Amsterdam Business School
MSc Business Economics, Finance Track
Master Thesis
The Effect of the Supply of CEOs on CEO
Turnover
Author:
Reinier H. de Groot
reinier.degroot@student.uva.nl
Student number 10188282
Supervisor:
Dr. Florian S. Peters
f.s.peters@uva.nl
July 6, 2016
1
Statement of Originality
This document is written by Student Reinier de Groot who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
2 Abstract
This empirical paper studies the effect of the supply of CEOs on the CEO dismissal
probability during a market downturn with use of a dataset containing CEO turnovers of a sample of North-American firms in the period of 1996 to 2014. By means of logit and probit regressions I show that the CEO dismissal probability increases during a market downturn. Furthermore, I provide supportive evidence of an imperfectly functioning relative performance evaluation. However, not enough evidence is found to conclude that the supply of CEOs affects the probability of a forced turnover.
3 Table of Contents
1. Introduction 4
2. Literature Review 6
2.1 Performance-based turnover in general 7
2.2 Performance measures 9
2.3 The labor market in context of executive turnovers 10
2.4 Hypotheses 11
3. Data description 11
4. Methodology 16
5. Results 19
5.1 Main logit regression results 19
5.2 Logit regression results with separate crisis indicators 21
6. Robustness Checks 26
6.1 Alternative performance measures 26
6.2 Alternative age threshold for forced turnover classification 26
6.3 Outside options proxy based on total assets 29
6.4 Adding industry fixed effects 32
7. Conclusion 32
4 1. Introduction
CEO turnover has been and still is a subject of interest and topic of finance-related studies. Not only because of scientific interest, but also because of the large media coverage of wealthy and successful CEOs. This may give the impression that in general CEOs do well and contribute to firm performance. This, however, is not always the case and reason enough to study the subject extensively. Moreover, the global business cycle has been fluctuating severely the last two decades due to financial crises. It is very likely that the CEO turnover and selection process has been affected as well. CEO turnover is the event in which the chief executive officer of a firm is replaced by a new one. The terms CEO turnover, executive turnover, CEO succession and turnover are used
interchangeably in this thesis and thus refer all to the same kind of event. CEO dismissal refers to a forced CEO turnover.
Executive turnover has been studied extensively within the field of finance and corporate governance from different points of view. An executive turnover may happen because of several reasons, including retirement, death, illness, voluntarily leaving the position or being removed by the firm due to poor firm performance. The last example of a forced turnover can be part of the corporate governance mechanism of a firm, i.e. creating shareholder value by replacing a poor performing CEO and reducing agency costs.
Assessing the relation between the firm’s performance and the CEO’s behavior remains to be a challenging task. An important aspect in this view is to establish which factors causing poor firm performance are within the CEO’s control and which are not. The ability to determine these factors and the willingness to act accordingly have a considerable impact on the executive turnover decisions. During an industry downturn, the ability, willingness or both are affected in a way that results in more frequent
turnovers (Jenter & Kanaan, 2015). Firms and investors are less capable of objectively evaluating the firm performance relative to others and thereby judging whether it relates to the quality of the CEO or the difficult circumstances in which the firm is in. Hence, they tend to replace him or her more easily because the relative evaluation of the CEO suffers from the collective poor performance of firms in the same industry (Jenter & Kanaan, 2015). This is a factor that increases executive turnover during an industry downturn. Adversely, a CEO himself can make the decision to resign during an industry downturn to explore the options in another industry. This is a second factor that may
5
increase turnover during an industry downturn. During an economy-wide crisis, however, the two mentioned factors increasing turnovers (poor firm performance evaluation and voluntary resignation by CEO) may have a smaller effect on the turnover rate.
The cause of the reduced effect may relate to the reduced functioning of the labor market as a whole, but more specifically the executive labor market. There are fewer job opportunities, or outside options, for executives when the complete economy
experiences a downturn, than when a specific industry experiences a downturn. The supply of well performing executives is low, while the demand is high. The consequence would be that firms may retain their poor performing CEOs due to a shortage of superior executives in the labor market. From the CEO’s perspective, a CEO might be less inclined to voluntarily leave his or her position, knowing that there are fewer job opportunities at other firms. In this sense, executive labor mobility is considerably lower during a crisis. The lack of outside options for both the firm and the CEO may disturb the
disciplining function of a turnover within governance mechanisms and may lead to new insights in this respect, which has not been extensively researched by previous studies.
Eventually there are two opposite forces driving executive turnover during a crisis. On the one hand there is the decreased monitoring and evaluation ability or willingness, leading to an increase of forced CEO replacements, even though all firms perform poorly. On the other hand there is the hampered functioning of the executive labor market, which decreases voluntary and forced CEO replacements. In this thesis I focus on the hampered functioning of the labor market in which there is a smaller supply of CEOs for firms to choose from. This line of reasoning leads to the research question of this paper: Does the supply of CEOs affect executive turnover during a crisis?
My research extends Jenter and Kanaan (2015) who find that industry
performance affects the executive turnover frequency. They further found that market performance in terms of stock returns has a similar influence on the turnover frequency, but to a small extent. Their research looks primarily into the lower degree of relative performance evaluation which drives the increase in turnovers. My research adds the functioning of the executive labor market as a second driver of the change in the number of turnovers. I perform logit and probit regressions on the probability of a turnover, using a dataset on U.S. companies ranging from 1996 to 2014. This time period
6
and the financial crisis of 2007-2008, extending into the Great recession in the years thereafter. Assuming that the two drivers have opposite effects would explain why Jenter and Kanaan found the increase to be significantly smaller during an entire market downturn.
I hypothesize that the two drivers for a change in the number of turnovers have opposite effects during a crisis. I posit that the lesser functioning labor market has a negative effect (fewer turnovers, i.e. smaller probability of a turnover) and that the hampered ability of relative performance evaluation has a positive effect (more turnovers during a crisis). The combined effect is ambiguous, although the study of Jenter and Kanaan (2015) suggests that the number of turnovers would increase in a market downturn. Their finding was, however, of such a small magnitude that it may differ when using a different dataset and different control variables.
The decision to replace a poor performing CEO is an important one since it can create or destroy shareholder value, depending on the CEO’s quality. Therefore it is beneficial to know when a CEO should be fired and when not. A CEO dismissal affects future decisions concerning investment, firm operations and firm financing. Considering all aspects of the market, the board should be able to distinguish which factors of poor firm performance are within the CEO’s control and which are caused by a market
downturn in the form of an economic crisis or an industry downturn. When considering the labor market as well, knowledge on this driver helps firms to decide if a replacement is appropriate at the time, given that there is a small or lacking supply of superior CEOs. If this is the case, a turnover would not be value enhancing. Knowledge on when the board should and should not fire the CEO is necessary to make the right decision.
The remainder of this thesis is structured as follows. Section 2 places my research in context of previous literature and offers the methodological basis to justify the
regression specification. Section 3 describes the dataset being used and shows how the variables are constructed. Section 4 elaborates on the methodological approach. Section 5 shows the empirical results, which are accompanied by an economic interpretation or explanation. Section 6 consists of several robustness tests. Section 7 concludes.
2. Literature review & Hypotheses
The following literature review is divided in three sections, presenting knowledge on various aspects of my research: Performance-based turnover in general; Performance
7
measures; and the labor market in context of executive turnovers. I conclude this section with a brief description of my hypotheses.
2.1 Performance-based turnover in general
An early research on the topic of executive turnover, which lays the foundation for many later studies on the subject, is done by Weisbach (1988). He establishes that there is an association between prior performance and CEO resignation, but, more importantly, that this association is even stronger for companies with outsider-dominated boards,
compared to companies with insider-dominated boards. Outside directors on boards seem to be able to effectively monitor top management and thereby taking value-increasing actions. One of the most important of these actions is firing the CEO. Positive abnormal stock returns on the day of announcing the removal of top management acknowledge the increase in value by the market. Monitoring management is an important duty of the board and outsiders seem to do a better job than insiders in this respect. Inside directors’ career opportunities often depend on the CEO and they would therefore not contradict the management, whereas the outside directors do not
experience such a bias. The latter group, however, must have some incentive to remove bad management and increase firm value. Hence, outside directors usually have a stake in the firm in the form of stocks or are otherwise incentivized by maintaining a good reputation by being a director at a well performing firm.
Note, however, that although an outsider-dominated board generally makes value-increasing decisions, the board members themselves should not occupy too many boards at a time, as this will worsen their decisions (Fich & Shivdasani, 2006).
Weisbach (1988) performs a logit regression to estimate the likelihood of a CEO leaving his position within a given firm-year, consistent with previous literature1, for
both firms with insider-dominated boards and firms with outsider-dominated boards. Probit and logit regressions are also the method of choice for my thesis, as it is a standard method to research effects on CEO turnover.
Borokhovich, et al. (1996) likewise use a maximum likelihood estimation and support and extend Weisbach (1988) by showing that there is a strong and positive relation between the fraction of outside directors on the board and the likelihood of selecting an outside CEO for both voluntary and forced turnovers. This indicates that
8
outsider-dominated boards not only prefer outside candidates as CEO successors in case of poor performance, but also in case of a routine turnover.
The preferred choice for an outside CEO brings the supply of executives into the equation and shows the importance of the labor market. Moreover, the likelihoods of turnover and outside succession depend on industry characteristics as well. A higher degree of homogeneity of an industry, i.e. firms in an industry are more similar to each other, increases the likelihood of (forced) turnover and outside succession (but within the same industry), compared to a heterogeneous industry (Parrino, 1997). Parrino further adds that new CEOs from outside a homogeneous industry, who are able to change the direction of a firm, are more attractive when the firm is not performing well. In good times, however, the firm prefers a CEO who can pursue the current strategies. This finding points out that during bad times, such as a crisis, the supply of outside executives plays an even more important role than during good times.
Besides the number of outsiders on the board, the number of women represented on the board also impacts firm value (Ahern & Dittmar, 2012). Using a recently accepted Norwegian law concerning the number of women on board of director positions as a natural experiment, the authors find a significant decrease in stock returns and a
decreasing Tobin’s Q over the following years, with an increase of the number of women on the board. This result indicates that firms that are affected by the quota are being constrained in maximizing value by choosing their own board members. This shows the effect of women on the board on firm performance, but does not show the effect on the probability of having a female CEO succession. This result stresses the importance of the board and executive selection in view of firm performance.
Bertrand & Mullainathan (2001) and more recently Kaplan & Minton (2012) and Jenter & Kanaan (2015) focus on the issue of whether CEOs and turnover decisions are affected by factors beyond their control. Underperforming CEOs, relative to their peers, are as expected more likely to be fired. In addition though, even if the entire industry or market takes a downturn, the CEO turnover rate increases. This effect is larger for firms with poor governance. This is the result of the apparent inability of the board to perfectly evaluate the CEO, when the peer group firms performs poor as well. In other words, firms and investors are less willing or less capable of objectively evaluating the CEO relative to other firms in the same industry. This evaluation is called relative
9
frequency during a downturn of the entire market is smaller than during an industry downturn. The research further establishes that CEO tenure has a negative effect on the probability of dismissal, supporting other papers such as Peters & Wagner (2014) and Kaplan & Minton (2012). The authors, however, do not incorporate the executive labor market in their study. Especially during a market downturn or crisis the supply and demand of executive labor might play a larger role in executive turnover. There is high demand for well performing CEOs, but a low supply of them. Firms perhaps hold onto their inferior CEOs due to the lack of supply of superior ones. My thesis adds to Jenter & Kanaan (2015) and turnover literature as a whole by considering and quantifying this extra dimension.
2.2 Performance measures
An aspect to deal with in this research is the use of appropriate performance measures in context of performance-based CEO turnover. Performance is generally evaluated relative to a peer group.
An often used performance measure is the difference between the firm’s stock return and the return on a value-weighted market portfolio for the year or quarter before the CEO left. Weisbach (1988) and others have used this successfully before him, including Coughlan & Schmidt (1985) and Warner, et al.(1988). All studies show a negative relation between the stock return and the probability of a CEO turnover. Also in more recent literature, the stock return remains a performance measure of choice. This measure is directly related to creating shareholder value and is thus appropriate, especially in the field of corporate governance. Quarterly returns are preferred over yearly returns as they possibly leave a large gap between the turnover and the
performance measure, which affects the accuracy of it. The return could alternatively be computed by taking the residuals of a regression of the capital asset pricing model, but this would bias the returns downwards for bad CEOs. The reason for this is that the model is very likely estimated using data of several years prior the turnover. It does not reflect the CEO’s performance during crisis times only.
The stock returns performance measure can in another way be adjusted with respect to the industry by subtracting the mean or median stock return of the industry at the time of succession (Parrino, 1997). The same can be done for accounting earnings as a performance measure in the form of the return on assets (ROA), which equals the
10
ratio of net income to book assets2. One subtracts the industry mean or median ROA to
adjust to the peer group. The use of accounting earnings as performance measure could be helpful because the decision to replace an executive is often made by the board of directors, which also looks at accounting earnings to assess the CEO quality. Among others, the ROA and stock return measures have been previously used by (Parrino, 1997), (Shivdasani & Yermack, 1999), (Bertrand & Mullainathan, 2001), (Huson, et al., 2001), (Kaplan & Minton, 2012), (Eisfeldt & Kuhnen, 2013), (Liu, 2014) and (Jenter & Kanaan, 2015).
2.3 The labor market in context of executive turnovers
The labor market for executives as a whole has not been researched as extensively as CEO turnover on its own. Eisfeldt & Kuhnen (2013) do however look into the link between industry conditions and the CEO labor market. The authors’ findings are consistent with other empirical results on the fact that relative performance evaluation does not function perfectly. Moreover, they introduce the labor market driver of
turnovers and show that the outside options for managers are affected by an industry downturn, which in turn influences the performance evaluation. This leads to different effects on turnover than the standard principal-agent model would predict.
Measuring labor market opportunities for managers is a virtually impossible task and finding a suitable proxy remains challenging as well. Still, Brookman & Thistle (2013) found a proxy which seemingly fulfills its task sufficiently. Per firm, they form a peer group consisting of firms within the same two-digit SIC number and within the range of 50%-200% of the firm’s sales. The authors subsequently distinguish between 12 different job titles, which are noted in ExecuComp, to indicate comparable jobs. Then the number of outside opportunities is measured by the median compensation of
executives with the same job title as the executive in the firm of interest. A similar method has been used before by Bizjak, et al. (2008).
Liu (2014) is one of few who examined the CEO’s network, which proxies as the number of outside options, and relates this to the probability of a CEO turnover. In this research, outside options are measured using four centrality measures borrowed from sociology. She researched executive turnovers from the point of view that CEOs
2 Return on assets can alternatively be calculated as earnings before interest and taxes divided by total
assets. Adjusting this measure to the industry entails subtracting the median or mean value of the industry. This can also be done for the entire market.
11
themselves may decide to resign. The larger the number of outside options of a CEO is, the larger is the probability of a turnover. Moreover, this relation is stronger for poor performers than for strong performers. These findings support my hypothesis that a CEO with fewer outside options during a crisis or market downturn, would have a smaller probability of resigning.
2.4 Hypotheses
On the basis of the concise literature review above, three hypotheses are formulated: H1. A market downturn increases the probability of a forced turnover.
H2. An increase in the supply of CEOs increases the probability of a forced turnover. H3. A decrease of the supply of CEOs due to a market downturn decreases the probability of a forced turnover.
Due to imperfectly functioning relative performance evaluation the board or shareholders are not able to leave a market downturn out of the equation when
considering a CEO dismissal and thus increasing the dismissal probability. Furthermore, if there are many outside options available in terms of other CEOs, a firm may tend to fire its CEO more easily as the likelihood of finding a better performing CEO is larger. Vice versa, if the supply of CEOs is low, the chance of finding a superior CEO is smaller and thus is the probability of CEO dismissal smaller as well.
3. Data description
In the empirical part of this study, I make use of annual panel data on Chief Executive Officers (CEOs) and their corresponding companies’ balance sheets and income
statements over the period of January 1995 until December 2014 - observations of the year 1995 are left out of the analysis, since these are merely used to calculate returns for 1996. All firms are based in North-America. The data includes dates of joining and
leaving the company by CEOs, their compensation and age, earnings data and S&P500 index returns. All data is downloaded through Wharton WRDS from three databases: Compustat ExecuComp – Annual Compensation; CRSP Stock Market Indexes; and Compustat Monthly updates – Fundamentals Annual. CEOs are identified with use of ExecuComp’s “ceoann” and “titleann” variables. Altogether, my combined dataset contains 33,280 firm-year observations and 3,403 turnovers, of which 1,799 forced. I
12
document a turnover if for a given company the CEO of year t is not the same as the CEO in year t-1. This is checked and if necessary corrected by means of the “became CEO” and “left as CEO” variables recorded in ExecuComp.
As this research deals with relative performance evaluation and thus investigates the consequences of poor performance for the CEO, it is necessary to determine whether a CEO turnover is forced by the board or not. This remains to be a challenging task. A known method of doing so was implemented by, among others, Parrino (1997) and Huson, Parrino, & Starks (2001) which involves reading press reports and establishing the reason for a turnover. This press-based procedure will probably yield the most accurate results but it is obviously a very laborious method and hence not the method of choice for my research. The method used by Peters & Wagner (2014), however, is a rather useful alternative since it only involves using the CEO’s age as a proxy for
whether the turnover is forced or voluntary (for example in case of a retirement). They have tested several age thresholds and, compared with the press-based method, the age of 56 yields the most similar results. Thus, in the main analysis I classify a turnover as forced if the CEO in question is below the age of 56 at the time of turnover. I do,
however, compare the results of different age thresholds in a range of 50 to 60 years old. Besides the simplicity of this age-based classification, the robustness against biases as a result of the amount of press coverage is another advantage of this method. The press-based method is likely to generate false negatives, but unlikely to generate false positives. In other words, some forced turnovers may not be recognized as such, but all forced turnovers reported by the media are usually indeed forced. The age-based method on the other hand will generate both false negatives and false positives and yields supposedly a more balanced classification of forced turnovers.
The second major variable in this research is the variable that accounts for the extent or amount of outside opportunities for the CEOs and firms. For the time being, there has yet to be found an exact measure to identify the supply and demand of the executive labor market. Hence, to include the effect of the labor market one has to rely on the use of a proxy for the outside opportunities. In this case, the proxy is based on the median compensation of managers in a peer group of firms. It is computed as follows. For each firm-year I create a peer group of firms that are in the same two-digit industry and is within the range of 50 to 200 percent of sales in dollars in the year in question. This group of firms serves as possible firms a CEO is able to work, based on the size of
13
his current employer. The proxy itself is the median of the total compensation of the CEOs in this peer group. For the CEO himself/herself, this proxy indicates if there is a small or large demand, or little or many outside opportunities, for CEOs of his caliber at that time. A higher amount of outside opportunities would make it relatively more attractive to voluntarily leave the current position and apply at another firm, ceteris paribus.
More importantly, from the firm’s perspective, if there are more outside
opportunities, it will be more attractive to dismiss the current CEO and hire another one, ceteris paribus. If the supply of CEOs is larger, there will probably be more
well-performing managers available. Or the other way around: in bad times, the amount of outside opportunities is smaller and firms may hold on to their average performing CEO, as there is no better alternative available. In short, the proxy of outside opportunities would be positively related to the probability of a (forced) turnover. To examine the relative performance evaluation in the context of turnovers, a relative performance measure is required. The measures being used are Stock return, Return on Assets (ROA) and Return on Equity (ROE). All three are adjusted for performance of the industry or the market. To adjust for industry performance, I subtract the median industry stock return from the individual stock return in a given year, which results in the industry adjusted return. The same procedure goes for the ROA and ROE measures. To adjust stock returns for the market, I subtract the return of the Standard & Poor’s market index excluding distributions. Some caution is needed with assigning firm performance to a certain CEO. If a turnover took place in the first half of the fiscal year, I assign the
measured performance of the current year to the new CEO and I assign the performance of the previous year to the dismissed CEO since the current year performance is then likely the result of the efforts of the new CEO.
The next step in the analysis is to link the effect of outside opportunities to crisis periods. This requires some indicator for crisis periods. The simplest way to do so is to classify certain years as “crisis years” and create a dummy variable for these years. The dataset comprises two major crises: the bursting of the Dotcom bubble and the financial crisis of 2007-2008. It is generally accepted that these occurred in the periods 1999-2001 and 2007-2008, respectively. Hence this dummy variable equals one if the observation is in one of the years 1999, 2000, 2001, 2007, or 2008.
14
All dollar values are converted to 1995 US dollars using the Consumer Price Index, available at the FRED database (Federal Reserve Bank of St. Louis, 2016). Variables in dollar values are winsorized at a one percent level.
Table 1 presents descriptive statistics. Panel A shows statistics on the frequency of turnovers and whether or not they happen during a crisis or not. Paying attention to the distribution of turnovers across crisis and non-crisis periods, note that turnovers – both forced and unforced – happen in 31 percent of the time during crisis periods, although the crisis observations only account for 27 percent of the dataset. This indicates a slight skewness of turnovers towards crisis periods. Panels B and C show firm and CEO characteristics, respectively.
Figure 1 graphically depicts the development of the number of (forced) turnovers and the outside options proxy per year. For scaling purposes, the average of the outside options proxy is divided by ten. This does not affect the interpretation of the shape of the graph. After the Dotcom bubble burst and during the ‘07/’08 financial crisis, the number of (forced) turnovers increases. The outside options, however, decrease during the ‘07/’08 financial crisis as expected, but increase after the Dotcom bubble burst. The latter is not consistent with my hypothesis.
15 Table 1: Summary Statistics
This table shows an overview of the used dataset in this research. Panel A shows the number of observations and (forced) turnovers as well as the distribution of turnovers and observations between crisis periods and non-crisis periods. Panel B shows
information on the distribution of several firm characteristics. Panel C shows information on the distribution of a number of CEO characteristics. ROA, ROE, Total Assets and Total Equity are winsorized at the 1 percent level. All dollar values are converted to 1995 dollars.
Panel A: General Turnover statistics Non-Crisis
Period
Crisis Period Total
# Firm-Year Observations 24,167 (73%) 9,113 (27%) 33,280 (100%)
# Turnovers 2,364 (69%) 1,039 (31%) 3,403 (100%)
- Of which forced 1,237 (69%) 562 (31%) 1,799 (100%)
- Of which non-forced 1,127 (70%) 477 (30%) 1,604 (100%)
Panel B: Firm Characteristics
Mean Median Std. Obs.
Stock return 0.14 0.05 0.99 30,336 ROA 0.03 0.04 0.10 33,088 ROE 0.09 0.11 0.42 33,087 Total Assets ($mln) 6,650.15 1,392.23 22,025.32 33,108 Total Equity ($mln) 1,771.94 504.66 3,857.05 33,109 Panel C: CEO Characteristics
Mean Median Std. Obs.
Total Compensation ($1000) 4,004.71 2,237.66 8,225.51 33,106
Outside options (proxy) 3,105.39 2,177.54 3,417.13 33,276
Tenure 6.77 5 7.46 32,430
Outsider 0.38 0 0.49 14,639
16 4. Methodology
The goal of my empirical research is to estimate the possible effect of the executive labor market on the forced turnover probability during a financial crisis by means of a proxy for the outside options for CEOs. I start however by estimating some base case
regressions without the influence of the labor market. The most frequently used and proven empirical method for estimating relations between turnover and performance measures is a probit or logit regression. Whether to choose probit or logit largely depends on personal preferences. I choose for logit regressions, since with use of the regression estimates, the probability of a forced turnover can be easily calculated in the following way with N variables (values):
𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = 1/[1 + 𝑒𝑥𝑝(−(𝛽0+ 𝛽1 ∗ 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒1+ 𝛽2∗ 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒2+ ⋯ + 𝛽𝑁 ∗ 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑁))].
In the most minimal form, this encompasses a dependent turnover variable and some independent performance measure. It is however customary to add some control
Figure 1:( Forced) Turnovers and Outside options
Turnovers and forced turnovers seem to behave rather similarly. The outside options proxy seems to be positively correlated in roughly the first half of the sample, but negatively correlated in the second half.
17
variables and possibly time fixed and/or industry fixed effects. Such a basic regression can be written in the form of:
Pr(𝐹𝑜𝑟𝑐𝑒𝑑𝑖,𝑡) = 𝛽0+ 𝛽1∗ 𝑃𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒𝑖,𝑡−1+ 𝛽2∗ 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖,𝑡+ 𝐹𝑖𝑥𝑒𝑑 𝐸𝑓𝑓𝑒𝑐𝑡𝑠𝑖,𝑡 + 𝜀𝑖,𝑡 In which “Forcedi,t” is a binary variable indicating a forced turnover for firm i in year t
and “Performancei,t-1” represents the performance of firm i in year t-1. “Controlsi,t”
represents a vector of control variables for firm i in year t, “Fixed Effectsi,t” indicates
year or industry fixed effects and, lastly, “εi,t” is a stochastic error term. My base case
regression, however, is adjusted by replacing idiosyncratic firm performance with industry-adjusted performance and adding an extra lag. Moreover, I add “Tenurei,t” and
“Outsideri,t” as control variables. “Tenurei,t” represents the tenure of the CEO of firm i in
year t and “Outsideri,t” is a binary variable indicating whether the CEO of firm i in year t
is an outsider or not. This results in the following base case regression (1):
Pr(𝐹𝑜𝑟𝑐𝑒𝑑𝑖,𝑡) = 𝛽0+ 𝛽1∗ 𝐼𝑛𝑑. 𝐴𝑑𝑗. 𝑃𝑒𝑟𝑓𝑖,𝑡−1+ 𝛽2∗ 𝐼𝑛𝑑. 𝐴𝑑𝑗. 𝑃𝑒𝑟𝑓𝑖,𝑡−2+ 𝛽3∗ 𝑇𝑒𝑛𝑢𝑟𝑒𝑖,𝑡 + 𝛽4 ∗ 𝑂𝑢𝑡𝑠𝑖𝑑𝑒𝑟𝑖,𝑡+ 𝑌𝑒𝑎𝑟 𝐹𝐸𝑡+ 𝜀𝑖,𝑡
In regression (1), the variable of interest is the industry-adjusted performance variable, of which the coefficient β1 and β2 are expected to be negative: better performance
decreases the probability of a forced turnover. I run two separate regressions (regression (1) and regression (2)) using the industry-adjusted stock return
performance measure in one and industry-adjusted return on assets in the other to compare results of the two measures. I expect β3 to be negative since longer tenure
decreases the probability of dismissal after bad performance. This is likely because of the firm’s confidence in an executive builds up over the years in his position. With the same line of reasoning concerning trust issues and knowledge of a CEO’s capabilities, β4 is expected to be positive. The regression does not include industry fixed effects, only year fixed effects. Year fixed effects are added to account for time trends and economy wide shocks. The lack of industry fixed effects is due to the fact that the performance measure is adjusted for industry returns.
I extend this base case regression with a proxy for the outside options a CEO has, concerning jobs as CEO at other firms. This regression (3) has the following form:
Pr(𝐹𝑜𝑟𝑐𝑒𝑑𝑖,𝑡) = 𝛽0+ 𝛽1∗ 𝐼𝑛𝑑. 𝐴𝑑𝑗. 𝑃𝑒𝑟𝑓𝑖,𝑡−1+ 𝛽2∗ 𝐼𝑛𝑑. 𝐴𝑑𝑗. 𝑃𝑒𝑟𝑓𝑖,𝑡−2+ 𝛽3∗ 𝑇𝑒𝑛𝑢𝑟𝑒𝑖,𝑡 + 𝛽4 ∗ 𝑂𝑢𝑡𝑠𝑖𝑑𝑒𝑟𝑖,𝑡+ 𝛽5 ∗ 𝑂𝑢𝑡. 𝑂𝑝𝑡𝑖,𝑡+ 𝑌𝑒𝑎𝑟 𝐹𝐸𝑡+ 𝜀𝑖,𝑡
18
In which “Out.Opt.i,t” represents the outside options proxy for the CEO of firm i in year t.
The fewer the amount of outside options is, the smaller the probability of a forced turnover would be, from the view of the firm. The opposite, however, is unlikely to be true in practice. Although the number of outside options from the view of the firm is larger (i.e. the supply of CEOs is larger), the firm would probably not suddenly replace the current CEO. Yet, theoretically this relation may exist. β5 is expected to be positive.
The next extension of the regression would be a binary variable that indicates whether the observation is during a crisis or not. This yields the following regression equation:
Pr(𝐹𝑜𝑟𝑐𝑒𝑑𝑖,𝑡) = 𝛽0+ 𝛽1∗ 𝐼𝑛𝑑. 𝐴𝑑𝑗. 𝑃𝑒𝑟𝑓𝑖,𝑡−1+ 𝛽2∗ 𝐼𝑛𝑑. 𝐴𝑑𝑗. 𝑃𝑒𝑟𝑓𝑖,𝑡−2+ 𝛽3∗ 𝑇𝑒𝑛𝑢𝑟𝑒𝑖,𝑡 + 𝛽4 ∗ 𝑂𝑢𝑡𝑠𝑖𝑑𝑒𝑟𝑖,𝑡+ 𝛽5 ∗ 𝐼𝑛𝑐𝑟𝑖𝑠𝑖𝑠𝑡+ 𝜀𝑖,𝑡
In which “Incrisist” indicates if the observation in year t is during a financial crisis. When
adding this variable, I leave out the year fixed effects since the Incrisis variable accounts for year-specific trends. Besides, having both in the regression results in collinearity issues. The coefficient of “Incrisist” is expected to be positive due to a non-perfect
relative performance evaluation.
This final addition results in a regression equation that includes all previously mentioned variables and an interaction term: “Out.Opt.i,t*Incrisist”:
Pr(𝐹𝑜𝑟𝑐𝑒𝑑𝑖,𝑡) = 𝛽0+ 𝛽1∗ 𝐼𝑛𝑑. 𝐴𝑑𝑗. 𝑃𝑒𝑟𝑓𝑖,𝑡−1+ 𝛽2∗ 𝐼𝑛𝑑. 𝐴𝑑𝑗. 𝑃𝑒𝑟𝑓𝑖,𝑡−2+ 𝛽3∗ 𝑇𝑒𝑛𝑢𝑟𝑒𝑖,𝑡
+ 𝛽4 ∗ 𝑂𝑢𝑡𝑠𝑖𝑑𝑒𝑟𝑖,𝑡+ 𝛽5 ∗ 𝑂𝑢𝑡. 𝑂𝑝𝑡𝑖,𝑡+ 𝛽6∗ 𝐼𝑛𝑐𝑟𝑖𝑠𝑖𝑠𝑡+ 𝛽7∗ (𝑂𝑢𝑡. 𝑂𝑝𝑡𝑖,𝑡 ∗ 𝐼𝑛𝑐𝑟𝑖𝑠𝑖𝑠𝑡) + 𝜀𝑖,𝑡
The addition of the interaction term can show if the effect of the outside options proxy exists during crisis periods. I expect β7 to be positive: during a crisis, the larger the
amount of outside options, the larger the probability of a forced turnover.
A further step in the analysis is to have two separate regressions with two separate crisis indicator variables: one for the Dotcom bubble crisis and one for the subprime financial crisis of ‘07/’08. This gives insight in whether each crisis affects the dismissal rate differently, or not at all. The two regression equations have the following forms:
19 Pr(𝐹𝑜𝑟𝑐𝑒𝑑𝑖,𝑡) = 𝛽0+ 𝛽1∗ 𝐼𝑛𝑑. 𝐴𝑑𝑗. 𝑃𝑒𝑟𝑓𝑖,𝑡−1+ 𝛽2∗ 𝐼𝑛𝑑. 𝐴𝑑𝑗. 𝑃𝑒𝑟𝑓𝑖,𝑡−2+ 𝛽3∗ 𝑇𝑒𝑛𝑢𝑟𝑒𝑖,𝑡 + 𝛽4 ∗ 𝑂𝑢𝑡𝑠𝑖𝑑𝑒𝑟𝑖,𝑡+ 𝛽5 ∗ 𝑂𝑢𝑡. 𝑂𝑝𝑡𝑖,𝑡+ 𝛽6∗ 𝐼𝑛𝑐𝑟𝑖𝑠𝑖𝑠𝐷𝑜𝑡𝑐𝑜𝑚𝑡+ 𝛽7 ∗ (𝑂𝑢𝑡. 𝑂𝑝𝑡𝑖,𝑡∗ 𝐼𝑛𝑐𝑟𝑖𝑠𝑖𝑠𝐷𝑜𝑡𝑐𝑜𝑚𝑡) + 𝜀𝑖,𝑡 Pr(𝐹𝑜𝑟𝑐𝑒𝑑𝑖,𝑡) = 𝛽0+ 𝛽1∗ 𝐼𝑛𝑑. 𝐴𝑑𝑗. 𝑃𝑒𝑟𝑓𝑖,𝑡−1+ 𝛽2∗ 𝐼𝑛𝑑. 𝐴𝑑𝑗. 𝑃𝑒𝑟𝑓𝑖,𝑡−2+ 𝛽3∗ 𝑇𝑒𝑛𝑢𝑟𝑒𝑖,𝑡 + 𝛽4 ∗ 𝑂𝑢𝑡𝑠𝑖𝑑𝑒𝑟𝑖,𝑡+ 𝛽5 ∗ 𝑂𝑢𝑡. 𝑂𝑝𝑡𝑖,𝑡+ 𝛽6∗ 𝐼𝑛𝑐𝑟𝑖𝑠𝑖𝑠07/08𝑡+ 𝛽7 ∗ (𝑂𝑢𝑡. 𝑂𝑝𝑡𝑖,𝑡∗ 𝐼𝑛𝑐𝑟𝑖𝑠𝑖𝑠07/08𝑡) + 𝜀𝑖,𝑡
In which “IncrisisDotcomt” is a binary variable that equals 1 if the observation is during
the Dotcom bubble crisis and “Incrisis07/08t” is a binary variable that equals 1 if the
observation is during the subprime financial crisis of 2007/2008.
5. Results
In this section I present the results of the logit regressions to estimate the previously described model. The results of the probit regressions are very similar and thus are not presented in this section to keep things concise. For completeness, the probit regression results are available in the appendix.
5.1 Main logit regression results
Table 2 presents the results of the various regressions to estimate the probability of a forced CEO turnover. The dependent variable in all regressions is the binary variable “Forced” which equals 1 for a forced turnover and 0 for no turnover. Regression (1) is the base case regression which is roughly identical to the second-stage logit regression of Jenter & Kanaan (2015). In this regression I use CEO Tenure and an Outsider dummy as controls and two lagged values of industry adjusted return as independent variables of interest. Year fixed effects are added to account for time trends and market-wide shocks. Apart from the second lag of the industry adjusted return, all variables are statistically significant at the five percent level at least. The second lag of industry adjusted return does not have any statistical significance. Moreover, the estimated coefficients have the expected signs. A higher lagged return decreases the probability of a forced turnover. The same goes for Tenure: the longer the CEO has been in his position the smaller the probability of being dismissed by the board. This is probably due to the trust and relationship a CEO builds with the board when exercising his job. If a CEO that has been in his position for a long time, the board and shareholders will likely allow him
20
to have a poor year in terms of firm performance without immediately firing him, compared to a CEO that achieves poor results during his first year as a CEO. One can imagine that a similar “trust issue” is applicable to an outside CEO: one that has not worked for the company before being appointed as CEO. Hence, the coefficient is
positive. For example, the probability of a forced CEO turnover in a given year for a CEO of a firm with a lagged industry adjusted return of 0.05, lagged industry adjusted ROAs of 0.04, a tenure of 5 years, who is not an outsider, is equal to: 1/[1+exp(-(-2.095-0.823*0.05-0.258*5+0.274*0))] = 3.1%.
Regression (2) is in principle the same as regression (1), but (2) utilizes the industry adjusted return on assets as the performance measure. The statistical
significance and the signs of the coefficients are almost equivalent to regression (1). The second lag of the Industry Adj. Return, however, is statistically significant at the one percent level and, surprisingly, positive. This would suggest that a higher Return on Assets two years ago would increase the probability of a forced turnover, which is not in accordance with economic theory. When estimating regression (2) without the first lag of industry adjusted ROA, the second lag is not statistically significant anymore.
Furthermore, if I estimate regression (2) with the addition of an interaction term between the first and second lags of industry adjusted ROA, the coefficient of the interaction term is positive and significant at the five percent level. This might indicate that a large swing from positive ROA in one year to negative ROA in the next year is considered undesirable and thus a reason for an increase in the probability of a CEO dismissal. Otherwise, the positive coefficient may be due to data errors. Using the same example as above, the probability of a forced turnover according to regression (2) equals: 1/[1+exp(-(-2.181-2.380*0.04+1.633*0.04-0.246*5+0.235*0))] = 3.1%.
Regressions (3) and (4) are the same as (1) and (2), extended with the outside options for CEOs proxy. This proxy variable is added to estimate the effect of the decreased supply of adequate CEOs and/or the decreased demand for CEOs on the probability of a forced turnover. The coefficient of this variable is not statistically
significant, nor would the coefficient itself been have economically large enough to make any judgments about the effect on dismissal probability.
Regressions (5) and (6) are again similar to (1) and (2), with the exceptions of not adding year fixed effects and including a dummy variable which equals 1 to indicate a financial crisis. In both regressions (5) and (6) the coefficients are positive,
21
economically large and statistically significant at the five percent level. This shows that the probability of a CEO dismissal is larger during a crisis period compared to a non-crisis period. This finding corresponds to my hypothesis. This further corresponds to Jenter & Kanaan’s (2015) suggestion that during a market downturn, the probability of CEO dismissal increases. During a market downturn the board seem to be less capable of correctly evaluating CEO performance, as other factors beyond the CEO’s control are apparently taken into account as well. In other words, the relative performance evaluation process is not functioning perfectly.
Lastly, regressions (7) and (8) are the fully specified regressions. The latter two include the outside options proxy, the crisis indicator, and an interaction term in addition to the variables from regressions (1) and (2). The interaction term
Out.Opt.*Crisis shows the effect of the number of outside options during a crisis. In both regressions, the interaction term is not statistically significant and not large enough to have a notable effect. This suggests that the number of outside options does not matter for the probability of dismissal, whether it is during a crisis or not.
5.2 Logit regression results with separate crisis indicators
Recall, however, that in graph 1 in the Data section the outside options and forced turnovers seemed to be positively related to each other during the Dotcom bubble crisis and negatively related during the financial crisis of ‘07/’08. This feature may cause the coefficient of the interaction term between outside options and crisis to be insignificant. Separately indicating the first and second crises may reveal a significant effect. Table 3 presents the results of these regressions. The single crisis indicator is replaced by two separate crisis indicators: one for the Dotcom bubble crisis of 1999-2001 and one for the later financial crisis of 2007-2008. One interaction term per regression has been added as well. The remaining independent variables are lagged industry adjusted returns, CEO tenure, an outsider dummy and the outside options proxy.
Both regression results in table 3 show that the first lag of the industry adjusted return, the CEO tenure and the outsider dummy are significant at the one percent level. The coefficients also have the expected signs, identical to the signs of the coefficients in table 2. However, neither of the crisis indicators and neither of the interaction terms are statistically significant. This result suggests that, even when treating the two financial crises differently, there appears to be no relation between the outside options proxy and
22
the probability of a forced CEO turnover during a crisis. In economic terms this means that, according to this empirical research, a higher or lower supply of CEOs does not affect the likelihood of a CEO dismissal. This result is not dependent on market or industry downturns.
23 Table 2: Main logit regression results
The eight logit regressions shown below all have Forced as the dependent binary variable which equals 1 in case of a forced CEO turnover. Two yearly lags of Industry Adjusted Return are added. This independent variable is computed as the annual stock return minus the median annual stock return of the industry the firm is in. The lags of the Industry Adjusted ROA are computed in a similar fashion. It is computed by subtracting the median ROA of the industry from the firm’s ROA. CEO tenure is the number of years a CEO has been in his position. A CEO is classified as an outsider if he has been working less than twelve months for the firm prior to becoming CEO. The outside options proxy variable follows the supply of CEOs. The binary variable “During a Financial Crisis” equals 1 if an observation is during the Dotcom Bubble crisis or the ‘07/’08 financial crisis. Out. Opt.*Crisis is an interaction term between the outside options proxy and the “During a financial crisis” dummy variable. T-statistics are reported in parentheses below the coefficients. Statistical significance at the 5% and 1% level are indicated by ** and ***, respectively.
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent: Forced Forced Forced Forced Forced Forced Forced Forced
Industry Adj. Return, t-1 -0.823*** -0.822*** -0.813*** -0.811***
(-5.516) (-5.522) (-5.553) (-5.562)
Industry Adj. Return, t-2 -0.0639 -0.0657 -0.0600 -0.0623
(-0.854) (-0.873) (-0.832) (-0.853)
Industry Adj. ROA, t-1 -2.380*** -2.351*** -2.323*** -2.277***
(-5.724) (-5.632) (-5.683) (-5.540)
Industry Adj. ROA, t-2 1.633*** 1.646*** 1.567*** 1.583***
(3.093) (3.128) (2.964) (3.018)
CEO tenure -0.258*** -0.246*** -0.257*** -0.245*** -0.263*** -0.250*** -0.262*** -0.249***
(-16.48) (-16.96) (-16.48) (-16.96) (-17.19) (-17.68) (-17.18) (-17.68)
CEO is an outsider 0.274*** 0.235** 0.267*** 0.231** 0.305*** 0.262*** 0.295*** 0.255***
(2.793) (2.480) (2.723) (2.435) (3.118) (2.758) (3.016) (2.691)
Outside options proxy -1.09e-05 -1.10e-05 -2.36e-05 -2.98e-05
(-0.821) (-0.853) (-1.044) (-1.329)
During a financial crisis 0.225** 0.224** 0.168 0.136
24
Table 2 - continued
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent: Forced Forced Forced Forced Forced Forced Forced Forced
Out. Opt.*Crisis 1.89e-05 2.88e-05
(0.694) (1.104)
Constant -2.095*** -2.181*** -2.063*** -2.149*** -2.096*** -2.140*** -2.020*** -2.044***
(-9.871) (-10.89) (-9.553) (-10.52) (-26.13) (-28.00) (-19.16) (-19.82)
Observations 10,957 11,703 10,957 11,703 10,957 11,703 10,957 11,703
25
Table 3: Logit regression results with separate crisis indicators
The four logit regressions shown below have Forced as the dependent binary variable which equals 1 in case of a forced CEO turnover. Two yearly lags of Industry Adjusted Return are added. This independent variable is computed as the annual stock return minus the median annual stock return of the industry the firm is in. The lags of the Industry Adjusted ROA are computed in a similar fashion. It is computed by subtracting the median ROA of the industry from the firm’s ROA. CEO tenure is the number of years a CEO has been in his position. A CEO is classified as an outsider if he has been working less than twelve months for the firm prior to becoming CEO. The outside options proxy variable follows the supply of CEOs. The binary variables “During Dotcom bubble Crisis” and “During '07/'08 financial crisis” indicate whether an observation is during the Dotcom Bubble crisis or the ‘07/’08 financial crisis, respectively. “Out.
Opt.*DotcomCrisis” is an interaction term between the outside options proxy and the “During Dotcom bubble crisis” dummy variable. “Out. Opt.*’07/’08 Crisis” is an
interaction term between the outside options proxy and the “During '07/'08 financial crisis” dummy variable. T-statistics are reported in parentheses below the coefficients. Statistical significance at the 5% and 1% level are indicated by ** and ***, respectively.
(1) (2) (3) (4)
Dependent: Forced Forced Forced Forced
Industry Adj. Return, t-1 -0.810*** -0.818***
(-5.562) (-5.495)
Industry Adj. Return, t-2 -0.0638 -0.0601
(-0.869) (-0.834)
Industry Adj. ROA, t-1 -2.253*** -2.303***
(-5.490) (-5.592)
Industry Adj. ROA, t-2 1.603*** 1.599***
(3.068) (3.043)
CEO Tenure -0.263*** -0.250*** -0.265*** -0.253***
(-17.12) (-17.63) (-17.36) (-17.88)
CEO is an outsider 0.299*** 0.258*** 0.303*** 0.261***
(3.057) (2.730) (3.092) (2.756)
Outside options proxy -1.77e-05 -2.66e-05 -1.24e-05 -9.93e-06
(-0.907) (-1.361) (-0.821) (-0.698)
During Dotcom bubble crisis 0.123 0.0403
(0.827) (0.294)
Out. Opt.*DotcomCrisis 1.16e-05 2.85e-05
(0.445) (1.198)
During '07/'08 financial crisis 0.150 0.248
(0.834) (1.425)
Out. Opt.*’07/’08 Crisis 1.70e-05 -3.79e-06
(0.410) (-0.0908)
Constant -1.990*** -1.999*** -2.000*** -2.054***
(-20.57) (-21.25) (-22.49) (-24.23)
Observations 10,957 11,703 10,957 11,703
26 6. Robustness Checks
6.1 Alternative performance measures
The first checks for robustness I perform have to do with using different performance measures. In previous literature, stock return and return on assets have been the main tools to measure firm performance, and thereby the CEO’s capability, but other
measures have also been mentioned. One of these is the stock return relative to the return of the market. For this purpose a market index is used and for this sample, containing merely North-American based firms, the S&P500 index is a suitable one. I compute market adjusted returns by subtracting the annual S&P500 index return from a firm’s annual stock return.
Another used performance measure is the return on assets, calculated as
EBIT/Total Assets. Using earnings before interest and taxes, or operating profit, instead of net income can filter out issues that may not be directly related to the CEO’s abilities. For example, a tax burden or financing issues are left out of the equation to assess the current CEO. This form of ROA is adjusted in the same way as the previously used ROA: I subtract the median industry ROA from the firm’s ROA. For the regressions with market adjusted returns and ROA’s constructed with EBIT, I expected no significant changes in the coefficients. Table 4 shows that all coefficients have the same sign and significance, except the dummy variable indicating a crisis in regression (5), which incorporates the market adjusted return as performance measure. The estimate of the coefficient is not statistically significant, whereas this is indeed the case when using other performance measures. This may be due to endogeneity issues regarding the correlation between the return of the market and whether it is during a crisis or not. One can expect that
whenever there is a financial crisis, the market return decreases. This could explain a loss in statistical significance for that particular regression. Besides this artifact, the main regression results still hold.
6.2 Alternative age threshold for forced turnover classification
It can be argued that my method for classifying a turnover as forced is quite arbitrary in the sense that I merely consider the CEO’s age as determinant. For the main results I choose an age threshold of 56 – i.e. all turnovers of CEOs younger than 56 years old are classified as forced. This exact threshold was chosen since Peters & Wagner (2014)
27
Table 4: Logit regressions with Market adjusted returns and ROA constructed with EBIT
The eight logit regressions shown below all have Forced as the dependent binary variable which equals 1 in case of a forced CEO
turnover. Two yearly lags of Market Adjusted Return are added. This independent variable is computed as the annual firm stock return minus the market index return. The (lags of) the Industry Adjusted ROA are computed by subtracting the median ROA of the industry from the firm’s ROA. In the regressions below, ROA is computed as Earnings Before Interest & Taxes divided by Total Assets. CEO tenure is the number of years a CEO has been in his position. A CEO is classified as an outsider if he has been working less than twelve months for the firm prior to becoming CEO. The outside options proxy variable follows the supply of CEOs. The binary variable “During a Financial Crisis” equals 1 if an observation is during the Dotcom Bubble crisis or the ‘07/’08 financial crisis. Out. Opt.*Crisis is an interaction term between the outside options proxy and the “During a financial crisis” dummy variable. T-statistics are reported in parentheses below the coefficients. Statistical significance at the 5% and 1% level are indicated by ** and ***, respectively.
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent: Forced Forced Forced Forced Forced Forced Forced Forced
Market Adj. Return, t-1 -0.797*** -0.793*** -0.772*** -0.767***
(-5.540) (-5.518) (-5.790) (-5.773)
Market Adj. Return, t-2 -0.0485 -0.0491 -0.0345 -0.0350
(-0.726) (-0.733) (-0.595) (-0.598)
CEO tenure -0.259*** -0.246*** -0.258*** -0.246*** -0.263*** -0.250*** -0.262*** -0.250***
(-16.50) (-16.96) (-16.50) (-16.96) (-17.19) (-17.71) (-17.18) (-17.71)
CEO is an outsider 0.273*** 0.257*** 0.269*** 0.252*** 0.308*** 0.288*** 0.300*** 0.280***
(2.788) (2.719) (2.739) (2.671) (3.137) (3.041) (3.056) (2.964)
Industry Adj. ROA(EBIT), t-1 -3.098*** -3.036*** -3.033*** -2.945***
(-4.039) (-3.956) (-3.999) (-3.891)
Industry Adj. ROA(EBIT), t-2 2.741*** 2.741*** 2.669*** 2.664***
(3.309) (3.324) (3.230) (3.255)
Outside options proxy -7.49e-06 -1.18e-05 -2.00e-05 -3.12e-05
(-0.574) (-0.913) (-0.893) (-1.384)
During a financial crisis 0.127 0.211** 0.0727 0.122
(1.253) (2.247) (0.553) (0.977)
Out. Opt.*Crisis 1.81e-05 2.92e-05
28
Table 4 - continued
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent: Forced Forced Forced Forced Forced Forced Forced Forced
Constant -2.248*** -2.173*** -2.226*** -2.138*** -2.096*** -2.124*** -2.032*** -2.025***
(-10.43) (-10.68) (-10.14) (-10.33) (-26.17) (-27.91) (-19.33) (-19.76)
Observations 10,957 11,640 10,957 11,640 10,957 11,640 10,957 11,640
29
found that this closely matches the dismissal rate of the press-based procedure of identifying forced turnovers. Yet, for robustness purposes, I estimate and present in table 5 the forced turnover probability, and the coefficients and t-statistics of the outside options proxy and financial crisis dummy variables for a range of age thresholds from 51 to 60. The continually increasing retirement age can be considered as a reason for
varying the age thresholds. Nowadays, people keep working until a higher age compared to 20 years ago. Picking one certain age and not considering further options would ignore this effect, if there is one.
All regressions in table 5 include two lags of the industry adjusted return, CEO tenure, an outsider dummy variable, a crisis dummy variable, the outside options proxy, and an interaction term between the outside options proxy and the crisis dummy
variable. For calculation of the probability of dismissal, I again use the aforementioned example values: the lagged industry adjusted return is 0.05, CEO tenure is 5 years and the CEO is not an outsider. The higher the age threshold is set, the larger the amount of turnovers is classified as forced and the higher the probability of a CEO dismissal. Hence it is not surprising that, overall, negative coefficients become more negative and positive coefficients become more positive. Varying the threshold hardly affects the statistical significance of the lags of industry adjusted return, CEO tenure and outsider dummy variables. The coefficient of Out. Opt.*Crisis is in none of the regressions statistically significant at the 5% level neither is it ever economically large enough to have a notable effect. Only at the extreme levels of the threshold is the significance affected
6.3 Outside options proxy based on total assets
The main outside options proxy is computed as the median of the total compensation of a given CEO’s or firm’s peer group of firms in the same industry. The firms for this peer group are selected on the basis of the sales volume. An alternative to this is to select the firms based on the total assets of a firm. A firm would be included in the peer group if it lies within the range of 50%-200% of total assets of the firm for which a peer group is constructed. A reason to compare results of the two different proxies is that CEOs and firms possibly consider firm size on the basis of total assets as well. Table 6 shows the same regressions as the main regressions which incorporate the outside options proxy in table 2, but with the outside options proxy now based on the total assets of a firm. There is no considerable change in the regressions results: all coefficients have the same
30
sign and are statistically significant at the same levels as the main logit regression results.
Table 5: Partial logit regression results across different age thresholds
The ten logit regressions below have the variable Forced as the dependent binary variable which equals 1 in case of a forced turnover and is varied by choosing different age thresholds. The industry adjusted return is computed as the annual stock return minus the median annual stock return of the industry the firm is in. The binary variable “During a Financial Crisis” equals 1 if an observation is during the Dotcom Bubble crisis or the ‘07/’08 financial crisis. The outside options proxy variable follows the supply of CEOs. Out. Opt.*Crisis is an interaction term between the outside options proxy and the “During a financial crisis” dummy variable. T-statistics are reported in parentheses below the coefficients .Statistical significance at the 5% and 1% level are indicated by ** and ***, respectively.
Age Threshold 51 52 53 54 55
Dismissal probability 2.4% 2.7% 2.7% 2.9% 3.1%
Industry Adj. Return, t-1 -0.791*** -0.779*** -0.802*** -0.847*** -0.851***
(-3.857) (-4.234) (-4.621) (-5.104) (-5.380)
During a financial crisis 0.125 0.0216 0.0508 0.104 0.124
(0.649) (0.127) (0.317) (0.687) (0.877)
Outside options proxy -0.000116*** -8.73e-05*** -4.72e-05* -4.29e-05 -3.45e-05
(-2.936) (-2.685) (-1.649) (-1.630) (-1.409)
Out. Opt.*Crisis 5.79e-05 6.98e-05* 3.10e-05 2.52e-05 2.20e-05
(1.118) (1.707) (0.847) (0.737) (0.714)
Age Threshold 56 57 58 59 60
Dismissal probability 3.3% 3.6% 3.8% 4.1% 4.7%
Industry Adj. Return, t-1 -0.811*** -0.861*** -0.904*** -0.916*** -0.933***
(-5.562) (-6.125) (-6.536) (-6.841) (-7.199)
During a financial crisis 0.168 0.193 0.247** 0.245** 0.207*
(1.280) (1.547) (2.037) (2.107) (1.826)
Outside options proxy -2.36e-05 -9.97e-06 -9.22e-06 -3.23e-06 -8.94e-06
(-1.044) (-0.464) (-0.441) (-0.163) (-0.460)
Out. Opt.*Crisis 1.89e-05 9.09e-06 7.85e-06 6.07e-06 1.23e-05
31
Table 6: Logit regression results with alternative outside options proxy
The eight logit regressions shown below all have Forced as the dependent binary variable which equals 1 in case of a forced CEO turnover. Two yearly lags of Industry Adjusted Return are added. This independent variable is computed as the annual stock return minus the median annual stock return of the industry the firm is in. The lags of the Industry Adjusted ROA are computed in a similar fashion. It is computed by subtracting the median ROA of the industry from the firm’s ROA. CEO tenure is the number of years a CEO has been in his position. A CEO is classified as an outsider if he has been working less than twelve months for the firm prior to becoming CEO. The outside options proxy (total assets) variable follows the supply of CEOs. The binary variable “During a Financial Crisis” equals 1 if an observation is during the Dotcom Bubble crisis or the ‘07/’08 financial crisis. Out. Opt.*Crisis is an interaction term
between the outside options proxy and the “During a financial crisis” dummy variable. T-statistics are reported in parentheses below the coefficients. Statistical significance at the 5% and 1% level are indicated by ** and ***, respectively.
(1) (2) (3) (4)
Dependent: Forced Forced Forced Forced
Industry Adj. Return, t-1 -0.824*** -0.814***
(-5.514) (-5.565)
Industry Adj. Return, t-2 -0.0629 -0.0608
(-0.842) (-0.836)
CEO tenure -0.258*** -0.246*** -0.263*** -0.250***
(-16.46) (-16.96) (-17.13) (-17.67)
CEO is an outsider 0.278*** 0.238** 0.306*** 0.262***
(2.820) (2.507) (3.114) (2.757)
Outside options proxy (total assets) 6.87e-06 8.12e-06 -7.84e-06 -1.22e-05
(0.520) (0.655) (-0.388) (-0.619)
Industry Adj. ROA, t-1 -2.400*** -2.327***
(-5.750) (-5.673)
Industry Adj. ROA, t-2 1.626*** 1.574***
(3.068) (2.985)
During a financial crisis 0.152 0.125
(1.151) (1.004)
Out.Opt.(TA)*Incrisis 2.14e-05 2.95e-05
(0.832) (1.234)
Constant -2.115*** -2.205*** -2.072*** -2.102***
(-9.791) (-10.81) (-20.40) (-21.36)
Observations 10,957 11,703 10,957 11,703
32
6.4 Adding industry fixed effects
By adding industry fixed effects to the main regression, I can control for potential industry-specific turnover characteristics. The results, however, are hardly affected. Statistical significance and coefficient signs have remained the same, thus I do not report the results in this section. The regression results are available in the appendix.
7. Conclusion
In this thesis I study the effect of the supply of CEOs on the probability of a forced turnover during a financial crisis. I use turnover and balance sheet data from North-American companies for the time period of 1996-2014 which includes two major
financial crises: the crisis after the bursting of the Dotcom bubble and the financial crisis of ‘07/’08 or Subprime crisis. The probability of a forced turnover is estimated using logit and probit regressions. The dependent binary variable equals 1 in case of a forced turnover (and 0 if no turnover took place) and the independent variables include, among others, a relative performance measure, a proxy for the supply of CEOs and a binary variable indicating a financial crisis. A turnover is classified as forced on the basis of an age threshold, similar to the procedure Peters & Wagner (2014) apply as an alternative to the press-based classification of Huson, et al.(2001) and Parrino (1997), among others. Industry adjusted stock return functions as the main performance measure. The outside options proxy following the supply of CEOs is constructed in the same way as Bizjak, et al.(2008) and Brookman & Thistle (2013). The proxy is based on the median compensation of CEOs within a peer group of firms. For the financial crisis binary variable, I simply declare an observation to be during a crisis if it is in one of the following years: 1999, 2000, 2001, 2007 or 2008.
The results support those of Jenter & Kanaan (2015) concerning the relative performance evaluation: a year lagged value of the industry adjusted return is negatively related to the probability of a forced turnover. Furthermore, these prove to be robust through several robustness checks. When distinguishing crisis years from non-crisis years, my results also strengthen the minimal evidence of Jenter & Kanaan (2015) of a market downturn affecting the dismissal rate of a CEO, confirming that relative
performance evaluation of CEOs is not perfect. The positive relation between the occurrence of a financial crisis and the probability of a forced turnover is statistically significant at the 5% level. Adding the outside options proxy and an interaction term
33
between the latter variable and the crisis binary variable, however, mitigates the statistical significance of the crisis variable. Moreover, the outside options proxy itself and the interaction term do not have any statistical significance, despite various
robustness tests. This leads to the conclusion that, according to this research, there is no significant relation between the supply of CEOs and the probability of a CEO dismissal during a financial crisis, in contrast with economic theory.
My result concerning the supply of CEOs likewise does not correspond to prior publications on the subject of CEO turnovers and the executive labor market (Brookman & Thistle (2013), Eisfeldt & Kuhnen (2013) and Liu (2014)). Both the supply side as well as the demand side of the executive labor market have been studied in which significant effects on the turnover process have been established. These effects are both from the point of view of the CEO and from the point of view of the firm.
The lack of corresponding results of this thesis might be due to multiple
limitations or issues. Firstly, I chose for a relatively simple age-based method to classify turnovers as forced. Although this has been used before in academic publications, it has always been implemented in conjunction with an additional classification method – in most of the papers this is the well-known press-based classification by Parrino (1997). For me, however, such a method was out of scope for an MSc thesis. Improving the forced turnover classification method may provide more robust results. Secondly, I use annual data on turnovers, stock returns and balance sheets. To assess performance, smaller time periods would perhaps result in more precise estimates. To this extent, quarterly or monthly intervals would enhance results because performance can be assigned to the CEO in question more precisely. Thirdly, the outside options proxy I use might not be the optimal proxy for the supply of CEOs. Studying the turnover literature, however, I come across very few alternatives that seem appropriate, unfortunately. The proxy is based on the total compensation of CEOs. Executive compensation, however, keeps increasing over the years – corrected for inflation – and this affects the reliability when handling observations that are several years apart. Lastly, I use a rather arbitrary method for indicating the periods in which there is a financial crisis. A more careful approach could make the results (more) robust. On the other hand, as I deal with data on a yearly basis, not much fine tuning was possible in this respect. In a follow-up research with monthly or quarterly data some measure of credit risk such as the TED-spread3
