidiaries of foreign internationals are not
consolidated. Financial data is presented by calendar year; only for a small share of
companies does the financial year not coincide with the calendar year.
6
In most of our regression specifications, we add (industry × year) fixed effects.
We use the first two digits of the 1993 version of the Dutch industry classification
codes (SBI), which aligns with the European NACE Rev.1 classification at the
4-digit level.
7
B. Employee-Level Labor Data
Information on employer–employee links is provided to SN by the Employee
Insurance Agency (EIA or, in Dutch, UWV), an administrative authority responsible
for implementing employee insurance and recording labor market data. From these
data, SN creates the databases BAANKENMERKENBUS, which records qualitative
job characteristics (e.g., the type of the job such as regular employment or internship,
and the start and end dates of an employment relation), and BAANSOMMENTAB,
which records quantitative job characteristics (such as salaries). We use the
information in these databases to link employees to firms, to construct our sample
(e.g., excluding interns), and to determine when employees separated from their
initial job (using the unique employment relationship identifier –baanid).
C. Individual-Level Antidepressant Use Data
The Netherlands has a universal health care insurance system where taking
out the basic health insurance is mandatory for all residents. Care consumers are
free to choose among multiple nationwide private health insurers who offer the
same regulated basic insurance package for an annual premium of approximately
1,000–1,200 EUR (subsidies are available for low-incomehouseholds). The package
covers general practitioner (GP) care, maternity care, hospital care, home nursing
care, pharmaceutical care, and mental healthcare, but does not cover for example
dentistry or physical therapy which may be covered by supplementary insurance
products. Care consumers must pay for their health consumption up to an annual
deductible (EUR 150 in 2008 and EUR 385 as of 2016; the deductible can be
voluntarily increased to lower the insurance premium). Certain care products such
as GP care and maternity care do not count toward the deductible.
6
We do not observe firms’financial years, but as per Bureau van Dijk Orbis data only 9% of the
900 firms in the Netherlands that meet our broadest sample selection criteria (Dutch-owned with at least
EUR 23 million in assets) had their 2010 financial year ending not on Dec. 31. The financial year is
labeled as the calendar year with which it has the most overlap (e.g., data of a financial year ending on
Sept. 30, 2019 are labeled as 2019 data).
7
We use the codes provided in the General Business Register (in Dutch: Algemeen Bedrijven
Register or ABR). The industry classification codes are registered at the Chamber of Commerce for
each legal unit. In the General Business Register, SN provides a code at the business unit level by using
the code of the legal unit within the business unit that has the most employees. Similar to this approach,
we use the code of the business unit with the most employees within an enterprise group as the enterprise
group-level code.
8 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

The initial point of contact for most medical complaints is the general prac-
titioner. All residents are registered with a local GP of their choice. GPs play a
gatekeeper role; their referral is necessary for (nonurgent) hospital and specialist
care. This holds for mental health problems as well; patients first approach their
GP (or in rare cases a so-called first-line psychologist), who may refer them to the
second-line specialist mental care in case of any serious problem. GPs frequently
employ antidepressant medication as the first line of treatment for mental health
complaints; in 2010, 30% of adult patients with any psychological diagnoses were
prescribed antidepressants (Nuijen et al. (2012)). Unjustified antidepressant use is
reported to be low (Piek et al. (2011)).
The individual-level medicine use database (MEDICIJNTAB) comprises
annual binary indicators for the use of medicines that are reimbursed under the
Dutch basic health insurance scheme. The indicators are grouped at the 4-digit ATC
(Anatomical Therapeutic Chemical) level. Therefore, we observe if a person was
reimbursed (any positive amount of) antidepressants (ATC-code N06A) in a par-
ticular year, but we do not observe the exact chemical substance (e.g., paroxetine,
N06AB05) nor do we observe the exact amounts (e.g., defined daily doses, DDDs).
As antidepressants are only available on prescription, and all antidepressants are
reimbursed under the basic health insurance,
8
the database gives a complete picture
of antidepressant use.
D. Other Databases
We use two additional databases provided by SN in the selection of our
employee sample. First, we determine employees’age and gender using the Munic-
ipal Personal Records Database (or in Dutch: Gemeentelijke Basis Administratie or
GBA). Second, we collect information on the position of each person in their
household from the Income of People (in Dutch; Integraal Persoonlijk Inkomen
or IPI) database.
In order to illustrate the strong (cross-sectional) correlation between worrying
about job loss and antidepressant use, we also use answers to the question “Are you
concerned of keeping your job?”from the National Labor Conditions Survey
(in Dutch: Nationale Enquête Arbeidsomstandigheden, NEA), an annual survey
on working conditions, accidents at work, work content, and industrial relations.
E. Attrition
The administrative databases on medicine use and employment do not
suffer from the attrition problems that surveys usually face (e.g., nonresponse).
Yet, attrition might occur if someone leaves the Dutch population, for instance, due
to emigration or death. Using the Wealth of Households (VEHTAB) data set, which
8
Reimbursements for medicines count tow ard the compulsory annual deductible. Some medicines
are only partially reimbursed and a personal contribution must be paid. We could also have studied
the use of anxiolytic drugs (N05B, such as benzodiazepines including alprazolam/Xanax); however,
starting from 2009 these drugs are only reimbursed in rarecases and are consequently missing from the
MEDICIJNTAB database. Furthermore, due to the side effects of benzodiazepines, Dutch guidelines
on the pharmacological treatment of anxiety disorders recommend the use of antidepressants
(Sonnenberg, Biennali, Deeg, Comijs, Van Tilburg, and Beekman (2012)).
Kárpáti and Renneboog 9
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

lists all households and household members that belong to the Dutch population
on Jan. 1 of each year, we find that attrition is similar for treated and control
employees.
9
Attrition affects the definition of our main outcome variable (antide-
pressant use): we assign a missing value to person-year observations where the
given person was missing from VEHTAB (we do this because MEDICIJNTAB
does not cover the antidepressant use of people who are not part of the Dutch
population).
F. Sample Composition
The starting point of our sample selection is the 1,204 firms (enterprise groups)
in the 2007 annual SFLE (Statistics of Finances of Large Enterprises). Because
repayment obligations of local subsidiaries may have limited financial consequences
as corporate groups can meet these obligations through their internal capital markets
(e.g., Desai, Foley, and Hines (2004)), we exclude Dutch subsidiaries of foreign
internationals. We identify these subsidiaries as firms with more than 50% of the share
capital owned by foreign-based companies or with an Ultimate Controlling Institu-
tional Unit (UCI) located outside the Netherlands.
Following Almeida et al. (2011), we also exclude firms with a low long-term
debt (excluding the current portion of long-term debt) to total assets ratio at the year-
end of 2007. This is because our treatment classification aims to contrast firms with
comparable debt profiles, for which long-term debt financing is a permanently
important source of funds. In the baseline specification, we only consider firms
with at least 10% of long-term debt to total assets. In robustness tests, we will vary
this cut-off.
Finally, we exclude firms that operate in government-controlled and heavily
regulated industries. These are government management (SBI code 7511), public
transport via railway (6010), national post with universal service obligation (6411),
and utilities (40–41). We further exclude outsourcing firms (74,501, 74,502) because
we cannot observe the actual company where outsourced employees work. The
resulting sample consists of 353 firms (enterprise groups).
We identify the business units of these enterprise groups on Jan. 1, 2008 using
the General Business Register (ABR). Subsequently, for the same date, we identify
the people employed by each business unit. We restrict our sample to employees
with a regular or on-call job contract
10
aged between 20 and 60 years in 2008,
11
and
who are the household head of their household or the partner of the household
head.
12
The final sample consists of 328,229 employees. The steps of sample
composition are presented in Table 1.
9
Untabulated; by 2016 treated employees are 0.16% more likely to be in the Dutch population
(t-stat. = 0.74).
10
On-call employees only work when the employer calls them up, as they do not have fixed working
hours. We exclude employees classified as interns and outsourced workers. We also exclude director-
major shareholders, who are people with a considerable ownership in the firm they manage.
11
In the sample period, early retirement was widespread in the Netherlands with 80% of employees
retiring before reaching the state pension age of 65, mostly at the age of 60. As labor market shocks have
arguably limited impact on employees close to retirement, we exclude them.
12
SN classifies people as either household head (person with the highest socioeconomic position),
partner (married or unmarried) of the household head, children of the household head, or other/unknown
10 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

G. Treatment Classification and Summary Statistics
We classify firms as “treated”or “control”based on the share of long-term
debt that they were required to repay in 2008. Unlike most databases comprising
European firms’financial data (e.g., Bureau van Dijk’s Orbis), which report the
current portion of long-term debt aggregated with all other current liabilities, the
SFLE database reports these itemsseparately. This is important because other current
liabilities, such as short-term bank loans, may be correlated with the business outlook
that the company faced preceding the crisis, and may thus fail tobe exogenous to the
outcomes we study. We calculate our “forcing”variable as
SHARE_OF_CURRENT_PORTION_OF_LT_DEBT
=CURRENT_PORTION_OF_LONG_TERM_DEBT
CURRENT_PORTION_OF_LONG_TERM_DEBT+TOTAL_LONG_TERM_DEBT ,
where TOTAL_LONG_TERM_DEBT is the part of long-term debtmaturing beyond
1year.
13
In our baseline specification, we classify firms as treated if the SHARE_OF_
CURRENT_PORTION_OF_LT_DEBTratio is at least 25%. Thisresultsin23 treated
and 329 control firms. In robustness tests, we will vary this cut-off point.
Table 2 presents summary statistics for treated and control firms (Panel A)
and for their employees included in our sample (Panel B). The last three columns
show a comparison between treated and control; the column “Raw Δ”presents the
TABLE 1
Steps of Sample Composition
Table 1 presents the steps taken to arrive at the final sample of firms and employees.
No. of Enterprise
Groups
No. of Business
Units
No. of
Employees
Total in 2007 SFLE 1,204
Excluding foreign-owned firms 609
Excluding firms with <10% LT debt on total assets 378
Merging with business units 378 3,018
Merging with employees 376 2,106 801,297
Excluding government-controlled and regulated industries 353 1,936 464,447
Restricting to age 20–60 years 352 1,917 388,539
Restricting to household head and partner 352 1,914 331,899
Excluding interns, outsourced employees, and director-
major shareholders
352 1,899 328,229
(e.g., children of the partner from a previous marriage). We only include people of the first two
categories, excluding children and other/unknown household members because we aim to limit our
sample to people for whom an employment shock (or a threat thereof ) has high stakes.
13
The SFLE database differentiates between five categories of long-term debt, i) Debt to group
companies, ii) Subordinated loans, iii) Bonds outstanding, iv) Loans from domestic financial institu-
tions, and v) Other long-term debt (a residual category that includes, inter alia, loans from private
individuals, derivatives, and lease obligations). Ideally, we would only consider bonds outstanding and
bank loans (and the current portion thereof ) because these financing forms are the hardest to renegotiate.
However, the SFLE reports the current portion of all five debt categories combined. Although the scope
of this problem is limited as bank loans constitute by far the largest share of long-term debt for most
sample companies, in Table 8, we also execute a robustness test where we exclude firms with any long-
term intercompany debt on their 2007 opening balance sheet (the type of long-term debt presumably the
least binding).
Kárpáti and Renneboog 11
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

TABLE 2
Pre-Treatment Summary Statistics
Table 2 reports pre-treatment descriptive statistics for the treated and control firms, and their employees. The column Raw Δpresents the difference in means. The column Adj. Δpresents the difference in means
estimated in a regression where we control for 2-digit SBI 1993 industry fixed effects, and in case of Panel B also for the four financial variables included in our main specification (liquid assets to TA, LT debt to TA, cash
flow, and log TA). The column t-stat. presents the tstatistic on the regression coefficient reported in column Adj. Δ. No 10th and 90th percentiles are reported for treated firms, following Statistics Netherlands guidelines,
because these values would refer to fewer than 10 companies. Variable definitions are presented in Appendix B.
Treated Control Raw ΔAdj. Δt-Stat.
Panel A. Firm Characteristics N Mean Std. Dev. p10 p50 p90 NMean Std. Dev. p10 p50 p90
LIQUID_ASSETS_TO_TA, 2007 23 0.03 0.04 . 0.00 . 329 0.05 0.08 0.00 0.01 0.12 0.02 0.01 0.93
LT_DEBT_TO_TA, 2007 23 0.43 0.19 . 0.43 . 329 0.32 0.19 0.14 0.27 0.59 0.11 0.07 1.55
CASH_FLOW, 2007 23 0.17 0.13 . 0.14 . 329 0.11 0.09 0.03 0.10 0.20 0.06 0.06 2.12
TOTAL_ASSETS, 2007 (EURm) 23 489 1,753 . 82 . 329 663 2,457 33 92 1,139 174 51 0.11
SHARE_OF_CURRENT_PORTION_OF_LT_DEBT 23 0.34 0.09 . 0.30 . 329 0.06 0.06 0.00 0.03 0.15 0.28 0.27 13.41
No. of employees in sample 23 1,552 5,554 . 242 . 329 889 1,920 70 285 1,993 663 909 0.74
Industry composition:
Wholesale and retail trade 12 88
Other 11 233
Panel B. Employee Characteristics
ANTIDEPRESSANT_USE, 2007 (%) 35,692 4.75 21.26 0.00 0.00 0.00 292,537 3.99 19.57 0.00 0.00 0.00 0.76 0.00 0.01
TENURE_IN_YEARS, 2008 35,692 9.26 8.80 0.00 7.00 22.00 292,537 8.79 8.85 0.00 6.00 21.00 0.47 0.99 1.91
AGE, 2008 35,692 39.60 10.85 24.00 40.00 55.00 292,537 42.10 9.98 28.00 42.00 56.00 2.50 1.46 2.50
FEMALE 35,692 0.50 0.50 0.00 0.00 1.00 292,537 0.32 0.47 0.00 0.00 1.00 0.18 0.07 1.50
12 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

difference in means, the column “Adjusted Δ”presents the difference in means
adjusted in a regression setting for industry fixed effects (and in Panel B also for the
firm financial controls that we include in our baseline regressions: log total assets,
liquid assets to total assets, long-term debt to total assets, and cash flow), and the last
column presents the significance of the regression coefficient “Adjusted Δ”by
means of a t-statistic.
As Panel A of Table 2 shows, treated firms (at the end of 2007) were slightly
smaller in terms of total assets but larger in terms of number of employees (column
Raw Δ). Treated firms had a somewhat lower cash ratio (liquid assets to total assets)
and more long-term debt outstanding relative to their total assets, but they exhibited
a more positive cash flow in 2007. As the “Adjusted Δ”and “t-stat.”columns
reveal, once controlling for industry composition, only this latter difference (and
the difference in our forcing variable, share of current portion) is statistically
significant at the 5% level. We control for any (remaining) difference in these
variables in our regression models.
Turning to employee characteristics in Panel B of Table 2, ANTIDEPRESSANT_
USE, our main outcome variable, is an annual binary indicator that takes the value
1 if a person was reimbursed for antidepressant medications in the given year
(we multiply the indicator by 100 hence our results are in %). 4.1% of our sample
used antidepressants in 2007, comprising 4.75% of treated employees and 3.99% of
control employees. This difference practically disappears in the regression setting in
column “Adjusted Δ.”While a similar pre-treatment level of the dependent variable
is not required for identification in a difference-in-differences setting, it is reassur-
ing that the included control variables adequately explain any differences in treated
and control employees’antidepressant use in 2007. Regarding other employee
characteristics, treated and control employees have similar tenure, although treated
employees are slightly younger and more likely to be female. Once controlling for
industry fixed effects and the financial control variables, in column “Adjusted Δ,”
these differences in age and gender diminish and even reverse. In a robustness test,
we will also control for year effects interacted with these employee characteristics to
account for any time trends that might depend on these characteristics (e.g., older
employees might be more affected by the crisis).
While Table 2 presents the pre-treatment characteristics of treated and control
employees, we present the mean of our binary outcome variables (antidepressant
use, job separation) at the bottom of the respective regression tables.
H. Institutional Detail on Corporate Information Flows to Employees
The validity of our research question hinges on employees who do not belong
to the management being able to gain knowledge of the firm’s (financial) situation
and strategy. If the average employee is oblivious of the firm’s financial health,
corporate financial (di)stress cannot induce mental stress.
Although the Netherlands does not have a “Mitbestimmung”(codetermina-
tion) corporate governance system as is the case in Germany (where half the
supervisory board seats are reserved for union and employee representatives and
employees can hence weigh on corporate decision-making), employees in the
Netherlands have strong rights to acquire corporate information and to be consulted
Kárpáti and Renneboog 13
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

on important corporate decisions. Furthermore, employees are also asked for
approval in case of social corporate policies. The body to which corporate law
grants these rights is the “Ondernemingsraad”(OR), the Works Council. Every firm
with at least 50 employees is legally obliged to have an OR, and the employee
representatives at the OR are elected by the employees. The number of OR members
depends on firm size (for a small firm with 50 employees, the OR counts 5 mem-
bers). The OR has several important rights: i) Information rights: The law requires
the management to provide the OR information on a range of financial/economic
issues. On an ongoing basis, the management must provide information on the
activities and financial results of the business, and on future prospects. The man-
agement is also obliged to give the OR copies of the annual report and accounts,
including consolidated accounts of the group, and, where this is relevant, details of
the specific results for the part of the business the works council covers (this would
the case if the annual report only gives consolidated information). ii) Consultation
rights: Corporate law states that consultations with and representation of the
employees are in the interests of the sound functioning of the enterprise in all its
objectives. iii) Social rights: The management is legally required to ask for
the approval of the OR when corporate decisions have social consequences. This
would, for example, be the case when the corporation needs to be restructured when
changes in working conditions are to be introduced (e.g., more/less overtime,
introduction of more labor flexibility, changes to a shift system).
The OR receives new information timely as the OR members meets once a
month and every 2 months the OR meets the management (in about 20% of firms,
the OR-management meets more frequently). At least twice a year, the OR and the
management join for consultation meetings whereby the general strategy of the
enterprise is to be discussed. More specifically, this includes decisions to attract
large loans; to set up, take over, or sell other organizations; or to terminate some
corporate activities. The information is dissipated to the employees through various
information channels: for example, every OR publishes a newsletter distributed to
all employees that includes the agenda and the minutes of the OR meetings.
I. Empirical Specification
Our empirical specification compares the time-trend of antidepressant use
of employees of high-repayment firms (treated) and other sample employees
(controls), accounting for employee fixed effects, industry-specific year effects,
and year effects that depend on pre-crisis firm (or employee) characteristics. We
estimate a linear probability model:
ANTIDEPRESSANT_USEi,f,j,t=αi+Tfβt+γj,t+x0
f,2007λt+ϵi,f,j,t,(1)
where ANTIDEPRESSENT_USEi,f,j,tis a binary indicator variable capturing
whether individual iwho worked on Jan. 1, 2008 for firm fbelonging to industry
jwas reimbursed for any antidepressant use in year t,αiare employee fixed effects,
Tfis the treatment indicator that takes the value 1 for treated firms and 0 for control
firms, γj,tare (year × industry) fixed effects, and xf,2007 is a 4-by-1 column vector of
firm f’s 2007 financial characteristics, comprising log total assets, liquid assets to
14 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

total assets, long-term debt to total assets, and a measure of cash flow [=(net income
plus depreciation and amortization)/total assets].
14
These financial characteristics
are derived from Almeida et al. (2011), who argue that industry fixed effects and
these financial characteristics
15
capture a lot of otherwise unobserved firm hetero-
geneity that is important both for the treatment classification (i.e., maturity structure
of long-term debt) and firms’business conditions prior to and during the crisis. βt
are the differential year effects for the treated firms, our main coefficients of interest.
We estimate model (1) using data from 2006 to 2013, equivalent to 2 years prior to
and 5 years following the 2008 financial crisis. Due to the presence of individual
fixed effects, we normalize β2007 to 0. We cluster standard errors at the firm (enter-
prise group) level because the treatment variation is at the firm level.
We quantify the average treatment effect over the period of 2008 to 2012
16
using a difference-in-differences model:
ANTIDEPRESSANT_USEi,f,j,t=αi+TfPOSTβ+x0
f,2007 POSTλ+ POSTγj+ϵi,f,j,t,(2)
where POST is an indicator for the post-treatment period (2008–2012), γjare
industry fixed effects, xf,2007 is a 4-by-1 column vector of the same 2007 firm
financial characteristics as in model (1), and consequently λis a 4-by-1 column
vector of coefficients. The included periods are 2006 to 2012.
In order to study treatment heterogeneity, we also use a version of model (2)
where we interact the treatment indicator with pre-treatment employee characteristics:
ANTIDEPRESSANT_USEi,f,j,t=αi+z0
i,2007 TfPOSTβ+Tf× POSTδ1
+z0
i,2007 ×POSTδ2+POSTγj
+x0
f,2007 POSTλ+ϵi,f,j,t,
(3)
where zi,2007 is an n-by-1 column vector of 2007 employee characteristics such as
age, gender, or having a partner, and βis an n-by-1 column vector of coefficients.
III. The Effect of Financial Constraints on Employee
Mental Health
Figure 3 presents the estimated treatment effects (βt) from model (1).
Employees ofthe treated firms, relative to employees of the control firms, experience
an increase in antidepressant use starting from 2008; the treatment effect reaches its
peak in 2011 and it is not statistically significantly different from 0 in 2013 anymore.
The relatively fast increase in antidepressant use in 2008 may reflect increased
job insecurity (as discussed in Section IV), and is in line with the findings of
14
In a robustness test (column 4 of Table 4), we also include in xa set of 2007 employee charac-
teristics to account for the differences between treated and control employees reported in Panel B of
Table 2.
15
Almeida et al. (2011) further control for Tobin’s Q and credit ratings, variables unavailable in our
data set.
16
We opt to quantify the average treatment effect for the 2008–2012 period as the estimation of
model (1) will reveal that treatment effects in 2013 are not statistically significant anymore.
Kárpáti and Renneboog 15
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

Schaller and Stevens (2015) who document that displaced US workers exhibit
depression or anxiety within months after the loss of their jobs. The immediacy
of the treatment effect is further supported by results from the psychology literature.
Kendler, Karkowski, and Prescott (1999) study 15 different stressful life events
and find that 11 of them, including job loss, and financial or housing problems, are
significantly associated with the onset of major depression in the month of occur-
rence. The onset of depression may have a swift effect on antidepressant use due
to the prescription preferences of Dutch general practitioners: Van Marwijk, Bijl,
Adèr, and De Haan (2001) report that in 1998 Dutch GPs prescribed antidepressants
in 73% of the first consults for depressive symptoms.
The persistence of the treatment effect can be partially explained by the
persistence of depression. Depression (medical term: major depressive disorder)
is a lifelong illness that is categorized by recurrent depressive episodes. The
majority of patients recover (i.e., are no longer symptomatic) within 12 months
following a depressive episode; however, long-term recovery (lack of recurrence) is
low, approximately 30% at a 6-year horizon, and almost 80% of patients experience
at least one further episode in their lifetime. Furthermore, a large proportion (up to
27%) of patients never recover and develop chronic depressive illness (Malhi
and Mann (2018)). The long-lasting nature of depression is also supported in our
medicine use data, 57% of people in our sample who used antidepressants in 2006
continued to do so in 2012.
We also estimate treatment effects for 2006, to investigate “parallel trends”
before the treatment. Ideally, we would present trends for multiple pre-treatment
periods, but the medicine use database is only available from 2006. As Figure 3
illustrates, treated and control employees demonstrated a similar change in antide-
pressant use between 2006 and 2007, conditional on the control variables.
The coefficient estimates from Figure 3 and the corresponding standard errors
are presented in Table 3. The table also shows the estimates of the control variables
FIGURE 3
Treatment Effects on Antidepressant Use
Figure 3 shows the estimated treatment effects on antidepressant use (in percentage points) and 95% confidence interval.
The corresponding coefficient estimates and t-statistics are also presented in Table 3.
–0.2
0
0.2
0.4
0.6
0.8
1
1.2
2006 2007 2008 2009 2010 2011 2012 2013
16 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

(interacted with the year indicators). Employees of firms with higher long-term debt
to total assets ratio in 2007 exhibited a greater increase in antidepressant use during
the crisis, whereas employees of larger firms exhibited a smaller increase. Both the
2007 cash ratio (liquid assets to total assets) and cash flow appear to diminish
growth in antidepressant use, although these estimates are mostly not statistically
significant.
Table 4 presents the average 2008–2012 treatment effects from model (2),
a difference-in-differences model. All specifications control for employee fixed
effects. The baseline specification in column 1 further controls for (industry × year)
fixed effects and year fixed effects that depend linearly on 2007 firm financial
characteristics. Columns 2–4 present variations on these additional controls. Col-
umn 2 drops the (2007 firm financials × year) fixed effects, while column 3 defines
industries at a coarser (sectoral) level instead of using 2-digit Dutch SBI93 industry
codes. Finally, because the descriptive statistics in Table 2 show that treated and
control employees exhibited some pre-treatment differences in age, salary, and
tenure, column 4 includes the interaction of these characteristics with year
dummies. All four specifications show qualitatively similar results, with 2008–
2012 treatment estimates ranging between 0.26 pp and 0.47 pp (a 5.4% to 9.7%
effect relative to the 4.8% baseline probability of antidepressant use).
To better understand the magnitude of these treatment effects, we can compare
them to estimates from the literature on the mental health effects of wealth and
employment shocks. Our main result of a 5.4% to 9.4% relative increase in the
TABLE 3
Effects on Antidepressant Use Over Time
Table 3 shows estimates of the effect of a firm having to repay at least 25% of long-term debt in 2008 (Treated) on employees’
antidepressant use, based on model (1). All columns belong to a single regression; each column shows coefficient estimates
on the year × Treated and year × Controls interactions for the given year. 2007 is the omitted year. As specified in model (1), the
regression includes employee fixed effects as well as 2-digit SBI93 industry × year fixed effects. Antidepressant use is
originally a binary variable that takes the value 1 if a person was reimbursed for (any) antidepressant use in the given year; we
multiply this variable by 100 and therefore all coefficients in the table are expressed in %. Antidepressant use is only defined
for people who lived in the Netherlands on Jan. 1 of the given year. The row “unconditional mean”presents the sample mean of
the dependent variable for the given year. Variable definitions are presented in Appendix B. The t-statistics, reported in
parentheses, are based on standard errors clustered at the firm (i.e., enterprise group) level. *, **, and *** denote significance
at the 10%, 5%, and 1% levels, respectively.
Dependent Variable: ANTIDEPRESSANT_USE (Binary, ×100)
2006 2008 2009 2010 2011 2012 2013
TREATED 0.0849 0.412*** 0.445*** 0.486*** 0.667*** 0.407** 0.295
(0.88) (3.47) (3.39) (2.81) (3.52) (2.25) (1.47)
LIQUID_ASSETS_TO_TA, 2007 0.501 0.317 0.505 0.156 0.892 1.028 0.751
(1.34) (0.97) (1.25) (0.28) (1.31) (1.34) (1.10)
LT_DEBT_TO_TA, 2007 0.0199 0.301 0.731*** 0.745*** 0.858*** 0.606** 0.625**
(0.12) (1.53) (3.17) (3.23) (3.55) (2.20) (2.27)
log(TOTAL_ASSETS), 2007 0.00556 0.0236 0.0205 0.0710*** 0.0780*** 0.0766*** 0.120***
(0.32) (1.25) (0.91) (2.72) (2.71) (2.74) (3.87)
CF, 2007 0.180 0.436 0.723 0.661 0.566 0.397 0.884
(0.45) (1.15) (1.42) (0.86) (0.64) (0.44) (0.99)
Unconditional mean (%) 3.72 4.31 4.58 4.83 5.14 5.33 5.60
Employee FE Yes
Industry × year FE Yes
No. of firms (=clusters) 352
No. of obs. 2,603,121
Kárpáti and Renneboog 17
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

probability of antidepressant use due to firm-level (re)financing difficulties is
similar to the effect of the 2006–2009 US housing price shock (7.51% rise in
antidepressant prescription volume) reported by Lin, Ketcham, Rosenquist, and
Simon (2013), but smaller than the effect of job loss (22% increase in the probability
of depression/anxiety) calculated by Schaller and Stevens (2015) or the effect of
losing on average USD 220,000 during the Oct. 2008 market crash (35% relative
increase in the probability of antidepressant use) estimated by McInerney, Mellor,
and Nicholas (2013). This benchmarking exercise shows that firm-level (re)financ-
ing difficulties had a serious impact on employee mental health, although not
directly comparable to the effects of job loss.
IV. The Transmission Channel of Job Insecurity
A. Increased Job Insecurity of Treated Employees
How can firm-level refinancing difficulties lead to an increase in employees’
antidepressant use? The estimated 0.44 pp increase in antidepressant use is a
TABLE 4
Average Treatment Effect for 2008–2012
Table 4 shows mean 2008–2012 treatment effect estimates of a firm having to repay at least 25% of long-term debt in 2008
(Treated) on employees’antidepressant use, based on model (2). Antidepressant use is originally a binary variable that takes
the value 1 if a person was reimbursed for (any) antidepressant use in the given year; we multiply this variable by 100 and
therefore all coefficients in the table are expressed in %. The interaction of the Treated treatment indicator and the control
variables with the Post indicator (which takes the value 0 in 2006–2007 and the value 1 in 2008–2012) are tabulated. All models
also control for employee fixed effects and 2-digit SBI93 industry × Post fixed effects. Column 1 presents the baseline
specification . Column 2 does not inc lude the Post × firm- level control var iables. Column 3 use s a coarser, sector al-level,
industry classi fication. Final ly, compared to col umn 1, column 4 also inc lude pre-treatm ent employee char acteristics
interacted with t he Post indicator. Th e row “unconditio nal mean”presents the sample mea n of the dependent var iable.
Variable definitions are presented in Appendix B.Thet-sta tistics, repor ted in parentheses , are based on stand ard errors
clustered at the f irm (i.e., enter prise group) le vel. *, **, and ** * denote signific ance at the 10%, 5 %, and 1% levels,
respectively.
ANTIDEPRESSANT_USE (×100)
POST × …Baseline No Covariates Coarser Industry Employee Covariates
TREATED 0.440*** 0.260*** 0.333*** 0.473***
(3.09) (3.54) (2.91) (3.35)
LIQUID_ASSETS_TO_TA, 2007 0.261 0.141 0.326
(0.56) (0.33) (0.72)
LT_DEBT_TO_TA, 2007 0.637*** 0.627*** 0.618***
(3.48) (4.12) (3.40)
log(TOTAL_ASSETS), 2007 0.0565*** 0.0538*** 0.0486**
(2.66) (2.85) (2.41)
CF, 2007 0.470 0.0912 0.473
(0.67) (0.17) (0.72)
AGE, 2008 0.0014
(0.50)
FEMALE 0.288***
(3.81)
TENURE, 2007 (years) 0.011***
(3.41)
Unconditional mean (%) 4.84 4.84 4.84 4.84
Employee FE Yes Yes Yes Yes
Industry × post FE SBI93 SBI93 Sections SBI93
No. of firms (clusters) 352 352 352 352
No. of obs. 2,282,057 2,282,057 2,282,057 2,282,057
18 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

weighted average treatment effect on employees who left their job during the
sample period (leavers) and on those who stayed in their jobs (stayers). We argue
that an important transmission channel from refinancing difficulties to employee
mental health is job loss for leavers and decreased job security for stayers.
Previous work demonstrated that financial constraints can negatively affect
firms’labor demand. Chodorow-Reich (2014) finds evidence, among 2,000 non-
financial US firms, that companies reduced employment more in the period of 2008
to 2009 if they had pre-crisis relationships with banks that were in a less healthy
condition during the financial crisis. Huber (2018) shows that German firms fully
dependent on Commerzbank, a bank severely affected by the 2008 financial crisis,
reduced their employment on average by 5.3% between 2009 and 2012 compared
to firms with no Commerzbank relationship. Giroud and Mueller (2017) report
that employment in more highly levered US firms was more sensitive to declines in
local consumer demand during the Great Recession. Giroud and Mueller argue
that financing constraints may dampen labor demand by impairing firms’ability to
engage in labor hoarding, a practice of retaining temporarily unnecessary employees
to preserve firm-specific human capital and to avoid firing/re-hiring costs.
During the first crisis years, labor hoarding was widespread in the Netherlands;
indeed, several studies credit to this phenomenon the relatively mild increase in
Dutch unemployment rates between 2008 and 2010 (e.g., Van den Berge, Erken,
de Graaf-Zijl, and van Loon (2014)). Nonetheless, as highlighted by Giroud and
Mueller (2017), labor hoarding requires financial resources, which are scarcer for
financially constrained firms. In the Netherlands, financial resources are particu-
larly important for labor hoarding due to the inflexible employment terms regarding
both working hours and wages. Over 80% of Dutch employees are covered by
collective labor agreements (CLAs), which largely prevent companies from adjust-
ing nominal wages downward. Adjustments in the number of working hours are
also not straightforward to implement because Dutch CLAs, unlike for instance
German ones, do not contain provisions for temporary shorter working hours
(Tijdens, van Klaveren, Bispinck, Dribbusch, and Öz (2014)).
Given these observations, we hypothesize that firms that had to repay a
larger share of their long-term debt in 2008 had relatively fewer resources to
engage in labor hoarding, and consequently, employees of these firms suffered
from decreased job security. The adverse mental health effects of job loss are well-
documented (Browning and Heinesen (2012), Ganster and Rosen (2013), and
Schaller and Stevens (2015)), which could explain treatment effects on leavers.
However, decreased job security can damage employee mental health even in the
absence of actual job loss (Witte (1999), Burgard et al. (2009), Reichert and
Tauchmann (2011), and Kim and Von Dem Knesebeck (2015)).
17
Green (2011),
for instance, concludes that for an employee of average employability the mental
health effect of extreme job insecurity is similar to the effect of unemployment.
Therefore, decreased job security could also explain treatment effects on stayers.
17
We also illustrate this negative association between job insecurity and mental health in the
Netherlands, using the National Working Conditions Survey. Although we cannot establish causality,
Table 1 of the Supplementary Material shows that employees answering “yes”to the question “Are
you concerned of keeping your job?”are substantially (~2 pp or 44% relative to the 4.5% baseline)
more likely to use antidepressants, even after controlling for a broad range of personal and firm
characteristics.
Kárpáti and Renneboog 19
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

In line with this hypothesis, we find evidence that employees in treated firms
were more likely to separate from their job during the crisis period. Panel A of
Table 5 presents the treatment effects on the (cumulative) probability of job sepa-
ration. We assume that an employment relationship ended in a given year (“job
separation”) if the job is not registered anymore in the “Quantitative characteristics
of employment relationships”database (BAANSOMMENTAB) in the following
year. We only consider the initial employment relationships that existed on Jan. 1,
2008. The results in columns 1–3 show that treated employees had a 4.6, 6.7, and 6.2 pp
higher probability of job separation by the end of 2008, 2009, and 2010, respectively.
This is an economically significant increase compared to the unconditional means
TABLE 5
Treatment Effects on Cumulative Job Separation
Table 5 reports treatment effects on cumulative job separation (Panel A), UWV dismissal (Panel B), and cumulative job
separation with a gap (Panel C). CUMULATIVE_JOB_SEPARATION is a binary indicator that takes a value of 1 if an
employee’s initial ( Jan. 1, 2008) job ended, for any reason, by the end of the year in consideration (the year indicated in
the header of the column). CUMULATIVE_UWV_DISMISSAL is a binary indicator that takes the value 1 if an employee’s initial
(Jan. 1, 2008) job ended with a dismissal permit from the Dutch Employee Insurance Agency (UWV), by the end of the year
in consideration. CUMULATIVE_JOB_SEPARATION_WITH_A_GAP is also a binary indicator that takes the value 1 if an
employee’s initial (Jan. 1, 2008) job ended, for any reason, by the end of the year in consideration and there was any gap
(>1 day) between the end date of that job and the beginning date of any following job. We multiply all dependent variables by
100 and therefore all coefficients in the table are expressed in %. In all panels, columns 1–3 control for 2007 firm financial
characteristics (including industry fixed effects), whereas columns 4–6 further control for 2007 employee characteristics (age,
gender, tenure). As on Jan. 1, 2008 all employees –by definition –worked at their initial job, the regressions do not include
employee fixed effects. In Panel A, the row “unconditional mean”presents the sample mean of the dependent variable (in %).
The t-statistics, reported in parentheses, are based on standard errors clustered at the firm level. *, **, and *** denote
significance at the 10%, 5%, and 1% levels, respectively.
Firm Controls Firm and Employee Controls
2008 2009 2010 2008 2009 2010
Panel A. CUMULATIVE_JOB_SEPARATION (Binary, ×100)
TREATED 4.60*** 6.66** 6.20** 5.60*** 7.93*** 7.47**
(2.97) (2.29) (2.01) (3.95) (2.92) (2.58)
LIQUID_ASSETS_TO_TA, 2007 4.50 9.26 14.0* 4.50 8.74 13.24*
(0.84) (1.22) (1.69) (1.01) (1.36) (1.85)
LT_DEBT_TO_TA, 2007 3.41 8.64* 8.29 3.38 8.57** 8.25*
(1.16) (1.90) (1.64) (1.31) (2.10) (1.80)
log(TOTAL_ASSETS), 2007 0.529* 0.991** 1.24** 0.150 0.524 0.769*
(1.90) (2.10) (2.52) (0.61) (1.19) (1.65)
CF, 2007 3.53 9.31 11.4 7.86 13.29 14.33
(0.63) (0.93) (1.02) (1.65) (1.49) (1.43)
AGE, 2008 0.356*** 0.344*** 0.270***
(6.51) (4.38) (2.94)
FEMALE 0.091 1.443* 1.974**
(0.18) (1.77) (2.01)
TENURE, 2007 (years) 0.489*** 0.681*** 0.755***
(7.37) (7.55) (6.93)
Unconditional mean (%) 14 23 30 14 23 30
Panel B. CUMULATIVE_UWV_DISMISSAL (Binary, ×100)
TREATED 0.224 0.667** 1.25** 0.220 0.662** 1.24**
(1.34) (2.15) (2.42) (1.31) (2.12) (2.37)
Panel C. CUMULATIVE_JOB_SEPARATION_WITH_A_GAP (Binary, ×100)
TREATED 3.07*** 5.03** 4.83** 3.53*** 5.53*** 5.17**
(3.32) (2.50) (2.19) (3.95) (2.85) (2.39)
Industry FE Yes Yes Yes Yes Yes Yes
Age, gender, tenure No No No Yes Yes Yes
No. of firms (clusters) 352 352 352 352 352 352
No. of obs. 328,229 328,229 328,229 328,229 328,229 328,229
20 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

of job separation (e.g., 30% by end-2010). Job separation is statistically signifi-
cantly lower in larger firms, whereas it appears to be larger in firms with higher
end-2007 long-term debt to total assets ratio and cash ratio. In columns 4–6, we
further control for the same pre-treatment employee characteristics that we also
controlled for in column 4 of Table 4. After adding these additional controls, the
treatment effects increase slightly with treated employees facing a 7.2 pp greater
probability of having separated from their job by the end of 2010. Age and tenure
statistically significantly decrease the probability of job separation.
Job separation, the dependent variable in Panel A of Table 5, does not differ-
entiate between voluntary and involuntary departures. Arguably, involuntary job
separation would be a stronger indicator of decreased job insecurity, although
in practice employees might also quit “voluntarily”when they face a poor work
environment. In Panels B and C, we perform two additional analyses that suggest
that the increased rate of job separation among employees of treated firms is at least
partly due to dismissals. In Panel B, we study the treatment effects of dismissal for
economic reasons with the permit granted to an employer by the Employee Insur-
ance Agency (UWV). In the Netherlands, employers have multiple legal possibil-
ities to dismiss employees for economic reasons, the three main ways being: i) by
mutual consent, ii) with a dismissal permit from the UWV, and iii) with the permit of
the subdistrict court. Based on the study of Hoevenagel and Engelen (2013)on
dismissal routes in the Netherlands, we estimate that dismissals via a UWV permit
capture around one-half of all dismissals for economic reasons. The estimates in
Panel B show that employees in treated firms faced an increased probability of
being dismissed for economic reasons with an UWV permit in the period of 2008 to
2010. As UWV dismissals only represent a part of all dismissals for economic
reasons, these numbers likely provide a lower bound for the treatment effects on
such dismissals.
In Panel C of Table 5, we define involuntary job separation as any separation
where there is a gap between the end date of the terminating contract and the start
date of any new employment contract of the given person. Although this definition
might capture some voluntary departures as well, we find that it is a strong predictor
of receiving any unemployment benefits in the year of job separation.
18
The
estimates in Panel C also suggest that employees in treated firms suffered from
greater involuntary job separation.
Finally, we address two potential concerns to the estimated treatment effects
on job separation. First, it might be that treated firms in general have higher
employee turnover, even after controlling for industry fixed effects and 2007 firm
characteristics. To address this issue, we study the 2005–2007 job separation rate of
2005 employees of the treated and control firms. First, we match the 352 firms in
our sample to their employees on Jan. 1, 2005.
19
We then estimate three regressions
for (cumulative) job separation up until end-2005, end-2006, and end-2007 using
18
Among the about 1 million 20- to 60-year-old individuals with a job separation in 2011, about 60%
had a job separation with a gap in employment. 31.5% of these individuals received unemployment
benefits versus only 9% among those who had job separation without a gap in employment.
19
We can match 325 of the 352 enterprise groups to their business units and employees on Jan.1,
2005. Out of the 27 nonmatched enterprise groups, 24 could not be matched because their identifier in the
General Business Register changed between 2005 and 2008 due to restructuring, split, restarting, or
Kárpáti and Renneboog 21
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

the same controls as in columns 1–3ofTable 5 (industry fixed effects and firm
financial characteristics, but measured in 2007). Table 2 in the Supplementary
Material presents the results from this placebo test. If anything, the employees of
treated firms in 2005 were less likely to be separated from their jobs than employees
of control firms, although the estimates are economically and statistically close to 0.
Second, it might also be possible that employees with a generally weaker
labor force attachment select into the treated firms, for any reason. The placebo
test discussed above cannot address this potential concern as the employees in our
sample, by definition, are with their firm on Jan. 1, 2008. To establish that treated
and control employees exhibited parallel trends in labor force attachment before
the crisis, we study the differences in the (log of) the annual calendar days worked
(summed across all employers) for the two groups. Working fewer calendar days
may indicate that an employee is less attached to the labor market (has gaps in
employment) or that he or she had suffered from job loss (which often leads to gaps
in employment). We regress the log of the annual calendar days worked on the
same controls, firm characteristics interacted with year fixed effects and employee
fixed effects, as in our baseline antidepressant regression. The results, presented in
Figure 1in the Supplementary Material, reveal that there were no differential trends
in the labor force attachment of treated and control employees before the crisis.
In contrast, starting from 2008, treated employees experienced a drop of about 2%–
3% in the number of calendar days worked. This latter result is in line with the
increasing job separation in treated firms during the crisis that we document above.
The fact that treated and control employees show parallel trends in the number
of calendar days worked before the crisis is also supportive of the parallel trends
assumption for our main outcome variable, antidepressant use. This is because
unemployment and job loss are two of the most important employment-related
causes of mental health problems (e.g., Tefft (2011), Schaller and Stevens (2015)).
In summary, the results in Table 5 and the additional analyses in Table 2 in
the Supplementary Material and Figure 1 in the Supplementary Material provide
evidence that while the job security of treated and control employees was similar
prior to the crisis, during the crisis employees in treated firms experienced increas-
ing job insecurity.
B. Increased Antidepressant Use of Employees Who Kept Their Jobs
Can the greater propensity of job loss and its negative effects on employees
who lost their jobs fully explain the deteriorating mental health of employees of
firms with refinancing difficulties? We argue that this is not the case, and that
employees who managed to keep their jobs also suffered from an increased prev-
alence of mental health problems.
First, in a back-of-the-envelope calculation, we multiply the job loss estimates
(with an upper bound of 6.2 pp) in Table 5 with the effects of job loss on self-
reported depression/anxiety estimated by Schaller and Stevens (2015) (1.6 pp). The
resulting treatment effect (6.2 pp × 1.6 pp = 0.1 pp) is clearly smaller than the
mergers. Two enterprise groups were created in the period, and we find no information on one remaining
enterprise group.
22 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

0.44 pp overall increase in antidepressant use we estimate in Table 4. This suggests
that it is not merely mental illness caused by job loss that is driving our results.
Next, in Table 6 we restrict our sample to employees who kept their jobs at
least till the end of the year in which we measure antidepressant use. In the first
7 columns, we repeat the analysis of Table 3 for this restricted sample of employees,
while in the last column, we estimate the average 2008–2012 treatment effect as
in column 1 of Table 4. We observe a similar trend in treatment effects as for the
complete sample, the excess probability of antidepressant use rapidly increases
in 2008 and remains statistically significant until 2013. The average 2008–2012
treatment effect amounts to 0.28 pp, or 6.5% compared to the unconditional mean
antidepressant use of 4.3% in this sample. Although we cannot interpret these
results causally due to the possibly endogenous nature of job separation, discussed
in footnote 4, they support a negative effect of refinancing constraints on the mental
health of employees who kept their jobs.
Finally, we study the moderating effect of personal, household, and employ-
ment characteristics that are expected to increase the effects of firm-level refinan-
cing difficulties on experienced job insecurity and the effects of job insecurity on
employee mental health. We consider eight moderating characteristics: age, gender,
partnership status, having children in the household, the share of salary in total
household income, job tenure, wage, and job loss experienced by peer employees.
Personal characteristics (age and gender) may influence reemployability upon job
loss. In the Netherlands, Deelen, de Graaf-Zijl, and van den Berge (2018) show that,
TABLE 6
Treatment Effects on Employees Keeping Their Job
Table 6 shows estimates of the effect of a firm having to repay at least 25% of long-term debt in 2008 (Treated) on employees’
antidepressant use. Antidepressant use is originally a binary variable that takes the value 1 if a person was reimbursed for
(any) antidepressant use in the given year; we multiply this variable by 100 and therefore all coefficients in the table are
expressed in %. Contrary to Tables 3 and 4, which follow employees over time even if they leave their initial job, this table only
considers employees who stay in their initial job (the job on Jan. 1, 2008), that is, stayers, until at least the end of the year of the
observation. The first seven columns present the treatment effects over time, based on model (1), in a similar manner as in
Table 3. The last column presents the average 2008–2012 treatment effect, based on model (2), in a similar manner as in Table
4(a difference is that this specification controls for industry × post-fixed effects instead of industry × year FE). The t-statistics,
reported in parentheses, are based on standard errors clustered at the firm level. *, **, and *** denote significance at the 10%,
5%, and 1% levels, respectively.
Dependent Variable: ANTIDEPRESSANT_USE (Binary, ×100)
2006 2008 2009 2010 2011 2012 2013
2008–
2012
TREATED 0.0849 0.323*** 0.326** 0.362** 0.342 0.406** 0.144 0.278**
(0.88) (2.60) (2.03) (1.98) (1.63) (2.27) (0.73) (2.15)
LIQUID_ASSETS_
TO_TA, 2007
0.501 0.226 0.181 0.129 0.704 0.688 0.899 0.0897
(1.34) (0.71) (0.37) (0.22) (0.96) (0.82) (1.51) (0.20)
LT_DEBT_
TO_TA, 2007
0.0199 0.282 0.745*** 0.974*** 0.965*** 0.469 0.368 0.636***
(0.12) (1.30) (3.18) (3.63) (2.98) (1.35) (1.08) (3.24)
log(TOTAL_
ASSETS), 2007
0.00556 0.00615 0.00388 0.0326 0.0378 0.0215 0.0632* 0.0170
(0.32) (0.30) (0.16) (1.26) (1.19) (0.73) (1.90) (0.81)
CF, 2007 0.180 0.273 0.668 0.430 0.155 0.760 0.914 0.280
(0.45) (0.70) (1.10) (0.56) (0.18) (0.94) (1.25) (0.47)
Unconditional mean 3.72 4.10 4.22 4.30 4.50 4.53 4.67 4.31
Employee FE Yes Yes
Industry × year FE Yes Yes
No. of firms (=clusters) 352 352
No. of obs. 1,986,249 1,817,359
Kárpáti and Renneboog 23
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

following a dismissal, older (ages 45–54 in their sample) men are more negatively
affected in terms of reemployment probability than either prime-age men (age 35–
44) or older women. The mental health effects of unemployment and job loss also
appear to be stronger for men than for women (e.g., Kuhn et al. (2009), Paul and
Moser (2009)). This suggests that job insecurity may be more stressful for older
and male employees. Having no partner may represent a lack of a familial support
and could increase the risk of developing mental illness (Teo, Choi, and Valenstein
(2013)), whereas having (a) child(ren) could indicate that job loss is more conse-
quential due to a higher number of dependents. Earning a salary that represents a
greater share of the total household income (conditional on having a partner or not)
may imply a more detrimental effect of an eventual job loss on the family budget.
Indeed, Marcus (2013) suggests that the mental health effects of job loss are worse
if the dismissed employee had a higher pre-dismissal share of household income.
The potential moderating role of tenure is motivated by Caggese et al. (2019), who
find that financially constrained firms may find it optimal to dismiss short-tenured
employees when facing economic distress. Regarding wage, we consider an indi-
cator that takes the value of 1 if an employee’s hourly wage is in the top quartile of
all hourly wages in the firm. The idea here is to create a proxy for managerial
employees, who are likely more aware of the financial and economic circumstances
of the firm.
20
Finally, we also consider the job insecurity experienced by the
peersofstayeremployees.We calculate the difference in job separation rates
(as defined in Panel A of Table 5) in the business unit of the employee between
2005–2007 and 2008–2010. We expect stronger adverse effects on the mental
health of employees who work in business units that experienced a greater increase
in job separation rates.
Table 7 shows the moderating effects of the above characteristics on the 2008–
2012 average treatment effect of antidepressant use, based on model (3). The table
presents the coefficient estimates of the triple interactions Post × Treated × Char-
acteristic and of the double interactions Post × Treated and Post × Characteristic.
Columns 1–8 show results from eight separate univariate specifications where we
interact Post × Treated with a single characteristic measured pre-treatment (during
2007 or on Jan. 1, 2008), while column 9 presents results from a model where all
the eight triple interactions are included. As the results reveal, treatment effects are
larger both in the univariate and in the multivariate regressions for employees
who have at least one child in their households (column 2), whose salary constitutes
a large share of their total household income (column 4), who have 5–10 years
of tenure
21
(column 6), and who work in a business unit where job separation
20
Nonmanagerial employees can also learn about their employer’s financial and economic circum-
stances, either by observing the outcomes of their peer employees (e.g., dismissals) or from information
shared by the employee representatives. Firms in the Netherlands that employ at least 50 employees are
required to establish a “Works Council,”an employee representative body, which has extensive infor-
mation and consultation rights.
21
The estimated treatment effect heterogeneity in tenure merits further discussion as it is slightly
different from what we would hypothesize based on Caggese et al. (2019). Although we do find that
employees with more than 10 years of tenure are less affected than employees with 5–10 years of tenure,
it is employees with less than 5 years of tenure whose mental health is the least affected. A potential
explanation is that although short-tenured employees do indeed face a higher risk of job loss
(in untabulated regressions we do find evidence of this), they already internalize this risk (e.g., because
they work in fixed-term contracts, and their mental health is less sensitive to increasing job insecurity).
24 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

TABLE 7
Treatment Heterogeneity Among Stayers
Table 7 reports estimates of treatment heterogeneity for the 2008–2012 average treatment effect, based on model (3). All specifications
include the same controls as column 1 of Table 4.AsinTable 6, we restrict observations to stayers, employees who keep their jobs at least
till the end of the year in consideration. Columns 1–8 present results from five univariate specifications where we interact Post × Treated
with a single characteristic measured pre-treatment (during 2007 or on Jan. 1, 2008). Column 9 presents a multivariate specification
where Post × Treated is interacted with each characteristic. NO_PARTNER is 1 if a person lived without a partner (unmarried or married).
HAS_CHILD(REN) is 1 if a person had at least one child in his/her household. HIGH_SHARE_IN_HOUSEHOLD_INCOME is 1 if the share
of a person’s salary in his/her total household income was in the top half of the distribution (conditional on having or not having a partner).
AGE_ABOVE_44 refers to the age of a person in 2008. TENURE_5_10_YEARS and TENURE_10+_YEARS are indicators of having a
tenure between 5 and 10 years or above 10 years, respectively. Employeeswith a tenure lower than 5 years serve as the omitted category.
WAGE_IN_TOP_25% is an indicator if the employee’s hourly wage was in the highest 25% of all wages within the firm in 2007.
DIFFERENTIAL_JOB_SEPARATION_IN_BUSINESS_UNIT refers to the difference of job separation rates (see Appendix B for a
definition) between 2005–2007 and 2008–2010 in the business unit of the employee. The t-statistics, reported in parentheses, are
based on standard errors clusteredat the firm level. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.
ANTIDEPRESSANT_USE (×100)
POST × …123 456 789
TREATED 0.252** 0.090 0.333** 0.186 0.262* 0.016 0.224* 0.284* 0.359
(2.04) (0.61) (2.27) (1.48) (1.82) (0.10) (1.68) (1.82) (1.59)
NO_PARTNER 0.288*** 0.330***
(3.28) (2.97)
TREATED × NO_PARTNER 0.114 0.398**
(0.87) (2.33)
HAS_CHILD(REN) 0.171*** 0.072
(2.86) (0.89)
TREATED × HAS_
CHILD(REN)
0.330*** 0.394***
(3.27) (3.79)
FEMALE 0.341*** 0.379***
(4.06) (3.98)
TREATED × FEMALE 0.0681 0.083
(0.60) (0.62)
HIGH_SHARE_IN_
HOUSEHOLD_INCOME
0.142** 0.072
(2.42) (0.94)
TREATED ×
HIGH_SHARE_IN_
HOUSEHOLD_INCOME
0.223*** 0.229**
(2.49) (2.22)
AGE_ABOVE_44 0.0988* 0.015
(1.93) (0.22)
TREATED × AGE_
ABOVE_44
0.0512 0.032
(0.47) (0.26)
TENURE_5_10_YEARS 0.0231 0.069
(0.26) (0.63)
TENURE_10+_YEARS 0.206*** 0.199*
(2.91) (1.96)
TREATED ×
TENURE_5_10_YEARS
0.731*** 0.710***
(5.45) (4.70)
TREATED × TENURE_
10+_YEARS
0.213** 0.087
(2.01) (0.66)
WAGE_IN_TOP_25% 0.404*** 0.300
(7.21) (3.75)
TREATED × WAGE_
IN_TOP_25%
0.219** 0.171
(2.32) (1.17)
DIFFERENTIAL_
JOB_SEPARATION_
IN_BUSINESS UNIT
0.00 0.198
(0.00) (0.65)
TREATED × DIFF_
JOB_SEPARATION_
IN_BUSINESS_UNIT
2.046** 2.186**
(2.13) (2.26)
Industry × post FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
2007 firm variables × post Yes Yes Yes Yes Yes Yes Yes Yes Yes
Employee FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
No. of firms (=clusters) 352 352 352 352 352 352 352 316 315
No. of obs. 1,817,338 1,817,338 1,817,338 1,777,198 1,817,338 1,817,338 1,777,242 1,237,772 1,217,531
Kárpáti and Renneboog 25
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

increased more during the crisis (column 8). Employees without a partner (col-
umn 1) and those with an hourly wage in the top quartile (column 7) also appear to
be more affected, although these differences are only statistically significant in the
multivariate and the univariate specification, respectively. Finally, we do not find
differences in treatment effects for employees of different age and gender.
22
Taken together, these treatment heterogeneity estimates provide suggestive
evidence that job insecurity may be a potential factor behind the increase in
antidepressant use for employees who did not lose their jobs. On the other hand,
the causal interpretation of these treatment heterogeneity tests needs to be made
cautiously as other (omitted) moderators might drive some of the results. It is
possible that other factors related to stress at work, and not related to job sepa-
rations, are responsible for the adverse effects on mental health. For example,
Popov (2014) shows that credit constraints are associated with significantly lower
investment in on-the-job training, which could also result in lower productivity
and mental strain.
In summary, our results suggest that the adverse mental health effects of
refinancing constraints are present both for employees who lose their jobs due to
these constraints and for employees who manage to keep their jobs albeit possibly
suffer from greater job insecurity. From the firm’s perspective, the mental illness of
nondismissed employees is of particular importance as it may negatively affect firm
productivity. Therefore, our results illustrate that the mental health costs of financial
constraints are not restricted to dismissed employees but also affect the firm.
V. Robustness and Placebo Tests
There are several assumptions regarding treatment specification and sample
selection that underpin our results of a greater post-2007 increase in antidepressant
use in high-repayment share firms. In this section, we present estimates where
we relax/alter these assumptions. We also perform placebo tests to verify that our
results are not driven by the excess sensitivity of treated firms to the economic
downturn in 2008–2009 (i.e., macroeconomic effects unrelated to the credit supply
shock) and to assert that the treatment effect does not apply in firms where the
repayment share is not expected to be binding because of internal capital markets.
We also address possible ex ante sorting by employees in control and treated firms
prior to the 2008 shock. In addition, we show evidence that the negative effects of
financial difficulties on employee mental health also hold in a much larger sample
22
We also study treatment heterogeneity in education. Higher-educated employees may face better
prospects of reemployability; they may in general serve in different positions (e.g., more likely in
managerial roles) than lower-educated employees. We have information on education for a subset of
the employees in our sample (about 16,000 employees out of the total 330,000) from the GEMON 2012
survey. We code if someone is college educated by means of a binary variable, and interact this variable
with the POST × TREATED indicator. Among stayers, higher-educated employees appear to be less
affected by the debt refinancing shock, but this difference is not statistically significant. We also study
treatment heterogeneity in medicine use: we create an indicator variable capturing if a person has used
any medicines (apart from antidepressants) in 2007, and interact it with our treatment variable. Our
results show that employees who were in worse physical health (as proxied by the use of medication
other than antidepressants) were more affected by the treatment, but this difference is not statistically
significant.
26 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

where we proxy for financial difficulties by means of a high leverage ratio. Finally,
as CLAs may be an additional source of uncertainty, we control for renegotiations
of CLAs in the treatment period.
A. Sample Selection Criteria and the Definition of Treated Firms
We present robustness tests based on changes in the sample selection criteria
(Table 8). Column 1 presents the baseline estimate of the 2008–2012 average
treatment effect on antidepressant use from Table 4. Column 2 excludes firms with
any long-term debt resulting from intragroup loans on their opening 2007 balance
sheet. Ideally, we would restrict our 2008 repayment share variable to the repay-
ment of long-term debt that is most probably binding and hard-to-renegotiate, such
as bank loans and bonds. Due to data limitations, this is not possible, but excluding
firms with intercompany loans would alleviate concerns that our repayment share
variable picks up nonbinding repayment obligations within the group. The point
estimate from column 2 is very close to the baseline estimate, although the sample
size decreases.
Columns 3–5 investigate alterations on the long-term debt to total assets
selection criterion to capture the degree to which firms rely on long-term debt.
When we include firms with lower long-term debt to total assets ratios (columns
3 and 4), the treatment effects become smaller, while restricting the sample to firms
with at least 15% long-term debt to total assets increases the estimate. This is as
expected, as the size of the refinancing shock is arguably proportional to the share of
long-term debt on the balance sheet.
Column 6 retains all industries (i.e., the state-controlled and heavily regulated
industries such as utilities are not excluded). Adding firms that belong to regulated
industries or form part of the state administration yields a slightly lower (0.34 pp vs.
0.44 pp) 2008–2012 treatment estimate, indicating that for such firms the refinancing
TABLE 8
Variations on Sample Selection
Table 8 reports the 2008–2012 average treatment effects (from model 2) when the sample selection criteria are changed. All
specifications include the same controls as column 1 of Table 4. Column 1 shows the baseline (repeats column 1 in Table 4). Column
2 excludes firms that had any long-term group lending on their 2007 opening balance sheet. Columns 3–5 vary the minimum long-term
debt to total assets ratio (excluding the current portion). Column 6 also includes firms from government-controlled and highly regulated
industries. Columns 7 and 8 exclude the top 5% and 10% largest firms (based on the number of employees in our sample), respectively.
The t-statistics, reported in parentheses, are based on standard errors clustered at the firm (i.e., enterprise group) level. *, **, and ***
denote significance at the 10%, 5%, and 1% levels, respectively.
Baseline
Excluding
Firms with
Group Lending
LT Debt to
TA > 0%
LT Debt to
TA ≥5%
LT Debt to
TA ≥15%
No Industry
Restrictions
Excluding
Top 5%
Firms
Excluding
Top 10%
Firms
12345678
AVERAGE_TREATMENT_
EFFECT_2008_2012
(binary, ×100)
0.440*** 0.377** 0.225** 0.357*** 0.497** 0.338*** 0.410** 0.406**
(3.09) (2.47) (2.48) (3.03) (2.50) (4.04) (2.57) (2.28)
Employee FE Yes Yes Yes Yes Yes Yes Yes Yes
Industry × post FE Yes Yes Yes Yes Yes Yes Yes Yes
2007 firm variables ×
post FE
Yes Yes Yes Yes Yes Yes Yes Yes
No. of treated firms 23 22 41 31 20 25 22 21
No. of control firms 352 304 485 408 301 375 335 317
No. of obs. 2,282,057 1,279,136 3,092,926 2,622,039 1,944,757 3,210,309 1,205,793 855,764
Kárpáti and Renneboog 27
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

problems may be more easily addressed without repercussions on the job security of
the employees.
Finally, in columns 7 and 8, we reestimate the treatment effects after control-
ling for a firm size effect. We exclude the largest 5% and 10% of firms, respectively,
where we measure size as the number of employees in the sample. Excluding these
large firms yields almost identical treatment effects, indicating that the results are
not driven by a handful of the largest firms.
Next, we turn to alternative thresholds for the current portion of long-term
debt to define treated and control firms. Table 9 repeats the analyses of Table 3 and
column 1 of Table 4, and presents the estimated treatment effects on antidepressant
use when we use a lower (20%) or higher (30%) cut-off value. The effects for a
higher (30%) cut-off, which results in only 10 treated firms, are slightly larger than
the baseline estimates. If we choose a lower cut-off value (20%), our estimation
results are expectedly attenuated due to the fact that the refinancing (di)stress may
be somewhat lower.
TABLE 9
Alternative Treatment Classifications
Table 9 presents alternative treatment specifications and variations on the 2008 repayment threshold of long-term debt. Panel
A shows the treatment effects over time; column 1 of Panel A corresponds to the first row of Table 3. Panel B shows the average
2008–2012 treatment effect; the estimate in column 1 corresponds to the treatment estimate in column 1 of Table 4.All
specifications include the same controls as Table 3 and column 1 of Table 4. In column 2, we classify firms as treated if they
had to repay at least 20% of their long-term debt in 2008, whereas in column 3if they had to repay at least 30% of their long-term
debt. The changing number of treated and control firms is presented at the bottom of the table. The t-statistics, reported in
parentheses, are based on standard errors clustered at the firm (i.e., enterprise group) level. *, **, and *** denote significance
at the 10%, 5%, and 1% levels, respectively.
Dependent Variable: ANTIDEPRESSANT_USE (Binary, ×100)
25% (Baseline) 20% 30%
123
Panel A. Dynamic Effects
2006 0.0849 0.133 0.0459
(0.88) (1.32) (0.38)
2008 0.412*** 0.291** 0.362***
(3.47) (2.04) (3.34)
2009 0.445*** 0.270* 0.400**
(3.39) (1.66) (2.56)
2010 0.486*** 0.339* 0.561**
(2.81) (1.88) (2.53)
2011 0.667*** 0.487*** 0.696***
(3.52) (2.63) (2.76)
2012 0.407** 0.274 0.488**
(2.25) (1.42) (1.99)
2013 0.295 0.163 0.198
(1.47) (0.80) (0.77)
Panel B. Average Effects
2008–2012 0.440*** 0.265* 0.479**
(3.09) (1.78) (2.52)
Employee FE Yes Yes Yes
Industry × post FE Yes Yes Yes
2007 firm variables × post FE Yes Yes Yes
No. of treated firms 23 37 10
No. of control firms 329 315 342
No. of obs. (Panel B) 2,282,057 2,282,057 2,282,057
28 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

Finally, we address a set of additional potential concerns (stratifying replace-
ment shares, industry effects, eliminating very low repayment share benchmarks,
repayment share definitions), and discuss the results (tables are not shown for
reasons of parsimoniousness but are available upon request).
First, we study the treatment effects using two nonoverlapping treatment
indicators: having a repayment share between 25% and 30% and a repayment share
over 30%. In both specifications, control firms are those with a lower than 25%
repayment share (as in the above analysis). Our results in Table 3 in the Supple-
mentary Material show that both of these groups of treated firms have experienced
increasing antidepressant use in the period of 2008 to 2012, and thus that the results
are not confined to only those firms that had to repay at least 30% of their long-
term debt.
Second, given the possible concern that facing high repayment obligations
may only have an effect on specific industries, we exclude industries one by one
(from both treated and control firms) and estimate very similar treatment effects in
all subsamples.
23
Third, another potential concern is that firms with low repayment shares might
be systematically different from other firms, for instance, because they have nego-
tiated debt contracts with bullet-type repayment and/or use different debt instru-
ments. Many firms in our sample have very low (<5%) 2008 repayment shares. We
therefore exclude in two regression models the firms with the lowest repayment
shares: i) those with a zero repayment share, and ii) firms a repayment share lower
than 5%. We find that the baseline results on anti-depressant are upheld in these
alternative models.
24
Fourth, there are two main ways of defining the threshold of the long-term debt
that needs to be refinanced. In our baseline specification, we define firms with a
high refinancing need based on the ratio of the repayment obligation and total
outstanding long-term debt. Although we had already excluded firms from our
sample that have a low debt to total assets ratio (below 10% in our baseline), it might
be possible that a firm with a lower share of long-term debt to total assets and a large
fraction of it needing to be refinanced is classified as treated, while it may not really
be affected by the refinancing needs. We therefore study another possibility to
define the refinancing needs threshold and consider the ratio of repayment obliga-
tions and total assets. Under this definition, we classify firms as “treated”if the ratio
of their 2008 repayment obligations to their total assets reached at least 7.5%
(case 1), 10% (case 2), or 12.5% (case 3). Using total assets as the denominator
has the advantage that we do not need to restrict our sample to firms that have a
larger share of long-term debt to total assets on their balance sheets to ensure that the
refinancing needs are significant compared to the size of the firm. Abandoning this
restriction increases the size of our sample to 568 firms. The results of this analysis
show that higher repayment obligations in 2008 as a proportion of total assets are
also related with an increase in antidepressant use among employees.
25
23
Table available from the authors.
24
Table available from the authors.
25
Table available from the authors.
Kárpáti and Renneboog 29
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

B. Economic Recessions With and Without Financing Constraints
We perform a placebo test to verify that our results are not driven by the
excess sensitivity of the treated firms to the economic downturn in 2008–2009
(i.e., macroeconomic effects unrelated to the credit supply shock). We exploit the
fact that after a short-lived recovery in 2010–2011, the Dutch economy fell back
into recession in the second half of 2012 (“double-dip recession”). Importantly,
Duchi and Elbourne (2016) show that the effect of credit supply shocks on corporate
lending growth and corporate investments is negligible in this period, noting that
“when we look at the double-dip recession in 2012, adverse credit supply shocks
play no role”(p. 65). Therefore, the 2012 recession presents a negative economic
shock without a strong corporate credit supply component: if our results are indeed
driven by disruptions in corporate credit supply, we would expect to find no positive
treatment effects in this period. Indeed, we demonstrate in Table 10 that, following
the 2012 recession, growth patterns in antidepressant use are similar between
employees of firms that had to repay a large share (>25%) of their long-term debt
in 2012 and employees of other firms in our sample. If anything, employees of high-
repayment firms exhibited a slight decrease in antidepressant use. Column 2 repeats
the job separation analysis from Table 5 in this placebo setting. The results reveal
that employees of 2012 high-repayment firms did not suffer from increased job
insecurity. In summary, the placebo test shows that repayment of a high share of
long-term debt has no detrimental effects on employees during an economic down-
turn when credit constraints were not binding.
TABLE 10
Placebo Test: Firms with High 2012 Debt Repayment Share
Table 10 presents the results of a placebo test where we define financially constrained (treated) firms as those that had to
repay at least 25% of their long-term debt in 2012. Column 1 presents treatment effect estimates on antidepressant use, as
defined in Table 3; the coefficient estimates on the treatment indicator × year interaction terms are shown. The omitted year,
due to employee fixed effects, is 2011. Column 2 presents treatment estimates on cumulative job separation, as defined in
Table 5. Control variables are similar to those in Tables 3 and 5, respectively, but are defined using 2011 data. The t-statistics,
reported in parentheses, are based on standard errors clustered at the firm (i.e., enterprise group) level. *, **, and *** denote
significance at the 10%, 5%, and 1% levels, respectively.
ANTIDEPRESSANT_USE CUMULATIVE_JOB_SEPARATION
12
2010 0.125
(1.35)
2012 0.107 0.0146
(1.30) (0.43)
2013 0.252** 0.000868
(2.38) (0.03)
2014 0.188 0.0193
(1.34) (0.54)
2015 0.355** 0.0361
(2.58) (0.96)
2016 0.284
(1.43)
Employee FE Yes No
Industry × year FE Yes Yes
2011 firm variables × year FE Yes Yes
No. of firms (clusters) 406 406
No. of obs. 2,485,867 1,433,912
30 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

C. Additional Controls Capturing Financial Constraints
It is possible that firms that had to repay a larger share of their long-term debt in
2008 might differ in terms of, for example, debt maturity structure because a shorter
debt maturity could (mechanically) lead to a larger annual repayment share. Firms
with shorter debt maturities might have been differently/more strongly affected by
the financial crisis.
26
So, this could suggest that treated firms could be weaker or,
more broadly, choose a shorter debt maturity within their long-term tranches for
other endogenous reasons. Consequently, it could be that employees sort them-
selves toward the treated firms. Let us first point out that, following Almeida et al.
(2011), we have already controlled for flexible time trends in pre-crisis firm
characteristics that aim to capture firm heterogeneity. In our baseline estimation
(Table 4), these characteristics include 2-digit industry codes, size (log total assets),
long-term debt to total assets, cash flow, and liquid assets (cash) to total assets.
Almeida et al. (2011) argue that “it is commonly accepted that these covariates
capture a lot of otherwise unobserved firm heterogeneity.”
27
Still, we further
examine potential concerns that so far unobserved (quality) differences between
treated and controls firms, and not financing frictions, drive our results.
First, we extend the set of control variables using 2007 accounting information
in Panel A of Table 4 in the Supplementary Material. Column 1 presents the average
treatment effect on antidepressant use in our baseline specification, while columns
2–8 add to the baseline regressions the following flexible time trends based on
additional 2007 firm characteristics: firm profitability measured as ROA (column
2), leverage measured as total debt to total assets (column 3), interest coverage
defined as EBIT divided by interest expense (column 4), a binary indicator captur-
ing dividend payments in 2007 (equal to 1 in case of payments) (column 5), a binary
indicator capturing whether the firm has any bonds outstanding (equal to 1 in case of
bonds) (column 6), an indicator equal to 1 if the firm is a public liability company
28
(column 7), and finally all these controls together (column 8). We choose these
characteristics of firm heterogeneity based on the literature on measuring financial
constraints (Kaplan and Zingales (1997), Lamont, Polk, and Saa-Requejo (2001),
Whited and Wu (2006), and Hadlock and Pierce (2010)). As the results in Panel A of
Table 4 in the Supplementary Material reveal, our baseline estimate of a 0.44
percentage points increase in antidepressant use among treated employees in the
period of 2008–2012 hardly changes (it remains between 0.39 pp and 0.51 pp) when
we include these additional controls (interacted with the Post indicator).
26
Alternatively, theoretical and empirical evidence suggests that firms with shorter debt maturity
may have higher investment opportunities and may signal higher quality by choosing more short-term
debt (Myers (1977), Flannery (1986)). Barclay and Smith (1995) estimate that firms with more growth
options have, ceteris paribus, less long-term debt and find a negative, albeit economically small,
correlation between a measure of firm quality and debt maturity. Stohs and Mauer (1996) document
that debt maturities are on average shorter for higher-quality firms (proxied by more positive future
earnings surprises).
27
Almeida et al. (2011) also control for firms’market-to-book ratio (Q); however, this control cannot
be used for the private firms in our sample (and is not observed for the public firms in our data set).
28
This is a necessary but not sufficient condition for the firm to be publicly traded; it should be noted
that given the confidential nature of the data, Statistics Netherlands does not reveal which firms are in the
sample of treated and control firms.
Kárpáti and Renneboog 31
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

Second, as for a subsample of firms additional past (financial) information is
available, we control for these characteristics in Panel B of Table 4 in the Supple-
mentary Material.
29
While column 1 repeats the results of our baseline specifica-
tion, columns 2 and 3 control for two measures of firm age, a characteristic often
used in measures of financial constraints (e.g., in the Hadlock and Pierce (2010)
index).
30
Column 4 controls for revenue growth, a component of the Whited and
Wu (2006) index of financial constraints, during the period of 2005 to 2007. As this
measure requires balance sheet data from 2005, our sample is reduced to 257 firms.
Column 5 controls for a set of binary indicators (as usual interacted with the year
fixed effects) of business events that affected the firm in 2007. The source of this
information is the National Working Conditions Survey, which is filled out by about
30,000 employees in the last quarter of each year.
31
The question we rely on asks if
“any of the following changes occurred in your company (plant/location) in the past
12 months?”i) large reorganization, ii) takeover by another firm, iii) takeover of
another firm, iv) downsizing without forced dismissals, (v) downsizing with forced
dismissals, vi) merger with another firm, vii) outsourcing of support services,
viii) relocation of business activities abroad, and ix) automation of business oper-
ations. Among these additional control variables, only the indicators for “takeover
by another firm”(0.25 pp) and “takeover of another firm”(0.27 pp) are statistically
significant.
32
Column 6 controls for the share of long-term debt that the firm had to
repay in 2007. If treated firms indeed have a systematically shorter debt maturity
structure, their 2007 repayment share should also be higher on average. Therefore,
by controlling for the 2007 repayment share we can partially control for the maturity
of corporate debt in our sample. We find that adding this additional control variable
does not affect the estimated treatment effect. Finally, column 7 combines all these
preceding controls, while column 8 also includes the control variables that we
added in Panel A of Table 4 in the Supplementary Material. The results in Panel
B of this table show that our baseline estimate of a 0.44 percentage points increase in
antidepressant use among treated employees in the period of 2008 to 2012 is robust
29
Past financial information is not available for all sample firms due to two main reasons: i) firms are
not part of the “Annual Statistics of Finances of Large Enterprises”if they are not considered to be a large
enterprise in the given year and ii) data on some firms cannot be traced back to earlier years because the
firm’s unique identifier changes (e.g., due to mergers, reorganizations or a change in the tax unit
structure).
30
Although there is no direct information on firm age in the data sets of Statistics Netherlands, we can
still match 326 of the 352 firms in our sample by means of the historical General Firm Register. The
register contains in some cases the year of establishment and in others the year of the first appearance in
the register. As the register starts in 1994, age is right censored at 14 years for many sample firms.
Consequently, in column 2, we control for abinary indicator if the firm is at least 14 years old (about 60%
of the sample firms). Another proxy for firm age is the longest tenure among all employees in the firm
(available for all sample firms) (column 3).
31
We use data from the 2007 survey wave. There are 218 firms in our sample in which at least one
employee filled out the questionnaire.
32
For reasons of parsimoniousness, we do not show all these parameter estimates in Table 4 in the
Supplementary Material, but they are available upon request. The significance of the takeover coeffi-
cients suggests that experiencing a takeover, either on the target or on the acquirer side, can be
detrimental for employee mental health. This is in line with the results of Bach et al. (2021), discussed
in the introduction.
32 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

to the addition of these further control variables. The estimates range between
0.37 pp and 0.57 pp.
D. Matching
Third, as we want to address possible ex ante sorting by employees in control
and treated firms before the 2008 shock and alleviate any concerns that these
differences drive our results, we control for employee traits in our models and
apply additional matching strategies. The summary statistics presented in Table 2
have revealed some differences between the employees of treated and control firms.
For example, employees of treated firms were more likely to use antidepressants
before the crisis and are a bit younger. Importantly, these unconditional differences
either disappear or are substantially reduced once we control for the same charac-
teristics that we use in our regression models: industry fixed effects, firm size (total
assets), cash flow, cash ratio, and long-term debt to total assets ratio. Table 5 in
the Supplementary Material presents the differences in pre-crisis employee char-
acteristics between treated and control employees unconditionally in Panel A, and
conditional on the included control variables (Panel B). For example, although
the unconditional mean of antidepressant use is 0.76 pp higher among treated
employees (column 1), once we account for differences in industry composition,
size, cash flow, cash ratio, and long-term debt ratio, the difference disappears.
Unconditionally, treated employees are younger (column 2), more likely female
(column 3), less likely to have a partner (column 4), less likely to have dependent
children (column 5). Also, they take on average 0.15 more types of medications
(ATC-4 code groups) (column 6), earn a lower salary (column 7), and have a lower
household income (column 8). However, most of these differences arise from the
different industry composition and other differences in firm characteristics of
treated and control firms. Once accounting for the control variables in Panel B,
these differences disappear or greatly shrink. Although treated employees still earn
a lower salary (column 7), the difference is only weakly statistically significance
(at the 10% level). We onlystill observe some differences in age and the probability
of having dependent children.
To account for possible imbalances between the characteristics of employees
in treated and control firms, we have also performed the regression analysis on
a matched sample of treated and control firms. Panels C and D of Table 5 in the
Supplementary Material present the differences in characteristics of treated and
control employees after applying matching.
33
The results in Panel C show that, even
without adjusting for the firm-level characteristics in a regression setting, matching
33
There are different ways to perform matching, based on how to calculate the similarity between
observations and how many similar observations (neighbors) to use. We exact match on the industry
codes and use nearest-neighbor matching based on the Mahalanobis distance measure on the four
financial characteristics (cash ratio, log total assets, long-term debt to total assets, and cash flow). We
use matching with replacement and for each treated firm we select the 3 “nearest”neighbor control firms.
We estimate the regression models on these “matched sample”that consists of the treated firms (with any
matched controls) and the nearest neighbor control firms. We also implement matching with 5 nearest
neighbors and/or based on propensity scores.
Kárpáti and Renneboog 33
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

eliminates almost all (mean) differences between treated and control employees.
An exception is the difference in tenure, which slightly increases. The results in
Panel D show that controlling for the firm-level characteristics, there remain
no significant differences between treated and control employees, including
differences in tenure. We obtain similar results with different matching method-
ologies, using the five instead of the three nearest neighbors and/or matching on
the propensity score. We then estimate treatment effects on 2008–2012 average
antidepressant use in the matched sample(s). The estimates, presented in Table 6
in the Supplementary Material, range between 0.42 pp and 0.46 pp and are all
statistically significant at the 1% or 5% levels. Because matching greatly reduces
the pre-treatment differences between treated and control firms and their
employees, these results alleviate potential concerns that the estimated treatment
effect on antidepressant use is driven by imbalances between treated and control
employees.
E. Refinancing Frictions and Internal Capital Markets
We implement an additional placebo test relating repayment share with
antidepressant use on firms for which the channel of refinancing difficulties is
not expected to be binding. In this test, we study the treatmenteffects for foreign-
controlled firms. When such firms have to repay a large share of (intragroup or
external) debt, they may not face refinancing difficulties because their corporate
groups could meet repayment obligations through their internal capital markets.
34
The results, available upon request, show that employees of foreign firms that had
to repay more than 25% of their outstanding long-term debt in 2008 did not exhibit
higher antidepressant use during the crisis (the coefficient estimate is 0.06 with a
t-value of 0.4) compared to employees of other foreign firms.
F. Leverage
Studying the repayment share of long-term debt is not the only way to
investigate the effects of employer firms’financial difficulties on employees’
mental health during the crisis. We had argued that focusing on the repayment
share offers the possibility of identifying the causal effects of refinancing diffi-
culties on employee mental health because repayment obligations are pre-
determined by debt contracts. A drawback of focusing on repayment share is that
our sample is reduced to the largest Dutch firms for which we have information on
the repayment share. Adopting leverage (long-term debt plus short-term debt over
total assets) as a measure of financial vulnerability substantially increases the
sample size: leverage information is available for most nonfinancial firms in the
Netherlands such that our sample increases to about 94,000 firms and more than
34
For instance, Almeida, Kim, and Kim (2015) argue that internal capital markets of Korean business
groups helped them mitigate the negative effects of the Asian crisis on investment and performance.
Desai et al. (2004) study the internal capital markets of multinational corporations and find that
multinational firms employ internal capital markets to overcome the imperfections in external capital
markets. Gopalan, Nanda, and Seru (2007) study Indian business groups and find that intragroup loans
are being used as a means of support for firms in distress.
34 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

2,000,000 employees.
35
On the negative side, a firm’s leverage does not neces-
sarily capture only its financial position. Leverage might be correlated with the
firm’s ownership and governance structure (e.g., family firms), its recent eco-
nomic performance, its future growth opportunities, and other factors. Employees
of firms with a high leverage ratio might have behaved differently even in absence
of the credit supply shock presented by the Global Financial Crisis. Due to these
limitations, we regard this analysis as a robustness tests on our main specification.
The motivation for this robustness analysis is found in the study of Giroud and
Mueller (2017) who provide evidence that high-leverage firms decreased their
employment more during the Great Recession in response to local demand
shocks. Financing constraints may dampen labor demand by impairing firms’
ability to engage in labor hoarding, a prevalent practice in the Netherlands during
the crisis (Van den Berge et al. (2014)).
Our results show that employees of firms with a high leverage ratio (at the
year-end of 2007) increased their use of antidepressants more during the period of
2008 to 2012. In all specifications, our key independent variable is an indicator
capturing whether the firm had higher than median (0.47) leverage ratio at
end-2007.
36
This indicator is statistically significantly (at the 1% level) correlated
with the increase in antidepressant use in the period of 2008 to 2012. The size of the
effect (0.08 pp in the baseline specification) is smaller than in our main specifica-
tion. This is as expected because having a high leverage ratio does not necessarily
lead to refinancing difficulties. Our results also provide evidence that one of the
possible transmission channels of high leverage to worsening mental health is
higher job uncertainty as employees in treated firms were also more likely to be
separated from their jobs in 2008/2009.
37
G. Collective Labor Agreements
CLAs could affect the degree of protection of employees in relation to their
working hours and conditions. If a new CLA were to be negotiated in the context of
an economic recession and financial distress at the corporate level, additional
uncertainty could weigh on employees, bringing about additional stress. A status
quo (the nonnegotiation) of a CLA may hence provide stability to employees and
eliminate a source of uncertainty. Ouimet and Simintzi (2021) state that “firms […]
locked in by wage agreements during the crisis outperform their peers. Implicit in
35
While our baseline study is based on firms present in the SFGO (Annual Statistics of Finances
of Large Enterprises) data set, these analyses are based on the much larger NFO (Annual Statistics of
Finances of Non-Financial Enterprises) data set. We apply the same sample selection criteria as in our
baseline analysis with two differences. We do not exclude firms with a low long-term debt to total assets
ratio (as we consider all forms of leverage).
36
In a first set of regressions, we include the same control variables (cash flow, liquid assets to total
assets, and log total assets) as our baseline specification (with exception of the long-term debt to total
assets ratio as this is used to defined leverage ratio and hence the treatment). In subsequent regressions,
we include additional firm-level (ROA, paying any dividends, interest coverage ratio, being a public
limited company) and employee-level (age, gender, tenure, pre-tax salary) characteristics as controls
(interacted with the Post indicator).
37
Table available from the authors.
Kárpáti and Renneboog 35
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

our findings is the assumption that managers of the firms not bound by the agree-
ments made decisions that ex post were not value maximizing.”
We would expect that the renegotiation of firm-level CLAs would have less
impact in the Netherlands than in the US or UK for two reasons. First, in the
Netherlands, most firms and employees fall under a sectoral CLA and only a
minority of firms (mainly the larger ones) have a firm-level CLA. Indeed, in our
sample about 60% of employees are covered by a sectoral CLA, even though our
sample firms are large. Even if a sectoral CLA were to be renegotiated in 2008,
employees of a treated firm may react less to this uncertainty because the outcome
of the renegotiation will be affected by the (financial) situation of the whole sector
and not by the heath of their own firm. Second, reducing what is considered by
employees as “acquired rights”would be very rare. For instance, CLAs would not
agree to reduce salaries, pay packages, fringe benefits. In addition, many employ-
ment conditions are legally determined: for example, employers in the Netherlands
are not allowed to fire employees for reasons of illness for a period of 2 years and are
required to offer gradual reintegration tracks. Laying off people is strictly regulated
and in the wake of a restructuring an employer faces limitations (see above).
Consequently, the CLAs in the Netherlands are mainly the result of negotiations
between employers and unions about wage increases, potential bonuses, or labor
flexibility. Finally, we should point out that in our analyses, we control for industry-
time fixed effects, which should mostly account for the effects of changes in
sectoral CLAs.
We test the impact of CLAs that are (not) renegotiated in the year of financial
stress, considering both firm-level and sectoral CLAs. We have collected the list of
company and sectoral CLAs that were agreed upon prior to Sept. 2008 and that have
not expired prior to Jan. 2010.
38
Of the about 330,000 employees in our sample,
23% fell under a CLA that was renegotiated just before the crisis (sectoral CLAs or
firm-level CLAs). In our basic model, we add the CLA variable interacted with the
year-fixed effects. Controlling for the pre-2008 renegotiation of firm and sectoral
CLAs only slightly diminishes the positive relation between antidepressant use and
repayment share (from 0.44 pp to 0.37 pp). Interestingly, the CLA indicator has a
positive effect on 2008–2012 antidepressant use (albeit only weakly significant –at
the 10% level): Employees of firms/sectors that had renegotiated their CLAs just
before the crisis experienced a slightly higher increase in antidepressant use during
the crisis (Table 7 in the Supplementary Material). A possible explanation could be
that CLAs agreed upon before the crisis reduced the firms’abilities companies to
respond to the crisis by cutting working hours or introducing flexible labor.
VI. Conclusion
This article argued that corporate financial constraints can have adverse effects
on employee mental health and that these effects are not restricted to employees
38
We follow the treatment definition of Ouimet and Simintzi (2021): “Our treated firms include firms
that agreed to a multiyear settlement before September 2008 and this settlement expired only after
January 2010.”
36 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

who lose their job due to these constraints. To identify the causal effects of financial
constraints, we exploited the plausibly exogenous variation in firms’need to
refinance their long-term debt in 2008, a period when refinancing became more
difficult due to a severe tightening of bank lending standards. Using administrative
data from the Netherlands on the antidepressant medicine use of 330,000
employees in 352 firms, we estimated that employees of firms that were facing
the repayment of at least 25% of their long-term debt in 2008 were 0.44 pp more
likely to consume antidepressants in the period of 2008 to 2012. This is an eco-
nomically significant 9% increase relative to the 5% unconditional prevalence of
antidepressant use, comparable to the 7.5% rise in antidepressant prescription
volume following the 20% decline in US housing prices between July 2006 and
Feb. 2009 estimated by Lin et al. (2013).
These results are qualitatively robust to alternative industry classifications,
variations in control variables, restricting or broadening the sample of firms,
altering the 25% refinancing cut-off, and using pre-regression matching to remove
any imbalances between employees of treated and control firms. Placebo tests
suggest that the results are not driven by the excess sensitivity of treated firms to
the economic downturn in 2008–2009 (i.e., macroeconomic effects unrelated to the
credit supply shock) and that the relation between financial constraints and mental
health does not apply in firms where financial constraints are not expected to be
binding because of internal capital markets.
Although the estimated effects can be partially explained by higher job loss in
constrained firms, much of the increase in antidepressant use occurs at employees
who manage to keep their jobs. Studies of employee-level heterogeneity in the
treatment effect, among employees who keep their jobs, suggest that antidepressant
use grows more for employees who may experience a larger increase in job
insecurity, or for whom job insecurity may represent a greater mental health burden:
employees with children, employees without a partner, employees whose salary
constitutes a greater share of family budget, and employees who work in business
units where job separation increased more during the crisis. Although we lack direct
data on employee perceptions of job security, these results suggest that increased
perception of job insecurity is a transmission channel for deteriorating mental
health.
Given the important role of mental health in employee productivity, these
results provide evidence that deteriorating mental health represents a hitherto
undocumented cost of financial constraints for firms. Furthermore, they also illus-
trate that crisis periods can have an adverse mental health effect even on employees
who manage to keep their jobs, as these employees may still suffer from decreased
perceptions of job security.
Kárpáti and Renneboog 37
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

Appendix A. Databases Used
Appendix B. Variable Definitions
Appendix B reports the description of all the variables used in the analysis. The Annual
Statistics of Finances of Large Enterprises (SFLE) contains both opening and
closing balance sheet values; in the main analysis we use 2007 closing values,
therefore we refer to these variables in the table below. “Initial job”refers to the
employment relationship that existed on Jan. 1, 2008, the date on which employees
were matched to employer firms in our sample. All monetary values are in nominal
EURs.
TABLE A1
Statistic Netherlands Data Sets Used
Table A1 reports the Statistics Netherlands(SN) data sets used in the analysis.
Name in English SN Name Description
Annual Statistics of Finances of
Large Enterprises, SFLE
Statistiek Financiën van
Grote Ondernemingen,
SGFO
Annual survey on the finances (balance sheet, income
statement) of the largest nonfinancial enterprises in the
Netherlands. As of 2007, all enterprises are sampled
with total assets over EUR 23 million. Close to 100%
response rate for the largest 300 enterprises
General Business Register, GBR Algemeen Bedrijven
Register, ABR
Continuously updated database of companies
registered in the Netherlands, with information on
corporate/legal structure (enterprise group, business
units, legal entities), industry classification codes and
events (e.g., mergers, liquidation)
Qualitative characteristics of
employment relationships
BAANKENMERKENBUS Information on, inter alia, start and end date of
employment relationship, type of employment (e.g.,
regular employee, on-call, outsourcing, manager-large
shareholder), social security insurance indicators (e.g.,
insured for unemployment benefits)
Quantitative characteristics of
employment relationships
BAANSOMMENTAB Information on, inter alia, taxable salary, calendar days
worked and payroll tax withheld
Annual dispensations of
medicines per ATC-4 code
per person
MEDICIJNTAB All medicines dispensed that are reimbursed under the
basic health insurance policy to persons who are
registered in the Municipal Personal Records Database
(GBA). No quantities are recorded; merely the 4-digit
ATC codes (e.g., N06A) are listed that were dispensed
for a given person in the statistical year
Extract from the Municipal
Personal Records Database
Gemeentelijke Basis
Administratie,
GBAPERSOONTAB
Demographic background data (that do not or hardly
change) of all persons who appear in the Municipal
Personal Records Database from Jan. 1, 1995 (e.g.,
gender, year of birth, migration background)
Income of People, Income of
Households
IPI/IHI Annual income components (such as labor income,
subsidies, income from entrepreneurship) of people
resident in the Netherlands on Jan. 1 of the statistical
year, and their households. Information on the position
of the person within the household with respect to the
head of the household
Wealth and household
composition
VEHTAB/
KOPPELTABELVEHTAB
Annual wealth components (assets and liabilities) of
households in the Netherlands on the Jan. 1 of the
statistical year. KOPPELTABELVEHTAB contains
information on household members
National Labor Conditions
Survey
Nationale Enquête
Arbeidsomstandigheden,
NEA
Annual survey of workers (excluding self-employed)
between 15 and 74 years old on working conditions,
work content, labor relations and employment
conditions
38 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

Long-Term Debt Structure
LONG_TERM_DEBT_TO_GROUP_COMPANIES: Both in the Netherlands and
abroad, maturity > 1 year. Source: SFGO/B65.
SUBORDINATED_LOANS: Maturity >1 year. Source: SFGO/B67.
BONDS_OUTSTANDING: Maturity >1 year. Source: SFGO/B69.
LT_BANK_LOANS: Loans from domestic financial institutions, including mortgages,
maturity >1 year. Source: SFGO/B71/.
OTHER_LONG_TERM_DEBT: Other unclassified long-term debt, including loans
from private parties, financial leasing, derivatives, and member loans (for
cooperatives). Source: SFGO/B73.
CURRENT_PORTION_OF_LONG_TERM_DEBT: Repayment obligation of long-
term debt (including bonds and other debt) due within 1 year. Source: SFGO/B85.
TOTAL_LONG_TERM_DEBT: = Long-term debt to group companies + Subordinated
loans + Bonds outstanding + Loans from domestic financial institutions + Other
long-term debt.
TOTAL_LONG_TERM_DEBT_INCLUDING_ITS_CURRENT_PORTION: = Current
portion of long-term debt + Total long-term debt.
SHARE_OF_CURRENT_PORTION_OF_LT_DEBT: = (Current portion of long-term
debt)/(Total long-term debt including its current portion).
Firm Characteristics
log(TOTAL_ASSETS): Natural logarithm of the total assets of the company. Source:
SFGO/B37.
LIQUID_ASSETS_TO_TOTAL_ASSETS_RATIO: Liquid assets are the sum of Cash
and cash equivalents, Term deposits with financial institutions, and Receivables
from financial institutions (current account). The ratio is defined as (Liquid assets)/
(Total assets). Source: SFGO/B31-B35; SFGO/B37.
LONG_TERM_DEBT_TO_TOTAL_ASSETS: = Total long-term debt/Total assets.
CASH_FLOW: = (Net income + depreciation and amortization)/Total assets. Source:
SFGO/R20, R05.
SBI93_1993_VERSION_OF_THE_DUTCH_INDUSTRY_CLASSIFICATION_
CODES: The industry classification codes are registered at the Chamber of Commerce
for each legal unit (e.g., B.V.). In the GBR, SN provides a code at the business unit
level by using the code of the legal unit within the business unit that has the most
employees. Similar to this approach, we use the code of the business unit with the
most employees within an enterprise group as the enterprise group level code.
Source: ABR/RBE_SBI93.
ROA (RETURN_ON_ASSETS): Return on assets (=Net income divided by Total
assets). Source: SFGO/R20, B37.
LEVERAGE_RATIO: = (Long-term debt incl. Current portion + Short-term debt)/Total
assets. Source: SFGO/B65-B87.
INTEREST_COVERAGE: = EBIT/Interest expense. Source: SFGO/R07, R12.
Kárpáti and Renneboog 39
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

PAID_DIVIDENDS: Indicator if the firm paid any dividends during the year. Source:
SFGO/R21, R22.
HAS_BONDS: Indicator if the firm had any bonds outstanding at the end of the year.
Source: SFGO/B69.
PUBLIC_LIMITED_COMPANY: Indicator if the firm’s legal form is public limited
company (N.V.). Source: ABR/RECHTSVORMCODE.
FIRM_AT_LEAST_14_YEARS_OLD: Indicator that the firm’s age is at least 14 years
based on Statistics Netherlands’firm registry. Source: ABR.
FIRM_AGE(FROM_TENURE): The tenure (in years) of the employee with the longest
tenure in the enterprise group. Source: BAANKENMERKENBUS.
REVENUE_GROWTH_2005_2007: Percentage change in the revenues of the enter-
prise group between 2005 and 2007. Source: SFGO/R01.
DIFFERENTIAL_JOB_SEPARATION: For a given business unit (=a part of the firm/
enterprise group that carries out similar economic activities according to Statistics
Netherlands), the share of Jan. 1, 2008 employees who separated from their jobs by
Jan. 1, 2011 minus the share of Jan. 1, 2005 employees who separated from their
jobs by Jan. 1, 2008. Job separation includes both voluntary and involuntary
separation, see the definition below. Source: BAANSOMMENTAB, ABR.
Employee Characteristics
INITIAL_JOB: The job (employment relationship) that existed on Jan. 1, 2008 and
based on which the employee was selected into the sample (employees with
multiple jobs on Jan. 1, 2008 are excluded).
TENURE_IN_YEARS: Integer part of number of days since the employment relation-
ship exists (on Jan. 1, 2008) divided by 365 (e.g., tenure in days = 400, tenure in
years = 1). Source: BAANKENMERKENBUS/DATUMAANVANGBAANID.
HAS_A_PARTNER: Takes the value 1 if person iis recorded as household head with
(married or unmarried) partner, or as partner of the household head in the 2007
Income of Households data set; otherwise takes 0. Source: IPI/POSHHK.
DEPENDENT_CHILD: Takes the value 1 if there is an underage child in the household
of the individual on Jan. 1, 2008. Source: IPI.
NUMBER_OF_MEDICINES: The number of different medicines (=ATC4 codes) the
individual was reimbursed for under the basic health insurance policy in 2007.
Source: MEDICIJNTAB.
SALARY: Pre-tax salary from the employment relation (which forms the basis of being
selected into the sample) of the individual in 2007. Source: BAANSOMMENTAB.
WAGE_IN_TOP_25%: Takes the value 1 if the employee’s hourly wage was in the
highest 25% of all wages within the firm in 2007. Source: BAANSOMMENTAB.
HOUSEHOLD_INCOME: Pre-tax household income; sum of all income components
(such as labor income, subsidies, income from entrepreneurship) of all members of
the individual’s household. Source: IHI.
40 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

Outcome Variables
ANTIDEPRESSANT_USE: Takes the value 1 if person iis listed as an antidepressant
(ATC4 code: N06A) user in year t. Takes the value 0 if person iis not registered as
antidepressant user and person iis in the Supplementary Table (KOPPELTABEL-
VEHTAB) of the Wealth of Households (VEHTAB) data set, which contains all
residents on Jan. 1. The variable is set to missing otherwise. Source: MEDICIJN-
TAB, KOPPELTABELVEHTAB.
CUMULATIVE_JOB_SEPARATION: Takes the value 1 if the initial job of person i
terminated by the end of the given year. A job is considered terminated in year tif
there is no salary received from the job in year t+ 1 (more precisely if the job
identifier baanid cannot be matched to year t+1’s BAANSOMMENTAB datafile);
otherwise equal to 0. Source: BAANSOMMENTAB/BAANID.
CALENDAR_DAYS_WORKED: The sum of all days in the year when the employee
had an employment contract. In rare cases, if the employee has multiple jobs, the
value could exceed 365/366. Source: BAANSOMMENTAB/KALDG.
CUMULATIVE_UWV_DISMISSAL: Takes the value 1 if an employee’s initial (Jan.
1, 2008) job ended with a dismissal permit from the Dutch Employee Insurance
Agency (UWV), by the end of the year in consideration. Source: UWVON-
TAANVTAB.
CUMULATIVE_JOB_SEPARATION_WITH_A_GAP: Takes the value 1 if an
employee’s initial job ended, for any reason, by the end of the year in consideration
and there was any gap (>1 day) between the end date of that job and the beginning date
of any following job. Source: BAANSOMMENTAB, BAANKENMERKENBUS.
Supplementary Material
Supplementary Material for this article is available at https://doi.org/10.1017/
S0022109023000595.
References
Almeida, H.; M. Campello; B. Laranjeira; and S. Weisbenner. “Corporate Debt Maturity and the Real
Effects of the 2007 Credit Crisis.”Critical Financial Review, 1 (2011), 3–58.
Almeida, H.; C. S. Kim; and H. B. Kim. “Internal Capital Markets in Business Groups: Evidence from
the Asian Financial Crisis.”Journal of Finance, 70 (2015), 2539–2586.
Bach, L.; R. Baghai; M. Boss; and R. Silva. “How Do Acquisitions Affect the Mental Health of
Employees?”Available at SSRN 3947957 (2021).
Baghai, R.; R. Silva; V. Thell; and V. Vig. “Talent in Distressed Firms: Investigating the Labor Costs of
Financial Distress.”Journal of Finance, 76 (2021), 2907–2961.
Barclay, M. J., and C. W. Smith Jr. The Maturity Structure of Corporate Debt.”Journal of Finance,50
(1995), 609–631.
Benmelech, E.; N. Bergman; and A. Seru. “Financing Labor.”NBER Working Paper No. 17144 (2011).
Benmelech, E.; C. Frydman; and D. Papanikolaou. “Financial Frictions and Employment During the
Great Depression.”Journal of Financial Economics, 133 (2019), 541–563.
Bradford, W.D., and W. D. Lastrapes. “Prescription for Unemployment? Recessions and the Demand for
Mental Health Drugs.”Health Economics, 23 (2014), 1301–1325.
Brown, J., and D. A. Matsa. “Boarding a Sinking Ship? An Investigation of Job Applications to
Distressed Firms.”Journal of Finance, 71 (2016), 507–550.
Browning, M., and E. Heinesen. “Effect of Job Loss due to Plant Closure on Mortality and
Hospitalization.”Journal of Health Economics, 31 (2012), 599–616.
Kárpáti and Renneboog 41
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

Bubonya, M.; D. A. Cobb-Clark; and M. Wooden. “Mental Health and Productivity at Work: Does
What You Do Matter?”Labour Economics, 46 (2017), 150–165.
Burgard, S. A.; J. E. Brand; and J. S. House. “Perceived Job Insecurity and Worker Health in the United
States.”Social Science & Medicine, 69 (2009), 777–785.
Burton, W. N.; A. B. Schultz; C. Y. Chen; and D. W. Edington. “The Association of Worker Productivity
and Mental Health: A Review of the Literature.”International Journal of Workplace Health Man-
agement, 1 (2008), 78–94.
Caggese, A.; V. Cuñat; and D. Metzger. “Firing the Wrong Workers: Financing Constraints and Labor
Misallocation.”Journal of Financial Economics, 133 (2019), 589–607.
Carvalho, D. “Financing Constraints and the Amplification of Aggregate Downturns.”Review of
Financial Studies, 28 (2015), 2463–2501.
Charles, K. K., and P.DeCicca. “Local Labor Market Fluctuations and Health: Is There a Connection and
for Whom?”Journal of Health Economics, 27 (2008), 1532–1550.
Chodorow-Reich, G. “The Employment Effects of Credit Market Disruptions: Firm-Level Evidence
from the 2008–9 Financial Crisis.”Quarterly Journal of Economics, 129 (2014), 1–59.
Deelen, A.; M. de Graaf-Zijl; and W. van den Berge. “Labour Market Effects of Job Displacement for
Prime-Age and Older Workers.”IZA Journal of Labor Economics, 7 (2018), 3.
Desai, M. A.; C. F. Foley; and J. R. Hines, Jr. “A Multinational Perspective on Capital Structure Choice
and Internal Capital Markets.”Journal of Finance, 59 (2004), 2451–2487.
DNB. Quarterly Bulletin March 2009. De Nederlandsche Bank (2009).
Duchi, F., and A. Elbourne. “Credit Supply Shocks in the Netherlands.”Journal of Macroeconomics,
50 (2016), 51–71.
Duijts, S. F. A.; I. Kant; G. M. H. Swaen; P. A. van den Brandt; and M. P. A. Zeegers. “A Meta-Analysis
of Observational Studies Identifies Predictors of Sickness Absence.”Journal of Clinical Epidemi-
ology, 60 (2007), 1105–1115.
Duval, R.; G. H. Hong; and Y. Timmer. “Financial Frictions and the Great Productivity Slowdown.”
Review of Financial Studies, 33 (2020), 475–503.
Flannery, M. J. “Asymmetric Information and Risky Debt Maturity Choice.”Journal of Finance,41
(1986), 19–37.
Ganster, D. C., and C. C. Rosen. “Work Stress and Employee Health: A Multidisciplinary Review.”
Journal of Management, 39 (2013), 1085–1122.
Garcia-Gomez, P.; E. G. Maug; and S. Obernberger. “Private Equity Buyouts and Employee Health.”
European Corporate Governance Institute-Finance Working Paper No. 680 (2020).
Gardarsdottir, H.; E. R. Heerdink; L. van Dijk; and A. C. G. Egberts. “Indications for Antidepressant
Drug Prescribing in General Practice in the Netherlands.”Journal of Affective Disorders, 98 (2007),
109–115.
Giroud, X., and H. M. Mueller. “Firm Leverage, Consumer Demand, and Employment Losses During
the Great Recession.”Quarterly Journal of Economics, 132 (2017), 271–316.
Gopalan, R.; V. Nanda; and A. Seru. “Affiliated Firms and Financial Support: Evidence from Indian
Business Groups.”Journal of Financial Economics, 86 (2007), 759–795.
Green, F. “Unpacking the Misery Multiplier: How Employability Modifies the Impacts of Unemploy-
ment and Job Insecurity on Life Satisfaction and Mental Health.”Journal of Health Economics,
30 (2011), 265–276.
Hadlock, C. J., and J. R. Pierce. “New Evidence on Measuring Financial Constraints: Moving Beyond
the KZ Index.”Review of Financial Studies, 23 (2010), 1909–1940.
Hoevenagel, R., and M. Engelen. “Ontslagkosten van Werkgevers in 2012.”Technical Report (2013).
Huber, K. “Disentangling the Effects of a Banking Crisis: Evidence from German Firms and Counties.”
American Economic Review, 108 (2018), 868–898.
Kalara, N., and L. Zhang. “The Changing Landscape of Firm Financing in Europe, the United States and
Japan (No. 383).”Working Paper. CPB Discussion Papers (2018).
Kaplan, S. N., and L. Zingales. “Do Investment-Cash Flow Sensitivities Provide Useful Measures of
Financing Constraints?”Quarterly Journal of Economics, 112 (1997), 169–215.
Kendler, K. S.; L. M. Karkowski; and C. A. Prescott. “Causal Relationship Between Stressful Life
Events and the Onset of Major Depression.”American Journal of Psychiatry, 156 (1999), 837–841.
Kim, T. J., and O. Von Dem Knesebeck. “Is an Insecure Job Better for Health than Having No Job at All?
A Systematic Review of Studies Investigating the Health-Related Risks of both Job Insecurity and
Unemployment.”BMC Public Health, 15 (2015), 1–9.
Kuhn, A.; R. Lalive; and J. Zweimüller. “The Public Health Costs of Job Loss.”Journal of Health
Economics, 28 (2009), 1099–1115.
Lamont, O.; C. Polk; and J. Saá-Requejo. “Financial Constraints and Stock Returns.”Review of
Financial Studies, 14 (2001), 529–554.
42 Journal of Financial and Quantitative Analysis
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press

Lin, H.; J. D. Ketcham; J. N. Rosenquist; and K. I. Simon. “Financial Distress and Use of Mental Health
Care: Evidence from Antidepressant Prescription Claims.”Economics Letters, 121 (2013), 449–453.
Malhi, G. S., and J. J. Mann. “Depression.”Lancet, 392 (2018), 2299–2312.
Marcus, J. “The Effect of Unemployment on the Mental Health of Spouses - Evidence from Plant
Closures in Germany.”Journal of Health Economics, 32 (2013), 546–558.
McInerney, M.; J. M. Mellor; and L. H. Nicholas. “Recession Depression: Mental Health Effects of the
2008 Stock Market Crash.”Journal of Health Economics, 32 (2013), 1090–1104.
Myers, S. C. “Determinants of Corporate Borrowing.”Journal of Financial Economics, 5 (1977), 147–
175.
Nuijen, J.; M. Emmen; D. Smit; I. Stirbu-Wagner; M. Veerbeek; and P. Verhaak. “GGZ in de eerste lijn:
de rol van de huisarts en de eerstelijnspsycholoog.”Trimbos Instituut. Technical Report (2012).
Ouimet, P., and E. Simintzi. “Wages and Firm Performance: Evidence from the 2008 Financial Crisis.”
Review of Corporate Finance Studies, 10 (2021), 273–305.
Paul, K. I., and K. Moser. “Unemployment Impairs Mental Health: Meta-Analyses.”Journal of Voca-
tional Behavior, 74 (2009), 264–282.
Piek, E.; K. van der Meer; W. J. G. Hoogendijk; B. W. J. H. Penninx; and W. A. Nolen. “Most
Antidepressant Use in Primary Care is Justified; Results of the Netherlands Study of Depression
and Anxiety.”PLOS One 6 (2011), 1–8.
Popov, A. “Credit Constraints and Investment in Human Capital: Training Evidence from Transition
Economies.”Journal of Financial Intermediation, 23 (2014), 76–100.
Popov, A., and J. Rocholl. “Do Credit Shocks Affect Labor Demand? Evidence for Employment and
Wages During the Financial Crisis.”Journal of Financial Intermediation, 36 (2018), 16–27.
Reichert, A. R., and H. Tauchmann. “The Causal Impact of Fear of Unemployment on Psychological
Health.”Unpublished Working Paper, Ruhr Economic Papers (2011).
Salm, M. “Does Job Loss Cause Ill Health?”Health Economics, 18 (2009), 1075–1089.
Schaller, J., and A. H. Stevens. “Short-Run Effects of Job Loss on Health Conditions, Health Insurance,
and Health Care Utilization.”Journal of Health Economics, 43 (2015), 190–203.
Simon, G. E.; C. Stewart; A. Beck; B. Ahmedani; K. J. Coleman; R. Whitebird; F. Lynch; A. A. Owen-
Smith; B. Waitzfelder; S. B. Soumerai; and E. M. Hunkeler. “National Prevalence of Receipt of
Antidepressant Prescriptions by Persons Without a Psychiatric Diagnosis.”Psychiatric Services,
65 (2014), 944–946.
Sonnenberg, C. M.; E. J. M. Biennali; D. J. H. Deeg; H. C. Comijs; W. Van Tilburg; and A. T. F.
Beekman. “Ten-Year Trends in Benzodiazepine Use in the Dutch Population.”Social Psychiatry and
Psychiatric Epidemiology, 47 (2012), 293–301.
Stohs, M. H., and D. C. Mauer. “The Determinants of Corporate Debt Maturity Structure.”Journal of
Business, 69 (1996), 279–312.
Tefft, N. “Insights on Unemployment, Unemployment Insurance, and Mental Health.”Journal of Health
Economics, 30 (2011), 258–264.
Teo, A. R.; H. Choi; and M. Valenstein. “Social Relationships and Depression: Ten-Year Follow-Up
from a Nationally Representative Study.”PLOS One, 8 (2013), e62396.
Tijdens, K.; M. van Klaveren; R. Bispinck; H. Dribbusch, and F. Öz. “Wage and Workforce Adjustments
in the Economic Crisis in Germany and the Netherlands.”European Journal of Industrial Relations,
20 (2014), 165–183.
Van den Berge, W.; H. Erken; M. de Graaf-Zijl; and E. van Loon. “The Dutch Labour Market During the
Great Recession.”CPB Background Document (2014).
Van der Veer, K. J. M., and M. M. Hoeberichts. “The Level Effect of Bank Lending Standards on
Business Lending.”Journal of Banking and Finance, 66 (2016), 79–88.
Van Marwijk, H. W. J.; D. Bijl; H. J. Adèr; and M. De Haan. “Antidepressant Prescription for Depression
in General Practice in the Netherlands.”Pharmacy World and Science, 23 (2001), 46–49.
Whited, T. M., and G. Wu. “Financial Constraints Risk.”Review of Financial Studies, 19 (2006), 531–
559.
Witte, H. De. “Job Insecurity and Psychological Well-Being:Review of the Literature and Exploration of
Some Unresolved Issues.”European Journal of Work and Organizational Psychology, 8 (1999),
155–177.
Kárpáti and Renneboog 43
https://doi.org/10.1017/S0022109023000595 Published online by Cambridge University Press