CEO turnover in a financial crisis
Abstract
This paper investigates whether there is a relationship between CEO turnover and firm
performance and whether CEO turnover is different during a financial crisis compared to a non-crisis period. It uses logit regressions with CEO turnover as dependent variable. It uses the financial crisis started at the end of 2007 as the crisis period. This crisis is known as the Global Financial Crisis. Data is used for a time period from 1996 to 2012 for the S&P1500 firms. It finds evidence at a 1% level that there is a negative relationship between CEO turnover and firm performance one year prior to the CEO turnover. It finds no evidence for a relationship between CEO turnover and firm performance of more than one year prior to the CEO turnover. Moreover, it finds that the influence of firm performance on CEO turnover is large when the performance is either extremely good or extremely bad. This paper also finds that a CEO is replaced less often in a crisis compared to a non-crisis period. Moreover, it finds that the difference in CEO turnover between a non-crisis and a non-non-crisis period is the largest for firms with average performance.
Key words: CEO turnover; firm performance; Global Financial Crisis
Name: Robin van Kleef
Student number: 10049932
Thesis: Master thesis
Specialization: Finance
2
1. Introduction
The CEO turnover has become important in recent decades. One of the most important decisions of the board of directors is the decision whether to retain or to replace its CEO. The Global Financial Crisis which started at the end of 2007 creates interesting research opportunities to investigate whether CEO turnover is different in a financial crisis compared to a non-crisis period.
This paper investigates whether there is a relationship between CEO turnover and firm performance. CEO turnover is defined as a change in the CEO position within a firm. This paper examines whether a firm changes its CEO after disappointing firm performance. Several researches about this subject are done before. Warner et al. (1988) and Weisbach (1988) find a negative relationship between CEO turnover and firm performance one year prior to the CEO turnover. They did not find a relationship between CEO turnover and firm performance of more than one year prior to the CEO turnover. They find that firms react relatively quick with changing their CEO after disappointing firm performance. Denis and Denis (1995) find a negative relationship between CEO turnover and firm performance of three years prior to the CEO turnover. They find that firms don’t react quickly with changing their CEO after disappointing firm performance since the performance already has been disappointing for several periods.
Then this paper investigates whether there is a relationship between CEO turnover and a financial crisis independent of its performance. It asks whether a CEO is changed more or less often in a financial crisis. In a financial crisis, firm performance is on average worse compared to a non-crisis period. This paper will control for this influence on CEO turnover so it can examine the non-crisis period. Jenter and Kanaan (2006) find that CEO turnover is higher when industry performance is low. The CEO may be fired when the whole industry is performing badly, even when the
performance of a firm is not bad relative to the other firms within its industry. This may lead to a higher CEO turnover in a financial crisis. On the other hand, Morck et al. (1988) and Cornelli et al. (2013) find that CEO turnover is not higher when industry performance is low. This may lead to the same CEO turnover in a crisis compared to a non-crisis period. It’s also possible that a CEO is less likely to be replaced during a financial crisis, because firms may act more risk averse during a financial crisis. Replacing a CEO is risky because the new CEO may perform even worse.
The models that will be used to investigate these relationships are logit models with CEO turnover as dependent variable. It will use return on assets, return on equity and share return as performance measures. The financial crisis will be included in the model with a dummy variable that equals 1 when the observation is in the year of the financial crisis and 0 otherwise. The Global
3
Financial Crisis which started at the end of 2007 will be used as crisis period. This paper also includes several control variables and these will be discussed extensively. The models will be used to answer the following research questions: Is there a relationship between CEO turnover and firm performance? Is CEO turnover different in a financial crisis compared to a non-crisis period?
This paper contributes to the literature in several ways. Firstly, there is little research done about the relationship between CEO turnover and a financial crisis. This is probably the most
important contribution of this paper. This paper investigates whether the Global Financial Crisis has an effect on the CEO turnover of firms. This hasn’t been investigated before. Secondly, this paper uses the time period from 1996 to 2012 in its sample and this hasn’t been used before. For the relationship between CEO turnover and firm performance, most research is done before 2000. Thirdly, this paper uses a large sample. Most researches have under 1000 CEO turnovers in its sample (Warner et al. 1998, Denis and Denis, 1995 and Weisbach, 1988) . This paper has over 2700 CEO turnovers in its sample. Finally, the models that are used to answer the research questions are different from the models used before. It uses the performance measures that are commonly used and it will also use different performance measures. The model also includes control variables like firm size, board size, board independence, age of board members and CEO characteristics.
This paper is structured as follows. Section 2 discusses the theoretical framework of this research and it will build hypotheses. Section 3 describes the data, methodology and descriptive statistics of this paper. Section 4 shows the results and robustness and multicollinearity checks. Finally, section 5 concludes and discusses the results.
2. Related Literature
This section describes a theoretical framework about the relationship between CEO turnover and a financial crisis. It starts with describing the composition and the role of the board of directors in a firm. Then the section continues with several theories about the relationship between CEO turnover and firm performance. Next the focus will be on the relationship between CEO turnover and relative firm performance. Then the section examines theories and empirical findings about other factors that may affect CEO turnover. Finally, the section will build hypotheses based on past research.
4
2.1 The composition and the role of the board of directors
Most US firms are not managed by its owners but by a management team. This separation of ownership and control creates an agency problem. Jensen and Meckling (1976) find that managers in a firm not always make value maximizing decisions. Fama and Jensen (1983) conclude that this separation of ownership and control leads to decision systems that separate decision management from decision control. The most used system within a firm is a firm with an one tier board structure. The firm has a board of directors with executives and non-executives. The board of directors has the power to hire, fire and compensate the management (Baysinger and Butler, 1985). According to Baysinger and Butler (1985), a board of directors may be classified into three broad components; the executive, monitoring and instrumental component. Directors in the executive component are closely aligned with top management. The executive directors provide expertise to the company decision making progress. Examples of executive directors are former employees and CEO’s. The monitoring component consists of directors who are independent, are not employed by the firm and have no relationship with the management. The main function of the monitoring component is to check whether the management acts in the interest of shareholders. They also provide advisory services. Board members, who are part of the instrumental component of the board, are on the board for functional reasons. They often are lawyers, financiers and consultants who have a lot of knowledge in their field of interest. They advise the management with making important decisions.
An important role of the board of directors is to monitor the management. Without the monitoring component, managers can use shareholders’ funds on private benefits. They can buy themselves extravagant perks, for example, expensive cars, boats or a company airplane. Barry Diller of Expedia and IAC Corp spend nearly one million dollar in personal flight time with the company airplane (McIntyre, 2012). Managers can also give themselves excessive compensation. Even when they did not reach their goals, managers often get their bonuses. Managers can also build an empire around them so they are hard to fire. With a large empire managers have much power and this is often not in the interest of shareholders. The monitoring component of the board controls that managers don’t abuse their power in their own interests but acts in the interest of shareholders. This diminishes the agency problem between shareholders and managers.
Another important role of the board of directors is that they appoint the CEO and the top management and that they have the ability to replace them when they are underperforming. The decision whether to retain or replace the CEO is an important decision that can have a great impact on the daily operations and the performance of the firm.
5
2.2 Relationship between CEO turnover and firm performance
This section continues with several studies about the relationship between CEO turnover and firm performance.
Warner et al. (1988) study the relationship between firm’s stock returns and subsequent top management changes. They use a random sample consisting of 269 firms listed on the New York and American Stock Exchanges on July 2, 1962. For these firms they examine top management changes for the time period from 1962 to 1980. They find a total of 567 management changes. For these changes Warner et al. (1988) try to find the reason why the CEO is replaced. The most commonly reported reason is retirement. However, they are aware that this term could be a euphemism for a forced departure. Firms prefer to avoid negative media attention and therefore a CEO is more likely to get an exit package when he quietly resigns or retires. This could lead to a situation where the reason of the CEO replacement on paper is different from the real reason of the CEO replacement. After controlling for death, health issues and retirements, Warner et al. (1988) come to 64 top management changes as forced departures. They use logit models to test the relationship between stock returns and top management changes. They use different samples consisting of all management changes, CEO changes and forced departures. They find evidence at a significance level of 5% that there is a negative relationship between management changes and stock returns one year prior to the management change. The relationship is the strongest for forced departures. They also use lagged values of performance for three years. Only the performance one year prior to the management change is significant and this indicates that the board of directors acts relatively quick with its decision whether to retain or to replace its CEO. When Warner et al. (1988) divide the sample in ten groups based on its performance. They find that the models only have predictable abilities when the performance is extremely good or extremely bad. The results are only significant when the worst 10% and the best 10% of the firms based on their performance are included. Summarized, Warner et al. (1988) find evidence that there is a negative relationship between management changes and firm performance one year prior to the management change when the performance of the firm is extremely good or extremely bad.
Denis and Denis (1995) examine whether management turnover is preceded by declines in operating performance and whether this leads to improved firm performance. The sample consists of 908 top management changes announced in the Wall Street Journal for the time period from 1985 to 1988. They classify 107 departures as forced. They start their research with examining the effect of announcements of management changes to abnormal return. Denis and Denis (1995) find
6
that the effect of announcements of management changes is associated with abnormal returns that are significantly positive for the forced departures. The results are not significant for the departures that they classify as normal retirements. Then they continue with an event study that examines firm performance prior to and after the management change. They find that the ratio of operating income to total assets decreases in the three years prior to a management change and increases following the change. These results differ between forced departures and normal retirements. Forced departures exhibit large and significant decreases in operating performance prior to
management changes and this significantly improves following these changes. Normal retirements don’t exhibit performance declines prior to the management change, but they do exhibit small performance improvements following the change. Summarized, Denis and Denis (1995) conclude that there is a negative relationship between a forced management change and performance of three years prior to a forced management change and that forced management changes are relatively rare. This is not completely in line with Warner et al. (1988) because they only find a negative relationship between a management change and performance one year prior to the management change.
Weisbach (1988) examines the relationship between CEO turnover and firm performance with different compositions for the board of directors. The focus in this part of the paper will be on the relationship between CEO turnover and firm performance. The effect of board composition will be examined later this literature review. Weisbach (1988) uses data for 495 publicly held firms listed on the New York Stock Exchange for the time period from 1977 to 1980. This data has 286 CEO departures in total. Only nine of these departures have disappointing performance as reason of the CEO change. However, he finds that firm performance prior to a CEO replacement is often disappointing. This suggests that companies don’t announce the true reason behind its CEO resignations. Therefore, Weisbach (1988) ignores the stated reasons for resignation in his sample. He also excludes reasons of death and preceding a takeover from the sample because those CEO changes are certainly no forced CEO changes. Weisbach (1988) also excludes CEO replacements if the CEO has an age of 64, 65 or 66, because these replacements are likely to be retirements. Weisbach (1988) uses logit regressions as models to estimate the relationship between CEO turnover and firm performance. The dependent variable is CEO turnover and equals 1 when the CEO is replaced. He uses stock returns and earnings as performance measures. Weisbach (1988) finds that there is a negative relationship between CEO turnover and firm performance one year prior to the CEO replacement. He finds no evidence for a relationship between CEO turnover and
7
firm performance of more than one year prior to the CEO replacement. These results are in line with Warner et al. (1988).
Summarized, most past researches find a negative relationship between CEO turnover and firm performance one year prior to the CEO turnover. The results differ about a possible
relationship between CEO turnover and firm performance of more than one year prior to the CEO turnover. Past research also finds that it’s hard to find the real reason of a CEO change. Therefore, past research uses all management changes in its sample or makes several assumptions to control for the CEO changes where the reason of the CEO change is different than “fired” as a result of disappointing firm performance.
2.3 Relationship between CEO turnover and relative firm performance
This section discusses several studies about the relationship between CEO turnover and non-firm specific factors.
An important role of the board of directors is the decision whether to retain or fire the CEO of the firm. However, it’s not easy to fully distinguish CEO performance from firm performance. Jenter and Kanaan (2006) investigate whether CEO’s are fired after disappointing firm performance caused by factors that are out of the control of the CEO. For example, a war in the Middle East could lead to high oil prices. This may lead to high costs for some industries, for example the industrials industry. As a result, the performance for firms in the industrials industry may be disappointing. Jenter and Kanaan (2006) investigate whether factors like these affects CEO
turnover. For their research they collect a sample of 1206 voluntary and 384 forced CEO turnovers for the time period from 1993 to 2001. They classify a CEO turnover as forced if the press reports state that the CEO is fired, forced out, or retires or resigns due to policy differences. They use a two-stage least squares logit regression. The first stage of the model predicts company stock
returns using industry and market stock returns. The second stage of the model predicts forced CEO turnover using the company stock returns predicted in stage one. They find that low industry stock returns and low market returns significantly increase the likelihood of forced CEO turnovers. A decline of the industry component of firm performance from its 75th to its 25th percentile increases the probability of a forced CEO turnover by 50 percent. The board of directors tries to distinguish performance that’s in the control of the CEO and performance that’s not in the control of the CEO. They want to make the decision whether to retain or replace the CEO based on performance that’s in the control of the CEO. However, it’s not always possible to fully distinguish performance that’s
8
in the control of the CEO and performance that isn’t in the control of the CEO. Jenter and Kanaan (2006) find that this leads to situations where the CEO is fired by factors that are not in their
control. This leads to a higher CEO turnover when industry performance is low. Summarized, Jenter and Kanaan (2006) find that CEO’s are significantly more likely to be dismissed from their jobs after bad industry performance. This leads to a higher CEO turnover when industry performance is low and this may lead to a higher CEO turnover in a financial crisis.
Morck et al. (1988) investigate the relationship between firm performance and
management characteristics of firms experiencing internally whole management turnover, hostile takeover and friendly takeover. They use a multinomial logit model for Fortune 500 firms for the time period 1980 to 1985. The dependent variable is the performance measure. The performance measures used are Tobin’s Q, abnormal return and employment growth. They find that firms that replaced their whole top management perform poorly relative to other firms in their industry. They also find that a replacement of the whole top management doesn’t occur more often in a bad performing industry compared to a healthy industry. They also find that firms that are run by a member of the founding family are less likely to experience a top management replacement. Summarized, Morck et al. (1988) find that CEO turnover is higher when a firm is performing badly relative to their industry but not higher when the whole industry is performing relatively bad. This result is in contrast to Jenter and Kanaan (2006). They find that CEO turnover is higher when industry performance is low and Morck et al. (1988) find that CEO turnover is not higher when industry performance is low.
Cornelli et al. (2013) investigate whether there is a relationship between CEO turnover and firm performance and under what circumstances the board of directors fire its CEO. They use a sample of 473 firms in central and Eastern Europa and central Asia for a period from 1992 to 2005. They use an instrumental variable regression to investigate the relationship between CEO turnover and firm performance. Cornelli et al. (2013) find a negative relationship between CEO turnover and firm performance. However, they find that firm performance is not the most influencing factor on CEO turnover. The most important factor that influences CEO turnover is information about the firm’s operations and the CEO competence (daily decisions, communication, etc.). The EBRD monitoring reports contain information about the competence of the CEO. Cornelli et al. (2013) include a dummy variable in their model that equals 1 when the reports state that the board views the CEO as incompetent and 0 otherwise. They find that this variable has a larger influence on CEO turnover than firm performance. With this result they show that the board can distinguish bad luck
9
from mistakes of the CEO. They show that the boards do not fire their CEO by mistake and that relative performance is more important. For example, when industry performance is low CEO’s will not be replaced when the firm is performing reasonably compared to the other firms within its industry. This result is in line with Morck et al. (1988), but in contrast to Jenter and Kanaan (2006).
Summarized, the results of past researches differ when emphasising on the relative performance. Jenter and Kanaan (2006) find that CEO turnover is higher when industry
performance is low. On the other hand, Morck et al. (1988) and Cornelli et al. (2013) find that CEO turnover is not higher when industry performance is low. In a financial crisis, the performance of firms is lower compared to a non-crisis period and this may lead to different CEO turnover.
2.4 Other factors influencing CEO turnover
This section shows empirical findings about factors that may affect CEO turnover and factors that may influence the relationship between CEO turnover, the crisis and firm performance. This information will be used in the methodology to develop empirical models. This paper classifies the variables in board, CEO and firm characteristics.
2.4.1 Board characteristics
This section discusses the factors that may influence CEO turnover or firm performance and are classified in board characteristics.
2.4.1.1 Inside and outside directors on the board
Borokhovich et al. (1996) investigate whether there is a relationship between the percentage of outside directors and the decision whether to appoint an inside or an outside CEO. They also investigate what this effect is on stock returns. They use a sample that consists of 969 CEO
turnovers at 588 US firms between 1970 and 1988 from the COMPUSTAT database. They use probit models with a dependent variable that equals 1 when an outsider is appointed CEO and 0
otherwise. They control for firm performance measured by the return on assets. They find strong evidence that the frequency that an outside director is appointed CEO increases with the
percentage of outside directors. Moreover, they find that appointments of outside directors as CEO increases stock returns and that stock returns decreases when and insider replaces a fired CEO.
Weisbach (1988) investigates whether there is a relationship between the
10
firms between 1977 and 1980. He uses logit equations with CEO turnover as dependent variable and stock returns as performance variable. He also creates a dummy variable that equals 1 when the percentage of outsiders on the board is between 40% and 60% and a dummy variable that equals 1 when the percentage of outsiders on the board is at least 60%. He finds that the probability of a CEO replacement after disappointing firm performance is higher for firms with outsider dominated boards compared to firms with insider dominated boards.
There are several studies about the different decisions that the managers of board members take in a crisis compared to a period of economic growth. Schaor and Zuo (2011) investigate the characteristics of CEOs who started in a recession compared to those who didn’t. They find that a CEO who started in a recession is about one year younger and is more likely to be an inside CEO. A board of directors could attract more inside directors because they act more risk averse during a financial crisis. The abilities of inside CEO’s is known for the board of directors and therefore it’s less risky to nominate an inside manager as CEO compared to an outside CEO.
Hermalin and Weisbach (1988) find that when a CEO is about to retire, more inside directors are attracted by the firm. After a CEO change, inside directors are more likely to leave the board. They also find that after disappointing firm performance, inside directors are more likely to leave the board and outside directors are more likely to join the board. Firms do this to improve the monitoring of managers. On the other hand, inside directors have on average more expertise and experience and they can provide valuable information to outside directors and are therefore important for a firm (Baysinger and Butler, 1985). However, Rosenstein and Wyatt (1990) find no clear evidence that outside directors are more or less valuable than inside directors.
Overall, several researches find a relationship between CEO turnover and the percentage of outside directors. There is also evidence that the percentage of outside directors affects firm performance.
2.4.1.2 Board size
Coles et al. (2008) investigate the relationship between firm value and board size. They use a sample of US firms for a time period from 1998 to 2001 and a OLS regression model with the logarithm of the number of board members as dependent variable. They find that either small or large boards are optimal. They find that large firms have larger boards because they have greater advising requirements than small firms. They also find that this relationship is driven by the number of outside directors. Eisenberg et al. (1998) study the relationship between board size and financial
11
performance. With return on assets as performance variable, they find a significant negative relationship between board size and ROA. They find that small boards are more effective. Board size also seems to have a relationship with firm performance and inside or outside dominated boards. Guest (2009) examines the impact of board size on firm performance for a sample of 2746 UK listed firms over 1981-2002. He uses an ordinary least-squares (OLS) regression model with ROA as dependent variable and the logarithm of board size as most important independent variable. Moreover he controls for industry and firm size. He finds that board size has a strong negative relationship with ROA and share returns. He also finds that the relationship is the strongest for large firms. His evidence supports the argument that problems of poor communication and decision-making undermine the effectiveness of large boards. Overall, several researches find a relationship between board size and firm performance. This may also lead to a different CEO turnover.
2.4.1.3 CEO duality
CEO duality is a factor that may influence CEO turnover and firm performance. A firm has CEO duality if the CEO of the firm is also the chairman of the board. Opponents of CEO duality state that the CEO gets too much power. Moreover, the main role of the board is to monitor the management and mitigate the agency conflict. When the CEO has to monitor himself and his own management, than this could lead to a bigger agency conflict. This may lead to a lower CEO turnover and worse firm performance. On the other hand, firms with CEO duality may be more efficient because the communication between the management and the board might be easier.
Rechner and Dalton (1991) investigate whether CEO duality affects firm performance. They use a sample of 250 Fortune 500 firms for the period from 1978 to 1983. They use a multivariate covariance analysis, a regression model with multiple dependent variables. They use return on equity, return on investment and profit margin as dependent variables. CEO duality is the most important independent variable and equals 1 when the CEO is also chairman of the board. They find that firms with a different CEO and Chairman outperform firms with CEO duality. Baliga et al. (1996) also investigate whether there is a relationship between CEO duality and firm performance. They uses a sample of Fortune 500 firms for a time period from 1980 to 1991. They measures firm performance by the return on assets and the return on equity. They find only weak evidence that CEO duality has a negative relationship with firm performance. On the other hand, Brickley et al. (1997) find that the costs of separation of the CEO and Chairman are larger than the benefits for
12
most firms.
2.4.1.4 Gender diversity in the board
The majority of the board members are men. Several researches investigate whether this is optimal. Shrader et al. (1997) investigate the effect of women on top management positions and firm performance. Shrader et al. (1997) hypothesis is that firms with a higher percentage of women, perform better. They base this on the fact that a firm with more women reflects the composition of the existing market better. Nowadays, about 25% of the top managers are woman, but the
composition of the market has about 50% women. They use a sample of 200 US firms for the year 1992 and these observations are obtained from the Wall Street Journal. They use a OLS regression model with return on assets as dependent variable. They measure gender diversity as the
percentage of women managers, the percentage of women top managers and the percentage of women on the board. They didn’t find strong evidence that a higher percentage of female board members increases firm performance. They give as explanation for this that there are only 4.5% female board members in the sample. However, they find an indication for a negative relationship between female board members and firm performance. Nowadays, there are more female board members.
Adams and Ferreira (2009) find indications that firms with relatively many women perform worse on the board of directors. They use a sample of S&P1500 firms for the period from 1996 to 2003. They use a OLS regression model with a dependent variable that equals 1 when a directors attended less than 75% of the meetings. They control for firm performance measured by ROA, board size, compensation, and the percentage of independent directors. They find that women monitor more. This may lead to different firm performance because monitoring may align the interest of managers with the interest of shareholders. On the other hand, monitoring is costly and this may decrease the performance. Adams and Ferreira (2009) also use a OLS regression model with CEO turnover as dependent variable and the percentage of female directors as most important variable. They control for board size, percentage of independent directors, CEO age and CEO
gender. They find an indication that there is a negative relationship between CEO turnover and the percentage of female directors, but the results are insignificant.
Summarized, there is some evidence that women on the board of directors influence CEO turnover and firm performance. Therefore, this factor should be taken into account.
13
2.4.1.5 Age of the board members
Another factor that may influence CEO turnover is the age of the board members. Core et al. (1999) find that CEO compensation rises with the number of directors aged above 69. They use a sample of 205 publicly trades US firms for a time period from 1982 to 1984. They use a OLS regression model with CEO compensation as dependent variable. They use the percentage of outside directors over age 69 as independent variable. Moreover, they control for firm performance, measured by ROA, percentage of inside directors, board size and CEO duality. They find that CEO compensation rises with the percentage of directors aged above 69. Higher compensation for the CEO may lead to higher incentives and this may result to better performance and to a lower CEO turnover. Moreover, old directors are on average more experienced compared to young directors and therefore they’re probably better in their advisory role and this may also lead to higher performance.
Ferris et al. (2003) find that the presence of multiple directorships might generate higher agency costs. They use a OLS regression model with a sample of Forbes 500 firms in 1995.
Moreover, they find that the average age of board members in boards with multiple directorships is higher. The higher agency costs probably decreases firm performance and may affect CEO turnover. Therefore, old directors may have a relationship with firm performance or CEO turnover.
2.4.2 CEO characteristics
This section discusses the factors that may influence CEO turnover or firm performance and are classified in CEO characteristics.
2.4.2.1 Age of the CEO
One CEO characteristic that will be included in the model is the age of the CEO. The age of the CEO when they leave the firm plays an important role. For CEO’s older than 64, retirement plays an important role (Lausten, 2002). They did this by splitting the sample into a group with CEO’s aged less than 64 years and a group with CEO’s of at least 64 years of age. The CEO turnover was higher for the group with CEO’s of at least 64 years of age.
Jensen and Murphy (1990) find a positive relationship between CEO turnover and CEO age. They use logit regressions with CEO turnover as dependent variable and a sample of over 1000 CEO departures for US firms for a time period from 1965 to 1989. Chevalier and Ellison (1999) find that the relationship between CEO turnover and firm performance is stronger for younger CEO’s. Morck
14
et al (1988) find that firms that are run by a young top executive are more likely to experience a hostile takeover. According to Morck et al. (1988), that leads to a higher CEO turnover. Young CEO’s have less experience and this may lead to worse performance. Overall, the age of the CEO could have a relationship with CEO turnover or firm performance.
2.4.2.2 Gender of the CEO
The gender of a CEO is also a factor that may affect CEO turnover and firm performance. Davis et al. (2010) find indications that firms with a female CEO perform better due to stronger market
orientation relative to a male CEO. They use a sample of 1083 US firms. The methodology they use, is a multivariate covariance analysis. They use firm growth and firm performance as dependent variables. The most important independent variable is CEO gender and that equals 1 when the CEO is male and 0 when the CEO is female. The results are insignificant, but they indicate that female CEO’s perform on average slightly better than male CEO’s.
Shrader et al. (1997) finds indications for a positive relationship between women in the management and firm performance. They use the same methodology as they use for the relationship between firm performance and the percentage of female directors. There is also evidence that female CEO’s have lower wages (Skalpe, 2007). This may lead to lower incentives for females and this may lead to lower performance of the firm. This may also affect CEO turnover.
Summarized, past researches find different results for a relationship between CEO gender and firm performance. However, some researches find a relationship between CEO gender and firm performance and therefore this should be taken into account.
2.4.2.3 CEO compensation
CEO compensation could have a relationship with CEO turnover and firm performance. When the compensation of a CEO is aligned with the interest of shareholders, than this may lead to better performance. The CEO has more incentives to perform well. Jensen and Murphy (1990) find a positive relationship between performance and CEO compensation. They use a sample of 1295 firms published in Forbes from 1974 to 1986. They use a OLS regression model with CEO
compensation as dependent variable. This compensation includes base salary and bonus. They use the percentage change of market value as performance measure. They find that the compensation of the CEO increases by $3.25 for every $1000 change in shareholders wealth. The compensation of the CEO may also affect CEO turnover.
15
Mehran (1995) finds that there is a positive relationship between firm performance and CEO compensation. He uses a sample of 153 US manufacturing firms for the period from 1973 to 1983. He uses a OLS regression model with CEO compensation as dependent variable and return on assets as performance measure.
Overall, most past researches find a positive relationship between CEO compensation and firm performance. CEO compensation may also affect CEO turnover. When the salary of a CEO is high, than shareholders probably expects more from the CEO. This could lead to more
replacements of the CEO. On the other hand, when the compensation of a CEO is relatively high, than the CEO is probably good. Good CEO’s are replaced less often and this could lead to a negative relationship between CEO compensation and CEO turnover. High compensation may also lead to higher incentives for the CEO and this may lead to better performance.
2.4.3 Firm characteristics
This section discusses the factors that may influence CEO turnover or firm performance and are classified in firm characteristics.
2.4.3.1 Firm size
The first firm characteristic that will be discussed is firm size. The size of a firm may have a
relationship with firm performance. On the one hand, large firms experience economies of scale. As a result, large firms can produce cheaper per product than small firms. Large firms also have more market power and a competitive advantage against small firms. On the other hand, small firms experience higher growth on average (Evans, 1987). He uses in his research a sample of 42,339 US firms between 1976 and 1980. He uses an OLS regression analysis with the growth rate, measured by the percentage change in employment, as dependent variable. Firm size is measured by the logarithm of the number of employees. With this regression he finds a negative relationship
between the growth rate and firm size and this is significant at a 1% level. A higher growth rate may lead to higher performance and to a lower CEO turnover.
Philippon (2006) finds that managers sometimes expand their firms beyond the profit-maximizing size and that shareholders are more likely to tolerate this in good times. In a crisis period, the performance of most firms is bad and the shareholders may not tolerate this. Therefore, this may lead to firing the CEO in bad times because the board is less likely to fire the CEO in good times. Therefore, there may be a relationship between firm size and CEO turnover, the crisis
16
dummy or firm performance.
2.4.3.2 Firm industry
The variable industry may also affect firm performance or CEO turnover. Market-wide effects may affect some industries more than other. For example, when the oil price increases, the industrials industry is affected more than the telecommunication industry. Therefore, the industry of a firm affects the performance. Industries may also have different CEO policies. Firms in a growing industry perform on average well. This could lead to a lower CEO turnover. The performance of industries also differs when the economy is in a crisis. Most industries perform badly in a crisis but there are also industries that performs well. Summarized, the industry of a firm seems to have a role with firm performance and a financial crisis and may even affect CEO turnover.
2.5 Hypotheses
This section will build hypotheses based on the empirical literature mentioned above. First, the paper will investigate whether there is a relationship between CEO turnover and firm performance. Most researches find a negative relationship between CEO turnover and firm performance (Warner et al., 1988; Weisbach, 1988 and Denis and Denis, 1995). There is also research done about firm performance of several periods prior to a CEO replacement. Warner et al. (1988) and Weisbach (1988) find that CEO turnover is only sensitive to firm performance in the year prior to the CEO replacement and not to periods before that. Denis and Denis (1995) find that there is a negative relationship between CEO turnover and firm performance of three year prior to the CEO turnover. Based on these researches, this paper comes to the following hypothesis:
Hypothesis 1: There is a negative relationship between CEO turnover and firm performance one year prior to the CEO replacement. There is no relationship between CEO turnover and firm performance more than one year before the CEO
replacement.
After testing hypothesis 1, this paper investigates whether CEO turnover is different in a financial crisis while there will be controlled for firm performance. According to Jenter and Kanaan (2006), CEO turnover is higher when industry performance is low. Industry performance in a financial crisis is on average lower than in a non-crisis period. On the other hand, Morck et al. (1988) find that CEO’s will be replaced only when they perform relatively bad. Cornelli et al. (2013)
17
find that they incompetency of the CEO is more important than firm performance. They show that the board can distinguish bad luck from mistakes of the CEO. Boards do not fire their CEO by mistake, for example when industry performance is low. It’s also possible that the board of directors is more reluctant to replace its CEO in a financial crisis. Hiring a new CEO is risky and the board of directors may act more risk averse during a financial crisis. Based on these researches, this paper comes to the following hypothesis:
Hypothesis 2: The CEO turnover is lower in a financial crisis compared to a non-crisis period independent of firm performance.
3. Data, Methodology and Descriptive statistics
This section starts with the data that will be used to answer the hypotheses. Then it continues with describing the methodology. Finally, it presents the descriptive statistics.
3.1 Data
This paper will use data from the Riskmetrics, COMPUSTAT and EXECUCOMP databases in WRDS. These databases contain information about CEO turnover, firm performance and about the control variables of this research. The Riskmetrics database contains information about the composition of the board of directors for firms. The database will be used to measure the number of board
members, average age of directors, percentage of women on the board, percentage of
independent directors and CEO duality. The COMPUSTAT database will be used to find information about accounting data. With this data this the performance measures return on assets, return on equity and the share return are calculated. The database will also be used to find data about the size of a firm and the industry. The EXECUCOMP database will be used to find data about the CEO. With this information, the paper makes the CEO turnover variable. The database also contains information about the age, gender and compensation of the CEO. The sample of firms are firms listed on the S&P 500, S&P500 Midcap or S&P500 small cap firms for the time period from 1996 to 2012. The sample contains a non-crisis period, namely from 1996 to 2007 and 2012 and a financial crisis, namely from 2008 to 2011.
The time period from 1996 to 2012 of the sample includes a financial crisis. This financial crisis is known as the Global Financial Crisis and started at the end of 2007. It’s the biggest financial
18
crisis since The Great Depression in 1930. According to several experts and newspapers, the Global Financial Crisis covers the time period from 2007 to 2011 (Elliot, 2011). In this period the growth GDP rate was low or negative. The GDP growth rates are -0.3% and -2.8% in 2008 and 2009
respectively (Trading Economics, 2014). Therefore, this paper will assume that in the period of 2008 to 2011, there was a financial crisis.
3.2 Methodology
This section starts with a theoretical explanation of logit models. Then it continues with the methodology of this paper.
3.2.1 Logit regressions
Logit regressions are nonlinear regression models specifically designed for binary dependent variables (Stock and Watson, 2012). The dependent variable of a logit regression can only take the value 0 or 1. Therefore, a regression with a binary dependent variable Y models the probability that Y=1. The population logit model of the binary dependent variable Y with multiple regressions is shown in equation 1.
Equation 1: Pr(Y=1|X1, X2,…,Xk)=F(β0+ β1X1+ β2X2+…+ βkXk)
F is the cumulative standard logistic distribution function. The outcome of the coefficients can only be interpreted with a relationship. To calculate the probability that Y=1, equation 2 can be used. Equation 2: Pr(Y=1|X1, X2,…,Xk)=
This paper will use equation 2 to calculate the probability that a CEO will be replaced for a given observation.
3.2.2 Methodology
This section describes the models that will be used to answer the hypotheses. Model 1 shows the model that will be used to answer hypothesis 1 about the relationship between CEO turnover and firm performance.
19
Model 1:
CEO turnover = β0*β1 *ROAt-1 + β2 *ROAt-2 +β3 *ROAt-3 +β4 *Log(revenues)+ β5 *Average age of
directors+ β6 *Number of board members+ β7 *percentage of women on board +
β8 *Percentage of independent directors+ β9 *CEO duality+ β10 *CEO gender + β11 *CEO Age+
β12 *Log(CEO compensation)+ β13 *Industry+ uit
This model is a logit model. CEO turnover is the dependent variable that equals 1 when the CEO is replaced and 0 otherwise. All CEO changes are included in the dataset. Weisbach (1988) and Warner et al. (1988) also use logit models in their research. The independent performance variable is measured by the return on assets (Shrader, 1997 and Borokhovich et al., 1996). It’s defined as net income divided by total assets. Lags for three years will be used. Past literature also uses
performance measures of three years (Warner et al, 1988 and Weisbach, 1988). As robustness check ROA will be replaced by return on equity and stock return (Warner et al., 1988 and Weisbach, 1988). ROE is defined as net income divided by market value of equity. Stock return is measured as the return of the average share price for a given year. The paper uses returns to prevent
multicollinearity between the independent variables. The paper will check whether the problem of multicollinearity is solved.
Performance is not the only variable that affects CEO turnover. Therefore, this paper also includes several control variables. Section 2.4 discusses factors that may influence CEO turnover. Those factors will be included in the model. The paper categorizes the control variables in board, CEO and firm characteristics.
Board characteristics that are included in the model are the size of the board, average age of board members, percentage of independent directors, percentage of women on the board and CEO duality. Coles et al. (2008), Eisenberg et al. (1998) and Guest (2009) find that small boards are the most effective boards. Therefore, board size probably affects performance and this may lead to a different CEO turnover. On the other hand, large boards may monitor better because they have more members to monitor. The board member variable is measured by the number of members on a board. Weisbach (1988) finds that independent directors monitor better and this may lead to better performance. Hermalin and Weisbach (1988) find that when a CEO is about to retire, more inside directors are attracted by the firm. The percentage of independent directors is measured by the number of independent directors divided by the total number of directors on a board. Shrader (1997) finds that women on the board lead to better performance because it reflects the existing
20
market better. Adams and Ferreira (2009) find that women monitor better and this may lead to better performance. The percentage of women is measured by the number of women on the board divided by the total number of directors on the board. CEO duality gives a CEO more power. It makes them harder to fire and therefore CEO turnover may be lower. On the other hand, communication between the board and management is easier and this may lead to better performance. Rechner and Daltion (1991) and Baliga et al. (1996) find evidence that firms with a different CEO and Chairman outperform firms with CEO duality. The variable CEO duality equals 1 when the CEO is also chairman of the board and 0 otherwise.
CEO characteristics that are included in the model are age, gender and compensation of the CEO. Jensen and Murphy (1990) find that CEO’s are more likely to be fired when they are young. Retirement plays an important role for old CEO’s (Lausten, 2002). Therefore, the age of a CEO may affect CEO turnover. Davis et al. (2010) find that female CEO’s perform better due to stronger market orientation relative to men. The variable CEO gender equals 1 when the CEO is female and 0 otherwise. The compensation of a CEO may also affect CEO turnover because a high compensation may lead to higher expectations from shareholders. This may lead to a higher CEO turnover. Moreover, Jensen and Murphy (1990) find a positive relationship between performance and CEO compensation. The variable compensation is the dollar amount of the base salary and the bonus of a CEO in a given year.
Firm characteristics that are included in the model are firm size and firm industry. Firm size may affect CEO turnover because small firms experience higher growth on average (Evans, 1987). This may lead to a different CEO turnover. Philippon (2006) finds that managers sometimes expand their firm beyond the profit-maximizing size and this may also affect CEO turnover or firm
performance. Firm size is measured by the logarithm of revenues of a firm. Firm industry may affect CEO turnover because there may be different CEO policies across industries. Moreover, market performance may affect the performance of the firm. Then this performance may lead to a different CEO turnover. Firm industry is classified according to the Global Industry Classification Standard sectors.
This paper continues to answer hypothesis 2 about the relationship between CEO turnover and a financial crisis. Model 2 shows the model that will be used to answer hypothesis 2.
21
Model 2:
CEO turnover = β0+ β1 *Crisis dummy +β2 *ROAt-1 + β3 *ROAt-2 +β4 *ROAt-3 +β5 *Log(revenues)+ β6
*Average age of directors+ β7 *Number of board members+ β8 *percentage of women on board +
β9 *Percentage of independent directors+ β10 *CEO duality+ β11 *CEO gender + β12 *CEO Age+ β13
*Log(CEO compensation)+ β14 *Industry+ uit
This model is a logit model. The dependent variable is the probability that a CEO leaves the job (Weisbach, 1988 and Warner et al., 1988). The crisis dummy variable equals 1 when the
observation is in the period of 2008-2011 and 0 otherwise. In this period the economy is in a crisis or recovering from it. This crisis is known as the Global Financial Crisis. The idea behind this variable is that the CEO turnover may be different in a financial crisis compared to a non-crisis period. In a financial crisis, firm performance is lower on average. According to Jenter and Kanaan (2006), this leads to a higher CEO turnover. On the other hand, Morck et al. (1988) and Cornelli et al. (2013) find that CEO turnover is not higher when industry performance is low. It’s also possible that the board of directors acts more risk averse and are more reluctant to fire its CEO. If the coefficient of the crisis dummy is positive, than the probability that a CEO is replaced in a crisis, given all other variables, is higher. The variables of firm performance and board, CEO and firm characteristics are the same as described in model 1.
Then the paper continues with robustness checks. For a robustness check, the performance measure ROA will be replaced by the return on equity and the share price. Then the paper
examines whether multicollinearity is a problem. It starts with examining the correlation to check whether there are indications for multicollinearity. Then the paper checks whether multicollinearity leads to insignificant coefficients. Finally, the paper calculates the variance inflation factor what is a common measurement of multicollinearity. If the VIF is above 10, then multicollinearity can be considered as a problem.
3.3 Descriptive statistics
This section examines the descriptive statistics of the variables. It starts with describing how this paper adapts the data to prevent that the sample contains incomplete observations or outliers. Then it will examine the mean, standard deviation, minimum and maximum of the variables and it gives an interpretation of them. Finally this paper examines the correlations between the variables. It will focus on the CEO turnover and the crisis dummy.
22
3.3.1 Adjusting the sample and variables
Some observations have very high values for the performance measures. These outliers create misleading coefficients. Therefore, an observation is excluded from the sample when ROA or ROE is above 10 or below -10 and share return is above 5 or below -5.
Variables like revenues and compensation have high values and this lead to a non-normal distribution. To get a normal distribution the paper takes the logarithm of the variable. Figure 1 shows the histogram for the variable revenues and the logarithm of the revenues. The histogram of revenues is positively skewed and the logarithm of revenues is more normally distributed.
Therefore, this paper takes the logarithm of the revenues. CEO compensation is also non-normal distributed and Figure 1 shows that the logarithm of CEO compensation solves this problem. Figure 1. Distribution with and without logarithms
3.3.2 Descriptive statistics
Table 1 shows the descriptive statistics of CEO turnover, firm performance and firm size. CEO turnover has a mean of 0.1130. This means that 11.30% of the observations have a CEO change in that year. This leads to a total of 2,714 CEO changes.
23
Table 1. Descriptive statistics for CEO turnover and firm performance
Variable Observations Mean Std. Dev. Min Max
CEO turnover 24018 0.1130 0.3166 0 1 Crisis Dummy 24018 0.2845 0.4512 0 1 Revenues 24002 5836.56 18119 0 467231 Log(revenues) 24003 3.3285 0.7564 -0.8297 6.3728 Share Price 23472 34.8528 71.2822 0.0173 4111.95 Performance Measures ROAt-1 22966 0.0385 0.1443 -5.7785 1.3278 ROAt-2 21895 0.0385 0.1435 -5.7785 1.3278 ROAt-3 20654 0.0383 0.1555 -5.7785 4.8328 ROEt-1 22713 0.0047 0.3165 -7.8968 2.3863 ROEt-2 21477 0.0055 0.3123 -7.8968 2.3663 ROEt-3 20109 0.0052 0.3280 -9.2174 4.6304 Share Return t-1 19399 0.0539 0.4704 -0.9745 28.8090 Share Return t-2 17424 0.0436 0.4714 -0.9745 28.8090 Share Return t-3 15502 0.0147 0.4601 -0.9411 28.8090
Figure 2 shows the CEO turnover over time. Figure 2 shows a growth of CEO replacements around the year 2000 and a strong decline thereafter. There is also a decline in CEO turnover after 2008. In the periods of decline in CEO replacements the economy was often in a recession or in a crisis.
Figure 2. CEO turnover (1996-2012)
Figure 3 shows the relationship between CEO turnover and the S&P500 index. This indicates a positive correlation between CEO turnover and S&P 500. The correlation between the S&P500 and CEO turnover is nearly 35%. Figure 3 also indicates that CEO turnover grows when the S&P500
0 0,02 0,04 0,06 0,08 0,1 0,12 0,14 0,16 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Fr act io n C EO t u rn o ve r Year CEO turnover
24
index increases and declines when the S&P500 index decreases. The correlation between the S&P500t-1 and CEO turnover is even 37.7%. These results indicate that there is a negative
relationship between CEO turnover and a financial crisis. CEO turnover is low in periods when the economy is in a crisis and CEO turnover is high when there is economic growth.
Figure 3. CEO turnover and S&P 500 (1996-2012)
Table 1 also shows that the crisis dummy has a mean of 0.2845. This means that 28.45% of the observations are from the crisis period and 71.55% of the observations are from the non-crisis period. The revenues of the firms range from 0 to 467 billion a year with an average of 5.8 billion. The return on assets performance measure has an average of 3.85%. This means that the net income is on average 3.85% of the total assets. The return on equity has an average of 0.5%. The share price ranges from $0.01 to $4,112 and has an average share price of $34.85.
Table 2 shows the descriptive statistics of board and CEO characteristics. On the board of directors are on average 9.43 members ranging from 1 to 34 members per board. The average age of the directors is 60.55 years. The youngest member is 34 years old and the oldest member nearly 79 years old. In the sample, 10.32% of the board members are women. There are boards with 0% women on its board and the firm with the largest part of women on its board has 62.5% women. Of the directors 71.56% is independent and they didn’t have any relationship with the company or management before they were hired. Of the board members, 28.44% is an insider and is employed or has another relationship with the firm before they entered the board. Of the CEO’s, 11.38% is
0 200 400 600 800 1000 1200 1400 1600 0 0,02 0,04 0,06 0,08 0,1 0,12 0,14 0,16 0,18 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 S&P 5 0 0 Ind ex Fr act io n o f C EO t u rn o ve r Year
25
also chairman of the board of directors. The CEO is in 97.7% of the observations a man and CEO’s earn on average 1.24 million a year ranging from $0 to $ 77.9 million.
Table 2. Descriptive statistics for board and CEO characteristics
Variable Observations Mean Std. Dev. Min Max
Board Characteristics
Board size 18614 9.4343 2.6804 1 34
Average age of directors 18614 60.5505 4.1385 34 78.875
Percentage of female directors 18614 0.1032 0.0954 0 0.625
Percentage independent directors 18614 0.7156 0.1632 0 1
CEO duality 18614 0.1138 0.3176 0 1
CEO Characteristics
CEO gender 24018 0.0225 0.1484 0 1
CEO age 24018 53.9598 11.6034 0 96
CEO compensation 24018 1243.03 1684.25 0 77926
CEO compensation (log) 23908 2.9468 0.4556 -3 4.8917
Table 3 shows the descriptive statistics of the industry variable. The industry variable divides the sample in ten different industries. The largest industry from the sample is the technology
industry and this contains 4661 observations. The smallest industry in the sample is the
Telecommunication Services industry and counts 258 observations for the time period from 1996 to 2012. This is an industry with fewer firms because it’s expensive to create and support a
telecommunication network. Therefore, those firms are also relatively large. Table 3. Descriptive statistics for industry variable
Variable Observations GIC code Mean Std. Dev. Min Max
Industry 24018 0 1 Energy 1367 10 0.0569 0.2317 0 1 Materials 1646 15 0.0685 0.2527 0 1 Industrials 3480 20 0.1449 0.3520 0 1 Consumer Discretionary 4196 25 0.1747 0.3797 0 1 Consumer Staples 1260 30 0.0525 0.2230 0 1 Health Care 2642 35 0.1100 0.3129 0 1 Financials 3339 40 0.1390 0.3460 0 1 Information Technology 4661 45 0.1941 0.3955 0 1 Telecommunication Services 258 50 0.0107 0.1031 0 1 Utilities 1169 55 0.0487 0.2152 0 1 3.3.3 Correlation statistics
This section starts with evaluating the correlations between the performance measures, CEO turnover and the crisis dummy. Then it continues with examining the correlations between the
26
other variables.
Table 4 shows the correlations between the performance measures, CEO turnover and the crisis dummy. The performance measure ROAt-1 has a negative correlation with CEO turnover of
6.33%. This result indicates that there is a negative relationship between CEO turnover and performance. The correlation between ROAt-2 and CEO turnover is smaller and therefore is the
relationship between CEO turnover and ROAt-2 probably weaker. This is also the case for ROAt-3.
These results indicate that there is a negative relationship between CEO turnover and firm performance one year prior to the CEO turnover and that there is no or a weak relationship
between CEO turnover and firm performance of more than one year prior to the CEO turnover. This result is in line with Warner et al. (1988) and Weisbach (1988). This result is partly in contrast to Denis and Denis (1995) because they find a relationship between CEO turnover and firm
performance three years prior to the CEO turnover. The relationship between CEO turnover and ROE is similar to relationship between CEO turnover and ROA. There is a correlation of -5.14% between ROEt-1 and CEO turnover. The relationship between ROEt-2 and CEO turnover and between
ROEt-3 and CEO turnover is weaker. The correlation between CEO turnover and Share returnt-1 is
-5.19%. This also indicates a negative relationship between CEO turnover and firm performance. Overall, the correlations indicate that there is a negative relationship between CEO turnover and firm performance. However, there is still no significant evidence for this and therefore this will be investigated further at the results.
Table 4. Correlations between CEO turnover, firm performance and the crisis dummy
Performance Measure CEO turnover Crisis Dummy
ROAt-1 -0.0633 -0.0215 ROAt-2 -0.0175 -0.0039 ROAt-3 -0.0076 0.0338 ROEt-1 -0.0561 -0.0620 ROEt-2 -0.0005 -0.0526 ROEt-3 -0.0049 -0.0206 Share Return t-1 -0.0519 -0.0603 Share Return t-2 -0.0206 -0.1482 Share Return t-3 -0.0001 -0.0071 CEO turnover 1.0000 -0.0329
The correlation between the crisis dummy and the performance of last year is in all cases negative. The correlations between the crisis dummy and the performance of two or three years back are less negative and in one case even positive. This result indicates that there is a negative relationship between the crisis period and performance. This result is not surprising because in a
27
crisis most firms perform worse compared to a non-crisis period. The correlation between ROAt-3
and the crisis dummy is positive. However, this result doesn’t say much about this relationship and there is an explanation for this. For example, when the observation is in the year 2008, than the crisis dummy has a value of 1. Than ROAt-3 is the performance of the year 2005. In that year there is
no financial crisis.
The correlation between the CEO turnover and the crisis dummy is -3.3%. This result indicates that there is a negative relationship between CEO turnover and the crisis period. This indicates that a CEO is replaced less often in a financial crisis. This result is in contrast to the research of Jenter and Kanaan (2006). However, there are no significant results yet and therefore this will be discussed further in the results.
Appendix 1 shows the correlations of the control variables. The size of a firm has a small positive correlation with CEO turnover. Board characteristics also have relationships with CEO turnover and the crisis dummy. The average age of the board of directors have a negative correlation of 4.15% with CEO turnover. This indicates that older boards, which are on average more experienced, are less likely to replace their CEO. The average age of the board of directors have a positive correlation with the crisis dummy of 22.41%. This reasonable strong correlation indicates that firms attract older and more experienced directors in a crisis compared to a non-crisis period. The number of board members variable has a small positive correlation of 2.90% with CEO turnover. This result probably comes from the strong correlation of 60.07% between the number of board members and firm size. The number of board members has nearly no correlation with the crisis dummy. CEO duality has a negative correlation with CEO turnover of 4.50%. This means that when the CEO is also chairman of the board, the CEO is less likely to be replaced.
Table 5 shows the correlations between industries, CEO turnover and the crisis dummy. Overall, the results show little or no correlation between industries, CEO turnover and the crisis dummy. The Financials industry has a negative correlation of 2.64% with CEO turnover and a positive correlation of 4.94% with the crisis dummy. This indicates that CEO is replaced less often when the firm operates in the Financials industry.
28
Table 5. Correlations for Industries
Industry CEO turnover Crisis Dummy
Energy -0,0037 0,0100 Materials 0,0020 -0,0089 Industrials -0,0095 -0,0152 Consumer Discretionary 0,0054 -0,0094 Consumer Staples 0,0115 -0,0122 Health Care -0,0133 -0,0025 Financials -0,0264 0,0494 Information Technology 0,0290 -0,0000 Telecommunication Services -0,0040 0,0032 Utilities 0,0060 -0,0234 4. Results
This section starts with the regression results to answer the first hypothesis about the relationship between CEO turnover and firm performance. Then it continues with the second hypothesis about the relationship between CEO turnover and a financial crisis. Finally, it continues with robustness and multicollinearity checks.
4.1 Results for the relationship between CEO turnover and firm performance
Appendix 2 shows the regression results to answer the first hypothesis. The paper expects that CEO turnover and firm performance one year prior to the CEO replacement have a negative relationship and that there is no relationship between CEO turnover and firm performance of more than one year prior to the CEO replacement. The regressions used to answer this hypothesis are logit
regressions with CEO turnover as dependent variable. All CEO turnovers are included in the model. Column 1 shows the regression with the performance variable ROA t-1 regressed to CEO turnover.
The ROAt-1 has a negative relationship with CEO turnover and is significant at the 1% level. Further,
the coefficient is hard to interpret. To estimate a change in probability, the model from equation 3 will be used. This equation is derived from the standard logit models discussed in section 3.2.1.
Equation 3: ( )
(
)
An increase in performance from the 25th percentile to the 75th percentile decreases the change that the CEO will be replaced by 5%. An increase in performance from the 5th percentile to the 95th percentile decreases the change that the CEO will be replaced by 22%. Figure 4 shows the
29
value of 1 or 0. The red dots represent the observations. The line represents the estimated probability that a CEO will be replaced for a given performance. Figure 4A shows a negative relationship between CEO turnover and ROAt-1.However, there is only a strong relationship with
extremely good or extremely bad performance. ROAt-1 in figure 4A differs from -600% to 200%.
With these probabilities the change that a CEO will be replaced is 94.5% and 2.5% respectively. However, these values of performance are very rare or even unrealistic. Figure 4B shows more realistic results with ROAt-1 ranging from -50% to 50%. There is still a negative relationship between
CEO turnover and ROAt-1 but the influence of ROAt-1 on CEO turnover is smaller. Therefore, this
paper finds that there is a negative relationship between CEO turnover and firm performance one year prior to the CEO turnover. This paper also finds that the influence of ROAt-1 on CEO turnover is
only large when the performance is either extremely good or bad. A decrease in performance from the 70th to the 30th percentile increases the probability of a CEO change by 4.19%. A decrease in performance from the 99th to the 1st percentile increases the probability of a CEO change by 63.65%.
Figure 4. Relation between CEO turnover and ROAt-1
(A) (B)
The results in column 1 are in line with the results of Warner et al. (1988), Denis and Denis (1995) and Weisbach (1988). This paper finds that firm performance influences CEO turnover partly but not very strong. Cornelli et al. (2013) find that the decision of a CEO replacement is only partly on information about firm performance. This regression also indicates that the CEO replacement is only based partly on past performance.
30
Column 2 of Appendix 2 adds the firm performance variables ROAt-2 and ROAt-3 to the
model. Those variables are statistically insignificant. Therefore, this result indicates a negative relationship between CEO turnover and firm performance one year prior to the CEO turnover but no relationship between CEO turnover and firm performance of more than one year prior to the CEO turnover. This indicates that board members react quickly to disappointing performance by replacing their CEO. This result is in line with Warner et al. (1988) and Weisbach (1988) who also find this. This result is in contrast to Denis and Denis (1995).
Column 3 adds board characteristics to the model. The performance variable ROAt-1 is still
significant at a 1% level and the coefficient is even more negative. Average age of the board members has a negative relationship with CEO turnover and is significant at a 1% level. This result indicates that old board members replace their CEO less often compared to young board members. This result is in line with Core et al. (1999). Moreover, young directors are likely to take more risk. Replacing the CEO is risky because the new CEO may perform even worse. The number of board members is also significant at a 1% level with CEO turnover and there is a positive relationship. Large firms often have larger boards compared to small firms. The correlation confirms this. This result indicates that large firms replace their CEO more often, but the results are not significant. The result is in line with Evans (1987). They find that small firms have on average higher
performance. Large firms have on average more members on the board and this leads on average to lower performance and to a higher CEO turnover. Column 3 also finds a negative relationship between independent directors and CEO turnover at a 10% level. This result indicates that a board with many independent directors replace their CEO less often. This result is surprising because this is in contrast to Weisbach (1988). A possible explanation for this may be that a board with relative many independent directors values a CEO with a lot of expertise and knowledge more and this may lead to less CEO replacements. CEO duality has a negative relationship with CEO turnover and is significant at a 1% level. This result is not surprising and indicates that a CEO who is also chairman at the board is replaced less often. This result is in line with Rechner and Dalton (1991). There is no evidence that relative many women on a board affects CEO turnover.
Column 4 adds CEO characteristics to the model. There is a negative relationship between CEO turnover and CEO age and this is significant at a 1% level. This is a surprising result because the sample also includes CEO turnover where the reason is retired or death. This is more likely to occur when CEO age is high (Lausten, 2002). Jensen and Murphy (1990) find that CEO’s are more likely to be fired when they are young. This negative relationship between CEO turnover and CEO age is
