1
Bachelor Thesis on Corporate Governance
BSc Business Administration: Finance
Gender Diversity within the Boardroom: An Analysis of the
Economic Rationale behind Gender Quotas
by Lea Katharina Kasper
supervised by Florencio López de Silanes Molina Abstract
Gender equality is still a major imperative all over the world. While the topic is commonly morally and ethically justified, it becomes increasingly popular to be a key component to economic growth. Since Norway announced a first gender quota for corporate boards in 2003, a major rise of governmental initiatives and mandated quotas took place, based on this economic rationale. This paper examines the widely discussed debate behind the relationship of gender diversity on boards and firm performance. Through an extensive empirical analysis on a sample of 156 European firms, over the past 10 years, I do not find a significant relationship between gender diversity and firm performance after controlling for endogeneity and omitted variable problems. Further, no interaction effect of mandated quotas with this relationship is supported. My findings underline the empirical importance to account for endogeneity problems within the relation of gender diversity and firm performance. Further, the study challenges the economic rationale and stresses the implementation of necessary factors when investigating gender diversity and firm performance.
2
Statement of Originality
This document is written by Lea Katharina Kasper 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.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3 Contents
1 Introduction ... 5
2 Literature Review ... 8
2.1 Board Diversity and Firm Performance ... 9
2.2 Gender Diversity Quotas ... 11
2.3 Empirical Findings ... 12
2.4 Hypothesis ... 14
3 Data and Methodology ... 15
3.1 Population and Data Description ... 15
3.2 Variables ... 16
3.3 Empirical Methodology ... 19
4 Results and Analysis ... 22
4.1 Descriptive Statistics ... 22
4.2 Multicollinearity and Correlation Matrix ... 26
5 Empirical Results ... 27
5.1 The effect of gender diversity on firm performance ... 27
5.2 The Moderating Effect of Gender Quota ... 35
6 Discussion & Conclusion ... 39
References ... 42
4 List of Tables
Data Selection Process ... 16
Summary Statistics of the Fraction of Female Directors per Country ... 23
Summary Statistics of the Fraction of Female Directors per Year ... 23
Frequency of Firms with 30% or more women on boards per Year ... 23
Frequency of Firms with 30% or more women on boards per Country ... 24
Descriptive Statistics of Variables ... 25
Descriptive Statistics for Quota or no Quota imposed ... 26
Pooled OLS and Firm Fixed Effects Regression ... 29
Moderation with Fixed Effects and Pooled OLS ... 37
Statutory Gender Quotas and Gender Diversity Recommendations in Corporate Governance Codes in Europe ... 47
Variable Definition ... 47
Categorization of Firms into Industries ... 48
Sample of Firms ... 48
List of Figures Fig. 1 The links between Gender Diversity and Firm Performance ... 19
List of Abbreviations
EIGE European Institute for Gender Equality
EU European Union
OECD Organization for Economic Co-Operation and Development
OLS Ordinary Least Squares
ROA Return on Assets
ROS Return on Sales
5 1 Introduction
“Gender equality is not only a fundamental human right, but a necessary foundation for a peaceful, prosperous and sustainable world”. This is the 5th out of 17 Sustainable Development
Goals of the United Nation (United Nations, 2020). Gender equality has not only been one of the top priorities of worldwide organizations such as the United Nations and the OECD but also received high attention in international policy forums over the recent years. The topic is increasingly argued to not only be a moral imperative but also key to economic growth. A report by Goldman Sachs (2019) for instance examines the increase in female labor participation to 71%, and reports that listed firms with higher female manager ratios tend to deliver higher Return on Equity (ROE) and sales growth (Matsui et al., 2014). Moreover, Löfström (2009) shows a positive correlation between GDI/GEM (Gender-related Development Index and Gender Empowerment Measure, accordingly) and GDP in European countries, suggesting that enhanced gender equality in the labor market increases economic growth.
Organizations such as the European Commission base a great share of their arguments for gender equality on an economic rationale. While the European Commission is to date still reporting a Gender Equality Index score of 67.4 out of 100, with an increase of only 5.4 points during the last 15 years, the Commission is dedicated to establish this fundamental value (EIGE Gender Equality Index, 2019). One demand established by the European Commission to reach this goal has been gender balance on company boards. Based on this, a in 2012 proposed directive to introduce a 40% quota for underrepresented women in boards of directors (European Commission, 2012), initiated a rising debate about the topic. Even though the Council of the EU rejected the proposed directive by a qualifying majority, several European countries have enacted gender quotas since.
Extensive research has been focusing on the relationship of women on boards and firm performance and firm value (i.e. Terjesen et al., 2016; Isidro & Sobral, 2015; Carter et al., 2010; Gul et al., 2011; Byron & Post, 2015). As this literature provides mixed evidence about the impact greater gender diversity has on firm performance, and has not yet focused on the role of mandated quotas in this relationship, the goal of this paper is to offer fact-based insights on the economic rationale behind gender quotas and whether the reasoning behind most quota based policies is appropriate.
Findings by Post & Byron (2015) argue that the legal environment, such as strong shareholder protection and gender parity enhances the positive effect on accounting returns and
6
market performance respectively. Besides that, a paper by Leszczyńska (2018) suggests that mandatory quotas might impose burdens and hence negatively affect a company’s performance, which makes it crucial to research alternative measures to efficiently introduce more women on boards. For these reasons, this paper will be focusing on multiple European countries and contrast the legal nature of each countries’ gender quotas and its effects on the relationship between female directors and firm performance. Hence, this paper follows a twofold purpose. First, I aim to empirically examine the effect of female directors on firm performance. Second, I distinguish between statutory gender quotas and recommendations by countries effecting the relationship between women on boards and firm performance.
As the number of countries implementing a quota has been growing over time, I collect data from 8 different European countries over the period 2010-2019 on a yearly basis for 156 firms. The database consists of 1,560 observations of 17 variables, resulting in a total of 26,520 observations. I collect market data through Bureau van Dijk’s Orbis database.
Due to limited database access, I achieve the assembling of board related data through the study of 1,560 firm-year based annual reports. These reports are examined to collect data on 5 variables, namely board size, the percentage of female directors, an indicator variable, identifying whether a minimum of 30% female directors is reached, on CEO/Chair Duality, particularly whether the executive management acts as the chairman of the board and identifying whether an individual firm has a supervisory board or not. As annual reports are often not yet focusing on gender diversity, observations on gender diversity are selected through counting names and or pictures of individual directors to identify the gender of each. In total 7,800 observations are hand-collected.
The selected dataset is firstly used to resolve the effect of gender diversity on firm performance. Within this relation, gender diversity is measured as the percentage of female directors and the above-mentioned indicator variable. Firm performance is combined of Tobin’s Q, a measure of firm value, and ROA, an accounting-based measure. Through focusing on a comparison of different analysis to underline the importance of endogeneity and reverse causality problems within this branch of research, this study is able to obtain unbiased result of the relationship in question. Specifically, I firstly focus on an Ordinary Least Squares (OLS) regression of gender diversity on firm performance, implementing three different stages: (1) accounting for industry related effects with the implementation of industry dummies based on two-digit SIC codes, (2) adding year dummies to the model, to control for yearly differences, and (3) further including country dummies to control for country specific differences.
7
Next, I make use of the panel structure of my data and use a firm fixed effects regression to control for unobserved time-invariant heterogeneity of individual firm characteristics. This analysis is also be approached in two steps. First, no time fixed effects are taken into account, while these are implemented in a second step.
Lastly, I run a Two-Stage-Least-Squares (2SLS) regression to account for reverse causality. This is again performed through including both a baseline OLS regression and a fixed effects regression, to achieve a unique set of comparisons. Implementing such a complex analysis, this study is able to highlight omitted variable bias, as well as the problem of biased results due to endogeneity and reversed causality, within a broad European sample over a significant period of time.
The second question addressed in this paper, focuses on the interaction effect of statutory gender quotas on the afore investigated relationship of gender diversity on firm performance. This question is analyzed through the use of a moderated OLS regression accounting for industry, time and country fixed effects, and a firm fixed effects model including year dummies.
In a panel of 156 publicly listed European companies from 2010 to 2019, I find that gender diversity has little to no impact on firm performance, measured by both Tobin’s Q and ROA, once accounted for problems of endogeneity and omitted variable bias. Considering the comparison of results within the first model of interest, it becomes evident that the implementation of omitted variables, shows differences in results, especially considering the direction of the coefficients on gender diversity. Once controlling for firm fixed effects including year dummies, the estimated coefficients of gender diversity variables show a negative causal relationship to firm performance.
Concerning the moderating impact of mandated gender quotas, although there is no statistically significant relationship, it can be argued that gender quotas affect the relationship of gender diversity on firm performance negatively, as coefficients are negatively connotated.
These results are important, as they challenge the statements made earlier literature and particularly organizations such as the European Commission, refuting an economic rationale within the gender quota debate. Neglecting endogeneity problems within the relationship between gender diversity and firm performance, leads to distorted implications which can result in decision-making based on false grounds and hence jeopardizes the incentives behind an achievement of gender diversity within the upper echelon.
Ultimately, my analysis contributes to two fields of research: corporate governance and gender-based policies. Its primary contribution is to present evidence on a wrongly established
8
economic rationale within the field of gender diversity in boardrooms and to secondly shed light on mitigating factors reflected by time and constraining policies such as gender quotas.
Other papers, examining the impact of female directors on firm performance, have been mostly focusing on single country samples, such as the U.S. (Adams & Ferreira, 2009), Spain (Campbell and Mínguez-Vera, 2007), and Norway (Ahern & Dittmar, 2012). Even though papers like Isidro and Sobral’s (2015) and Lending and Vähämaa’s (2016) focus on a European sample, they do not research significant and recent time periods and the interaction effect of mandated gender quotas or governmental comply-or explain recommendations.
To the best of my knowledge, this paper is the first paper directly examining in depth analytical differences when investigating the relationship of women on boards and firm performance, as well as focusing on a European sample over the past ten years, comparing countries with gender quotas in place and countries only engaging in comply-or explain recommendations.
The remainder of the paper is organized as follows. Section II provides a profound review of existing literature, examining the role between gender diversity and firm performance, while critically questioning the effectiveness of gender quotas and setting the empirical research question. Section III describes the data and methods used in this empirical study. Section IV introduces the results of my analysis, and ultimately Section V discusses the established findings and concludes.
2 Literature Review
The importance of increased board diversity has been rising worldwide. Catalyst (2020) reports a global increase in the percentage of women on boards, yet states that women still remain underrepresented and overall progress is slow. Board diversity is not only argued to achieve equality but to also have economic benefits. While it is argued that, an economic reasoning in this relationship is necessary as a board of director’s foremost task is to maximize wealth through safeguarding the success of the firm and increasing shareholder benefits (Choudhury, 2015). Choudhury nevertheless stresses, that an economic justification through female director’s positive impact on firm financial performance does not hold, as empirical evidence is too divided.
A study by Ahern & Dittmar (2012) focusing on the introduction of gender quotas in Norway, for instance demonstrates that the constraint imposed by a mandatory quota negatively affects firm value. The researchers substantiate their findings with a comparison of personal
9
characteristics of directors and a deterioration in capabilities of the board. From there, the authors suggest that quotas are effective at improving gender diversity, yet impose significant cost to shareholders if a lack of experience in female directors is evident.
To take a closer look at the impact of gender quotas around Europe through quantitative analysis in this paper, I initially analyze the present relationship theoretically and go further by discussing the existing empirical research.
2.1 Board Diversity and Firm Performance
Improving gender diversity within the corporate world, has not only been argued at an ethical level, through the idea of equal opportunities and equal representation of genders, but also has gained a lot of attention through arguments based on economic benefits. Based on this, some of the most relevant theories are discussed to investigate the relationship between board diversity and firm performance further.
Considering the board of directors, agency theory is the most frequently mentioned theoretical framework when linking board diversity and firm value or firm performance. Jensen and Fama (1983) suggest agency problems arise once decision managers do not bear any of the wealth effects resulting from an undertaken decision and therefore stress the necessity for a board to control and monitor managers. Consequently, board composes an independent function to act in the best interest of the shareholders to ensure separation of decision management and control.
Even though research has been conclusive regarding the effectiveness of inside and outside directors in board compositions (e.g. Baysinger & Hoskisson, 1990) and Adams and Ferreira (2009) note that female directors more frequently have a similar role to that of independent directors, which leads to a higher board effectiveness, no valuable evidence could be provided that gender diversity increases firm performance.
In particular from an agency-theoretic standpoint, the generalized rationale that women bring a fresh perspective, thus improve problem solving, seems to overlook the several duties of the corporate board. By pointing out various studies on female directorship’s impact on different tasks, Francoeur et al. (2008) conclude that the overall impact of gender diversity on corporate governance and hence financial performance is impossible to anticipate.
In light of this, it is however also important to determine what good corporate governance and thus the corporate board is supposed to achieve. If the main purpose of good corporate governance is to increase firm performance, and such is not achieved, then increasing
10
the number of female directors rather has symbolic than practical value (Brown et al., 2002). The symbolic value of women can be viewed in consideration of stakeholder theory, stating the appointment of women as directors is caused by the pressure of shareholder activists, large institutional investors, politicians and consumer groups (Francouer et al., 2008). Based on this, stakeholder theory does not exclude the financial benefits of women on boards, but stresses the promotion of women regardless of improved financial performance as good policy (Francoeur et al., 2008).
As another important theory in this context, resource dependency theory (Pfeffer, 1972; Pfeffer & Salancik, 1978), emphasizes the interdependence of organizations and their external environment and the benefits firms can accrue from three specific linkages. As legitimacy is one of the linkages pointed out by Pfeffer & Salancik (1978), and is directly given by a highly visible function of the board of a corporation, societal pressure to more diverse leadership also plays a significant role here. Hillman et al. (2007) point out, that including women on the board might strengthen the reputation and credibility of a firm in both internal and external labor markets, and subsequently holds the potential to produce unique information sets available to management for improved decision-making (García-Meca et al., 2015), creating a wider range of channels for communication with external organizations (Hillman et al., 2007).
Another theory relevant for the relationship between gender diversity and firm performance is human capital theory (Becker, 1964). Human capital theory is concerned with higher investments in education and knowledge as resulting in development of independent thinking, which is stated as a key characteristic required for non-executive directors (Roberts et al., 2005). As consistent findings in previous research has shown that female directors have high levels of education, Singh et al. (2008) provide empirical evidence that human capital theory plausibly explains why a particular set of women are newly appointed to corporate boards. Particular international diversity such as the possession of an MBA degree, differentiate women from men appointees. In this regard, it can be clearly rejected that women lack adequate human capital and can bring a high degree of human capital resources to corporate boards.
In the next section, I will turn on policy’s, such as quota’s, effectiveness and impact on women on boards and organizational performance.
11 2.2 Gender Diversity Quotas
Generally, it is widely agreed upon a need for gender equality in the corporate upper ranks, yet there is substantial discordance regarding the implementation of quota-based policies. An empirical paper by Terjesen and Sealy (2016), reviews the extant literature on the debate around gender quotas and outlines future actions. The researchers structure their paper based on motivations, legitimacy and outcomes to get a multi-level understanding of the ethical dilemma and tensions around quotas. Considering a business perspective, it is on the one hand emphasized that a quota as a radical change agenda, can be poorly informed and leading to impulsive responses. On the other hand, it is argued to effectively leverage female talent, enhance creativity, innovation and decision making.
A quota-opposing argument has been substantiated by for instance findings of Ahern & Dittmar (2012), who investigated the pre- and post-quota effect on affected firm’s Tobin’s Q and stock prices. Their findings show an instant drop in stock prices after the Norwegian quota announcement. Additionally, the study demonstrates that the constraint imposed by a mandatory quota negatively affects firm value. The researchers substantiate their findings with a comparison of personal characteristics of directors and a deterioration in capabilities of the board. From there, the authors suggest that quotas are effective at improving gender diversity, yet impose significant cost to shareholders if a lack of experience in female directors is evident.
Critical mass theory, as social identity theory however suggest the necessity of a threshold value to be reached to shift group dynamics, and create substantive rather than symbolic change (Kanter, 1977; Joecks et al., 2015; Konrad et al., 2008). Next to that, Torchia et al. (2011) studied the effectiveness of female directors on firm organizational innovation through their involvement in board strategic task, focusing on the critical mass theory. Their findings stress the importance of women being a consistent minority (at least three) in the boardroom to have a positive impact on firm organizational innovation, and lead to a change in perceptions of female directors’ legitimacy, which could support a necessity for gender quotas to secure a certain critical mass in boardrooms.
Apart from this, Terjesen and Sealy (2016) discuss whether gender quotas resolve the present gender inequality. The scholars find that the Norwegian quota increased female directors and political leadership but did not lead to an increase in females at lower positions in corporations nor improve the gender pay gap. They also underline, that some countries’ voluntary systems led to the desired increase in women on boards and that a prolonged cultural change is more likely assured through voluntary measures, such as recommendations than
12
through statutory rules. As gender quotas do not lead to an increase in women in executive leadership positions, which reflects a direct interplay with the availability of qualified female directors, Adams and Ferreira (2009) state that female board members elected based on an imposed quota, lacked experience and were younger than their male counterparts. This resulted in more acquisitions, worse accounting returns and fewer takeover defenses, decreasing overall firm performance. Related to this, Beaman and et al. (2009) remark that a gender quota could backlash and reinforce negative stereotypes in case unqualified women are recruited resulting from a limited supply.
Finally, a study by Leszczyńska (2018) examines the Impact Assessment issued by the European Commission in regard to the directive of a 40% statutory female quota on corporate boards in European publicly listed companies. The scholar challenges the assumption made by the European Commission that increased gender diversity improves corporate governance and company performance. The paper thereby focuses on the means of how gender diversity is achieved, rather than the direct relationship between gender diversity and firm performance. In this context, the author illustrates that mandated quotas might impact the overall functioning of the board. Cooperation between incumbent and new members might be negatively affected by a statutory quota, as has been found in an experiment by Mollerstrom (2012). This reduced cooperation remains even if participants are given a rationale by appealing to a fairness argument. It is therefore important to evaluate the negative side effects mandated quotas might bring to board and company performance and consider those when stressing an economic rationale in a gender diversity debate (Leszczyńska, 2018).
2.3 Empirical Findings
Previous research on the relationship between gender board diversity and firm performance has provided mixed evidence, which could be explained by differences in research settings.
Adams and Ferreira (2008) investigate a sample of US firms and stress that increased gender diversity in the boardroom leads to tougher monitoring and more incentive alignment, which relating to agency theory, means the stronger governance increases shareholder value. Through an analysis accounting for the potential endogeneity of gender diversity, they however conclude that an over monitoring of gender-diverse boards decreases value and thus averagely does not add to firm value or increase performance. The researchers therefore argue, that there is no present evidence that quota-based policies would improve firm performance.
13
A paper by Campbell and Mínguez-Vera (2007), examines a sample of Spanish firms, motivated by the focus on a civil law system compared to earlier research focusing on common law countries (such as Adams & Ferreira, 2008). The researches find a positive relationship between board diversity (measured by the percentage of women and by the Blau and Shannon indices) and firm value (measured by Tobin’s Q), implying that an important focus for Spanish companies is the balance between men and women rather than the sole presence of women.
Compared to the above-mentioned research, García-Meca et al. (2015) focus on the effect of board diversity (gender and nationality) on performance in banks across nine countries. The cross-country analysis allows the researches to account for environmental and institutional differences, which increases external validity. The authors find a positive relationship between gender diversity and bank performance, which they suggest is caused by improved governance.
As another important empirical paper, Isidro and Sobral (2015), investigate the direct and indirect effects of women on boards on firm value, in a cross-country sample, focusing on 16 European countries. The scholar’s empirical findings suggest that greater female director representation can enhance firm value (measured by Tobin’s Q) through better financial performance (measured by ROA and ROS) and greater ethical and social compliance.
In contrast to Isidro and Sobral’s study, Ahern and Dittmar’s (2012) paper measures a pre-post quota effect of Norwegian firms on firm value (Tobin’s Q). This shows a negative effect of a 40% quota on firm value. The researchers substantiate their findings with a comparison of personal characteristics of directors and a deterioration in capabilities of the board. From there, the authors suggest that quotas are effective at improving gender diversity, yet impose significant cost to shareholders if a lack of experience in female directors is evident. Moreover, Lending and Vähämaa (2017), take into account the changes in the nominating process of directors through large shocks such as the enforcement of gender quotas. The sample includes European countries with and without quotas, comparing a set of Nordic companies to southern European companies. The relationship the researchers measure is however focused on female representation, quotas and their impact on board structure and expertise. Their findings differ per region and specific board characteristics. As expected, the scholars find a greater percentage of women on boards in countries with either pending or required quotas, Nordic countries do not show an effect of female directors on board expertise, and also the imposition of quotas does not affect board expertise. In southern European countries, board expertise however increases with a higher female percentage of board members. In contrast to that, a pending quota decreases board expertise in southern Europe.
14
The paper therefore suggests that the country in which the company operates affects board structure beyond legislated quotas, as gender equity differs greatly in Southern compared to Nordic Europe.
2.4 Hypothesis
As studies have mainly been focusing on individual countries, when examining the impact a quota has in terms of economic justifications, this study tries to shed light on the economic differences between voluntary governmental recommendations and mandated quotas in regard to gender diversity and firm performance. Following these empirical findings, there is a need to understand the interaction between the nature of measures taken to increase gender diversity on boards and firm performance to be able to justify gender diversity on economic means.
More specifically the following research question needs to be addressed:
“Are mandatory quotas economically effective in increasing gender diversity in boardrooms?”
This paper contributes to the existing literature in different ways. First, it tries to resolve the ambiguous relationship between board gender diversity and firm performance, measured by both firm value and financial performance, through looking at a cross-country sample. Second, my analysis makes an important contribution to European corporate governance literature, through focusing on country-level quota regulations and whether these impact the relationship between female directors and firm performance. Through the investigation of the moderating effect of quota-based policies, this study can shed light on not only corporate governance issues but also the debate within public policy.
It finally helps to understand how to achieve an increase in gender diversity on boards more effectively without jeopardising economic outcomes such as firm performance. This would provide major evidence for political decision making and economic influence governmental initiatives might have.
The research will be conducted through focusing on two stated hypotheses:
H1: More gender diverse boards are associated with better financial performance. H2: A mandatory gender quota weakens the effect between a higher percentage of women on
15 3 Data and Methodology
3.1 Population and Data Description
The sample used in this study consists of 1,560 observations for 156 listed companies (see Tabel A.4) in eight European countries (namely Norway, France, Italy, Germany, Sweden, Finland, Denmark, United Kingdom). To ensure a significant and recent time period, the data is collected over 10 years (2010-2019), although at times there is no available information for some firms, resulting in an unbalanced character of the sample.
The initial sample includes the ten largest and midsize companies by market capitalization of each country. Financial firms , except Real Estate Firms, will be excluded, as the role of their boards might be constraint through special inspection by financial authorities and due to their particular accounting practices (Pucheta-Martínez, 2018; La Porta et al., 2002). This sample ensures the scope of the research and external validity.
The data have been collected through several means (Table 1). Accounting data and Tobin’s Q were collected through Bureau van Dijk’s database Orbis, and is reported in U.S. Dollars to account for differences in currencies, whereas board related data has been hand-collected, due to no access to WRDS’s BoardEx database. Even though the Orbis database provides a comprehensive range on detailed firm information, data on historical board composition is not available. Therefore, the collection of board related data has been conducted through internet research, focusing on annual reports to identify the Board Size, the number of female directors, and whether CEO/Chair Duality is apparent. To gather this information, I firstly search for a photograph of board members in the annual report to identify the gender of board members, if such is not available and no information is given regarding the gender of the board member (such as Miss, Mr, Mrs), I search the names individually to reveal the gender of the respective person. In case the exact name did not lead to feasible solutions, I decided on gender given by the frequency of one gender assigned to the given name. Orbis was additionally used for information on the firm’s year of incorporation (to compute Firm Age) and SIC industry classifications. As an indication of the present status on quota legislation, I constructed a table which highlights the legislative differences per country, the table is provided in Appendix Table A.1
16 Table 1
Data Selection Process
Years 2010-2019
Countries Germany, Norway, Italy, Sweden, France, Finland, Denmark, United Kingdom
Variables Obs Source
Quota 1560 see Table A.1; Arndt & Wrohlich (2019), Ahern & Dittmar (2012), European Commission(2016) Female 1483 Annual Reports of individual firms from 2010-2019 IndFemale 1483 Annual Reports of individual firms from 2010-2019 BoardSize 1483 Annual Reports of individual firms from 2010-2019 BoD/Supervisory 1494 Annual Reports of individual firms from 2010-2019 CEODuality 1489 Annual Reports of individual firms from 2010-2019
ROA 1464 Bureau van Dijk’s Orbis
Tobin’s Q 1414 Bureau van Dijk’s Orbis FirmSize 1465 Natural Logarithm of Total Assets Leverage 1462 Long-Term Debt / Total Assets TotalAssets 1465 Bureau van Dijk’s Orbis LongTermDebt 1462 Bureau van Dijk’s Orbis
SIC 1560 Bureau van Dijk’s Orbis
FirmAge 1543 Year of Incorporation – current year Year of Incorporation 1560 Bureau van Dijk’s Orbis FSeniorMiddleMgmt 1404 The World Bank Data Bank FIndustry 1560 The World Bank Data Bank Total 25,351
Missing Values 1169
3.2 Variables
The dependent variable, Firm Performance, will be measured as Tobin’s Q as a market valuation indicator and Return on Assets (ROA) as an accounting based indicator. Both measures have been extensively used as a proxy for firm performance (Isidro & Sobral, 2015; Adams & Ferreira, 2009; Post & Byron, 2015) and complement each other (García-Meca et al., 2015). Tobin’s Q (Tobin’s Q) provides a measure of the market’s expectation of future firm performance and is defined as the sum of total assets minus the book value of equity plus the
17
market value of equity, divided by total assets (Terjesen et al., 2016). Firms reporting a Tobin’s Q greater than one are expected by investors to have a greater worth than its book value, while a value smaller than one suggests insufficient utilisation of available resources by the firm (Campbell & Mínguez-Vera, 2007). In comparison to Tobin’s Q, ROA focuses on past financial performance of the firm and does only reflect financial performance captured by accounting rather than other aspects of the business. The use of both measures is chosen to on the one hand account for possible market anomalies that may constraint available information being reflected in the stock price, on the other hand to include both financial and non-financial effects of the presence of women on boards (Isidro & Sobral, 2015).
The first independent variable Gender Diversity is measured as the percentage of female directors (Female), calculated as the number of women divided by total number of directors times one hundred. Gender Diversity is additionally measured as an indicatory variable coded 1 if there are at least 30% of female directors and 0 otherwise (IndFemale) (Lending & Vähämaa, 2017; Carter et al., 2003). The additional indicator variable is chosen by drawing on critical mass theory, which argues that a certain threshold has to be reached for minorities to act more distinctively within groups (Kanter, 1977). Particularly, Torchia et al. (2011) find a critical value of three women on boards to benefit from the diversity in perspectives, backgrounds and skills in the boardroom. As the present study however entails large variations in board size, ranging from 3 to 27 members, a critical mass of 3 women on boards would have significantly different effects considering board size. As proposed by Isidro and Sobral (2015), I use a related indicator of 30%. To resolve possible nonlinearity of Female, I additionally square Female (as sqFemale) and add this variable to the model.
Quota, as the moderating variable, is measured as an indicator variable on country-level and is coded 1 for statutory quotas and 0 for recommendations per country.
To avoid coefficient bias and account for omitted variable bias, I implement a set of control variables, which have previously shown an effect on firm performance. Firm Size, approximated by the natural logarithm of total assets, is a typical representative of firm value and performance. With reference to earlier findings, one can expect a negative relationship between the firm size and firm value and financial performance (Adams & Ferreira, 2009; Carter et al., 2010).
As firm age increases, firms might lose their flexibility and deteriorate into organizational rigidities which lead to declining R&D activities, rising costs and slowing growth. In line with this, Loderer and Waelchli (2010) report a negative relationship between
18
firm age and performance. Based on this, age (lnAge), as the logarithm of the number of years of the firm, is included as a control variable.
The capital structure of firms has been a widely discussed topic after Modigliani and Miller’s irrelevance proposition (Modigliani & Miller, 1958). Even though the relationship between leverage and firm performance continues to report controversial findings, it is necessary to account for a correlation between those variables in the present study (Jermias, 2008). Leverage is included as the ratio of long-term debt to total assets. As board related control variables, I include Board Size (lnBoardSize) and CEO/Chair Duality (CEODuality).
Considering the argument that larger boards facilitate key board functions, it has contrarily been noted that too large boards suffer from coordination problems which lead to ineffectiveness and hence to a decline in firm performance (Guest, 2009), which indicates a negative relationship of board size and firm performance once a certain threshold is exceeded. In addition to the board size alone, implementation of an independent chair is an important factor of board structure for instance in view of agency problems. The independence of the board is chosen to separate decision management and decision control to align conflicting interests of shareholders and top management (Boyd, 1995). Nevertheless, do Boyd’s findings indicate a slightly positive relationship between CEO/Chair Duality and firm performance, which enhances the necessity of implementing it as a control variable.
To comprise industry specific trends in financial performance, industry indicator variables are included. Furthermore, Hillman et al. (2007) argue that the nature of an industry is likely to affect the benefits of female representation. Hence, it is necessary to add an industry indicator variable based on a three-digit standard industrial classification (SIC) to control for cross-industry differences (Isidro & Sobral, 2015).
To account f reverse causality in the model, a 2SLS regression is applied. For running a 2SLS, the model requires an instrumental variable, which is correlated with the fraction of female directors on the board, but is exogenous to firm performance. In the context of governance regressions, it can be difficult to find a valid instrument, as factors which are argued to be most correlated with the regressor are other characteristics already included in the model (Adams & Ferreira, 2009). In line with Marinova et al. (2016), I will make use of the percentage of women within present industries in my sample per year as an instrumental variable. A table including all variables and definitions is provided in the Appendix Table A.2.
19 3.3 Empirical Methodology
The aim of this paper is to firstly investigate the impact of female directors on corporate boards on firm performance, and secondly to identify the moderating effect of statutory gender quotas on this relationship.
The relationship is graphically shown in Figure 1. The first link illustrates the relationship between women on boards and firm performance, while the second link is representing the impact of gender quotas on the first relationship.
Fig. 1 The links between Gender Diversity and Firm Performance
The first relationship reflected in Figure 1, is estimated by the following regression model: !!"/#$%!" = (!"+ ∑(! +,-./01-23!"$%+ ∑(! 45!"$%+ 6! + 7"+ 8!"
where firm performance is measured by Tobin’s Q or ROA (Qit/ROAit) as described in Section
III.2. for firm i in year t. GDiversity is measured by a binary variable, equal to one indicating a minimum of 30% female directors (IndFemale), the percentage of board members that are female (Female), and the percentage of female directors squared (sqFemale) for firm i in year t. The control variables, specified by CV, are FirmSizeit, Ageit, Leverageit, BoardSizeit,
CEODualityit and SICit, where the latter two are binary variables. One-year lag is taken for
FEMALE and all control variables to account for reverse causality. Furthermore, 72 and 6- are used to control for time- and firm fixed effects respectively.
As the second hypothesis is focusing on the moderating effect of statutory quotas on the relationship between firm performance and female board representation, the second relationship is estimated by the following model:
!!"/#$%!" = (!" + ∑(! +,-./01-23!"$%+ (! !9:2;!"$% + ∑(! +,-./01-23!"$%!9:2;!"$%+ ∑(! 45!"$%+ 6! + 7"+ 8!" Gender Diversity Firm Performance Gender Quota H1: + H2: -
20
where βj ∑GDiversityitQuotait reflects the interaction between Gender Diversity and Quota.
The first formulated hypothesis in Section II will be investigated by the means of a Pooled OLS as a simple baseline model. I examine the relationship further using panel data regression. This enables me to assess performance in the sample of different entities over serval consecutive years and to differentiate between companies and their imposition to gender quotas over time. Moreover, the use of panel data allows to eliminate any unobservable heterogeneity that may be present among firms. As a correlation between unobservable heterogeneity with the regressor would bias the obtained results, panel data allows me to undertake a conditional inference, estimated by fixed effects (Himmelberg et al., 1999). If such correlation is not visible, unconditional inference can be carried out using the random effects method (Campbell & Mínguez-Vera, 2007).
To test these assumptions, the Hausman test (Hausman & Taylor, 1981) is used. This test compares the equality of the coefficients with both fixed effect estimations and random effect estimations. As the null hypothesis states, that the coefficients of both models are similar, rejecting this hypothesis implies a significant difference between estimated coefficients with only the fixed effects estimation being consistent.
As the first regression, estimating the effect of Gender Diversity (IndFemale/Female/sqFemale) on Tobin’s Q, reports a p-value of the Hausman test of 0.0046, 0.0002 and 0.0010 respectively, I can reject the null hypothesis at a 5% significance level, suggesting systematic difference between estimators. The same procedure yields a p-value of 0.1471, 0.1423 and 0.1235 for ROA as a dependent variable accordingly. This means the null hypothesis cannot be rejected and a random effects estimation could be used.
Considering the specific problem of my research question, one could however assume that the firm specific effects are correlated with the independent variables, which would violate one of the main assumptions of the random effects model (Wooldridge, 2012). For simplicity and the strong assumptions a random effects model is based on, I perform a fixed effects model for both dependent variable Tobin’s Q and ROA.
Endogeneity and Reverse Causality. As Hermalin and Weisbach (2001) emphasize endogeneity problems in examining board composition and firm performance, an Ordinary-Least-Squarea method could produce biased coefficient estimates. Such endogeneity problems can arise through omitted variables that affect both the fraction of female directors and firm
21
performance. The authors note that, firm performance results both from actions of previous directors and is itself a factor which has potential to influence future board composition.
Additionally, considering reverse causality, an increased number of women on boards might not only affect firm performance but could also be caused by better performing firms, as those firms might be more likely to hire female directors but also directly incentivize women to join them (Adams & Ferreira, 2009). To control for the problem of possible endogeneity of variables, the estimation is carried out through a two-stage least-squares model in addition to using one-year lags of the endogenous regressor and control variables to account for reverse causality. Performing a 2SLS, an instrumental variables (IV) estimator is needed. By using an instrumental variable, the unobserved correlation can be identified, which enables me to see the true correlation between dependent and explanatory variables. For an instrument to be valid, it needs to be relevant, hence correlated to the instrumented endogenous variable, and exogenous, thus uncorrelated with the error term.
To specify whether the instrument is strongly correlated with the endogenous variables, I run the first stage regression to see whether the share of women in industry is statistically significant and whether the F-statistic is large enough (rule of thumb >10) for the instrument to be relevant (Stock & Staiger, 1997). Making use of the instrumental variables regression, I implement the percentage of females per industry per country and year, and the percentage for women in senior or middle management positions per country per year, as instrumental variables.
The equation of the first-stage regression is as follows:
</=;>/!"$% = (!"+ ∑(! ?5!"$%+ ∑(! 45!"$%+ 6! + 7"+ 8!"
where IV reflect the instrumental variables used.
The second stage of the 2SLS model, reflects the actual relationship this paper aims to examine. A replacement of all explanatory variables reflecting Women on Boards is made by the fitted values of the first-stage regression.
22 4 Results and Analysis
4.1 Descriptive Statistics
Table 2 reports descriptive statistics for women on corporate boards across the countries in the sample, and Table 3 reports across time. It becomes evident, that an overall average of 27.3% female directors is apparent in the sample, which confirms that women are still a minority and do not pass a critical value of 30% yet. Not surprisingly, the fraction of female directors is the largest in Norwegian firms with an average of 38.7%, which are inflicted by a 40% quota for the entire sample period. The share of female directors does not necessarily seem to be larger for countries imposed by a quota in applicable years. It is however notable that Nordic countries, in particular Sweden and Finland have considerably high number of women on boards with almost 30% on average without being subject to a mandated quota. On the contrary is Germany’s share of female directors significantly lower, while the country’s publicly listed companies are subject to a gender statue of 30%. This low mean of women on boards could be explained by the fact that non-compliance is not directly pursued by sanction, but “just” an empty chair (see Table A.1). As the minimum proportion of female directors is zero, it can be assumed that in most European countries, there are still major obstacles for women to reach the boardroom (Isidro & Sobral, 2015). These reported results are in accordance with results published by the European Women on Boards in their 2019 report. The organization reports an average of 33% of female board members in STOXX Europe 600 companies. The report stresses that a significant increase took place during the recent years, but there is still a lack of gender equality on other levels of corporate governance. Considering the reported results, one can clearly see a continuous growth in the share of women on corporate boards from 2010 until 2019, especially Italy and France experienced an increase of around 30% in female directorship. It is yet noteworthy that Italy and France both passed quota legislation with sanctions in 2011 (see Table A.1) and the increase can hence be concluded as a direct result of such mandated quotas. Countries only engaged in voluntary initiatives such as Sweden, Finland, the United Kingdom and Denmark, still have an observable increase of women on boards during the last 10 years. In connection to that, the frequency of firms with 30% or more female directors is not necessarily higher for countries imposed by a quota. In particular, examining the frequency of Germany (26.8%), Italy (44.44%) and France (54%) as quota mandated countries, compared to the frequency of countries such as Sweden (50%), Finland (53.1%), and United Kingdom (26.6%), points out that statutes to achieve a critical mass might not be absolutely needed (Tabel 5).
23 Table 2
Summary Statistics of the Fraction of Female Directors per Country
Country N Mean Std. Dev. Min Max
DE 190 .202 .114 0 .444 DK 187 .215 .111 0 .5 FI 192 .293 .102 .091 .5 FR 189 .32 .144 0 .636 GBR 173 .208 .133 0 .6 IT 189 .241 .141 0 .467 NO 195 .387 .075 0 .571 SE 168 .309 .142 0 .667 Total 1483 .273 .137 0 .667 Table 3
Summary Statistics of the Fraction of Female Directors per Year
Year N Mean Std. Dev. Min Max
2010 138 .183 .138 0 .583 2011 139 .2 .137 0 .583 2012 141 .218 .137 0 .583 2013 146 .235 .132 0 .545 2014 150 .263 .122 0 .583 2015 152 .281 .117 0 .583 2016 152 .309 .123 0 .636 2017 155 .334 .119 0 .667 2018 156 .337 .112 0 .636 2019 154 .344 .11 0 .667 Total 1483 .273 .137 0 .667 Table 4
Frequency of Firms with 30% or more women on boards per Year
Year N Mean Std. Dev.
2010 138 .217 .414 2011 139 .252 .436 2012 141 .305 .462 2013 146 .329 .471 2014 150 .427 .496 2015 152 .48 .501 2016 152 .533 .501 2017 155 .619 .487 2018 156 .699 .46 2019 154 .734 .443
24 Table 5
Frequency of Firms with 30% or more women on boards per Country
Country N Mean Std. Dev. Germany 190 .268 .444 Denmark 187 .267 .444 Finland 192 .531 .5 France 189 .54 .5 United Kingdom 173 .266 .443 Italy 189 .444 .498 Norway 195 .887 .317 Sweden 168 .5 .501
Table 6 presents descriptive statistics of the variables used in the empirical estimations. The mean Tobin’s Q being greater than one indicates investors’ expectations to be positive about the future value of the firm, in particular that the invested resources will generate future value. As the mean is closer to the minimum of 0 than to the maximum of 13.99 one can expect outliers in the sample.
The mean ROA is 5.3%, with a minimum value of -52.95% and a maximum of 51.25%, a large variability of profitability between firms can be concluded. As discussed above, the fraction of women on boards has a mean of 27.3%, with a maximum of 66.7% female directors and firms with no female directors in their boardrooms. As indicated by the binary variable reflecting boards with a minimum of 30% females, on average 46,7% of times firms achieved such a critical mass. The quota variable is also constructed as an indicator variable and coded 1 for a quota apparent in country/company in a certain year and 0 if otherwise. With a mean of 38.8%, more than one third of our sample over time is subject to such a gender mandate, which reports the growing number of countries implementing gender quotas over time. Firm Size is measured as the natural logarithm of total assets. Size has an average of 6.77 and with the smallest and largest value being 2.74 and 8.74 respectively. The mean age of companies in my sample is 61.65, with value ranging from 0 years, which reflects companies which were not yet incorporated in a given year, to 365 years, which is reflected by a certainly high standard deviation (54.37). The mean leverage is 0.185, being in line with values reported by Lending and Vähämaa (2017), who focus on Southern as well as Nordic European countries, and find mean leverage values of 0.19 and 0.16 respectively.
The number of directors on boards ranges from 3 to 27 directors, with a mean of 10.33. This number is possibly increased by a concentration of larger boards in countries such as Germany, France, and Italy, reporting a mean value greater than 10. As CEO/Chair Duality is measured as a binary variable, the mean reflects the fraction of firms over time having a CEO
25
as the chairman of the board. In my sample on average 11.1% of the boards are concerned by this. To account for values largely deviating from a preferred skewness and kurtosis of zero and three accordingly, I computed the natural logarithm of Age and Board Size to use the variables in my regression analysis. Although, the kurtosis of Tobin’s Q and ROA are also higher compared to other variables which indicates large outliers and a non-normal distribution of the variables, the large sample size (1,560) and the use of robust standard errors, satisfies OLS assumptions and one can assume an asymptotic standard normal distribution based on the Central Limit Theorem (Stock & Watson, 2015).
Table 6
Descriptive Statistics of Variables
Variables Mean Median St.Dev Min Max Skewness Kurtosis N Tobin’s Q 1.318 .811 1.588 0 13.996 3.435 18.828 1414 ROA 5.336 4.183 7.727 -52.946 51.248 .309 13.507 1464 Female .273 .286 .137 0 .667 -.199 2.696 1483 IndFemale .467 0 .499 0 1 .134 1.018 1483 Quota .388 0 .487 0 1 .46 1.212 1560 FirmSize 6.766 6.663 .892 2.742 8.739 -.13 3.209 1465 FirmAge 61.651 40 54.367 0 365 1.674 7.708 1543 Leverage .185 .17 .143 0 1.118 .995 4.747 1462 BoardSize 10.329 10 3.899 3 27 .859 3.781 1483 CEODuality .111 0 .315 0 1 2.469 7.095 1489 SIC 4.205 3 2.118 2 9 1.144 2.706 1560
Further Table 7 Panel A and B, report a comparison of firms imposed by a quota or not imposed by such accordingly. As becomes evident, the Tobin’s Q of firms not being inflicted by a quota is higher (1.41) than the measure of quota mandated firms (1.19). This result could suggest that market expectations are lower, when firms are imposed by a quota compared to firms who enjoy a comply-or explain arrangement when choosing their board composition. Besides that, the mean share of women on boards is almost 10% lower in firms not imposed by a quota. This underlines the fact that quotas do lead to an increase of women in boardrooms. As the relationship between firm performance, female directors and quota mandates, is a highly complex one and influenced by several factors, these suggestions will be further investigated using a regression analysis.
26 Table 7
Descriptive Statistics for Quota or no Quota imposed Panel A: Quota imposed
Variable Obs Mean Std.Dev. Min Max
Tobin’s Q 566 1.185 1.274 0 8.167 ROA 578 4.625 5.342 -11.444 29.043 Female 593 .327 .123 0 .636 IndFemale 593 .661 .474 0 1 FirmSize 578 6.845 .837 4.914 8.739 lnAge 605 3.888 .824 1.609 5.9 Leverage 578 .186 .145 0 .665 lnBoardSize 593 2.335 .367 1.099 3.258 CEODuality 593 .206 .405 0 1
Panel B: No Quota imposed
Variable Obs Mean Std.Dev. Min Max Tobin’s Q 848 1.407 1.762 0 13.996 ROA 886 5.8 8.919 -52.946 51.248 Female 890 .236 .133 0 .667 IndFemale 890 .337 .473 0 1 FirmSize 887 6.714 .923 2.742 8.651 lnAge 931 3.617 1.034 0 5.278 Leverage 884 .183 .141 0 1.118 lnBoardSize 890 2.218 .383 1.099 3.296 CEODuality 896 .049 .216 0 1
4.2 Multicollinearity and Correlation Matrix
As multicollinearity between variables arises when variables are highly correlated, and causes difficulties in ascertaining the effect of any single variable (Hair et al., 1998), a correlation analysis is performed.
Pearson’s correlation analysis is performed and results are reported in Table A.5. The correlation matrix shows that most variables are not significantly correlated. Strong significant correlation exists between Female and IndFemale, the indicator variable, which is expected and does not hurt, as regressions with those variables are run separately.
Possible multicollinearity between variables is further tested by estimating variance inflation factors (VIF test), resulting in all factors being lower than 8.125, passing the recommended standard (Hair et al., 1998). The VIF test suggests no multicollinearity is apparent and affecting results.
27 5 Empirical Results
In this section, I perform my empirical analysis to shed light on the relationship between gender diversity and firm performance using the methodology described in Section III and the impact a mandated quota has on this relationship. By doing this, I firstly present the estimations of the pooled OLS and the firm fixed effects regression between gender diversity and firm performance. Secondly, I go forward introducing the 2SLS model to account for endogeneity and reverse causality within the model. As a third step, I investigate the impact gender quotas have on the aforementioned relationship by making use of a moderated regression. This regression will also be both performed on the basis of a Fixed Effects regression and a pooled OLS.
5.1 The effect of gender diversity on firm performance
A Baseline Model and Firm Fixed Effects. Table 8 reports the regression outputs of the relationship between the dependent variables Tobin’s Q and ROA, reflecting firm performance, the independent variables, Female, IndFemale and sqFemale (reflecting gender diversity) and a set of control variables extensively described in Section III.
As the main independent variable, Gender Diversity, is measured in three different ways, Model (1) introduces an indicator variable, indicating a minimum of 30% of female directors Model (2) measures the share of women on a corporate board, while Model (3) is reflected by the share of female directors squared. Both the OLS and the Fixed Effects Model made use of robust and clustered standard error, adjusting for heteroskedasticity and autocorrelation. Furthermore, all models account for industry-specific characteristics through two-digit SIC dummies.
In the pooled OLS regression, one can see that firm performance measured by Tobin’s Q (Table 8 Panel A) is dependent on the firm size and leverage, particularly negatively affected in both cases. Further Tobin’s Q is reported to be independent of all remaining variables. Contrarily to Tobin’s Q, ROA is significantly affected by both Gender Diversity measured as the fraction of female directors and the indicator variable IndFemale. It is however important to note, that this significant effect vanishes, once we account for year and country fixed effects.
Moreover, the board size negatively influences ROA, meaning that an increase in the number of directors in the boardroom, decreases Return on Assets within a firm.
28
To address omitted variables problems, I add firm fixed effects in Panel C and D. In comparison to the pooled OLS results, the firm fixed effects model documents a statistically significant effect of IndFemale on firm performance measured by Tobin’s Q (Panel C Column (1b)). Notably, the coefficient reported is negatively impacting Tobin’s Q by -0.191 (5% significance level), even when adjusting for year fixed effects. This presents that once we account for differences across firms and years, the change from a below 30% margin of female directors to a above 30% share, leads to a decrease of 0.191 in Tobin’s Q, holding all other factors constant. To get a better notion of this effect economically, the coefficient (-0.191) can be multiplied by the standard deviation (1.588) of the variable, resulting in an economical significance of -.303. Compared to a mean of 1.31, this result stems an almost 25% change of the mean and can be said to be economically significant.
The substantial changes in coefficients within the two consecutive regressions can be explained by the fact, that the OLS estimation does not account for unobserved time-invariant confounding factors across firms (Stock & Watson, 2015). Such factors could include differences in organizational culture or general business practices, which are constant over time but vary across entities, which is considered using a firm fixed effects model. These findings are in line with Adams and Ferreira (2009), who report a change in coefficient signs once firm effects are added, concluding that the positive correlation between gender diversity and performance in the OLS regression is driven by omitted firm specific factors. Further, seeing an almost doubling increase of R-squared across firm fixed effect models, when including year fixed effects, one can assume that the significant effect of changes in time take on a large share when explaining changes in Tobin’s Q, wherefore firm year dummies should be included when the dependent variable is Tobin’s Q.
Considering the fixed effects regression on ROA (Panel D), one can examine overall significant fixed effects models with a statistically significant F-statistic, however an extremely low R-squared, suggesting only a small part of the variance in ROA is explained by the model. Contrasting to the fixed effects regression on Tobin’s Q (Panel C), the R-squared does not increase heavily once time dummies are included, which reflects a rather steady development of ROA over time.
29
29 Table 8
Pooled OLS and Firm Fixed Effects Regression
Panel A: Pooled OLS Regression Dependent Variable Tobin’s Q
(1a) (1b) (1c) (2a) (2b) (2c) (3a) (3b) (3c) IndFemale 0.048 -0.078 0.041 (0.135) (0.155) (0.146) Female 0.720 0.167 0.866 (0.633) (0.785) (0.755) sqFemale 1.554 0.689 2.084 (1.184) (1.422) (1.325) FirmSize -0.532*** -0.521*** -0.373** -0.547*** -0.532*** -0.394*** -0.551*** -0.537*** -0.402*** (0.137) (0.136) (0.145) (0.137) (0.136) (0.148) (0.137) (0.136) (0.147) lnAge -0.035 -0.045 -0.058 -0.033 -0.043 -0.053 -0.034 -0.043 -0.053 (0.088) (0.090) (0.095) (0.088) (0.089) (0.094) (0.088) (0.089) (0.094) Leverage -3.161*** -3.185*** -3.460*** -3.125*** -3.170*** -3.424*** -3.108*** -3.155*** -3.398*** (0.926) (0.926) (0.889) (0.932) (0.932) (0.886) (0.928) (0.930) (0.877) lnBoardSize 0.338 0.310 0.195 0.356 0.328 0.188 0.383 0.346 0.225 (0.249) (0.248) (0.264) (0.248) (0.247) (0.266) (0.247) (0.246) (0.259) CEODuality -0.273 -0.248 -0.253 -0.271 -0.251 -0.246 -0.285 -0.258 -0.258 (0.256) (0.255) (0.501) (0.253) (0.253) (0.493) (0.252) (0.252) (0.494) _cons 4.400*** 4.048*** 3.016*** 4.266*** 4.006*** 3.077*** 4.290*** 3.993*** 3.129*** (0.834) (0.807) (0.918) (0.832) (0.812) (0.931) (0.827) (0.805) (0.933) Obs. 1254 1254 1254 1254 1254 1254 1254 1254 1254 R-squared 0.219 0.238 0.306 0.223 0.238 0.309 0.224 0.238 0.311 F 4.966*** 4.925*** 4.276*** 4.909*** 4.889*** 4.185*** 4.899*** 4.881*** 4.175***
SIC Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes
Year Dummies No Yes Yes No Yes Yes No Yes Yes
Country Dummies No No Yes No No Yes No No Yes
Standard errors are in parenthesis
30
Panel B: Pooled OLS Regression with Dependent Variable ROA
(1a) (1b) (1c) (2a) (2b) (2c) (3a) (3b) (3c) IndFemale 0.921* 0.791 0.526 (0.549) (0.614) (0.700) Female 5.441** 5.125* 3.566 (2.377) (2.786) (3.510) sqFemale 9.650** 9.006* 6.203 (4.546) (5.356) (6.569) FirmSize -0.136 -0.094 0.517 -0.206 -0.167 0.466 -0.205 -0.163 0.467 (0.707) (0.712) (0.792) (0.707) (0.710) (0.812) (0.714) (0.719) (0.818) lnAge 0.135 0.124 0.236 0.139 0.133 0.244 0.132 0.125 0.239 (0.393) (0.394) (0.361) (0.391) (0.392) (0.362) (0.392) (0.393) (0.361) Leverage -10.086*** -10.149*** -12.421*** -9.875*** -9.951*** -12.303*** -9.824*** -9.911*** -12.274*** (3.668) (3.691) (3.562) (3.627) (3.645) (3.534) (3.625) (3.644) (3.530) lnBoardSize -2.664** -2.721** -1.989* -2.613** -2.655** -2.027* -2.478** -2.533** -1.912* (1.184) (1.183) (1.135) (1.154) (1.151) (1.127) (1.170) (1.171) (1.140) CEODuality 0.407 0.451 0.302 0.419 0.442 0.331 0.335 0.369 0.288 (1.006) (1.014) (1.412) (1.011) (1.020) (1.393) (1.009) (1.023) (1.403) _cons 11.894** 11.363** -0.170 11.148** 10.844** -0.113 11.488** 11.084** -0.049 (4.621) (4.624) (5.387) (4.622) (4.628) (5.448) (4.621) (4.630) (5.483) Obs. 1275 1275 1275 1275 1275 1275 1275 1275 1275 R-squared 0.077 0.082 0.185 0.082 0.086 0.186 0.081 0.086 0.186 F 2.391 3.259 6.937 2.420 3.336 6.989 2.372 3.324 7.060
SIC Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes
Year Dummies No Yes Yes No Yes Yes No Yes Yes
Country Dummies No No Yes No No Yes No No Yes
Standard errors are in parenthesis
31
31 Panel C: Fixed Effect Model with Dependent Variable Tobin’s Q
(1a) (2a) (3a) (1b) (2b) (3b)
IndFemale -0.074 -0.191** (0.068) (0.074) Female 0.357 -0.482 (0.315) (0.351) sqFemale 0.302 -0.972 (0.660) (0.712) FirmSize -0.449 -0.442 -0.438 -0.448 -0.418 -0.429 (0.345) (0.345) (0.344) (0.294) (0.293) (0.292) lnAge 0.918** 0.733** 0.798** -0.204 -0.238 -0.235 (0.370) (0.365) (0.380) (0.447) (0.447) (0.449) Leverage -1.776*** -1.711*** -1.727*** -1.716*** -1.680*** -1.681*** (0.636) (0.630) (0.633) (0.597) (0.590) (0.591) lnBoardSize -0.285 -0.292 -0.284 -0.264 -0.273 -0.296 (0.448) (0.448) (0.451) (0.434) (0.435) (0.437) CEODuality -0.285** -0.252** -0.259** -0.213 -0.185 -0.188 (0.129) (0.117) (0.121) (0.132) (0.127) (0.125) _cons 1.964 2.488 2.269 5.699** 5.670** 5.751** (2.832) (2.867) (2.848) (2.636) (2.610) (2.601) Obs. 1254 1254 1254 1254 1254 1254 R-squared 0.060 0.060 0.059 0.128 0.123 0.124 F 4.168*** 4.379*** 4.404*** 7.498*** 7.299*** 7.446*** SIC Dummies No No No No No No
Year Dummies No No No Yes Yes Yes
Country Dummies No No No No No No
Standard errors are in parenthesis
32
Panel D: Fixed Effects Model with Dependent Variable ROA
(1a) (2a) (3a) (1b) (2b) (3b)
IndFemale 0.084 -0.025 (0.447) (0.483) Female 1.087 0.106 (2.025) (2.313) sqFemale 1.324 -0.235 (4.276) (4.656) FirmSize -2.088* -2.096* -2.089* -1.851 -1.846 -1.850 (1.223) (1.229) (1.225) (1.220) (1.230) (1.220) lnAge 3.588* 3.346 3.474 2.823 2.811 2.822 (2.157) (2.087) (2.197) (2.214) (2.190) (2.208) Leverage -10.020*** -9.977*** -9.995*** -10.303*** -10.297*** -10.299*** (3.269) (3.212) (3.222) (3.356) (3.326) (3.330) lnBoardSize -1.395 -1.401 -1.366 -1.743 -1.745 -1.749 (1.708) (1.715) (1.693) (1.743) (1.747) (1.735) CEODuality -1.604*** -1.578*** -1.593*** -1.446** -1.440** -1.444** (0.523) (0.519) (0.515) (0.562) (0.562) (0.561) _cons 11.191 11.905 11.466 12.907 12.896 12.923 (10.673) (10.806) (10.878) (11.143) (11.174) (11.019) Obs. 1275 1275 1275 1275 1275 1275 R-squared 0.031 0.031 0.031 0.039 0.039 0.039 F 5.290*** 5.060*** 5.089*** 2.663*** 2.765*** 2.770*** SIC Dummies No No No No No No
Year Dummies No No No Yes Yes Yes
Country Dummies No No No No No No
Standard errors are in parenthesis
*** p<0.01, ** p<0.05, * p<0.1
Two-Stage-Least-Squares Regression. To address reverse causality concerns, I estimate both models using instrumental variables techniques.
I present the results of the 2SLS regression in Table 9. Model (1a) and (2a) reflect a firm fixed effects model, while Model (1b) and (2b) reflect a pooled OLS regression of gender diversity on Tobin’s Q (Panel A) or ROA (Panel B). I make use of both models to facilitate a more meaningful comparison of the results. However, I refrain from reporting a regression output using sqFemale in this case, as the instruments did not fulfill the relevance criteria discussed in Section 3.3. Following the results of the Hansen J statistic, I include both instruments into all regressions, since the null hypothesis cannot be rejected, meaning that overidentification of instruments is not apparent.
Table 9 Panel A reports the output of the second stage regression on Tobin’s Q. First of all, both implemented regressions report insignificant results considering the regressor
