The influence of board gender diversity on corporate risk- taking: a moderating role of director’s age
Name: Sander Schimmel Student number: 12931470 Thesis supervisor: Réka Felleg Date: August 15, 2021
Word count: 12697,
MSc Accountancy & Control Specialization: control
EBEC code: EC 20210127100130
Faculty of Economics and Business, University of Amsterdam
Statement of Originality
This document is written by student Sander Schimmel who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
UvA Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
In recent years, there has been a lot of discussion about the proportion of women in top management at companies. Some countries already have a mandatory female quota in the board of directors, but it is unclear what the financial effects are for companies. Studies in the social psychological field have examined that women are more risk averse than men. However, this difference in risk-taking behavior applies more for young men and women than for older men and women. Therefore, in this study the direct effect of board gender diversity on corporate risk-taking has been examined, but also the moderating effect of director’s age on this direct relationship. Social role theory predicts a negative effect of board gender diversity on corporate risk-taking and risk-sensitivity theory and young male syndrome predict a negative moderating effect of director’s age on the relationship between board gender diversity and corporate risk- taking. This study contains risk data about variability of return on assets, leverage and R&D intensity from 230 U.S. listed manufacturing and services firms. The regression results show that there is no significant effect of board gender diversity on corporate risk-taking. The moderating effect of director’s age on the relationship between board gender diversity and corporate risk-taking is negative and significant, therefore this hypothesis is accepted. Because this moderating effect had not yet been investigated in the corporate field, my results make it an interesting topic for future research.
Keywords: Board gender diversity, corporate risk-taking, director’s age
Contents
1 Introduction ... 5
1.1 Background ... 5
1.2 Research question ... 6
1.3 Contribution ... 6
2 Literature review and hypothesis development ... 9
2.1 Agency theory ... 9
2.2 Resource dependence theory ... 9
2.3 Social role theory ... 10
2.4 Risk-sensitivity theory and young male syndrome ... 11
2.5 Gender diversity ... 11
2.6 Age diversity ... 12
3 Research method ... 14
3.1 Sample ... 14
3.2 Corporate risk-taking... 15
3.3 Independent variables ... 16
3.4 Control variables ... 16
3.5 Empirical design ... 18
4 Results ... 20
4.1 Descriptive statistics ... 20
4.2 Correlation matrix ... 22
4.3 Regression results ... 22
4.4 Summary of findings ... 27
4.5 Robustness tests... 29
5 Conclusion and discussion ... 33
References ... 36
1 Introduction
1.1 Background
In the past few years there have been many developments about mandating gender quotas at corporate boards. The politics is involved in this and some countries already have a compulsory female quota in the board of directors (Bertrand, Black, Jensen, & Lleras-Muney, 2019). The question is what kind of effects this has on risk-taking behavior within organizations. More gender diversity in the board of directors leads to less firm risk, because it decreases the volatility of the stock market return (Lenard, Yu, York, & Wu, 2014). Based on this, companies may consider hiring more women for the board of directors to reduce the company’s risk-taking behavior. Furthermore, companies may consider hiring older people to reduce corporate risk- taking, because a higher average age of the board of directors leads to less corporate risk-taking (Bernile, Bhagwat, & Yonker, 2018; Harjoto, Maretno & Laksmana, 2018). Prior literature and several theories suggest that age influences risk-taking behavior for men differently than for women. However, the existing literature on this topic is about individuals’ risk-taking behavior in daily life. Nevertheless, people behave differently in organizational and group settings (March & Shapira, 1987; Perryman, Fernando, & Tripathy, 2016). Therefore it is still unclear whether the theories hold up in the corporate field.
The board of directors of a company is responsible for managing the organization and serving and protecting shareholders’ interest (Zahra & Pearce, 1989). The board consists of several persons and they have multiple roles. Literature describes a control role, service role, strategy role and access to resources role (Nicholson & Newton, 2010; Zahra & Pearce, 1989;
Johnson, Daily, & Ellstrand, 1996). The control role includes managing the CEO and other managers and executives, to ensure that they are acting in the best interest of shareholders. The strategy role in conjunction with the service role entails advising the executives about management problems and defining the strategy of the organization (Johnson et al., 1996). The access to resources role is about providing resources to the organization with respect to improving multiple sections within a company (Hung, 1998). The control role is central to this research, because managers make risky or less risky decisions that should be monitored by the directors. In this way, the directors can influence the actions of the managers, which means that the board of directors has an impact on the risk-taking behavior of the firm.
Besides gender diversity, another factor that can influence the corporate risk-taking is the age of the directors. Based on research from Barker & Mueller (2002), age has a negative
impact on risk-taking behavior (Barker & Mueller, 2002). This means that older people are more likely to be prudent and risk averse. However, the effect of age may vary by gender. This means that the effect of board gender diversity on corporate risk-taking can be influenced by the board age diversity. Byrnes, Miller & Schafer (1999) argue that the effect of gender on risk- taking behavior is negatively moderated by age.
1.2 Research question
The research question of this thesis is: ‘Does board gender diversity have a negative impact on corporate risk-taking and is this negatively moderated by director’s age?’. The study initially focusses on the number of female directors in the board and the impact on several financial figures which reflect risk-taking behavior. Social role theory predicts that women are more risk averse, which implies a negative relationship between women representation on the board and corporate risk-taking. However, there is no overall consensus in the literature, as some studies contradict the theory and find no relationship or a positive relationship. Therefore it is of added value to investigate this topic further. The terms corporate risk-taking and risk-taking behavior are used interchangeably, but it covers the same meaning in this study. The impact of director’s age are tested, this includes the effect of director’s age on corporate risk-taking and the moderating effect of director’s age on the relation between board gender diversity and corporate risk-taking. Most of the literature finds a negative effect of board gender diversity on corporate risk-taking (Khaw, Liao, Tripe, & Wongchoti, 2016; Jeong & Harrison, 2017; Loukil
& Yousfi, 2016; Abou-El-Sood, 2019; Lenard et al., 2014; Faccio, Marchica, & Mura, 2016).
Also the moderating effect of age is predicted to be negative based on other studies (Rolison, Hanoch, Wood, & Liu, 2014; Byrnes, Miller, & Schafer, 1999; Gardner & Steinberg, 2005).
1.3 Contribution
The topic of board gender diversity and the effect on corporate risk-taking has been investigated less than the effect of board diversity in its entirety. However, the effect of board gender diversity on corporate risk-taking has been investigated by multiple researchers. There is no overall consensus about what the effect looks like, but most of the researchers have found evidence for a negative relation between board gender diversity and corporate risk-taking, as mentioned in the previous paragraph.
The moderating effect of director’s age on the relationship between board gender diversity and corporate risk-taking has not been investigated yet. However, according to the risk-
sensitivity theory and the young male syndrome I predict that young males take more risk than young females, but the difference in risk-taking between older males and older females is smaller than the difference between young males and young females. The risk-sensitivity theory explains that people adapt their risk preference or risk aversion to their personal needs (Mishra, 2014). The young male syndrome substantiates that young males take risks because of competition with other men (Fischer & Hills, 2012). Combining the risk-sensitivity theory and the young male syndrome leads to the statement that the needs of young males are more about increasing and maintaining their social status and competing with other males (Wilson
& Daly, 1985). Females are less sensitive to social status disparities compared to young males, resulting in less risk-taking behavior (Mishra, 2014). When the young males become older, the competitive disadvantage decreases because the older males have gained skills and social status, ultimately resulting in less risk-taking behavior (Mishra, 2014). In conclusion, the risk- taking behavior of males decreases over time, but the risk-taking behavior of females remains the same regardless of age. Thus, people’s age influences how males and females take risks such that young males take more risk than females and old males, but young females take about the same amount of risk as old females and old males. The gender differences at young ages are greater than the gender differences at older ages, so the prediction is that age negatively moderates the effect of gender on risk-taking behavior.
This moderating effect has been examined in risk-taking behavior about topics in the ethical, financial, health, recreational and social domain (Rolison et al., 2014; Byrnes et al., 1999; Gardner & Steinberg, 2005). These studies prove a negative and significant moderating effect of age on the relationship between gender and risk-taking behavior. However, by means of these studies the effect of director’s age on the relationship between board gender diversity and corporate risk-taking is unclear. First, the risk tendency in organizational settings is different from the classical or individual perceptions of risk (March & Shapira, 1987). The classical perception of risk entails that people evaluate risks based on variance, probabilities and values of possible outcomes (March & Shapira, 1987). Executives and managers are ignoring probability estimates and they give more priority to meeting targets instead of assessing risks (March & Shapira, 1987). Secondly, individuals make their own choices, but in a group the risk attitude is dependent on the influence and the power of a group leader (Perryman et al., 2016). Based on these distinctions between individual risk-taking behavior and corporate risk-taking, it is not possible to predict the moderating effect of director’s age with one hundred percent certainty. From an academic perspective, based on the risk-sensitivity
theory and the young male syndrome I predict that director’s age has a negative moderating effect on the relationship between board gender diversity and corporate risk-taking.
Analyzing my regression results, I find no significant effect of board gender diversity on corporate risk-taking. This is inconsistent with my prediction, but in line with the study of Loukil & Yousfi (2013). However, I find a moderating effect of the average age of directors on the relationship between board gender diversity and corporate risk-taking. The effect of board gender diversity on leverage and R&D intensity is weakened by average age of directors, which is in line with the prediction. This provides new insights into the literature on diversity and corporate risk-taking. Companies can take this information into account when composing their board of directors. The study also contributes to shareholders and investors being able to analyze the board’s diversity and estimate whether or not the company will be a risky investment.
Lastly, this study is designed as follows. Section 2 includes the literature review and hypothesis development. Section 3 is about the research method, which includes the sample description and an overview of the variables. Then the results are described in section 4. Section 5 entails the conclusion and discussion of the results.
2 Literature review and hypothesis development
2.1 Agency theory
The agency theory addresses an agency relationship when someone acts on behalf of another (Shapiro, 2005). In other words, a principal delegates work to an agent who carries out the work. In a firm, the executives/managers are the agent and the shareholders are the principal.
The principal cannot always completely monitor the performance of the agent, which leads to asymmetric information. Also the incentives of the principal and the incentives of the agent can differ from each other. The directors on the board are the representatives of the shareholders. They should control and monitor the actions of the managers. The shareholders want high performance and high returns and the managers want high compensation and personal welfare. This causes a discrepancy between the interests of the shareholders and the interests of the managers. In addition, the risk attitude of an agent can vary from the risk attitude of the principal. This risk attitude depends on the capability to diversify the employment (Eisenhardt, 1989). The divergent risk attitude and interests of the managers cause an agency problem. The role of the board of directors is to mitigate the agency problem, resulting in a better alignment of managers’ risk attitudes and interests with those of the shareholders.
A board consisting of people with different characteristics and different backgrounds can be seen as a creative group (Loukil & Yousfi, 2016). The board members supply knowledge, expertise and experience to make the right decisions. In a diverse board this supply will be of higher quality, which increases the effectiveness of the board. Because of this, the board of directors are more successful in controlling and monitoring the performance of the managers.
As a result, the managers are not able to engage in excessive risk-taking or excessive risk avoidance, making the returns more stable. Thus, based on agency theory, I conclude that a more diverse board of directors is able to effectively control the actions of managers, meaning that they influence the risk-taking behavior of the firm.
2.2 Resource dependence theory
The resource dependence theory elaborates on the fact that organizations act in such a way that it is relevant to their independence on several resources (Bryant & Davis, 2012). The agency theory explains the monitoring role of the board, but the resource dependence theory is about directors providing resources to the firm (Hillman & Dalziel, 2003). The four most important resources that directors provide to the firm are: advice and counsel, legitimacy, information
about the firm and environmental factors, and preferential access to resources outside the firm (Pfeffer & Salancik, 2003; Hillman & Dalziel, 2003). A larger board can be seen as a board with a lot of resources, resulting in a positive relationship between board size and firm performance (Dalton, Daily, Johnson, & Ellstrand, 1999). However, more directors is not always the best option, because few directors with access to many resources are more valuable than many directors with access to few resources (Boyd, 1990).
The resource dependence theory also describes that companies want to decrease their dependency on resources and maintain their independency. Compared to male directors, female directors are perceived more as independent directors (Adams & Ferreira, 2009). That is why high independency can be achieved by creating a diverse board of directors. This results in organizations being able to deal better with uncertainty. Different types of directors can supply different kinds of information and resources (Hillman, Cannella, & Paetzold, 2000). This means that heterogeneous groups, like a diverse board with males and females, provide multiple solutions from different perspectives (Hillman, Shropshire, & Cannella Jr, 2007). This should improve the directors’ monitoring quality, by which the managers do not engage in excessive risk-taking. Overall, the resource dependence theory implies that board gender diversity ultimately has a negative impact on risk-taking behavior.
2.3 Social role theory
The social role theory is about the differences in behavior between males and females (Eagly, Wood, & Diekman, 2000). People have an idea about what characteristics are typical for men or women. These are the gender stereotypes. Dependent on their cultural and environmental circumstances, men and women behave according to the gender roles (Eagly et al., 2000). The gender stereotypes include that men are more competitive than women. This is caused by the picture that has been outlined by people that women are more communal and men are more agentic (Eagly et al., 2000). Communal includes nurturant and safe behavior and agentic includes assertive and competitive behavior. In addition, men produce more testosterone and this results in more risky behavior (Booth, Granger, Mazur, & Kivlighan, 2006). On the other hand, females are associated with less risk-taking behavior. Thus, this theory predicts that more board gender diversity leads to less corporate risk-taking.
2.4 Risk-sensitivity theory and young male syndrome
The risk-sensitivity theory covers the concept of adapting risk-taking behavior based on personal needs (Mishra, 2014). In situations of high need, people will take more risk. For example, a person with a low social status has a high need to improve his/her status and therefore will take more risk. Furthermore, this theory implies that people do not want to reach the highest attainable goal, but rather their own desired goal (Mishra, 2014). The risk- sensitivity theory combined with the young male syndrome entails that young men are more likely to find themselves in situations of high need (Wilson & Daly, 1985). They experience a competitive disadvantage relative to older men in terms of social status or accumulated resources (Mishra, Gregson, & Lalumiere, 2012). Therefore their desired state is considerably higher than their current state. Females are less involved in comparing themselves against other females and they experience less competition (Fischer & Hills, 2012). In conclusion, this theory predicts that young males take more risks than females and older males.
2.5 Gender diversity
Gender is a factor used in many diversity studies. Also the association between gender and risk-taking behavior has been examined several times. The percentage of females present in the board has a significant negative impact on return volatility (Bernile et al., 2018). Higher diversity in terms of gender leads to a lower volatility of stock market return, which means that board gender diversity has a negative effect on corporate risk-taking (Lenard et al., 2014).
Another study also has found evidence for a negative relationship between the number of female directors and corporate risk (Perryman et al., 2016). However, there are also studies that find no significant relationship or a positive relationship between board gender diversity and corporate risk-taking. Sila, Gonzalez & Hagendorff (2016) found that there is no relationship between board gender diversity and equity risk (Sila, Gonzalez, & Hagendorff, 2016). In Tunisian listed companies, there is no relationship between board gender diversity and strategic and financial risk-taking (Loukil & Yousfi, 2016). Adams & Funk (2012) state that female directors do not have a negative impact on corporate risk-taking (Adams & Funk, 2012).
Furthermore, in US listed firms, board gender diversity has a positive impact on firms’ financial risk (Chen, L. H., Gramlich, & Houser, 2019). Based on all studies mentioned in this section, I conclude that there is no consensus about this topic in the literature. Therefore I test whether there is a positive, negative or no relationship between board gender diversity and corporate risk-taking in my sample. In prior literature, the largest proportion of articles found a negative
association between board gender diversity and corporate risk-taking. This is in line with the prediction of the social role theory, which advocates that men take more risks than women.
Based on the social role theory and previous findings (Bernile et al., 2018; Lenard et al., 2014;
Perryman et al., 2016), hypothesis 1 is as follows:
H1: Board gender diversity has a negative impact on corporate risk-taking.
-
2.6 Age diversity
Several studies have elaborated on the effect of director age on corporate risk-taking. Older people stick to traditional investment styles which are less risky (Farag & Mallin, 2018). This is the direct effect of age on risk-taking behavior, but there is no literature about the moderating effect of director’s age on the influence of board gender diversity on corporate risk-taking.
However, Byrnes, Miller & Schafer (1999) investigated the moderating effect of age on the relationship between gender and risk-taking behavior. Here, the risk-taking behavior is about people’s risk attitude towards activities in daily life, like smoking, drinking, driving and gambling. There is a significant effect of gender on risk-taking behavior, but the effect decreases over time (Byrnes et al., 1999). This means that the difference of risk-taking behavior between young males and females is higher than between older males and females (Byrnes et al., 1999). This has been addressed in the thesis as the moderating effect of age. There is no evidence yet for this moderating effect in terms of corporate risk-taking. Rolison, Hanoch, Wood, & Liu (2014) stated that risk-taking behavior decreases over time, but this decline is higher for men than for women. This means that between young people, the effect of gender diversity on risk-taking behavior is higher than between older people (Rolison et al., 2014).
Their study is about risk-taking behavior of individuals in the ethical, financial, health, recreational and social domain. Based on these studies, the moderating effect of age on the relationship between board gender diversity and corporate risk-taking cannot be predicted with one hundred percent certainty. This is because of the difference between individual risk-taking behavior and corporate risk-taking. The studies of Byrnes et al. (1999) and Rolison et al. (2014) are about how people individually take risks, but this thesis is about risk-taking in groups and
Corporate risk-taking Board gender
diversity
Figure 1: Hypothesis 1
organizational settings. The risk tendency in organizational settings is different from the classical or individual perceptions of risk (March & Shapira, 1987). In general, managers ignore probability estimates and they give more priority to meeting targets instead of assessing risks (March & Shapira, 1987). Additionally, in a group the risk attitude is dependent on the influence and the power of a group leader (Perryman et al., 2016). This is why the moderating effect of age cannot be predicted with one hundred percent certainty. However, combining the results of the two studies (Byrnes et al., 1999; Rolison et al., 2014) with the social role theory, risk-sensitivity theory and young male syndrome, I constructed the following hypothesis about the moderating effect of director’s age:
H2: The influence of board gender diversity on corporate risk-taking is negatively moderated by board of director’s age.
-
Figure 1: Hypothesis 1
Corporate risk-taking Board gender
diversity
Average director’s age
Figure 2: Hypothesis 2
3 Research method
3.1 Sample
The research method for this study is archival research. This means that secondary data has been used. Data is collected from databases of Wharton Research Data Services (WRDS). The research is focused on companies listed in North America. The databases of Compustat and Institutional Shareholder Services (ISS) has been used and combined. In the first place, I consult Compustat to collect data about financial figures, to calculate the corporate risk-taking.
The ISS database is used for gathering data about board gender and age diversity. Also the control variables are collected from these databases. Board size and data about independent directors is available in ISS. Figures about firm size are gathered from Compustat.
The sample for this research consists of data from 2010-2019. Data for 2009 and earlier is not included because of the financial crisis in the United States from 2007 to 2009. This recession caused an irregular presentation of the financial figures that reflect corporate risk- taking (Shahzad, Lu, & Fareed, 2019). Observations from the years 2013-2019 have been used for data about director’s age, gender, independency and financial figures. One exception is the delta of the return on assets (ROA). For this variable, data about previous years is necessary to calculate the variability. That is why also financial figures of 2010-2012 have been used. Data about company information, director’s information, earnings, R&D expenses, total debt and total assets are queried. Then the two different databases are merged based on the CUSIP code and the year. The CUSIP code identifies a company and every company contains observations by year. The CUSIP code is included in the Compustat database and in the ISS database.
Following prior literature, the unit of analysis is company-year (Ahern & Dittmar, 2012; Bao, Fainshmidt, Nair, & Vracheva, 2014; Bernile et al., 2018). Every observation of a variable reflects information about a company in a specific year. According to previous studies, companies in the financial sector are excluded because these firms follow specific regulations and this has an impact on their investment/risk approaches (Bhat, Chen, Jebran, & Memon, 2019; Harjoto, Maretno A., Laksmana, & Yang, 2018; Farag & Mallin, 2018; Loukil & Yousfi, 2016). Lastly, firms with missing values about relevant variables for this research have been eliminated.
Table 1 shows the adjustments in the number of observations:
Table 1: Adjustments in observations
Selection criteria Observations
Compustat data 20,298
Companies with missing value in Compustat 14,712 -
Data without missing values 5,586
Financial services companies 102 - Data excluding financial services companies 5,484 No match between ISS and Compustat 4,538 -
Merged data 946
Outliers deleted 4 -
Total observations 942
Before the data from ISS was merged with data from Compustat, I deleted the companies with missing values for one or more variables. After having all the variables in one dataset, I searched for outliers with the help of box plots, histograms and information about the first percentiles and last percentiles. The variable RDINT contained a few major outliers which I have deleted. RDOA included some outliers, but they did not consist of a big deviation from the minimum and maximum. Therefore, I winsorized these outliers. This means that the values of the outliers were replaced by the values of the first percentile and last percentile. As a result, the impact of the outliers is reduced, which makes the data more suitable for statistical testing.
3.2 Corporate risk-taking
Corporate risk-taking is a well-known concept in the existing literature. In this study, it has been measured in three different ways. Based on other studies, the first measure is the delta of return on assets (ROA) (Khaw et al., 2016; Saeed, Belghitar, & Yousaf, 2016). ROA is calculated through earnings before interest and taxes (EBIT) divided by total assets. Then the delta has been calculated by dividing the ROA of a certain year by the ROA of 3 years before.
More risk-taking leads to a higher variability of returns. Therefore, I associate a high variability with more corporate risk-taking and a low variability with less corporate risk-taking.
Secondly, based on other studies, leverage is a measure which reflects corporate risk-taking (Ahern & Dittmar, 2012; De Cabo, Gimeno, & Nieto, 2012; Faccio et al., 2016; Huang, J. &
Kisgen, 2013). This includes the ratio of total debt divided by total assets. A higher leverage reflects more corporate risk-taking. Lastly, research and development (R&D) intensity is a measure for corporate risk-taking (Chan, Lakonishok, & Sougiannis, 2001; Chen, S., Ni, &
Tong, 2016). This has been calculated by dividing the R&D expenditures by total sales. A higher R&D intensity represents more corporate risk-taking.
3.3 Independent variables
The independent variables in this study are board gender diversity and director’s age diversity.
In the ISS database data is available about the gender of the directors. This is recorded as a yes or no on the question whether the director is a female. The age of the director is recorded as a number, e.g. 51. These are single values about a person, but the data in this research is about risk-taking at company-level. To be able to merge the dataset of ISS with the dataset of Compustat, I translated data about gender and age to company-level. That is why, in line with prior literature, the number of female directors in the board of a company has been calculated as a ratio (Chen et al., 2016; De Cabo et al., 2012; Dong, Girardone, & Kuo, 2017; Dowling &
Aribi, 2013). For example, a board with 1 female and 3 males gets a value of 0.25. In accordance with other studies, age is reported as the average age of the directors on the board (Grove, Patelli, Victoravich, & Xu, 2011; Harjoto & Laksmana, 2018; Bernile et al., 2018;
Bonn, Yoshikawa, & Phan, 2004). For example, a board with three people of 41, 53 and 68 years old gets a value of 54.
The moderating effect of age on the gender-risk taking relationship has been tested by means of an interaction term. This includes the multiplication of the percentage of females in the board of directors with the average age of the board of directors. The coefficient of the interaction indicates whether the average age of the board of directors has an impact on the effect of board gender diversity on corporate risk-taking and whether this impact includes a positive or negative direction.
3.4 Control variables
Corporate risk-taking can be affected by other factors than board gender diversity and director’s age. Therefore these factors are included in the regression. According to previous studies the
following variables are controlled for: board size, independent directors and firm size (Harjoto et al., 2018; Loukil & Yousfi, 2016; Bhat et al., 2019).
Companies with larger boards perform less volatile than organizations with smaller boards (Cheng, 2008). Thus, the prediction is that larger boards have a negative impact on corporate risk-taking. The number of directors is an absolute value. This implies that the data about this variable is not normally distributed. To adjust this variable to normally distributed values, it has been measured as the natural logarithm of the number of directors on the board. This is calculated using the natural logarithm function (ln) (Bruna, Dang, Scotto, & Ammari, 2019).
Independent directors act in the best interest of the shareholders, which results in less risk- taking behavior (Loukil & Yousfi, 2016). This variable is available in the database of ISS, in which an independent director has a value of I. Based on prior literature, the independency of a board has been calculated as a ratio (Bhagat & Black, 2001). A board with 3 independent directors and 5 employee/inside directors gets a value of 0.375. A higher ratio indicates lower corporate risk-taking.
Firm size can have a negative impact on corporate risk-taking, because larger companies might be more stable and their performance is less volatile (John, Litov, & Yeung, 2008). As a result, they are more reserved and cautious in risk-taking. Total assets has been used to measure firm size. This contains absolute values which can vary a lot. As a result, the values are not normally distributed. However, firm size is measured, in line with prior literature, as the natural logarithm of a company’s total assets (Chong, Ong, & Tan, 2018).
Table 2: Description of variables
Variable Measure
LEV DROA
Leverage, total debt / total assets
Delta of ROA, EBIT / total assets of year n divided by EBIT / total assets of year n-3
RDINT R&D intensity, R&D expenses / total sales GENDER % female directors over total directors AVAGE Average board of directors’ age GENDER*AVAGE Interaction term, GENDER * AVAGE
INDEP % independent directors over total directors LNSIZE Natural logarithm of total assets
LNBSIZE Natural logarithm of board size
3.5 Empirical design
The hypotheses of this study are about the direct effect of board gender diversity on corporate risk-taking and about the moderating effect of directors’ age on the relationship between board gender diversity and corporate risk-taking. The direct effect of board gender diversity (hypothesis 1) is tested in model 1, using a multiple regression with all independent variables and control variables included. In other words, model 1 includes the effect of all independent and control variables on LEV, DROA and RDINT. The regression equation for the three different measures of corporate risk-taking is based on the study of Gulamhussen & Santa (2015) and it is adapted to the variables in this model:
LEV = α + β1GENDER + β2AVAGE + β3INDEP+ β4LNSIZE+ β5LNBSIZE + ε
DROA = α + β1GENDER + β2AVAGE + β3INDEP+ β4LNSIZE+ β5LNBSIZE + ε
RDINT = α + β1GENDER + β2AVAGE + β3INDEP+ β4LNSIZE+ β5LNBSIZE + ε
The moderating effect of director’s age (hypothesis 2) is tested with multiple linear regression analysis, in which the independent variables, control variables and the interaction term are included (Fairchild & MacKinnon, 2009). Thus, model 2 adds the interaction term to model 1.
In model 1 the direct effects of the independent variables are tested, but in model 2 only the interaction effect is evaluated. Based on prior literature, the regression equation includes the betas from model 1 plus the interaction term (Berger, Kick, & Schaeck, 2014; Gulamhussen &
Santa, 2015).
LEV = α + β1GENDER + β2AVAGE+ β3GENDER*AVAGE+ β4INDEP+ β5LNSIZE+ β6LNBSIZE + ε
DROA = α + β1GENDER + β2AVAGE+ β3GENDER*AVAGE+ β4INDEP+ β5LNSIZE + β6LNBSIZE + ε
RDINT = α + β1GENDER + β2AVAGE+ β3GENDER*AVAGE+ β4INDEP+ β5LNSIZE + β6LNBSIZE + ε
The outcomes of the multiple linear regression analysis are compared to several robustness tests. To test for robustness, I conduct a robust regression and a multiple linear regression where the standard errors are clustered at firm level. The latter I refer to as the clustered regression.
This method has been used by many other papers, because of the use of panel data sets (Petersen, 2009). Panel data means that the data consists of observations from several firms and multiple years. In cases like this, there can be a correlation between the residuals of various observations. This biases the results of the regression analysis. The multiple linear regression with standard errors clustered at firm level ensures that the significance of the effect is based on unbiased standard errors (Petersen, 2009). The second robustness test is the robust regression. This regression I execute with the rreg command in Stata. It deals with outliers and variables that are not normally distributed (Brinda, Rajkumar, Enemark, Attermann, & Jacob, 2014; Rajkumar, Brinda, Duba, Thangadurai, & Jacob, 2013).
4 Results
This section includes the representation of the data that has been used for this research. Also multiple statistical tests have been executed to test the hypotheses. First, I present the descriptive statistics and test for multicollinearity. After that, the correlation matrix and regression analysis have been reviewed. The results are evaluated to conclude whether there is a relationship between board gender diversity and corporate risk-taking and how this relationship is influenced by directors’ age.
4.1 Descriptive statistics
Table 3 shows the distribution of the sectors included in the sample. The total number of companies is 230. Most of the listed U.S. companies in the sample are in the manufacturing industry. Another substantial part of the sample are the listed U.S. firms in the services industry.
The other sectors comprise a minor fraction of the sample. Therefore my results only apply to listed U.S. firms in the manufacturing and services sectors.
Table 3: Sample per sector
Number of companies 230
Mining 1.30%
Construction 0.43%
Manufacturing 80.00%
Transportation & Public Utilities 0.88%
Wholesale Trade 0.43%
Retail Trade 0.88%
Services 15.65%
Public Administration 0.43%
Total 100%
Table 4 shows the descriptive statistics of the independent and dependent variables. All the variables include a total of 942 observations. The information about gender reveals that, on average, 17.6% of the directors on the board of listed U.S. firms are female directors. The standard deviation is 0.107 and the table shows that there are no companies with more than 50% female directors on the board. The average age of the board of directors is 62.37. The standard deviation is 3.635 and the average board of directors’ age ranges from 48.875 to 78.667. The statistics on independent directors show that listed U.S. manufacturing and services firms on average have a large proportion of independent directors, 82.1%. The standard deviation is relatively low, 0.095. All boards in this sample exist of at least 37.5%
independent directors and in one or more cases all the directors are independent. The natural logarithm of total assets includes a mean of 8.394 and a standard deviation of 1.792. The range of this variable is between 4.165 and 13.395. The natural logarithm of board size has a mean of 2.205. The standard deviation is 0.229, the minimum is 1.386 and the maximum is 2.996.
The mean leverage is 0.255. This means that on average total debt is 25.5% of total assets. The associated standard deviation is 0.153. In one or more cases, total debt is zero and the maximum shows that there is at least one observation in which total debt is almost equal to total assets, i.e. 97.7%. The delta of return on assets has a mean of -0.029 and a standard deviation of 1.442.
Observations can vary considerably from one another and this is reflected in the minimum of - 5.511 and the maximum of 8.654. R&D intensity is a ratio and in this sample the observations are within a range of 0 to 0.487. This means that the R&D expenses are always less than 50%
of total sales. The mean is 0.075 and the standard deviation is 0.085.
Table 4: Descriptive statistics
Variable Obs Mean Std. Dev. Min Max
GENDER 942 .176 .107 0 .5
AVAGE 942 62.37 3.635 48.875 78.667
INDEP 942 .821 .095 .375 1
LNSIZE 942 8.394 1.792 4.165 13.395
LNBSIZE 942 2.205 .229 1.386 2.996
LEV 942 .255 .153 0 .977
DROA 942 -.029 1.442 -5.511 8.654
RDINT 942 .075 .085 0 .487
One assumption for conducting multiple regression analysis is that independent variables are not highly correlated with each other (Graham, 2003). A high correlation is a correlation above 0.8. This assumption is called multicollinearity. If there is multicollinearity, it is difficult to analyze the direct effect of one independent variable on the dependent variable. This is because in that case the effect of the independent variable also depends on another independent variable.
In table 5, I show the results from the multicollinearity test. The variable inflation factor (VIF) measures multicollinearity, in which a value of more than 5 indicates collinearity (Paul, 2006).
The VIFs in table 5 are close to 1, which means that there is no multicollinearity. Also the tolerance values reflect no multicollinearity, because a tolerance value greater than 0.4 means that there is no multicollinearity (Adeboye, Fagoyinbo, & Olatayo, 2014). Based on the values in table 5, I conclude that, in terms of multicollinearity, the dataset is suitable for multiple regression analysis.
Table 5: Multicollinearity
Variable VIF Tolerance
GENDER 1.27 0.785
AVAGE 1.07 0.937
INDEP 1.10 0.910
LNSIZE 1.03 0.975
LNBSIZE 1.23 0.811
4.2 Correlation matrix
The first test to analyze whether the independent variables influence the dependent variables is the correlation matrix. In table 6 I show whether there is a positive or negative relationship between two variables and to what extent this relationship is significant. First, the fact that there is no multicollinearity between variables is confirmed, because none of the correlations is above 0.8 (Chong et al., 2018). However, there are multiple significant correlations between independent and dependent variables. Reviewing the influence on the dependent variables, GENDER is positively and significantly correlated with LEV at the 5% significance level.
AVAGE is significantly correlated with all dependent variables. It has a positive correlation with DROA and a negative correlation with LEV and RDINT. INDEP is also significantly correlated with all dependent variables, in which it has a positive correlation with LEV and DROA. LNSIZE only has a significant and positive correlation with LEV. LNBSIZE is not significantly correlated with LEV, DROA or RDINT.
Table: 6 Pairwise correlations
Variables (1) (2) (3) (4) (5) (6) (7)
(1) GENDER 1.000
(2) AVAGE -0.200*** 1.000
(3) INDEP 0.237*** 0.017 1.000
(4) LNSIZE 0.066** -0.142*** 0.021 1.000
(5) LNBSIZE 0.398*** -0.035 0.253*** 0.076** 1.000
(6) LEV 0.072** -0.100*** 0.085*** 0.173*** 0.000 1.000
(7) DROA -0.021 0.062* 0.073** 0.022 0.002 0.004 1.000
(8) RDINT 0.016 -0.094*** -0.060* -0.048 0.032 -0.086*** -0.057*
Note: ***, **, * denote significance level of 1%, 5%, 10%, respectively.
4.3 Regression results
To test whether the independent variables have a positive or negative effect on the dependent variables, I executed multiple linear regressions. As described in section 3.5, Model 1 includes
the effects of the independent variables and model 2 includes the effect of the interaction term of GENDER*AVAGE. The regression analysis are presented in three different tables, in which each table represents the regression on one of the three dependent variables. Table 7 reports the regression results with leverage as the dependent variable.
Table 7: Dependent variable = LEV
(1) (2)
VARIABLES Model 1 Model 2
GENDER 0.0713 2.214***
(0.0516) (0.729)
AVAGE -0.00296** 0.00204
(0.00138) (0.00219)
GENDER*AVAGE -0.0347***
(0.0118)
INDEP 0.136** 0.155***
(0.0536) (0.0537)
LNSIZE 0.0138*** 0.0128***
(0.00275) (0.00276)
LNBSIZE -0.0378 -0.0317
(0.0236) (0.0236)
Constant 0.283*** -0.0488
(0.106) (0.155)
Observations 942 942
R-squared
Adjusted R-squared F-value
0.046 0.041 9.04***
0.055 0.049 9.04***
Standard errors in parentheses
Note: ***, **, * denote significance level of 1%, 5%, 10%, respectively.
Table 7 shows the results of a multiple linear regression with dependent variable LEV and the independent variables of model 1 and model 2. Model 1 includes the effect of board gender diversity and director’s age as well as the control variables. The R-squared for this model is 0.046 and the adjusted R-squared is 0.041. This means that 4.6% of the variance of the dependent variable can be explained by the independent variables of model 1. The adjusted R- squared takes into account the number of predictors in the model. As a result, this value is a bit lower than the R-squared. The F-value and its significance level reflect the overall significance of the model. The F-value in table 7 is 9.04 and significant at the 1% level. This means that at least one of the independent variables has an effect on the dependent variable. Model 2 represents a R-squared of 0.055 and an adjusted R-squared of 0.049. The F-value for model 2
is also significant. Based on that, I conclude that the regression equation from both models has some validity in fitting the data.
Prior literature suggests that board gender diversity has a negative impact on corporate risk-taking (Bernile et al., 2018; Lenard et al., 2014; Perryman et al., 2016). Coherent with the social role theory, hypothesis 1 predicts that board gender diversity negatively influences corporate risk-taking. However, the results in table 7 show that GENDER does not have an impact on LEV in this sample. In accordance with prior literature (Barker & Mueller, 2002;
Farag & Mallin, 2018), AVAGE has a negative and significant effect on LEV. This means that an older average age on the board of directors leads to less risk-taking in terms of leverage.
Table 7 shows that control variable INDEP has a positive significant effect on LEV, which is in line with prior literature (Khaw et al., 2016; Cheng, 2008). The control variable LNSIZE has a positive significant effect on LEV, which is inconsistent with existing literature (John et al., 2008). LNBSIZE does not have a significant impact. This is in accordance with the study of Bruna et al. (2019), but inconsistent with the study of Cheng (2008). Based on table 7, I reject hypothesis 1, because the regression results show no significant effect of board gender diversity on corporate risk-taking.
Based on prior literature and the risk-sensitivity theory and young male syndrome, hypothesis 2 predicts that the effect of board gender diversity on corporate risk-taking is moderated by director’s age. This has been tested in model 2, using the interaction term GENDER*AVAGE. Table 7 shows that there is a negative and significant interaction effect.
This means that the effect of GENDER on LEV is dependent on the value of AVAGE. This is consistent with previous studies on the moderating effect of age in research areas other than the corporate field (Byrnes et al., 1999; Rolison et al., 2014; Gardner & Steinberg, 2005). The interaction coefficient is negative, which means that if AVAGE increases, the effect of GENDER on LEV decreases. Thus, based on this multiple linear regression I accept hypothesis 2. The effect of board gender diversity on corporate risk-taking decreases as the average age of the board of directors increases.
Table 8: Dependent variable = DROA
(1) (2)
VARIABLES Model 1 Model 2
GENDER -0.353 -3.222
(0.496) (7.043)
AVAGE 0.0238* 0.0171
(0.0133) (0.0211)
GENDER*AVAGE 0.0464
(0.114)
INDEP 1.202** 1.177**
(0.516) (0.519)
LNSIZE 0.0250 0.0264
(0.0265) (0.0267)
LNBSIZE -0.0530 -0.0613
(0.228) (0.229)
Constant -2.529** -2.085
(1.024) (1.494)
Observations 942 942
R-squared
Adjusted R-squared F-value
0.011 0.005 2.03*
0.011 0.005 1.72 Standard errors in parentheses
Note: ***, **, * denote significance level of 1%, 5%, 10%, respectively.
Table 8 presents the results of a multiple linear regression with DROA as the dependent variable. The R-squared for model 1 is 0.011, which means that 1.1% of the variance in DROA can be explained by the independent variables in model 1. This is a relatively low number, but the F-value is significant at the 10% level. This means that at least one of the independent variables has an effect on DROA. The R-squared for model 2 is equal to the value for model 1, but the F-value is not significant.
Table 8 shows that GENDER does not have an impact on DROA, which is inconsistent with the prediction of hypothesis 1. However, AVAGE does have an effect on DROA. This is equal to the results of the regression with LEV as dependent variable. There is 1 difference between the two results, because AVAGE has a negative effect on LEV, but it has a positive effect on DROA. The latter is not in line with results from prior literature. Regarding to the control variables, LNSIZE and LNBSIZE do not have an effect on DROA. INDEP does have a significant and positive effect on DROA, which is inconsistent with the study of Loukil &
Yousfi (2016). Model 2 represents the interaction effect. The interaction effect on DROA is not significant, which means that there is no interaction effect. In other words, the effect of
GENDER on DROA is not moderated by AVAGE. Based on these regression results, I reject hypothesis 1 and hypothesis 2.
Table 9: Dependent variable = RDINT
(1) (2)
VARIABLES Model 1 Model 2
GENDER -0.00309 0.905**
(0.0290) (0.410)
AVAGE -0.00235*** -0.000229
(0.000778) (0.00123)
GENDER*AVAGE -0.0147**
(0.00662)
INDEP -0.0613** -0.0533*
(0.0301) (0.0302)
LNSIZE -0.00303* -0.00345**
(0.00155) (0.00156)
LNBSIZE 0.0192 0.0218
(0.0133) (0.0133)
Constant 0.255*** 0.115
(0.0598) (0.0870)
Observations 942 942
R-squared
Adjusted R-squared F-value
0.018 0.013 3.49***
0.023 0.017 3.74***
Standard errors in parentheses
Note: ***, **, * denote significance level of 1%, 5%, 10%, respectively.
I executed another multiple linear regression with the third dependent variable, RDINT. Table 9 displays a R-squared of 0.018 and an adjusted R-squared of 0.013 for model 1. For model 2, the R-squared is 0.023 and the adjusted R-squared is 0.017. The F-value of both models is significant, which means that at least one of the independent variables has a significant impact on RDINT.
In accordance with the results of the first regressions in table 7 and 8, table 9 reports no effect of GENDER on RDINT. This is inconsistent with previous studies (Abou-El-Sood, 2019; Jeong & Harrison, 2017; Khaw et al., 2016; Lenard et al., 2014). Based on the regression results of all dependent variables, I reject hypothesis 1. Table 9 presents a significant effect of AVAGE on RDINT. The effect of AVAGE on RDINT is negative and this is in line with the expectation that board of directors’ age has a negative impact on corporate risk-taking. The control variables INDEP and LNSIZE significantly have a negative effect on RDINT, which is
consistent with prior literature (John et al., 2008; Loukil & Yousfi, 2016). The control variable LNBSIZE does not have an effect on RDINT. This is not in line with the study of Cheng (2008).
Table 9 reports for model 2 a significant interaction effect. The effect is negative, which means that the higher the AVAGE, the lower the effect of GENDER on RDINT. This is consistent with the results in table 7, where LEV was the dependent variable. Based on this, I accept hypothesis 2, which means that the effect of board gender diversity on corporate risk-taking is negatively moderated by the average age of the board of directors.
4.4 Summary of findings
In table 10 I have merged the results of model 1. In this way, it gives an overview of which independent variables have an impact on which dependent variable, which makes it possible to conclude whether the variables affect corporate risk-taking.
Table 10: Regression results for model 1
(1) (2) (3)
VARIABLES LEV DROA RDINT
GENDER 0.0713 -0.353 -0.00309
(0.0516) (0.496) (0.0290)
AVAGE -0.00296** 0.0238* -0.00235***
(0.00138) (0.0133) (0.000778)
INDEP 0.136** 1.202** -0.0613**
(0.0536) (0.516) (0.0301)
LNSIZE 0.0138*** 0.0250 -0.00303*
(0.00275) (0.0265) (0.00155)
LNBSIZE -0.0378 -0.0530 0.0192
(0.0236) (0.228) (0.0133)
Constant 0.283*** -2.529** 0.255***
(0.106) (1.024) (0.0598)
Observations 942 942 942
R-squared 0.046 0.011 0.018
Standard errors in parentheses
Note: ***, **, * denote significance level of 1%, 5%, 10%, respectively.
Table 10 shows no significant effect of GENDER on all the dependent variables. Therefore I conclude that, in this sample, board gender diversity does not have an impact on corporate risk- taking. Thus, I reject hypothesis 1. This is a different outcome from the results of most other studies that address this topic. However, AVAGE has a significant impact on all the dependent variables. It has a negative impact on LEV and RDINT, but a positive impact on DROA.
Because the significance of the effects on LEV and RDINT is greater than the significance of
the effect on DROA, I conclude that overall, in this sample, AVAGE has a negative effect on corporate risk-taking. This is in line with prior literature (Farag & Mallin, 2018). INDEP has a significant positive impact on LEV and DROA, but a significant negative impact on RDINT. I conclude that INDEP has an effect on corporate risk-taking, but the direction of the effect depends on the risk-taking measure. The results of LNSIZE vary from a positive significant effect to a negative significant effect and no significant effect. Therefore it is not possible to draw one unambiguous conclusion. Table 10 shows that LNBSIZE has no effect on corporate risk-taking, which is inconsistent with prior literature (Nakano & Nguyen, 2012; Huang, Y. S.
& Wang, 2015).
Table 11: Regression results for model 2
(1) (2) (3)
VARIABLES LEV DROA RDINT
GENDER 2.214*** -3.222 0.905**
(0.729) (7.043) (0.410)
AVAGE 0.00204 0.0171 -0.000229
(0.00219) (0.0211) (0.00123) GENDER*AVAGE -0.0347*** 0.0464 -0.0147**
(0.0118) (0.114) (0.00662)
INDEP 0.155*** 1.177** -0.0533*
(0.0537) (0.519) (0.0302)
LNSIZE 0.0128*** 0.0264 -0.00345**
(0.00276) (0.0267) (0.00156)
LNBSIZE -0.0317 -0.0613 0.0218
(0.0236) (0.229) (0.0133)
Constant -0.0488 -2.085 0.115
(0.155) (1.494) (0.0870)
Observations 942 942 942
R-squared 0.055 0.011 0.023
Standard errors in parentheses
Note: ***, **, * denote significance level of 1%, 5%, 10%, respectively.
Table 11 shows the regression results for model 2. This models has only been used to evaluate the interaction effect of GENDER*AVAGE. The results of the multiple linear regression reveal that the effect of GENDER on LEV and RDINT depends on the value of AVAGE, because the interaction effect is significant. Based on the dependent variables LEV and RDINT, I conclude that the average age of the board of directors has a negative impact on the effect of board gender diversity on corporate risk-taking. The higher the AVAGE, the lower the effect of GENDER on LEV and RDINT. However, when executing the regression with dependent variable DROA,
there is no significant interaction effect. Therefore I conclude that the average age of the board of directors does not always lead to a lower effect of board gender diversity on corporate risk- taking. However, based on the significant interaction effects on LEV and RDINT, I accept hypothesis 2.
4.5 Robustness tests
To test whether the results of the multiple linear regression are robust, I conducted two different regressions. The first one is a multiple linear regression, with standard errors clustered at firm level. Below I refer to this as the clustered regression. The second one is the robust regression.
This method gives more weight to observations that are more well behaved than to observations which resemble outliers or high leverage data points.
Table 12: Clustered regression results for model 1
(1) (2) (3)
VARIABLES LEV DROA RDINT
GENDER 0.0713 -0.353 -0.00309
(0.0883) (0.595) (0.0571)
AVAGE -0.00296 0.0238** -0.00235*
(0.00221) (0.0118) (0.00141)
INDEP 0.136 1.202** -0.0613
(0.0845) (0.568) (0.0577)
LNSIZE 0.0138** 0.0250 -0.00303
(0.00565) (0.0287) (0.00311)
LNBSIZE -0.0378 -0.0530 0.0192
(0.0375) (0.264) (0.0287)
Constant 0.283 -2.529*** 0.255**
(0.172) (0.892) (0.106)
Observations 942 942 942
R-squared 0.046 0.011 0.018
Robust standard errors in parentheses
Note: ***, **, * denote significance level of 1%, 5%, 10%, respectively.
Table 12 shows the results from a multiple linear regression, with standard errors clustered at firm level. The output displays no direct effect of GENDER on all the dependent variables, which is in line with the results of the multiple linear regression in table 10. When clustering at firm level, the effect of AVAGE on DROA and RDINT remains significant, but the effect of AVAGE on LEV is not significant anymore. In table 10, the effect of INDEP was significant on all dependent variables. But in table 12, only the effect of INDEP on DROA is significant.
The only change in effects of LNSIZE on the dependent variables is that the effect on RDINT is not significant anymore in the clustered regression. The effects of LNBSIZE on all dependent variables remain insignificant, which is consistent with the regression results in table 10. Based on this clustered regression, I still reject hypothesis 1, because the regression results show no significant effect of GENDER on the dependent variables.
Table 13: Clustered regression results for model 2
(1) (2) (3)
VARIABLES LEV DROA RDINT
GENDER 2.214** -3.222 0.905
(1.021) (6.867) (0.741)
AVAGE 0.00204 0.0171 -0.000229
(0.00328) (0.0165) (0.00196) GENDER*AVAGE -0.0347** 0.0464 -0.0147
(0.0165) (0.112) (0.0117)
INDEP 0.155* 1.177** -0.0533
(0.0837) (0.571) (0.0589)
LNSIZE 0.0128** 0.0264 -0.00345
(0.00569) (0.0295) (0.00308)
LNBSIZE -0.0317 -0.0613 0.0218
(0.0368) (0.268) (0.0284)
Constant -0.0488 -2.085* 0.115
(0.226) (1.200) (0.141)
Observations 942 942 942
R-squared 0.055 0.011 0.023
Robust standard errors in parentheses
Note: ***, **, * denote significance level of 1%, 5%, 10%, respectively.
Table 13 presents the results of the multiple linear regression for model 2, with standard errors clustered at firm level. Comparing this to the output in table 11, the effect of the interaction term on LEV decreases from the 1% significance level to the 5% significance level. This still means that I accept hypothesis 2, based on this negative effect. The interaction effect of GENDER*AVAGE on DROA remains insignificant in table 13. However, table 11 presents a negative and significant interaction effect on RDINT, but the results in table 13 show that there is no significant interaction effect on RDINT. Based on these insignificant effects on DROA and RDINT, I reject hypothesis 2.
Table 14: Robust regression results for model 1
(1) (2) (3)
VARIABLES LEV DROA RDINT
GENDER 0.0730 -0.249 -0.0233
(0.0511) (0.161) (0.0203)
AVAGE -0.00377*** 0.00571 -0.00222***
(0.00137) (0.00434) (0.000546)
INDEP 0.140*** 0.0230 0.00611
(0.0531) (0.168) (0.0211)
LNSIZE 0.0163*** 0.0201** -0.00297***
(0.00273) (0.00863) (0.00109)
LNBSIZE -0.0347 0.117 -0.0131
(0.0234) (0.0740) (0.00932)
Constant 0.293*** -0.854** 0.240***
(0.105) (0.333) (0.0419)
Observations 942 942 942
R-squared 0.062 0.012 0.027
Standard errors in parentheses
Note: ***, **, * denote significance level of 1%, 5%, 10%, respectively.
Table 14 provides the results of a robust regression for model 1. This regression has been executed using the rreg command in Stata. In line with previous regression analysis, there is no effect of GENDER on all the dependent variables. This means that I reject hypothesis 1, which implies that board gender diversity does not affect corporate risk-taking. In contrast to the multiple linear regression and the clustered regression, the robust regression shows no effect of AVAGE on DROA. However, the robust regression reveals that AVAGE has a negative and significant impact on LEV and RDINT. This is in line with the multiple linear regression, but inconsistent with the clustered regression. Furthermore, the robust regression provides a significant and positive effect of INDEP and LNSIZE on LEV, which is consistent with the multiple linear regression results in section 4.3. In all the different regression analysis, LNBSIZE does not affect corporate risk-taking.
Table 15: Robust regression results for model 2
(1) (2) (3)
VARIABLES LEV DROA RDINT
GENDER 2.145*** -2.830 0.964***
(0.721) (2.284) (0.290)
AVAGE 0.00103 -0.000157 -0.000144
(0.00216) (0.00686) (0.000872) GENDER*AVAGE -0.0334*** 0.0419 -0.0159***
(0.0116) (0.0369) (0.00469)
INDEP 0.151*** -0.00224 0.0113
(0.0531) (0.168) (0.0214)
LNSIZE 0.0152*** 0.0208** -0.00326***
(0.00273) (0.00867) (0.00110)
LNBSIZE -0.0300 0.109 -0.0129
(0.0234) (0.0741) (0.00942)
Constant -0.0169 -0.459 0.109*
(0.153) (0.485) (0.0616)
Observations 942 942 942
R-squared 0.070 0.013 0.041
Standard errors in parentheses
Note: ***, **, * denote significance level of 1%, 5%, 10%, respectively.
Table 15 provides the results of a robust regression for model 2. This includes the test for hypothesis 2, the moderating effect. The results show a negative and significant interaction effect of GENDER*AVAGE on LEV and RDINT. This is consistent with the multiple linear regression results in table 11. The significant interaction effect on LEV is also in line with the clustered regression. Based on these results, I accept hypothesis 2, which means that the effect of board gender diversity on corporate risk-taking is negatively influenced by the average age of the board of directors.
5 Conclusion and discussion
This research examines the relationship between board gender diversity and corporate risk- taking, and how this relationship is influenced by age diversity in the board of directors. The sample for this study exists from data about U.S. listed manufacturing and services companies.
It includes 942 observations from 230 different companies. Board gender diversity has been measured as the percentage of female directors over total directors. Board age diversity has been measured as the average age of the board of directors of a company.
The social role theory predicts that the higher the proportion of women on the board of directors, the lower the level of corporate risk-taking. Many studies have shown that this prediction is sufficient (Bernile et al., 2018; Lenard et al., 2014; Perryman et al., 2016; Khaw et al., 2016; Jeong & Harrison, 2017). However, other studies found a positive or no significant effect of board gender diversity on corporate risk-taking (Adams & Funk, 2012; Sila et al., 2016). To test hypothesis 1, the effect of board gender diversity on corporate risk-taking, I conducted multiple linear regressions. Hypothesis 1 has been presented in model 1. The dependent variables to measure corporate risk-taking were leverage, delta of return on assets and R&D intensity. The independent variables in model 1 are board gender diversity and the average age of the board of directors. The model also includes control variables, which are the percentage of independent directors over total directors, the natural logarithm of total assets and the natural logarithm of board size. The regression results showed no significant effects of board gender diversity on all the dependent variables. Also the clustered regression and robust regression, checking for robustness, showed no significant effects. Therefore I reject hypothesis 1. The fact that the theory and prior literature suggests a significant effect and I find no effect means there is a type II error. The regression results show that the average age of the board of directors has a significant impact on all the dependent variables. When checking for robustness, most of the effects remain significant. However, some effects are positive and some effects are negative. Therefore it is not possible to draw a conclusion about the direction of the direct effect of director’s age on corporate risk-taking.
Hypothesis 2, the moderating effect of director’s age on the relationship between board gender diversity and corporate risk-taking has not been investigated in this organizational/corporate setting. The risk-sensitivity theory and young male syndrome predicts that the effect of gender on risk-taking behavior decreases as age increases. This prediction has been confirmed in in the ethical, financial, health, recreational and social domain (Rolison et
al., 2014; Byrnes et al., 1999; Gardner & Steinberg, 2005). I conducted multiple linear regressions with an interaction term to examine whether the average age of the board of directors influences the effect of board gender diversity on corporate risk-taking. This is presented in model 2. The independent variables and control variables are equal to those of model 1, but the interaction term has been added. The regression results provide a negative and significant interaction effect on leverage and R&D intensity. The robust regression also shows a negative and significant interaction effect on R&D intensity and leverage, but the clustered regression only reports a significant effect on leverage. In other words, my results provide some evidence to accept hypothesis 2.
The operationalization of the variables is based on prior literature. However, board gender diversity has not always been measured as the percentage of female directors over total directors. In addition, in this study the proportion females on the board represents the concept of board gender diversity. If there is a significant effect of board gender diversity on corporate risk-taking, this means that every women extra causes an increase or decline in corporate risk- taking. However, the critical mass theory explains that a minority of 1 or few females in a group are not considered as full members, but just as representatives of the gender (Kanter, 1977). This means that the significant effect of board gender diversity may be questioned due to a lack of contribution from the minority in the board of directors.
In the literature, board age diversity has been measured in ways other than the average age of the directors. In my research it has been measured as a continuous variable, but Bhat et al.
(2019) measured age in five categories, 40 and younger, 41-49, 50-59, 60-69 and 70 and above.
This may be more accurate if for example a person of 51 years old acts the same as a person of 59 years old. However, most of the existing literature treats age as a continuous variable, by which I conclude that this construct is valid (Berger et al., 2014; Chai & Sikandar Mirza, 2019;
Xu, Zhang, & Chen, 2018; Anderson, Mansi, & Reeb, 2004).
A limitation of this study is that the sample only contains listed U.S. firms in the manufacturing and services industry. Therefore the results might not be generalizable to companies in other countries or companies in other industries. Because there are many more manufacturing and services companies listed in the U.S. than the sample covers, the results might not be applicable to all listed U.S. companies in this industry. In this study, corporate risk-taking has been measured by leverage, delta of return on assets and R&D intensity. This is based on prior literature, but there are also many more measures for corporate risk-taking.
That is why my conclusions might not apply when using other risk measures. In my research,
