The impact of female non-executive directors on
corporate risk-taking in the EU
Master’s thesis final version
Rob van de Molengraaf S1688944
MSc Finance
First supervisor: Prof. Dr. Hermes Second supervisor: Dr. Mierau
Final version: 13-06-2014
2
1. Introduction
Despite much recent attention, corporate boards in the EU still consist of only 16.6% women (Report on women and men in leadership positions and Gender equality strategy mid-term review – European Commission, 2013). This percentage, however, has been showing a steeply upward sloping trend for the last couple of years; as can be seen in Figure 1, the share of women in the boardroom in the EU was only 11.0% in September 2010. This percentage is expected to accelerate even more in the coming years since the European Commission (EC) adopted the women on boards’ proposal which sets a 40% objective by 2020 for members of the under-represented sex for non-executive members of the boards of publicly listed companies, or 2018 for listed public undertakings (Women on Boards: Commission proposes 40% objective – European Commission, 2013).
Figure 1: Share of women on the boards of large listed companies, EU-27, 2010-2013 Continuing Progress
From: Report on women and men in leadership positions and Gender equality strategy mid-term review (European Commission), 2013. (www.europa.eu/rapid/press-release_MEMO-13-882_en.htm)
3 strong organizational and financial performance. For instance, in a study carried out by Catalyst1, companies with more women on their boards were found to outperform their rivals with a 42% higher return in sales, 66% higher return on invested capital and 53% higher return on equity.” (Questions and Answers: proposal on increasing gender equality in the boardrooms of listed companies – European Commission, 2012)
These outstanding results are, however, not univocally supported by leading academic research. Adams and Ferreira (2009) show that female directors have a significant impact on board inputs and firm outcomes. Their results, however, also suggest that mandating gender quotas for directors can reduce firm value for well-governed firms. Bøhren and Staubo (2013) investigated the consequences of the first mandatory minimum fraction of female directors’ law in Europe, which was introduced in Norway. The new gender balance law requires firms to be liquidated if they fail to meet the 40% objective of its directors of each gender. The authors suggest that mandatory gender quotas may produce firms with inefficient organizational forms or inefficient boards since half the firms exit to an organizational form not exposed to the law. In line with these results, Ahren and Dittmar (2012) exploit the unique opportunity to address the endogeneity problems concerning the relation between board composition and firm outcomes since the imposed Norwegian law is clearly an exogenous shock.2 They find a significant drop in stock prices at the announcement of the law and a large decline of Tobin’s Q over the following years. This is consistent with the theory that firms choose boards in order to maximize value. Consequently, imposing laws that limit the freedom of choice would deteriorate firm value.
These contradicting results and the notable objective of 40% directors of each gender by the EC are a cry for additional insights about the influence of more women in corporate boards on firm outcomes in the European Union. Adams and Ferreira (2009) also state that still relatively little research links diversity and corporate governance, despite its importance in the policy debate.
Before we can create more consensus and understanding in the debate about the influence of gender diversity on firm value or firm performance, we should have a look on the relationship between gender diversity and the underlying drivers of firm value or firm performance. Whereas others try to link gender diversity directly to firm performance, this approach will gain more insight into the reasons why firm performance would be influenced. This should help to gain more knowledge about the possible
1
Catalyst Inc., 'The Bottom Line: Corporate Performance and Women's Representation on Boards', 2007
2
4 causal relationship between gender diversity and firm value or firm performance. This approach is in line with Adams and Kirchmaier (2012) who conclude: “It may be more important to address the underlying causes of relative female underrepresentation than to target boardrooms directly.”
Corporate risk-taking has been identified as one of the fundamental drivers of firm performance by previous literature. (Boubakri, Cosset, & Saffar, 2013; Bromiley, 1991; Li, Griffin, Yue, & Zhao, 2013) Therefore, this paper focuses on the relationship between the proportion of female directors and corporate risk-taking in the EU. Based on existing literature, females are associated with less risk-taking in general. This drives us to the main research question whether more female directors in a corporate board are also related to less corporate risk-taking.
It contributes to the literature in several important dimensions. First, to the best of my knowledge, this paper is the first to relate the ratio of female directors in a corporate board to corporate risk-taking in the EU so it will provide better insights of this relationship. Second, it can contribute to the current debate about the relevance of gender quota and the accuracy of these governance proposals. Third, it can also give us more insight whether the impact of female directors on corporate outcomes is prevented by tokenism. Fourth, most corporate governance research is based on US samples (Kang, Ding, & Charoenwong, 2010). This paper sheds its light on a European sample and therefore contributes to the international generalizability of prior research.
The main results give a clear negative relation between the ratio of female directors and corporate risk-taking. This relation becomes stronger for higher levels of board independency. Thereby, the results indicate that only one female director is not enough to be influential on corporate risk-taking.
The next section starts with an overview of current literature regarding intrinsic differences between males and females. It gives more insight about prior governance research which is related to gender diversity. Consequenly, this section sheds its light on the differences in risk-taking between men and women in financial decision-making and finally, the hypothesis will be discussed. Section three gives an overview of the data and variables used for the analyses. Section four presents the empirical analyses about the relation between corporate risk-taking and female directors. Finally, section five presents some concluding remarks.
2. Literature review and hypothesis
5 women regarding to risk exist. There is a vast amount of literature available which tells us that men and women respond differently to risk. This is driven by intrinsic differences between men and women, which are mainly discussed in socio-cultural, psychological and biological literature. Some basic explanations will be mentioned from a wide area about why the difference in risk-seeking may be observed between genders, but more in-depth research is beyond the scope of this paper.3 Beyer and Bowden (1997) argue that social inputs may be influential in creating the observed gender differences. Social desirable behavior already plays an important role from a young age; boys should be stalwart and play army boy, whereas girls should be sweet and play with their dolls. These desirable behaviors would develop greater prudence and less self-confidence for females and more risk-seeking behavior for males. The biologists Wilson and Daly (1985) suggest that risk-taking in males has evolved due to the advantage of risk-taking behavior for males in reproducing itself. Based on research in the animal kingdom, males need to exercise risky behavior to be successful to be able to mate with several females which is essential to defend their position within the group. Females, however, do not have the capability to mate with multiple males at the same time. This would drive the higher risk-taking behavior of males. Consequently, Buss (1999) states that risk-taking in females has evolved less in response of different evolutionary influences because of their greater biological dedication to their children. Zuckerman (1994) gives a more psychiatric explanation and states that women tend to possess higher levels of a type of enzyme4, which can serve to reduce risk seeking. Finally, Byrnes et al. (1999) perform a leading meta-analysis about gender differences in risk-taking and they show that risk-taking for male participants is indeed significantly larger. They refer to a broad area of research for explanations of these differences in risk-taking, which are already mainly covered in this section. However, they are not able to present a summarizing explanation which explains the differences most concisely.
The intrinsic difference in risk-taking between men and women also clearly comes forward in prior research about the role of women and their impact on governance and firm performance. This research has shown significant differences between men and women. Adams and Ferreira (2009) show that women have less attendance problems than men. Thereby, the greater the fraction of women on the board is, the better the attendance behavior of men. Women are also more likely to join monitoring committees. These results suggest that gender diverse boards allocate more effort to monitoring. CEOs are also more likely to be hold accountable for stock price underperformance with gender diverse
3
For more research about why these differences exist, see for example: Gneezy et al. (2009), Croson and Gneezy (2009), Booth and Nolan (2012) and Booth et al. (2014)
4
6 boards and directors receive relatively more equity-based compensation. Their results with respect to firm performance are more difficult to interpret. When strong external governance5 is present, a more gender diverse board would be value diminishing. A more gender diverse board would be value enhancing, however, with weak shareholder rights, because it would increase board monitoring. An important notion to their research, however, is that they link their results mostly to gender diversity but they are not able to prove that these results are not mainly driven just by the level of independence which rises on average with higher gender diversity. Females are seen as more independent, since they usually do not belong to the “old boys club”. Jurkus et al. (2011) investigate the effect of gender diversity among top management of Fortune 500 firms on agency costs. The general effect is large, however the model becomes insignificant when controlling for the endogenous nature of gender diversity. They conclude that the effect is only sizable for firms with weak governance which confirms Adams and Ferreira’s (2009) findings on firm performance: their results suggest that increasing gender diversity in the board can have beneficial effects for firms where strong external governance is absent. In line with these findings, Gul et al. (2011) dive into the relationship between gender-diverse boards and firm-specific information of stock prices. Higher transparency results from better firm-specific information. They define this as “the public disclosure of more firm-specific information by managers; and greater incentives for the collection of private firm-specific information by investors.” They provide evidence that stock prices of firms with a gender-diverse board reflect more firm-specific information after controlling for corporate governance, institutional ownership, earnings quality and acquisition activity. This relationship is stronger for firms with weak governance structures which lead the authors to the most important finding that gender diversity in corporate boards could act as a substitute mechanism for corporate governance for these firms. Kang et al. (2010) relate this and other research about the appointment of female directors to investor reactions. Based on an event study, they present evidence that investors generally respond positively to the appointment of female directors in Singaporean firms. Regression analyses show that investors are most receptive when women are independent directors. This could suggest that female directors are expected to act more in name of the shareholders. A recent study of Levi and Zhang (2013) point out that firms with female directors are less likely to acquire other firms and if they do, they pay lower bid premia. Based on S&P 1500 data they find that each additional female director is associated with 7.6% fewer acquisition bids and the bid premium
5
7 paid reduces by 15.4%. These results suggest that more female directors are associated with a less risk-taking corporate finance policy. Based on these studies, among others, it can be concluded that a higher ratio of women in the board clearly has an impact on corporate governance and firm outcomes.
Next to the impact of the intrinsic difference in risk-taking between men and women on corporate governance and firm outcomes in general, the difference in risk-taking is also discussed in financial literature and men are again more related to risk-taking behavior. Jianakoplos and Bernasek (1998) find that single women exhibit relatively more risk aversion in financial decision-making than single men. They do not give clear explanations for the difference, but they mainly explain that their results are in line with the proven systematic risk-aversion difference between men and women. Barber and Odean (2001) demonstrate that men are more active in trading common stock than women, because they are more confident. They base their research on psychological research, which proves that in areas such as finance men are more overconfident than women. The relative overconfidence of men in finance could be an important driver of a possible difference in risk-taking by male directors versus female directors as well. Felton et al. (2003) suggest that females make less risky investment choices than males. This difference is primarily due to the riskier investment choices of optimistic males. Thereby, final portfolio values of males demonstrated greater variability than did portfolio values of females. Their results suggest that the gender difference in risk-taking of men and women may be due to a specific subgroup of men. More recently, Huang and Kisgen (2013) relate overconfidence to gender diversity and corporate finance. They examine corporate financial and investment decisions made by female executives compared with male executives such as acquisitions and debt issues. Their evidence suggests that men exhibit relative overconfidence in significant corporate decision-making compared with women. They state that the reasoning behind these results lie in behavioral differences between men and women which are studied in psychological and other fields. More explanation is not provided. Based on these financial papers regarding differences in risk-taking behavior, females are less associated with risk-taking behavior than males.
8 publication bias also creates incentives for researchers to design studies such that it will increase the chance of generating differences. Charness and Gneezy (2012) tackle these issues by testing the robustness of the hypothesis that men are more risk-taking than women based on an accumulated dataset of 15 different researches.6 Yet, they are still able to report a consistent gender difference which confirms the previously mentioned results.
Driven by the discussed literature which is almost entirely focused on the relation with firm performance, I will focus on one of the most important underlying drivers of firm performance: corporate risk-taking, a relation which is largely unexplored in current literature. The main research question is whether more female directors in a corporate board is also related to less corporate risk-taking, which results in the following hypothesis:
Hypothesis 1. A higher ratio of female non-executive directors in a corporate board is negatively
associated with corporate risk-taking in the EU
Previous discussed research provides evidence that women are associated with less risk-taking both in general and in financial related decision-making. Based on these findings, I expect the hypothesis to hold. The definition of a non-executive director will be discussed in next chapter.
3. Sample and variables
In this section, I begin by describing the sample of EU firms. Second, all the variables used in the analysis will be explained including their descriptive statistics. Third, a correlation matrix will be presented to check for possible multicollinearity issues.
3.1. Sample
The dataset is constructed by combining two data sources; the BoardEx Europe database for the governance variables and Thomson Reuters’ Datastream for the financial variables. The calculation of my proxy for risk is based on the volatility of the firm’s current and future earnings over a 5 year period (T over 0 to +4). This proxy and timeframe follow the current literature, as will be discussed in section ‘3.2.1. Corporate risk-taking variable’. Because I need 4 additional years of data after the firm-year observation, 2009 is the most recent possible year to include in my sample. However, data for 2009 is not included in the BoardEx database. Thereby, the BoardEx Europe database does not contain data
6
9 prior to 2006, therefore my dataset includes data from the years 2006, 2007 and 2008. BoardEx Europe provides 9,198 firm-year observations in total for these years. The corresponding financial variables including the necessary input to determine the risk variables are extracted from Thomson Reuters’ Datastream.
Many observations are deleted from the final sample because of missing data for one or more variables. Moreover, all observations from Norway7 and Switzerland are deleted as well, since this research focuses on the European Union. I also follow the convention in the literature and delete financial firms (one-digit SIC code ‘‘6’’)8. The financial data of firms from the UK, Denmark and Sweden is converted into euros by Datastream, the rest of the firms have the euro as their domestic currency. All these corrections lead to a final sample of 2,289 firm-year observations. The distribution of the final sample per country and per industry is presented in Table 1.
Table 1
Distribution of all 2,289 included firms in the sample, classified by country and by industry. Distribution of firms
Panel A: By country Panel B: By industry
Country 2006 2007 2008 Total Percentage Industry Number Percentage
Austria 5 17 20 42 1.83 Basic industry 275 12.01
Belgium 18 18 15 51 2.23 Capital goods industry 272 11.88
Denmark 9 9 6 24 1.05 Construction industry 162 7.08
Finland 2 5 4 11 0.48 Consumer durables industry 334 14.59
France 113 119 116 348 15.20 Food/tobacco industry 121 5.29
Germany 73 76 77 226 9.87 Leisure industry 147 6.42
Greece 10 16 14 40 1.75 Petroleum Industry 79 3.45
Italy 36 41 39 116 5.07 Services industry 472 20.62
Netherlands 37 39 37 113 4.94 Textiles/trade industry 117 5.11
Portugal 7 15 16 38 1.66 Transportation industry 123 5.37
Republic Of Ireland 23 28 21 72 3.15 Utilities industry 187 8.17
Spain 24 24 28 76 3.32 Total 2,289 100.00
Sweden 33 31 29 93 4.06
United Kingdom 357 352 330 1,039 45.39
Total 747 790 752 2,289 100.00
The sample is equally divided over the years as can be seen in panel A from Table 1. It does, however, show a strong bias towards UK firms (45.39%). France and Germany together account for 25% of the sample. I will include country-specific control variables and I will control for country fixed effects
7 Norway is also not suitable for this research because there is already a gender quota of 40%, as previously discussed.
8
10
to limit the country-specific influences. Panel B from Table 1 shows that most of the firms belong to the services industry (20.62%). A substantial percentage of firms also belong to the consumer durables industry (14.59%), the basic industry (12.01%) and the capital goods industry (11.88%).9
3.2. Variables
3.2.1. Corporate risk-taking variable
Riskier corporate operations have more volatile returns to capital. So following previous studies (Acharya, Amihud, & Litov, 2011; Boubakri et al., 2013; Faccio, Marchica, & Mura, 2011; Hilary & Hui, 2009; John, Litov, & Yeung, 2008), the proxies used for the degree of corporate risk-taking are based on the volatility of corporate earnings. Based on Boubakri et al. (2013), I calculate the volatility of a firm’s earnings over five years as my primary measure of corporate risk-taking (RISK1):
where,
RISK1i,t indexes the firm i and year t, and EBITi,t is defined as the earnings before interest and taxes for firm i in year t; Ai,t is equal to the total assets; T over (0 to +4; +1 to +5; +2 to +6). So for a firm-year observation from 2006, the volatility of the firm’s ROA over 2006 to 2010 determines the corporate risk-taking behavior. In other words, future corporate behavior determines the volatility at T=0.
The firm’s earnings are defined by the return on assets (ROA) which is equal to the ratio of earnings before interest and taxes (EBIT) to total assets. Concerns regarding outliers are mitigated, because I winsorize ROA at the 1% level on both sides of the sample distribution. Still in line with Boubakri et al. (2013), I estimate four other proxies for corporate risk-taking to test for robustness of
RISK1. These variables are discussed in section ‘4.2.2. Robustness of RISK1’. In short, RISK2 is based on
the difference between the maximum ROA minus the minimum ROA over five years; RISK3 is similar to
RISK1, but is corrected for the average country risk-taking behavior over the same five years; RISK4 is
9 Industry classification is based on Campbell (1996).
11 based on the volatility of return on sales (ROS) instead of ROA; and R&D is the ratio of research & development costs over total assets. For a detailed explanation of these variables, I refer also to the Appendix.
Panel A of Table 2 shows descriptive statistics of the proxies for corporate risk-taking. Although Boubakri et al. (2013) uses a US sample based on the years 1981-2007, my descriptive statistics of the proxies for corporate risk-taking are still quite in line with their statistics. RISK1, the five-year volatility of ROA, shows a mean (median) of 0.052 (0.035). Negative values are impossible for RISK1, since it is measured as the volatility. Therefore, the minimum value of 0.000 is a logical value. The values for RISK2 are obviously higher with a larger spread as well, since it concerns the difference between the minimum and maximum ROA over 5 years instead of the volatility. The values of RISK3 are comparable to RISK1. The volatility of ROS is a bit higher just as in Boubakri et al. (2013), which results in a mean for RISK4 of 0.066 (0.034). The spread of RISK4 is twice the spread of RISK1 and this is also in line with the descriptive statistics of Boubakri et al. (2013). R&D shows a mean of 0.037 (0.018) with a relatively high spread and standard deviation.
3.2.2. Definition of a director in general
As stated before, this paper focuses only on non-executive female directors. Some research focuses on directors in general, executive directors only or on both separately. Therefore, it is important to keep a clear distinction between these types of directors to compare my findings with current literature. To avoid misunderstandings, I will shortly describe the differences for the sake of clarity.
12 Based on this structure and the crucial rights and obligations concerning the control rights, the risk-appetite of the non-executive directors is clearly expected to influence the risk-taking behavior of the firm.
An important notion to make, however, is the difference between unitary and two-tier boards in Europe. In a unitary board structure, the non-executive directors (NEDs) are seated together with the executive directors (EDs) in one board. In a two-tier board structure, however, the supervisory board purely consists out of NEDs and the executive board is formed by the EDs.10 In both cases, firms are obliged to report their number of EDs and NEDs who are seated in the board. These numbers are collected by BoardEx, which is my data source for my governance variables.
3.2.3. Female director variables
In line with previous research (Adams & Ferreira, 2009; Gul et al., 2011; Jurkus et al., 2011; Kang et al., 2010; Levi et al., 2013) about the influence of gender diversity on firm outcomes, I use two types of definitions of the presence of female directors. My primary measure is the fraction of female NEDs of the total NEDs (FEMALE1). For robustness purposes, I also test for another type of definition which consists of three different dummy variables. I also include these different female dummies in order to investigate whether a minimum number of female directors might be required to affect corporate risk-taking. The dummies are: FEMALEDUMMY1 which equals one for boards with at least one female NED and zero otherwise; FEMALEDUMMY2 which equals one for boards with at least two female NEDs and zero otherwise; and FEMALEDUMMY3 which equals one for boards with at least three female NEDs and zero otherwise.
In panel B of Table 2 a mean (median) of 0.077 (0.000) is shown for FEMALE1. Compared to the proposed EU quota of 40%, the mean but especially the median is remarkably low. Yet, the mean is in line with prior research such as Adams and Kirchmaier (2012), who present specific means for each European country in specific. All my included countries are present in their sample and their specific means fluctuate around my total mean. The maximum of my total sample is 0.667. When looking at the female dummies, panel B of Table 2 shows that 0.387 of total firm-years have at least one female NED and that this percentage falls dramatically to 0.129 for firm-years with at least two female NEDs. Only 0.039 of the total firm-years have at least 3 NEDs seated in their board.
13
3.2.4. Control variables
Next to the female directorship variables, I control for differences for firm-level and country-level influences. I follow the convention of prior literature and include standard control variables that prior studies show to be associated with corporate risk-taking (Boubakri et al., 2013; John et al., 2008). I will describe the specific association of each variable with corporate risk-taking in the next two sections.11
Table 2
Descriptive statistics for the regression variables
This table displays descriptive statistics for the key regression variables used to examine the relationship between the fraction of female directors and corporate risk-taking. The variables are explained in section '3.2. Variables' and are outlined in the Appendix as well.
Mean Median Maximum Minimum Std. Dev. Observations
Panel A: Corporate risk-taking variables
RISK1 0.052 0.035 0.335 0.000 0.049 2,289
RISK2 0.127 0.086 0.698 0.000 0.119 2,289
RISK3 0.051 0.037 0.321 0.002 0.044 2,289
RISK4 0.066 0.034 0.632 0.000 0.089 2,289
R&D 0.037 0.018 0.691 0.000 0.061 1,032
Panel B: Female directorship variables
FEMALE1 0.077 0.000 0.667 0.000 0.119 2,289
FEMALEDUMMY1 0.387 0.000 1 0 0.487 2,289
FEMALEDUMMY2 0.129 0.000 1 0 0.336 2,289
FEMALEDUMMY3 0.039 0.000 1 0 0.193 2,289
Panel C: Firm-level control variables
BOARDINDEP 0.608 0.667 1.000 0.000 0.359 2,289 SALESGROWTH 0.157 0.094 1.861 -0.438 0.314 2,289 SIZE 4.788 4.837 6.904 2.381 0.977 2,289 ROA 0.109 0.091 0.399 0.007 0.074 2,289 LEVERAGE 0.229 0.208 0.693 0.000 0.170 2,289 CAPEX 0.052 0.039 0.272 0.001 0.048 2,289
Panel D: Country-level control variables
HERFINDAHL 0.064 0.052 0.316 0.018 0.046 2,289 GDPGROWTH -0.001 0.017 0.070 -0.156 0.067 2,289 ECONFREEDOM 7.610 7.670 8.020 6.840 0.331 2,289 ECONDEVELOP 1.469 1.467 1.613 1.170 0.069 2,289 11
14
3.2.4.1. Firm-level control variables
15 higher leverage increases the financial risk which consequently could inflate earnings volatility (González, 2013). Sixth, I include the ratio of capital expenditures to total assets (CAPEX). High capital expenditures are associated with higher earnings volatility, because investments are often high and returns are uncertain (Bhandari, 1981). Finally, I add country, year and industry dummies to control for the different fixed-effects. The industry-categorization is based on Campbell (1996). I winsorize all firm-level control variables at the 1% firm-level in each tail of the sample distribution to limit the influence of outliers in the variables.
Panel C of Table 2 shows descriptive statistics of the included firm-level control variables. It shows that, on average, 0.608 of the NEDs are classified as independent. Furthermore, it shows that firms in the sample exhibit a high level of SALESGROWTH with a mean (median) of 0.157 (0.094) and they appear to be profitable with a mean ROA of 0.109 (0.091). The LEVERAGE ratio of the firms ranges from unlevered firms to highly levered firms with a maximum of 0.693 and an average of 0.229 (0.208). The average capital expenditures (CAPEX) are 0.052 (0.039). Compared to other studies, especially
SALESGROWTH and ROA seem to be quite high. Additional analysis shows that SALESGROWTH is inflated
by the economic boom of 2006 and 2007; the mean (median) for 2006, 2007 and 2008 are respectively: 0.195 (0.122), 0.193 (0.113) and 0.083 (0.033). ROA seems to be less influenced by the economic cycle: 0.113 (0.096), 0.114 (0.093) and 0.101 (0.085). The remaining variables are more in line with other studies.
3.2.4.2. Country-level control variables
16 This should capture the influence of economic growth of a country on managerial investment decisions. Next, Boubakri et al. (2013) argue the importance of controlling for the economic freedom (ECONFREEDOM) and economic development (ECONDEVELOP), because they both are important determinants for investors to allocate their funds. In the end, risky projects require investors. Though all included countries are from the EU, there are still differences between national regulations and legal systems. Higher economic freedom and development should lead to a more friendly environment for investors and managers. Innovative (riskier) projects could be implemented more easily in these countries which contribute to higher earnings volatility. The economic freedom index of Gwartney, Lawson and Hall (2013) includes several regulation, institutional and investment assessments of all included countries.12 Since I am not interested in all these effects individually, this index suits perfectly for this research since it controls for many different country effects together in one index. Finally, the economic development is approximated by the log of the GDP per capita.
The Herfindahl-index shows a mean of 0.064 (0.052). The spread between the minimum and maximum value of this index is quite large. Finland has the least competitive capital market of the sample. GDP growth is negative with a mean of -0.001 (0.017) which is probably caused by the crisis at the end of 2007. This is mostly influenced by the GDP of the UK which decreased by 0.156 in 2008. The economic development variable has a mean of 1.469 (1.467). Finally, economic freedom shows a mean of 7.610 (7.670).
3.2.5. Correlation coefficients
Table 3 shows the correlation coefficients matrix and the possible multicollinearity issues seem to be limited. Obviously, FEMALE1 and the female dummies are highly correlated but they are used as substitutes so multicollinearity between these variables will not be an issue. The control variables
ECONFREEDOM and ECONDEVELOP are, however, also correlated with a coefficient of 0.543, which may
result in multicollinearity problems. Therefore, I also estimate the model by not putting them both in the same model in the next chapter.13
12
This index, scaled from 1-10, has 5 main indicators: i) size of government: taxes, subsidies, investment; ii) legal system and property rights: e.g. legal enforcement, integrity of the legal system; iii) sound money: inflation, money growth; iv) freedom to trade internationally: e.g. tariffs, trade-barriers; v) regulation: credit-market regulation, labor-market regulation, business regulation.
13
17 Table 3
Correlation coefficients matrix
This table report correlations for the regression variables for the full sample.
RISK1 FEMALE1 FEMALE- FEMALE- FEMALE- BOARD- SALES- SIZE ROA LEVERAGE CAPEX HERFIN- GDP- ECON- ECON-
DUMMY1 DUMMY2 DUMMY3 INDEP GROWTH DAHL GROWTH FREEDOM DEVELOP
RISK1 1.000 FEMALE1 -0.113 1.000 FEMALEDUMMY1 -0.156 0.816 1.000 FEMALEDUMMY2 -0.121 0.618 0.485 1.000 FEMALEDUMMY3 -0.092 0.366 0.253 0.522 1.000 BOARDINDEP -0.010 -0.011 -0.048 -0.127 -0.081 1.000 SALESGROWTH 0.175 -0.100 -0.117 -0.079 -0.063 -0.076 1.000 SIZE -0.303 0.183 0.379 0.249 0.190 0.012 -0.149 1.000 ROA 0.219 0.010 -0.013 -0.015 0.032 0.051 0.026 0.144 1.000 LEVERAGE -0.182 0.073 0.136 0.063 0.020 0.003 -0.060 0.334 -0.160 1.000 CAPEX -0.041 0.013 0.030 0.004 -0.006 0.026 -0.019 0.163 0.124 0.210 1.000 HERFINDAHL -0.050 0.088 0.079 0.069 0.065 0.174 -0.035 0.103 -0.008 0.131 0.050 1.000 GDPGROWTH -0.027 0.036 0.073 0.087 0.059 -0.124 0.163 0.163 0.030 0.041 0.035 0.220 1.000 ECONFREEDOM 0.242 -0.078 -0.159 -0.164 -0.126 0.305 0.069 -0.366 0.112 -0.233 -0.082 -0.238 -0.150 1.000 ECONDEVELOP 0.167 0.090 0.074 0.051 0.040 0.183 0.057 -0.157 0.115 -0.197 -0.128 0.117 0.230 0.543 1.000 Table 4 Univariate analyses
This table reports the univariate analyses of corporate risk-taking (RISK1) by the fraction of female directors (FEMALE1) and the presence of respectively 1,2 or 3 female directors FEMALEDUMMY1,
FEMALEDUMMY2 and FEMALEDUMMY3.
Dependent Variable: Corporate risk-taking, RISK1
Variable Coefficient Prob. Variable Coefficient Prob. Variable Coefficient Prob. Variable Coefficient Prob.
C 0.056 0.000 C 0.058 0.000 C 0.054 0.000 C 0.053 0.000
18
4. Empirical results
In this section, I present the results of the empirical analyses of the relationship between the fraction of female directors and corporate risk-taking. It starts with a univariate analysis about the plain relationship of these two variables without controlling for the other potential determinants of corporate risk-taking. Second, I discuss a multivariate analysis where all the control variables are included. Third, I address the potential endogeneity issues. Then, I discuss the interaction effect of board independency on the relationship between the fraction of female directors and corporate risk-taking. Finally, I address some limitations of the generalizability of the results.
4.1. Univariate analysis
The results of the univariate analysis are presented in Table 4. For all proxies of the presence of female directors there is a highly significant negative relationship between corporate risk-taking and the presence of female directors. The coefficient for FEMALE1 of -0.046 is also economically significant: the mean for RISK1, from Table 2, is 0.052 and if the fraction of female NEDs would rise with 10% , the dependent variable, RISK1, will drop by -0.004614. This reduces RISK1 from 0.052 to 0.0474, which is a decrease of 8.85%. This is quite a difference in terms of risk-taking behavior. The coefficients for the dummy variables for at least one, two or three female directors are, respectively, -0.016, -0.018 and -0.023. As expected, this indicates that the presence of more female directors involves more influence on corporate risk-taking.
Although these results provide preliminary support for the hypothesis, the univariate analysis is no more than an indication for a possible relationship. Other explanatory variables for corporate risk-taking are not included in the univariate analysis. Obviously, there are, other variables which could have an influence on corporate risk-taking as well. Therefore, I will now proceed to the multivariate analysis.
4.2. Multivariate analysis
In this section, I report the results from the multivariate analysis about the relationship between corporate risk-taking and the fraction of female directors. The variables are discussed previously in this paper. I construct a panel framework and I estimate the regressions using ordinary least squares (OLS).
14
19 The primary regression is the following:
The definitions of the controls have been previously described in section ‘3.2. Variables’ and are discussed in the Appendix. CONTROLS consist of all presented control variables (BOARDINDEP,
SALESGROWTH, SIZE, ROA, LEVERAGE, CAPEX, HERFINDAHL, GDPGROWTH, ECONFREEDOM and ECONDEVELOP) and ε is an error term. The focus lies on because it measures the sensitivity of corporate risk-taking with respect to a change in the fraction of female directors.
4.2.1. The impact of female directors on corporate risk-taking
The results of the primary regression analysis are presented in Table 5. Model 1 suggests that
FEMALE1 is negatively associated with corporate risk-taking and that this association is statistically
significant at the 1% level. These results are in line with the univariate analysis and support hypothesis H1, which states that a higher fraction of female NEDs results in more conservative corporate risk-taking behavior. This association is also economically significant with a coefficient estimate for FEMALE1 of 0.022; increasing the fraction of female directors from the first to the tenth decile (0.000 to 0.250)15 results in a 10.9% decrease of the proxy for corporate risk-taking (0.060 to 0.055), holding all other independent variables at their mean values.16
The control variables show several significant relations that are consistent with John et al. (2008) and Boubakri et al. (2013). Next to the intercept, there are five significant control variables at the 1% level (BOARDINDEP, SALESGROWTH, SIZE, ROA, and ECONFREEDOM) and one control variable at the 5% level (ECONDEVELOP). Except for ROA, these variables are related to corporate risk-taking according to the expected sign. It is difficult to determine the reason for the unexpected sign for ROA which is also not consistent with the theory that poor performance could lead to excessive risk-taking to meet performance targets. Theories of overconfidence or excessive risk-taking to live up to the expectations of shareholders to maintain the high levels of ROA could explain the positive sign. A limitation of the sample is the relatively short time period of three years. Therefore, the economic boom of 2006 and
15
First decile of FEMALE1 is 0.000 and the tenth decile of FEMALE1 is 0.250 16
20 2007 with excessive ROAs and prosperity might have influenced the unexpected sign of ROA. The other four control variables are not significant and only HERFINDAHL is showing the expected sign.
In Models 2-4 of Table 5, I replace FEMALE1 with one of the three female dummies to check for robustness of my proxy for the presence of female directors and to check whether a minimum number of female directors is required to be influential on corporate risk-taking. Model 2 suggests that there is no significant association between the presence of at least one female director and corporate risk-taking, although, it reports a negative coefficient. Huse and Solberg (2006) report the same outcome as follows:”When a board member is appointed because she is a woman, and she is the only woman on the board, it is easy for her ideas to be swept aside, ignored, or generally dismissed.” My results might present evidence for this theory. Model 3 reports the results for the regression with FEMALEDUMMY2 as proxy for the presence of at least two female directors. Based on these outcomes, there appears to be a significant negative association at the 10% level when there are at least two female directors seated on the board. In Model 4, I report the results for the regression with at least three female directors. The association becomes stronger with respect to FEMALEDUMMY2 considering the higher coefficient (-0.005 to -0.009) and higher t-statistic (-1.809 to -1.911). From Model 2-4 it can be concluded that one female director is not enough to influence the corporate risk-taking behavior, but the dummies for at least 2 or 3 female directors provide additional evidence that more female directors are associated with lower corporate risk-taking. Therefore, my basic model is robust for alternate proxies of the presence of female directors on the board, because all three dummies present negative signs of which 2 are significant. Thereby, it provides additional insights that only one female director is not enough to have a significant influence. Liu et al. (2013) use the same female dummies to detect the influence of female directors on firm performance in China. Although they do not link female directors to corporate risk-taking, it is interesting to see that they find the same pattern; no significant influence for only one female director, but a rising coefficient and significance for at least two and three female directors. My results are also in line with Konrad and Kramer (2006), who state that at least three women on an average board are necessary to avoid tokenism and actually derive benefit from their representation. According to Konrad and Kramer (2006), there is a high risk of tokenism when a woman serves alone. This risk decreases when two women are seated and dynamics improve as well, but tokenism may still exist. Only when three or more women are present in a board, they can start to make a difference in improving corporate governance.
21 coefficient of 0.543. This could lead to multicollinearity problems and therefore Model 5 leaves the
ECONFREEDOM variable out of the regression and Model 6 leaves ECONDEVELOP out of the regression.
22
Table 5
Regressions of corporate risk-taking on the presence of female directors This table reports the OLS estimation of the following model:
where RISK1 is a proxy for corporate risk-taking, FEMALE1 is the ratio of female NEDs over total NEDs and CONTROLS is a set of control variables (BOARDINDEP, SALESGROWTH, SIZE, ROA, LEVERAGE, CAPEX, HERFINDAHL, GDPGROWTH, ECONFREEDOM and ECONDEVELOP). Model 1 is my basic model and shows the regression for my primary measure: FEMALE1. In model 2-4 the female variable is replaced by a dummy for respectively at least 1,2 or 3 female NED(s) in the board. Model 5 and 6 deal with possible multicollinearity issues since ECONFREEDOM and ECONDEVELOP show a relatively high correlation (0.543). The definitions of the variables are available in section ‘3.2. Variables’ and are outlined in the Appendix. Beneath the estimate, I report the t-statistic in parentheses. The superscript asterisks ***, ** and * denote statistical significance at, respectively, the 1%, 5% and 10% levels.
Dependent Variable: Corporate risk-taking, RISK1
Basic model Female dummies Multicollinearity
23 Though countries in the EU are relatively uniform, it is likely that the results are still influenced by country effects. The financial crisis is also likely to create heterogeneity between the years. The pooled regression does not distinguish between the different countries except for the country-level control variables, nor does the pooled regression account for differences between years or industries. In other words, by combining all countries, years and industries by pooling, the model denies for heterogeneity that may exist among these countries, years or industries. Therefore, I control for these effects by estimating a fixed effects panel data regression. The Hausman test points out that fixed effects estimation is preferred over random effects estimation; thereby the Wald test shows that fixed effects estimation is preferred over pooled regression. So, based on both the Hausman test and the Wald test, fixed effects estimation is the most appropriate model.
To estimate the fixed effects panel data regression, I created dummy variables for the countries, industries and years which lead to the following model:
∑ ∑ ∑
Table 6 displays the results and it can be observed that there is still a significant negative association between female directors and corporate risk-taking. The coefficient of FEMALE1 dropped, however, from -0.022 to -0.018 and the significance level dropped from 1% to 5%. Still, based on the outcomes of Table 6 it can be concluded that the negative association between female directors and corporate risk-taking is robust for heterogeneity among countries, years and industries. Thereby, the control variables show similar results compared to the Model without the fixed effects. BOARDINDEP,
SALESGROWTH, SIZE and ROA have nearly the same coefficient and are still significant at the 1% level.
24
Table 6
Regression of corporate risk-taking on the presence of female directors including fixed effects This table reports the OLS estimation of the following model:
where RISK1 is a proxy for corporate risk-taking, FEMALE1 is the ratio of female NEDs over total NEDs and CONTROLS is a set of control variables (BOARDINDEP, SALESGROWTH, SIZE, ROA, LEVERAGE, CAPEX, HERFINDAHL, GDPGROWTH, ECONFREEDOM and ECONDEVELOP). COUNTRY, INDUSTRY and YEAR are the fixed effects terms. Model 1 is my basic model and shows the regression for my primary measure: FEMALE1. The definitions of the variables are available in section '3.2. Variables' and are outlined in the Appendix. Beneath the estimate, I report the t-statistic in parentheses. The superscript asterisks ***, ** and * denote statistical significance at, respectively, the 1%, 5% and 10% levels.
Dependent Variable: Corporate risk-taking, RISK1 Basic model Variable (prediction) (1) C 0.084 (0.644) FEMALE1 (-) -0.018** (-2.144) BOARDINDEP (-) -0.009*** (-2.838) SALESGROWTH (+) 0.017*** (5.623) SIZE (-) -0.013*** (-10.218) ROA (-) 0.148*** (10.935) LEVERAGE (+) -0.005 (-0.713) CAPEX (+) -0.013 (-0.597) HERFINDAHL (+) 0.005 (0.037) GDPGROWTH (+) -0.009 (-0.176) ECONFREEDOM (+) -0.028 (-0.336) ECONDEVELOP (+) 0.007 (0.171)
COUNTRY FIXED EFFECTS YES
INDUSTRY FIXED EFFECTS YES
YEAR FIXED EFFECTS YES
25
4.2.2. Robustness of RISK1
In this section I discuss the robustness of my primary proxy for corporate risk-taking: RISK1. In Models 3 to 10 from Table 7, I use alternative proxies for corporate risk-taking to mitigate concerns that the evidence would be driven by the measurement of RISK1. Table 7 presents the regressions with respectively RISK1, RISK2, RISK3, RISK4 and R&D as dependent variables. Thereby, for every proxy it presents a regression with and without taking account for country, industry and year fixed effects which results in ten models. As described before, RISK2 is based on the difference between the maximum ROA minus the minimum ROA over five years; RISK3 is similar to RISK1, but is corrected for the average country risk-taking behavior over the same five years; RISK4 is based on the volatility of return on sales (ROS) instead of ROA; and R&D is the ratio of research & development costs over total assets. For a detailed explanation of these variables, I refer again to the Appendix.
In Models 3 and 4, I use RISK2 as a proxy for corporate risk-taking. My initial results remain unchanged and are still highly significant. The coefficient for corporate risk-taking even doubles for these Models, which indicates an even stronger relationship. Models 5 and 6 concern the country-adjusted proxy. This proxy is based on Boubakri et al. (2013) and John et al. (2008), who confirm that this method is challenging for a variety set of reasons. First, a limited number of observations is used to calculate the average ROA for a given country-year which is questionable for some countries. This especially holds for countries with a small contribution to the sample such as Finland or Denmark.17 Second, I already control for economic conditions in my regressions by including country-level control variables and country-fixed effects. Nevertheless I include RISK3 to offset my results more easily against the results of Boubakri et al. (2013) and John et al. (2008). Table 7 indicates that FEMALE1 is negatively associated with corporate risk-taking for both Models 5 and 6. However, only Model 5 shows a significant relation at the 10% level. Models 7 and 8 are calculated on the basis of the volatility of ROS. According to Boubakri et al. (2013), the use of ROS mitigates concerns that ROA is sensitive to accounting conventions and inflation. Model 7 shows a significant negative association at the 5% level, but Model 8 presents an insignificant negative association. In Model 9 and 10, I measure corporate risk-taking as the ratio of R&D costs over total assets based on previous studies highlighted by Boubakri et al. (2013). The probability of success for these investments is low and their benefits are uncertain, thereby their success (if achieved) materializes in the long run. This proxy for corporate risk-taking does not present a significant relation between FEMALE1 and risk-taking.
17
27
Table 7
Robustness tests for the proxy of corporate risk-taking This table reports the OLS estimation of the following model:
where RISK is one of the five proxies for corporate risk-taking, FEMALE1 is the ratio of female NEDs over total NEDs and CONTROLS is a set of control variables (BOARDINDEP, SALESGROWTH, SIZE, ROA, LEVERAGE, CAPEX, HERFINDAHL, GDPGROWTH, ECONFREEDOM and ECONDEVELOP). COUNTRY, INDUSTRY and YEAR are the fixed effects terms, which are only included in model 2, 4, 6, 8 and 10. Model 1 is my basic model and shows the regression for my primary measure: FEMALE1. The definitions of the variables are available in section '3.2. Variables' and are outlined in the Appendix. Beneath the estimate, I report the t-statistic in parentheses. The superscript asterisks ***, ** and * denote statistical significance at, respectively, the 1%, 5% and 10% levels.
Dependent Variable: Corporate risk-taking
Variable (prediction) RISK1 RISK1 RISK2 RISK2 RISK3 RISK3 RISK4 RISK4 R&D R&D
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) C -0.061** 0.084 -0.141** 0.250 -0.101*** 0.038 -0.038 0.479* 0.076 0.048 (-2.103) (0.644) (-1.999) (0.794) (-3.932) (0.327) (-0.683) (1.952) (1.166) (0.172) FEMALE1 (-) -0.022*** -0.018** -0.055*** -0.044** -0.013* -0.011 -0.031** -0.023 -0.009 -0.004 (-2.780) (-2.144) (-2.810) (-2.220) (-1.862) (-1.471) (-1.997) (-1.454) (-0.540) (-0.233) BOARDINDEP (-) -0.007*** -0.009*** -0.018*** -0.023*** -0.006** -0.008*** -0.009 -0.021*** 0.003 0.006 (-2.560) (-2.838) (-2.623) (-2.865) (-2.410) (-2.729) (-1.617) (-3.265) (0.561) (0.901) SALESGROWTH (+) 0.018*** 0.017*** 0.042*** 0.041*** 0.015*** 0.015*** 0.047*** 0.043*** -0.001 -0.002 (5.805) (5.623) (5.637) (5.451) (5.534) (5.198) (7.975) (7.391) (-0.192) (-0.235) SIZE (-) -0.013*** -0.013*** -0.032*** -0.031*** -0.012*** -0.012*** -0.014*** -0.013*** -0.010*** -0.007*** (-11.487) (-10.218) (-11.395) (-10.161) (-12.157) (-10.925) (-6.243) (-5.445) (-5.106) (-3.206) ROA (-) 0.159*** 0.148*** 0.384*** 0.356*** 0.088*** 0.077*** 0.046* 0.031 0.115*** 0.103*** (11.958) (10.935) (11.918) (10.884) (7.403) (6.403) (1.789) (1.231) (4.241) (3.760) LEVERAGE (+) -0.004 -0.005 -0.012 -0.014 0.000 -0.003 0.019 0.009 -0.070 -0.055 (-0.656) (-0.713) (-0.815) (-0.889) (-0.031) (-0.494) (1.606) (0.769) (-5.208) (-3.885) CAPEX (+) -0.007 -0.013 -0.013 -0.028 -0.019 -0.021 0.168*** 0.066 -0.122** -0.061 (-0.348) (-0.597) (-0.275) (-0.546) (-1.083) (-1.133) (-4.288) (1.644) (-2.420) (-1.141) HERFINDAHL (+) 0.020 0.005 0.050 -0.031 0.014 -0.005 0.024 -0.114 -0.041 0.065 (0.878) (0.037) (0.892) (-0.088) (0.679) (-0.036) (0.594) (-0.418) (-0.931) (0.234) GDPGROWTH (+) -0.013 -0.009 -0.038 -0.020 -0.017*** -0.022 -0.020 -0.088 -0.080** -0.079 (-0.809) (-0.176) (-0.991) (-0.168) (-1.258) (-0.511) (-0.678) (-0.960) (-2.325) (-0.805) ECONFREEDOM (+) 0.014*** -0.028 0.032*** -0.071 0.016*** 0.025 0.028*** -0.276* -0.030*** 0.136 (3.333) (-0.336) (3.188) (-0.358) (4.446) (0.334) (3.477) (1.786) (-3.421) (0.780) ECONDEVELOP (+) 0.039** 0.007 0.097** 0.012 0.055*** 0.003 -0.042 0.004 0.167*** -0.024 (2.099) (0.171) (2.174) (0.356) (3.340) (0.220) (-1.178) (0.170) (3.388) (0.340)
COUNTRY FIXED EFFECTS NO YES NO YES NO YES NO YES NO YES
INDUSTRY FIXED EFFECTS NO YES NO YES NO YES NO YES NO YES
YEAR FIXED EFFECTS NO YES NO YES NO YES NO YES NO YES
28
4.2.3. Endogeneity issues
A potential concern with my analyses is that FEMALE1 may not be exogenous. Prior corporate governance research indicates the presence of endogeneity issues. 18 Indeed, Adams et al. (2010) state that there are both theoretical arguments and empirical evidence to regard board structure as endogenous. Some omitted unobserved determinants may both explain corporate risk-taking and the fraction of female directors (FEMALE1). Such omitted variables could lead to spurious correlations between corporate risk-taking and the fraction of female directors, which would lead to biased and inconsistent OLS estimates. Duchin et al. (2010) also state that board composition is endogenous, which makes it difficult to present reliable evidence that board composition would influence corporate outcomes. It could be possible, for instance, that some firms in my sample are prone to a ‘cowboy culture’19 which is both associated with high risk-taking and with males. In this case the ‘cowboy culture’ would be an omitted variable in my analysis, which would distort my results. Or, for example, a firm with high risk-taking behavior that needs to fulfil the requirements of regulation, governance principles or society20 of a minimum of female directors, could specifically look for female directors with a more risk-seeking profile. This would also bias the results of this paper. Adams et al. (2010) and Duchin et al. (2010) clearly point out the possible endogeneity issues regarding board composition. Since the gender diversity of NEDs is part of board composition, I need to control for these possible endogeneity issues as well.
The possibility of reverse causality is also part of these endogeneity issues. In my analyses, I presume that directors influence the corporate risk-taking behavior. But firms with a higher risk-taking profile may also be less likely to hire female directors. According to Adams et al. (2010), much of what is learnt from research about corporate boards is about equilibrium associations. Causality is in most cases almost impossible to determine. Directors are nearly always selected through a certain endogenous equilibrium selection process; a random allocation of directors is not very likely. A newly imposed gender quota, which was the case in Norway for instance, is the best approximation possible of a natural experiment. But these cases are rare. My sample is not based on a newly imposed gender quota or a comparable exogenous shock and therefore, an endogenous selection process could be possible in my
18 For an extensive research about the issues of endogeneity regarding the role of the board of directors on governance structures, I refer to Adams et al. (2010)
19
Corporate cowboys are associated with excessively risk-taking and are mostly men. For example, the former leaders of Enron are regarded as corporate cowboys.
20
29 sample. This makes it, again, hard to exclude the possibility that both corporate risk-taking behavior and
FEMALE1 are driven by some omitted variable. Therefore, I perform an additional analysis to control for
possible endogeneity issues.
To address the possible endogeneity in my basic model of Table 5, I reestimate the Model using IV techniques. Including fixed effects could partly mitigate endogeneity problems, but is not sufficient. Adams and Kirchmaier (2012) present evidence that low labor force participation by females is one of the main reasons few women sit on corporate boards. Consequently, they show that female labor force participation (PARTICIPATION) is positively related to the fraction of female directors on corporate boards. Because labor force participation is highly unlikely to be related to corporate risk-taking, I may use this as an instrumental variable (IV) to correct for the endogeneity of FEMALE1. Table 8 presents the two-stage least squares (2SLS) regression with female labor force participation as IV. The first-stage model is reported in Model 1, where the correlation between FEMALE1 and PARTICIPATION is estimated. Based on a t-statistic for PARTICIPATION of -6.101, the used IV is clearly correlated with
FEMALE1 at the 1% significance level. In Model 2, I report the second stage regression. Though the
coefficient still reports a negative sign, the coefficient is not significant anymore after controlling for endogeneity. Therefore, I must conclude that the negative association found in chapter 4 between RISK1 and FEMALE1 is not robust to this method of addressing the endogeneity of FEMALE1, because I cannot provide evidence that FEMALE1 is exogenous. This highlights the relevance of addressing endogeneity in corporate governance research.
30
Table 8
IV regressions of corporate risk-taking on the presence of female directors This table reports the OLS estimation of the following model:
where RISK1 is a proxy for corporate risk-taking, FEMALE1 is the ratio of female NEDs over total NEDs and CONTROLS is a set of control variables (BOARDINDEP, SALESGROWTH, SIZE, ROA, LEVERAGE, CAPEX, HERFINDAHL, GDPGROWTH, ECONFREEDOM and ECONDEVELOP). Model 1 reports the first stage of an instrumental variable (IV) regression with PARTICIPATION as an instrument for FEMALE1. Model 2 reports the results of the second stage IV estimation. The definitions of the variables are available in section '3.2. Variables' and are outlined in the Appendix. Beneath the estimate, I report the t-statistic in parentheses. The superscript asterisks ***, ** and * denote statistical significance at, respectively, the 1%, 5% and 10% levels.
Dependent Variable
FEMALE1 RISK1
Variable (prediction) first stage second sage
31
4.2.4. Female directors and board independence
In this section, I address the interaction of board independence and female directors regarding the influence on corporate risk-taking. All the previously discussed results indicate that board independence (BOARDINDEP)21 is a highly significant variable with respect to corporate risk-taking. Based on current literature, this is no surprise. As discussed in section 2 ‘literature review’, Adams and Ferreira (2009) state that female directors ought to be more independent, since they usually do not belong to the “old boys club”. Their influential paper presents evidence of the impact of women on governance and performance. However, their paper is not able to present clear evidence that their results are not mainly driven just by the level of independence which rises on average with more female directors. This raises the question whether the impact of female directors on corporate risk-taking would change with different levels of board independence.
According to Dalton and Dalton (2010), the value added by independent directors can be seen as analogous to the value added by female directors; both types of directors bring fresh perspectives to the table. However, boards need to be open for fresh perspectives to benefit from different ideas. Fitzsimmons (2012) states that a diverse and open-minded culture is an essential condition for organizations to benefit from board diversity: one that welcomes fresh perspectives instead of training every board member to think and act the same way. The risk to marginalize the input of female directors would therefore be reduced by diverse board cultures. Important to note is that these cultures do not necessarily need to be national cultures. Cultures from other industries or resulting from demographic differences are crucial for the type of culture in the board.
I will test for the moderating effect of board culture regarding the influence of female directors on corporate risk-taking by using the board independency ratio as a proxy for the diverse and open-minded culture. Since independent directors are not affiliated with the firm’s business, they ought to provide more fresh perspectives. If these fresh perspectives are not desired, the number of independent directors will probably be lower. Therefore, I include an interaction term: FEMALE1*BOARDINDEP. In my sample, BOARDINDEP is not positively correlated with FEMALE1.22 This is somewhat contradicting to the hypothesis that female directors are more independent.23 Most of prior research is based on US samples, however my sample is based on European firms and in Europe there are still many family firms. A possible explanation could be that the small negative correlation is caused by a high representation of
21 For a definition, I refer back to ‘3.2.4.1. Firm-level control variables’ 22
Table 3 reports a correlation coefficient between BOARDINDEP and FEMALE1 of -0.011 23
32 family tied female directors, who are not independent. The absence of a correlation suits well for this moderation term, because the term is not affected by women in the board. According to the previously discussed theory, the influence of FEMALE1 should be smaller for a lower level of BOARDINDEP and bigger for a higher level of BOARDINDEP; a positive sign would be expected for this interaction term. This would imply that the higher the board independency is (as a proxy for a diverse and open-minded board culture), the greater the effect of female directors on corporate risk-taking would become. The results of this regression are presented in Table 9.
33
Table 9
Regression of corporate risk-taking on the presence of female directors including the interaction term This table reports the OLS estimation of the following model:
where RISK1 is a proxy for corporate risk-taking, FEMALE1 is the ratio of female NEDs over total NEDs, FEMALE1*BOARDINDEP is an interaction term and CONTROLS is a set of control variables (BOARDINDEP, SALESGROWTH, SIZE, ROA, LEVERAGE, CAPEX, HERFINDAHL, GDPGROWTH, ECONFREEDOM and ECONDEVELOP). COUNTRY, INDUSTRY and YEAR are the fixed effects terms. Model 1 is my basic model and shows the regression for my primary measure: FEMALE1. The definitions of the variables are available in section ‘3.2. Variables’ and are outlined in the Appendix. Beneath the estimate, I report the t-statistic in parentheses. The superscript asterisks ***, ** and * denote t-statistical significance at, respectively, the 1%, 5% and 10% levels.
Dependent Variable: Corporate risk-taking, RISK1
Variable (prediction) (1) C 0.079 -0.608 FEMALE1 (-) -0.056*** (-3.733) FEMALE1*BOARDINDEP (+) 0.064*** (3.055) BOARDINDEP (-) -0.014*** (-3.788) SALESGROWTH (+) 0.017*** (-5.577) SIZE (-) -0.013*** (-10.328) ROA (-) 0.148*** (-10.977) LEVERAGE (+) -0.004 (-0.556) CAPEX (+) -0.011 (-0.527) HERFINDAHL (+) 0.016 (-0.110) GDPGROWTH (+) -0.009 (-0.181) ECONFREEDOM (+) -0.027 (-0.332) ECONDEVELOP (+) 0.007 (0.548)
COUNTRY FIXED EFFECTS YES
INDUSTRY FIXED EFFECTS YES
YEAR FIXED EFFECTS YES
34
4.2.5. Generalizability
Next to the implications of endogeneity for econometric analysis, Adams et al. (2010) stress the implications of how to interpret actual governance practices. Heterogeneity between firms plays a key role in this issue. Models must abstract away from many features of real-life corporations, because corporations are too complex to be fully modelled. This increases the difficulty to understand the complicated and multifaceted solutions which firms apply to solve their governance issues. As Adams et al. (2010) describe: “The optimal governance structure might involve a certain type of board, operating in a certain fashion, having implemented a particular incentive package, and responding in certain ways to feedback from the relevant product and capital markets.” Including all those features in a model is obviously infeasible. Therefore, Adams et al. (2010) ask themselves whether one can expect the assumption of ceteris paribus with respect to the non-modeled features of the situation to be reasonable. Holding all other aspects constant leads to a constrained answer, which may differ from the unconstrained answer. This would lead to inaccurate conclusions. I will exemplify this principle based on a simple model from Adams et al. (2010).
Economic theory prescribes survival of the fittest through competition, so inappropriate governance structures will fail in the end. Therefore, if a ‘poor’ governance structure is observed, why did the firm deliberately choose for this structure? The observed ‘sub-optimal’ governance structure might represent the right solution to the constrained optimization problem the organization faces. This issue makes it hard to generalize outcomes of governance research. Adams et al. (2010) illustrate this
Figure 2a: Relation between a specific firm attribute Figure 2b: The real decision faced by firms and firm financial performance
35 issue with the relation of a governance attribute (board size) and firm performance. Figure 2a underscores the apparent negative relation between board size and financial performance. Without further analysis, it is tempting to conclude that shrinking a firm’s board would lead to better financial performance. However, this conclusion fails to consider why firm 2 might have chosen for a larger board. Figure 2b shows the real decisions faced by the firms. It shows that the relation is non-monotonic; in this case it is concave. From Figure 2b it can be observed that both firms are at their maximum and firm 2 would do worse by shirking their board, which would be the logical conclusion from Figure 2a. This situation illustrates the limited ability to generalize the results regarding board composition and firm outcomes.
The previously described situation is also applicable on my research, where the ratio of female directors is the governance attribute and firm performance is replaced by corporate risk-taking. Based on previous results it is not likely that additional female directors will increase corporate risk-taking. In other words, a monotonic relationship is expected in contrast to the example of Adams et al. (2010). Based on Figure 3a, an additional female non-executive director would lead to equally less corporate risk-taking for every firm. In other words, the relationship is linear with the same coefficient for every firm. However, the sensitivity of corporate risk-taking with respect to additional female directors may vary between firms, as discussed in the previous section. Therefore, Figure 3b shows the real decision faced by firms. This Figure illustrates that the individual curves could vary per firm and that they are not linear, which is important to understand for firm policymakers. In line with the results presented in Table 5, the influence of female directors on corporate risk-taking is expected to increase with more
Figure 3a: Relation between FEMALE1 Figure 3b: The real decision faced by firms
and corporate risk-taking
