Firm Performance Induced CEO Turnover An Empirical Study

Firm Performance Induced CEO Turnover

An Empirical Study

S.J. Bijsterbosch

Faculty of Economics and Business, University of Groningen

August 2013

---

ABSTRACT

This thesis studies the change in probabilities of CEO turnover based on firm performance of individual firms during a fifteen-year timeframe. Firm performance is being measured as the change in stock price, and is compared to both the S&P500 index as well as to the industry mean. By doing so, the relationship between performance and CEO turnover is specified into specific components, leading to different outcomes. The main conclusion from this research is that the excess return compared to the overall industry performance does not have the expected large negative influence on CEO turnover, if at all. However, the industry return has a large negative effect on CEO turnover. This is surprising, as one would expect the opposite effect residing from excess return and industry return. This inconsistency with previous literature and up-front expectations might be due to market sentiment.

Keywords: CEO turnover, firm performance, industry-adjusted stock price return JEL Classifications: G3, G34, L2

1. Introduction

According to Bloomberg, CEO turnover has risen tremendously in the past years, hitting almost 14% up from a 15-year low in 2010.12 For the main part, this can be described as being related to the recent economic crises that swept the U.S. and Europe. However, these figures are still noteworthy when comparing them with turnover figures in historical financial crises (Kaplan and Minton, 2012). During previous crises, stock price returns were affected and showed a downwards direction, before many CEO turnovers where realized. This suggests that having negative stock price returns increases the chances of a CEO layoff, which is the primary focus of this thesis. I therefore expect there to be a negative relationship between stock price return and CEO turnover. Aside from making this comparison with prior events, it might be interesting to study the determinants of these relative high turnover figures. More importantly, what consequences do these actions represent for investors? And what do these findings represent for investors? Basic textbook knowledge would suggest that an increased possibility of CEO turnover would attract new investors as new leaders would keep the company in line and lead them to a road of prosperity (Virany and Tushman, 1986), which is obviously beneficial for the rate of return for the investors. Should they invest in companies with diminishing stock price returns, as these companies have a higher probability of firing the CEO, making way for a new leader and a prosperous future? As being a major event for the company and those involved, CEO turnover will determine and shape the upcoming future course of the firm, as well as future performance. However, the latter is of minor importance within this study, as the primary focus of this thesis is the stock price return and the effects it has on the likelihood of a CEO turnover. The definitions describing these two variables will come back in section 2 and will be further elaborated upon, as well as clearly described, in that particular part.

Figure 1. – The primary focus of this paper is to study the effects of stock price performance on the likelihood of CEO turnover. Regardless of any other possible explanations for the turnover event, this study solely researches

the effects of stock price returns.

1

http://www.bloomberg.com/news/2011-09-01/ceo-turnover-at-six-year-high-as-apple-joins-pg-e-in-transition.html

Within this paper I will primarily examine the effect of stock price performance and its volatility towards the likelihood of CEO turnover. This research will be based on a literature study, complemented by empirical research. CEO turnover as well as its relationship with firm performance can be influenced by a vast number of variables, of which the majority however will not be discussed (in length) within this paper. The remaining variables will be discussed briefly and explained in the literature review section as to come to a clear understanding of underlying variables and relationships concerning CEO turnover rate. The actions of a CEO are the basis for the decision whether or not to dismiss the CEO. In other words; the extent to which the CEO has a direct influence on the course of actions and outcomes, such as firm performance, determine for a large part the decision to initiate CEO dismissal. The actions beyond the control of the CEO are not to be taken into account, as the CEO has no influence on these and should not be held responsible for their outcomes. With this in mind, the relationship between firm performance and CEO turnover thus has to be controlled for some sort of industry average or other kind of industry benchmark. These variables will be further defined and discussed in coming sections (2). Although the relationship between CEO turnover (Brickley, 2003; Eisfeldt and Kuhnen, 2011) and possible explanations such as corporate governance (Lausten, 2002) and firm performance and others related to these events are broadly discussed in scientific literature, this study will solely research this relationship and not any additional explanatory variables.

During this thesis, the relationship between stock price return and the probability of CEO turnover is tested on a sample size of 4995 firms and 589 different CEOs during a 15-year time period and this relationship is expected to be negative. The results turn out to be consistent with these expectations. However, when separating stock price return into two components, known as industry return and excess return, results are inconsistent with expectations and prior literature. One of the key findings resulting from this research is that although it is expected that the negative excess returns influence the likelihood on CEO turnover, it is the industry return that has the most influence on this relationship. This result might indicate that the decision to trigger a CEO turnover is not based on manager performance (excess return), but predominantly determined by industry performance (industry return). This might be explained due to market sentiments.

predict future CEO layoffs, these investors obtain more knowledge on which companies to invest in. This way they can form their investment portfolios while being better informed. The remainder of this study will proceed as follow: In the next section, I will discuss and interpret scientific literature relevant to this particular field of research. Aside from the literature review, the key empirical studies from the past as well as their results based on secondary data will be discussed. Within this literature section, the main variables used are defined and discussed. In the following section, section 3, the dataset used is being described plus the research methods are explained, detailing the databases used, but also the techniques for the statistical analysis. The empirical results are shown and summarized in section 4, whereas section 5 contains the results from further analysis. The conclusion, summary of the results, final remarks and research limitations and possible recommendations for future research are shown in section 6.

2. Literature review

suggests two possible outcomes. At one side, a rational investor could argue that a firm’s decision to fire a CEO is a result of economic down-times, poor performance or negative outcasts for the near future. This might signal the investors that the company is currently in a bad place, facing difficulties or future estimates are not as optimistic as forecasted, making the shares of the company less attractive in the short term. Investors will think twice when considering obtaining shares of this company in its portfolio. On the other side however, investors may reason that the board of a company has decided to fire the current CEO in order to replace him with a better suited person, one who can provide company growth and meet the performance targets. With this replacement of CEO, prosperity might find its way to the company, potentially raising the share price of the company, thereby making the company an interesting investment for investors. Leaving basic reasoning behind, studies and corresponding empirical results show that the general tendency after a CEO turnover represents a rise in stock price of the company involved (Clayton et al., 2003). Earlier work suggests that being a CEO has become increasingly riskier over time. Jensen et al. (2004) report an increase in turnover in the 1990s, albeit small, compared to the 70s and 80s. These findings are consistent with that of Murphy and Zbonjik (2004) and Khurana (2002), reporting a turnover percentage of 12 % in the 1990s versus 10% in the 1970s and 1980s. Furthermore, the average tenure of a CEO has declined dramatically, compared to tenures reported in previous studies for the 1980s and 1990s (Kaplan and Minton, 2012).

stronger related, to stock price movements then is previously assumed in prior literature. For example, there are a number of studies (Coughlin and Schmidt, 1985; Warner et al., 1988; Weisbach 1988; Denis et al., 1997; Murphy, 1999) which only find a very limited relationship between CEO turnover and stock price performance or other measures of firm performance. The general tendency in these studies is the result that a CEO in a top decile performing company has an estimated two to six per cent lower probability of being fired, compared to the CEO of a bottom decile performer. This spread is dependent on the sample taken and the performance indicator used, but overall is within these boundaries, which is rather low. As a result, these authors conclude that the threat of dismissal is not a major incentive for the CEO, as performing good or bad does not make that big a difference. Thus, firm performance provides little incentives for CEOs as it provides little explanation in CEO turnover variation (Brickley, 2003).

A second point worth mentioning and up for debate is the question what really defines a CEO turnover? As mentioned, a lot of research has already been done on this matter. Looking at those specific researches, one finds two general “streams” of definitions concerning CEO turnover; disciplinary turnover, also defined as internal turnover (board initiated) and retirements labeled as normal. Some literature even distinguish a third, the external turnover, which is a consequence of acquisitions (Murphy, 1999). The definition where a CEO turnover is classified as disciplinary comes forth in the papers of Parrino (1997) and Huson et al. (2001). Zhao and Lehn (2003) summarize the definitions used in both researches as follow: “If the news reports that the CEO is fired, forced to step down, or departs due to unspecified policy differences, this turnover is classified as disciplinary. For other cases, if the departing CEO is under the age of 65, and the news announcement reports that the CEO is retiring, but does not announce the retirement at least six months before the effective date, or if the announcement does not report the reason for the departure as involving death, poor health, or the acceptance of another position elsewhere, the CEO turnover is still classified as a disciplinary turnover”.

CEO changes, this event is labeled as a CEO turnover. This includes the possibility that the new CEO was formerly chairman of the board. This latter view is supported by Jenter and Lewellen (2010). By making use of this definition of CEO turnover, they discover opposing results in a sample of publicly traded US firms in the time interval 1992 – 2005, showing large effects of firm performance related to CEO turnover, contrary to prior studies. The major difference in their work and that of prior studies is the fact that Jenter and Lewellen (2010) do not make a clear distinction between voluntary and forced turnover, but simply puts all CEO turnovers in the same bracket. They argue that there is no direct need to distinguish whether a turnover might be voluntary or on a forced basis, as almost all departures are somehow performance-induced. Furthermore, another argument why Jenter and Lewellen (2010) do not distinguish between the two is that they reason that the use of algorithms for classifying turnovers is sensitive to misclassifications and may be the cause of a downward bias in estimating the performance-turnover slope. This view is supported by Kaplan and Minton (2012), who also conclude that many turnovers are potentially misclassified, thereby biasing the results. The turnover-performance sensitivities they find are approximately the same for both forced and voluntary classifications. Earlier, Warner et al. (1988) concluded that although turnovers are slightly more concentrated in the lowest performance decile, the algorithms used are unable to fully capture and model the relationship between performance and turnover. In my opinion, Jenter and Lewellen (2010) have a valid and solid point and in order to avoid the before mentioned potential bias due to any misclassifications, this paper will follow Jenter and Lewellen’s (2010) reasoning and make no distinction between voluntary and forced CEO turnovers. Aside from the misclassifications, the shareholders are primarily concerned whether or not an underperforming CEO departures and not so much if this departure is voluntary or explicitly forced by the board of directors. As a result, their study shows that the relationship between CEO turnover and bad performance is very sensitive, especially when comparing these results to prior research. This focus is consistent with the argument Jenter and Lewellen (2010) provide for not making a distinction between the two departures and I concur with this view and apply it within this paper as well.

available assets. Yet others believe that taking the accounting earnings as a proxy for performance instead of stock returns show superior results because they are less noisy due to the fact they are not subject to industry biases and prove more accurate when comparing and evaluating individual firm performance (Warner et al., 1988). However, most commonly used among researchers is the height of stock prices, both for its easy understanding and applicability, and second for its easy accessibility. This study will follow the majority of researchers, and measure company performance based on stock price. Of course, this has its limitations as well, as it restricts the research to listed firms, which narrows the dataset. Furthermore, one can question if this has any consequences for the results and conclusions, as listed firms might experience more pressure and would therefore increase the likelihood of CEO turnover.

Overall, the two main variables this paper studies are stock price performance and CEO turnover. I expect to find a negative relationship between the two variables, with CEO turnover being the dependent variable and the stock price performance the main independent variable. This makes up for the following hypothesis:

H0: There is a negative relationship between the height of stock price return and the

likelihood of CEO turnover

This research will add explanatory power to the determinants of CEO turnover and its likelihood of occurrence. While the consequences of CEO turnover on stock price performance have been studied thoroughly, the stock price performance as a determinant of CEO turnover is not. Therefore, I expect this thesis to contribute to that particular field of research.

3. Data & methodology

Data collection and sample

In order to conduct the empirical analysis I had to construct a database consisting of several variables. To ensure that the analysis and resulting outcomes are useful for a broader generalization, I included a time-interval of 15 years, spanning from 1996 until 2010.

The dependent variable throughout this research paper entails the event of a CEO turnover. As mentioned before, this event is characterized by a change in CEO identity in two sequential years within the same company. The data on this event is taken from the ExecuComp database, which in turn is part of the COMPUSTAT database, covering all sorts of data with respect to directors and executives. Within the ExecuComp database, the annual compensation file contains information on different specific characteristics regarding the CEO, such as salary and shares held. More importantly, it displays who is CEO at a specified time or time interval per firm (Executive_ID). This can be either through CEO name or unique person-specific identification code per firm-CEO combination. In order to be precise and consistent, thus functioning as a double-check, both are being used in this thesis. The turnover event is constructed as a dummy variable. If the person-specific ID did not match that of the same firm in the previous year, this event was labeled as a CEO turnover, resembled by a 1, 0 if otherwise. As mentioned earlier in the literature review, this study does not make a difference between voluntary or forced turnover, consistent with the work of Jenter and Lewellen (2010) and Kaplan and Minton (2012).

Primary goal of this thesis is the analysis of the relationship between firm performance and changing probabilities in CEO turnover. Being the main independent predictor of this relationship, firm performance is defined within the contents of this paper as the stock price performance of the firm. Whether it is just the stock price of the shares, the change in the overall market or the industry-adjusted excess returns as the subject of study, all necessary data on this part are obtained from the Center for Research in Security Prices (CRSP). This web based database provides information regarding stock prices at a specific time or time interval and their volatility per firm.

When extracting the monthly stock price returns for the S&P 500 firms in the stated time interval, I also included the corresponding SIC code per firm. This was necessary to create the different industry classes according to Fama and French (1997). These authors use a classification of four-digit SIC codes, which in turn leads to 12 different industries.3 See appendix A for this list. This classification of industries is used as a benchmark to measure relative firm performance, compared to its core industry performance. Having done this, I can measure if the performance, be it either good or bad, is a consequence of industry and/or economic performance or merely attributable to its CEO or team of directors. Controlling for this is necessary, as otherwise all industries would be treated equally, while in reality their performance is very different. This is consistent with prior literature. For example, Warner et al. (1988) argue that relative firm performance is a superior predictor of CEO turnover as the firm performance is adjusted for industry specific (thus external) events for which the CEO should not be held responsible as it is outside his or her control. Combining the industry averages and the monthly firm specific stock returns, yields the yearly industry adjusted stock return for each individual company. This measure thus consists of the average annual stock return of a company minus the average annual stock return of the industry.

The total sample consists of fifteen years of data about 4995 firms, corresponding with 589 CEOs. Furthermore, it includes 564 of turnover events and thus 4431 of non-turnover events of which the definition of both is specified above. On average, a firm experiences three CEO turnovers during this fifteen year interval, with a minimum of zero turnovers and a maximum of six turnovers.

To assess whether the validity of the constructed database is acceptable, I had to test its quality. This is done by checking the data for normal distribution and applying corrections

where necessary. These corrections are for example exclusion of outliers or using the natural logarithms of certain variables which might show abnormal values for skewness or kurtosis. In order to study the impact of poor stock price performance on the change in probability of CEO turnover, I use the independent and dependent variables as described above, as well as some control variables as to study other endogenous effects from within the firm-CEO relationship. Furthermore, I use a fixed effects model to capture time invariant and firm invariant observed and unobserved heterogeneity between firms. It is worth mentioning that the primary aim of this thesis is not to analyze all the variables which might influence the relationship between CEO turnover and stock price performance; it merely studies the relationship between these two variables and to which extent stock price performance influences the probability of CEO turnover.

Methodology

This paper studies panel data. In combining cross-sectional data with time-series data, the potential problems of multicollinearity are being mitigated. Since this dataset contains multi-dimensional data over time, the dataset can be characterized as being longitudinal data or panel data; it holds observations on multiple events over multiple periods of time for the same firms. To be more specific, I have a balanced panel data set as the panel data has the same number of time-series observations for each cross-sectional unit (Brooks, 2010). Furthermore, since the dependent variable has a binary outcome (0 or 1), the best applicable model which can be used for this type of panel data is the use of binary logistic regression models, or logistic regressions model.

The outcomes of the logistic regression can either be in log odds, odds ratios or probabilities. These different measurement outcomes are widely used and common in scientific literature, but in order to fully understand these outcomes and their differences I explain the relationship between these different measurements in appendix D.4 During this thesis, the outcomes will be pronounced in terms of odds ratios, while in addition the final results will also be specified in terms of probabilities.

As mentioned before, I also use a fixed effect model in order to control for unobserved heterogeneity between firms. This controls for certain firm-specific aspects which do not vary

over time. These characteristics are not observed, but they might affect the relationship. It is thus of importance to control for these unobserved heterogeneity effects.

During this thesis I will make references to Ordinary Least Squares regressions (OLS) as to make some of the techniques I use or outcomes up for debate better understandable in more common language. The logistic regression for example is run based loosely on the following OLS regression formula:

CEO turnoveri,t = α + β1stock price performancei,t + εi,t

Besides testing the main independent variable, I will also test some control variables. These control variables will be added within this formula to test whether or not these control variables affect the primary relationship between CEO turnover and stock price performance.

Control variables

Aside from stock price performance, CEO turnover might be influenced by other variables in the direct environment as well. The relationship between CEO turnover and firm performance therefore has to be adjusted for these effects. The following variables are used as control variables to test the effect on the stock price return – CEO turnover relationship.

FIRM SIZE – Firm size is being measured as the total assets in millions of dollars. Larger firms are expected to have an increased probability of CEO turnover (Fredrickson et al., 1988).

EMPLOYEES – This measure is a weaker proxy for firm size, as more employees represent a larger firm. This control variable is measured as the total average number of employees within the company in the past year.

SALARY – Salary is being measured as the base salary a CEO receives, expressed in dollars (Fredrickson et al., 1988).

4. Results

This section describes and presents the estimation results corresponding with the empirical analysis of the sample dataset. By taking it step by step, beginning with a simple glance at the data, we end up with the estimation outcomes resulting from the logistic regression model. First of all, let it be perfectly clear again that the primary aim of this research paper is to study the effect of stock price performance on the probability of CEO turnover. All else is not of interest for now, and I am not trying to explain all possible variables of influence on CEO turnover. Having clarified that, this relationship remains the primary focus throughout the analyses.

Figure 2. – Return per event. – This figure shows the mean stock price return if there is a CEO turnover (1) and when there is no CEO turnover (0), respectively 0.1235 and 0.1672. Hence, the non CEO turnover event is shown on the left, yielding a 0.1672 stock price return during that year. While on

the right the figure displays the CEO turnover event, accompanied by an average stock price return of 0.1235.

Figure 2 shows that in a year where there is a CEO turnover event, the mean return of the stock price is 0.1235, which obviously is lower than the mean stock price return in years when there is no CEO turnover (0.1672). This figure displays a discrepancy in the mean returns between the two events.

p-value to the alternative hypothesis is 0.0413, which is significant at the 5 % significance level. This is enough evidence to reject H0 and to conclude that there is a difference between the occurrence of a CEO turnover and the height of the stock return; that is, a CEO turnover is triggered more by lower stock price returns.

Table 1. - The baseline logistic regression model. – This table displays the logistic regression model for CEO turnover. All data are from COMPUSTAT (executive compensation file) and CRPS. The sampled firms include all listed firms from the S&P 500 and the sample period is 1996 through 2010. Turnover is classified as 1 if the ID of the CEO in the current year is different from the CEO ID of the previous year within the same company.

Stock price return is expressed in percentage units. CEO age is in years, CEO salary in $. Company size is expressed as total assets in million $. The number of employees explains itself. Column 1 represents the regression outcomes for only the primary independent variable; stock price return. Column 2 to 5 are as 1 plus one added control variable, while column 6 represents the regression results of all control variables added to the

main independent variable. The regression outcomes are shown in terms of odds ratios, the corresponding p-values are shown in parentheses, whereas the ***, **, * indicate significance at the 1%, 5% and 10% level

respectively.

Dependent variable:

CEO turnover (1) (2) (3) (4) (5) (6)

Stock price return 0.7901 0.7622 0.7505 0.8013 0.7842 0.7318 (-2.02)** (-2.34)** (-2.46)** (-1.90)** (-2.06)** (-2.64)*** CEO age 0.9132 0.9163 (-12.32)*** (-11.66)*** CEO salary 0.9992 0.9993 (-5.95)*** (-5.28)*** Company size 1.00 1.00 (2.12)** (3.02)*** # of employees 1.00 1.0008 (0.79) (2.07)*** # obs 4859 4859 4859 4859 4859 4859

By adding more control variables, it becomes clear that the underlying primary relationship between turnover and return is barely affected. In fact, the odds ratio slightly changes to 0.7318. The results of adding different control variables are also shown in figure 3, in columns 2-5 and, whereas column 6 shows the logistic regression analysis outcomes when all control variables are added. By looking at the logistic regression outcomes, it is obvious that the odds ratio of the primary relationship barely changes and is robust. This has economic significance because an increase in stock price return yields a lower probability of CEO turnover, as expected; a 1 per cent higher return in stock price return will show that the odds of having a CEO turnover is 0,7318 as high as not having a CEO turnover.

treats the explanatory variables as if they were non-random. These non-time varying effects are unobservable, but do have an influence on the primary relationship of study. These fixed effects can be controlled for by performing an analysis based on the panel fixed effect model. Having run this type of regression, the results show different outcomes than under the previous tests, as can be seen in table 2.

Table 2. - Fixed effect panel logistic model. – This table displays the logistic regression model controlled for fixed effects for CEO turnover. All data are from COMPUSTAT (executive compensation file) and CRPS. The sampled firms include all listed firms from the S&P 500 and the sample period is 1996 through 2010. Turnover

is classified as 1 if the ID of the CEO in the current year is different from the CEO ID of the previous year within the same company. Stock price return is expressed in percentage units. CEO age is in years, CEO salary in $. Company size is expressed as total assets in million $. The number of employees explains itself. Column 1

represents the regression outcomes for only the primary independent variable; stock price return. Column 2 to 5 are as 1 plus one added control variable, while column 6 represents the regression results of all control variables

added to the main independent variable. The regression outcomes are shown in terms of odds ratios, the corresponding p-values are shown in parentheses, whereas the ***, **, * indicate significance at the 1%, 5% and

10% level respectively.

Dependent variable

CEO turnover (1) (2) (3) (4) (5) (6)

Stock price return 0.83659 0.8848 0.7624 0.8458 0.8427 0.8127

(-1.53) (-1.01) (-2.28)** (-1.44) (-1.45) (-1.65)* CEO age 0.8038 0.8161 (-15.77)*** (-14.75)*** CEO salary 0.9973 0.9972 (0.000)*** (-10.07)*** Company size 1.00 0.9999 (1.28) (-1.16) # of employees 1.00 1.0042 (1.04) (2.25)** # obs 4465 4465 4465 4465 4465 4465

normal circumstances of the standard logistic regression analysis, the odds ratio was 0.7318 (-2.64). Having controlled for unobserved heterogeneity by using the fixed effects model, the odds ratio became 0.8127 (-1.65). This ratio is both statistically significant as well as economically relevant, as the odds of a CEO turnover under increasing stock price returns are less than the odds of a CEO not getting fired under these circumstances. A 1 per cent higher return in stock price return will yield that the odds of CEO turnover is 0,8127 as high as the odds of not having a CEO turnover. This is consistent with this thesis’ expectations that the relationship between turnover and stock price performance would prove to be negative.

5. Further analysis

proves us that the primary underlying relationship between turnover and stock return is not just a coincidental outcome of logistic regression analysis.

During this entire analysis this thesis has looked solely at the stock price return versus CEO turnover. However, when reasoning further, one can argue whether or not this measure of performance is adequate and justifiable in order to layoff CEOs. Is it not slightly unfair, to say the least, to fire a CEO if the company’s stock return is doing poorly, but compared to the overall market is outperforming this market? The issue that arises is whether or not the turnover is just a means to cover up weak performance, or is it that the CEO can truly be held responsible for the firm’s poor performance? In order to properly assess this question it is important to make a clear distinction between manager inflicted poor performance or due to extraneous forces such as poor economic tides (Furtado and Karan, 1990). Clearly, if the firm performs poorly due to the latter, it is far from reasonable to hold the CEO (solely) accountable. Aside from the different measures of (firm) performance, Kaplan and Minton (2012) found that the relationship between firm performance and CEO turnover has three different dimensions. They assess the performance of the industry relative to the stock market, the performance of the firm relative to the industry and third, the performance of the overall stock market. The sensitivity of turnover related to these dimensions of performance show a statistical significance and is economically meaningful. This has led to some interesting results in a sample from the 1980s, where Morck et al. (1989) interpret their results as CEO turnovers being sensitive to performance relative to its industry, but a low relationship with poor industry performance. Consistent to the studies of Kaplan and Minton (2012) and Furtado and Karan (1990), this paper will also compare the performance of the individual firm with the overall performance and that of its industry.

According to basic textbook reasoning, one might expect this to happen, as CEOs underperform if the excess return is proven to be negative. On the reverse, it is highly unlikely a CEO would be laid off if he outperforms the industry or overall market as this clearly benefits the company. Having made a clear distinction between return and excess return, figure 3 shows the results when putting these variables into a graph. From this graph it is obvious that when there appears to be a CEO turnover, the mean of the excess return of the stock price is negative (-0.016). The opposite is true when there is no CEO turnover, as the mean of the excess return of the stock price is then positive (0.0068). This indicates that a negative excess stock performance might trigger a CEO turnover.

Figure 3. – Excess return per event. - This figure shows the excess return per event. We can see that when there is a turnover event (1), the excess return in that year is negative. While on the other hand, the excess return

is positive when there is no CEO turnover. This makes a clear distinction between positive and negative excess returns and how this information is being used in decision making regarding the dismissal of the CEO.

Table 3. - Logistic regression model with excess return and industry return. – This table displays the logistic regression model for CEO turnover, while the main independent variable stock price return is now split into excess return and industry return. All data are from COMPUSTAT (executive compensation file) and CRPS. The

sampled firms include all listed firms from the S&P 500 and the sample period is 1996 through 2010. Turnover is classified as 1 if the ID of the CEO in the current year is different from the CEO ID of the previous year within the same company. Stock price return is expressed in percentage units. CEO age is in years, CEO salary

in $. Company size is expressed as total assets in millions of $. The number of employees explains itself. Column 1 represents the regression outcomes for only the primary independent variables; excess return and industry return. Column 2 to 5 are as 1 plus one added control variable, while column 6 represents the regression

results of all control variables added to the main independent variable. The regression outcomes are shown in terms of odds ratios, the corresponding p-values are shown in parentheses, whereas the ***, **, * indicate

significance at the 1%, 5% and 10% level respectively.

Dependent variable CEO turnover (1) (2) (3) (4) (5) (6) Excess return 0.8489 0.794 0.8235 0.8575 0.8389 0.7687 (-1.18) (-1.66)* (-1.40) (-1.11) (-1.24) (-1.86)* Industry return 0.7028 0.7145 0.6403 0.7173 0.7052 0.6758 (-1.91)** (-1.82)* (-2.39)** (-1.79)* (-1.88)* (-2.08)** CEO age 0.9134 0.9165 (-12.28)*** (-11.62)*** CEO salary 0.9992 0.9992 (-5.95)*** (-5.26)***

Company size 1.00E+00 1.00E+00

(2.1)** (2.98)***

# of employees 1 1

-0.82 (2.10)**

# obs 4845 4845 4845 4845 4845 4845

Furthermore, this finding is highly economically relevant. Since it was expected that the excess return would be a perfect indicator of future CEO turnover, it appears that not the excess return, but the industry return has greater explanatory power for changing CEO turnover probabilities. This is somewhat striking, as this finding suggests that CEOs are being fired due to, or despite, market performance instead of being judged solely on their endogenous efforts. It is being reasoned that the CEO and his or her actions are responsible for realized excess returns (be it positive or negative). Instead of basing the judgment whether or not to replace the CEO on excess return (thus his/her actions), the results suggest the contrary. Let us consider the analysis based under the assumption of controlling for unobserved heterogeneity first, before we discuss this matter further.

Table 4. - Logistic regression model with excess return and industry return. – This table displays the logistic regression model for CEO turnover, while the main independent variable stock price return is now split into excess return and industry return. All data are from COMPUSTAT (executive compensation file) and CRPS. The

sampled firms include all listed firms from the S&P 500 and the sample period is 1996 through 2010. Turnover is classified as 1 if the ID of the CEO in the current year is different from the CEO ID of the previous year within the same company. Stock price return is expressed in percentage units. CEO age is in years, CEO salary in $. Company size is expressed as total assets in million $. The number of employees explains itself. Column 1 represents the regression outcomes for only the primary independent variables; excess return and industry return. Column 2 to 5 are as 1 plus one added control variable, while column 6 represents the regression results of all control variables added to the main independent variable. The regression outcomes are shown in terms of odds ratios, the corresponding p-values are shown in parentheses, whereas the ***, **, * indicate significance at the 1%,

5% and 10% level respectively.

Dependent variable CEO turnover (1) (2) (3) (4) (5) (6) Excess return 0.897 0.8956 0.8964 0.9034 0.897 0.8855 (-0.79) (-0.76) (-0.80) (-0.74) (-1.45) (-0.83) Industry return 0.7446 0.8751 0.5482 0.7573 0.7608 0.685 (-1.58)* (-0.68) (-3.03)*** (-1.49) (-1.45) (-1.81)** CEO age 0.8043 0.817 (-15.73)*** (-14.67)*** CEO salary 0.9972 0.9972 (-11.29)*** (-10.05)***

Company size 1.00E+00 1.00E+00

(-1.26) (-1.18)

# of employees 1 1

(-1.02) (2.23)**

As can be seen from the table above, controlling for fixed effects still leaves the industry return statistically significant when adding the control variables. Although the previous graph showed similar results, the outcomes are unexpected. As it turns out, it is not the excess return that is major predictor of future CEO turnover. Rather, it is the industry return which has the greatest effect on the likelihood of CEO turnover. Being statistically significant at the 5 % significance level, the odds ratio shows 0.685. This ratio is highly economically significant, as it shows that the probability of a CEO turnover is almost one-third as high as the odds of no CEO turnover when the industry return is lacking in performance. A 1 per cent increase in industry return leads to the odds of a CEO turnover is 0,685 as high as not having a CEO turnover. In other words, the probability of a CEO turnover decreases as the industry return increases. It is rather puzzling if this termination is considered disciplinary, considering the perspective of theoretical literature on relative performance evaluation. This literature suggests that exogenous industry shocks should be cut out of the turnover decision (Gibbons and Murphy, 1990). This result is interesting, as it suggests that the chances of a CEO being fired due to poor performance is related to the overall industry performance, instead of the CEO’s own actions and efforts. No matter how hard she has worked to beat the market, if the overall market is doing poor, the probability of a CEO turnover is higher than it would be if she has performed relatively poorly compared to normal or above average industry performance. This finding is of economic relevance, as the results suggest the CEOs are being weighed against industry performance instead of excess returns.

to the new industry conditions. Hence, CEOs are not judged based on excess returns, but rather on industry performance.

6. Conclusion

The aim of this thesis is to research the relationship between stock price performance and CEO turnover. It is expected that relative poor stock price performance will increase the likelihood of CEO turnover. As this is the primary relationship under study, I have not tried to list all possible explanations for a CEO turnover. Instead, I have only focused on these two variables and to which extent stock price performance influences the likelihood of CEO dismissal. This has led to testing the following hypothesis:

H0: There is a negative relationship between the height of stock price return and the

likelihood of CEO turnover

In order to examine this relationship, data of all listed companies on the S&P 500 have been researched during the time interval 1996-2010. This has led to a dataset consisting of 4995 firms and 585 CEOs. As CEO turnover is a binary dependent variable and the dataset has panel characteristics, I have used a logistic regression model for the data analysis.

The regression results show outcomes consistent with previous literature (Jenter and Lewellen, 2010; Furtado and Karan, 1990). When testing a logistic regression with solely the dependent variable and main independent variable – stock price performance and CEO turnover – the results show a statistically significant results and the odds ratio show 0.79. This outcome tells us that the odds of a CEO turnover are higher than the odds of not having a CEO turnover if the stock price performs relatively poor. The same results are shown when controlling for the different control variables such as CEO age, firm size and number of employees. In fact, the outcomes become highly statistically significant, while the odds ratio remains roughly the same. This means that the stated hypothesis can be accepted, as a poor stock price performance increases the likelihood of a CEO dismissal, as is consistent with previous studies (Warner et al., 1988) .

meaningful, as poor stock price performance increases the likelihood of CEO dismissal. Again, the null hypothesis can be accepted.

Lastly, I performed an analysis where the stock price performance is split into two components; industry performance and excess return. It was expected that the decision whether or not to dismiss the CEO is dependable on the excess return, as the excess return expresses the stock price return compared to the market. If the excess return is positive, it might indicate that the CEO has done a remarkable effort to boost company growth and should be rewarded. On the other hand, if the excess return is negative, it is to be expected that the CEO has done a poor job and lags the industry average, which in turn will lead to her dismissal. Therefore, I expected the excess return to be a perfect predictor of CEO turnover. However, the results from the regression analysis show otherwise. Instead of the excess return being the primary independent variable, it is the industry return that has a major influence on CEO turnover. The results under the normal logistic regression as well as under the fixed effects panel regression model are largely the same and show statistical significance. Perhaps even more important, the results are highly economically relevant, as the decision to dismiss a CEO is largely based on industry performance. This particular field of research is relatively new and has not been studies sufficiently in scientific literature. It is only recently that researchers has started to study the cause of this relationship (Eisfeldt and Kuhnen, 2013; Kaplan and Minton 2012).

References

Adams, R. B., and Ferreira, D., 2008, A theory of friendly boards, Journal of Finance, 62, 217-250

Brickley, J.A., 2003, Empirical research on CEO turnover and firm-performance: A Discussion, Journal of Accounting and Economics, 36, 227-233

Brooks, C., 2010, Introductory econometrics for finance, Second edition, Cambridge University Press

Clayton, M., Hartzell, J., and Rosenberg, J., 2003, The Impact of CEO Turnover on Equity Volatility, Federal Reserve Bank of New York, Staff Report no. 166

Coughlan, A., and Schmidt R., 1985, Executive compensation, management

turnover, and firm performance: An empirical investigation, Journal of Accounting and Economics, 7, 43-66.

Denis, D.J., Denis, D.K., and Sarin, A., 1997, Ownership structure and top executive turnover, Journal of Financial Economics, 45, 193-221

Eisfeld, A.L., and Kuhnen, C., 2013, CEO Turnover in a Competitive Assignment Framework, Journal of Financial Economics, 109, 351-372

Eisfeldt, A., and Rampini, A., 2008, Managerial incentives, capital reallocation, and the business cycle, Journal of Financial Economics, 87, 177–199

Fama E.F., and Jensen M.C., 1983, Separation of Ownership and Control, Journal of Law and Economics, 301- 325

Fredrickson, J., W., Hambrick, D., C., and Baumrin, S., 1988, A model of CEO dismissal, The Academy of Management Review, 13(2), 255-270

Furtado, E.P.H., and Karan, V., 1990, Causes, consequences, and shareholder wealth effects of management turnover: a review of the empirical evidence, Financial Management, 19 (2), 60-75

Gibbons, R., and Murphy, K.J., 1990, Relative performance evaluation for chief executive officers, Industrial and Labor Relations Review, 43, 30S–51S

Hillier, D., Grinblatt, M., and Titman, S., 2008, Financial markets and corporate strategy, McGraw-Hill

Huson, M.R., Parrino, R., and Starks, L., 2001, Internal monitoring mechanisms and CEO turnover: a long-term perspective, Journal of Finance, 56, 2265–2297

Jensen, M., Murphy K., and Wruck E., 2004, CEO Pay . . . and How to Fix It, Working paper, Harvard Business School

Jenter, D., and Kanaan, F., CEO turnover and relative performance evaluation, Journal of Finance, forthcoming.

Jenter, D., and Lewellen, K., 2010, Performance-induced CEO turnover. Unpublished working paper. Stanford University and Dartmouth College.

Kaplan, S., and Minton, B., 2012, How Has CEO Turnover Changed? Increasingly performance sensitive boards and increasingly uneasy CEOs, International Review of Finance, working paper series (National Bureau of Economic Research), no. 12465

Khurana, R., 2002, Searching for a corporate savior: The Irrational Quest for Charismatic CEOs, Princeton University Press.

Lausten, M ., 2002, CEO turnover, firm performance and corporate governance: empirical evidence on Danish firms, International Journal of Industrial Organization, 20, 391–414

Mikkelson, W.H., and Partch, M.M., 1997, The decline of takeovers and disciplinary managerial turnover, Journal of Financial Economics, 44, 205-228

Morck, R., Shleifer, A., and Vishny, R., 1989, Management Ownership and

Market Valuation: An Empirical Analysis, Journal of Financial Economics, 20, 293– 315

Murphy, K.J., 1999, Executive Compensation, In O. Ashenfelter and D. Card (eds.), Handbook of Labor Economics. Volume 3, North Holland, 2485-2525

Murphy, K.J., and Zabonjik, K., 2004, Managerial Capital and the Market for CEOs, working paper, USC.

Patro, S., Lehn, K., and Zhao, M., 2003, Determinants of the size and structure of corporate boards: 1935-2000, Financial Management, 38, 1-59

Parrino, R., 1997, CEO turnover and outside succession: A cross-sectional analysis, Journal of Financial Economics, 46, 165-197

Viranya B., and Tushman M.L., 1986, Top Management Teams and Corporate Success in an Emerging Industry, Journal of Business Venturing, 261-274

Warner, J.B., Watts R.L., and Wruck K.H., 1988, Stock prices and top management changes, Journal of Financial Economics, 20, 461-492

Weisbach, M.S., 1988, Outside directors and CEO turnover, Journal of Financial Economics, 20, 432-460

Introduction to SAS. UCLA: Statistical Consulting Group.

Appendices

Appendix A – Fama and French industry list

This appendix lists the 12 industries I use in this thesis. The table is composed based on different industry classifications and the corresponding SIC codes as on the website of French.

http://mba/tuck/dartmouth.edu/pages/faculty/ken.french/data_library.html Fama and French 12 industry

portfolio's

Industry Types of companies SIC codes 1. Consumer non-durables Food, Tobacco, Textiles, Leather 0100-0999, 2000-2399, 2700-2749, 2770-2799, 3100-3199, 3940-3989 2. Consumer

durables Cars, TV's, Furniture

2500-2519, 2590-2599, 3630-3659, 3710-3711, 3714, 3716, 3750-3751, 3792, 3900-3939, 3990-3999 3. Manufactoring Machinery, Trucks, Planes, Paper 2520-2589, 2600-2699, 2750-2769, 3000-3099, 3200-3569, 3580-3629, 3700-3709, 3712-3713, 3715, 3717-3749, 3752-3791, 3793,3799, 3860-3839, 3860-3899 4. Energy

Oil, Gas, Coal

extraction 1200-1399, 2900-2999 5. Chemicals

Chemicals and allied

products 2800-2829, 2840-2899 6. Business equipment Computer, Software, Electronic Equipment 3570-3579, 3660-3692, 3694-3699, 3810-3829, 7370-7379 7. Telecom Telephone and Television Transmission 4800-4899 8. Utilities Utilities 4900-4949

9. Shops Wholesale, Retail 5000-5999, 7200-7299, 7600-7699 10. Healthcare

Medical equipment,

Healthcare, Drugs 2830-2839, 3693, 3840-3859, 8000-8099

11. Money Finance 6000-6999

12. Other

Mines, Hotels, Serv.

Appendix B – T-test

This appendix shows the test outcomes resulting from performing a t-test. From the table is it clear that there are 4308 non turnover events, while there are 551 CEO turnovers. Furthermore, it shows the mean stock price returns for both events, respectively 0.1672 and 0.1236. The test results show that the null hypothesis that there is no significant difference between the two events can be rejected. The alternative hypothesis states that a higher

stock price return leads to a lower probability of CEO turnover. This results is 0.0413 which is significant at the 5% significance level.

Pr(T < t) = 0.9587 Pr(|T| > |t|) = 0.0826 Pr(T > t) = 0.0413 Ha: diff < 0 Ha: diff != 0 Ha: diff > 0 Ho: diff = 0 degrees of freedom = 4857 diff = mean(0) - mean(1) t = 1.7363 diff .0437016 .0251696 -.0056423 .0930455 combined 4859 .162298 .0079824 .5564259 .1466489 .1779471 1 551 .123552 .0186958 .4388532 .0868282 .1602759 0 4308 .1672536 .0086778 .5695683 .1502408 .1842665 Group Obs Mean Std. Err. Std. Dev. [95% Conf. Interval]

Appendix C – OLS regression

The table below shows the Ordinary Least Square regression results. As can be seen, the underlying relationship between stock price performance and CEO turnover remains negative.

The table below shows the Ordinary Least Square regression results. As can be seen, the underlying relationship between stock price performance and CEO turnover remains negative, even when controlling for robust standard

errors.

_cons .6361416 .0426286 14.92 0.000 .5525697 .7197135 Employees .0000713 .0000509 1.40 0.161 -.0000285 .0001712 size 1.11e-07 4.29e-08 2.60 0.009 2.73e-08 1.96e-07 salary -.0000604 .0000115 -5.26 0.000 -.0000829 -.0000379 age -.0083369 .0006991 -11.93 0.000 -.0097074 -.0069664 return -.014686 .0087715 -1.67 0.094 -.0318823 .0025103 ceo_turnov~1 Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .30932 R-squared = 0.0412 Prob > F = 0.0000 F( 5, 4724) = 36.11 Linear regression Number of obs = 4730

Appendix D – Log odds, odds ratios and Probabilities explained

Log odds, odds ratios and probabilities are all related to each other, but represent the outcomes in different forms. For example, let’s start with the probability. Suppose the probability of success p (turnover or any other

kind of event) is 0.8, meaning that the probability of failure q is 0.2 (=1-p). Next we can define the odds of success as the ratio between success and failure. Hence, the odds of success are p/q = 0.8 / 0.2 = 0.4. In describing this outcome, it is said that the odds of success are 4 to 1, while the odds of failure would be 1 to 4

(0.2 / 0/8 = 0.25). As we can see, the odds of success and failure are reciprocals of one another. The last step towards the odds ratios is slightly more difficult. First the different probabilities and odds have to be calculated,

to come up with the odds ratios. Suppose that 7 out of 10 CEOs will get fired when the stock price return is negative while 2 out of 10 CEOs will be laid off if the stock price return is positive. The probability p for a CEO turnover during negative stock returns is 7 / 10 = 0.7, and the probability q for a non- CEO turnover event during

negative stock returns is 1 – p = 0.3. The probability p for a CEO turnover during positive stock returns is 2 / 10

= 0.2, while the probability q for a non-CEO turnover event during positive stock return is 1 – p = 0.8. With the calculated probabilities, the odds for both scenarios can be derived. The odds for a CEO turnover during negative

stock returns is 0.7 / 0.3 = 2.333, while the odds for a non-CEO turnover during positive stock returns is 0.2 / 0.8 = 0.25. To compute the odds ratio for CEO turnover, the odds are divided between each other; 2.333 / 0.25 = 9.33. In common language, this odds ratio outcome means that odds for a CEO turnover during negative stock