Feminist share prices, a report on empowered returns;
Does the effect of a newly appointed CEO’s sex influence a firm’s short-term performance?
By Sjoerd Läubli, 11000961
Bachelor Thesis Finance & Organisation
Thesis Supervisor: J. D. Jagau
26 June 2018
Abstract:
Board diversification and female representation in the corporate world in general is the subject of many discussions worldwide. This paper tries to establish if gender bias can be found during the announcement period of the Chief Executive Officer succession. The database is comprised of firms with CEO succession announcements after 2010 and that are indexed in the S&P500. An empirical study using simple linear regression has been performed, where a relationship between the dummy variable for a newly appointed CEO’s sex and a firms Excess Returns on the market was found. This result of this study indicates a positive effect of announcing a Female CEO on the short-term performance of a company.
CONTENT
INTRODUCTION 3
LITERATURE REVIEW 5
Regression Model 7
Dependent Variables 7
Independent Variables 8
Control Variables & Bias Control 8
Peer Group Establishment & Data Gathering 9
Robust Regression Analysis 12
Stata Progress 12
RESULTS 13
Linearity 13
Normality 13
Multicollinearity 13
Descriptive Statistics 14
Regression Results & Discussion 14
CONCLUSION & DISCUSSION 17
Conclusion 17
Discussion 17
BIBLIOGRAPHY 19
APPENDICES 26
Appendix 1 26
Appendix 2 27
Appendix 3 28
Appendix 4 29
Appendix 5 30
Introduction
Four years after the instalment of a 40% female board members quota for large Norwegian companies in 2008, a European commission proposed a similar regulation which also contained a quota of at least 40% female board members for large companies in the entire European Union (The Economist, 2018) (European Commission, 2012). This regulation still hasn’t been approved, however, in 2015 Germany introduced a quota for large German companies with more than 2000 employees to have at least 30% female board members, ensuring female representation in the corporate field (Smale & Miller, 2015). The debate in both the USA and Europe on the instalment of gender quota in corporate law and the consequences for the firms involved has been ongoing until today (Stalet, 2016) (Boffey, 2017). Female representation in the corporate environment is growing. Between 2007 and 2017 the percentage of S&P500 companies board seats occupied by women rose from 16 to 22 (Lublin, 2018). The amount of female Chief Executive officers (CEO’s) rose from 12 in 2007 to 23 in January 2017 (Lublin, 2017) (Catalyst, 2018).
Although the effect on firm performance of appointing female board members and CEO appointments in general have been researched multiple times, the effect of the succession from a male to a female CEO has not yet been studied thoroughly. This raises the question if gender bias can also be found in the progress of CEO succession, and if the sex of a CEO is a relevant variable for firm performance in the short run. This thesis will consider the following research question:
“Does the sex of a newly announced CEO influence the firm performance in the short run during the succession period for firms, indexed in the S&P500, after 2010?”
Note that the terminology of sex and gender are very different. In this paper gender refers to behaviour and capabilities expected of or demonstrated by an individual that comply with the expectations regarding their sex of both themselves and their surroundings. The sex of an individual refers to the anatomy of a person’s reproductive system (Muehlenhard & Peterson, 2011, pp. 795-799). This thesis will only employ a person’s sex as a differentiating factor between male and female CEOs.
The Chief Executive Officer could be regarded as the single most important position of a company. A CEO must be able to analyse the company, to accurately allocate responsibilities and recognize causalities. The CEO’s role is unique due to its broad monitoring and controlling, where as other position in the company tend to have a more narrowed focus. Because of this the Chief Executive Officer is usually held accountable for changes in the company’s performance (Lafley e.a., 2009). Therefore, analysing the effect of a newly announced CEO’s sex on firm performance is crucial for understanding both a CEO’s responsibility and the consequences of gender bias.
Based on the Literature studied, a negative correlation between a female CEO announcement and firm performance is expected, because both diversification of boards and CEO succession is negatively correlated to short term stock returns, as will be discussed below. The two hypotheses of this paper are:
Hypothesis 1: A significant negative effect of a female CEO succession announcement on firm performance measured in a percentage change of returns, of two months prior and after the announcement, will be found.
Therefore, the dummy for a female CEO succession announcement (dummy_FemSucc) will have a negative coefficient that significantly differs from zero.
Hypothesis 2: A significant difference can be found between firm’s performance, after the announcement, of companies with a newly announced female CEO relevant to companies with a newly announced male CEO.
Therefore, the coefficient of the dummy for a newly announced female CEO (dummy_FemCEO) will also negatively differ from zero.
This Paper will consist of 5 chapters; The next chapter will consist of a summary and review of all the studied literature that is relevant to this research. The third chapter will include both the theoretical framework, were the models, terminology and statistics used are explained and the research methods applied to gather data and achieve results. After this the results of these methods will be presented and discussed. In the final chapter a short summary will be given and a conclusion formed. A discussion on improvements and further research will also be included there.
Literature review
In this chapter an outlay of previous research related to this topic is given. Starting with multiple papers written on board diversification, after which CEO succession effects and a control variable will be reviewed.
Several studies have been done on the diversification of boards. In 2005 Farrel & Hersch published a paper on the drivers of board diversification and the characteristics of the firm in which a woman is most likely to be appointed a board member. The research suggests that, even though more women serve on well performing firms, the most important driver is an internal or external demand for diversification. Thus, the reasoning behind this decision is not performance based (Farrel & Hersch, 2005). Multiple effects of a diversified board on the company have been found. A paper in 2009 concluded that diversification leads to better attendance rates of board members, an increase in effort allocation and higher willingness to join committees from the board (Adams & Ferreira, 2009). Dobbin & Jung (2011) concluded that the decision to hire female board members usually has a neutral or negative impact on firm performance. This could either be caused by decreased monitoring and controlling capabilities of the board itself or because of institutional investors’ gender bias. This would mean that the performance expectations of a female board member differ from the expectations of a male board member’s performance (Dobbin & Jung, 2011). Based on data from Norwegian companies after the implementation of the women quota, hiring female board members because of regulations or as a reaction to demands for diversity can also have a negative impact on firm performance due to the lack of experience of female candidates (Ahern & Dittmar, 2012). However, in 2014 Gregory-Smith, Main & O’Reilly found that although diversified boards behave differently, the bottom line performance of a firm does not change.
The effects of diversification in boards on companies has been researched many times with contradictory results, on the other hand the amount of papers on the effect of appointing a female Chief Executive Officer (CEO) is limited. In 1987 Beatty & Zajac concluded that Chief Executive Officer succession announcements in general have a negative impact on the firm’s value, measured in stock prices. This is supported by a paper in 2017 that found a negative correlation of CEO succession with the company’s stock prices in the short run (Schepker, Kim, Patel, Thatcher & Campion, 2017). The nature of this relationship between CEO succession and firm performance had already been researched in 2010 by Chang, Dasgupta & Hilary, who concluded that the firm’s performance is mainly influenced by the expectations of the new CEO’s abilities. This would support the notion that short run effects of a CEO’s sex on firm performance could be caused by gender bias. In other words, the investors have different expectations of male and female CEO’s behaviour and capabilities and this would also keep in line with Dobbin & Jung’s findings on the effects of appointing female board members. An actual difference in behaviour was found by Eahab & Ursel (2011) who concluded that female Chief Executive Officers tend to be more risk averse than their male counterparts and therefore are associated with a decrease in risky investments.
Another variable that should be considered when researching the effect of a newly appointed CEO on firm performance is whether the executive was chosen from within the company or its industry, as insider successions tend to have a better market responds in the short run (Jalal & Prezas, 2012).
Research Methodology:
In this chapter the theoretical framework is given. First the model used is presented and the variables explained. Thereafter the control variable and different actions to eliminate bias are introduced. Following this the research methods applied for the peer group establishment and data gathering are stated. Finally, the two kinds of regression analyses and the Stata procedures are given.
Regression Model:
The regression models used in this paper is a simple linear regression model, with different specifications for both hypothesis, the basic model is:
〖Firm Performance〗_i= α_i+ β_1*D_Female+ β_2*D_Outsider+ ε_i
To be able to test both hypotheses, Firm Performance is specified in two ways. Both are based on the same data, however, for the first hypothesis the data is remodelled into a ratio. The specific models for the hypotheses are:
Hypothesis 1:
〖Change in Average Returns Ratio〗_i= α_i+ β_1*D_Female+ β_2*D_Outsider+ ε_i
Where 〖Change in Average Returns Ratio〗_i (ratio_change) is the dependent variable, D_Female (dummy_FemSucc)is the independent dummy variable for a female CEO succession announcement and D_Outsider (dummy_Outsider) is the dummy control variable for Outsider CEO succession announcement.
Hypothesis 2:
〖Daily Excess Returns〗_i= α_i+ β_1*D_Female+ β_2*D_Outsider+ ε_i
Where 〖Daily Excess Returns〗_i (r_daily) is the dependent variable, D_Female (dummy_FemCEO) is the independent dummy variable for a female CEO succession announcement and D_Outsider (dummy_Outsider) is the dummy control variable for Outsider CEO succession announcement.
Dependent Variables:
The dependent variable is different for each hypothesis. In the first hypothesis, in order to take both the succession and the CEO’s sex effect into account a time ratio of the average daily return had to be included as the dependent variable. For each company the average return before and the average return after was calculated, after which the ration was computed by the standard method of: (〖Average Daily Return〗_(after )- A〖verage Daily Return〗_before)/〖Average Daily Return〗_before . By using the ratio, it is possible to analyse the difference in returns caused by the succession and the influence of the dummies, dummy_FemCEO and dummy_Outsider, in the same regression.
For the second hypothesis the unchanged daily excess returns of each company can be used as the dependent variable, because the influence of the succession itself is not included into the model. For this regression only, the daily excess returns after the succession announcement of each company are used, therefore only testing whether a company with an announced female CEO does better than a company with an announced male CEO after the market is informed of the planned succession.
Independent Variables:
The Independent variable of the model is a dummy which indicates the sex of an announced CEO. For a female CEO announcement D_Female= 1, for a male CEO announcement D_Female=0.
For the first hypothesis model the dummy, dummy_FemSucc, is applied to all the company’s ratios. For the second hypothesis model the dummy, dummy_FemCEO, is applied to all the daily excess returns of each company after the succession announcement.
Control Variables & Bias Control:
The Control Variable used in both regressions is the dummy variable dummy_Outsider) indicating the announcement of a succession with an Outsider CEO. An Outsider succession is defined as a CEO succession were the candidate is not promoted from within the company but appointed from outside the company and sometimes even from outside the industry (Jalal & Prezas, 2012). When a CEO is selected from within the company dummy_Outsider = 0 and for an Outsider succession announcement dummy_Outsider = 1.
As individual returns would also be affected by different systematic risk of the market, the dependent variable is based on the excess returns, limiting the differences between the companies to idiosyncratic risk (Berk & DeMarzo, 2014). This would undermine the effect of changes in the market’s economy over time and therefore, render the use of a control variable for time in the model unnecessary.
Another factor that might be of influence on the dependent variable Firm Performance could be significant differences in both bias and average excess returns between firms. A, for example, a sample with more male to male CEO successions announcements of hedge fund companies could give a skewed image of returns between the two peer groups. Also, the bias might differ, due to more male dominated industries. As the sample for male to female CEO successions is relatively small, it makes sense to match the industries of each company in this peer group with two companies within the same industry with male to male CEO announcements. Thereby undermining the industry effects on both the possible gender bias and the excess returns. Thus, each company with an announced female CEO is “matched” with two companies with an announced male CEO. A secondary principle of this matching was the level of competition between companies, where the biggest competitors within a company’s industry were prioritised and their CEO successions were examined first.
A final restriction on the CEO announcements used in the sample was the omission of appointments caused by the forced departure of the previous CEO. As Lafley e.a. (2009) pointed out the CEO is often held accountable for a change in the performance of the company. Therefore, a CEO might be forced to leave due to a company’s decreased performance. On top of that public scandals regarding the CEO’s decisions may also influence the company’s performance while also causing a forced departure (Lafley e.a., 2009).
Peer Group Establishment and Data Gathering:
As no CEO succession database could be accessed, the information was gathered manually. Firstly, a list of Female CEO’s in the S&P500 by Catalyst was the basis for the male to female CEO successions peer group. Thereafter four main information sources were used to attain reliable information on the successions. The announcement date is used as the event date for succession, as it is likely that the market will respond as soon as the announcement is made (Jalal & Prezas, 2012, p. 409). Dates and general info on the succession were gathered from either the Wall Street Journal, PR Newswire or a press release by the company. When the necessary time frame and reasons for succession were established, all executive profiles were regarded in the online Bloomberg database to establish if the newly appointed CEO was promoted from within the company or not.
The peer group for male to female CEO announcements contains 22 companies. After applying the matching principles and gathering information using the same sources the male to male CEO succession announcements peer group contains 44 companies. The total sample for this research contains 66 companies in 14 different Industries.
Th excess returns of each company are obtained from CRSP, a partner of Wharton Research Data Services. CRSP offers a direct download of a company’s excess return on the S&P500 index, which in this paper is used as the market index. Daily excess returns of two months prior and two months after the announcement were obtained for each company, providing between 81 and 88 datapoints per firm. The data was collected in excel and lined by the announcement date. The change in average excess returns ratio was also calculated in excel. All companies’ information on the announced CEO’s sex and Insider or Outsider appointment was then compiled into columns for both the Excess Returns after announcement and the ratio’s, creating the basis for the dummy regressions in Stata.
Table 1; The peer group of Companies who announced a succession of a female CEO between 2010 and 2018. Information Sources can be found in the bibliography.
Table 2; The peer group of Companies who announced a succession of a male CEO between 2010 and 2018. Information Sources can be found in the bibliography.
Robust Regression Analysis
Using an Ordinary Least Squares (OLS) analysis to establish the coefficient of an independent variable is problematic when the data sample includes outliers or influential observations. Outliers can be checked by creating a scatter plot and observing if largely deviating data are present. Influential observations are detected in the same way (Keller, 2012, pp. 650-651).
As the scatter plot of the Change in Average Returns Ratios data shows at least one definite Outlier, namely the ratio of the 40th company of the graph. Applying the standard OLS regression would give a biased result. A robust regression analysis sets barriers for the value of data points and eliminates outliers and influential observations, ensuring the fit of the OLS line to all the data in the sample (Keller, 2012, 651-652). The data in the scatter plot of the excess returns on the S&P500 after the announcement is not that easily analysed. Therefore, to ensure a non-biased result, both standard and robust regression analysis are run on the data in Stata for both hypothesis model 1 and hypothesis model 2. Scatter Plots can be found in the Appendix.
Stata process
In Stata the two regressions models are tested. For both models two dummies are generated. For the first regression the two dummies are dummy_FemSucc and dummy_Outsider, and for the second regression the dummies generated are dummy_FemCEO and dummy_Outsider.
After this the independent variables are regressed with their dummies as the dependent variables together and again without the control variable (dummy_Outsider). Both robust and standard regression are performed for both regressions. In total eight regressions are run.
The barrier used is an α of 5%. As it is not known if the effect might be positive instead of negative, the test is two-sided the coefficients are considered significantly different from zero when the given p-value<0.025, ((α=0.05)/2=0.025).
Results
In this chapter the results will be presented and discussed. First the first three OLS assumption of the data will be tested. Second the descriptive statistics from the robust and standard regressions will be given per dependent variable. Third the results of the regressions will be discussed, and significance analysed.
Linearity
Linearity could be assumed from a time series data of companies returns. Linearity is checked by creating a scatterplot with a trend line. As our independent variable is a dummy, a trend line prognosis has been added to get a clearer view of the linear relationship. To make sure the trend line isn’t biased, the outlier has been omitted from the Ratios data. As seen from the scatterplots of both the Change in Average Returns Ratios and the Excess Returns on the Market a linear relationship does exist. Therefore, linearity is ensured.
The Scatter Plots can be viewed in the Appendix 1.
Normal distribution
To check if the datasets are normally distributed a Histogram of both datasets is computed. A normal distribution has the typical “wave” shape. As the Change in Average Returns Ratios are calculated from the Excess Returns on the Market it can be assumed that if the condition of normality holds for the second, the first will also be normally distributed.
The histogram of the Excess Returns on the Market has a very distinct normally distribution shape. The Change in Average Return Ratios histogram has a less clear distribution; however, it does not seem to conflict with normality and as it is a computation of the Excess Returns normality can be assumed.
The Histograms can be viewed in the Appendix 2.
Multicollinearity
To make sure the independent variables are not correlated with each other to a degree that would cause biased results, correlation matrixes for both sets of dummies have been created. Multicollinearity exist if the two dummy variables have a correlation that exceeds 0.8. As can be seen in the two matrixes, for both models, multicollinearity does not exist.
The Correlation Matrixes can be found in the Appendix 3.
Descriptive Statistics
Table 3; The number of observations is 66 for the first model. Both the dependent and the two independent dummy variables are included, representing the mean values, standard deviations, minimum and maximum sample values. The Outlier has not been omitted.
Table 4; The number of observations is 2979 for the second model. Both the dependent and the two independent dummy variables are included, representing the mean values, standard deviations, minimum and maximum sample values.
Table31 gives the summary statistics for model of the first hypothesis. The mean of the dependent variable (0.5516323) implies an average increase of returns after the succession of 55%. The means of the dummies give a portion of 33.3 % female CEO announcements, exactly 〖1/3〗^rd) and 16.7% outsider CEO announcements (exactly 〖1/6〗^th). The maximum given in the table is an increase in return of 2242.2%, this would be regarded as an Outlier.
The fourth table gives the summary statistics for the second hypothesis model. The mean of the dependent variable (0.0004337) shows an average daily excess return above the market of 0.043% after the succession announcement. The percentages for the dummies slightly differ from the first table. This is caused by the different amounts of daily returns per company due to the succession announcement having taken place on different dates for each company. However, the differences do need exceed 0.3%, thus the portions of female CEO and outsider CEO announcements would still be close to respectively 〖1/3〗^rd and 〖1/6〗^th of the sample. The minimum value of -6.7% deviates a lot from the average daily return, this again advocates the use of robust regressions.
Regression Results and Discussion
The results will be discussed for each Hypothesis model separately. For both models the robust regressions gave significantly different outcomes. Therefore, only the robust regression results will be discussed. The standard regression results for hypothesis models 1 and 2 can be found in Appendix 4 and 5 respectively.
– Hypothesis 1;
Table 5; The robust regression results with dependent variable Change in Average Returns Ratios (ratio_change). Independent dummy variables for female CEO announcement and Outsider CEO announcement.
Table 6; The robust regression results with dependent variable Change in Average Returns Ratios (ratio_change). Independent dummy variable for female CEO announcements. The control variable is omitted.
Table 5 and 6 show the results of the regressions run for the first Hypothesis model, where the Change in Average Returns Ratios is the dependent variable. As shown in the table, the dummy variable for a female CEO announcement has a coefficient of 1.1206 with the control variable and 1.1371 without the control variable. Both coefficients are of significant influence, as (p=0.008<0.025) and (p=0.007<0.025). Therefore, there is a definite effect of the sex of an announced CEO on the ratios. However, the coefficients have a positive influence on the returns ratios. This is the opposite to what was expected in the hypothesis. In Table 3 the control variable does not have a significant effect on the independent variable (p=0.371>0.025). This will be discussed in the final discussion of the results.
– Hypothesis 2;
Table 7; The robust regression results with dependent variable Daily Excess Returns on the S&P500 (r_daily). Independent dummy variables for female CEO announcement and Outsider CEO announcement.
Table 8; The robust regression results with dependent variable Daily Excess Returns on the S&P500 (r_daily). Independent dummy variable for female CEO announcements. The control variable is omitted.
As the first regression model also took into account the returns of a company before announcement, the second regression gives a narrower image of a company’s performance after the announcement only. The dummy for a company with an announced female CEO has a coefficient of 0.0008246 with the control variable and 0.008156 without the control variable. Again, the control variable itself does not have a significant coefficient as p=0.603>0.025). In this case adding the control variable does not even change the significance of the independent variable. The coefficients for a female CEO announcement are significantly different from zero, suggesting a definite effect of the sex of an announced CEO on the short-term performance of a company.
– General Results;
For both hypothesis the coefficients deciding the influence of an announced CEO’s sex are significantly different from zero and positive. Thus, a definite advantage of female CEO succession announcement on the company’s short-term performance is implied. This result is contradictory to the literature studied about board diversification and gender bias at time of appointment for other positions.
The control variable for outsider CEO succession announcement is not significant, and thus cannot be regarded as different from zero. This result is contradictory to the findings of Jalal & Prezas (2012), who found a definite negative influence of outsider CEO succession on a company’s returns.
No evidence is found for both the first and second hypothesis. However, the null-hypothesis of there being no influence of a CEO’s sex is also no longer supported, as we have found significantly positive coefficients. These results call for more research into this topic, as will be suggested in the discussion.
Conclusion and Discussion
The final chapter is separated into two parts. First a summary of the research topic and results is provided, from which a conclusion is drawn. Second a discussion on this research and suggestions for further research are given.
Conclusion
This thesis tried to find an answer to the question if the sex of a newly announced CEO influences the short-term performance of companies, indexed in the S&P500, after 2010. Two hypotheses were formed to answer this question. The first hypothesis regarded the effect of a CEO’s sex on the difference in performance of each company before and after the succession announcement. The second hypothesis analysed the difference in performance between firms with an announced female or an announced male CEO succession. The expectations for both hypothesis, based on literature studied, was a negative effect of a female CEO succession announcement on the firm’s short-term performance. Robust regressions were run with a dummy variable for the CEO’’s sex. The coefficients found for hypothesis 1 and hypothesis 2 were significantly different from zero. However, both coefficients had a positive value, indicating a positive gender bias for announcing a female CEO. These results are contradictory to the expectations contrived from the studied literature, and no evidence could be found for hypothesis 1 and 2. Nevertheless the null-hypothesis of there being no effect of an announced CEO’s sex on firm performance is also rejected. To be able to conclude with certainty that announcing a female CEO has a positive effect on short term firm performance more female CEO succession must be studied.
Discussion
As the amount of female CEO successions in the S&P500 is still very limited this research can only be considered an indication of the relationship in reality. By filtering the successions used in the database the internal validity was improved. However, this has a negative effect on the external validity, as CEO succession in industries were no female CEO announcements were made were left out of the regression analyses. To be able to extend this external validity research could be done by getting access to a Chief Executive Officer succession database and regressing the female CEO announcements on male CEO announcements of all industries in the S&P500. One could even go to the extent of including both female and male CEO succession in multiple market Indexes from different regions, for example the indexes of the largest companies on the Asian or European stock exchanges.
The control dummy variable for outsider CEO succession was not significantly different from zero for all the regression in this thesis. This may also be due to the relatively small portion of outsider CEO announcements included into the database, as Jalal & Prezas (2012) did find a large and significant coefficient when testing this variable on a larger database. Excluding the variable from the results of this thesis would therefore have been a dubious decision, even though it is common to omit non-significant regressors.
To extend the inside in the effects of a CEO’s sex on firm performance one could also compute the excess returns over both the short and the long term, to analyse if the bias found time-resistant. The literature studied on diversified board suggests no change in the bottom line performance of a company in the long run, this could be tested for the CEO position too.
A final suggestion for future research could be to analyse if there is a correlation between the measure of diversification of a company’s board and the appointment of a female CEO, as one of the main responsibilities of a bard is the selection and appointment of a new Chief Executive Officer.