Women on top as it schould be?

(1)

WOMEN ON TOP

AS IT SHOULD BE?

Name: W.P.H. Kramer Student number: 10105018

Bachelor: Economics and Business

Specialization Finance and Organization

Bachelor thesis (12 EC)

Supervisor: Ms. Eszter Czibor, MSc. 1

(2)

1

1. Introduction

A recent change of regulations in The Netherlands resulted in a new diversity quota, which has been implemented on January 1st 2013. This new law states that big firms are obligated to have a board that consists of at least 30% of each gender. If this target is not met, there is an obligation for firms to give an explanation in their annual report why the target is not achieved. Furthermore, they need to answer the question what measures the firm takes to realize this target in the future. There are some reasons given by the institutions for this new law. The reason of this quota is that there are few women on top-positions in The Netherlands compared to other countries and this number would not grow without intervention. Secondly, it has been mentioned that research showed that less diversity in the board results in worse financial results (PWC, 2013).

Firms always want to provide shareholders optimal results and therefore they look at all aspects and try to find more efficient solutions when needed. Knowing that the board makes the most important decisions for a firm, the board should be composed in the best way. The question is whether more women in the board actually improve financial results. The board and other important functions were traditionally filled in by males but the percentage of women increased over the last decades. Sometimes this was enforced, sometimes not. In several European countries a gender diversity quota already exists, such as in Iceland, Norway and Spain. Also, in many other countries the debate about this topic has been started as well. But is there clear evidence that a higher number of women in the board leads to a better performance of firms? The answer on this question regarding previous research is ambiguous.

Several researches about this question have been done in the past all over the world. These gave no clear unanimous answer and the discussion about the actual relationship continued until today. Why are there ambiguous results for this question? A lot of the researches were conducted in the U.S. and less in Europe, so different countries can be a reason. Another reason is that these researches have been done in different time-periods which could influence results. Finally, there are different estimation methods and control variables used in the studies, which makes the comparison of results difficult. Looking towards The Netherlands, there is almost no empirical research done at all for this topic and therefore this is a good argument to make this thesis which could give new and extra insights for the situation in The Netherlands. Another good argument is the recent change in regulation stimulating the number of women in the board. The assertion that this gender quota will give positive effects can be tested now. In

(4)

3

order to complement the ambiguous results of past research, the following research question will be analyzed:

“How is the performance of Dutch firms affected by the number of females in the board?”

The research method to find an answer for this question is based on a model which will be built and based on the ordinary least squares method. The firm performance measures are the return on assets (ROA) and the natural logarithm of Tobin’s Q, which are used in many other previous studies as well. Using both an accounting measure (ROA) and a non-accounting measure (Tobin’s Q) gives better results. To account for omitted variable bias, some control variables are included in the model, such as board size, firm size and the age of the firm. Another new relevant variable introduced by Hoogendoorn et al. (2013), the %-share of females in the board squared, will be included too. The final variable included is based on the study of Smith et al. (2006) and is the squared age of the firm. These two squared variables are included because they allow the possibility of a non-linear relationship of these two variables with firm performance. An important explanatory variable, the dummy variable for female presence, is included too. Comparisons are made between this variable and the other explanatory variable, the female share, to see whether there are significant different results between these two. The models are also analyzed with a fixed effects regression. Some variables are “invisible” and cannot be measured. To control for this, company and time fixed effects are added, which are discussed more detailed in part 3. Finally, the estimation of the quota-effect is discussed. This research is built up in the following way. Section 2 provides insight in previous literature considering diversity and the relevant aspects that belong to diversity. At the end the hypothesis is discussed. Part 3 explains the research method including the model and data used and which methodology is used to test the hypothesis. Part 4 shows the descriptive results, empirical results and gives an analysis of these results. Finally, the conclusion and discussion is included in part 5.

(5)

4

2. Literature review

2.1 The board and the agency problem

The topic gender diversity in the board of Dutch firms will be discussed in this research. There are different types of diversity for the board: gender, ethnicity, age, schooling, size and independent board members are examples of variables that can be used to find out how diverse a board actually is. The board is an important aspect of every firm because it influences the firm value through the decisions it implements. The tasks of the board are described by Fama and Jensen (1983) in two parts. The first task mentioned is to give advice and help the management to direct them towards the right decisions aligned with the firm strategy. Secondly the board needs to minimize the agency problem between shareholders and management. Another definition of the task of a board is given by Hendrikse (2003), which implies the board represents the shareholders and should manage in the interest of them.

The connection between the board and firm value is most of the times understood by applying the agency theory (Carter et al., 2003). Therefore, the agency theory is explained more in detail below. Agency theory states that there is a relationship between persons in which the principal (the owners: the shareholders) uses an agent (the management) to do specified tasks. This relationship can create undesirable behavior which comes from the fact that there is a separation of ownership and control (Hendrikse, 2003, p. 91). The consequence of this separation is that a principal often has no time to supervise or control the agent. Because the agent has to do the task, he receives decision rights. Both the agent and the principal want to maximize their payoff which leads to inefficient decisions for the overall firm performance. This is often displayed through decisions by the agent because his goal is to maximize his own payoff and not to maximize the value of the company he works for (Klein, 1998). Hence, both parties are self-interested which prevents an optimal cooperation with optimal outcomes. A situation of conflicting interests and asymmetric information can be solved by controlling the agent. Alignment of the interests of both parties is therefore very important. If alignment is missing, the agent needs to be controlled and monitored. This means the agent is being monitored when he is working to assure the desired behavior. The agent knows this and then he provides desirable behavior or he quits the job. However, monitoring employees gives extra costs so the net result of it might be negative in the end (Hendrikse, 2003, p. 92).

The goal of agency theory is to solve the agency problem by maximizing alignment of interests and minimizing agency costs that result from asymmetric information. As stated above,

(6)

5

the alignment should be done by the board to maximize the firm value (Fama and Jensen, 1983). The link with board and firm value is provided now. The management is monitored by the board for the shareholders (Tirole, 2005, p. 29). Hence board members want to become an expert in monitoring. The composition of the board is crucial to solve agency problems and create the optimal firm value. The question is whether a more diverse board is the solution to get a board that is a better monitoring unit and is less likely to be influenced by shareholders’ interests (Carter et al., 2003). Hence, is diversity in the board actually better for the performance of a firm?

2.2 Board diversity

As stated in the first paragraph the composition of a board is very important for every firm. The board is an important aspect of every firm because it influences the firm value through the decisions it implements. There are several studies that look at the question whether diversity of the board has a significant influence on the firm performance. Different studies give also different definitions of the word diversity so we should be careful with making any statements about diversity. Researchers often look at the variables gender and minority to check for board diversity (Carter et al., 2003). Diversity is explained by Rose (2007) by looking at the variables gender, education, ethnicity, proportion of foreigner’s etcetera; and board diversity is also connected to the debate about equal chances for men and women. Finally, in corporate governance diversity of the board can be seen as a combination of other variables. Quoting Van der Walt and Ingley (2003): “The concept of diversity relates to board composition and the varied combination of attributes, characteristics and expertise contributed by individual board members in relation to board process and decision making” (p. 219). There has been some research to the connection of board diversity and firm performance. Taken all together, these studies show ambiguous results. However, this thesis will not further concentrate on diversity in general but will analyze a more specific aspect of diversity: gender diversity and with that the relationship between gender diversity and firm performance. Previous results about this more specific topic are included in part 2.5: Gender diversity in boards: empirical results.

(7)

6

2.3 A theoretical review of the advantages and disadvantages of board

diversity

Many theories and arguments are given about why a firm should have diverse board or why not. A lot of these researches state that no diversity brings along ethical and economic concerns. One of them is that no diversity is a waste of all the potential talent that is available. Because of the fact firms have an ideal picture of a board-member; a part of the society is always excluded because they do not fulfill the asked profile. Some types of people with specific characteristics which might be beneficial are not present in board. In that case firms could perform better (Grosvold et al., 2007). Diversity could increase the creative and innovative ideas when a decision needs to be made because diversity brings along differences in views, opinions, characteristics, experiences and competencies among the board members (Flood et al., 1999). Another advantage of diversity is the argument that different types of people (ethnicity, gender, cultural background) bring different input which else would not have been the case. Also, a more diverse board is better able to solve problems. Although solving the problem could take longer, which is a disadvantage, the higher diversity in the board gives more views and perspectives. The decision-making process then looks at several alternatives and consequences, resulting in more effective solutions (Carter et al., 2003).

On the other hand there are several arguments against a more diversified board. One of them is that due to diversity it takes longer to take decisions because of conflicts within the board. These conflicts can be caused by more critical members or diverging opinions, and are less likely to arise when there is a less diversified board. It could be that over time a more gender diverse board creates more efficient decisions, but this diversification could be at the expense of firm performance if a quick response is wanted (Hambrick et al., 1996). Another argument against diversity, found from social psychology, is that the share of a minority in the board can be too small to have any influence on decisions. For example this can be a lonely outside-member. This can be caused due to specific internal group dynamics. The role of the minority is most of the times useless in this case and decisions made in board are less efficient (Carter et al., 2010).

2.4 A theoretical review of the advantages and disadvantages of gender

diversity

There can be several types of arguments to decide whether to include women in the board. Think about economical but also social and ethical arguments. Perhaps the enforcement of

(8)

7

placing women in boards is wanted by the society but that might affect firm performance negatively. Which argument is than preferred; the social or economic? An ethical reasons can be for example that women should not meet a “glass ceiling’’ and receive the same opportunities as men have. Economic reasons discuss the impact of more women in the board on firm performance.

Smith et al. (2006) stated that women in the board can be role models for other women who are active in the firm outside the board. Then performance could go up because of the fact that women become more productive. Another advantage of female board members is that the probability of being absent at meetings is lower than for males, which enhances efficiency during meetings (Adams and Ferreira, 2004). A final argument in favor of gender diversity is that it could give a better image to consumers when there is a more diversified board. This could attract more customers and thus positively influence firm performance (Smith et al., 2006).

Disadvantage of diversification could be that it leads to decrease in firm performance if a quick response is wanted. A more diversified board includes more opinions and views which delays the decision-making process (Hambrick et al., 1996). Women are overall more risk-averse than men. This could give negative results towards firm performance when increasing the female share of the board because decisions are less risky and often not optimal (Jianakoplos and Bernasek, 1998). Another disadvantage is provided by Ahern and Dittmar (2012) and already mentioned before: forced diversification may be harmful if the women are not capable to fulfill the tasks of a board member; for example because of none or not enough experience.

2.5 Gender diversity in boards: empirical results.

We discussed the board diversity and some arguments but now we are looking at the more specific topic which is included in the research question of this paper: gender diversity. Gender diversity focuses on the gender composition of the board. Several researches all over the world have been done to investigate whether gender diversity has an impact on the firm performance. But overall it cannot be concluded that there is a single specific relation because of opposite findings. This can have several reasons but some terms are often mentioned: omitted variable bias and endogeneity.

When investigating a causal relationship between two or more variables, you must keep in mind that there might be other variables that are important determinants of the dependent variable, which are not included in the model. If not included, these variables influence the error term of the model and the model is wrongly estimated: the coefficients are over/underestimated

(9)

8

and drawing valid conclusions about causal relationships is impossible. When the missing variable is important for the dependent variable and correlated with an included independent variable, there is omitted variable bias (Stock and Watson, 2012, p. 222). Endogeneity simply means that a variable in the regression model is correlated with the error term, which actually identifies there is omitted variable bias (Stock and Watson, 2012, p.462). These types of problems are mentioned in most articles below but not always dealt with, as is shown now. The articles below are categorized in three groups; negative, positive and no relationship between the female share of the board and firm performance.

Shrader et al. (1997) were unable to find any positive relationship but did find a negative relationship between the number of women in board and firm performance by using OLS and the financial measures ROA and ROE. This was analyzed with a sample of around 200 Fortune 500 firms.

An often mentioned case is that of Addams and Ferreira (2009) which investigated, for the period 1996-2003 with US firms, the relationship of women in the board and firm performance. The performance measures were the natural logarithm of the Tobin’s Q and the ROA and the control variables included firm size, board size and director characteristics. Also a dummy variable for female board members is included, meaning whether they were present or not in a board. A negative relationship is found but these findings are statistically insignificant. However, if the level of shareholder rights is added in the model, there is a positive one for firms with weak shareholder protection. Adams and Ferreira (2009) look especially at differences in behavior across gender. Women for example have a lower probability with respect to attendance problems. Also, increasing the number of women in the board gives better monitoring and higher attendances at meetings. Women apparently influence the behavior of men because a higher representation of women in the board gives better attendance results of male board members. Endogeneity is a problem, and is dealt with by including firm fixed effects to control for unidentifiable variables such as firm culture which might influence the relationship between female representation in the board and firm performance. With firm fixed effects, there is an assumption made that these kind of variables remain the same over time and thus have no significant influence on the panel data.

Also positive relationships have been found between gender diversity and firm performance (Carter et al., 2003). He defined diversity in the board as percentage of women, African Americans, Asians and Hispanics and used Fortune 1000 companies to investigate his question. He also looked at the specific form of endogeneity between firm performance and diversity in the board because they could affect each other. Firm performance could affect board

(10)

9

diversity and vice-versa (reverse causality). Carter et al. (2003) used Tobin’s Q to calculate firm value. Results showed a significant positive relationship between the number of women/minorities in the board and the firm value. Smith et al. (2006) investigated the impact of gender diversity in Denmark and tried to find out the relationship between number of women in the board and firm performance by using the 2500 largest Danish firms in a panel study. The data of firms were from the period 1993-2001. The hypothesis was based on literature and stated an expected positive relationship. In this research there can be shown that there is a positive relationship, even after controlling for some characteristics of firms. It shows also that qualifications of females are an important determinant of the positive effect of women in top management. In Spain there has been found a positive effect of gender diversity on the value of the firm by checking 68 Spanish firms in the period 1995-2000 using panel data (Campbell and Minguez-Vera, 2008). The firm performance measure is Tobin’s Q and included in the regression is a dummy variable which stated 1 if there were one or more women in the board and otherwise 0. Furthermore the control variables debt level, board size, firm size and ROA are used. To control for endogeneity they run a causality test, to see whether gender diversity is positively related with firm performance and not vice-versa (which was not the case). There has been showed board diversity has a positive effect on the firm value. But having women in the board has no significant effect on firm value. So a note should be made: companies should concentrate on board diversity, but not by purely focusing on adding more women (Campbell and Minguez-Vera, 2008). Other research in the US however found no relationship at all such as in Carter et al. (2010). The relation between firm performance and between the number of women and the number of ethnic minorities in the board is investigated in this study. Financial measures included Tobin’s Q and ROA. This research suggested endogeneity of firm performance and board diversity (in terms of gender and ethnicity). These results are explained such that effects of diversity differ under times and differ under circumstances. On average the result would state there is no effect at all.

Finally, the research done by Hoogendoorn et al. (2013) showed some additional and remarkable results. They did a field experiment in The Netherlands where they created 45 teams of around 12 first-year students, in which the female representation in the teams were not always the same. The group had to do tasks for their firms which lived for only one school year. All kind of tasks similar to that of real-life business were performed like for example prepare meetings for shareholders. The profit and sales of the teams were the performance measures. Concluded was that there is a non-linear relationship between the number of women in the team and performance. Those teams with one dominant sex appeared to perform worse compared

(11)

10

with teams in which the division of male/female was more balanced. In the end, a team with a female share of 0.55 showed the best results (in terms of profits and sales) in this field experiment. Hence, there is no possibility to draw any conclusions on past research about the relationship between gender diversity and firm performance across the world.

2.6 Gender quota: prevalence and theory

This part focuses on the global perspective of gender diversity. Although the trend is that women become more active on the labor market, there are still significant inequalities comparing with men. Different countries and continents are mentioned to see how all of them handle this issue. Do they intervene in the labor market with quotas? Figures and numbers are given to see how mankind handles this issue in several ways. Lab and field experiments concerning quota are discussed too and in the end the Ahern and Dittmar (2012) research is reviewed which investigated the effects of the implemented Norwegian quota.

Norway was the first country that introduced gender quotas for boards in 2005. In December 2003 the Norwegian government made a law for public listed companies to include a female representation in the board of at least 40%. The implementation remained voluntarily and gave not the desired effect. Therefore the law became compulsory in 2005 (Ahern and Dittmar, 2012). Companies who did not comply were dissolved. The deadline was the 1st

January 2008 to imply this rule. In the end no Norwegian firm was dissolved because they all complied. Spain also enforced a kind of a quota in 2007, which states that 40% of the board of listed companies is recommended to be female. No punishments are stated yet if the goal is not met. In a sample of the 35 largest listed companies in Spain, only 11.9% of the board members were female in 2011 (Deloitte, 2013). The Netherlands implemented a law too, which has been explained in the introduction. Of 85 Dutch stock-exchange listed firms, 13.7% of the directors is female in 2013 (Dutch Female Board Index, 2013). An increase of 3.2% compared to 2012. Strict quota laws are not an important topic all over the world. China does not make any issues about gender diversity. In their corporate governance code, gender is not seen as an important feature for board members. Hence, no gender quotas are implemented at all. A female representation of 8.5% in 2012 is given for Chinese listed companies. Looking at their neighbors, Hong Kong, the story is quite different. Although the percentage is just a bit higher, 9.4%, other measures are taken to increase this number. An implementation of amendments on the 1st of September 2013 requires that companies should mention whether they adopt a diversity policy or not, and explain why if not. An annual update concerning progress about

(12)

11

gender diversity is also required. The United States does not have any gender quotas implemented although some organizations tried to promote it by some projects to stimulate female representation. In 2012, only 16.1% of the board members were female (Deloitte, 2013).

Balafoutas and Sutter (2012) looked with a lab experiment at effects of affirmative action programs on individual incentives. These programs enhances an equal distribution of males and females in upper-level positions. They concluded that to encourage women, minimum quotas do not give any loss in efficiency. The interventions stimulated women to enter competitive environments and the researchers found that performance is not decreasing compared with no intervention. Affirmative action can also have negative consequences because it could give reverse discrimination. It also might result in a lower level of quality of employees who are selected because of the new policy (Niederle et al., 2013). In their lab experiment, Niederle et al. (2013) stated that another reason why female representation is low could be because of the competitive aspect, which women like less, and therefore do not apply for these jobs. They investigated whether implementing a quota stimulates women to participate more to apply for jobs and what the consequences are for the expected costs of implementing the new quota. Women appeared to participate more, even more than expected. The number of high-performing applicants did not change a lot, because a lot of women now participated while some men stayed out. Reversed discrimination seemed (almost) not to appear and the consequences were that the costs of this new policy are eliminated thanks to the huge increase of the supply (participation) of women. A field experiment of De Paolo et al. (2010) in Italy in 1993-1995 showed some consequences of a gender quota. A law was implemented that states a minimum number of female candidates for each party for the local elections. This law was removed in 1995, but there were some municipalities who voted during the period 1993-1993 and some not because these elections were not all at the same time. Hence, there was the possibility to investigate differences between these two “groups”. There has been found that the law stimulated females more to get on the party list for the municipalities which were affected. And after the law was removed in 1995, the group of municipalities remained to have more females than those municipalities which were not affected by the quota law during the period 1993-1995. Because of this result there can be stated that gender quotas help to diminish negative stereotypes. The same conclusion is drawn by Beaman et al. (2008).

A real-life example is that of Ahern and Dittmar (2012), who investigated what the consequences of the diversity quota were for Norwegian firms. As mentioned before, the law of 2005 stated that at least 40% of the board should be female by 2008. On average, this meant that more than 30% of the board had to be replaced because the female part first was

(13)

12

approximately 9%. First, the announcement of a compulsory quota gave a significant decline in the stock prices of the firms. The replacements in the board appeared to have serious negative consequences over time, as Tobin’s Q decreased the following years. This was because of the fact the board became younger and less experienced which seemed not be optimal for the firm performance due to their changes in decision-making.

2.7 Hypothesis

The previous overview shows that empirical results from the past do not give a definitive and clear answer on the direction of the relationship between gender diversity in the board and firm performance. As is shown in the past, both a positive and a negative relationship between gender diversity and firm performance belongs to the possible outcomes. However, the following hypothesis is given:

H1: The relationship between the number of females in the board and firm performance is positive for Dutch firms.

This hypothesis is based on the earlier mentioned advantages of diversity which are described in section 2.3 and 2.4, such as that of Smith et al. (2006), Adams and Ferreira (2004), Carter et al. (2003) and Flood et al. (1999). Overall, these reasons provide a solid base for this hypothesis.

(14)

13

3. Research method

In this section the research method is discussed. The data, methodology and all variables are described which are used for answering the research question.

3.1 Data

This empirical research is based on the 85 Dutch firms which are included in the Dutch Female Board Index (DFBI) 2013. We check whether these 85 are present in the previous editions of the DFBI too; for the period 2008-2012. We find that not all 85 are included in the sample because some provide data for only 1 year and for this reason 5 firms2 are excluded. The sample now consists out of 80 firms, but another issue decreased the sample too. Due to a split up of TNT N.V. into two independent firms in 2010 (Post NL N.V. and TNT Express N.V.), important financial data is missing and therefore they cannot be included. The sample size is thus a balanced data-panel of 78 firms with a total of 468 firm-year observations. In 50 out of 78 cases, the share of women changes over time, so there is sufficient variation in this share to identify the effect of diversity on firm performance. The statistical analysis is carried out using the software Stata.

The financial data (net income, total assets and revenues) is found in COMPUSTAT and the total market value in the Thomson Reuter’s DataStream database. The age of the firm is found in the annual reports. Finally, the Dutch Female Board Index is used to find the needed board data of Dutch firms (number of women in board and board size). The data of the Dutch Female Board Index is provided by the Erasmus Instituut Toezicht & Compliance and Nyenrode Business University (Lückerath-Rovers M., 2013).

In the Netherlands the corporate governance structure is provided by the so called ‘’Structure Act’’ from 1971. This implies that the board consists of a supervisory board and a management board, known as the two-tier board structure (Maassen and Van Den Bosch, 1999). Both boards have their own duties: the supervisory board has decision-control while the management board has decision-management as is stated by Fama and Jensen (1983). However, in this thesis we do not have the possibility to look for differences in both types because the DFBI looks only at the total number of women in the board and thus makes no distinction between the supervisory and management board. Hence, in this thesis the board means the sum of the supervisory and management board.

(15)

14

3.2 Methodology

The regression is made with the use of the program Stata. The log of Tobin’s Q and ROA are used as performance measures and are the dependent variables in the estimated models. The formulas that are used to calculate ROA and Tobin’s Q:

ROA: Net Income

Total Assets

and Tobin’s Q:

Total market value Total Assets

The total market value for Tobin’s Q is normally calculated by multiplying the total number of shares outstanding with the share price. However, as stated in paragraph 3.1, the total market value is found directly in the Thomson Reuter’s DataStream database. The models below aim to explain firm performance with the share of female board members, controlling for the age of the firm, board size and firm size, as well as company and time fixed effects.

Ln (Tobin’s Q) it =

Β0+ β1*(female share of board)it + β2*(female share of board)2 + β3*(Age firm)it +β4*(Age

firm)2+ β5*(Board size)it+ β6* Ln(Firm size)it+ β7*(Dummy female presence)it +(ϴi)+(Zt) +ɛit.

ROA it =

Β0+ β1* (female share of board)+ β2*(female share of board)2 + β3*(Age firm) + β4*(Age

firm)2_{+ β}

5*(Board size)+ β6* Ln(Firm size)+ β7*(Dummy female presence)it +(ϴi) +(Zt)+

ɛit

As stated in the hypothesis, the coefficient β1 is expected to be positive and significant because

this coefficient refers to the influence of gender diversity on the firm performance. The first performance measure is the natural logarithm of Tobin’s Q. Tobin’s Q is a measure of the firm assets in relation towards the firm’s market value. For values above 1, more investments in the firm are advised because a positive net result is possible. For values below 1, assets should be sold (BusinessDictionary). For the Tobin’s Q, a log-linear model is thus used. A log-linear model implies that a change in an independent variable of one unit is connected with a percentage change of the independent variable. To see this, if there is a one unit change in the board size for example, the coefficient of board size is multiplied by 100 percent to find the effect on Ln (Tobin’s Q). If the coefficient of B6 is discussed, which is also expressed in a

natural logarithm, a 1% change in revenues is associated with a B6% change in Tobin’s Q (Stock

and Watson, 2012, p. 310-312).

(16)

15

profitability of the total assets in terms of revenues. It is calculated by dividing the net income with the total assets (Investopedia). In this model, B6 is expressed again in a natural logarithm.

Now a linear-log model appears. Hence, if there is looked towards B6, a change of one unit in

sales with 1% changes the ROA with B6 * 0.01. For all the other variables in with no logarithms

holds the following reasoning: a one unit change in the variable gives a change of ROA with an amount of the corresponding coefficient. This study makes use of panel data. A fixed effects regression is included too to cope with endogeneity which is crucial to mention. It is included because there are omitted variables which have influence on the outcome but are for some reason unobservable. Therefore fixed effect regressions are included, which has two aspects. Omitted variables might vary for each firm, but do not change in time. An example of this is firm culture. These are called the firm fixed effects. Another aspect of fixed effects regressions is time fixed effects. These effects represents omitted variable bias that does not differ across firms but does change over time, like an economic crisis for example. These omitted variables are correlated with the performance measures and the error term ɛ. Fixed effects regression handles with observable and unobservable variables that could affect firm performance. To control for these unobservable variables, fixed effects deletes these unobservable variables and gives the possibility to measure the net effects of dependent and independent variables. The last part of the results includes only data of 2012 and 2013. Because of missing data for the ROA in 2013, only the Ln (Tobin’s Q) is included. The comparison in results between these two years then provides some extra insight of the so called “quota-effect”, which might be responsible for any differences in performance. This will be discussed in more detail in part 4.2.3.

Finally, heteroskedasticity plays a potential role in this study. Heteroskedasticity means that the variance of the error term

ɛ

it is not constant when it depends on the regressors in the

model. To work with this and avoid wrong estimates in the regressions, the use of heteroskedasticity - robust standard errors is an option in the ordinary least squares regression (Stock and Watson, 2012, p. 808). However, an additional threat for the fixed effect regression is autocorrelation: a variable is correlated with itself over time for a random firm. To circumvent this, there are heteroskedasticity-and autocorrelation-consistent standard errors are used (HAC). The option clustered standard errors is therefore used in Stata (Stock and Watson, 2012, p. 405-406) and the variable clustered is company. A check is performed to find out whether using the clustered standard errors instead of heteroskedasticity - robust standard errors gives any significant differences in results, which are not found. Hence, all regressions include the standard clustered errors.

(17)

16

3.3 Variables

The independent variable is gender diversity. Gender diversity is defined as the share of female board members, thus (nr. of females in the board / board size). This is used to analyze whether a higher share of female presence in the board has a positive effect on the firm performance measures ROA and Ln (Tobin’s Q). Also the squared female share of the board is used. This is based on Hoogendoorn et al. (2013), who finds that a gender mix could give better results. Additionally, an equal division of males and females do not produce less good results than in the case of a female majority. Hence, an inverse u-shaped relation is suggested which means there is some optimal point in the gender mix. There is a dummy variable included which looks whether there is female presence in the board. This dummy equals 1 if there is at least one female board member and otherwise equals 0. It is included for a robustness check; to see whether these estimations differ compared with using share of female variables.

The composition of the board is crucial to solve agency problems and create the optimal firm value. Yermack (1996) finds a convex shaped inverse relationship when he compares board size with firm performance. This actually says that smaller boards perform better. If a firm starts with a relative small board and let it grow in time, performance decreases. Hence, board size can be a crucial aspect and therefore it is included as a control variable.

The size and age of the firm are other variables that need to be used in the estimated model. Larger firms are expected to perform better because they possess more employees and thus revenues are expected to be larger. The size of a firm is in the estimated model measured with revenues. The revenues are included with a logarithm as in Adams and Ferreira (2009). The size of a firm could also be measured with the number of employees, but due to missing data it is not an option in this sample. Old firms can become more relaxed in time and earnings therefore could decline. On the other hand, young firms encounter start-up problems, high costs and low earnings. Therefore the relation between performance and firm age is argued to be inverse u-shaped (Smith et al., 2006). The age of a firm is found in annual reports through looking towards the foundation year. However, because of mergers/split-ups or other reasons, the foundation year can be hard to estimate. In this thesis the foundation year is therefore set at the year the oldest part of the current firm is founded.

The variable board independence will not be used in the estimated model. The reasoning behind this is that there is the expectation that all companies in the sample have a two tier structure. This means there is not more than one dependent director in the board. Hence, including board independence seems to be insignificant in this case.

(18)

17

There are also variables which are unobservable but do influence estimation results. Adams and Ferreira (2009) noted corporate culture as one of those variables. The use of panel data brings a solution to circumvent any problems with this kind of omitted variable bias. This is done with including a fixed effects regression (Stock and Watson, 2012, p. 396). The factor Zt is therefore included which refers to the year fixed effect. It is built through creating the

dummy variables for time (Dummy 2009, Dummy 2010, Dummy 2011, Dummy 2012 and Dummy 2013). The Dummy 2008 is not included to prevent the dummy variable trap. The dummy variable trap means that including a full set of binary variables with a constant term in a regression gives perfect multicollinearity (=one regressor is a linear function of another regressor) (Stock and Watson, 2012, p. 243). The company fixed effect, ϴi, is not included in the OLS regressions, because the necessary dummy variables for all firms would give a significant loss in degrees of freedom. In that case only 5 or 6 observations would be possible for each firm. The company fixed effect is therefore incorporated in the fixed effects regression. Additionally, each firm is labelled with a number for the fixed effect regression in Stata. Both the company and year fixed effect are noted in brackets on page 14 in the formulas.

(19)

18

4. Results

This section starts with the descriptive statistics in 4.1. The regression analysis are presented in part 4.2.1 and 4.2.2. The quota effect is discussed in part 4.2.3. Finally, robustness checks are described in part 4.3.

4.1 Descriptive statistics and correlation matrix

The total number of Dutch companies included in the analysis is 78, which are listed in appendix A. They give 468 firm-year observations over the period 2008-2013. Over this period, 3550 board positions existed, of which 311 were occupied by females. As is shown in the table 1, for the variables Net Income and ROA the number of observations are decreased towards 390. This is because the net income data over the year 2013 could not be included in this sample due to missing data.

Table 1. Descriptive statistics.

Variable Obs. Mean Std. Dev. Min Max

Board size 468 7.585 2.997 1 23

Nr. of females in board 468 0.665 0.898 0 5

Dummy female presence 468 0.442 0.497 0 1

Female share of board 468 0.071 0.089 0 0.3636

(Female share of board)2 ₄₆₈ _0.013 _0.021 ₀ _0.132

Age firm 468 95.910 84.208 10 397

(Age firm)2 ₄₆₈ _16274.680 _29248.340 ₁₀₀ ₁₅₇₆₀₉

Net Income (Millions) 390 160.887 657.824 -2169 5027

Total Assets (Millions) 468 23415.250 140605.500 0.708 1331663

Total Market Value (Millions) 468 2855.359 6481.782 2.67 50320.710

Firm size (Revenues) (Millions) 468 4902.423 14627.300 0.034 154598

Log (Revenues) 468 6.197 2.579 -3.381 11.949

ROA 390 0.008 0.267 -1.686 0.519

Tobin's Q 468 0.788 1.004 0.013 12.572

Regarding the financial values, there seem to be some outliers. These variables show a considerable high deviation from the mean if we look at the maximum values. This refers to the variables Net Income, Total Assets and Total Market Value. The extreme values appeared not

(20)

19

to influence the results significant as a result of a regression without these outliers (the R2_did

not change significantly). The companies with those extreme values are therefore not excluded. The percentage of firms that have at least one female (Dummy female presence = 1) over the whole observation period 2008-2013 is 44.23%. The maximum number of females in the board is 5 and on average there are 0.66. The maximum female share of the board is 0.3636 (36.36%), while on average this is 0.071 (7.1%). As we can see in figure 1, there is over time a constant increase and especially from 2012 towards 2013, probably because all firms knew that the official implementation date of the quota law was the 1st January 2013. Figure 2 gives the same numbers as figure 1, but is now shown in a histogram which outlines the significant increase of female presence in boards when approaching 2013.

Figure 1 &2. The mean female share of the board over the years.

From table 2 we can see that 311 females are present in the board in this sample. This can be calculated by multiplying the first row with the second row and sum them. For 261 firm-year observations, no females are included in the board.

Table 2. Frequency nr. of females in board.

Nr. of females in board Frequency Percent (%)

0 261 55.77 1 128 27.35 2 60 12.82 3 14 2.99 4 4 0.85 5 1 0.21 Total 468 100.00

(21)

20

The age of the firm in this sample is on average 96 years with a maximum of 397 years. The average board size is 7.58 with a minimum of 1 and a maximum of 23 positions, which was for ING Groep N.V. in 2008. Figure 3 shows that the average board size remains on a fairly constant level over time.

Figure 3. Average board size over time

Table 3 compares the means of variables for firms with and without a female member in the boards. There is almost always a significant difference between the two groups. Firms with at least one female in the board are bigger in terms of revenues and have larger boards. Also, these firms perform better on average, when we look at ROA. Table 3 suggests that including at least one female is more beneficial. However, there cannot be concluded that there exists a causal relationship between female presence and firm performance when we look to these averages. To control for these differences, we do include the apparently important control variables in the regressions when explaining the relationship between female board members and firm performance.

Table 3. Overview averages with/without female presence

Dummy female presence Average if 0 Average if 1 Difference

Average BS 6.088 9.473 3.385***

Average Nr. fem. 0 1.502 1.502***

Average FS board 0 0.161 0.161***

Average (FS board)2 ₀ _0.029 _0.029***

Average Age firm 79.414 116.710 37.296***

Average (Age firm)2 _13261.651 _20073.715 _6812.064**

Average NI 26.985 352.146 325.161*** Average TA 1227.102 51391.607 50164.505*** Average TMV 692.202 5582.817 2855.359*** Average FS (Rev.) 845.398 10017.800 9162.402*** Average Ln (Rev.) 5.131 7.541 2.410*** Average ROA -0.010 0.032 0.042** Average Tobin's Q 0.759 0.824 0.065

(22)

21

A normal distribution of a continuous random variable is a key assumption in statistics. There is checked for all variables whether this assumption holds. This does not hold for the firm size, expressed in revenues, and therefore the histograms below are provided to give more information.

Figure 4A. Distribution Revenues Figure 4B. Distribution Ln (Revenues)

The distribution of the revenues is presented in figure 4A. Figure 4A is definitely not normally distributed (bell-shaped) but skewed in a positive way. To get rid of this skewness, the natural logarithm can be taken and in figure 4B we can indeed see a more bell-shaped form when the log is applied. Considering the two performance measures, Ln (Tobin’s Q) shows normality but ROA does not. Creating Ln (ROA) is not an option in this study. The natural logarithm does not accept any negative values. In this sample, 46 companies (59%) have at least one negative value of ROA in the period 2008-2013, which means a potential significant loss in information. Hence, the natural logarithm is not used for the ROA.

Correlation is an important aspect when interpreting the results from a regression. This is why on the next page a correlation matrix is included. In this matrix are abbreviations of all variables. The description and definitions can be found in appendix B. The command “pwcorr” is used in Stata because this excludes missing values and only calculates the correlation of the data inserted. Given the correlation matrix, we need to look at the most relevant and important values. The most important are that of the included variables in the estimation model.

(23)

22

Table 4. Correlation matrix.

The individual correlation is statistically significant at: ***_1%,**_{5% or}*_{10% significance level (1-tailed). In brackets below the correlations are the}

p-values given.

BS Nr.Fem Dummy FS Board (FS

Board)2

Age firm (Age firm)2 _NI _TA _TMV _FS(Rev.) _Ln(Rev.) _ROA _Tobins'Q _Ln(TQ)

BS 1.0000 Nr. Fem. 0.6589*** (0.0000) 1.0000 Dummy 0.5616*** (0.0000) 0.8321*** (0.0000) 1.0000 FS Board 0.4692*** (0.0000) 0.9264*** (0.0000) 0.8979*** (0.0000) 1.0000 (FS Board)2 0.3571*** (0.0000) 0.8678*** (0.0000) 0.6844*** (0.0000) 0.9275*** (0.0000) 1.0000 Age firm 0.2245*** (0.0000) 0.2273*** (0.0000) 0.2202*** (0.0000) 0.2281*** (0.0000) 0.1844*** (0.0000) 1.0000 (Age firm)2 0.1032** (0.0256) 0.1073** (0.0203) 0.1158** (0.0122) 0.1307*** (0.0046) 0.1075** (0.0200) 0.9447*** (0.0000) 1.0000 NI 0.3100*** (0.0000) 0.4204*** (0.0000) 0.2441*** (0.0000) 0.3176*** (0.0000) 0.3374*** (0.0000) 0.0534 (0.2924) 0.0025 (0.9611) 1.0000 TA 0.3923*** (0.0000) 0.2196*** (0.0000) 0.1774*** (0.0001) 0.1101** (0.0172) 0.0644 (0.1643) 0.1574*** (0.0006) 0.0956** (0.0388) 0.1923*** (0.0001) 1.0000 TMV 0.6585*** (0.0000) 0.5908*** (0.0000) 0.3751*** (0.0000) 0.4034*** (0.0000) 0.3975*** (0.0000) 0.1717*** (0.0002) 0.0780* (0.0918) 0.6188*** (0.0000) 0.5082*** (0.0000) 1.0000 FS (Rev.) 0.5914*** (0.0000) 0.4442*** (0.0000) 0.3077*** (0.0000) 0.2781*** (0.0000) 0.2484*** (0.0000) 0.1723*** (0.0002) 0.0791 (0.0918)* 0.4231*** (0.0000) 0.8726*** (0.0000) 0.7630*** (0.0000) 1.0000 Ln(Rev.) 0.7221*** (0.0000) 0.5393*** (0.0000) 0.4702*** (0.0000) 0.4101*** (0.0000) 0.3235*** (0.0000) 0.2398*** (0.0000) 0.0767 (0.1045) 0.3483*** (0.0000) 0.3047*** (0.0000) 0.5414*** (0.0000) 0.5105*** (0.0000) 1.0000 ROA 0.1000* (0.0485) 0.1321*** (0.0090) 0.1093** (0.0309) 0.1173** (0.0205) 0.1053** (0.0377) 0.0912* (0.0720) 0.0575 (0.2569) 0.1482*** (0.0034) -0.0008 (0.9882) 0.1039** (0.0402) 0.0353 (0.4928) 0.2904*** (0.0000) 1.0000 Tobin's Q -0.0158 (0.7330) 0.0863* (0.0620) 0.0321 (0.4887) 0.0834* (0.0716) 0.0998** (0.0309) -0.0880* (0.0572) -0.0762* (0.0996) 0.0210 (0.6789) -0.1106** (0.0167) 0.0381 (0.4106) -0.1153** (0.0125) -0.1260** (0.0372) 0.2405*** (0.0000) 1.0000 Ln(TQ) -0.1995*** (0.0000) -0.0390 (0.4003) -0.0818* (0.0771) 0.0075 (0.8712) 0.0656 (0.1567) -0.2453*** (0.0000) -0.1676*** (0.0000) -0.0056 (0.9118) -0.4475*** (0.0000) -0.0636 (0.1693) -0.3717*** (0.0000) -0.1903*** (0.0000) 0.1210** (0.0176) 0.6525*** (0.0000) 1.0000

(24)

23

The most important correlations are discussed now in more detail. Firm size is significantly positive correlated with board size which confirms that bigger firms indeed have larger boards. This can be because bigger firms have more jobs, cases and projects which needs to be managed and controlled. The age of the firm is significantly positive correlated with both board size and firm size. An older firm appears to be larger in terms of revenues and board compared to younger firms. Firm age is also positive correlated with the female share of the board, which enhances that older firms are less reluctant to incorporate women in their board. Board size shows a positive correlation with the female share of the board too. It is probably the case that a larger board has a higher probability of presence of women in the board. Board size is significantly negative correlated with Ln (Tobin’s Q), while positive with ROA. It is not clear why this is the case, although it might be in some way indeed supported by Yermack (1996). Regarding the two performance measures, the correlation of 0.1210 is significant and positive. From this we can state that firms with higher profits also more often have a larger firm value. The same holds for ROA and Tobin’s Q, which show a correlation of 0.2405. Finally, the performance measures are both positively correlated with the female share of the board, although only ROA shows a significant value.

Some values in the matrix have less significance for this study and it is quite obvious why they are (too) high. For example, the Nr. Fem. is highly and almost perfect correlated with FS board (92.64%). This will however not influence the results, because only FS board is included in the model and formulated as (Nr. Fem. / Board size).

4.2 Regression analysis

The research method is similar to that of Adams and Ferreira (2009). This part starts with an overview of the results calculated by the method Ordinary Least Squares (OLS). Also, the implications of this output is discussed. The second part includes the fixed effects regression with all results and implications. The period 2008-2012 is used to get comparable samples and hence we are able to make a better comparison between the two performance measures and their corresponding results. Therefore, the number of firm-year observations in the regressions is 390. Recall that the ROA of 2013 can be not be calculated due to missing data. Finally, in part 3 there is tried to estimate the influence of the quota on firm performance.

(25)

24

4.2.1 Ordinary least squares regression

The hypothesis is tested and performed with two types of regressions, as explained in the methodology part 3.2. Both the natural logarithm of Tobin’s Q and the ROA are used as performance measures. In this part is the ordinary least squares output discussed and analysed. When the coefficients of gender diversity are positive and significant, the hypothesis is accepted. If a negative coefficient or a coefficient value of zero appears, the hypothesis is rejected.

Table 5 starts to show the ordinary least squares output for multiple regressions with the use of the female share variables and thus exclude the dummy female presence. Column 1a and 2a only regress the performance measures on the gender diversity variable. The columns 1b and 2b include the variable female share of board squared, based on the study of Hoogendoorn et al. (2013). If looked to the output in columns 1b and 2b, opposite results are found. The coefficient of the squared variable is positive in 1b which imposes a u-shaped relationship between gender diversity in boards and firm performance. Looking to the performance measure ROA in column 2b, a negative value of the squared variable is found and this is in line with Hoogendoorn et al. (2013), who found an inverse u-shaped relationship between firm performance and gender diversity too.

The concern is that the proposed relationship is misleading and subject to omitted variable bias. Therefore, columns 1c and 2c include the control variables and represents the total model. The variables for female share of the board change significantly for both performance measures compared to columns 1a and 2a. There can be said that there is indeed omitted variable bias in the models of column 1a and 2a. The R2 rises significantly after adding the control variables. The R2_{is defined as the fraction of the sample variance of the independent}

variable that is explained by the variables. An aspect of the R2_{is that it always rises when a}

variable is added. The adjusted R2_{is therefore a better measure to compare regressions. Those}

values are not shown in the tables, because they are almost identical values as R2_{. The}

significant rise in R2 gives indicates that including these variables substantially improves the fit of the regression. The variable age firm squared variable is not included, which will be further discussed in part 4.3: robustness checks. The coefficient of age firm appears to be significant at the 5% level in column 1c. This can be interpreted as follow: when a firm gets one year older, the Ln (Tobin’s Q) decreases with 0.26%. The natural logarithm is significant at the 10% level for ROA. Regarding the board size, only negative (non-significant) coefficients are included, similar to the findings of Yermack (1996). The year-dummies are included too. Column 1c includes only significant year-dummies, while column 2c has none. These mean that in those years with a negative sign, performance was lower than in the baseline year, the dummy of

(26)

25

2008, which is omitted. The value of the firm expressed in Ln (Tobin’s Q) is negative for all years. The value of the firm might be negatively influenced by the financial crisis which explains these negative coefficients. On the other hand, ROA shows insignificant small coefficients for the year-dummies. Both regressions of column 1c and 2c are significant, respectively at the 1% and 5% level, when looking towards the F-values. However, in 5 out of 6 regressions of table 5, no significant values are found for the variable female share of board and when all control variables are included, none remains significant. The hypothesis is therefore not accepted and apparently gender diversity in boards is not significantly related with firm performance.

Table 6 has regressions in which the gender diversity variables are replaced with the dummy female presence. This dummy equals 1 if at least one women is present in the board during the period 2008-2012, otherwise the dummy equals 0. In column 3b and 4b, the dummy is excluded. For Ln (Tobin’s Q), column 3b gives a regression model which is statistically significant at the 1% level with 16.74 as F-value. The model which includes the dummy in 3c is significant at the 1% level with an F-value of 14.02. All year dummies for both performance measures give similar findings as in table 5. Overall, the coefficients of control variables do not differ significantly when using the dummy female presence instead of the gender diversity variables. The signs for control variables remain the same too. For table 6, the same can be concluded regarding the hypothesis: rejection. Firm performance and the presence of women in boards is not positively related because none of these dummies show statistical significant coefficients. Overall, the ordinary least squares regressions give no support concerning the hypothesis that there is a positive relationship between gender diversity and firm performance.

(27)

26 The individual coefficient is statistically significant at: ***_1%,**_{5% or}*_{10% significance level.}

Table 5 OLS (FS.)

Independent variable Ln (Tobin's Q) ROA

1a 1b 1c 2a 2b 2c Constant -0.655 (0.103) -0.606 (0.098) 0.470 (0.319) -0.019 (0.025) -0.020 (0.026) -0.096 (0.052)

Female share of board 0.145 (0.946) -4.260 (2.911) -0.080 (2.066) 0.315* (0.162) 0.422 (0.334) 0.044 (0.300)

(Female share of board)2 _20.511**

(10.448) 10.814 (6.866) -0.497 (0.880) 0.256 (0.923) Age firm -0.0026** (0.0015) 0.00007 (0.00011) Board size -0.049 (0.052) -0.015 (0.012) Ln (Revenues) -0.046 (0.053) 0.033* (0.017) Dummy 2009 -0.500*** (0.065) 0.027 (0.031) Dummy 2010 -0.359*** (0.067) 0.021 (0.031) Dummy 2011 -0.238*** (0.066) -0.008 (0.033) Dummy 2012 -0.526*** (0.076) -0.007 (0.033) N 390 390 390 390 390 390 SER 0.98757 0.97693 0.91017 0.26576 0.26608 0.18009 F-statistic 0.02 3.66** 11.56*** 3.80** 3.32** 1.91** R2 _0.0002 _0.0241 _0.1683 _0.0101 _0.0103 _0.1171

(28)

Table 6 OLS (Dummy)

Independent variable Ln (Tobin's Q) ROA

3a 3b 3c 4a 4b 4c Constant -0.579 (0.096) 0.369 (0.318) 0.462 (0.328) -0.010 (0.023) -0.101 (0.052) -0.096 (0.052)

Dummy female presence -0.164 (0.203) 0.234 (0.169) 0.042 (0.026) 0.012 (0.025) Age firm -0.0024* (0.0014) -0.0025* (0.0015) 0.00008 (0.00012) 0.00007 (0.00011) Board size -0.034 (0.056) -0.052 (0.054) -0.015 (0.011) -0.016 (0.012) Ln (Revenues) -0.037 (0.053) -0.042 (0.054) 0.033* (0.017) 0.033* (0.017) Dummy 2009 -0.488*** (0.061) -0.498*** (0.064) 0.028 (0.031) 0.028 (0.031) Dummy 2010 -0.337*** (0.059) -0.344*** (0.060) 0.021 (0.031) 0.021 (0.031) Dummy 2011 -0.218*** (0.057) -0.234*** (0.061) -0.008 (0.033) -0.008 (0.033) Dummy 2012 -0.480*** (0.064) -0.502*** (0.069) -0.005 (0.031) -0.006 (0.033) N 390 390 390 390 390 390 SER 0.98433 0.92841 0.92487 0.18854 0.17979 0.17996 F-statistic 0.65 16.74*** 14.02*** 2.57 2.08** 1.83* R2 _0.0067 _0.1300 _0.1389 _0.0119 _0.1154 _0.1161

(29)

28

4.2.2 Fixed effects regression

This study makes use of panel data. Therefore fixed effects regressions are included, which coops with endogeneity. Table 7 includes all fixed effects regressions. If the ordinary least squares models with included control variables and the fixed effects models (column 1c, 2c, 3c and 4c versus 5c, 6c, 7c and 8c) are compared, there can be noticed that all variables have different coefficients. These differences indicate there is the problem of omitted variable bias. The gender diversity variables are first analysed. The female share of the board shows a positive relationship with both performance measures in column 5c and 6c. However, none of them is significant. The coefficient of the squared gender diversity variable only provides negative values, which confirms the suggestion by Hoogendoorn et al. (2013) that there is an inverse u-shaped relationship between gender diversity in the board and firm performance. Apparently, there is an optimal gender-mix for boards. The dummies for female presence give positive values in models with control variables, what also happened in the ordinary least squares regressions. Board size has negative coefficients for all ordinary least squares regressions. The coefficients of board size switch towards positive values for Ln (Tobin’s Q) in the fixed effect regressions. The effect of board size on firm performance measured by Ln (Tobin’s Q) could be influenced by variables that are omitted. The natural logarithm of revenues only changes in coefficient values when comparing the two estimation methods. The sign remains negative for the Ln (Tobin’s Q) and positive for the ROA.

The control variables influence the relationship between gender diversity variables and firm performance, both in the ordinary least squares and fixed effects regressions. This can be concluded from the fact that all coefficients of the variables female share of board and dummy female presence significantly differ when control variables are included compared to when control variables are excluded (all a-columns versus all c-columns). For example, column 5a predicts that when all board members are women, the Ln (Tobin’s Q) decreases with 39.2%. But doing the same test in column 5c, Ln (Tobin’s Q) increases with 35.4%. There can be concluded that all diversity variables show different values when we compare the ordinary least squares and fixed effects regressions. The fixed effects regression takes care of these unobservable variables that influence the gender diversity variables, while ordinary least squares does not. Hence, omitted variable bias is present due to some unobserved variation in years and firms in this research.

(30)

Table 7 Fixed Effects

Independent variable

Ln (Tobin’s Q) ROA Ln (Tobin's Q) ROA

5a 5b 5c 6a 6b 6c 7a 7b 7c 8a 8b 8c Constant -0.621 (0.028) -0.622 (0.034) 6.420 (1.373) 0.008 (0.005) 0.006 (0.006) 0.238 (0.536) -0.632 (0.033) 6.364 (1.348) 6.444 (1.366) 0.004 (0.006) 0.248 (0.514) -0.020 (0.026) Female share of board -0.392 (0.434) -0.320 (0.921) 0.354 (0.925) -0.003 (0.079) 0.123 (0.174) 0.015 (0.227) (Female share of board)2 -0.314 (2.656) -0.9787 (2.462) -0.551 (0.556) -0.177 (0.643) Dummy female presence -0.034 (0.079) 0.041 (0.082) 0.008 (0.014) 0.422 (0.334) Age firm -0.072*** (0.0137) -0.005 (0.005) -0.072*** (0.013) -0.072*** (0.014) -0.005 (0.005) -0.005 (0.005) Board size 0.003 (0.023) -0.0014 (0.0064) 0.003 (0.022) 0.002 (0.023) -0.001 (0.006) -0.001 (0.007) Ln (Revenues) -0.034 (0.045) 0.036 (0.024) -0.033 (0.045) -0.034 (0.045) 0.036 (0.024) 0.036 (0.024) N 390 390 390 390 390 390 390 390 390 390 390 390 F-statistic 0.81 0.67 5.88*** 0.00 0.58 1.12 0.18 9.61*** 7.24*** 0.30 1.50 1.12 R2 _0.0002 _0.0005 _0.0643 _0.0138 _0.0004 _0.0006 _0.0067 _0.0642 _0.0643 _0.0119 _0.0006 _0.0006

(31)

30

Similar to the ordinary least squares regression there can be stated that gender diversity in the board is not significantly related to firm performance when looking towards the fixed effects regressions. Again the hypothesis is rejected and no significant support is found that the relationship between gender diversity and firm performance is positive. As stated on page 9, reverse causality might occur. Reverse causality takes a closer look in how firm performance and gender diversity in the board could affect each other. This study investigates whether gender diversity in the board has any influence on firm performance. But there is also the option that firm performance affects board diversity. For example that firms that take less risk attract more women. This test is not incorporated in this thesis but it should be mentioned that it belongs to one of the possibilities concerning the relationship between firm performance and gender diversity in boards.

4.2.3 Quota-effect

A recent change of regulations in The Netherlands resulted in a new diversity quota, which has been implemented on January 1st 2013. This new law states that big firms are obligated to have a board that consists of at least 30% of each gender. It seems from figure 1 that firms do try to obey the implemented law due to the sharp increase in 2012. The question is whether the quota-effect on firm performance can be measured like in Ahern and Dittmar (2012).

The individual coefficient is statistically significant at: ***_1%,**_{5% or}*_{10% significance level.}

Table 8 2012-2013 OLS

Independent variable Ln (Tobin’s Q)

9a 9b

Constant -0.324

(0.375)

-0.400 (0.369)

Female share of board -1.990

(2.467)

(Female share of board)2 _16.214*

(8.565)

Dummy female presence 0.114

(0.227) Age firm -0.003 (0.002) -0.002 (0.002) Board size -0.081 (0.073) -0.073 (0.074) Ln (Revenues) 0.044 (0.073) 0.046 (0.074) Dummy 2013 0.117** (0.057) 0.157*** (0.049) N 156 156 F-statistic 3.43*** 3.56*** R2 _0.1044 _0.0634

Women on top as it schould be?

WOMEN ON TOP

AS IT SHOULD BE?

Table of contents

1. Introduction

2. Literature review

2.1 The board and the agency problem

2.2 Board diversity

2.3 A theoretical review of the advantages and disadvantages of board

diversity

2.4 A theoretical review of the advantages and disadvantages of gender

diversity

2.5 Gender diversity in boards: empirical results.

2.6 Gender quota: prevalence and theory

2.7 Hypothesis

3. Research method

3.1 Data

3.2 Methodology

ɛit

ɛ

3.3 Variables

4. Results

4.1 Descriptive statistics and correlation matrix

4.2 Regression analysis