1
Male executives underperform than female
executives on investment decision on firm level from
2006-2014——An Empirical Study in US
Amsterdam Business School
Name Rui Tang Student number 10994947
Program Economics & Business Specialization Finance
Number of ECTS 15
Supervisor Ilko Naaborg Target completion 15 / 8/ 2016
1
Abstract
Many studies have shown that gender diversity plays an important role in corporate
governance. In the recent decade, the world entered into economic recession and experienced a large scale of financial crisis in the 2000s. It is important to investigate whether female executives behave better than male executives concerning financial decision making for their companies.I analyze how the listed firms in the United States raise debt over the past decades to empirically test whether male executives underperform than female executives on
investment decision on firm level. My regression results show that gender difference affects the investment decisions and using leverage as the the standard proxy for investment amount, female executives tend to take less leverage and are more financial risk-averse than male executives especially during economic downturn.
Keywords:
Corporate governance, Overconfident, Gender, Executives, Financial decision, Leverage.
Table of Contents
1. Introduction PAGE3-8
2. Literature review
PAGE8-13
2.1 Theories regarding research topic PAGE8 -10
2.2. Empirical findings in the literature PAGE10-11
2.3. Conclusion on the literature PAGE11 -13
2.4 Summary of main literature findings PAGE11-17
3. Methodology and Data
PAGE13-22
3.1. Methodology PAGE13-18
3.2. Data and descriptive statistics PAGE18-22
2
4. Analysis
PAGE22-33
4.1. Empirical Results PAGE22-30
4.2 Robustness check PAGE30-33
5. Conclusion and discussion
PAGE33-34
References
PAGE 34-35
Appendix A&B PAGE35-37
1
1.
Introduction
Previous studies have shown the board characteristics influence firms’ performance. Rhode and Packel (2014) shows that gender is one of the key elements when it comes to evaluate the investment amount, return of stock performance, or operation performance. Several other studies show that male and female executives behave differently from each other when making individual investment decisions (Charness and Gneezy, 2010，Barber and Odean, 1999). However, little has been done to link the managerial behavioral biases between male and female executives to the equally important firm investment decisions.
Lack of gender diversity in the boardroom is a phenomenon often overlooked despite its importance in the board characteristics. Data from Catalyst, ( a Non-Profit Organization NPO) committed to expanding business opportunities for females, show that board members are predominantly consisted by males. Among board members of US’s Fortune 500 companies, only 16% are females The corresponding rate in Germany, Britain and France is around 12%. Among Asian countries, the rate is even lower, 8.5% in China, 5.3% in India and less than 1% directors are women in Japan ( Catalyst, 2014).
As a result of the globalization and economic growth, recent proposals for improving corporate governance have paid attention to the importance of gender diversity in the boardroom. In 2003, a new rule passed in Norway provided that at least 40% of the board members shall be women by July of 2005. This proportion was the highest rate imposed by any government at that time in the world. The law became
2
compulsory in 2006 and firms that were not in compliance with the law by January of 2008 would be forced to dissolve. (Ahern and Dittmar，2011). While the proportion of female board members is still very low in most of countries. By now, sixteen countries have stated quotas of female representatives on boards in an increasing development. In the United States, support for diversity has grown in principle but progress has lagged in practice (Rhode and Packel, 2014).
After the legislation are put into practice and more and more countries promote the world gender equality situation, the demand for female executives increased. As a result, women have become an inevitable force in boards of firms all over the world. It is important to further explore the impact of female executives for regulatory purposes and long term industry development.
There are growing evidences in the theoretical studies that having female board members has a positive effect on company performance (Adams and Ferreira, 2008). Other evidence also suggest that forcing companies to do so can have a negative effect on company performance (Ahern and Dittmar，2011). Given there is no conclusive answer on the question whether female executives have a positive or negative effect on firm performance from the corporate governance point of
view(Smith and Verner, 2006), I propose re-examine the impact of gender diversity on firm performance from a behavioral finance perspective. In particular, my study will focus on behavioral biases between male and female executives during the financial crisis because more implications drawn about managerial capabilities during the period of financial crisis (Croson and Gneezy, 2004).
3
As we all know, in the past decade, the world entered into economic recession and experienced a large scale of financial crisis especially around 2008. The global economic crisis from 2008 is defined as the most severe economic crisis in the last 60 years globally. Its profound impact upon the global market has not yet come to an end until today. Meanwhile, this widespread economic crisis is also the first outbreak of the economic crisis since women start to enter into the economic field. From then on, people started to think of what kind of roles gender plays in the companies and what effects gender would cause economically.
According to the latest data reported by Rothstein Kass Company (a business consulting company), from January 1, 2013 to the end of November, 2013, the rate of return from the world's hedge funds administered by women was 9.8%, while the rate of return from the global hedge fund index (HFRX Global Hedge Fund Index) was 6.13%. From January 2007 to June 2013, this data seems to give a more obvious implication: Hedge funds administered by women recorded a rate of return at 6%, while the global hedge fund index only recorded a rate of return at 1.1%. Rothstein Kass' survey also showed that women played a more and more important role in the male-dominated hedge fund industry, especially after the 2008’s financial crisis. With the consciousness of financial risk management arising, they found that female were more prudent in making investment decisions and in risk management. Female have the advantage of flexibility which put them at the better position to managing risk: when facing sudden financial crisis, compared to men, women are more capable of wading the financial difficulties in the world crisis. It is a unique quality women own by themselves. (Rothstein Kass Company, Women's Alternative Investment: a marathon, not a sprint, 2013)
In the behavioral finance perspective, empirical studies find that individuals do not behave rationally, the behavioral biases impact investor decisions and affect financial area. There are some discussions about different behavioral biases. Gervais and Odean (1998) propose a model to address overconfidence is the result of self-serving attribution bias, which govern investors’ own tendency to trade or not. However they tend to be overconfident due to taking too much credit. Deaux and Farris (1977), Meehan and Overton (1986), and Beyer (1990) find that the self-serving attribution bias is greater for male than for female, thus male are willing to behave more overconfident than female.
4
Like individual make investment decisions in the exchange market, whose decisions may be influenced by their behavioral biases, executives making investment decisions for their firms may suffer from this behavioral biases too. Barber and Odean (2001) empirically proved that male behave more overconfident than female in investment decision in financial area, so I expect this assertion can be used to explain the hypothesis in this study that male executives behave worse than female executives on investment decision.
Apart from the corporate governance theory and behavioral finance, other fields also
gradually pay their attention on the issue of gender diversity in business world such as social psychology and biology. They offer different angles to explain the effect of gender diversity on firm performance, trading behavior or risk preferences. Therefore, it is necessary to combine different theories from different fields to find a conclusive answer and to make a regulatory progress. In the social psychology field, based on the theory of German sociologist Ulrich Beck, Skelton, Christine, (2005) provide the evidence that postmodern society is a risky society since socially produced risks are threatening the survival of
humanity. In the 21st century, with the further deepening of economic globalization followed by the risks of globalization, the financial sector has to pay more attention to risk. Gender will play an important role in reducing the risk, which is an unprecedented historical opportunity for women in the new century. In the biology field, Cambridge University neuroscientist Coates,Gurnell and Sarnyai, (2010) conducted an experiment showing that female body cortisol is only 10% of men’s, which indicated that female are less prone to make irrational exuberance. That is why normally departments responsible for trading hire more women than other departments and thus reduce the possibility of financial catastrophe.
The aim of this paper is to study whether male executives underperform than female
executives on investment decision on firm level during the financial crisis. It is not yet known on firm level whether the behavioral biases influence male or female executives on corporate investment decisions.
The previous debates from Adams and Ferreira (2008) and Ahern and Dittmar (2011) are based on the different view about whether the women on boards could affect the governance
5
of companies in significant ways. This paper provides new evidence that is relevant to the debate by investigating the hypothesis to test the relation between gender effects and corporate investment decisions. In particular, this paper extends the theory by using a behavioral finance perspective to explain the impact from behavioral biases on the corporate investment decisions.
To evaluate the managerial capabilities between male and female executives, I need to define the standard account for underperformance of male executives compared to female
counterparty; I use market leverage, capital expenditure and the firm size as proxies for checking the performance. The market leverage and capital expenditure represent composition of a firm’s investment. If firm decide to put more capital expenditure into investment, or they raise debt to do the investment. I can treat them the same decision of interest in my case to do further investigation. If the market leverage and capital expenditur e decline in the tenure of female executive, I can claim that female executive have better performance than male executives. Firm size is another proxy which represents debt and liabilities of a firm. If the firm size decline in the tenure of female executive, I can also claim that female executive have better performance than male executives.
The contribution this paper adds to the existing literature is as follows:
First, this paper tries to fill the gap in the literature by linking the managerial beha vioral biases between male and female executives to the equally important firm investment decisions. I use overconfident as a proxy for behavioral biases to explain the gender difference on corporate investment decisions.
Second, This study focuses the sample period on financial crisis to examine whether the behavioral biases influence the corporate investment decisions. The conclusion drawn from the crisis period can offer crucial implication to policy makers about future economic burst circle.
In this study, I use difference in difference approach to test whether male executives underperform than female executives on investment decision on firm level. Then I use instrumental variable approach to test the endogeneity issue. I check the multicollinearity through correlation matrix. Other than that, I also use two methods to do the robustness check.
6
One is traditional panel data regression with industry identifier SIC (Sta ndard Industrial Classification). The other one is to replace the dependent variable (market leverage) by capital expenditure which is another standard proxy for the investment, and add the net plant, property, and equipment (PPENT), and the gross plant, property, and equipment (PPEGT) as control variables for the new model to do the robustness check.
This paper collects data mainly from three databases. ExecuComp database provides executives information and Compustat-Capital IQ database provides related financial information of the selected firms. This paper also uses certain domestic economic index of the United States as the instrument variable to omit the potential endogeneity problems. According to Sugarman and Straus’s (1988) paper, they collect data from all over the states in the US and give each state a score for its gender equality status which has closely positive relationship with the independent variable, a fact that a firm hires a female executive. So here I use the economic index, the median earnings on state level for female full-time and year-round workers (in inflation-adjusted dollars) in the United States, to represent my
independent variable similar as what Sugarman and Straus (1988) did. I hand collect this median earnings on state level data from American FactFinder database. All the methodology and analysis are shown in section 3 and 4.
This article argues that male executives underperform than female executives on investment decision on firm level. Section 2 begins the discussion by reviewing related previous studies. Section 3 provides the methodology and data statistics of my test. Section 4 presents the analysis and results of my main model, instrumental variable approach and other tests . Section 5 comes to the conclusion and explores suggestions for policy makers.
1
2. Literature review
This topic is a combination of corporate governance and behavioral finance. The related literature in both fields is discussed below:
2.1 Corporate governance theories
There is extensive literature existing on the relationship between the characteristics of board members and performance of their companies. So I start my research from the corporate governance perspective. Adams and Ferreira (2008) conducted a survey based on the United States firms, and tested three hypotheses to examine whether gender diversity affects CEO turnover rate and their compensation, corporate performance and the committee assignments. They find that female executives have better attendance records than male executives,
meaning that female executives have fewer attendance problems. However, their findings on the relationship between gender diversity and firm performance are ambiguous. In their case, the correlation between gender diversity and firm performance may be positive at first, but this positive correlation disappears when they use some methods to solve reverse causality and omitted variables problems. No solid evidence shows that the policy of hiring more female will enhance the level of firm performance. This paper will re-examine one of their hypothesis on the relation between gender diversity and corporate performance and further investigate if there is a significant relation to see if I can get a more comprehensive
conclusion than theirs’.
Adams and Ferreira (2008) provide an analysis of how gender diversity in boards affects CEO turnover rate. They have direct empirical evidence shown that if there are more female directors on boards, the CEO turnover rate will be higher. In addition, they conclude that the number of female directors on boards is negatively related to the stock performance. Since CEO turnover rate is very sensitive to stock performance. So gender difference can affect the stock performance through CEO turnover rate. This paper studies the CEO turnover rate from a different point of view. The theory of Huang and Kisgen (2013) can be considered as a supplementary to the above mentioned findings, which indicates that there is a relation between the date female or male become CEO and firm acquisition activities. So I choose the date a female or male becomes CEO as a proxy to examine the relation.
2
For more political settings support, Ahern and Dittmar (2011) exploited a natural experiment in Norway to identify the impact of corporate boards on firm value. They look at the effect of the legislation that obliges companies to hire female executives in Norway, which can be useful for reference for policy makers in other countries in the world. They use panel data of 248 publicly listed Norwegian firms for 8 years to test. They provide new evidence to show that the firm values will change along with the new board structure mandated by the law. Although both positive and negative effect could be key evidence that board characteristics affect firm value, they find that there is significantly negative relation between board characteristics and firm value when the new board structure mandated by the law. Their conclusion is opposite to that of Adams and Ferreira (2008). The debates are very interesting, because the boards is designed to maximize the shareholder value, but the restricted choice of board member due to mandatory gender quota imposed by law actually hurts the shareholder value. Reflected in their test, the firm performance becomes worse with too much gender diversity.
They also find that in practice, new female directors are different from the existing male directors. New female directors are younger, highly educated but less managerially skilled. Compared to male directors, they prefer to be employed as non-executive managers. These characteristics provide the idea of adding age as a control variable into this research. Ahern and Ditmar’s (2011) results may provide policymakers guidance on how to maintain value while providing greater gender equality in the boardroom. This paper combined with Adams and Ferreira (2008) provide evidence for policy makers to choose the right proportion of board members for their companies.
2.2. Behavioral finance theories
As there are a lot of researches have already done on the relation between board structure and firm performance, we look for other perspectives such as behavioral finance field to further explore gender effects on firm level. Bailey, Kumar and Ng (2010) examine the effect of behavioral biases on the investment decisions in mutual fund industry. They use a large sample of data of the US market to confirm that behaviorally-biased investors make terrible investment decisions. As they predict, if investors have higher income, more investment experiences and higher educational level, they are more likely to invest in mutual fund and
3
benefit from the mutual fund market. Furthermore, if investors have strong behavioral biases, they prefer to use high expense index funds. When they invest in mutual fund, they trade more frequently and successfully.
There are two differences between Bailey, Kumar and Ng (2010) and this paper. On one hand, they use data from mutual fund industry, this paper choose to use the sample of all listed firms in the US. On the other hand, they focus their study on individual investors who trade individual stocks in mutual fund market, while this research focuses on the firm level.
Bailey, Kumar and Ng (2010) also describe biased investors of five stereotypes as “gambler”, “smart”, “overconfident”, “narrow-framer”, and “mature”. This paper chooses overconfident as a proxy for behavioral biases that may affect the corporate investment decisions.
These characteristics are used to describe different types of investors between male or female individuals. It is called the factor analysis that they apply to define these characteristics (Factor analysis is a statistical method which is used to describe variability among observed, correlated variables in terms of a potentially lower number of unobserved variables called factors. Wikipedia). They find that when using factor analysis to describe investor
stereotypes, all these characteristics can appear in the same individual. They may also can be reflected in the same executive. Thus, it will be difficult to distinguish the real factors which influence the corporate managerial decisions.
“Overconfident” is an important concept from the empirical behavioral finance literature, it means investors prefer to trade frequently but unsuccessfully. Since male investors are more likely to behave over confidently, Bailey, Kumar and Ng (2010) use a male dummy as a proxy for overconfidence. Bunch of previous researches such as Odean (1999), Barber and Odean (2001) use overconfident as proxy to study how behavioral biases influence
investment decisions of individual investors on individual stocks.
Based on a sample of 35000 individual accounts over a six-year period, Barber and Odean (2001) found evidence that males are not only more overconfident about their investing abilities but also trade more often than females. Normally, men tend to bear higher trading costs than women because women usually prefer a "buy and hold " approach while men often sell their stocks at the incorrect time. A 45% more trading than women indicates that men are more likely to be active portfolio managers.
4
Based on the model developed by Gervais and Odean (1998), they conclude that it is the self-seeing attribution bias that result in investor overconfidence. In the model, investors build up the knowledge of their abilities relying on their successes and failures. They become
overconfident because they tend to take too much credit of their successes. In addition, Deaux and Farris (1977), Meehan and Overton (1986), and Beyer (1990) find that men spend more time and money on security analysis, rely less on their brokers, make more transactions, believe that returns are more highly predictable, and anticipate higher possible returns than do women. All these existing literature provide support that men behave more like
overconfident investors than do women.
Bailey, Kumar and Ng (2010) also provide empirical evidence for the fact that behavioral biases affect investment decisions from psychology perspective. They find that female are more risk averse than male in the aspect of raising debt. Note that the types we construct are only proxies for behavioral biases. They do not represent accurate definitions of behavioral biases in psychology studies. However, the measure “overconfident” is used to describe executives in terms of raising debt in this study.
2.3. Corporate governance combined with behavioral finance theories
Huang and Kisgen (2013) combine behavioral finance with corporate governance theory in studying the topic of gender effects on investment decisions. I also apply these two
perspectives to this paper.
They propose a theory to explain behavioral biases between male executives and female executives in corporate settings. Note that this study is also based on corporate investment decisions, not on individual decisions. They examine not only executive decision making but also the market’s reaction to that decision-making to explain that male executives are
overconfident than female executives in financial investment decisions. Based on Bailey, Kumar and Ng (2010), they further investigate the overconfidence proxy to imply that female will undertake fewer projects, or are more willingly to make fewer significant decisions than male, holding other factors constantly.
5
Since overconfident male executives will overestimate the net present values of a new project, they will undertake more transactions. Male executives prefer to expand the acceptable
transaction pool which includes some deals with negative net present values. They test this implication by examining the frequency of acquisitions, debt issuance, and equity issuance decisions for male executives versus female executives. Overconfidence also implies that the investment decisions ultimately made by female executives will have more positive market reactions. If male executives are more overconfident in their financial decision making, the market will respond more negatively on average since a portion of the deals undertaken by male executives are value-destroying. They test this implication by examining announcement returns surrounding acquisitions, equity issuance, and debt issuance decisions for male executives versus female executives.
Huang and Kisgen (2013) use difference-in-difference empirical framework to study how the executive transitions (the transition activity starts from the date a firm hires a CEO) of male executives or female executives influence the corporate investment decisions. They find that if a firm hires female executive, it will grow slowly and less tend to make acquisitions. Furthermore, female executives often get higher announcement returns of investment compared to male executives for their firms respectively. The outcomes for capital structure decisions are also similar, that is, female executives are less tend to issue debt and often get a higher announcement returns of investment. The higher announcement returns around
acquisitions and debt offerings of a firm with a female executive indicate that there is a consistency with the result that male executives are more overconfident than female executives. They provide more empirical evidence on earnings forecasts and late exercise options to confirm the result. They also conclude that male executives are more likely to conduct value destroying activities. However, they also conclude that female executives do not make significantly difference to the total debt compared to male executives. From the market reaction side, investors also prefer to choose the corporate investment decisions made by firms with female executives, because they also prefer taking less risk.
I construct this difference-in-difference empirical method and instrumental variables approach based on the model of Huang and Kisgen (2013) to further investigate the
behavioral bias between male and female executives during the financial crisis period on the corporate investment decisions. Here is an extension to their study.
6
Gender has also been examined in other business settings, including stock trading behavioral (Barber and Odean, 2001), start-up firms (Verheul and Thurik, 2001), and competitive environments in laboratory settings (Niederle and Versterlund, 2007). In addition, Graham, Harvey, and Puri (2012) use a survey-based approach to identify differences in CEO risk aversion and optimism, and then they relate these differences to corporate investment decisions.
2.4 Summary of main literature findings
First, Adams and Ferreira (2008) find the relation between gender diversity and firm performance is ambiguous. The correlation between gender diversity and firm performance may be positive at first, but this positive correlation disappears when they use some methods to solve reverse causality and omitted variables problems.
Second, Ahern and Dittmar (2011) provide the empirical evidence that there is significantly negative relation between board characteristics and firm value when the new board structure mandated by the law, which is opposite to the findings of Adams and Ferreira (2008).
Third, Bailey, Kumar and Ng (2010) introduce several proxies for behavioral biases based on factor analysis. They find that males are not only more overconfident about their investing abilities but also trade more often than females.
Fourth, Huang and Kisgen (2013) propose a theory to explain behavioral differences in gender in corporate settings. They find that male executives are overconfident than female executives in financial investment decisions.
Finally, although many previous researches have been done, there is still gap I can fill such as linking the managerial behavioral biases between male and female executives to the firm investment decisions. To combine corporate governance and behavioral finance theory, this paper focuses on behavioral biases between male and female executives on firm level. Furthermore, this study focuses the sample period on financial crisis to examine whether the behavioral biases influence the corporate investment decisions. The conclusion drawn from the crisis period can offer crucial implication to policy makers about future economic burst circle.
8
3. Methodology and Data
3.1. Methodology
The research approaches are divided into two parts. First, I use a difference-in-difference approach to evaluate male executives underperform than female executives on investment decision on firm level. Second, I use instrumental variable approach to test the potential endogeneity problems.
Female and male executives in most cases self-select into firms that are different in corporate environment, degree to which gender discrimination exhibits and other board characteristics. Endogeneity due to selection is therefore inevitable. If the empirical study wants to be randomly assigned, there may have some reasons, such as the age of an executive or the same first character of the first name of an executive. But even these random characteristics are based on gender; there are still possibilities of endogeneity. Because other characteristics may also be correlated with the behavior that is discriminated by gender, these characteristics can also affect the result in this study.
Except that, according to Bergmann’s (1974) overcrowding hypothesis, female executives are more likely to show up in retail industries. Bergman argued that female executives tend to or are forced to take jobs that male executives are not good at or not interested in. If retail industry is at a steady growth, female executives can also seek out positions in fast growing industries, especially where competition between males is furious.
First, I choose a difference-in-difference method for this empirical study. Using panel data with firm fixed effect and time fixed effect can get similar result, but in this case, difference-in-difference method is better. Because I want to compare the behavioral biases between a male executive and a female executive before and after transitions. In this way, I can control a sample except the male and fefemale transition firms, only keep male-to-female firms (male-to-female-to-male is the same as male-to-male-to-female). In difference-in-difference method, it is necessary for the executives to be effective during their tenure of office because the unobserved effects of a transition of an executive should be omitted. Note that effective means that the executive makes significant changes to their companies. Furthermore, I assume that the executive should be effective for at least three years (effective from the
9
transition year and the next two following year) to keep the executive have enough time to impact his/her decision making. Due to the comparison of the behavioral biases between a male executive and a female executive before and after transitions.
I omit all the time invariant unobserved effects. So in accordance with the changes of the transition year, all the other characteristics have to be changed as well and be independent at the same time.
However, the endogeneity problem will still exist after the difference-in-difference test. So I need to omit the unobserved effects caused by the discriminatory choices of the firm when it hires a female executive. As discussed before, a firm knows that it should not take too mu ch risk when the economic environment becomes worse, so it is likely to hire a female executive to apply prudent financial policy. As a matter of fact, this circumstance is consistent with this case of study. Recognizing that gender effects are unlikely to be the only determinant of the firm’s debt, I also need to construct control variables for other drivers of investment decisions suggested by the corporate governance and behavioral finance literatures, such as the
education level, age or nationality (Bailey, Kumar and Ng, 2010). After adding these control variables, I can clarify the relationship between gender effects and investment decisions. Second, to omit the unobserved effects of the discriminatory choices of the firm when it hires a female executive, I use instrumental variable approach. The instrumental variable I choose for a firm is a measurement of gender equality based on the state level of the United States. Refer to Sugarman and Straus (1988) who collected data from all over the states of the United States and gave each state a score for gender equality. I use median earnings in the past 12 months (in inflation-adjusted dollars) by sex for the whole population as a
measurement of gender equality. It is used to announce that the more earnings a state has, the more likely a firm built in the state is to hire a female executive, that is, the more equal female will be in this state. The instrument variable can be economic, political or in some other aspects, Sugarman and Straus’s (1988) score is political while this data is economical. For a country like the United States, the median earnings of female versus male through the whole year can be treated as important domestic economy indicator. I download data from American FactFinder database and keep it consistent with the time period of my sample for the difference-in-difference model. After merging the two sample data, I do the 2SLS
10
instrumental variable regression to check whether the male-to-female transition activity have significant effect on investment decisions.
Huang and Kisgen (2013) test their theory in the US market. Refer to their hypothesis, I test the relation between gender effects and corporate investment decisions. It reflects whether overconfident male executives have worse performance than female executives. The main objective of this thesis is to provide an extended theory that the investment decision made by male executives and female executives are caused by their behavioral biases. So this
hypothesis is male executives underperform than female executives on investment decision on firm level in the US firms.
Empirical model:
Y it+1 = μ + νi + τt + β1 Posti,t+1 + β2 Femalei*Postit + θXit+1 + εit+1 (1)
Where Y i,t+1 is the decision variable of interest (e.g.market leverage as proxy for
investment amount/debt level) measured at the end of year t+1; νi are firm fixed effects, τt are year fixed effects; Female i is an indicator variable for whether firm i is switch to a female excutive; Posti t is an indicator variable, Posti t=1 if the firm has a female executive at year t;
Xi,t+1 is a set of control variables (firm size, market-to-book ratio, and PPE) for firm i measured at the end of year t+1 and the residual term ε_it+1 also is measured at the end of year t+1. Note that I define all the variables except Posti t to be measured at the end of year
t+1 because the outcome of a firm will not be effective at the the first year a male or female becomes executive. The result of investment decisions made by executives will be effective at the end of next year. No dummy for female executive is needed since it will be included in the firm fixed effects.
The coefficient 𝛽2 is going to reflect the relationship between debt level and whether the firm has a female executive. I use autocorrelation adjustment at firm level to account for serial correlation within firms, cluster the standard errors at industry level and use robust standard errors to account for heterogeneity.
11
The significantly negative of coefficient β2 indicates is negative and significantly, I can conclude that Femalei*Posti t has a negative impact on firm decisions in terms of market leverage affects the decision variable of interest (market leverage) significantly. In other words, It means that female executives make significantly worse corporate investment decisions than male executives.
Note that there is two limitations of using DID approach. One is because of potential serial correlation bias (Betrand, Duflo, and Mullainathan, 2004). I therefore conduct a correlation matrix to test potential multicollinearity problems by assessing the dependence between two or multiple variables at the same time. The result of the correlation matrix is a table
containing the correlation pairwise coefficients between each variable and the others. The value of correlation coefficient close to+1 or -1means almost perfect positive or negative linear relationship. I will discuss this result of multicollinearity in Section 4.1.
However, the other one is more likely to be the main limitation of DID, it is called
Ashenfelter dips (Heckman and Smith, 1999). Ashenfelter(1978) noted a potentially severe limitation of his experiment when he observed that the mean earnings of people who took part in the government training programme decline in the time prior to programme entry. In addition, in the case of training program for a better job acquisition, people who joins the plan are those with better learning abilities and intelligence capacity, they are likely to succeed regardless of whether they are being trained or not. In this case, DID will be
undermined if firms switch to a female because there are some firm attributes that will make the female executive succeed. So it is firm's capacity/traits that lead to performance
improvement rather the other way around. The instrumental variable approach is used to solve this problem.
The potential endogeneity problems due to self-selection is not fully accountable for by DID. I then use the instrumental variable approach to rule out any lingering concerns (Huang and Kisgen, 2013). As discussed in section 3.1, I develop the instrumental variable following Sugarman and Straus (1988) where the index of gender equality on the state level in the United States is used to conceptualize women's equality. Following their approach, I use the index of median earnings of female versus male on the state level as an IV for whether a more friendly state will hire more female executives. The state level of gender equality can
12
be assigned to each firm based on where their headquarters are located in. The higher the median earnings are, the friendlier the state is towards female.
I expect I can obtain a strong instrumental variable because a strong instrumental variable can provide unbiased parameter estimates and robust standard errors. A strong instrumental variable need two requirements: one is the instrumental variable should be closely correlated with the firm which hires a female executive and has no relationship with other unobserved firm characteristics. In other words, it can only directly affect through the gender effects of our executives. The investment decision of a firm does not depend on whether the firm will hire a female executive or not,
The other one is the instrumental variable has no independent effect on the dependent
variable. The investment performance of the firm can affect the income of the state where the firm located at. So there is a potential problem with my choice of instrumental variable. In this case, median earnings on the state level can be changed along with the investment performance of the firms in its state. So the second requirement is not fully satisfied. But as we know, it is hard to find a valid instrumental variable in practice. Instead of using mean earnings on the state level which is more reliable on the investment performance of the firms, I choose the median earnings which is much more independent. So this instrumental variable is suitable to test in my study, it is the best instrument variable under current conditions. Two-stage least squares (2SLS) models:
First stage: Femaleit+1 = φ + τt + γFemalei Earningsit + θXi,t+1 + ηit+1 (2a)
Second stage:Y it+1 = μ + νi + τt + β1 Posti,t+1 + β2 Instrumented Femalei*Posti t + ωXit+1 + εit+1 (2b)
Where Yi,t+1 is market leverage of firm i; Femaleit+1 is a dummy variable that equals one if
the firm has a female CEO; Instrumented Femalei is the fitted probabilities of the female indicator from the first-stage regression; Femalei Earningsi is the state-level Female full-time, year-round workers with earnings; Posti t is an indicator variable, Posti,t=1 if the firm has a female executive at year t; Xi,t+1 is a set of control variables (firm size, market-to-book ratio, and PPE) for firm i measured at the end of year t+1 and the residual term ε_it+1 also is measured at the end of year t+1.
13
Note that Femaleit+1 is a dummy variable as dependent variable, I conduct the logit model
for the first stage. Logit model is used to model dichotomous outcome variables. In the logit model the log odds of the outcome is modeled as a linear combination of the predictor variables (Hosmer, D. & Lemeshow, S. 2000).
If the coefficient between median earnings and female dummy is positive and significant, it means the instrumental variable satisfy the first requirement. For the first stage regression, I use autocorrelation adjustment at firm level to account for serial correlation within firms, cluster the standard errors at industry level and use robust standard errors to account for heterogeneity.
And if the coefficient between the instrumented female variable and market leverage is negative and significant on the second stage, the same conclusion remains: male executives underperform than female executives on investment decision on firm level. I use
autocorrelation adjustment at firm level to account for serial correlation within firms, cluster the standard errors at industry level and use robust standard errors to account for
heterogeneity. I discuss the economic implication behind the instrumental variables approach and its limitation in Section 4.
Finally, I conduct two robustness checks. The first one is the traditional panel data
regression using cluster analysis within industry group with firm fixed effects. Because the objective of cluster analysis is to assign observations into homogeneous and distinct groups. In previous analysis I have already clustered the standard errors by using company name as a proxy for the groups. More specifically, it improves the chances of a good response rate to the ultimate result. So I conduct the cluster analysis by industry identifier.
The other way to test is to use capital expenditure as an alternative dependent variable instead of the market leverage of the firm. Then I run the panel regression similar to the main equation(1) with few changes in dependent variable and control variables. I add PPENT and PPEGT as new control variables because a firm uses capital expenditure to acquire or
increase physical assets such as property, plant and equipment or industrial buildings. This method can check the robustness of gender effects on financial decision making in this sample model. I will report the results and conclusion in Section 4.1.
14
3.2. Data and descriptive statistics
Now I describe the data sets needed for the empirical tests.
3.2.1 Data collection
I select the CEOs as the research objective in this paper because of its importance in all the characters of a firm. A CEO is the chief executive officer of a firm, who the position of the most senior corporate officer, executive, or administrator in charge of managing an
organization. A CEO of a corporation or company typically reports to the board of directors and is charged with maximizing the value of the entity. Titles are also often given to the holder of the CEO position include president and chief executive (CE).(Wikipedia), so I collect CEOs data for this analysis. The CEO data provide me a large sample of 1030 firms with female executive for this empirical analysis. In the meantime, CEOs have important influence on corporate financial decision and policy making, which also require to be tested. In capital structure decisions, CEOs make the ultimate decision later than the CFOs.
For the female executives I use my dataset on the ExecuComp database (which only includes the largest firms), I only pick firms with total assets more than $500 million which represent large firms. The ExecuComp database includes executives information either from company proxy statements or from annual reports, such as executive age, gender the date beca me CEO, tenure as executive,rank of the executive for all firms.and the number of other executives. I also define the firm should be a listed firm on NYSE, AMEX or NASDAQ in Compustat. Compustat-Capital IQ database will be used to collect financial information, like book assets, total debt, and capital expenditure and so on. I also get standard industrial classification (SIC) codes and business segment data from Compustat.
Because I want to study how the gender effect and gender bias change over the last decade, especially in the financial crisis, my sample annual data range is from 2006 to 2014. I collect the gender of executives, the full name of CEOs, and the rank of executives. This sample criterion identifies executives for 13,272 firms’ observations. My number of observations changes within regressions because of potential multicollinearity. The analysis of correlation matrix will be discussed in Section 4.1
15
I set this sample of executive transition from the company name/ year list using the following conditions:(1) I require the company reports at least three consecutive years as a standard of an executive; (2) The transition year started from 2006 to 2014, and the total assets of a firm must be greater than $500 million; (3) If the predecessor executive is a male, which means that the transition either would be a male to female transition or a male to male transition. After filtering, this sample covers 1030 cases of male to female transitions and 12,242 cases of male to male transitions. I assume the transition year as the first year when a new
executive became CEO in a firm.
I also collect the biographical information of executive like age. Control variables include size, market-to-book ratio, and PPE which defined in Appendix A. In the set of firm level controls I include one firm outcome measure, market leverage of last year. All of above control variables include year dummy variables. To solve heteroskedasticity problem and the regression without firm fixed effects, all standard error are adjusted.
I hand collect the median earnings on state level for female full-time, year-round workers (in inflation-adjusted dollars) in the United States from 2006 to 2014 which is accordance with my financial data and executive sample period.
(http://factfinder.census.gov/faces/tableservices/jsf/pages/productview.xhtml?src=bkmk)
3.2.2 Summary statistics
For this sample, the summary statistics of executives’ transitions are shown in Table 1. This table presents the distribution of this sample executives by gender and other characteristics, including the transition year (the first year when a new executive became CEO in a firm). I require that the executive has to be in office consecutively for at least three years. Table 1 indicates that in general firms hired more female executives in financial recession
years(especially from 2008 to 2010) in the last decade with peak value difference of 12% of female versus 11.4% of male executives in 2008. This growing number may be a symbol as increase in the supply of highly qualified female executives, and the fact female executives are better at prudent financial decision making in financial crisis.
16
Table 1
The distribution of executives by gender and transition year as follow:
This table presents the distribution of the sample executives by gender and other
characteristics, including the transition year (the first year when a new executive became CEO in a firm). I require that the executive has to be in office consecutively for at least three years. Transition Year Total 2006 2007 2008 2009 2010 2011 2012 2013 2014 12,24 2 Male 1315 1478 1397 1387 1374 1378 1353 1356 2226 10.7 % 12.1 % 11.4 % 11.3 % 11.4 % 11.3 % 11.1 % 11.1 % 18.2 % Female 80 101 124 100 122 115 132 128 119 1,03 7.7% 9.8% 12 % 9.7% 11.8 % 11.2 % 12.8 % 12.5 % 11.6 %
In Table 2, the table presents the mean statistics of market leverage, and size of the sample firms around the year which firms hire female or male executives (transition year t). I
calculate the mean market leverage and size of all the sample firms and compare them under different transition condition. I select market leverage and the firm size as proxies to do a simple analysis before the main regression. It’s easy to get the unadjusted results without adding any control variables. The unadjusted results shows that female executives reduce market leverage after transition year and they have lower leverage level compared to male executives no matter before or after a transition. Meanwhile, Table 2 shows the mean size of firms which hires a female executive is much greater than that of male firms the year both before and after the transition year. They are consistent with the hypothesis that male executives underperform than female executives on investment decision on firm level
17
Table 2
Firm leverage and size around executive transitions as follow:
The table presents the mean statistics of market leverage, and size of the sample firms around the year of transition. . I use dummy variable gender equals to 0 represents the firm which hires a male executive (Male CEO), gender equals to 1 represents the firm which hires a female executive (Female CEO); t-2 represents two years before the transition year, t-1 represents one year before the transition, year t represents the transition year and t+1 represents one year after the transition year. See Appendix A for the definition of the variables.
Market Leverage Size
Male CEO Female CEO Male CEO Female CEO t-2 29.65% 29.64% 17088.54 37013.24 t-1 28.16% 22.92% 8721.84 42287.96 t 30.06% 28.87% 28814.22 43170.96 t+1 27.57% 22.98% 16183.99 19026.87 Note: all statistics are summarized based on current year.
Table 3 presents regression analysis of the likelihood of hiring a female executive. The dependent variable is a binary variable which equals 1 if a transition firm hires a female executive and 0 otherwise. I define female as the dummy variable. It shows a probit regression of whether a firm will hire a female executive, I use control variables including firm size, market to book ratio, total assets capital expenditure, the market leverage of lag transition year, debt in current liabilities, deferred taxes, investment tax credits and year. Only in regressions with all the control variables can I get a significant effect on the control variable year, which is not representative at the firm level. So I come to the conclusion that on these characteristics, there is no preference for firms to hire female executives. It’s impossible for us to use all the characteristics to predict what kind of firms prefer to hire
18
female executives. But considering the control variable year, whether the firms would hire female executives are correlated with time. For example, it’s different for firms to hire or fire people in financial crisis or economic booming period. See table 3 in Appendix B.
4. Analysis
To examine the hypothesis that male executives underperform than female executives on investment decision on firm level. I examine the relation between the market leverage of the firm and whether the firm will hire a female executive. I begin by analyzing the effect of gender on firm leverage which is important for a firm’s capital investment, so I use it to represent investment decisions. I analyze male-to-female transition because I wonder if the firm’s investment decisions will be changed after a male-to-female transition. Next I conduct an Instrumental variable approach to rule out potential endogeneity problems due to DID or any lingering concerns. Further I do a correlation matrix to check multicollinearity problems in the main regression. Last, I report the results of various robustness check.
4.1. Analysis and Empirical Results
As I discuss the main regression of the model in section 3.1, I produce Table 4. Table 4 presents the results on capital structure decisions using two specifications. The dependent variables are the decision variables of interest (capital structure decisions/debt level). Column (1) reports difference-in-difference regression results from Eq. (1). Column (2) reports the regression results using the full panel of firms. Results from these regressions are reported in Table 4. Column (1) reports the results of the main difference-in-difference method, Column (2) reports the results using with market leverage as the dependable variable. I do not use book leverage because it is the market leverage which can reflect the effect operated by CEOs. In the difference-in-difference method, I use Femalei*Posti t to represent whether the firm hires a female executive.
Table 4
19
Model1 Model2 Model3
Post 0.00140 (0.13) -0.00111 (-0.13) 0.00126 (0.14) Female*post -0.0403* (-2.01) -0.0258** (-2.86) -0.0281** (-2.97) Market leverage of last year 0.335*** (12.29) 0.340*** (12.46) Firm size -0.00000138*** (-3.42) Capital expenditure 0.00000990 (1.93) Total assets 0.00000139* (2.52)
Note: All regressions include year fixed effects and firm fixed effects. Autocorrelation adjustment is applied to account for serial correlation. Robust standard errors are further clustered at industry level.
This table presents results on capital structure decisions using two specifications. The
dependent variables are the decision variables of interest (capital structure decisions/ leverage level). Column (1) reports difference-in-difference regression results from Eq. (1). Column (2) reports the regression results using the full panel of firms. See Appendix A for the definition of the independent variables. Numbers in parentheses are t-statistics based on robust standard errors. Significance on a five percent (*), ten percent (**), or one percent level (***) is indicated.
20
In Column 1 (Model 1), I find that coefficient on Femalei*Posti t is negative and statistically significant (at 0.05%), and I can also find the marginal effect of the Femalei*Posti t,
evaluated at the means of the data, beneath its t-statistics in Column 1. I can use this to assess whether the gender effects are economically significant, which means that the female
executives raise much less debt than male executives.
My results may have omitted variables problem due to unobserved firm characteristics or reverse causality. For a lot of firms, they prefer to hire female to their boards and assign them position related to finance field. I need to add some control variables to solve this problem, in Column 2 (Model 2), I add some firm fixed effects and firm level control variables still using the original regression model. The coefficient estimate on Femalei*Posti t changes a little than before but becomes more significant at the 1% level than before(in Column 1, the significant level is at the 0.05% level). Additionally, the results are robust including the firm fixed effects. In Column 3, I randomly choose control variables based on Model 1 to check the results. The coefficient estimate on Femalei*Posti t is similar to that of Model 2 and is significant at the 1% level.
The conclusion is that, even after controlling for firm level characteristics such as size, capex expenditure, market leverage of last year and total assets, female executives appear to have better performance than male executives. It is consistent with a large empirical theory stating that female are intrinsically different from male (Croson and Gneezy, 2004).The results from above tests state that female executives firms increase slowly, which are consistent with those male executives prefer to invest more amount than female executives. The result provide evidence that female executives make significantly different corporate investment decisions from male executives due to their behavioral biases. It is consistent with the hypothesis that male executives underperform than female executives on investment decision on firm level. The behavioral biases do affect the corporate investment decisions.
As I know that on the transition year, the outcome has not effectively affected by CEOs, so I need to do robust check by set a new variable called leadtransyear which represents three years after the transition year to further test, I will discuss it later in this section.
I report the instrumental variable method in Table 5. There are two requirements for valid instrumental variables as I discussed in Section 3. Thus, I can base on economic
21
theory or conduct natural experiments to find instrumental variables that satisfy the requirements.
Table 5
Instrumental variable approach
Table 5 presents the 2SLS regression results from Eq. (2a) and (2b). It reports the results from the first-stage OLS regressions with the Female Earningsi dummy as the independent variable. F-statistics and Wald test statistics from the first-stage
regressions are reported at the bottom of the table. It also reports the results for the second-stage regressions with market leverage of the firm as the dependent variables.
Instrumented Femalei dummy is the fitted value of the female indicator from the
first-stage regression. Female Earningsi is the state-level Female full-time, year-round workers with earnings (refer to the state-level gender equality index proposed by Sugarman and Straus (1988)). See Appendix A for the definition of other variables. Numbers in parentheses are t-statistics based on robust standard errors. Significance on a five percent (*), ten percent (**), or one percent level (***) is indicated.
Market leverage Femalei 0.342 (1.02) Capital expenditure 0.0000102* (2.05) Firm size -0.000000382 (-1.55)
Market leverage of last year 0.909***
22
Posti -0.0697
(-1.56)
Table 7
23
Note: All regressions include year fixed effects and firm fixed effects. Autocorrelation adjustment is applied to account for serial correlation. Robust standard errors are further clustered at industry level.
In Table 5, I use chi-square test to examine the 2SLS regression, the chi-square is used to test whether two categorical variables are independent (no relationship), thus in my case, to test whether there is an association between Instrumented Femalei and market leverage of the
24
firm. The chi-square value with five degrees of freedom is 686.06, which is statistically significant at the 99.99+% level. However, there are some limitations of chi-square test that if the test comes out significant that means there is some associations but no further information. I produce Table 6 which presents the result of the first stage of instrumental variable
approach. I use logit model which I discuss in Section 3 to get the results. The results show that the coefficient of Female Earningsi t is not significant. See the Table 6 in Appendix B. In table 7, in the first stage regression, I should conduct a F-test on all instruments to see if instruments are jointly significant in the endogenous variable Femalei. The F-statistics is 2.3729 which is very small and lower than 10, which means the instrument is weak instrument. It is consistent with the results of Table 6. When the instrumental variable is weak, the two stage regression estimators could be inconsistent or have large standard errors. However, Table 5 shows the standard errors are not large. I conclude the IV or 2SLS estimators are inconsistent. This research result need to reject the null hypothesis that male executives underperform than female executives on investment decision on firm level. The behavioral biases do not affect the corporate investment decisions. There are several other reasons such as the CEO experience age or education level that account for the corporate investment decisions.
Staiger and Stock (1997) suggest that to ensure that the maximum bias in 2SLS estimators to be less than 10 %, the F-statistics of instrumental variables should be larger than 10. If the maximum bias in 2SLS estimators is no more than 20 %, the F-statistics of instrumental variables should be larger than 5. If there is only one instrumental variable, I should use the t-statistics to replace the F-t-statistics.
It also reports whether the instrumental variable is a valid instrument. In my case, median earnings on the state level can be changed along with the investment performance of the firms in its state. So the second requirement is not fully satisfied. But as we know, it is hard to find a valid instrumental variable in practice. I find it very time consuming to searching for an adequate instrumental variable. Instead of using mean earnings on the state level which is more reliable on the investment performance of the firms, I choose the median earnings which is much more independent. Sugarman and Straus (1988) proposed the
state-25
level gender equality index, while I use the real economic index as the instrumental variable. So this is the limitation of IV regression to test the consistency of the 2SLS estimators. Because the F-statistics is not valid in the presence of heteroscedasticity, I need to use the heteroskedasticity-robust Wald statistic. I can get a heteroskedasticity-robust t statistic by dividing an OSL estimator by its robust standard error for zero null hypotheses that
instrument are weak. Stock and Yogo (2005) proposed the null hypothesis that the
instruments are weak. I need to use either the maximized 2SLS relative bias I want to tolerate or the maximized rejection rate of nominal 5% Wald test I want to tolerate. I can conclude the instrumental variables are valid if the test statistic is greater than the critical value. However, in the model, the critical values under each relative bias are not available. Thus I come to the rows states “2SLS size of nominal 5% Wald test” and “LIML Size of nominal 5% Wald test”. Because the critical values in the rows represent the second
characterization of weak instruments in Stock and Yogo’s (2005). I call the instruments weak instruments if the Wald test is at 5% level by proposing the rejection rate less than 10%, 15%, 20%, or 25%. Suppose I want the rejection rate of at most 10%, I report the e the test statistic of 2.3729 < 16.38, I can not reject the null hypothesis that the instrument is weak. So it is consistent with the usual F-statistic test before.
As I discussed in Section 3, I need to conduct a correlation matrix to solve the potential multicollinearity problems in the main regression. I report the results in Table 8.
Table 8
Correlation matrix of the main regression, see Appendix A for the definition of the independent variables. See Appendix A for the definition of the independent variables. Numbers in parentheses are t-statistics based on robust standard errors. Significance on a five percent (*), ten percent (**), or one percent level (***) is indicated.
26
By definition, any variable correlated with itself always has a correlation of 1. It applies to the report. To examine the correlation between multiple variables in the main regression, the pearson correlation coefficient and the level of statistical significance can show us the
strength and direction of a linear relationship between two variables. It reports in Table 7 that capital expenditure is positively correlate with size and market leverage of last year, it is because size is defined by long term debt plus debt in current liabilities market capitalization at the end of the fiscal year plus preferred stock liquidating value, while capital expenditure is part of firm’s capitalization. As the size of the firm increases, the capital expenditure will also rise in practice. Same as the correlation between capital expenditure and market leverage of last year, the positive correlation reveals that both of them reflect the capital structure of the firm, that is, capital expenditure increases as market leverage of last year increases.
The largest significant correlation is between the total assets of the firm and the size of the firm. It is a fact with the growing development with the size of a firm, the assets of the firm will increase, too. Higher values of the size of a firm are associated with greater assets of the firm.
But the large correlation effects only exist in control variables. The independent variable Femalei*Posti t has a weak correlation with post because post is part of the Femalei*Posti t variable. So I can conclude that the multicollinearity problems can be neglected.
27
As I discussed in section 3.1, there are two ways of testing robust check.
First, I find that the more female executives the firm hires in transition year, the less risks of raising debt the firm will face in general. The male-to-female transition has a negative correlation with market leverage of the firm. I did cluster analysis in the main regression by firm identifier, however, to be more specifically and improve the chances of a good response rate to the ultimate result, I need to conduct the cluster analysis using industry identifier. This is the first way of clustering the standard errors. I get the result in table 8.
Table 9
Traditional panel regression uses industry identifier SIC ( Standard Industrial Classification ) code which is a system for classifying industry areas. With this code, I run the previous panel regression keeping other variables unchanged. See Appendix A for the definition of the independent variables. Numbers in parentheses are t-statistics based on robust standard errors. Significance on a five percent (*), ten percent (**), or one percent level (***) is indicated.
Market leverage
Female*post -0.0258*
(-2.51)
Post -0.00111
(-0.11)
Market leverage of last yearr 0.335***
(10.91)
Firm size -0.00000138**
(-3.22)
Total assets 0.00000139*
28
Capital expenditure 0.00000990
(1.43)
Note: All regressions include year fixed effects and firm fixed effects. Autocorrelation adjustment is applied to account for serial correlation.
Table 9 shows that after clustering the standard errors by industry identifier, I can still get the coefficient of the firm which hires a female executive which is negative at 0.05% significant level. I can conclude that the test is claimed as robust because it still provides consistent result with the previous regression despite the condition changes.
Second, I conduct the second way to do the robust test by having the decision variables of interest and assumptions altered. I design a similar model as the equation(1). I choose capital expenditure instead of market leverage of the firm to test whether the firm hires a female executive has significant effect on capital expenditure. I have the result in table 10.
Table 10
This table presents results on capital expenditure. It reports the results from panel regression using capital expenditure as dependent variable, adding PPENT and PPEGT as control variables to improve the efficient and get unbiased coefficient estimator. See Appendix A for the definition of the independent variables. Numbers in parentheses are t-statistics based on robust standard errors. Significance on a five percent (*), ten percent (**), or one percent level (***) is indicated.
Capital expenditure
Female*post -86.62*
29
The net plant, property, and equipment 0.0000869 (0.30) The gross plant, property, and equipment -0.0000703
(-0.14)
Total assets 0.0000167
(0.84)
Firm size 0.0131***
(686.94)
Note: All regressions include year fixed effects and firm fixed effects. Autocorrelation adjustment is applied to account for serial correlation.
I choose capital expenditure as the variable of interest because capital expenditure represents the expense of a firm where the benefit continues over a long period. Because both capital expenditure and market leverage can reflect a firm’s financial condition especially in investment decision. If firm decide to put more capital expenditure into investment, or they raise debt to do the investment. I can treat them the same decision of interest in the case to test the hypothesis. The coefficient between firms which hire female executives and capital expenditure is negative at 0.05% significant level, which is consistent with the result of Table 4 except the sign of prior interaction is at 0.1% significant level. Thus, the test is robust and remain effective.
My results of above two robust tests suggest that firms should not hire female executives with the expectation that female will automatically improve firm financial decision level. The results are consistent with previous research such as Huang and Kisgen (2013).
5. Conclusion and discussion
Using the sample of thousands of listed firm data in the United States, I have shown that behavioral factors do not impact the investment decisions of female executives or male executives of a firm. As I might expect, firms with female executives are less likely to take too much risk to lend money and issue debt than the firms with male executives. To be more
