Employee representation and financial leverageI
Chen Lina , Thomas Schmida,, Yuhai Xuanb
a
University of Hong Kong, Faculty of Business and Economics
b
Harvard Business School
Abstract
German law mandates that firms’ supervisory boards consist of an equal number
of employees’ and owners’ representatives. This law, however, applies only in firms
with over 2,000 domestic employees. Exploiting this discontinuity for identification,
we find that employee power increases financial leverage. We explain this by a supply side effect: as banks’ interests are similar to those of employees, higher employee
power reduces agency conflicts between firms and debt providers. Supportive evidence comes from bank ownership: using the capital gain tax reform as exogenous
shock, we find that equity holdings of banks and employee codetermiation are substitutes. Lower cost of debt, longer debt maturities, smaller loan syndicates, fewer
covenants, and less financial constraints of codetermined firms also point towards
an alignment effect. Analyzing M&A decisions, cash flow and profit stability, and
idiosyncratic risk reveals lower firm risk as a possible channel for this alignment.
Keywords: Capital structure, financial leverage, employee representation, labor
rights, bank ownership, tax reform
JEL: G30, G32
I
We thank Tarun Chordia, Craig Doige, Andrew Ellul, Jarrad Harford, Macin Kacperczyk,
Ross Levine, Kai Li, Randall Morck, Paige Ouimet, Francisco Perez-Gonzalez, Sheridan Titman,
Neng Wang, and participants of the 2015 WFA conference for helpful comments. Please send
correspondence to Thomas Schmid, E-Mail: [email protected].
August 5, 2015
1. Introduction
Modern corporate finance often assumes a limited influence of employees on firm
policy. At most, an indirect influence via labor unions which can, for instance, organize strikes to enforce their interests is considered. In this paper, we exploit a
unique setting that grants employees a direct voice in the firm: the German law on
codetermination. This law mandates that the supervisory board of a firm, which
is similar to the board of directors, has to consist of an equal number of owners’
and employees’ representatives. We refer to this as parity employee representation
(PER).1 This, however, only applies if the firm has more than 2,000 domestic employees. The discontinuity around this threshold allows us to identify the effect of
employee power on financial leverage.
Due to their presence on corporate boards, employee representatives are well
informed and able to exert direct pressure on executive managers. We hypothesize
that their direct voice reduces agency conflicts between banks and firm’s owners
and managers. Although employee representatives aim to protect the interests of
the firm’s employees in the first place, they may unintentionally also represent the
interests of banks. The reason for this is that both employees and debt providers
have a strong interest in the long-term survival and stability of the firm. If this
hypothesis holds true, we expect firms with PER to have higher financial leverage
and more favorable financing conditions.
A graphical analysis provides first evidence for a positive jump in financial leverage around the threshold of 2,000 domestic employees. Our main empirical method
is a regression discontinuity design around this threshold. In particular, we compare
firms that are slighting above the threshold with those slightly below. We assume
that these firms are similar in all dimensions except their level of employee representation. The regression discontinuity approach confirms that firms with PER have a
higher financial leverage if compared to firms without strong employee power. Both
the statistical and economic significance of this result are strong. In particular, we
find that the financial leverage is about 10% higher in firms with PER.
Three potential threats to our identification strategy arise. First, firms may try
to strategically stay below the threshold of 2,000 domestic employees. However, an
analysis of the distribution of domestic employees in our sample provides no indication for any discontinuity around this threshold. Second, firms may need some time
to adjust their boards when growing above or shrinking below this threshold. Explicitly considering this timing lags, however, leads to even stronger results. Third,
1
An alternative term for parity employee representation is parity codetermination. We use both
terms interchangeably in this paper.
1
the results may be related to an increase in firm size around the threshold. Although
we consider this as unlikely as we control for the number of domestic employees up to
polynomial four on both sides of the threshold, we also perform a placebo test with
randomly chosen thresholds. These tests show that financial leverage only increases
around the threshold of 2,000 domestic employees which triggers PER.
To investigate whether this finding is indeed related to an interest alignment
effect between debt providers and firms, we perform several additional tests. The
first test focuses on a possible substitute for PER which also allows banks to directly
enforce their interests: bank ownership. As ownership itself is very likely endogenous, we exploit the capital gain tax reform in 2000 as an exogenous shock which
reduced equity holdings of German banks. Our results indicate that employee power
and bank ownership are substitutes because the impact of PER increased after the
tax reform. This effect is especially pronounced for firms which had bank ownership
beforehand. Second, we analyze cost of debt and find—in line with the view that
employee power reduces agency conflicts between firms and banks—that firms with
higher employee power have significantly lower loan spreads. Furthermore, these
firms have longer debt maturities, smaller lender syndicates, and fewer covenants.
We also provide evidence that parity codetermiation relaxes financial constraints,
i.e., it decreases firms’ cash flow sensitivity of cash.
A possible channel for this interest alignment effect is firm risk. To shed light on
this possible channel, we start by analyzing firms’ investment decisions, i.e., their
M&A deals and capital expenditures. In line with our expectations, we find that
codetermined firms are less likely to conduct M&A deals. Furthermore, they tend
to focus on value-increasing deals if they engage in M&A activities. We also present
evidence for lower capital expenditures in firms with more employee power. After
that, we demonstrate that cash flows and profits of these firms are less volatile.
Lastly, firms with PER also have lower idiosyncratic risk. Thus, higher stability of
codetermined firms provides a possible explanation for their interest alignment with
debt providers.
The remainder of this paper is structured as follows. The legal situation in
Germany, our empirical strategy, and theoretical consideration are summarized in
Section 2. The dataset is presented in Section 3. In Section 4, we show the results for the relationship between PER and financial leverage. We also investigate
possible identification threats and the general robustness of the results. We shed
light on interest alignment between banks and firms with PER in Section 5 and
possible channels for such alignment in Section 6. Finally, Section 7 concludes by
summarizing the findings and their implications.
2
2. Setting
2.1. Parity employee representation in Germany
The regulation of labor differs across countries (Botero et al., 2004). In this
paper, we focus on Germany as a country with a rather stringent labor regulation.
In general, public companies in this country have a two-tier board system with a
management and supervisory board (“Vorstand” and “Aufsichtsrat”). The former
consists of executive managers which are responsible for the daily business. The
members of the supervisory board have, among others, the duty to supervise the
executive managers (§111, 1 AktG). Furthermore, they elect the members of the
management board for at most five years, with the possibility of re-election (§84, 1).
Overall, the supervisory board has a similar role as the board of directors.
Employee representation in corporate boards of German firms is regulated by
the law on codetermination (“Mitbestimmungsgesetz, MitBestG) which came into
force in 1976. This law applies for public and private firms with more than 2,000
domestic employees (§1).2 For the calculation of the number of domestic employees,
all firms within one group of companies, i.e., those of the parent company and
all subsidiaries are considered (§5, 1). For firms which exceed this threshold, the
law mandates that the supervisory has to consist of an equal number of owners’
and employees’ representatives (§7).3 We refer to this equality as parity employee
representation (PER) or parity employee codetermination in this paper.4
Several other paper also focus on the effects of codetermination in Germany.
One strand of this literature argues that codetermination has no or a negative impact on firm valuation and performance (e.g., Benelli et al., 1987; FitzRoy and
Kraft, 1993; Gorton and Schmid, 2004). Gurdon and Rai (1990), however, conclude
that productivity declined for firms with codetermination, but their profitability increased. Evidence for an inverse U-shaped relation between firm value and employee
representation is reported by Fauver and Fuerst (2006). Renaud (2007), who also
2
Some firms are exempt from the law (§1, 4). In our dataset, this is only relevant for firms
which focus on news reporting, e.g., publishing companies, which we drop from the sample (see
Section 3.1).
3
The chairman of the board is elected by the owners’ representatives, whereas the employee
representatives elect the vice-chairman (§27). As the chairman has two votes in case of a tie,
owners have slightly more power than employees (§29).
4
Companies with less than 2,000, but more than 500 domestic employees are regulated by the onethird codetermination law (“Drittelbeteiligungsgestz”, DrittelbG). This law states that one-third of
the supervisory board has to consist of employee representatives. However, the power of this law
is much weaker if compared to the MitBestG due to several reasons: (i) one-third codetermination
can more easily be ignored by owners representatives than parity codetermination, (ii) the law does
not apply for some legal forms, (iii) employees of other firms in the group of companies are not
necessarily considered (§2, 2 DrittelbG). The last two aspects allow firms to strategically avoid onethird codetermination. These possibilities do not exist for parity codetermination (Rieble, 2006).
3
provides an overview on other empirical studies which exploit the German codetermination, finds that codetermination increases both productivity and profitability.
With regard to corporate policy decisions, Kraft et al. (2011) finds no evidence that
codetermination affects innovation activities. Kim et al. (2014) show that higher
employee power protects skilled workers against layoffs during industry shocks.5
2.2. Empirical strategy
Our empirical strategy is based on a regression discontinuity design. Such designs
have been frequently applied in empirical finance research (e.g. Chava and Roberts,
2008; Bakke and Whited, 2012; Cuñat et al., 2012; Natividad, 2013). A detailed discussion of this methodology is, among others, provided by Lee and Lemieux (2010).
In this paper, we exploit for identification that German firms have to establish parity codetermination by law as soon as the number of domestic employees exceeds
2,000. This setting ensures that the establishment of parity codetermination is not a
voluntary decision of the firms’ owners or manager.6 Rather, German law mandates
that:

1 if DEi,t > 2000
PERi,t =
0 if DE ≤ 2000
i,t
where i indicates firms and t years. Methodologically, we follow a parametric
strategy and limit the sample to firms with around 2,000 domestic employees for
most of our tests. In particular, we use a range of 1,000 to 3,000 employees for our
empirical tests. The choice of this range is a tradeoff between accuracy and number
of observations. This range results in about 160 firms and 700 firm-years in our main
models. The fact that we have only few observations which are exactly at the cutoff point of 2,000 domestic employees is also the reason why we do not apply local,
non-parametric estimation methodologies. Instead, we estimate the causal effect of
5
Valuation and performance consequences of labor power have also been studied outside Germany. For example, Faleye et al. (2006) investigate various dimensions and find, among others,
that labor-controlled firms have lower firm value and less risk taking. Atanassov and Kim (2009)
document a dark side of employee power as strong union laws protect not only workers but also
under-performing managers. Ginglinger et al. (2011), however, find that employee directors increase
firm valuation and corporate performance. With regard to firm policy, Bradley et al. (2013) find
that unionization leads to a decline in patent quantity and quality, whereas Acharya et al. (2014)
reports that wrongful discharge laws spur innovation. Chyz et al. (2013) provide evidence for a
negative relationship between firms’ tax aggressiveness and union power and Chen et al. (2014) report that firms with stronger labor power repurchase fewer shares. Another strand of the literature
focuses more generally on the relationship between labor regulation and economic indicators (e.g.,
Botero et al., 2004; Besley and Burgess, 2004; Propper and Van Reenen, 2010).
6
Firms with less than 2,000 domestic employees could voluntarily establish parity codetermination. This is, however, very unlikely as it would require that the firms’ owners voluntarily reduce
their power in the supervisory board.
4
parity employee representation on leverage with the following empirical model:
−→
Levi,t = α + β ∗ PERi,t−1 + ν∗ X i,t−1 +
4
P
p=1
γp ∗ Api,t−1 +
4
P
p=1
δp ∗ Api,t−1 ∗ T0|1 + i,t
Ai,t = DEi,t − 2000
where Levi,t is the leverage of firm i in year t, PER is a dummy which equals one
for parity employee representation, X is a vector of standard control variables (i.e.,
firm-specific factors like size, industry, and year dummies), and A is the assignment
variable. In our case, this is the number of domestic employees minus 2,000, i.e.,
we center it at the threshold. We include the assignment variable as polynomials
of degree four. This functional form controls for any direct impact of domestic
employees on leverage around the threshold. Lastly, we interact the assignment
variable with a dummy T which equals one if the number of domestic employees
exceeds 2,000 to control for a potentially different effect of the number of domestic
employees on leverage on both sides of the threshold. The coefficient which we are
mainly interested in is β. This captures the discontinuous effect of parity employee
representation on leverage at the threshold.
To summarize, we focus on firms with between 1,000 and 3,000 domestic employees (“bandwidth”). We include the centered number of domestic employees up
to polynomial four as controls (“functional form”). We also allow for a different
functional form on both sides of the threshold. This functional form captures the
continuous effect of the number of domestic employees (“assignment variable”) on the
outcome variable leverage. The discontinuous effect (i.e., the jump) at the threshold
is thus captured by the coefficient for PER. Lastly, we winsorize the variables at the
1 and 99% level to restrict the impact of outliers.
2.3. Theoretical considerations
Prior literature on financial leverage and employee power mainly argues that
firms use debt as strategic bargaining tool. For example, they may increase leverage
before negotiations with employees to improve their bargaining position. Theoretical
and empirical evidence for this view is provided by Bronars and Deere (1991), Perotti
and Spier (1993), and Matsa (2010). Similarly, Klasa et al. (2009) show that firms in
highly unionized industries hold less cash to improve their bargaining position and
Benmelech et al. (2012) finds evidence that firms use the threat of pension reduction
in negotiations with labor.7
7
Another related literature strand focuses on workers’ unemployment risk. For example, Agrawal
and Matsa (2013) show that firms choose conservative financial policies, e.g., low leverage, also to
mitigate workers’ exposure to unemployment risk. Thus, higher unemployment benefits increase
financial leverage. By contrast, Simintzi et al. (2014) conduct a cross-country study and document
5
These papers focus mainly on an indirect influence of labor on firm policy, e.g.,
via industry-level unionization. By contrast, we investigate a direct influence of
employees on corporate decision due to their representation in the firm’s supervisory
board. In this setting, the use of debt as strategic bargaining tool is unlikely due
to three reasons. First, employee representatives in the supervisory board have
the duty to monitor executive managers (§111, 1 AktG). If debt levels are high
and employees are concerned about their jobs, their representatives are more likely
to monitor executive managers tightly. Thus, executive managers have incentives
to choose conservative debt levels to escape tight monitoring. Second, employee
representatives can threaten executive managers with not re-electing them if they
disagree with the firm’s financial policy (§84, 1). Again, they are likely to prefer
conservative levels of debt to increase job safety. Third, the representatives of the
employees possess insider knowledge about the firm’s need for debt because they
have the right to inspect the firm’s internal documents (§111, 2). Thus, above
optimal debt levels in order to improve managers’ or owners’ bargaining position
are difficult to justify for them.
Rather, we hypothesize that another effect plays a major role in this setting:
reduced agency conflicts between the firm’s owners and managers on the one side
and banks on the other. While managers and owners have incentives for myopic
and/or selfish behavior, both banks and employee representatives are interested in
the long-term survival and stability of the firm to secure jobs and debt repayments.
Fauver and Fuerst (2006) argue in this context that “labor representation introduces
a highly informed monitor to the board that reduces managerial agency costs (such
as shirking, perk-taking, and excessive salaries) and private benefits of blockholder
control” (p. 680). Thus, labor representatives who aim to protect the interests of the
firm’s employees may (unintentionally) also represent the interests of banks. Due to
this interest alignment effect between banks and firms with employee representatives,
we expect that parity codetermination enables firms to access bank debt more easily
and at more favorable conditions. Among others, this may lead to higher debt ratios,
lower interest rates, and less financial constraints.
3. Data
3.1. Sample construction
Our dataset contains private and public German firms between 2005 and 2013.
The sample construction is based on data from Hoppenstedt GmbH, a commercial
that higher employment protection reduces leverage. Furthermore, Chen et al. (2012) find that
firms in more unionized industries have lower bond yields.
6
provider of business information for German firms. The final dataset is constructed
in several steps. First, we start with all medium-sized and large firms for which
consolidated financial statements are available in the Hoppenstedt database. We
identify a firm’s legal form and SIC code and subsequently exclude non-profit companies and firms from the financial services industry (SIC 6000-6999). In a second
step, ownership data are received directly from Hoppenstedt. As corporate policy
decisions are made on the holding company level and any decision-making on the
subsidiary level is not independent of the overall corporate strategy, we eliminate
companies that are subsidiaries of a domestic or foreign business group and companies for which ownership data is not available. Furthermore, we do not consider
firms if federal entities own more than 75% of voting rights. In a third step, we
construct the final panel dataset which includes data from firms’ balance sheets and
profit and loss statements, which is also obtained from Hoppenstedt. Overall, this
results in more than 10,000 firm-year observations.
Our identification strategy is based on the discontinuity around the threshold
of 2,000 domestic employees. Thus, our main focus lies on firms which are “close”
to this threshold. As explained in Section 2.2, most or our tests are conducted for
firms with between 1,000 and 3,000 domestic employees. For the construction of this
sub-sample, we have to identify the number of domestic employees for each firm-year
in the sample. Although there is a data field for the number of total employees in
the Hoppenstedt database, this information is missing for a considerable number
of firm-years. Thus, we manually complement this information for firms which are
likely within this range.8 After that, we manually check whether firms have parity
employee representation for this sub-sample and construct the dummy variable per.
For firms with more than 3,000 (less than 1,000) domestic employees, we assume
that per equals one (zero).
Next, we drop several observations from our sample. First, we eliminate firmyears in which a firm newly introduces or abolishes parity employee representation
(17 observations). After that, firms which focus on news reporting, e.g., publishing
companies, are dropped because they are exempt from parity employee representation (cf. §1, 4 MitBestG). The final sample for which we have data on domestic
employees contains 2,338 firm-years from 398 different firms. If we focus on our main
sample with firms around the threshold, i.e., with between 1,000 and 3,000 domestic
employees, we end up with 817 firm-years from 164 firms.
8
In particular, we search for this number in the firms’ annual reports for all firms with more than
1,000 and less than 6,000 total employees (because a fraction of less than 50% domestic employees
is very uncommon) if the information is missing. Firms’ annual reports are obtained from their
websites and from the Hoppenstedt database, which also includes reports from firms which do not
exist anymore.
7
3.2. Main variables
Data for the construction of our leverage measure and the control variables also
comes from Hoppenstedt. Due to the fact that our sample also includes private
firms, we apply a leverage measure based on book values of debt and equity. Thus,
leverage is defined as total debt divided by total debt plus book value of equity.
Total debt includes current and long-term liabilities and excludes provisions and
accruals. Several control variables are applied in our analysis. Their inclusion is
motivated by Frank and Goyal (2009). size is the natural logarithm of a firm’s
total assets. tangibility is long-term tangible assets divided by total assets. roa
measures profitability is is calculated as EBIT divided by total assets. As proxy
for growth opportunities, we include Tobin’s Q (tobq). As our sample also covers
private firms, we use the mean of this variable in an industry and year. Furthermore,
we also control for listing. This dummy equals one for public firms and zero for
private companies.9 In the German environment, private firms can (mostly) choose
to apply local or international GAAP. Thus, we include accounting standard
which equals one if a firm applies international accounting standards and zero otherwise. Year as well as industry-fixed effects based on the Fama/French 17 industries
classification are included in all regressions. As explained in Section 2.2, we also include the centered number of domestic employees as well as higher orders on both
sides of the cut-off point. The detailed definitions of all relevant variables as well as
their sources are summarized in Appendix A.
Descriptive statistics for our main sample with 1,000 to 3,000 domestic employees can be found in Table 1. The average firm in our sample has 1,785 domestic
employees, a leverage ratio of 0.53, and 744 million e total assets. Furthermore,
it has about 30% tangible assets and an EBIT to total assets ratio of 8%. About
one-fourth of all firms have parity codetermination and one-third are stock market
listed. Regarding accounting standards, about half of the firms follow international
guidelines. The average Tobin’s Q in our sample is 1.3.
— Table 1 about here —
4. Leverage
4.1. Parity employee representation and financial leverage
Before we investigate the relationship between parity employee representation
(PER) and leverage with a regression discontinuity design, we focus on a graphical
9
We define a firm as public if its shares are listed on a regulated European market or on an
exchange regulated market. All German stock exchanges are taken into consideration for the definition of a firm’s stock market listing status. The construction of this variable is based on data
from Hoppenstedt’s database, ad-hoc releases, and press reports.
8
evaluation. In Figure 1 we first display the mean financial leverage for bins of 50
employees between 1,000 to 3,000 domestic employees.10 As PER is mandated by
law for firms with more than 2,000 domestic employees, we expect a discontinuity
at this threshold. Indeed, we find an increase in leverage if the number of domestic
employees exceeds this threshold if we use quadratic (a) or linear fitting (b). We
observe a similar pattern if we use bins of 100 instead of 50 employees in (c) and (d).
Thus, these graphical evaluations provide first indication that PER leads to higher
financial leverage ratios.
— Figure 1 about here —
The main regression discontinuity analysis is presented in Table 2. As explained
in Section 2.2, we exploit that PER is required by law in firms with more than
2,000 domestic employees for identification. Thus, we include an indicator variable
for PER and use our main sample with firms between 1,000 and 3,000 domestic
employees. We indicate in each column of the table if and how we control for
the number of domestic employees. We start without any controls for domestic
employees in model I. After that, we subsequently add more controls. For example,
we include the centered number of domestic employees, the squared centered number,
and both variables interacted with a dummy for 2,000 domestic employees in model
IV. Our main model with polynomials up to order four on both sides of the threshold
is presented in model V. In all models, we find strong and consistent evidence that
PER increases financial leverage. The magnitude of this effect is above 10% percent
in absolute terms. Given the sample mean leverage of 0.5, this represents a relative
difference of about 20%. Thus, the positive impact of PER on financial leverage is
not only statistically significant, but also economically relevant.
For the control variables, we find—in line with Frank and Goyal (2009)—a negative sign for profitability. The other control variables are mainly insignificant, which
may be explained by the fact that we use a rather narrow sub-sample. A more
detailed discussion of the control variables is provided in Section 4.3.
— Table 2 about here —
4.2. Identification threats
Three concerns may arise with our identification strategy. First, firms could
strategically stay below the threshold of 2,000 domestic employees to avoid PER.
10
We exclude firm-years with more than 2,000 domestic employees but no parity employee representation and those with less than 2,000 but parity employee representation. The reasons for these
law deviations are in nearly all cases timing delays. A more detailed discussion of this aspect is
provided in Section 4.2.
9
This, however, would imply that both the firm’s owners and managers are willing
to forgo future growth. We argue that this is not very plausible. Rather, owners
and managers are well aware that they cannot avoid codetermination if they want
the firm to grow. Thus, there is no reason to strategically reduce firm growth just
to postpone PER. Nevertheless, we perform a graphical evaluation to investigate
whether there is a discontinuity in the distribution of domestic employees around
2,000. If firms stay below the threshold to avoid PER, we would expect to see a
disproportionally high number of firms just below the threshold and a low number
thereafter. The results are presented in Figures 2. Not surprisingly, the number of
firm-years slightly declines as domestic employees and thus firm size increases. We
find, however, no indication of any discontinuity around 2,000 domestic employees.
A kernel density estimate (green line) also provides no indication. Thus, we argue
that it is very unlikely that firms strategically stay below this threshold to avoid
PER.
— Figure 2 about here —
A second potential concern is related to timing issues. If a firm grows above
2,000 domestic employees, it may take some time before its supervisory board is
newly constituted to comply with PER. The same applies if a firm shrinks below
the threshold. To investigate this, we analyze how many percent of all firms have
PER dependent on the number of domestic employees. The results is shown in figure
3. As expected, there is a small number of firm-years with less than 2,000 domestic
employees but PER. Similarly, we find some observations above the threshold without PER. However, for the vast majority of firms the composition of the supervisory
board complies with the law on codetermination. To investigate if these rare deviations bias our findings, we re-run our regressions and exclude those cases where
a firm with more than 2,000 domestic employees has no PER (36 observations) or
a firm without 2,000 domestic employees has PER (14 observations). The result is
shown in Table 3, model I. As before, we find strong evidence that PER leads to
higher leverage. The coefficient for PER is even higher as before (0.20 vs. 0.15).
Thus, we conclude that timing issues are unlikely to bias our results.
— Figure 3 about here —
The third potential concerns is that our prior results capture a size effect due
to an increase in the number of (domestic) employees. Although we consider this
as unlikely as we include various controls for the number of domestic employees,
we also perform a placebo test. If prior results were not related to PER, but to a
general effect due to an increase in the number of domestic employees, we would
10
expect to find similar outcomes around other thresholds which have nothing to do
with PER as well. We start with 1,500 and 2,500 domestic employees as such
alternative thresholds for the placebo test in model II of Table 3 and restrict the
sample to firm-years with threshold +/- 1,000 domestic employees. We find no
indication that leverage changes around these two thresholds. We also perform a
placebo test with a threshold of 2,000 total employees. As explained before, the
German law on codetermination focuses only on the number of domestic, not total
employees. Consequently, we find no evidence that leverage changes around this
threshold. Thus, this placebo test provides further evidence that our prior result
is indeed related to PER which is triggered by the threshold of 2,000 domestic
employees.
— Table 3 about here —
4.3. Robustness tests
The choice of 1,000 and 3,000 as lower and upper bounds for our main sample
is to some extent arbitrary. To reduce concerns that their specification biases or
drives our results, we also present the outcome for different thresholds. Results are
shown in Table 4. We first consider firms with between 1,700 and 2,300 domestic
employees. After that, we extend the bounds to 1,400 and 2,600. In the third model,
we use an asymmetric range of 1,400 to 3,000 to account for the fact the number of
firms decreases with domestic employees. In the last model, we impose no specific
restrictions and include all sample observations. Overall, the results for all windows
confirm our prior findings. The coefficient is similar as before and ranges from 0.11
to 0.19; the statistical significance is 1% for all windows. Thus, we argue that the
choice of specific thresholds does not drive our findings.
— Table 4 about here —
As more general robustness test, we first adjust the way how we control for industry effects. In particular, we include industry-year fixed effects instead of industry
and year fixed effects; after that, we apply the more fine grained Fama/French 38
industries classification (instead of 17 industries). Results in model III of Table Appendix B show that neither of these two variations has a significant impact on the
outcome. We also restrict the sample to listed firms in model IV. As can be seen,
the PER effect also exists in this sub-sample. We thus argue that our approach
to consider also private firms, which are less frequently investigated in empirical
research, does not bias our results.
Lastly, we investigate whether our results depend on the database which we
use to construct our leverage measure and the control variables. As explained in
11
Section 3.2, we mainly rely on data from the Hoppenstedt database, which provides
data on balance sheets and profit and loss statements of public and private German
firms. Both coverage and quality of the data are guaranteed as Hoppenstedt relies
on several sources, e.g., direct contact with the companies and the electronic Federal
Gazette where firms have to file their annual reports (§325 HGB). Nevertheless, this
database is not common in empirical research. Thus, we perform several tests to
ensure that our results are not biased by using this database. In particular, we first
investigate whether the control variables are similar as in other studies; after that,
we replace Hoppenstedt data with data from Worldscope. In all models, we focus on
listed firms as this is more common in empirical capital structure research. Results
are shown in Table Appendix C. In model I, we include all firm-years, independent
of the number of domestic employees, to see whether the control variables are in line
with prior literature. We find positive and significant effects for size and tangibility;
the opposite is true for profitability and Tobin’s Q. All these findings are in line
with Frank and Goyal (2009). Next, we use data from the Worldscope database
which only includes information on public firms to measure leverage (model IIa)
or to construct all variables (IIb). The magnitude and significance of PER is very
similar as if we use Hoppenstedt data for public firms (model II, Table Appendix B).
As last variation, we analyze market leverage based on Worldscope data in model
III. Again, we find similar results for parity codetermination. Overall, we conclude
that our results are not biased by using the Hoppenstedt database.
5. Interest alignment
We hypothesize that our results can be explained by an interest alignment effect
between debt providers and firms with higher employee power. To shed light on
this, we now analyze how bank ownership affects the impact of PER on leverage.
After that, we investigate whether parity codetermination has an impact on firms’
interest rates, loan characteristics, and financial constraints. As before, we use a
regression discontinuity design around the threshold of 2,000 domestic employees.
5.1. Bank ownership
If our results can be explained by an interest alignment effect we expect that
PER is mainly important in firms in which banks have no direct voice. If banks have
other channels to directly influence firm policy, PER should be of minor importance.
One such channel, which is not uncommon in the German environment, is bank
ownership (Gorton and Schmid, 2000).11 Equity stakes enable banks to affect firm
11
Another channel which banks may use to exercise power during annual shareholder meetings
is proxy voting. Germany has a long-lasting tradition that small individual investors transfer their
12
policy more effectively than being just debt provider, e.g., via a seat as firm owner
in the supervisory board. Consequently, we hypothesize that the effect of PER is
reduced if a bank holds equity of the firm. The empirical challenge, however, is
that bank ownership is likely endogenously determined. Thus, simply analyzing
actual bank ownership may lead to biased results. Our main empirical strategy
to circumvent this problem is based on an exogenous event which reduced bank
ownership in Germany: the capital gain tax reform in 2000.
5.1.1. Capital gain tax reform
The first announcement of plans for a capital gain tax reform took place shortly
before Christmas 1999. The basic idea was to abolish the capital gain taxes if
corporates sell their equity stakes of other other corporates. In Germany, banks and
also insurance firms frequently held equity stakes in industrial firms. The origin
of these holdings often went back to the time after World War II (Edwards et al.,
2004). The announcement of plans to reduce capital gain taxes led to a massive
increase in stock prices of financial service companies. For instance, Edwards et al.
report that stock prices climbed 18% for Munich Re, 13.6% for Deutsche Bank,
and 12.9% for Allianz. The reduction of equity holdings of banks and insurance
companies, which were commonly regarded as the centerpiece of cross-links between
German cooperation (also called “Deutschland AG”) was one of the aims of the
tax reform.12 The law came into force on January 1, 2001; the capital gain tax,
however, was not abolished before January 1, 2002. After this date, German banks
and insurance companies could sell their equity stakes without paying taxes. As
capital gain taxes could be close to 50% of the value of the equity holding before
the reform, this represented a fundamental change and led to a huge wave of equity
divestitures (Von Beschwitz and Foos, 2013). We exploit this event as an exogenous
voting rights to the banks in which their shares are deposited. This channel, however, is less relevant
in our setting. First, our sample covers private and small to medium size public firms with rather
low free float (on average, the free float in our sample is below 20%). Proxy voting is mainly relevant
for large listed firms with high free float and no blockholders. Second, corporate governance reforms
before our sample period (e.g., the KonTraG in 1998) imposed restrictions for proxy voting and
made this less attractive and more costly for banks. In fact, many banks even stopped offering this
service to their customers (Winkler, 2006, p. 179). Third, the law requires banks to vote in the
best interest of stock owners. Fourth, it is not clear whether banks use proxy voting to enforce their
interest in practice. Gorton and Schmid (2000) find “no evidence that banks use proxy voting to
further their own private interests or, indeed, that proxy voting is used at all” (p. 70). They also
provide a more detailed overview on proxy voting in Germany.
12
From February until May 2000, the German Parliament discussed the draft law. Although the
opposition was against the draft, the majority of the governing parties led to an acceptance. Before
coming into force, the German constitution requires that not only the Parliament, but also the
Federal Council has to approve the law. Due to the fact that the Federal Council was dominated
by the opposition parties, the law was rejected in June 2000. After several adjustments, the Federal
Council finally approved the law on July 14, 2000.
13
shock on ownership of German banks in industrial firms.
5.1.2. Data
Because the event took place on January 1, 2002, we cannot use our main sample
which starts in 2005. Rather, we construct a second sample which covers 1996
to 1998 as pre-event period and 2002 to 2004 as post-event period. We do not
consider years between 1999 and 2001 as the tax reform was discussed but not yet
in force in this period. As data availability and quality for private firms in this
period is very poor, we consider only firms which were stock market listed. In
particular, our sample consists of firms which have been in the CDAX (the broadest
German stock index) for at least three years during our sample period. We then
exclude financial service companies and firms focusing on news publishing. For
the remaining firms, we retrieve ownership information from the yearly versions
of the Hoppenstedt Aktienfürer (either on CD-ROM or as book) and accounting
data from Worldscope. Unfortunately, there is no data item for the number of
domestic employees. Thus, we collect data on domestic employees and employee
representation manually from the firms’ annual reports (which are obtained from
the firms’ websites and the filings database of ThomsonReuters). Overall, we find
information on the number of domestic employees and employee representation for
565 firm-years from 125 firms.
5.1.3. Results
Our empirical strategy exploits that equity holdings of German banks decreased
after the capital gains tax reform. We find that the mean ownership of German
banks13 dropped significantly from about 3.5% to only 2% between the pre- and
post-event period (see Figure 4). To analyze the interplay between codetermination
and bank ownership, we interact PER (as of before the reform) with an indicator
variable for the post-event period. If PER and bank ownership were substitutes, we
expect that the influence of PER increases after the tax reform.
— Figure 4 about here —
Empirical results are presented in Table 5. We focus again on firms with between
1,000 and 3,000 domestic employees. We start with a OLS regression in model Ia.
As in our main models, we include the centered number of domestic employees up to
polynomial four on both sides of the threshold as controls. As expected, we find that
the impact of PER increased after the tax reform. As an alternative specification,
we also perform a firm-fixed effects regression in model Ib. In contrast to our main
13
We only consider German banks as foreign banks were not affected by the tax reform.
14
models, in which per is rather constant over time, we are now able to exploit timeseries variation due to the tax reform. Please note that PER is omitted in this
model because is measures before the reform and thus time constant. The results
are similar as for the OLS model. As further variation, we also interact the post
event indicator variable with the controls for the number of domestic employees in
model Ic. Again, the results point in the same direction. Thus, all specifications
clearly show that the impact of PER increased after the tax reform.
In a second step, we also take pre-event bank ownership into account. The
increased importance of PER after the reform should be especially pronounced for
firms which had a bank as owner in the pre-event period. Thus, we include a
triple interaction term between a dummy indicating if the firm had a bank as owner
before the reform, PER, and a post-event indicator variable. As the number of
firms which have bank ownership is limited, we consider all sample firms for this
test, i.e., also firms for which we do not know the number of domestic employees.
Consequently, we cannot control for the number of domestic employees in these
models. We start with a pooled OLS regression in model IIa and find that the
impact of PER increased especially for those firms that had bank ownership before
the reform. Alternatively, we again apply a firm-fixed effects design in model IIb in
which all time-constant variables are omitted; the results, however, are similar as
for the OLS model. Overall, interest alignment due to PER seems to be of reduced
importance whenever banks have a direct voice within the firm through their equity
stakes. This provides evidence for a substitution effect between bank ownership and
PER with regard to interest alignment.
— Table 5 about here —
5.2. Interest rate
If there is an interest alignment between banks and firms with higher employee
power, we further expect more favorable financing conditions for those firms. Thus
we analyze firms’ cost of debt and expect lower loan spreads for codetermined firms.
As empirical proxy for the cost of debt we use data on syndicated loans from
Dealscan.14 Unfortunately, the number of firms from our main sample for which
we find loan spreads is rather low. Thus, we do not impose any restrictions on the
number of domestic employees for this test. This results in about 253 loans from
61 different firms. Additionally to the variables in the leverage models, we include
controls for leverage and default risk. The latter is approximated by the Z-score.
All firm-specific independent variables are lagged one year (i.e., they refer to the
14
For this, we manually match the sample firms to Dealscan by firm name. For cost of debt we
focus on the all-in-drawn spread (spread).
15
last available annual report before the loan). Furthermore, we also control for deal
purpose and amount. Results are presented in Table 6. Due to the rather small
number of observations, we first control for industry and year effects by including
the industry-year mean spread instead of using industry and year fixed effects. In
the further models, we include year-fixed effects, industry-fixed effects, and then
both. All models provide strong and consistent evidence that employee representation reduces cost of debt. The magnitude of cost of debt reduction is around 1.5%.
Thus, the impact of PER on cost of debt is not only statistically significant, but also
economically relevant.
— Table 6 about here —
5.3. Loan characteristics
Besides the spread, we also investigate other loan-level information from Dealscan.
We focus on debt maturity, the number of lenders in a syndicate, and covenants. In
all regressions, we control for the deal purpose and amount. Based on the interest
alignment hypothesis, we first expect that PER increases debt maturity. Results
are shown in Table 7 support this view. With regard to the number of lenders,
we expect that risk sharing is less important if agency conflicts are lower. Indeed
we find that syndicates are on average smaller for firms with PER. With regard to
covenants, we find that the number of covenants is on average lower for firms with
parity employee representation. Overall, all these findings indicate that firms with
PER have more favorable financing conditions and provide support for the view
that higher employee power increases interest alignment between borrowing firms
and banks.
— Table 7 about here —
5.4. Cash flow sensitivity of cash
Based on the interest alignment hypothesis, we expect less financial constraints
for firms with PER. As empirical proxy for financial constraints we apply firms’
cash flow sensitivity of cash (Almeida et al., 2004). We expect a more pronounced
sensitivity for financially constraint firms. Thus, parity employee representation
should reduce firms’ cash flow sensitivity of cash. Empirical results are reported
in Table 8. We start with sub-sample regressions for firms with (model Ia) and
without (Ib) codetermination. We use the simple model of Almeida et al., i.e., we
control only for firm size and growth opportunities. As we can not use market-based
measures for the latter, we use the actual sales growth between years t-1 and t+1 as
proxy for growth opportunities. In line with our expectation, we find that changes
in cash holdings are not related to cash flow for firms with PER. For other firms, we
16
find a strong positive correlation. If we combine both sub-samples in model IIa, this
outcome is confirmed. The positive impact of cash flow on changes in cash holdings
drops to about zero for firms with strong employee rights. The results are even
more pronounced if we exclude observations with negative cash flow in model IIb.
Similar to the extended model of Almeida et al., we add ∆ short term debt, ∆ net
working capital, capex, and a dummy for M&A deals as further controls in model
IIIa. Lastly, we additionally include interactions between our controls for domestic
employees and cash flow. Overall, we find consistent evidence that PER reduces
firms’ cash flow sensitivity of cash.
— Table 8 about here —
6. Channels
In this section we investigate firm risk as possible channel for the interest alignment between banks and firms with PER. Risk averse employees may have strong
incentives to reduce firm risk due to their firm-specific human capital (Gorton and
Schmid, 2000). If stronger employee power reduces firm risk, this would provide an
intuitive explanation for lower agency conflicts between these firms and banks.
6.1. Investments
In general, M&A can increase or decrease firm-value. Debt providers, however,
may dislike M&A deals as they often increase firm complexity and managers’ and
owners’ possibilities for self-utility maximizing behavior (e.g., empire building). Furthermore, M&A deals can increase firm risk and reduce stability as the long-term
success of such transactions is often highly insecure. We collect data on M&A deals
for each sample year from SDC Platinum and match this information to our sample
firms (based on firm names).15 Results are presented in Table 9. We include the
same control variables as for leverage plus leverage itself. All independent variables
are lagged one year (i.e., they are based on the last available annual report before
the deal). We start by applying a dummy variable which indicates whether a firm
conducted any M&A deal in a specific year. After that, we investigate the annual
number of deals a firm conducts. For both models, we find strong evidence that
PER leads to significantly less M&A activity. Thus, firms with stronger employee
power tend to conduct fewer M&A deals.
15
In particular, we match the deals to the sample years based on their announcement dates and
consider all announced deals. We only consider years for which data for the whole year is available,
i.e., the time period from 2005 to 2013.
17
The next question we focus on is whether these firms select their M&A targets
more carefully. If employee representatives are effective monitors which reduce managerial agency cost, they may force managers to focus on value-increasing deals and
avoid those which are related to their utility maximization. For this purpose, we
analyze capital market reactions to the announcements of M&A deals. In terms
of methodology, we calculate cumulative abnormal returns (CAR) around the announcement dates. As event windows, we apply three days and use five days as
robustness tests. Normal returns are estimated during the 200 trading days ending
two months before the event (Cai and Sevilir, 2012). As market index, we use the
CDAX. Alternatively, cumulative excess returns (CER), i.e., the returns of the firm
which announces the deal above the market return, are applied. We control for the
same variables as in the main leverage regressions except for listing and accounting
standard (as listed firms have to apply international standards by law). Furthermore, we use the firm-specific TobQ as proxy for growth options. We also add free
float and market value as proxies for liquidity as well as controls whether the deal
is diversifying and international. Results are reported in Table 9. We find strong
evidence that the capital markets reacts more positively to M&A announcements of
firms with PER. This is in line with the view that these firms tend to conduct value
increasing deals and avoid transactions that are related to utility maximization of
managers.
Lastly, we analyze firms capital expenditures and find evidence for lower investments in firms which have parity codetermination.
— Table 9 about here —
6.2. Cash flow and profit stability
We next investigate the stability of firms’ cash flows and profits. If employee
power leads to more stable business decisions, we expect less profit and cash flow
fluctuations in firms with PER. We measure the standard deviation of cash flows
and profitability over windows of three, four, and five years and use the same control
variables as for our leverage regressions plus our proxy for firm growth (as growth
may be a primary determinant of stability) and leverage itself. For the explanatory
variables, we always use the value as of the beginning of the respective window. As
changes of employee representation may bias the outcome, we exclude such firms.
Results in Table 10 for cash flow (model I) and profitability (II) provide evidence
for more stability in firms with stronger employee power. The economic impact of
this effect is also substantial: the standard deviation of cash flow and profitability
is about 50% smaller in firms with codetermination. Thus, these findings indicate
that lower firm risk is a possible channel for the interest alignment between firms
with PER and banks.
18
— Table 10 about here —
6.3. Idiosyncratic risk
Besides the volatility of cash flows and profits, we also investigate idiosyncratic
firm risk. If firms with PER implement more stable investment policies, we expect
that these firms also have less idiosyncratic risk. To calculate idiosyncratic risk, we
follow Panousi and Papanikolaou (2012). In particular, we calculate σidio. for each
firm year based on weekly residuals of equity returns. The residuals are obtained
from regressing the firm’s equity returns on thesCDAX as market return. σidio for
i,t
firm i in year t is then calculated as σidio
=
52
P
2i,τ , where i,τ is the residual
τ =1
from the first regression in week τ . More details can be found in Appendix A. We
control for the same variables as in the main leverage regressions except for listing
and accounting standard (as listed firms have to apply international standards by
law). Furthermore, we use the firm-specific TobQ as proxy for growth options. We
also add free float and market value as proxies for liquidity. Results can be found in
Table 11. We find evidence that parity codetermination reduces idiosyncratic risk.
This is also true if we apply the log of σidio . Alternatively, we also investigate the
total volatility of firms’ equity returns and find similar results. This further adds
evidence that lower firm risk is a channel for the interest alignment effect between
firms with PER and banks.
— Table 11 about here —
7. Conclusion
In this paper, we analyze how employee codetermination affects financial leverage. We focus on employee representatives in the supervisory boards of German
firms. The supervisory board is comparable to the board of directors and has,
among others, the duty to monitor the executive managers. By law, the supervisory board of German firms has to consist of an equal number of employee and
owner representatives if the firm has more than 2,000 domestic employees (parity
employee representation, PER). Our identification strategy is based on a regression
discontinuity design around this threshold. Thus, we exploit that the appointment of
employee representatives is not a voluntary choice by the firms’ managers or owners,
but regulated by law. We find strong evidence that PER increases leverage.
Prior literature on employee power and financial leverage focuses mainly on
industry-level measures and explains higher leverage in more unionized industries
by collective bargaining. This explanation is, however, unlikely in our setting. Employee representatives who possess insider knowledge about the firm are more likely
19
to monitor managers tightly and less likely to re-elect them if debt levels are high.
Thus, maintaining high debt ratios as bargaining tool is unattractive for executive
managers if employee representatives are directly involved in the firm’s management.
We explain our finding by a supply side effect of debt capital, i.e., interest alignment
between banks and firms with PER. Employee representatives who aim to protect
the interests of the firm’s employees may (unintentionally) also help to protect the
interests of banks as both stakeholders are interested in the long-term survival and
stability of the firm. In line with this explanation, we find that bank ownership and
PER act as substitutes. As bank ownership per se is likely endogenous, we exploit
the capital gain tax reform of 2000 as exogenous shock on equity holdings of German
banks. Further analyses also reveal that firms with PER enjoy more favorable financing conditions, i.e., lower cost of debt. They also exhibit longer debt maturities,
smaller loan syndicates, and fewer covenants. Based on cash flow sensitivity of cash,
these firms have less financial constraints. Lastly, we identify lower firm risk due to
codetermination as possible channel for this interest alignment effect. In particular,
firms with strong employee power conduct less and better M&A deals, have lower
capital expenditures, more stable cash flows and profits, and lower idiosyncratic risk.
These findings indicate that employee power can be a powerful mechanism to
reduce agency conflicts between debt providers and firms’ owners and managers.
Our results show that laws that grant employees a direct voice in the firm’s management can improve firms’ financing opportunities and conditions. In this sense, this
paper identifies a bright side of employee codetermination. The question whether
firms can also profit by voluntarily granting employees such a direct voice in the
firm’s management is, however, not answered by this paper. We speculate that a
positive effect on financing requires a strong commitment of the firm for such rights.
Otherwise, the possibility that firms divest employees of their rights in certain situations, e.g., changes in the ownership structure, may wipe out the positive effects of
codetermination. A more detailed investigation in which situation employee rights
improve financing conditions is a promising area for future research.
20
References
Acharya, V. V., Baghai, R. P., Subramanian, K. V., 2014. Wrongful discharge laws
and innovation. Review of Financial Studies 27 (1), 301–346.
Agrawal, A. K., Matsa, D. A., 2013. Labor unemployment risk and corporate financing decisions. Journal of Financial Economics 108 (2), 449–470.
Almeida, H., Campello, M., Weisbach, M. S., 2004. The cash flow sensitivity of cash.
Journal of Finance 59 (4), 1777–1804.
Atanassov, J., Kim, E., 2009. Labor and corporate governance: International evidence from restructuring decisions. Journal of Finance 64 (1), 341–374.
Bakke, T.-E., Whited, T. M., 2012. Threshold events and identification: A study of
cash shortfalls. Journal of Finance 67 (3), 1083–1111.
Benelli, G., Loderer, C., Lys, T., 1987. Labor participation in corporate policymaking decisions: West Germany’s experience with codetermination. Journal of
Business 60 (4), 553–575.
Benmelech, E., Bergman, N. K., Enriquez, R. J., 2012. Negotiating with labor under
financial distress. Review of Corporate Finance Studies 1 (1), 28–67.
Besley, T., Burgess, R., 2004. Can labor regulation hinder economic performance?
Evidence from India. Quarterly Journal of Economics 119 (1), 91–134.
Botero, J. C., Djankov, S., La Porta, R., Lopez de Silanes, F., Shleifer, A., 2004.
The regulation of labor. Quarterly Journal of Economics 119 (4), 1339–1382.
Bradley, D., Kim, I., Tian, X., 2013. The causal effect of labor unions on innovation.
Working Paper.
Bronars, S. G., Deere, D. R., 1991. The threat of unionization, the use of debt, and
the preservation of shareholder wealth. Quarterly Journal of Economics 106 (1),
231–254.
Cai, Y., Sevilir, M., 2012. Board connections and M&A transactions. Journal of
Financial Economics 103 (2), 327–349.
Chava, S., Roberts, M. R., 2008. How does financing impact investment? the role
of debt covenants. Journal of Finance 63 (5), 2085–2121.
Chen, H. J., Kacperczyk, M., Ortiz-Molina, H., 2012. Do nonfinancial stakeholders
affect the pricing of risky debt? Evidence from unionized workers. Review of
Finance 16 (2), 347–383.
21
Chen, S.-S., Chen, Y.-S., Wang, Y., 2014. Does labor power affect the likelihood of
a share repurchase? Financial Management, forthcoming.
Chyz, J. A., Ching Leung, W. S., Zhen Li, O., Meng Rui, O., 2013. Labor unions
and tax aggressiveness. Journal of Financial Economics 108 (3), 675–698.
Cuñat, V., Gine, M., Guadalupe, M., 2012. The vote is cast: the effect of corporate
governance on shareholder value. Journal of Finance 67 (5), 1943–1977.
Edwards, C. H., Lang, M. H., Maydew, E. L., Shackelford, D. A., 2004. Germany’s
repeal of the corporate capital gains tax: the equity market response. Journal of
the American Taxation Association 26 (s-1), 73–97.
Faleye, O., Mehrotra, V., Morck, R., 2006. When labor has a voice in corporate
governance. Journal of Financial and Quantitative Analysis 41 (3), 489–510.
Fauver, L., Fuerst, M. E., 2006. Does good corporate governance include employee
representation? Evidence from German corporate boards. Journal of Financial
Economics 82 (3), 673–710.
FitzRoy, F. R., Kraft, K., 1993. Economic effects of codetermination. Scandinavian
Journal of Economics 95 (3), 365–375.
Frank, M. Z., Goyal, V. K., 2009. Capital structure decisions: which factors are
reliably important? Financial Management 38 (1), 1–37.
Ginglinger, E., Megginson, W., Waxin, T., 2011. Employee ownership, board representation, and corporate financial policies. Journal of Corporate Finance 17 (4),
868–887.
Gorton, G., Schmid, F. A., 2000. Universal banking and the performance of german
firms. Journal of Financial Economics 58 (1), 29–80.
Gorton, G., Schmid, F. A., 2004. Capital, labor, and the firm: A study of German
codetermination. Journal of the European Economic Association 2 (5), 863–905.
Gurdon, M. A., Rai, A., 1990. Codetermination and enterprise performance: empirical evidence from West Germany. Journal of Economics and Business 42 (4),
289–302.
Kim, E. H., Maug, E., Schneider, C., 2014. Labor representation in governance as
an insurance mechanism. Working Paper.
Klasa, S., Maxwell, W. F., Ortiz-Molina, H., 2009. The strategic use of corporate
cash holdings in collective bargaining with labor unions. Journal of Financial
Economics 92 (3), 421–442.
22
Kraft, K., Stank, J., Dewenter, R., 2011. Co-determination and innovation. Cambridge Journal of Economics 35 (1), 145–172.
Lee, D. S., Lemieux, T., 2010. Regression discontinuity designs in economics. Journal
of Economic Literature 48 (2), 281–355.
Matsa, D. A., 2010. Capital structure as a strategic variable: Evidence from collective bargaining. Journal of Finance 65 (3), 1197–1232.
Natividad, G., 2013. Financial capacity and discontinuous investment: Evidence
from emerging market multibusiness firms. Review of Financial Studies 26 (9),
2375–2410.
Panousi, V., Papanikolaou, D., 2012. Investment, idiosyncratic risk, and ownership.
Journal of Finance 67 (3), 1113–1148.
Perotti, E. C., Spier, K. E., 1993. Capital structure as a bargaining tool: The role of
leverage in contract renegotiation. American Economic Review 83 (5), 1131–1141.
Propper, C., Van Reenen, J., 2010. Can pay regulation kill? Panel data evidence on
the effect of labor markets on hospital performance. Journal of Political Economy
118 (2), 222–273.
Renaud, S., 2007. Dynamic efficiency of supervisory board codetermination in germany. Labour 21 (4-5), 689–712.
Rieble, V., 2006. Schutz vor paritätischer Unternehmensmitbestimmung. BetriebsBerater 37 (61), 2018–2023.
Simintzi, E., Vig, V., Volpin, P., 2014. Labor protection and leverage. Review of
Financial Studies, forthcoming.
Von Beschwitz, B., Foos, D., 2013. Banks’ equity stakes and lending: Evidence from
a tax reform. Working Paper.
Winkler, N., 2006. Das Stimmrecht der Aktionäre in der Europäischen Union. de
Gruyter, Berlin.
23
(a) bins of 50
(b) linear fit
(c) bins of 100
(d) linear fit
Figure 1: This figure shows regression discontinuity plots with quadratic (a, c) or linear (b, d)
fits and the corresponding 99% confidence interval. The x-axis displays the number or domestic
employees, measured in bins of 50 or 100 employees around 2,000. The y-axis shows the mean
leverage in the respective bin. Observations with more than 2,000 domestic employees but no parity
employee representation or less than 2,000 domestic employees but parity employee representation
are excluded.
24
Figure 2: Histogram of domestic employees
Figure 3: Employee representation and domestic employees
25
Figure 4: Bank ownership pre/post tax reform
26
Table 1: Descriptive statistics
Variable
PER
DE
Leverage
Listing
Size (mio Eur)
Tangibility
ROA
TobQind
Acc. Std.
N
Mean
SD
p25
p50
p75
min
max
768
797
797
791
797
797
797
797
797
0.25
1785.27
0.53
0.40
740.25
0.30
0.08
1.30
0.53
0.44
518.38
0.19
0.49
1042.76
0.17
0.08
0.30
0.50
0
1394
0.4
0
272
0.19
0.04
1.11
0
0
1662
0.56
0
452
0.27
0.08
1.24
1
1
2127
0.65
1
765
0.38
0.12
1.43
1
0
1000
0.04
0
43.2
0.02
-0.5
0.93
0
1
2998
1
1
8794
0.94
0.62
3.94
1
This table presents the number of observations (N), mean, standard deviation (SD),
25% percentile, median, 50% percentile, minimum, and maximum value for the most
relevant variables. Firm-years with ≥ 1,000 and ≤ 3,000 domestic employees are
considered. A detailed description of all variables can be found in Appendix A.
27
Table 2: Leverage
Model
I
II
III
IV
V
PER
0.068**
(2.16)
0.12***
(3.09)
0.11***
(2.91)
0.14***
(3.20)
0.15***
(3.22)
Size
-0.0038
(-0.17)
-0.11
(-0.96)
-1.05***
(-6.27)
-0.023
(-0.80)
0.028
(0.67)
-0.092**
(-2.21)
-0.0015
(-0.067)
-0.094
(-0.85)
-1.05***
(-6.21)
-0.014
(-0.55)
0.033
(0.79)
-0.096**
(-2.33)
-0.0022
(-0.098)
-0.082
(-0.73)
-1.06***
(-6.27)
-0.019
(-0.69)
0.035
(0.85)
-0.10**
(-2.46)
-0.0034
(-0.15)
-0.073
(-0.67)
-1.05***
(-6.24)
-0.012
(-0.43)
0.033
(0.81)
-0.098**
(-2.40)
-0.0030
(-0.13)
-0.071
(-0.65)
-1.04***
(-6.17)
-0.015
(-0.49)
0.031
(0.73)
-0.096**
(-2.31)
725
158
0.33
yes
yes
no
no
725
158
0.34
yes
yes
one
no
725
158
0.34
yes
yes
two
no
725
158
0.35
yes
yes
two
yes
725
158
0.36
yes
yes
four
yes
Tangibility
ROA
TobQind
Listing
Acc. Std.
Observations
Firms
R2
Industry FE
Year FE
Polynomial
/ either side
The dependent variable is leverage. All models are pooled OLS regressions. Independent variables except accounting standard and listing are
lagged one period. Firm-years with between 1,000 and 3,000 domestic
employees are included. per stands for parity employee representation in
the firm’s supervisory board. Polynomial indicates how we control for the
centered number of domestic employees. Either side means that the polynomials interacted with a dummy for 2,000 DE are included. T-statistics
based on Huber/White robust standard errors clustered by firms are presented in parentheses. ***, ** and * indicate significance on the 1%-, 5%and 10%-levels, respectively. A detailed description of all variables can be
found in Appendix A.
28
Table 3: Identification threats
Model
I
law deviations
PER
0.20**
(2.09)
DE1,500
IIa
IIb
placebo test
-0.0036
(-0.093)
DE2,500
0.037
(0.71)
TE2,000
Size
Tangibility
ROA
TobQind
Listing
Acc. Std.
Observations
Firms
R2
Industry FE
Year FE
Polynomial
/ either side
III
-0.051
(-1.03)
-0.011
(-0.47)
0.0018
(0.017)
-1.11***
(-6.53)
0.024
(0.74)
0.032
(0.76)
-0.097**
(-2.30)
-0.0011
(-0.055)
-0.20*
(-1.94)
-0.88***
(-7.16)
0.027
(0.75)
-0.013
(-0.34)
-0.099**
(-2.57)
0.013
(0.51)
-0.15
(-1.15)
-0.89***
(-5.30)
-0.092**
(-2.32)
0.029
(0.75)
-0.090**
(-2.21)
-0.0020
(-0.078)
0.034
(0.32)
-0.88***
(-6.14)
0.036
(0.68)
-0.035
(-0.98)
-0.044
(-1.17)
679
152
0.36
yes
yes
four
yes
934
187
0.32
yes
yes
four
no
557
142
0.32
yes
yes
four
no
653
155
0.33
yes
yes
four
no
The dependent variable is leverage. DE1,500(2,500) indicates
whether the number of domestic employees is above 1,500 (2,500).
TE2000 equals one if the firm has more than 2,000 total employees. All models are pooled OLS regressions. Independent variables
except accounting standard and listing are lagged one period. Firmyears with between 1,000 and 3,000 domestic employees are included
if not stated otherwise. In model I, observations with less (more)
than 2,000 domestic employees but (no) parity employee representation are excluded. A placebo test with different randomly chosen
thresholds for domestic employees is shown in model II. This model
includes firm-years with +/- 1,000 domestic employees. In model
III, we use a threshold of 2,000 total employees as placebo test. Tstatistics based on Huber/White robust standard errors clustered
by firms are presented in parentheses. ***, ** and * indicate significance on the 1%-, 5%- and 10%-levels, respectively. A detailed
description of all variables can be found in Appendix A.
29
Table 4: Alternative windows
Lower bound
Upper bound
1700
2300
1400
2600
1400
3000
full
sample
PER
0.16***
(4.12)
0.19***
(3.75)
0.15***
(3.16)
0.11***
(2.85)
Size
-0.0085
(-0.23)
-0.57***
(-3.53)
-0.96***
(-4.45)
0.033
(0.47)
-0.059
(-0.73)
-0.12
(-1.49)
-0.021
(-0.71)
-0.17
(-1.25)
-1.00***
(-4.63)
0.0058
(0.17)
-0.021
(-0.39)
-0.078
(-1.45)
-0.0098
(-0.37)
-0.10
(-0.81)
-1.00***
(-4.89)
-0.040
(-1.27)
0.022
(0.49)
-0.097**
(-2.14)
0.00013
(0.011)
-0.026
(-0.40)
-0.62***
(-6.45)
0.031
(1.27)
0.0012
(0.049)
-0.072**
(-2.53)
193
69
0.49
yes
yes
four
yes
439
113
0.34
yes
yes
four
yes
519
131
0.32
yes
yes
four
yes
2,102
378
0.21
yes
yes
four
yes
Tangibility
ROA
TobQind
Listing
Acc. Std.
Observations
Firms
R2
Industry FE
Year FE
Polynomial
/ either side
The dependent variable is leverage. All models are pooled
OLS regressions. Independent variables except accounting standard and listing are lagged one period. Upper and lower bounds
indicate the thresholds for the number of domestic employees.
T-statistics based on Huber/White robust standard errors clustered by firms are presented in parentheses. ***, ** and *
indicate significance on the 1%-, 5%- and 10%-levels, respectively. A detailed description of all variables can be found in
Appendix A.
30
Table 5: Bank ownership: tax reform
Model
Ia
PERpre
0.11
(1.00)
0.17*
(1.84)
PERpre x post
Ib
0.18**
(2.44)
Ic
0.35***
(4.40)
Bankpre
PER x bankpre
Post x bankpre
PERpre x post x bankpre
Size
Tangibility
ROA
TobQ
Acc. std.
Observations
Firms
R2
Industry FE
Year FE
Firm FE
Polynomial
/ either side
Polynomial x post
IIa
IIb
-0.088*
(-1.94)
0.041
(1.00)
0.13*
(1.72)
-0.089
(-0.99)
-0.18**
(-2.38)
0.21**
(2.39)
-0.12*
(-1.76)
0.20**
(2.44)
0.0044
(0.12)
0.033
(0.70)
-0.011
(-0.055)
-0.56*
(-1.94)
-0.097*
(-1.74)
-0.029
(-0.42)
-0.066
(-0.66)
0.033
(0.079)
0.014
(0.080)
0.0027
(0.076)
0.032
(0.57)
-0.11
(-1.51)
0.22
(0.92)
0.23*
(1.69)
-0.014
(-0.43)
0.042
(0.91)
0.032***
(3.34)
0.28***
(3.63)
-0.13
(-1.27)
-0.073***
(-5.24)
0.0071
(0.24)
0.072***
(2.74)
0.37***
(3.54)
-0.21***
(-3.43)
0.0039
(0.20)
0.010
(0.37)
160
57
0.32
yes
yes
no
four
yes
no
160
57
0.21
yes
yes
yes
four
yes
no
160
57
0.51
yes
yes
yes
four
yes
yes
1,103
264
0.23
yes
yes
no
no
no
no
1,103
264
0.13
yes
yes
yes
no
no
no
The event is a tax reform which made it more attractive for German banks to sell equity
stakes. The dependent variable is leverage. Models are pooled OLS regressions of
firm-fixed effects regressions. Only listed sample firms are considered. Firm-years with
between 1,000 and 3,000 DE are included in model I; all firm-years are considered in model
II. The time-constant variables bankpre and bankpre x perpre are omitted in the FE
models. Independent variables except accounting standard are lagged one period. The
sample period is 1996 to 1998 (pre-event) and 2002 to 2004 (post-event); years between
1999 and 2001 are excluded. post is one after 2001 as the tax reform became effective
on January 1st, 2002. PERpre is the value of per in 1998. Bankpre indicates if a firm
had a bank as owner before the reform (i.e., between 1996 and 1998). T-statistics based
on Huber/White robust standard errors clustered by firms are presented in parentheses.
***, ** and * indicate significance on the 1%-, 5%- and 10%-levels, respectively. A
detailed description of all variables can be found in Appendix A.
31
Table 6: Interest rate
Model
I
IIa
IIb
III
PER
-0.017***
(-5.67)
-0.013***
(-3.09)
-0.011*
(-1.82)
-0.018***
(-3.89)
-0.0023
(-0.49)
-0.0019**
(-2.21)
0.015**
(2.18)
-0.026*
(-1.87)
-0.0013
(-0.55)
0.0026
(1.64)
-0.0017
(-1.08)
0.010***
(9.35)
0.0051
(1.07)
-0.0053***
(-4.61)
0.00043
(0.078)
-0.018*
(-1.86)
0.00075
(0.31)
-0.00032
(-0.18)
-0.0012
(-0.72)
0.019**
(2.50)
-0.0028*
(-1.97)
0.0038
(0.40)
-0.0071
(-0.29)
-0.020***
(-2.97)
0.00089
(0.35)
-0.0033
(-1.20)
0.0099*
(1.73)
-0.0029**
(-2.58)
0.036***
(3.00)
-0.050**
(-2.54)
0.0066
(1.38)
0.0027
(1.29)
-0.0031*
(-1.71)
253
61
0.64
no
no
four
yes
253
61
0.63
no
yes
four
yes
253
61
0.53
yes
no
four
yes
253
61
0.68
yes
yes
four
yes
Leverage
Size
Tangibility
ROA
TobQind
Listing
Z-score class
Ind/year mean
Observations
Firms
R2
Industry FE
Year FE
Polynomial
/ either side
The dependent variable is spread of syndicated loans. All models
are pooled OLS regressions. Independent variables are lagged one period. All firm-years are considered. Deal controls include purpose and
amount. T-statistics based on Huber/White robust standard errors
clustered by firms are presented in parentheses. ***, ** and * indicate
significance on the 1%-, 5%- and 10%-levels, respectively. A detailed
description of all variables can be found in Appendix A.
32
Table 7: Loan characteristics
maturity
lenders
covenants
PER
0.22**
(2.60)
0.20**
(2.46)
-0.34*
(-1.84)
-3.10**
(-2.15)
Leverage
0.077
(0.43)
0.071*
(1.80)
0.28*
(1.75)
0.90**
(2.01)
-0.0061
(-0.048)
-0.10*
(-1.94)
-0.041
(-0.81)
-0.046
(-0.29)
0.070**
(2.12)
0.18
(1.29)
0.74*
(1.89)
0.057
(0.43)
-0.097**
(-2.14)
-0.026
(-0.57)
0.73***
(2.75)
0.028
(0.30)
-0.52
(-1.08)
0.51
(0.65)
0.61***
(2.75)
-0.075
(-0.61)
0.22
(1.64)
1.15
(0.35)
0.17
(0.28)
5.32
(1.57)
4.18
(0.62)
-1.36
(-0.60)
3.09***
(2.77)
-2.11***
(-2.64)
735
133
yes
yes
four
yes
OLS
735
133
yes
yes
four
yes
Poisson
763
134
yes
yes
four
yes
Poisson
763
134
yes
yes
four
yes
Poisson
Listing
Size
Tangibility
ROA
TobQind
Z-score class
Observations
Firms
Industry FE
Year FE
Polynomial
/ either side
Model
The dependent variables for maturity are ln maturity and
maturity. For lenders, the dependent variable is number of
lenders and for covenants the number of covenants. All firmyears are considered. The first model is a pooled OLS regression; all other models are Poisson regressions. Independent
variables are lagged one period. Deal controls include purpose and amount. T-statistics based on Huber/White robust
standard errors clustered by firms are presented in parentheses. ***, ** and * indicate significance on the 1%-, 5%- and
10%-levels, respectively. A detailed description of all variables
can be found in Appendix A.
33
Table 8: Cash flow sensitivity of cash
Model
Sample
Ia
PER
Ib
no PER
IIa
all
IIb
CF≥0
IIa
-0.023
(-0.30)
0.098***
(2.88)
0.015
(1.33)
0.10***
(2.91)
-0.12*
(-1.83)
0.024**
(2.27)
0.12***
(4.40)
-0.19***
(-2.85)
0.019*
(1.90)
0.064*
(1.91)
-0.13**
(-2.11)
0.027**
(2.37)
0.21
(1.46)
-0.24**
(-2.08)
0.0021
(0.44)
0.0023
(0.13)
0.0010
(0.38)
0.0034
(0.24)
0.0013
(0.62)
0.0040
(0.39)
-0.00048
(-0.25)
0.0011
(0.11)
0.0015
(0.64)
0.015*
(1.74)
-0.17*
(-1.82)
0.17***
(4.74)
-0.0015
(-0.048)
-0.00066
(-0.12)
0.0015
(0.60)
0.014
(1.59)
-0.16*
(-1.88)
0.17***
(4.74)
-0.0072
(-0.25)
-0.00044
(-0.079)
154
44
0.11
yes
yes
four
yes
no
436
106
0.084
yes
yes
four
yes
no
590
146
0.057
yes
yes
four
yes
no
567
144
0.056
yes
yes
four
yes
no
590
146
0.17
yes
yes
four
yes
no
590
146
0.19
yes
yes
four
yes
yes
PER
Cash flow
PER x cash flow
Size
Growth−1y,+1y
∆ Short debt
∆ NWC
Capex
M&A deal
Observations
Firms
R2
Industry FE
Year FE
Polynomial
/ either side
Poly. x cash flow
IIb
all
The dependent variable is ∆ cash. All models are pooled OLS regressions. Firm-years
with between 1,000 and 3,000 domestic employees are included. In model Ia (b), only firmyears with (without) PER are considered. PER is lagged one period. T-statistics based on
Huber/White robust standard errors clustered by firms are presented in parentheses. ***,
** and * indicate significance on the 1%-, 5%- and 10%-levels, respectively. A detailed
description of all variables can be found in Appendix A.
34
Table 9: Investments
PER
Leverage
Size
Tangibility
ROA
TobQind
Listing
Capex
# deals
-0.78***
(-3.00)
-0.84***
(-2.58)
4.16**
(2.45)
4.99***
(3.01)
3.98**
(2.54)
-0.059**
(-1.98)
-0.18
(-0.46)
0.56***
(4.76)
-0.37
(-0.65)
0.98
(0.90)
0.22
(0.52)
0.42**
(2.40)
-0.0023
(-0.0044)
0.79***
(6.42)
-1.12
(-1.48)
0.24
(0.17)
0.19
(0.31)
0.85***
(3.17)
2.93
(0.81)
-1.18
(-0.91)
-4.46
(-1.53)
-34.9**
(-2.59)
5.43
(1.47)
-2.07
(-1.41)
-4.78
(-1.53)
-32.4**
(-2.51)
0.60
(0.12)
-2.39*
(-2.04)
-4.05*
(-1.76)
-33.4
(-1.65)
0.026
(0.79)
0.026***
(2.90)
-0.0088
(-0.21)
0.23**
(2.17)
0.012
(0.57)
0.021*
(1.67)
0.85
(0.60)
-0.84
(-0.95)
-0.048***
(-3.19)
-1.21**
(-2.15)
1.50*
(1.70)
0.95
(0.70)
-0.22
(-0.21)
-0.046***
(-4.02)
-1.33**
(-2.22)
1.53*
(1.93)
0.82
(0.38)
0.10
(0.11)
-0.015
(-0.86)
-0.83
(-0.86)
1.69
(1.30)
182
30
0.42
yes
yes
four
yes
OLS
182
30
0.41
yes
yes
four
yes
OLS
182
30
0.38
yes
yes
four
yes
OLS
TobQ
Market cap.
Free float
Diversifying deal
International deal
Observations
Firms
R2
Industry FE
Year FE
Polynomial
/ either side
Model
M&A deals
CAR−1,+1 CER−1,+1
dummy
688
157
0.24
yes
yes
four
yes
Probit
692
158
yes
yes
four
yes
Poisson
CAR−2,+2
782
158
0.21
yes
yes
four
yes
OLS
The dependent variables are indicated in each column. Independent variables are lagged one
period for M&A analyses. The models includes firm-years with more than 1,000 and less than
3,000 domestic employees. T-statistics based on Huber/White robust standard errors clustered
by firms are presented in parentheses. ***, ** and * indicate significance on the 1%-, 5%- and
10%-levels, respectively. A detailed description of all variables can be found in Appendix A.
35
Table 10: Stability
Model
Variable
Period
Ic
IIa
3y
Ib
SD(cash flow)
4y
5y
PER
-0.030**
(-2.06)
-0.028*
(-1.96)
Leverage
-0.013
(-0.76)
0.0073**
(1.99)
0.029*
(1.67)
-0.018
(-0.41)
-0.0043
(-0.32)
0.032
(0.97)
-0.013*
(-1.88)
0.014*
(1.79)
344
97
0.40
yes
yes
four
yes
Size
Tangibility
ROA
TobQind
Growth−1y,+1y
Listing
Acc. Std.
Observations
Firms
R2
Industry FE
Year FE
Polynomial
/ either side
Ia
IIc
3y
IIb
SD(ROA)
4y
-0.038**
(-2.26)
-0.031
(-1.57)
-0.032*
(-1.85)
-0.048*
(-1.91)
-0.021
(-1.02)
0.0066
(1.65)
0.053**
(2.10)
-0.022
(-0.29)
0.0060
(0.29)
0.037
(1.10)
-0.013
(-1.45)
0.015
(1.59)
-0.018
(-0.84)
0.0083
(1.65)
0.052
(1.44)
0.052
(0.58)
0.0025
(0.11)
0.026
(1.06)
-0.011
(-1.06)
0.011
(1.00)
-0.013
(-0.76)
0.013***
(2.69)
0.030
(1.23)
-0.077
(-0.88)
-0.0031
(-0.20)
0.019
(0.68)
-0.0070
(-0.67)
0.012
(1.19)
-0.021
(-1.01)
0.013**
(2.32)
0.044
(1.46)
-0.051
(-0.58)
0.00012
(0.0056)
0.031
(1.02)
-0.0073
(-0.55)
0.012
(1.04)
-0.017
(-0.66)
0.016**
(2.14)
0.062
(1.33)
0.023
(0.22)
0.0030
(0.12)
0.022
(0.83)
0.0014
(0.088)
0.0014
(0.11)
256
90
0.44
yes
yes
four
yes
170
63
0.57
yes
yes
four
yes
347
98
0.41
yes
yes
four
yes
258
91
0.45
yes
yes
four
yes
171
64
0.54
yes
yes
four
yes
5y
The dependent variables are sd(cash flow) and sd(roa). Firms for which per changes
during the sample period are not considered. All models are pooled OLS regressions.
Independent variables are as at the beginning of the measurement period. T-statistics
based on Huber/White robust standard errors clustered by firms are presented in parentheses. ***, ** and * indicate significance on the 1%-, 5%- and 10%-levels, respectively.
A detailed description of all variables can be found in Appendix A.
36
Table 11: Idiosyncratic risk
Model
Variable
I
σidio
II
log(σidio )
III
σtotal
PER
-6.94**
(-2.47)
-0.15**
(-2.01)
-0.88**
(-2.11)
Leverage
6.57
(0.98)
7.08**
(2.18)
0.74
(0.099)
-57.1***
(-2.98)
8.25***
(4.71)
-6.79***
(-2.84)
0.014
(0.50)
0.13
(0.86)
0.18**
(2.18)
0.011
(0.050)
-1.03***
(-3.25)
0.17***
(3.27)
-0.18***
(-3.00)
3.4e-06
(0.0042)
0.93
(0.91)
1.32**
(2.46)
-0.014
(-0.011)
-9.93***
(-3.11)
1.73***
(5.47)
-1.11***
(-2.79)
0.0097*
(1.87)
258
58
0.66
yes
yes
four
yes
258
58
0.64
yes
yes
four
yes
258
58
0.70
yes
yes
four
yes
Size
Tangibility
ROA
TobQ
Market cap.
Free float
Observations
Firms
R2
Industry FE
Year FE
Polynomial
/ either side
The dependent variables indicated in each column. σidio is a yearly measure for idiosyncratic risk
based on equity returns (Panousi and Papanikolaou,
2012). σtotal captures total firm risk. All models are pooled OLS regressions. Independent variables are as at the beginning of the measurement
period. T-statistics based on Huber/White robust
standard errors clustered by firms are presented in
parentheses. ***, ** and * indicate significance on
the 1%-, 5%- and 10%-levels, respectively. A detailed description of all variables can be found in
Appendix A.
37
Appendix
Appendix A: Definition of variables
Variable
Description
Main variables
PER
DE
Leverage
Dummy which equals one if the firm has parity employee representation.
Source: hand-collected.
Number of domestic employees. Source: Hoppenstedt and hand-collected.
Total debt divided by total debt plus book value of equity. Total debt
includes current and long-term liabilities and excludes provisions and accruals. Source: Hoppenstedt (HS).
Control variables
Listing
Size
Tangibility
ROA
TobQ
TobQind
Acc. Std.
Z-Score
Z-Score class
Dummy that equals one if preferred or voting shares of the firm are listed
on any EU-regulated or exchange-regulated market in Germany. Source:
hand-collected.
Natural logarithm of total assets in Mio e . Source: HS.
Long-term tangible assets scaled by total assets. Source: HS.
Earnings before interest and taxes, scaled by total assets. Source: HS.
Market value of equity plus total liabilities divided by the sum of book
). Source: WC.
value of equity and total liabilities ( wc08001+wc03351
wc03501+wc03351
Mean of TobQ in an industry (based on the 17 industries classification)
and year. Source: Own calculation.
Dummy which equals one if a firm applies international accounting standards. Source: HS.
Z-Score is calculated as 0.717*x1 + 0.847*x2 + 3.107*x3 + 0.420*x4 +
earnings
liabilities
; x2 = retained
;
0.998*x5 with x1 = short-term assets−short-term
total assets
total assets
total equity
EBIT
sales
x3 = total assets ; x4 = total liabilities ; x5 = total assets ; Source: own calculation based on HS.
Equals minus one if Z-Score is above 2.9, one if Z-Score is below 1.22, and
zero if Z-Score is between 1.22 and 2.9. Source: own calculation.
Dealscan variables (source: Dealscan)
Spread
Maturity
Lenders
Covenants
Amount
Purpose
Spread all-in-drawn.
Ln of deal maturity in months.
Ln of number of lenders. If Dealscan reports no information on lender
shares, we assume that only one lender exists.
Number of covenants for the deal. If no information on covenants is availabe, we set the variable to zero.
Ln of the facility amount.
Dummy which equals one if primary purpose is “corporate purpose”.
Tax event
Post
ERpre−ref
Bankpre−ref
Dummy for post tax reform period. Equals one for years 2002 to 2004.
ER as of 1998. Source: hand-collected.
Dummy which equals one if a German bank owns voting rights in any year
between 1996 and 1998. Source: HS.
38
Definition of variables - continued
Variable
Description
M&A deals (source: SDC Platinum)
M&A deal
# deals
CAR
CER
Diversifying deal
International deal
Dummy which equals one if a firm conducts a M&A deal.
Number of M&A deals.
Cumulative abnormal return around the announcement of a M&A deal.
Normal returns are estimated with a market model; the CDAX is used
as market index. The estimation period is 200 trading days ending two
months before the announcement date (Cai and Sevilir, 2012). Events for
which we have less 150 observations during the estimation period are not
considered. The CAR is measured during a symmetric three or five day
window around the event.
Cumulative excess return. Excess return is defined as return above the
market index, i.e., the CDAX.
Dummy which equals one if the firm acquires a target outside its main
business segment (as indicted by the SDC macro industry).
Dummy which equals one if the target is located outside Germany.
Stability & risk
SD(cash flow)
SD(ROA)
σidio
σtotal
Standard deviation of cash flow over the specific time period. Source: HS.
Standard deviation of ROA over the specified time period. Source: HS.
Yearly measure for idiosyncratic risk (Panousi and Papanikolaou, 2012).
Calculated as volatility of weekly residuals. Residuals are obtained from a
regression of firm’s equity returns on the market return (CDAX). σidio for
pP 2
i,t
firm i in year t is then calculated as σidio
i,τ ; with i,τ being the
=
residual from the first regression in week τ . Equity returns are calculated
based changes of the Datastream item RI (return index) between week t-1
and t. Some adjustments are made for equity returns (i.e., deletion of all
observations with unadjusted price below one and observations where the
same price is reported for at least three consecutive weeks)
Yearly standard deviation of weekly equity returns. The same adjustments
as above are conducted.
Worldscope database
Leverage
Market leverage
Size
Tangibility
ROA
Total debt [wc03255] / (total debt + book value of equity [wc03501]).
Total debt [wc03255] / (total debt + market value of equity [wc08001]).
Ln of total assets [wc02999] in thousand e .
Property, plant, and equipment [wc02501] / total assets.
EBIT [wc18191] / total assets.
Other variables
Cash flow
Cash
Growth−1y,+1y
∆ Delta short debt
Sales minus total expenses plus depreciation divided by total assets.
Source: HS.
Cash and liquid assets divided by total assets. Source: HS.
salest+1
Sales growth between year t+1 and year t-1. Calculated as salest−1
− 1.
Source: HS.
Difference in trade liabilities between t-1 and t, divided by total assets.
Source: HS.
39
Definition of variables - continued
Variable
Description
Capex
Additions to fixed assets in year t divided by total assets at the end of year
t-1. Source: HS.
Difference in working capital, scaled by total assets. Source: HS.
Fraction of shares in free float. Source: DS.
Natural logarithm of one plus market capitalization in Mio e . Source: DS.
∆ NWC
Free float
Market cap.
HS stands for Hoppenstedt, DS for Dealscan, and WC for Worldscope.
40
Appendix B: General robustness tests
Model
Ia
ind-year FE
Ib
38 ind.
II
listed
PER
0.17***
(3.19)
0.17***
(4.15)
0.12**
(2.44)
Size
-0.0032
(-0.13)
-0.056
(-0.48)
-1.05***
(-5.72)
-0.053
(-0.83)
0.028
(0.61)
-0.093**
(-2.09)
-0.013
(-0.58)
-0.17
(-1.44)
-0.92***
(-5.94)
0.0063
(0.19)
0.036
(0.94)
-0.096**
(-2.40)
0.030
(1.29)
0.050
(0.38)
-0.88***
(-4.51)
Tangibility
ROA
TobQind
Listing
Acc. Std.
TobQ
Observations
Firms
R2
Industry FE
Year FE
Polynomial
/ either side
-0.014
(-0.44)
725
158
0.42
yes
yes
four
yes
725
158
0.41
yes
yes
four
yes
279
60
0.49
yes
yes
four
yes
The dependent variable is leverage. All models are
pooled OLS regressions. Independent variables except
accounting standard and listing are lagged one period.
In model I, we include firm-year fixed effects or apply
the 38 industries classification (instead of 17 industries). For model II, we restrict the sample to listed
firms. T-statistics based on Huber/White robust standard errors clustered by firms are presented in parentheses. ***, ** and * indicate significance on the 1%-,
5%- and 10%-levels, respectively. A detailed description of all variables can be found in Appendix A.
41
Appendix C: Robustness: database
Model
I
IIa
IIb
IIIa
IIIb
0.11**
(2.10)
0.12**
(2.08)
0.10*
(1.82)
0.100*
(1.79)
0.038***
(6.89)
0.14*
(1.75)
-0.35***
(-2.80)
-0.033**
(-2.22)
0.062**
(2.17)
0.28*
(1.73)
-1.01***
(-5.05)
-0.035
(-1.18)
0.058*
(1.97)
0.27
(1.66)
-0.98***
(-4.41)
-0.031
(-0.98)
0.043*
(1.67)
0.40**
(2.02)
-0.98***
(-6.79)
-0.064**
(-2.29)
0.039
(1.45)
0.37*
(1.93)
-0.97***
(-6.63)
-0.058**
(-2.06)
1,094
184
0.34
yes
yes
no
no
279
60
0.53
yes
yes
four
yes
280
60
0.52
yes
yes
four
yes
282
60
0.60
yes
yes
four
yes
283
60
0.58
yes
yes
four
yes
PER
Size
Tangibility
ROA
TobQ
Observations
Firms
R2
Industry FE
Year FE
Polynomial
/ either side
The dependent variable is leverage in models I and II and market
leverage in model III. Leverage in models II an III is based on Worldscope data. In model IIb and IIIb, control variables are also calculated
with data from Worldscope (instead of Hoppenstedt). All models are
pooled OLS regressions. Independent variables are lagged one period.
Only listed firms are considered. Firm-years with between 1,000 and
3,000 domestic employees are included in models II and III. In model, I all
firm-years independent of the number of domestic employees are included.
T-statistics based on Huber/White robust standard errors clustered by
firms are presented in parentheses. ***, ** and * indicate significance on
the 1%-, 5%- and 10%-levels, respectively. A detailed description of all
variables can be found in Appendix A.
42