Labors on the supervisory board : do smaller firms benefit or harm from it?

(1)

[I]

Labors on the supervisory board: Do smaller firms benefit or

harm from it?

-Empirical Evidence from Germany-

Written by: Jan Urbanec

Student number: 10225102

Supervisor: Professor Martin, Ph.D.

(2)

[II]

Abstract:

This research empirically investigates the composition of the German supervisory board as it contains both, shareholder and labor representatives. This research uses a quasi-experimental research design in order to analyze the impact of co-determination on firm performance. The findings show that small firms with less than 500 employees suffer from co-determination. Hence, co-determination is negatively associated with firm performance.

(3)

[III]

Inhaltsverzeichnis

I

Introduction ... 1

II Literature review & Hypothesis development ... 2

III Empirical Analysis ... 5

Research method ... 5

Sample Data ... 7

Findings ... 9

Discussion & Robustness checks ... 19

IV Conclusion... 23

V Literature ... 24

(4)

I

Introduction

Most companies are governed by a board of directors which are economic institutions that according to theory should help solving principal-agent problems (Hermalin and Weisbach, 2003). In recent years, there has been an ongoing debate about the efficacy of different systems of board of directors. There are two different systems which are mainly used: The one-tier system which is one board consisting of executive and non-executive directors and as second approach the two-tier system that consists of one supervisory board and one executive board. This research will focus on the two-tier system which is applied in Germany, more precisely, on the

supervisory board and its structure as the board contains both labor as well as capital

representatives. As both types of representatives differ in their aims, values and backgrounds, this study seeks to examine these differences and the potential impact on the firm performance. Therefore, the research question is formulated as follows: What type of supervisory board

structure and how does it affect firm performance? The purpose of this paper is to provide an

answer to the question stated above through which it contributes two important things: Firstly, it closes a gap in the literature as it has not been researched so far for small companies and

secondly, it helps emerging markets who want to employ the two-tier system how the supervisory board should be determined regarding co-determination and structure.

(5)

II Literature review & Hypothesis development

The impact of board composition on firm performance has been researched in the financial as well as the organizational economics literature (Adams et al., 2010). The majority of the studies, mostly US-based, concentrate on the optimal size and composition of the boards as it is assumed that this helps mitigating agency costs due to the separation of ownership and control (Shleifer and Vishny, 1997). However, this literature focuses mainly on the US market and thus, on the one-tier systems, which is one board consisting of executive and non-executive directors. German literature concentrates more on firms with a dual board structure. This two-tier system, as applied in Germany, consists of a supervisory board and an executive board. The main focus of German investigations, covering board composition, includes the role of bank and employee

representatives on the supervisory board (Frick and Lehmann, 2005). German law requires for many firms labor representation. Hence, the following paragraph will present the content of German law, followed by the aggregation of major findings and theory regarding

co-determination through which the hypothesis is developed.

Legal background of employee representation

In companies with a dual board structure such as required by German law, all members of the supervisory board are by definition independent. Furthermore, due to co-determination (”Mitbestimmung”), one third to one half of the directors of the supervisory board in German companies are appointed by the employees, depending on the size of the workforce and some other characteristics (e.g. family ownership).

In 2004 the One-third Codetermination Act (“Drittelmitbestimmungsgesetz”) was passed. This law revised the Works Constitution Act (“Betriebsverfassungsgesetz”) of 1952, which stated that all companies having between 0 and 2000 employees were required to appoint one-third of the supervisory board seats to labor representatives. This law applies to all but the metal industries. All companies in the metal-working industry fall under the Montan Co-determination Act from 1952 (“Montan-Mitbestimmungsgesetz”). These companies are compelled to half (50 %) labor representatives plus one neutral member. For all other firms above 2000 employees this ratio also

(6)

increases to 50%. The chairman of the board, who is a capital representative, has two votes in a tie.

In 1994, the law for small stock corporations1 (“Gesetz für kleine Aktiengesellschaften und zur Deregulierung des Aktienrechts”) loosened this regulation for small companies (less than 500 employees) which were founded after 1994 or are family owned firms2: they do not need to have any labor representation on the supervisory board. However, companies with less than 500 employees and founded before 1994 (who do not have the status of a family owned company) are still subject to co-determination.

Employee representation on the supervisory board

There is an ongoing debate about efficacy of labor representation (= co-determination) on the supervisory board and the resulting consequences as required by German law.

Furubotn (1988) argues that a positive argument in favor for co-determination is especially when labors make “durable reliance investments”. However, in the existence of information

asymmetry, this can never be proven. To the contrary, Furubotn (1988) summarizes the political economy argument against co-determination “(…) that the orthodox co-determined firm does not possess a truly efficient organizational structure.”3 Another, more market-oriented, argument against labor representation on the supervisory board is determined by Jensen and Meckling (1979) as they argue that “[if] co-determination is beneficial to both stockholders and labor[s], why do we need laws which force firms to engage in it? Surely, they would do so voluntarily. The fact that stockholders must be forced by law to accept co-determination is the best evidence we have that they are adversely affected by it.”4

Empirical Evidence present opposed findings like the contrary theoretical arguments mentioned above. Gorton and Schmid (2004) show that firms with parity co-determination trade at a

substantial discount relative to other companies. They conclude that co-determination provides a

1

The law for small stock corporations is still valid in the One-third Co-determination Act of 2004. 2

The legislator defines, according to the law for small stock corporations or One-third Co-determination Act, Family owned companies as firms where only one shareholder or a family owns all shares of the respective company. 3_{Furubotn (1988), page 178.}

4

(7)

binding constraint on the shareholders. These findings are supported by Schmid and Seger (1998) who also report a negative impact of labor representation. To the contrary, Fauver and Fuerst (2006) find that labor representation provides a powerful means of monitoring and reduces agency costs within the firm. Other research, as for instance by Benalli, Loderer and Lys (1987), finds no significant effect of the introduction of parity codetermination on stock prices.

These earlier studies provide only cross-sectional evidence by comparing firms with different labor representation constellations; hence, they suffer from huge endogeneity problems. A more convenient methodological approach is presented by Bermig and Frick (2010) who control for firm specific effects. However, they focus only on large companies and do not provide a consistent effect.

Therefore, one can summarize that prior research, concerning co-determination, provides

contrary results. Furthermore, most of the studies focus on the difference between parity and one third co-determination in large firms. However, no previous analysis has taken small firms (those with less than 500 employees) into account. Thus, this thesis will analyze co-determination in small firms for the first time using an exogenous shock in the law. Furthermore, this research will also make a methodological contribution as it will use a quasi-experimental research design which is more robust against endogeneity issues compared to previous cross-sectional studies. Although no prior research empirically examined the impact of co-determination on firm performance for smaller firms, hypothesis I is built on the market oriented view by Jensen and Meckling (1976). Hence, hypothesis I is formulated as follows:

Hypothesis I:

Labor representation on the supervisory board is negatively associated with firm performance.

This hypothesis is empirically tested and examined in the following part.

(8)

III Empirical Analysis

This chapter tests the aforementioned hypothesis and thus will answer the research question. The topic is analyzed using a quasi-experimental research method. That is creating two similar groups and investigating the treatment effect by comparing the “treated” with the control group.

The remainder of this part is as follows:

First, a detailed explanation of the respective econometric model used in the analysis is described, followed by the respective data collection. At the end of the chapter the corresponding findings and additional robustness checks are presented.

Research method

Quasi-Experimental Design in order to analyze the impact of

Co-determination on the firm`s performance

In this chapter two groups of very similar firms are created in order to determine the impact of co-determination on a firm`s performance.

As already stated above, the change in the law in 1994 (Law for small stock corporations) allows firms with specific characteristics to have no labor representation on their supervisory board. This rule counts for companies with less than 500 employees who were founded after 1994 or are family owned companies. However, firms with less than 500 employees and founded before 1994 (that were not family owned or refounded after 1994) are required to have labor representation on their supervisory board. This exogenous shock due to the change in the law allows forming two similar groups, namely group A with firms who do not need to have labor representation on their supervisory board and group B with firms who have labor representation on their board. Assuming that these two groups are quite similar due to their employee restriction and all other different firm-specific characteristics are randomly distributed in both groups, the impact of co-determination can be analyzed by comparing the “treated” (=co-determined) group B with the

(9)

control group A. Thus, the OLS5 regression estimation of the following equation is used in order to test the hypothesis that a company`s performance is negatively associated with labor representation on the supervisory board:

6

Several measures as proxies for firm performance (FIRMPERFORMANCE) are constructed, namely Return on Equity (ROE)7 and Return on Assets (ROA)8. Furthermore, PROFITABILITY9, a measure for a firm`s profitability is calculated. This latter measure is constructed in order to avoid that any financing or tax aspects of the respective firms will affect the results. Thus, this proxy measures a company`s sales (or revenues) less their respective costs of goods sold in relation to the required assets which where necessary to produce the output. Moreover, all regressions are performed using STATA`s robust function in order to achieve robustness against heteroskedasticity.

5_{OLS=Ordinary Least Squares.} 6

CONTROLS= Control variables. 7

Return of Equity (ROE) is defined as Net Income (NI) in 2006 divided by total average assets in 2006 (= [NI/ [total equity at the beginning of the year + total equity at the end of the year]/2]*100).

8

Return on Assets (ROA) is defined as Net Income in 2006 divided by total assets at the beginning of 2006 (= [[Net Income]/ [total assets at the beginning of the year]]*100).

9

Profitability is calculated by dividing a firm`s EBITDA (Earnings before Interests, Taxes, Depreciation and Amortization) by its total assets at the beginning of the year (= [[EBITDA] / [total assets at the beginning of the year]]*100).

(10)

Sample Data

This section will describe the sample data used for the empirical analysis.

The analysis of this quasi-experimental design will concentrate on the year 2006. Moreover, this investigation focuses on the German market with its five biggest indices: DAX, TECDAX, MDAX, SDAX and CDAX. Data from all companies belonging to one of these indices were taken into the final sample. The required data on the supervisory board structure (composition) was hand-collected from the Hoppenstedt10 database and annual reports. Accounting performance measures were obtained from the database Datastream and Thomson Reuters Worldscope and missing values were completed using the annual reports of the respective firm`s website or alternatively from the homepage of the Bundesanzeiger11 as according to the German publication-requirement “Publikationspflicht”, companies have to publish their annual reports in the Bundesanzeiger.

Additionally, all firms with more than 500 employees are dropped out of the sample as in the analysis two similar groups with less than 500 employees are investigated. Furthermore, firms with missing data variables are not taken into account in the respective analysis12. Moreover, the final sample is restricted to firms with a positive net income.13

A table of summary statistics of the final sample size can be seen in Table 1.

10

The data regarding the supervisory board structure were first obtained hand-collecting the required data items from the Hoppenstedt database. Afterwards, for all firms with missing data entries and companies who were founded before 1994, the required supervisory board structure data was collected using the annual reports. Thus, it was able to confirm the accuracy of the Hoppenstedt database as the information obtained from the database were in 95% of the cases in agreement with the annual reports.

11_{The Bundesanzeiger can be online reached via www.bundesanzeiger.de (01.08.2013 18:11).} 12

Stata 12 handles this problem automatically due to the “listwise deletion“ in, for instance, a regression analysis. Therefore, it is better to handle the problem of missing data in this way instead of dropping the whole firm out of the sample. An example could be that the data contains for a firm no entry about their respective Return on Equity while it contains an entry about their respective Return on Assets. Therefore, this firm will be considered in the analysis of Return on Assets but not in the analysis of Return on Equity.

13

The reason is that this research aims creating two groups as similar as possible. Thus, only firms with a positive net income will be taken into account as the amount of loss making firms was too low in order to be included without potentially distorting the results.

(11)

Table 1: Descriptive Statistics of the variables of firms used in the quasi-experimental design

Note: ROE, ROA and PROFITABILITY are in percentage terms and measured as described before. CODET is a dummy variable, which equals 1 if the respective firm has labor representation on the supervisory board. EMPLOYEES are measured as the logarithm of the average employees in 2006. ASSETS are measured as the logarithm of the average total assets in 2006. LEVERAGE is constructed by dividing a firm`s average total debt in 2006 by its average total equity in 2006.

VARIABLES N mean sd min max

ROE in % 107 16.23 21.77 0.53 196.29 ROA in % 108 10.70 13.11 0.24 109.94 EMPLOYEES 130 4.59 1.34 0.41 6.20 CODET 131 0.08 0.27 0.00 1.00 LEVERAGE 107 1.36 2.24 0.02 15.46 ASSETS 130 10.83 1.28 7.76 16.21 PROFITABILITY in % 128 17.13 16.20 0.40 98.37

(12)

Findings

This section will present the findings from the quasi-experimental design. The following findings will be described and analyzed separately from the different dependent variables which were taken as proxies for the firm`s performance.

ROE as dependent variable

In this analysis a firm`s return on equity (ROE)14 is taken as proxy for firm performance, which represents the dependent variable in the models. ROE is regressed against CODET, a dummy variable, which equals 1 if a firm has labor representation on their supervisory board. Furthermore, several control variables are included in the models in order to control for omitted variable bias. Theory, as demonstrated by Modigliani and Miller (1958), let suggest that ROE is positively associated with a firm`s leverage until the bankruptcy costs, i.e. the cost of financial distress, exceed the benefits of the leverage effect. Thus, a company`s leverage ratio (LEVERAGE)15 is included as control variable. Furthermore, a firm`s assets16 as proxy for the firm`s size is included into the model as well as a firm`s employees17 in order to control employee differences in the firm characteristics.

Table 2 provides an overview of the correlations of the variables used in this analysis. It is important to analyze the correlations in order to avoid multicollinearity.

14_{Return of Equity (ROE) is defined as Net Income (NI) in 2006 divided by total average assets in 2006 (= [NI/ [total} equity at the beginning of the year + total equity at the end of the year]/2]*100).

15

A firm`s leverage effect is determined by dividing the average debt in 2006 by the average equity in 2006 (=[[debt at the beginning of the year + debt at the end of the year]/2]/[[equity at the beginning of the year + equity at the end of the year]/2]).

16

A firm`s assets are measured as the logarithm of the average total assets in 2006 (= log [[total assets at the beginning of the year + total assets at the end of the year]/2]).

17_{A firm`s employees are measured as the logarithm of the average employees in 2006 (=log [[employees at the} beginning of the year + employees at the end of the year]/2]).

(13)

Table 2: Correlation overview of the respective dependent, independent as well as control variables used in the regression estimation, where ROE is taken as dependent variable, CODET as independent variable, and LEVERAGE, EMPLOYEES and ASSETS as control

variables.

ROE CODET LEVERAGE EMPLOYEES ASSETS

ROE 1,0000

CODET -0.0585 1,0000

LEVERAGE 0.4534 0.2349 1,0000

EMPLOYEES -0.1603 0.0757 -0.0477 1,0000

ASSETS -0.1746 0.1383 0.3726 0.0834 1,0000

The overview of the table of correlations shows that ROE is correlated with the leverage ratio (LEVERAGE) of the respective firm. Furthermore, LEVERAGE is correlated with ASSETS. Although this can be an indication of multicollinearity, excluding the leverage effect will lead to an omitted variable bias as it determines to an important extent a firm`s return on equity (Modigliani and Miller, 1958). Also the exclusion of a firm`s assets can lead to an omitted variable bias as this variable serves as a proxy for firm size. Thus, the regression estimation will take this indication carefully into consideration by running the regressions in different constellations in order to avoid that a possible multicollinearity problem will affect the results.18 Moreover, one can suggest that a company`s employees are highly correlated with a firm`s assets. However, this correlation is less than 10% which should indicate that here no problem of multicollinearity could arise. A higher correlation can be seen between the independent variable CODET and a firm`s assets (ASSETS). It relies around 25%, which will be taken into account during the regression estimation and analysis.

18

Although the correlation is not too high, a regression of LEVERAGE on ASSETS was performed in order to identify its coefficient of determination (which equals 0.2625). Using this result, the tolerance (T=1- coefficient of determination) is calculated: T=0.7375. This is an indication that no multicollinearity between these two variables should arise as this value (T) is above 0.2. This analysis has not provided any new insights and thus, is omitted to save space.

(14)

Table 319 presents the results of the OLS20 regression estimation which tests the following equation:

(1)

Equation 1 will be fully tested in model 4, while in model 3 a firm`s assets will be excluded to test if the findings are in this case also robust because there is a positive correlation between the variables ASSETS and LEVERAGE. Model 2 tests equation 1 as well but excludes EMPLOYEES to see if the estimation results would be affected as one can suggest a high correlation between a firm`s assets and employees. However, the findings are not affected by the aforementioned reasons and show a negative association between firm`s performance and Co-determination (CODET) in all models. Furthermore, model 1 takes just a firm`s leverage ratio into account and neglects the other control variables for firm size and characteristics.

The results indicate that Co-determination (CODET) is in all cases negatively associated with the firm performance if Return on Equity is taken as proxy for firm performance. Table 5 shows that if more control variables are included, the coefficient as well as the standard deviation of CODET slightly declines. However, the findings show that if a firm is co-determined, it has on average 15% less return on equity compared to firms with no co-determination on the supervisory board. These findings are statistically significant at 5% (for models 1 and 3) and 1% level (for models 2 and 4), depending on the regarded model.21

19_{The output of all regression tables were created using outreg2 as add-in for STATA 12.Outreg2 is based on outreg} written by John Gallup.

20

OLS= Ordinary Least Squares. 21

The analysis will not take the interpretation of the other independent variables into consideration. This is due to two reasons: First, this thesis aims to only examine the relationship between co-determination and firm performance and second, all other independent variables are just taken as control variables.

(15)

Table 3: OLS Regression estimation of Return on Equity on Codetermination

OLS regressions of Return on Equity (ROE) on Codetermination (CODET) and control variables.

Note: ROE is measured in percentage terms. Models 1 until 4 provide the results of the regression estimation of Return on Equity (ROE), defined as a firm`s net income divided by its average equity in 2006 ([total equity at the beginning of the year + total equity at the end of the year]/2), on Codetermination (CODET), a dummy variable which equals 1 if a firm has labor representation on their supervisory board, as well as several control variables. Model 1 includes only LEVERAGE, Model 2 includes LEVERAGE and ASSETS and Model 3 includes LEVERAGE and EMPLOYEES as control variables. Model 4 estimates the regression including all control variables. *, **, *** imply statistical significance at the 0.1, 0.05, and 0.01 levels, respectively.

(1) (2) (3) (4)

VARIABLES ROE in % ROE in % ROE in % ROE in %

CODET -17.78** -15.59*** -16.66** -14.92*** (7.67) (5.94) (6.99) (5.62) LEVERAGE 4.23** 5.42*** 4.15** 5.35*** (1.78) (1.42) (1.75) (1.41) ASSETS -8.09** -8.05** (3.44) (3.44) EMPLOYEES -2.47 -1.74 (1.51) (1.26) Constant 11.26*** 96.82** 22.58*** 104.51*** (3.11) (37.52) (8.33) (39.16) Observations 107 107 106 106 R-squared 0.23 0.36 0.25 0.37

(16)

ROA as dependent variable

In this analysis, Return on Assets (ROA) is taken as dependent variable in order to analyze the impact of co-determination on firm performance. The reason is, that Return on Equity (ROE) is determined to a huge amount by the respective firm`s leverage. Although the previous analysis has controlled for this fact, this part will analyze the relation between co-determination and a firm`s performance using the ratio Return on Assets (ROA) as proxy for firm performance as this variable does not depend on a company`s leverage ratio.22

Despite the dependent (ROA)23 and independent variable (CODET)24, a firm`s assets (ASSETS)25 as proxy for the firm`s size, as well as a firm`s employees (EMPLOYEES)26, in order to control for employee differences in the firm characteristics are included into the model as control variables.

Table 4 provides an overview of the correlations of the variables used in this analysis. This table is analyzed regarding high correlations in order to avoid mulicollinearity. However, Table 4 presents no high correlations between the independent variables. This indicates that no problem of multicollinearity may arise.

The results of the OLS regression estimation of equation 2 are presented in Table 5. Equation 2 will be fully tested in model 4, while model 1 contains no control variables. Moreover, model 2 includes ASSETS as control variable and model 3 EMPLOYEES, respectively. However, all models show a negative association between co-determination (CODET) and a firm`s Return on Assets (ROA).

22_{There may be a correlation between ROA and LEVERAGE as a higher LEVERAGE leads usually to higher paid} interests which indeed affects the net income. However,

23

Return on Assets (ROA) is defined as Net Income in 2006 divided by total assets at the beginning of 2006 (= [[Net Income]/ [total assets at the beginning of the year]]*100).

24

CODET: a dummy variable, which equals 1 if a firm has labor representation on their supervisory board. 25

A firm`s assets are measured as the logarithm of the average total assets in 2006 (= log [[total assets at the beginning of the year + total assets at the end of the year]/2]).

26_{A firm`s employees are measured as the logarithm of the average employees in 2006 (=log [[employees at the} beginning of the year + employees at the end of the year]/2]).

(17)

Table 4: Correlation overview of the respective dependent, independent as well as control variables used in the regression estimation, where ROA is taken as dependent variable, CODET as independent variable, and LEVERAGE, EMPLOYEES and ASSETS as control

variables.

ROA CODET EMPLOYEES ASSETS

ROA 1,0000

CODET -0.1601 1,0000

EMPLOYEES -0.1659 0.0927 1,0000

ASSETS -0.2752 0.1634 0.0895 1,0000

The results indicate that co-determination (CODET) is in all cases negatively related with the firm`s performance if Return on Assets (ROA) is taken as proxy for firm performance. Moreover, Table 5 shows that if more control variables are included, the coefficient of CODET slightly declines. However, the findings predict that if a firm is co-determined, it performs on average 5 until 6% (depending on which control variables are included into the model) worse compared to firms who do not have labor representation on their supervisory board. These findings are statistically significant at the 1% level for models 1 until 3 and at the 5% level for model 4, respectively.27

27

Note that all the regressions from Table 5 were also performed using the average assets (in 2006) in the denominator in order to determine the Return on Assets (ROA) ratio. The results were statistically significant and do not provide new insights. Thus, they are omitted to save space. However, Appendix 1 presents this table in order to show that the findings are not affected by the choice of measuring the denominator.

(18)

Table 5: Regression estimation of Return on Assets on Codetermination

OLS Regression Estimation of Return on Assets (ROA) on Codetermination (CODET) and Control variables.

Note: ROA is measured in percentage terms. Models 1 until 4 provide the results of the regression estimation of Return on Assets (ROA), defined as a firm`s net income divided by its total assets at the beginning of the year, on Codetermination (CODET), a dummy variable which equals 1 if a firm has labor representation on their supervisory board, as well as several control variables. Model 1 regresses ROA on CODET without any control variables, while model 2 includes ASSETS as control variable and model 3 includes EMPLOYEES, respectively. Model 4 estimates the regression including all control variables. *, **, *** imply statistical significance at the 0.1, 0.05, and 0.01 levels, respectively.

(1) (2) (3) (4)

VARIABLES ROA in % ROA in % ROA in % ROA in %

CODET -7.51*** -5.56*** -6.89*** -5.08** (1.74) (1.95) (1.56) (1.99) ASSETS -2.54 -2.53 (1.60) (1.63) EMPLOYEES -1.47 -1.29 (0.89) (0.82) Constant 11.32*** 38.60** 18.03*** 44.33** (1.36) (18.14) (4.89) (19.59) Observations 108 108 107 107 R-squared 0.03 0.09 0.05 0.11

(19)

PROFITABILITY as dependent variable

In this part of the analysis, PROFITABILITY28 is taken as dependent variable and therefore serves as proxy for firm performance. Both measures from the previous analysis, ROE and ROA, use a firm`s net income in the numerator in determining its respective ratio. By taking instead a firm`s EBITDA, this provides the advantage that it neglects a company`s financing decisions (no paid interests). Additionally, the measure neglects discrepancies in the effective tax rate due to differences between the tax and accounting books of the respective companies (e.g. accelerated depreciation in the tax balance sheet versus linear depreciation in the accounting balance sheet). All other independent variables (CODET and control variables) are taken analogous to the previous analysis where ROA is taken as dependent variable. Thus, the following equation is estimated:

(3)

Table 6 provides a correlation-overview of all variables used in this analysis. However, Table 6 presents no high correlations between the independent variables. This indicates that no problem of multicollinearity should arise.

The results of the OLS regression estimation of equation 3 are presented in Table 7. Equation3 will be fully tested in model 4, while model 1 contains no control variables. Moreover, model 2 includes ASSETS as control variable and model 3 EMPLOYEES, respectively. The results indicate that co-determination (CODET) is in all cases negatively associated with PROFITABILITY. Additionally, Table 5 shows, depending on the model (and thus, which control variables are included), that co-determined firms have between 7 and 9 % less PROFITABILITY compared to firms who do not have labor representation on their supervisory board. The findings predict again a worse performance compared to firms not-co-determined

28

Profitability is calculated by dividing a firm`s EBITDA (Earnings before Interests, Taxes, Depreciation and Amortization) by its total assets at the beginning of the year (= [[EBITDA] / [total assets at the beginning of the year]]*100).

(20)

firms. These findings are statistically significant at the 1% level for models 1, 3 and 4 and at the 5% level for model 2, respectively.2930

Table 6: Correlation overview of the respective dependent, independent as well as control variables used in the regression estimation, where PROFITABILITY is taken as dependent

variable, CODET as independent variable, and LEVERAGE, EMPLOYEES and ASSETS as control variables.

PROFITABILITY CODET EMPLOYEES ASSETS

PROFITABILITY 1,0000 CODET -0.1228 1,0000 EMPLOYEES 0.1272 0.0843 1,0000 ASSETS -0.1653 0.1765 0.0842 1,0000 29

Note that all the regressions from Table 7 were also performed using the average assets (in 2006) in the denominator in order to determine the PROFITABILITY ratio. The results were (in all models) statistically significant and do not provide new insights. Thus, they are omitted to save space. However, Appendix 2 presents this table in order to show that the findings are not affected by the choice of measuring the denominator.

30

Note that all the regressions from Table 7 were also run using a firm`s EBIT (Earnings before Interests and Taxes), instead of EBITDA, in the numerator in order to determine the PROFITABILITY ratio. The results were in all cases statistically significant and are presented in Appendix 3. However, no new insights can be taken from this analysis and therefore, omitted to save space.

(21)

Table 7: Regression estimation of PROFITABILITY (EBITDA/ASSETS) on Codetermination

OLS regression estimation of PROFITABILITY on CODET and control variables.

Note: PROFITABILITY is measured in percentage terms. Models 1 until 4 provide the results of the regression estimation of PROFITABILITY, defined as a firm`s EBITDA divided by its total assets at the beginning of the year, on Codetermination (CODET), a dummy variable which equals 1 if a firm has labor representation on their supervisory board, as well as several control variables. Model 1 regresses PROFITABILITY on CODET without any control variables, while model 2 includes ASSETS as control variable and model 3 includes EMPLOYEES, respectively. Model 4 estimates the regression including all control variables. *, **, *** imply statistical significance at the 0.1, 0.05, and 0.01 levels, respectively.

(1) (2) (3) (4)

VARIABLES PROFITABILITY in % PROFITABILITY in % PROFITABILITY in % PROFITABILITY in %

CODET -8.65*** -6.83** -9.53*** -7.61*** (2.44) (2.72) (2.40) (2.50) ASSETS -2.14 -2.38 (1.70) (1.73) EMPLOYEES 1.91 2.06 (1.78) (1.76) Constant 16.59*** 39.66** 8.00 32.92* (1.82) (18.51) (9.09) (18.20) Observations 128 128 127 127 R-squared 0.01 0.04 0.03 0.06

(22)

Discussion & Robustness checks

The findings from the previous part show a negative association between firm performance and co-determination, irrespectively of which dependent variable, as proxy for firm performance, is chosen. These findings are, regardless of the model, statistically significant at least at the 5% level (often also at the 1 % level). Thus, the null hypothesis, that labor representation is not negatively related to firm`s performance, can be rejected and the developed hypothesis can be accepted.

This part of the empirical analysis examines whether the findings are robust against small modifications. Therefore, table 8 presents the findings of all three different dependent variables when they are winsorized at the 10% level31. The intuition behind this robustness check is to eliminate the possibility that huge outliers have influenced the findings. Models 1 till 3 regress the respective dependent variable on CODET without any control variables, while models 4 until 6 provide the regression estimation of the respective dependent variables on CODET and all established control variables as in the previous analysis. Table 8 shows that especially the estimated Return on Equity (ROE) is highly overestimated taking the winsorized values into account. Here the effects of outliers have impacted the results to a high extent. Although the estimated regressions are in all models still statistically significant, winsorizing outliers reduces the assessed ROE by more than 10%. However, all winsorized models show still a negative association between firm performance and co-determination (CODET). Winsorizing outliers can reduce the estimated coefficients to a certain extent and also the other two winsorized dependent variables (W_ROA and W_PROFITABILITY) reduce the estimated coefficients, but here by less than 1%.32 Thus, this robustness check shows that the estimated coefficients for Return on Assets (ROA) and PROFITABILITY are robust against outliers while the impact of CODET on Return on Equity (ROE) is overestimated.33

31

This was done using STATA`s add in winsor. Winsorized at the 10% level means that the 10% highest and lowest values of the respective variable are replaced by the next highest/lowest data entry. It has the advantage that on the one hand it eliminates the influence of huge outliers while on the other hand it takes big values into account compared to simply eliminating outliers (for instance using STATA`s add-in function truncateJ).

32

Except of model 2 as here the estimated coefficient of W_ROA is reduced by 2.3%.

33_{The same regression table was also created winsorizing at the 5% level. However, the results do not provide any} new insights and are thus, omitted to save space. The results can be found in Appendix 4.

(23)

Table 8: OLS regressions of winsorized Firm performance on Codetermination

OLS regression estimation of winsorized (at the 10% level) firm performance on CODET and control variables.

Note: FIRMPERFORMANCE is measured in percentage terms. Models 1 until 6 provide the results of the regression estimation of winsorized (at the 10% level) Return on Equity (W_ROE), winsorized (at the 10% level) Return on Assets (W_ROA) and winsorized (at the 10% level) PROFITABILITY (all latter three variables are measured as in the previous analysis) on Codetermination (CODET), a dummy variable which equals 1 if a firm has labor representation on their supervisory board, as well as several control variables. Models 1 till 3 regresses the respective dependent variable on CODET without any control variables, while models 4 until 6 provide the regression estimation of the respective dependent variables including all established variables as in the previous analysis. *, **, *** imply statistical significance at the 0.1, 0.05, and 0.01 levels, respectively.

(1) (2) (3) (4) (5) (6)

VARIABLES W_ROE in % W_ROA in % W_PROFITABILITY in % W_ROE in % W_ROA in % W_PROFITABILITY in %

CODET -3.73* -5.21*** -7.89*** -4.84** -4.30*** -6.97*** (2.10) (1.14) (1.84) (2.18) (1.23) (1.90) LEVERAGE 0.61*** (0.18) ASSETS -0.68 -0.93** -1.65** (0.72) (0.46) (0.70) EMPLOYEES -0.52 -0.54 0.96 (0.65) (0.48) (0.68) Constant 13.72*** 9.41*** 16.02*** 22.91*** 21.94*** 29.49*** (0.83) (0.63) (1.04) (7.77) (5.19) (8.21) Observations 107 108 128 106 107 127 R-squared 0.01 0.05 0.04 0.07 0.11 0.08

(24)

Moreover, the law determines that firms founded after 1994 do not need to have labor representation on their supervisory board. However, firms founded before 1994 are required to have one third labor representation on the supervisory board except of family owned companies. This implication allows forming even more similar groups, namely group C with firms founded before 1994 who needed to have labor representation on their supervisory board and group D with companies founded before 1994 who are exempted from this rule as they have the status of a family owned company or were refunded after 1994.

None of the firms founded after 1994, although voluntary possible, are co-determined. Hence, one possible disadvantage of the formed groups from the previous analysis and sample can be that no co-determined firm was founded after 1994. Thus, the new created groups will not suffer from the problem that younger aged firms are in the group of not co-determined firms. Hence, the similarity of the new created groups rises. The intuition behind this robustness check is to show that the findings of the previous analysis are still valid and the same compared to those of the new and more identical formed groups with more restricted implications. Table 9 provide the results of the regression estimations of the respective dependent variables on CODET and the established control variables from the previous analysis with the restriction that only firms originally founded before 1994 are taken into account. Moreover, the previous robustness check has shown that outliers affect the regression estimation of the CODET coefficient when Return on Equity serves as dependent variable. Hence, this analysis takes also the winsorized values34 in order to avoid the aforementioned problem. Again models 1 until 3 regress the respective dependent variable on CODET without any control variables, while models 4 until 6 provide the regression estimation of the respective dependent variables on CODET and all established control variables as in the previous analysis.

The results of table 9 show that the findings are nearly identical to those from table 8. All models show, statistically significant, a negative association between firm performance and co-determination (CODET). Furthermore, the estimated coefficients for CODET differ on average less than +/-1% compared to those, estimated in table 8. Hence, the estimated regressions from this analysis verify the previous findings.

34_{At the 10% level. A regression estimation of the same constellation without winsorizing can be found in Appendix} 5. However, the results do not provide any new results compared to Table 9 and thus, is omitted to save space.

(25)

Table 9: OLS regressions of Firm performance on Co-determination with Firms founded before 1994

Note: FIRMPERFORMANCE is measured in percentage terms and the sample is restricted to firms founded before 1994. Models 1 until 6 provide the results of the regression estimation of winsorized (at the 10% level) Return on Equity (W_ROE), winsorized (at the 10% level) Return on Assets (W_ROA) and winsorized (at the 10% level) PROFITABILITY (all latter three variables are measured as in the previous analysis) on Codetermination (CODET), a dummy variable which equals 1 if a firm has labor representation on their supervisory board, as well as several control variables. Models 1 till 3 regresses the respective dependent variable on CODET without any control variables, while models 4 until 6 provide the regression estimation of the respective dependent variables including all established variables as in the previous analysis. *, **, *** imply statistical significance at the 0.1, 0.05, and 0.01 levels, respectively.

(1) (2) (3) (4) (5) (6)

CODET -4.39* -5.56*** -6.84*** -5.39** -4.96*** -5.76** (2.30) (1.35) (2.08) (2.35) (1.50) (2.22) LEVERAGE 0.39 (0.46) ASSETS -0.17 -0.93 -2.18*** (1.30) (0.62) (0.79) EMPLOYEES 0.09 0.01 1.21* (0.85) (0.59) (0.72) Constant 14.38*** 9.76*** 14.97*** 15.46 19.79*** 33.20*** (1.22) (0.95) (1.41) (13.08) (6.82) (8.82) Observations 60 61 73 59 60 72 R-squared 0.03 0.09 0.05 0.05 0.12 0.15

(26)

IV Conclusion

Prior Literature examines the impact of co-determination on firm performance in several studies. However, most of previous research focuses on the difference between one third and parity co-determination. The contribution of this thesis is twofold as it investigates on the one hand the impact of labor representation on the supervisory board for smaller firms for the first time and on the other, it uses a quasi-experimental research design that do not suffer much from endogeneity issues compared to most of the other cross-sectional studies.

The thesis has created two similar groups in order to determine the influence of co-determination on firm performance. Furthermore, several measures for firm performance are constructed in order to avoid limiting this research to one measurement. The results show a negative association between labor representation and firm performance regardless of which measure for the success of a firm is taken. Therefore, co-determined firms perform worse than companies without labor representation on the supervisory board. These results are robust against outliers except of if Return on Equity (ROE) is taken as proxy for firm performance. Although the winsorized regression estimations show an overestimation of the impact of co-determination on a firm`s Return on Equity (ROE), the results present also a negative relation with statistical significance. Additionally, the results are verified using even more restricted implications by forming the two groups. Hence, the investigation shows that co-determination has a negative influence on firm performance. Another indication which supports these findings is that no firm, in the sample, appoints labors to the supervisory board on a voluntary basis. Therefore, firms try to avoid co-determination which can also be seen by the fact that most of the firms founded before 1994 are refounded in a year after 1994. Thus, it is questionable why the legislator does not provide the same opportunity to all firms with less than 500 employees irrespectively of when the company was founded. One can bring forward the argument that those firms have learned to accept labors on the supervisory board, but this may just be an explanation that does not lie on well-grounded facts. Hence, one possibility for further research could be to investigate whether co-determination also provides advantages as for instance in economic crisis or in countries where the trust in supervisory boards lack of reliability and authenticity.

(27)

V Literature

Adams, R. B., Hermalin, B. and Weisbach, M. (2010): ‟The Role of Boards of Directors in Corporate Governance: A Conceptual Framework and Survey‟, Journal of Economic Literature 48, 58-107.

Baums T. and Frick B. (1999): „The Market Value of the Codetermined Firm‟ in Blair M. M. and Roe M. J. (eds.) Employees and Corporate Governance, Washington, DC: Brookings Inst. Press: 194–205.

Benelli G., Loderer C. and Lys T. (1987): „Labor Participation in Corporate Policymaking Decisions: West Germany‟s Experience with Codetermination‟, The Journal of Business 60: 553–575.

Bermig, A. and Frick, B. (2010): „Board size, board composition and firm performance: Empirical evidence from Germany„, working paper, available at

http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1623103 (02.08.2013 17:46).

Fauver L. and Fuerst M. E. (2006): „Does Good Corporate Governance Include Employee Representation? Evidence from German Corporate Boards‟, Journal of Financial Economics 82: 673–710.

Frick, B. and Lehmann, E.E. (2005): „Corporate Governance in Germany: Ownership, Codetermination and Firm Performance in a Stakeholder Economy„, in H. Gospel, and A. Pendleton, eds.: Corporate Governance and Labour Management: An International Comparison (Oxford University Press Oxford).

Furubotn, E. G. (1988): „Codetermination and the Modern Theory of the Firm: A Property-Rights Analysis‟, The Journal of Business 61, 165-181.

(28)

Gorton, G. and Schmid, F.A. (2004): ‟Capital, labor, and the firm: A study of German codetermination‟, Journal of the European Economic Association 2, 863-905.

Hermalin, B. and Weisbach, M. (2003): „Boards of directors as an endogenously determined institution: A survey of the economic literature„, Economic Policy Review, 7-26.

Jensen, M. C. and Meckling,W.H. (1979): „Rights and Production Functions: An Application to Labor-Managed Firms and Codetermination„, The Journal of Business 52, 469-506.

Modigliani, F. and Miller, M.H. (1958): „The cost of capital, corporation finance, and the theory of investment„, American Economic Review 48, 655–669.

Schmid, F. A. and Seger, F., (1998): „Arbeitnehmermitbestimmung, Allokation von

Entscheidungsrechten und Shareholder Value„, Zeitschrift für Betriebswirtschaft (ZfB) 453-473.

Shleifer, A. and Vishny, R.W. (1997): „A Survey of Corporate Governance„, The Journal of Finance 52, 737-783.

(29)

(30)

Appendix 1: Regression estimation of Return on Assets on Codetermination

OLS Regression Estimation of Return on Assets (ROA) on Codetermination (CODET) and Control variables.

Note: ROA is measured in percentage terms. Models 1 until 4 provide the results of the regression estimation of Return on Assets (ROA), defined as a firm`s net income divided by its average total assets ([total assets at the beginning of the year + total assets at the end of the year]/2), on Codetermination (CODET), a dummy variable which equals 1 if a firm has labor representation on their supervisory board, as well as several control variables. Model 1 regresses ROA on CODET without any control variables, while model 2 includes ASSETS as control variable and model 3 includes EMPLOYEES, respectively. Model 4 estimates the regression including all control variables. *, **, *** imply statistical significance at the 0.1, 0.05, and 0.01 levels, respectively.

(1) (2) (3) (4)

VARIABLES ROA in % ROA in % ROA in % ROA in %

CODET -5.55*** -3.71** -5.11*** -3.38* (1.43) (1.74) (1.29) (1.77) ASSETS -2.41** -2.40** (1.17) (1.19) EMPLOYEES -1.06 -0.89 (0.68) (0.61) Constant 9.13*** 34.95*** 13.96*** 38.99*** (1.02) (13.20) (3.73) (14.19) Observations 108 108 107 107 R-squared 0.02 0.12 0.05 0.14

(31)

Appendix 2: Regression estimation of PROFITABILITY (EBITDA/ASSETS) on Codetermination OLS regression estimation of PROFITABILITY on CODET and control variables.

Note: PROFITABILITY is measured in percentage terms. Models 1 until 4 provide the results of the regression estimation of PROFITABILITY, defined as a firm`s EBITDA divided by its average total assets ([total assets at the beginning of the year + total assets at the end of the year]/2), on Codetermination (CODET), a dummy variable which equals 1 if a firm has labor representation on their supervisory board, as well as several control variables. Model 1 regresses PROFITABILITY on CODET without any control variables, while model 2 includes ASSETS as control variable and model 3 includes EMPLOYEES, respectively. Model 4 estimates the regression including all control variables. *, **, *** imply statistical significance at the 0.1, 0.05, and 0.01 levels, respectively.

(1) (2) (3) (4)

CODET -4.91** -3.69* -5.58** -4.05* (2.29) (2.15) (2.28) (2.23) ASSETS -1.55 -2.21* (1.15) (1.16) EMPLOYEES 0.02 0.02* (0.01) (0.01) Constant 12.38*** 29.17** 9.34** 32.30** (1.75) (13.25) (3.75) (12.89) Observations 128 128 128 128 R-squared 0.01 0.02 0.02 0.05

(32)

Appendix 3: Regression estimation of PROFITABILITY (EBIT/ASSETS) on Codetermination

OLS regression estimation of PROFITABILITY on CODET and control variables.

Note: PROFITABILITY is measured in percentage terms. Models 1 until 4 provide the results of the regression estimation of PROFITABILITY, defined as a firm`s EBIT divided by its total assets at the beginning of the year, on Codetermination (CODET), a dummy variable which equals 1 if a firm has labor representation on their supervisory board, as well as several control variables. Model 1 regresses PROFITABILITY on CODET without any control variables, while model 2 includes ASSETS as control variable and model 3 includes EMPLOYEES, respectively. Model 4 estimates the regression including all control variables. *, **, *** imply statistical significance at the 0.1, 0.05, and 0.01 levels, respectively.

(1) (2) (3) (4)

CODET -6.44*** -5.48** -6.88*** -5.84** (2.28) (2.66) (2.14) (2.41) ASSETS -1.14 -1.29 (1.60) (1.65) EMPLOYEES 0.86 0.94 (1.77) (1.78) Constant 12.73*** 24.98 8.91 22.44 (1.69) (17.26) (9.10) (17.08) Observations 128 128 127 127 R-squared 0.01 0.02 0.01 0.02

(33)

Appendix 4: OLS regressions of winsorized Firm performance on Codetermination

Note: FIRMPERFORMANCE is measured in percentage terms. Models 1 until 6 provide the results of the regression estimation of winsorized (at the 5% level) Return on Equity (W_ROE), winsorized (at the 5% level) Return on Assets (W_ROA) and winsorized (at the 5% level) PROFITABILITY (all latter three variables are measured as in the previous analysis) on Codetermination (CODET), a dummy variable which equals 1 if a firm has labor representation on their supervisory board, as well as several control variables. Models 1 till 3 regresses the respective dependent variable on CODET without any control variables, while models 4 until 6 provide the regression estimation of the respective dependent variables including all established variables as in the previous analysis. *, **, *** imply statistical significance at the 0.1, 0.05, and 0.01 levels, respectively.

(1) (2) (3) (4) (5) (6)

CODET -4.72** -6.25*** -8.62*** -5.86** -5.05*** -7.58*** (2.35) (1.32) (2.03) (2.48) (1.41) (2.07) LEVERAGE 0.74*** (0.23) ASSETS -1.12 -1.07* -1.82** (0.88) (0.63) (0.86) EMPLOYEES -0.70 -0.96 1.01 (0.77) (0.67) (0.81) Constant 14.28*** 10.17*** 16.55*** 28.78*** 26.12*** 31.64*** (0.99) (0.81) (1.22) (9.80) (7.02) (9.96) Observations 107 108 128 106 107 127 R-squared 0.02 0.05 0.03 0.08 0.11 0.07

(34)

Appendix 5: OLS regressions of Firm performance on Co-determination with Firms founded before 1994

OLS regression estimation of firm performance on CODET and control variables.

Note: FIRMPERFORMANCE is measured in percentage terms and the sample is restricted to firms founded before 1994. Models 1 until 6 provide the results of the regression estimation of Return on Equity (ROE), Return on Assets (ROA) and PROFITABILITY (all latter three variables are measured as in the previous analysis) on Codetermination (CODET), a dummy variable which equals 1 if a firm has labor representation on their supervisory board, as well as several control variables. Models 1 till 3 regresses the respective dependent variable on CODET without any control variables, while models 4 until 6 provide the regression estimation of the respective dependent variables including all established variables as in the previous analysis. *, **, *** imply statistical significance at the 0.1, 0.05, and 0.01 levels, respectively.

(1) (2) (3) (4) (5) (6)

VARIABLES ROE in % ROA in % PROFITABILITY in % ROE in % ROA in % PROFITABILITY in %

CODET -6.13** -7.41*** -6.99** -6.90** -6.39*** -6.00* (2.79) (1.83) (3.12) (2.85) (1.89) (3.31) LEVERAGE 0.52 (0.52) ASSETS -0.87 -1.34 -3.09* (1.66) (1.01) (1.66) EMPLOYEES -0.20 -0.29 2.97 (1.20) (0.92) (2.37) Constant 15.52*** 11.23*** 14.93*** 25.28 27.04** 35.15*** (1.69) (1.48) (2.66) (17.20) (11.06) (13.21) Observations 60 61 73 59 60 72 R-squared 0.03 0.07 0.02 0.05 0.10 0.10