Diagnosing the gender effect : evidence from the forced CEO turnover rate

(1)

Diagnosing the Gender Effect:

Evidence from the Forced CEO Turnover Rate

University of Amsterdam, Amsterdam Business School Faculty of Economics and Business

MSc Finance

Master Specialization Corporate Finance Master Thesis

Author: Heran Fan

Student number: 11626569

Thesis supervisor: Dr. Torsten Jochem

(2)

Statement of Originality

This document is written by Heran Fan 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)

Abstract

This paper aims to investigate the role of a gender effect in the senior corporate firings among a large sample of S&P 500 firms during the period from 2000 to 2015. We document that female CEOs tend to face a higher probability of dismissals when they underperform absolutely or underperform relative to their industry peers, while there is no significant difference between female CEOs’ forced turnover rate and male CEOs’ forced turnover rate when they underperform relative to the competitive peers or relative performance peers. We also implement an instrumental variable (IV) approach to confirm results. Our findings provide new though yet weak evidence that a gender effect may exists in the senior management level.

Keywords: Gender effects, Forced CEO turnover, Peer performance.

(4)

“Qatar Airways has to be led by a man, because it is a very challenging position.” - Al Baker, the current CEO of Qatar Airways 1 “You’re the only woman. It’s very only. I was at a high level playing in a golf foursome with all high-level men.” - Jan Fields, the President of McDonald’s USA during 2010-20122

1. Introduction

In recent years, there has been an increasing interest in studying gender effects in corporations. Among those questions are whether CEO gender has an impact on firm valuation, or whether the market will react significantly differently towards CEOs’ gender and thus treat female CEOs differently from male CEOs (Martin, Nishikawa & Williams, 2009; Ahern & Dittmar, 2012; Bass & Avolio, 1994). As indicated by Al Baker and Jan Fields (above quotes), it is a harsher environment for women to participate in the senior management positions. However, the last two decades have seen a dramatic increase in newly designed government policies for addressing the gender imbalance problem in corporate boards. For example, the Norwegian Parliament announced a gender quota in late 2003 that all public-limited firms are required to reach at least 40 percent of female representation on the board. Spain and Iceland followed by introducing the quota of 40 percent female board representation in 2007 and 2010 respectively. Some governments set a different percentage requirement for the gender quota they were in the process of, such as Netherlands with a 30 percent female quota by 2016 and Denmark with a flexi-quota in 2012. Other countries like Canada and UK were contemplating it and might introduce the gender quota in the near future.

1_{Whitley, A., Katz, B.D., & Park, k. (2018, June 5). Qatar Air Boss Apologizes for Saying CEO Job Must Be Held by a}

Man. Bloomberg. Retrieved from https://www.bloomberg.com/news/articles/2018-06-05/women-look-away-qatar-airways-ceo-says-only-men-can-do-his-job

2_{Chira, S. (2017, July 21). Why Women Aren’t CEOs, According to Women Who Almost Were. The New York Times.}

(6)

Despite the great attention paid to this field, there is still a lack of academic consensus on whether male CEOs are valued more compared to female CEOs. Many researchers state that certain stereotyped biases and qualification problems remain in the upper level management under introducing the “glass ceiling” effect, which lead to an relatively underrepresentation of female in the top management team (Kaufmann et al., 1996; Singh & Vinnicombe, 2004; Oakley; 2000). Under the “glass ceiling” effect theory, the pay gap is expected to exist and to be larger at the top of the hierarchy, while the evidence from Bugeja, Matolcsy and Spiropoulos (2012) show there is no significant pay gap at the top management level. More conversely, some scholars suggest that the beneficial role of female CEOs in corporate operation is recognized and thus it is worthwhile to retain female CEOs as a valuable input for the firm (Francoeur et al., 2008). Besides, a large number of studies examine potential gender effects mainly through the hiring, wage and promotion channels (Bell, 2005; Bugeja, Matolcsy & Spiropoulos, 2012). So far, no research has been found to examine the gender effect problem through the firing channel - the forced CEO turnover event. Thus a detailed investigation of whether and how a gender effect exists in forced CEO turnover rates is still lacking. Hence, if there does not exist any gender effect, we would expect no difference in the forced turnover rate of female CEOs and male CEOs when CEOs are underperforming. The objective of this paper is to examine the role of a potential gender effect in senior corporate firings. We first develop a forced CEO turnover dataset by using a merged data set of two groups of scholars’ work: Jenter and Kanaan and Peters and Wagner provide data for forced turnovers from 2000 to 2015 (Jenter & Kanaan, 2015; Peters & Wagner, 2014). Both of them apply the same methodology to define forced CEO turnover events: that is, they primarily conduct a press-based approach and use an age criterion when the press does not report the underlying reason for CEO departures clearly. It contributes to a forced CEO

(7)

turnover dataset of 925 observations during the sample period of January 1, 2000 to December 31, 2015.

We then design a Logistic model by regressing the forced CEO turnover event on the firm’s absolute underperformance and compare whether female CEOs would face a higher forced turnover rate under a similar underperformance situation relative to male CEOs. It shows that female CEOs tend to be terminated faster when they underperform absolutely. Nevertheless, this model lacks the empirical explanatory power since it ignores the situation that firms might underperform absolutely but still outperform some selected peers, which lower the probability of CEO dismissals. Therefore, we improve this Logistic model by replacing the absolute underperformance with relative underperformance and further adopt three different definitions of peer performance to make the results more robust. The results from the first measure – underperformance relative to industry peers – help to reject the null hypothesis and indicate that a gender effect exists in the senior corporate firings. Two alternative measures – underperformance relative to product market peers and relative performance peers – do however not provide significant results. Furthermore, to overcome a potential endogeneity problem in the previous Logistic models we build a two-stage instrumental variable (IV) regression model using an industry shock as the instrument for firms’ absolute performance. An industry shock is considered as a good instrumental variable as the weighted average industry performance is unrelated to the CEO’s own skills or abilities but directly affect the firm performance and thus the CEO evaluation. Nevertheless, the relevance test fails to prove that this industry shock is a very relevant and strong instrumental variable and thus it is not surprising to see the findings from this IV model do not support a gender effect for the senior corporate firing channel.

(8)

those models and especially significantly negative under the first measure of underperformance. Overall, our results therefore provide some weak evidence for a gender effect to exist in forced CEO turnover events.

This paper contributes to the gender effects literature in several ways. First, to the best of our knowledge, it is the first paper that tries to study gender effects directly through linking it with corporate firings. Therefore, this paper helps to fill the gap and adds new evidence in this area. Second, our definition of performance peer groups, whose performance is beyond the CEO’s control, is not only limited to industry peers, but also introduce two other definitions of peer groups - the product market/competitor peers and relative performance peers. This brings more robustness to the result and sheds light on other related studies that use performance peers. Third, the weak evidence from our findings informs the debate of policy-makers that it is not worthwhile to consider designing policies towards managing the corporate gender imbalance problem in a very aggressive way.

The remainder of the paper is organized in the following way. Chapter 2 begins by laying out the theoretical research dimensions and then develops the related hypothesis. Chapter 3 is concerned with the data and provides the summary statistics. Chapter 4 describes the methodology and we discuss the main results in Chapter 5. Chapter 6 summarizes and concludes.

(9)

2. Literature Review and Hypothesis Development

2.1. CEO Underperformance and Dismissals

Empirical results show that firms tend to not compensate or punish their CEOs solely based on the observed CEOs’ or firms’ own performance but more focus on comparing CEOs’ or firms’ own performance with their performance peer groups. This is mainly due to the economic theory that some random factors and uncertainty exist and they are out of CEOs’ control but would directly impact on the CEO or firm performance (Gibbons & Murphy, 1990). Since using relative performance peer groups helps to remove these external shocks to some extent and further makes the CEO performance evaluation much fairer, modern firms widely adopt this method to evaluate their CEOs and make rewards or punishment decisions based on it. The worst punishment would be dismissing the current CEO.

To be specific, if the board observes that the CEO performs very poorly compared to the selected peers and his or her performance measurement is even under some setting threshold, the CEO would face a higher probability to be terminated. For example, Parrino (1997) find that CEOs who underperform relative to the industry level are more easily to be identified and thus to be fired.

However, as suggested by Albuquerque (2009), it is important to set the choice of performance peers and a misspecified peer group would lead to a potential bias in the relative performance measurement. For instance, Albuquerque (2009) points out that simply applying the industry peers method would ignore the situation that when the common exogenous shock affects some firms in the industry positively while influence others in the same industry negatively, then the mean industry peer performance would not reflect this external shock as these two effects cancel out simultaneously. In order to avoid this problem and improve the

(10)

robustness of the results, we implement three different definitions of peer groups in this paper, which we not solely focus on the most widely used- the industry peers, but also include the product market/competitive peers and the relative performance peers. The selection of peer groups in this paper will be discussed in details in the Chapter 3.

Additionally, even if the previous studies show that CEO underperformance is positively linked to the CEO dismissals, they do not further examine separately whether CEO under different gender would face a significant different turnover rate under this similar situation, which could be also considered as a potential gender effect study.

2.2. Forced CEO Turnover and Gender Effect Existence

As discussed in previous section, it remains unclear whether male CEO and female CEO would face a significant different turnover rate when they are underperforming relative to the selected peers. Therefore, the main prediction is that if there is no gender effect, male CEO dismissal rate should be indifferent from female CEO dismissal rate when the CEO is underperforming related to his or her peers.

However, even though many studies have conducted in figuring out whether there exists certain gender effect in corporate boards or not, the answer remains ambiguous.

A great number of scholars argue that the market treat female CEOs differently from male CEOs and generally has a passive attitude on female CEOs, which lowers female presence in upper level management. As discussed by Singh and Vinnicombe (2004), one of the key reasons why women are facing a “glass-ceiling” problem to reach the top management position comes from the qualification issue that they tend to be lack of ambition, line experience and commitment. Thus female are generally not qualified enough to compete with

(11)

male for the senior positions in the corporate world. Oakley (2000) points out that not only qualification issue, but also discrimination reason results in a significantly lower female representation on the board. She indicates that certain gender-based stereotypes remain in the labor market that women are usually stereotypically seen as less capable than men in most all the business-required traits or skills. This leads to unequal work opportunity for female in the senior management positions. Besides, female CEOs are found to be more risk-adverse and thus invest in less risk-involved projects with lower returns compared to male CEOs, which will distort capital allocation and further cause shareholder dissatisfaction (Faccio et al., 2016). Moreover, Ahern and Dittmar (2012) study the impact of gender quota on Norwegian firms by applying both OLS regressions on abnormal announcement returns and instrumental variables regressions with Tobin’s Q as the dependent variable. They argue that the market generally seems to have a negative view of the mandated female board representation policy, which is showed by a distinct decrease in announcement returns and Tobin’s Q. All of these studies display that the existence of gender effect problem restricts the number of female CEOs and the market perceives female CEOs negatively relative to male CEOs. Therefore, these studies might support the alternative hypothesis that female CEOs will be fired sooner when they are underperforming.

Nevertheless, many other studies show that female CEOs would be considered as a valuable input for the corporate organization and thus it is worthwhile to remain female at upper level management. Francoeur et al. (2008) find that a higher participation rate of female in senior management positions generates a positive and significant abnormal return and improves firms’ financial performance. Findings from Bass and Avolio (1994) also support this viewpoint and state that the benefits of female leadership have a positive influence on both group and organizational performance. Furthermore, greater gender diversity in board helps to raise the firms’ corporate social rating, which is beneficial for the enhancement of firms’

(12)

reputation (Bear et al., 2010). These findings are consistent with the current rise of legislative actions regarding the advancement of female in business, such as the mandated female quote in Norway. Based on them, we could argue that firms may become more patient towards female CEOs when they underperform, which cause a lower forced turnover rate for female CEOs rather than male CEOs.

Besides, there also have evidence that gender effect does not impact significantly on firm valuation and CEOs with different gender do not get treated differently, which could build our null hypothesis that there is no significant difference in forced CEO turnover rate for different gender. For example, Mohan and Chen (2004) show that IPO pricing is unrelated to CEO gender. By studying a matched group of 70 female CEOs’ announcement events with 70 male CEOs’ announcement events, Martin et al. (2009) find there is no significant difference of three-day cumulative abnormal returns between two sample groups and further conclude that financial market does not obtain certain gender discrimination on CEO announcement. Furthermore, the potential gender pay gap is not found at the top management level by comparing the compensation of female CEOs with male CEOs of US public companies over the period of 1998-2010 (Bugeja, Matolcsy & Spiropoulos, 2012). According to these quantitative analysis, we could expect female CEOs will be treated similarly in a way as male CEOs when they underperform compared to peers and facing the similar firing situation.

Consequently, the core hypothesis of this paper is:

H0: there is no difference between female and male CEOs’ forced turnover rate when they are

underperforming.

H1: there is a significant difference between female and male CEOs’ forced turnover rate

(13)

3. Data and Summary Statistics

3.1. Data Source

Our dataset consists of S&P 500 firms from January 1, 2000 to December 31, 2015.

We collect data on the forced CEO turnover from the ExecuComp database. Based on our study focus, it is vital to separate the forced CEO turnover event from the voluntary CEO turnover event. One of the most popular ways to define the CEO dismissals is using the press reports as suggested by Parrino (1997). Therefore, we use a merged version of forced CEO turnover dataset from 1993-2000 provided by Jenter and Kanaan and from 2001-2015 provided by Peters and Wagner (Jenter & Kanaan, 2015; Peters & Wagner, 2014). They mainly conduct the press-based procedure to classify the forced CEO turnover events as press reports that whether the CEO is forced out or not. When the press does not report the clear reason for some CEO departures, they further use the age criterion by defining those CEO departures under the age of 60 years old as the forced case while those who is leaving above or at the age of 60 years old are considered as the voluntary case (Jenter & Kanaan, 2015; Peters & Wagner, 2014). Thus, our dataset consists of 925 forced CEO events during 2000 to 2015. Then, we merge this dataset with the entire dataset from the ExecuComp database during the same period and define other turnover events except these 925 events are voluntary CEO turnover or non-forced turnover events. In the end our dataset contains 30,692 firm-year observations of which 925 observations are forced CEO events.

Another key variable in this paper is the peer group performance. We apply three different definitions of peer groups in order to avoid potential measurement bias and ensure the robustness of results. Firstly, we define the industry peers, which is one of the most widely used methods in relative peer performance literature. Because we assume firms in the same

(14)

industry would face similar exogenous shock and uncertainty that generate a reasonable comparison for CEO performance in the same industry. The data is obtained from Compustat. We implement the similar method to formulate the mean industry performance changes as Bertrand and Mullainathan (2001) do in their paper, which is to use the weighted average accounting return in the year in the two-digit industry (i.e. SIC2) that firm belongs to, excluding the firm itself. Secondly, the product market/competitor peers are also considered as a good proxy of performance peers with the available access on Hoberg Philipps database. The reason is that nowadays many firms not only face main competitors from their own industry but also from other industries, thus it is better for firms to set the competitors from their product market or based on the product similarity for their peers. In Hoberg Philipps database, it selects a distinct group of competitors for each firm during 2000-2015, while some “closer rivals” with a relatively higher score than others. Then we select the top five “closer rivals” and calculate the average accounting return (i.e. ROA) for these five key competitors and use it as the second measure of peer group performance. Thirdly, we use the relative performance peers that could be obtained in Institutional Shareholder Services (ISS) database. Since CEOs are normally rewarded by comparing their performance to a group of relative performance peers the board selected, instead of simply comparing with the mean industry level or the firm’s main competitors. However, we apply the average stock return of these relative performance peers rather than the average accounting return (i.e. ROA) of these relative performance peers for the third measure of peer group performance. This is based on the Bizjak and his colleagues’ (2017) finding that for firms, which select the relative performance peers as the benchmarking peer group, mainly use stock returns rather than accounting returns as the performance metric. In the last one decade, roughly 80% of these firms used stock return as the performance metric while only 30% of them chose the accounting return as the performance proxy for the relative performance peers.

(15)

CEO characteristics and compensation data come from Execucomp. We obtain accounting information from Compustat and board characteristics from ISS. Stock return information is obtained from CRSP.

3.2. Summary Statistics

Table 1 provides the cross-sectional mean values for firm and board characteristics for two subgroups, firms that have a female CEO and firms that have a male CEO, from 2000 to 2015. Panel A presents the firm characteristics and Panel B details on the board characteristics. Panel A shows that on average firms with a female CEO tend to be smaller under the measurement of total assets than firms with a male CEO. However, they do not display the similar pattern and there is no significant difference found in the other measurements of firm size: the cash they are holding, sales and the number of employees. Besides, firms under female leadership have a lower leverage ratio of 22% compared to firms under male leadership with a ratio of 24%, which is significant at 5% level. This is consistent with the previous literature we discuss above in the Chapter 2 that female CEOs tend to be more risk-adverse and not willing to obtain a high leverage ratio.

Panel B displays that the average board size is not different for firms under female leadership and firms under male leadership. Both of them contain approximately 8.5 members on the board. But the composition of these board members is significantly different under two cases. First of all, it is worthwhile to note that the percentage of female directors in the board of firms with a female CEO is roughly 35% while firms with a male CEO only reach an approximately 5% female representation on the board. One possible explanation is that more female directors on the board help to increase the probability for a female to win the top

(16)

position. Moreover, on average firms under male leadership tend to have a relatively higher percentage of Caucasian and insiders in the board compared to firms under female leadership.

Table 1.

Firm and Board Characteristics.

This table presents summary statistics for firm and board characteristics. Panel A shows the average firm characteristics under two different cases- when firms with a female CEO or firms with a male CEO. Panel B reports the statistics on the board characteristics. Column 1 reports estimates under the case that the firm has a female CEO in this year. Column 2 reports estimates under the case that the firm has a male CEO in this year. Column 3 displays the statistics from a t-test of the difference between these two cases. The mean and standard deviation (in parentheses) for each variable are reported separately for two groups of firms. *, **, and *** indicate significance level at 10%, 5% and 1% respectively. Female CEO (1) Male CEO (2) T-score of difference (3)

Panel A: Firm characteristics

Ln(Total assets) 7.68 (1.90) 7.81 (1.74) -1.94* Ln(Cash) 4.70 (1.78) 4.68 (1.83) 0.37 Leverage 0.22 (0.21) 0.24 (0.22) -2.08** Sales (thousand) 4115.90 (8337.72) 3794.93 (6707.94) 1.34 Number of employees (thousand) 12.56 (22.32) 11.51 (19.98) 1.04 Observations 733 29,326

Panel B: Board characteristics

Board size 8.53 (3.50) 8.51 (3.67) 0.18 Female (%) 35.43 (0.20) 4.59 (0.11) 46.37*** Caucasian (%) 87.04 (0.16) 88.89 (0.15) -3.31*** Insider (%) 15.79 (0.14) 17.37 (0.15) -3.17*** Observations 904 12,725

In Table 2 we report the summary statistics on CEO compensation and characteristics for the overall sample from 2000 to 2015 and the comparison of them across the two subgroups: one group is firms with a female CEO and the other is firms with a male CEO. It shows on

(17)

but the weight of composition in the total compensation package is different. Female CEOs generally obtain a higher salary but a significantly lower bonus and long-term incentive payouts. Additionally, it presents that female CEOs are significantly younger than male CEOs (about two years) and the average age of male CEOs for leaving the firm is much larger compared to the female CEOs. Furthermore, female CEOs on average have a shorter tenure with around seven years comparing male CEOs with approximately eight years.

Table 2.

CEO Characteristics and Incentives.

This table presents summary statistics for CEO characteristics and incentives. Column 1 reports estimates under the case that the firm has a female CEO in this year. Column 2 reports estimates under the case that the firm has a male CEO in this year. Column 3 displays the statistics from a t-test of the difference between these two cases. The mean and standard deviation (in parentheses) for each variable are reported separately for two groups of firms. *, **, and *** indicate significance level at 10%, 5% and 1% respectively. Female CEO (1) Male CEO (2) T-score of difference (3) Salary (thousand) 759.90 (336.29) 729.47 (327.91) 2.51** Bonus (thousand) 216.83 (578.10) 353.06 (728.21) -6.49*** Total compensation (thousand) 5228.87 (6654.44) 5322.39 (6860.56) -0.39 Long-term incentive payouts

(thousand) 33.38 (207.20) 33.04 (513.06) -6.20*** Age 53.55 (5.32) 55.62 (6.89) -10.49***

Age left firm 56.56

(5.18) 59.17 (6.77) -13.73*** Tenure 7.10 (3.70) 7.97 (3.99) -6.46*** Observations 791 29,323

(18)

4. Methodology

The paper will firstly test whether underperformance in absolute terms, which is measured by the accounting return (i.e. ROA), will lead to a higher probability of forced CEO turnover rate and whether female CEOs and male CEOs face a significantly different forced turnover rate under the same absolute underperformance situation. In other words, we test if, for a firm which generates a negative ROA (e.g. -5%), a female CEO or a male CEO will be more likely to be fired or not. To analyze it, we run a Logistic regression model as follows:

Forcedi,t = α + β1ROAi,t + β2Femalei,t + β3 (ROA*Female)i,t + Xi,t

+ Industry FE + Year FE + εi,t (1)

The dependent variable, Forcedi,t, is a dummy variable and equal to one if there is a

immediate forced turnover event for firm i at time t and zero otherwise. ROAi,t is the

accounting measure of performance for firm i at time t, and Xi,t refers to a set of control

variables (e.g. firm size). Femalei,t is a dummy that equals to one if the gender of CEO at time

t is female and zero otherwise. We also include the industry fixed effects and year fixed effects in order to capture the potential trends. β3 is the variable of interest. If there is no

gender effect, this coefficient will be not significant from zero. We would expect a negative coefficient when the gender effect exists and the male CEO is valued more. Because a negative coefficient refers to a higher probability of female CEOs to be terminated for an increasing negative ROA. On the other hand, we expect β3 to be positive if female CEOs are

valued more and thus men are fired more quickly than women.

However, this model does not consider the situation that even the firm underperforms in the absolute term; it does not underperform relative to the industry level or its selected peers. For example, if the firm generates a ROA of -5% at time t while the average ROA in its main

(19)

industry is -10%, the firm actually outperforms and thus the CEO will not face a threat of dismissal in this case. Therefore, we improve the Logistic model in (1) by replacing the absolute accounting measure ROA with the relative term of performance, which is also more consistent with the previous literature and helps to test the core hypothesis of this paper. Then the second Logistic model becomes:

Forcedi,t = γ + δ1Rel.perfi,t + δ2Femalei,t + δ3(Rel.perf*Female)i,t + Xi,t

+ Industry FE + Year FE + εi,t (2)

Where Rel.perfi,t refers to the difference between the firm i’s own performance and its peer

group performance at time t. Thus if the underperformance in relative term means a negative

Rel.perfi,t that firm i’s own performance is weaker compared to its peers. And as discussed

above, we apply three definitions of peers in this paper, which means we will conduct this model under each definition of peers.

δ3 is the coefficient of interests. If there is no gender effect, this coefficient will be not

significant from zero. If male CEOs are relatively valued more, the coefficient is expected to be significantly negative, since relative performance is a negative item and a negative coefficient refers to a higher forced turnover rate for female CEOs. If female CEOs are relatively valued more and the board becomes more patient on their poor performance, we expect a significant positive coefficient.

In order to address the potential endogeneity problem in the previous model setting, we further design a two-stage instrumental variable (IV) regression by using the industry shocks as the basis for the instrumental variable. The motivation behind is that the industry shocks or what we called the mean industry peer performance could be considered as the exogenous shock that are out of the CEO’s control and he or she has no capability to affect the mean

(20)

industry peer performance. But this industry shock might influence the relative performance evaluation directly; especially when the current CEO is underperforming compared to the mean industry peers. Then the board would make the related decision to terminate the current CEO. Therefore, we design the following two-stage regressions:

(i) First stage: Perfi,t = φ1 + ω1Ind.Shocki,t + ν1i,t

(Perf*Female)i,t = φ2 + ω2(Ind.Shock*Female)i,t + ν2i,t

(ii) Second stage: Forcedi,t = σ + ρ1𝑃𝑒𝑟𝑓i,t + ρ2Femalei,t + ρ3(𝑃𝑒𝑟𝑓 ∗ 𝐹𝑒𝑚𝑎𝑙𝑒)i,t

+ Xi,t + Industry FE + Year FE + εi,t (3)

Where Perfi,t in first stage refers to the firm i’s own performance at time t and we use the

accounting measure ROA as a proxy for measuring the firm performance. Ind.shocki,t is the

weighted average accounting return ROA for the industry at time t, excluding the firm i itself. 𝑃𝑒𝑟𝑓i,t and (𝑃𝑒𝑟𝑓 ∗ 𝐹𝑒𝑚𝑎𝑙𝑒)i,t in second stage refer to the predicted value of firm i’s

performance on exogenous component at time t and the estimated value of it times the dummy variable Femalei,t. Again, ρ3 is our variable of interest that if there is no gender effect

problem, this coefficient should be not significant from zero. If the gender effect problem exists and the male CEO is valued relatively more than the female CEO, this coefficient is expected to be negative. If the female CEO is valued relatively more, vice versa.

(21)

5. Results

In this chapter we conduct both multivariate OLS model and IV approach to test our core hypothesis on examining the potential role of gender effect in the senior corporate firing channel.

Table 3 presents the coefficient estimates from the first Logistic model in regression (1) based on the absolute accounting measure of underperformance. Column (1) is the simplest OLS regression results without considering the control variables and fixed effects. Columns (2), (3) and (4) include SIC2-level industry fixed effects and year fixed effects. Column (3) adds the firm size variable and column (4) add two more variables - CEO age and leverage to control CEO incentives. The reason for controlling these variables is that two subgroups from sample group displays the significant difference under these key firm characteristics and CEO characteristics variables as showed in Table 1 panel A and Table 2. We do not control for any board characteristics variables yet are mainly due to the less observations they have and the certain board characteristics might be the potential channel or explanation for the gender discrimination issue in the corporate firings, which could be tested further if we find some strong evidence for the existence of gender effect problem.

It shows all estimated coefficients on ROA is significantly negative at 1% level in all four columns, which refers to a strong positive and significant relationship between the absolute underperformance and the probability of CEO dismissals. This is due to a larger negative number of accounting return ROA with this negative coefficient will lead to an increase in the forced CEO turnover rate. We also see that a female CEO in general will face a relatively higher forced CEO turnover rate compared to a male CEO. This could be that female CEOs are more likely to be hired when the firm is facing some financial difficulities or in a crisis, and thus these female CEOs tend to be fired more often when their leadership could not

(22)

improve the firm performance during such a short time period. Or it could be explained by the female CEO’s qualification problem as mentioned in the literature review section that they are on average lack of certain criteria such as line experience for this senior position at the firm and therefore could not perform as good as the male CEO, which makes them to be more easily fired.

It is vital to note that all coefficients on the interaction term ROA*Female are negative and significant at 10% level except for column 4 with including more control variables. This indicates that a female CEO who underperforms absolutely will face a significantly higher chance to be terminated than a male CEO with equal underperformance. This could be considered as a relatively strong support for the existence of gender discrimination in the senior firing cases.

Nevertheless, these results under the absolute underperformance lacks the empirical explanatory power. The main reason is that firms in the real world do not evaluate the CEO simply according to the absolute term of performance, which does not provide the correct and fair measurement. For instance, even if the CEO is underperforming in absolute term, he or she could still outperform relative to the peers.

(23)

Table 3.

Forced CEO Turnover and the Absolute Underperformance.

This table presents the results of the logit model in equation (1) the absolute underperformance during the sample period 2000-2015. We use the accounting return ROA calculated using data from Compustat as the proxy of the absolute underperformance.

The dependent variable is the forced CEO turnover rate (Forced). Forced equals 1 for firms that fire their CEO in this year. Control variables include ln(assets), ln(age) and leverage. Column (2), (3) and (4) include all SIC2-level industry fixed effects and year fixed effects and column (4) include all control variables. All robust standard errors are clustered at the industry level and presented in parentheses. *, **, and *** indicate significance level at 10%, 5% and 1% respectively.

Dependent Variable: Forced CEO Turnover

(1) (2) (3) (4) ROA -0.040*** (0.012) -0.037*** (0.012) -0.043*** (0.012) -0.046*** (0.013) Female 0.018*** (0.006) 0.013** (0.006) 0.015** (0.006) 0.005 (0.006) ROA*Female -0.135* (0.072) -0.128* (0.075) -0.124* (0.076) -0.062 (0.076) Ln(Assets) 0.001* (0.0007) 0.001 (0.0008) Ln(Age) -0.033*** (0.008) Leverage 0.013* (0.007)

Year fixed effects No Yes Yes Yes

Industry fixed effects No Yes Yes Yes

(24)

Table 4 reports the coefficient estimates from the second Logistic model in regression (2) based on the three different measures of peer group performance, which are the mean industry peers performance, competitive peers performance and relative peers performance. These are presented in panel A, B and C respectively. These performance measures attempt to overcome the empirical problem mentioned before for the absolute underperformance measure.

In panel A, the coefficient on relative performance (Rel.Perf) is significantly negative at 1% level in all four columns, which is in line with the previous empirical findings that CEOs underperform relative to the industry peers would be terminated sooner. On average it also shows that female CEOs might face a much higher forced CEO turnover rate, which is about 1.5% to the unconditional average forced CEO turnover rate than male CEOs when they perform poorly compared to the industry level, while this significant pattern disappear when we control more variables. This 1.5% increase in the forced CEO turnover rate is actually quite large as we only find 925 observations as the forced CEO turnover events during 2000-2015 with overall 30,692 firm-year observations.

The most important variable in this study, which is the interaction factor (Rel.Perf*Female), has a negative coefficient at 10% significance level in Column (1) , (2) and (3). It helps to reject the null hypothesis and indicates that male CEOs are relatively valued more and female CEOs are more likely to be fired under the similar underperformance situation. However, this relationship is not very strong and if we include more control variables in the regression as showed in Column (4), this relationship becomes even insignificant.

In panel B, the coefficient estimates based on the second measure- the product market/competitive peers display the similar pattern as the first measure that the sign of coefficients are completely consistent with the panel A. However, the coefficient on the

(25)

interaction term (Rel.Perf*Female) is not significant in all columns, which fails to reject the null hypothesis and further implies female CEOs are not treated differently compared to male CEOs when they are underperforming compared to the competitive peers.

We also report the similar sign of coefficient for the first three variables in panel C based on the third measure of peer performance- the relative performance peers. Even if the estimated coefficients on the relative performance is negative, this association is much weaker and only significant in Column (3) and (4). Moreover, both coefficients on Female and the interaction term (Rel.Perf*Female) is not significant in all four columns under this measure. It means we get a similar conclusion as the second measure that female CEOs are not treated differently from male CEOs when they perform poorly compared to the relative peers. But this could also due to the significantly less amount of observations under the third measure and thus we could not simply give an explanation on the results from panel C.

In summary, results from these three different measures of peer group performance do not provide the consistent conclusion even though the sign of coefficient on the interaction term (Rel.Perf*Female) is same under all these three measures. Because only the first measure rejects the null hypothesis while the other two measures fail to show that a gender effect exists that leads to a higher female forced CEO turnover rate.

(26)

Table 4.

Forced CEO Turnover and the Relative Underperformance.

This table presents the results of the logit model in equation (2) based on the three different measures of peer group performance during the sample period 2000-2015. In panel A, the coefficient estimates of the first measure- the mean industry peers performance from Compustat are reported. The coefficient estimates of the second measure of peer group performance- the competitive peers performance from Hoberg-Phillips database are reported in panel B. And the third measure- the relative peers performance from ISS is displayed in panel C. First two measures use the accounting return and the third measure uses the stock return as the proxy of performance.

The dependent variable is the forced CEO turnover rate (Forced). Forced equals 1 for firms that fire the CEO in this year. Control variables include ln(assets), ln(age) and leverage. Column (2), (3) and (4) include all SIC2-level industry fixed effects and year fixed effects and column (4) include all control variables. All robust standard errors are clustered at the industry level and presented in parentheses3_{. *, **, and *** indicate significance level at 10%, 5% and 1% respectively.}

3_{We also try to test the equation (2) with clustering at the firm level. As it generates the similar results without improving}

the significance, we keep the results with clustering at the industry level for the table.

Panel A: first measure of peer performance- the mean industry peers performance

(1) (2) (3) (4) Rel.Perf -0.041*** (0.012) -0.041*** (0.013) -0.046*** (0.014) -0.051*** (0.014) Female 0.018*** (0.005) 0.015*** (0.006) 0.015*** (0.006) 0.008 (0.006) Rel.Perf*Female -0.133* (0.071) -0.136* (0.074) -0.132* (0.075) -0.067 (0.077) Ln(Assets) 0.0005 (0.001) 0.0003 (0.001) Ln(Age) -0.032*** (0.009) Leverage 0.020** (0.008)

(27)

4_{We also apply the accounting return ROA for the third measure of peer performance, however, the less powerful linking of}

accounting return with this measure group and the low amount of observations generates more noises in the results. Thus we could not provide the proper explanation for these results based on the accounting return ROA for this measure.

Panel B: second measure of peer performance- the product market/competitive peers performance

(1) (2) (3) (4) Rel.Perf -0.044*** (0.008) -0.046*** (0.008) -0.046*** (0.008) -0.037*** (0.008) Female 0.016** (0.006) 0.011* (0.007) 0.011* (0.007) 0.005 (0.007) Rel.Perf*Female -0.035 (0.030) -0.032 (0.031) -0.032 (0.031) -0.022 (0.029) Ln(Assets) 0.0002 (0.001) 0.00003 (0.001) Ln(Age) -0.031*** (0.008) Leverage 0.021*** (0.007)

Observations 29,282 29,282 28,995 27,768

Panel C: third measure of peer performance- the relative performance peers4

(1) (2) (3) (4) Rel.Perf -0.092 (0.097) -0.056 (0.126) -0.176* (0.097) -0.152* (0.091) Female 0.004 (0.022) 0.008 (0.046) 0.008 (0.034) 0.011 (0.037) Rel.Perf*Female -0.788 (0.987) -0.777 (1.438) -0.727 (1.440) -0.697 (1.444) Ln(Assets) -0.005 (0.005) -0.005 (0.005) Ln(Age) 0.045 (0.038) Leverage 0.024 (0.043)

(28)

There remains one key obstacle on interpreting our coefficient of interest δ3 in the previous

Logistic model, which is the endogeneity of firm’s selection of performance peers. For example, especially for the last two definitions of peer groups, firms have their own selection criteria for the choice of their peers and it is hard to investigate whether the included peers are relevant or not. For the firms with an entrenched CEO, no matter what gender they are, these CEOs have the ability to choose the peers who are always underperforming relative to themselves, which make them less likely to be fired even if under the situation that they do perform very poorly during that period. Therefore, it is not surprising to see an insignificant sign on the coefficient of interest δ3 for the last two measure of peer performance in this case.

In order to overcome this potential endogeneity problem, we further implement the IV methodology and both the Kleibergen-Paap F-stat results of the first-stage regression and the coefficient estimates from the second-stage regression in equation (3) are presented in Table 5. As described above, the firm’s performance (Perf) is instrumented by the industry shock (Ind.Shock) and the interaction term (Perf.Female) is instrumented by the Ind.Shock*Female. Column (1) of Table 5 firstly display the results on baseline version of the second-stage regression without controlling for fixed effects and some CEO or firm characteristics. It shows firms that confront the negative industry shock in the year are more likely to fire the current CEO in this year. This association remains its significance in other three Columns by including fixed effects and more control variables. However, the positively significant relationship between the female CEO in general with the forced CEO turnover rate is only be found in Column (1).

Notably, all estimated coefficients on the interation term (Perf.Female) are negative but not significant in Column (1) through (4). Even though the negative sign illustrates that female CEOs who face the negative industry shock and thus underperform tend to be fired sooner,

(29)

the insignificance of this sign shows that we fail to reject the null hypothesis. It means the potential gender effect problem does not exist in the senior corporate firing case and the board does not treat female CEOs significantly different from male CEOs. Nevertheless, all the F-stat of the first-stage in Column (1) through (4) are around 6 to 8, which are not sufficiently high (>9) and thus we are not able to infer the instrumental variable we make as a valid or strong instrument. Under this weak instrument setting, it would not be surprising that we do not find any significance with the IV methodology.

To sum up, the IV results conflict with the results from the previous absolute underperformance model and the first peer group performance measure under the relative underperformance model, while they are similar with the last two measures under the relative underperformance model. However, the instrument is not strong enough for interpreting these IV results validly. But by combining with the previous models’ results, we could conclude that there is only weak evidence on the gender effect existence in the corporate firing channel by studying the forced CEO turnover rate.

(30)

Table 5.

Forced CEO Turnover and the Gender Effect.

This table presents both the Kleibergen-Paap F-stat results of the IV first-stage regression and the results of the IV second-stage regression in equation (3) during the sample period 2000-2015. The instrumental variable is the industry shock, which is measured by the weighted average accounting return of the SIC2-level industry excluding the firm itself in the year.

The dependent variable is the forced CEO turnover rate (Forced). Forced equals 1 for firms which fire the CEO in this year. Control variables include ln(assets), ln(age) and leverage. Column (2), (3) and (4) include all SIC2-level industry fixed effects and year fixed effects and column (4) include all control variables. All robust standard errors are clustered at the firm level and presented in parentheses. *, **, and *** indicate significance level at 10%, 5% and 1% respectively.

(1) (2) (3) (4) Perf -0.086*** (0.032) -0.078*** (0.030) -0.102*** (0.029) -0.102*** (0.029) Female 0.017** (0.008) 0.012 (0.009) 0.011 (0.008) 0.005 (0.008) Perf*Female -0.089 (0.106) -0.093 (0.110) -0.069 (0.106) -0.057 (0.109) Ln(Assets) 0.003*** (0.001) 0.002*** (0.0009) Ln(Age) -0.031*** (0.008) Leverage 0.005 (0.008) F-stat of the first

stage

6.203 6.994 8.386 7.907

(31)

6. Conclusion

The main goal of this paper is to investigate the potential role of gender effects in senior corporate firings, which is measured by a forced CEO turnover event. We exploit those events by comparing the difference between the female forced CEO turnover rate with the male forced CEO turnover rate under the situation when both CEOs are equally underperforming.

We first build a simple Logistic model by using the absolute accounting return of underperformance as a proxy of CEO’s underperformance and find that female CEOs who are underperforming absolutely on average obtain a significant increase in the forced turnover rate compare to male CEOs under the same situation.

Next, we improve this model by introducing the relative version of underperformance, since the absolute version of underperformance contains some noise for the purpose of CEO quality assessment. The relative underperformance is measured by using three different kinds of peer groups, which are the industry peers, the product market/competitor peers and the relative performance peers respectively. Only the first measure- the industry peers generate similar results as the absolute underperformance. That is, male CEOs are valued relatively more than female CEOs and thus face a lower chance to be fired when they underperform equally to a female CEO relative to the industry peers. The other two measures of relative performance obtain different results in that they do not find any significant relationship between the forced CEO turnover rate and the gender of the CEO. This could be due to the endogeneous selection issue that the entrenched CEO, regardless of their gender, might choose the product market peers or the relative performance peers who are always underperforming worse compared to themselves in order to keep his or her position during the bad time. Both of these two measures have their own selection criteria that we could not

(32)

observe entirely and thus it is much easier for these entrenced CEOs to apply their power on selecting the peers they prefer, which lead to an insignificant relationship for their forced turnover rates and their gender. Or this insignificant relationship could be explained by the choice of the board of selecting the peer groups that it is more common for the board to select the industry peers as the peer group but the other two kinds of peer groups are not very relevant in the CEO evaluation.

Moreover, we improve one step further to take into account the potential endogeneity problem in the design of these Logistic models, which is to design a two-stage instrumental variable (IV) regression model and test the core hypothesis by implementing this IV approach. The results are not consistent with the findings from the absolute underperformance model setting and the first measure of relative underperformance model setting. Instead, they are similar with the Logistic model based on the two other kinds of definitions for relative underperformance, which states that female CEOs are not treated significantly different from male CEOs in turnover decisions. However, these insignificant results from the IV methodology come from a weak design of the instrument variable, which might lead to these interpretation less meaningful and powerful.

Therefore, the overall results show some new but weak support of the existence of a gender effect in dismissal decisions at the senior management level. Yet these results help to fill in the gap in the gender effects research and bring some new insights for the policy makers who would like to deal with the gender imbalance in the corporate world. Since there is no very strong results to show the existence of the gender effect in the corporate firings, policy makers are advised not to take aggressive actions on encouraging the female representation in the senior management level.

(33)

Moreover, the paper tests the hypothesis by assuming the board will fire the current CEO in the current year if he or she underperforms this year, while it does not consider the situation that the board might not terminate the current CEO immediately but would make the dismissal decision based on the last several years’ performance.

Besides, as the paper finds very weak support for a gender effect to exist in the corporate firing cases, it might also be worthwhile to further explore whether the CEO with more power would lead to different results. For instance, the female CEO with the longer tenure might also face a lower probability to be fired if she performs very poorly compared to the peers, since she might have built a long-term relationship with the board and the firm. Therefore, the potential gender effects would disappear under the situation that both male CEO and female CEO with more power. Future research could be done in this area.

(34)

7. Reference

Ahern, K. R., & Dittmar, A. K. (2012). The changing of the boards: The impact on firm valuation of mandated female board representation. The Quarterly Journal of Economics, 127(1), 137-197.

Albuquerque, A. (2009). Peer firms in relative performance evaluation. Journal of Accounting and Economics, 48(1), 69-89.

Bass, B. M., & Avolio, B. J. (1994). Shatter the glass ceiling: Women may make better managers. Human Resource Management, 33(4), 549-560.

Bear, S., Rahman, N., & Post, C. (2010). The impact of board diversity and gender composition on corporate social responsibility and firm reputation. Journal of Business Ethics, 97(2), 207-221.

Bell, L. (2005). Women-Led Firms and the Gender Gap in Top Executive Jobs. IZA Discussion Paper No. 1689. Available at SSRN: https://ssrn.com/abstract=773964

Bertrand, M., & Mullainathan, S. (2001). Are CEOs rewarded for luck? The ones without principals are. The Quarterly Journal of Economics, 116(3), 901-932.

Bizjak, J. M., Kalpathy, S. L., Li, Z. F., & Young, B. (2017). The role of peer firm selection in explicit relative performance awards.

Bugeja, M., Matolcsy, Z. P., & Spiropoulos, H. (2012). Is there a gender gap in CEO compensation?. Journal of Corporate Finance, 18(4), 849-859.

Faccio, M., Marchica, M. T., & Mura, R. (2016). CEO gender, corporate risk-taking, and the efficiency of capital allocation. Journal of Corporate Finance, 39, 193-209.

Francoeur, C., Labelle, R., & Sinclair-Desgagné, B. (2008). Gender diversity in corporate governance and top management. Journal of Business Ethics, 81(1), 83-95.

Gibbons, R., & Murphy, K. J. (1990). Relative performance evaluation for chief executive officers. ILR Review, 43(3), 30-S.

Jenter, D., & Kanaan, F. (2015). CEO turnover and relative performance evaluation. The Journal of Finance, 70(5), 2155-2184.

Kaufmann, G., Isaksen, S. G., & Lauer, K. (1996). Testing the “glass ceiling” effect on gender differences in upper level management: The case of innovator orientation. European Journal of Work and Organizational Psychology, 5(1), 29-41.

Martin, A. D., Nishikawa, T., & Williams, M. A. (2009). CEO gender: Effects on valuation and risk. Quarterly Journal of Finance and Accounting, 23-40.

Mohan, N. J., & Chen, C. R. (2004). Are IPOs priced differently based upon gender?. The Journal of Behavioral Finance, 5(1), 57-65.

(35)

Oakley, J. G. (2000). Gender-based barriers to senior management positions: Understanding the scarcity of female CEOs. Journal of Business Ethics, 27(4), 321-334.

Parrino, R. (1997). CEO turnover and outside succession a cross-sectional analysis. Journal of financial Economics, 46(2), 165-197.

Parrino, R., Sias, R. W., & Starks, L. T. (2003). Voting with their feet: Institutional ownership changes around forced CEO turnover. Journal of financial economics, 68(1), 3-46. Peters, F. S., & Wagner, A. F. (2014). The executive turnover risk premium. The Journal of Finance, 69(4), 1529-1563.

Singh, V., & Vinnicombe, S. (2004). Why so few women directors in top UK boardrooms? Evidence and theoretical explanations. Corporate Governance: An International Review, 12(4), 479-488.

Diagnosing the gender effect : evidence from the forced CEO turnover rate