1
Is the Hoberg and Phillips peer return benchmark is a more precise predictor
for forced CEO turnover than industry peer return and market peer return?
Institution: University of Amsterdam
MSc Finance Student: Mark Goes
Track: Corporate Finance Date: 1 July 2017
Type of document: Master Thesis Thesis supervisor: Florian Peters
Abstract
This paper shows that the Hoberg and Phillips peer return benchmark is a more precise indicator for CEO performance than industry and market return benchmarks. The forced turnover database from Peters and Wagner and the Hoberg and Phillips data library to create peer groups are used to create a sample of 938 forced CEO turnover observations and 35,128 no forced CEO turnover observations. The results show that turnover-performance sensitivity is much greater in the domain of negative turnover-performance. Forced CEO turnover is more sensitive to idiosyncratic performance during the non-crisis period than during the crisis period. Outperforming the benchmark has a negative effect on the probability of a forced CEO turnover.
2 Statement of Originality
This document is written by Student Mark Goes 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.
3 Inhoud
1. Introduction ... 4
2. Related Literature ... 6
2.1 Trends ... 6
2.2 Main existing theories ... 7
2.3 Advantages CEO turnover ... 10
2.4 CEO pay ... 11
2.5 Relation to previous research and contribution ... 12
3. Methodology ... 13
4. Data ... 16
5. Results ... 20
5.1 Explanatory power of different benchmarks ... 20
5.2 the effect of underperformance ... 21
5.3. the effect of the peer group size... 22
5.4 similarity score ... 22
5.5 Forced turnover during crisis period ... 23
5.6 Forced turnover during under- and outperformance ... 24
5.7 robustness ... 24
6. Conclusion... 25
References ... 28
4 1. Introduction
Whether or not to fire the CEO after bad firm performance is one of the most important decisions made by the corporate boards. Previous research shows that from 1971 until 2005 the number of CEO turnovers increased. This is an interesting development and raises the question how CEO turnover is determined. according to theory, CEOs should be fired or rewarded based on their performance. However, literature shows us that this is not always the case (Kaplan and Minton (2012); Jenter and Kanaan (2015). This indicates that the board of directors fails to exclude factors beyond the control of the CEO when estimating firm performance. The papers of Holmstrom (1979) and Diamond and Verrecchia (1982) implicate that it would be optimal for boards to filter all observable exogenous shocks from firm performance updating their assessments of the quality of the CEO . Consequently, Hölmstrom and Kaplan (2003) expressed the importance of a more effective benchmark for CEO compensation and performance in their paper.
The question that is attempted to answer in this thesis is if a Hoberg and Phillips peer return benchmark is a more precise predictor for forced CEO turnover than industry peer return and market peer return. This research question might give a partial solution to the inability of the board of directors to filter out exogenous shocks. Previous research already investigated the relation between forced CEO turnover and the (relative) market and industry return (Jenter and Kanaan (2015); Kaplan and Minton (2012)). However, research about the influence of a firm-specific peer group performance, using the Hoberg and Philips data library, on forced CEO turnover has not been done yet.
Previous literature found an increase in CEO turnover during the 1970-2015 period (Huson, Parrino and Starks (2001); Jensen et al. (2004); Kaplan and Minton (2012); Murphy (1999); Murphy and Zabojnik (2007)). Furthermore, different possible explanations for CEO turnover are tested in previous literature. The first who did research into the relationship between management turnover and firm performance where Coughlan and Schmidt (1985). They found a negative relationship between management turnover and firm performance. The results in this thesis confirm this result. Barro and Barro (1990), Gibbons and Murphy (1990) and Warner, watts and Wruck (1988) found that CEOs are more likely to be forced out of their job if their performance is poor relatively to the industry
5
average. This suggest that industry performance is filtered out when looking at the CEO performance. This finding is partly contradicted in this thesis. For all benchmarks tested, underperformance of the relative industry return has the lowest effect on the forced CEO turnover probability. Morck, Shleifer and Vishny (1989) found that internal turnover is related to industry-adjusted performance while external turnover is related to industry performance. Jenter and Kanaan (2015) found that when the performance of the peer group is good, the chance that an underperforming CEO is fired is significantly lower than when the performance of the peer group is bad. This indicates that the boards do not behave optimally and misattribute exogenous performance components to the CEO. In this thesis is confirmed that board do not behave optimally and fail to filter out exogenous performance components to the CEO. Bizjak, Lemmon and Nguyen (2011) found that firms, holding all else equal, are more likely to choose larger and better performing peers compared to potential peer firms that are smaller and have worse relative performance. Farrel and Whidbee (2003) found that the chance of a turnover is higher if the CEO deviates a lot from the expected performance. Jenter and Lewellen (2010) found that a large fraction of CEO turnover is performance induced which is confirmed in this thesis. Additionally Eisfeldt and Kuhnen (2013) show that the overall industry performance is critical to the decision that leads to forced CEO turnover. Furthermore, previous literature shows that although the board of directors has access to inside information, the board still relies on the stock market for additional information to make turnover decisions (Huson, Parrino, and Starks, 2001; Bushman, Dai, and Wang, 2010; Kaplan and Minton, 2012; Jenter and Kanaan, 2015). The first to make the link between the ability of the CEO and the CEO turnover where Hermalin and Weisbach (1998). In extension from this research Dikolli, Mayew and Nanda (2014) found that the likelihood of firing CEO due to performance based arguments declines with tenure. This indicates that learning about the skill of the CEO is indeed an argument for CEO turnover. Furthermore, Jenter and Kanaan (2006) argued that skill might be an important factor for CEO turnover. They argued that the reason that CEO turnover actually was higher when industry performance was low, is because firm performance in bad times may be more revealing about CEO skill than performance in good times. The results in this thesis indicate that during the crisis period for both industry and peer group returns are better filtered out when determining CEO performance than during the non-crisis period. This suggests that the increase in turnovers during bad industry performance is not causes by
6
exogenous shocks. Consequently, it indicates that it is possible that the increased revealing of CEO skill during bad times is the cause of increasing CEO turnover.
To test the hypothesis a logit regression on forced turnover is executed. To test for the effect of underperformance an interaction variable is included between the performance of the firm relative to the benchmark and dummy variable for underperformance. The results show that turnover-performance sensitivity is much greater in the domain of negative performance. Furthermore, the sample is tested for the crisis period (2007-2009) and non-crisis period (1996-2006 and 2010-2014) since performance during a crisis period does contain more information about the skill of the CEO. The results suggest that forced CEO turnover is more sensitive to idiosyncratic performance during the non-crisis period than during the crisis period for the relative market, industry and peer group benchmarks. Additionally, the effect of a change in peer group returns are tested for the explanatory power of the Hoberg and Phillips (relative) peer group returns. The results show that the size of the peer group does not affect the results. Moreover, the effect of the Hoberg and Phillips similarity score on the explanatory power of the Hoberg and Phillips (relative) peer group returns is tested. The results show that when the Hoberg and Phillips similarity score is higher the explanatory power of the Hoberg and Phillips peer group benchmark increases significantly. Finally the effect of outperformance of the relative benchmark is tested. The results show that outperforming the benchmark has a negative effect on the probability of a forced CEO turnover. This contradicts the results of Jenter and Kanaan (2015) who found that outperforming the market only has a small effect on the probability of a forced CEO turnover.
Section 2 of this thesis describes the existing literature about the subject and provides the contribution of this thesis relative to the other literature. Section 3 provides the hypothesis and the methodology of the thesis. Section 4 describes the data sources, sample selection, variable definitions and descriptive statistics. Section 5 presents the mai n empirical results and the robustness checks. Section 6 summarizes an concludes.
2. Related Literature
7
Standard agency theory states that there are benefits to evaluating agents on the basis of their relative performance when agents are affected by common shocks (Holmstrom (1979, 1982); Diamond and Verrecchia (1982)). The intuition behind this is that it is not optimal to punish or reward the agent for (bad) luck.
The first who did research into the relationship between management turnover and firm performance where Coughlan and Schmidt (1985). They found a negative relationship between management turnover and firm performance. This is in line with the findings of Weisbach (1988) and Zimmerman (1993) who found empirical evidence that suggests that there is an increased likelihood of CEO turnover after a period of poor firm performance. The reasoning behind these papers is that firm performance reveals information about the ability of the CEO to create shareholders value. When the firm performance is (relatively) poor, the CEO is replaced because the firm’s owners infer that he is unable to effectively formulate and implement strategies and policies that enhances firm value.
Since the first research of Coughlan and Schmidt (1985) there have been certain trends related to CEO turnover. Huson, Parrino and Starks (2001) found that during the 1971-1994 period the CEO turnover increased. Additionally, they found that board characteristics influence the likelihood of CEO turnover. This is in line with the findings of Murphy and Zabojnik (2007) and Jensen et al. (2004) who also found an increase of CEO turnover during this period. Furthermore, Kaplan and Minton (2012) found an increase in the probability of a CEO turnover in the period 1992-2005. They attribute this finding to the increased sensitivity of boards to CEO performance. This increase is relative to the findings of Murphy (1999) in the 1970-1995 period. The results found by Kaplan and Minton (2012) suggest that during the period in which boards have been criticized, boards have become increasingly sensitive to firm stock performance. Moreover, Peters and Wagner (2014) suggest that the increase in turnovers during the mentioned periods has led to a significant increase in CEO pay, as executives face a higher risk of losing their jobs.
2.2 Main existing theories
The papers of Hölmstrom (1979) and Diamond and Verrecchia (1982) implicate that it would be optimal for boards to filter out all observable exogenous shocks from firm performance when updating their assessments of the quality of the CEO. There is extensive
8
literature that supports this reasoning. Barro and Barro (1990), Gibbons and Murphy (1990) and Warner, Watts and Wruck (1988) found that CEOs are indeed more likely to be forced out of their job if performance is poor relatively to the industry average. This suggest that when CEO performance is determined, industry performance is filtered out.
This is in line with the findings of Morck, Shleifer, and Vishny (1989). They examine turnovers of entire top management teams during time period 1980-1985 and found them to be equally likely in troubled and healthy industries. This suggests that when dismissal decisions are made, industry performance is filtered out. Furthermore, Morck, Shleifer and Vishny (1989) found that internal turnover is related to industry-adjusted performance while external turnover is related to industry performance. These results are consistent with the findings of Murphy (1999) and Jensen et al. (2004) who found a modest relationship between internal (board initiated) turnover and firm stock performance. This can be interpreted as an indication that boards respond well to poor performance relative to the industry, but do not respond well to poor industry performance.
The findings of Jenter and Kanaan (2015) contradict these results. They found that when the performance of the peer group is good, the chance that an underperforming CEO is fired is significantly lower than when the performance of the peer group is bad. This indicates that the boards do not behave optimally and misattribute exogenous performance components to the CEO. A possible explanation for this phenomenon is the inability of boards to set firm specific benchmarks. These results suggest that the assumption that boards optimally filter out observable exogenous shocks, which is used in much of the existing literature, is not that reliable for explaining real-world CEO dismissals.
This is in line with the findings of Bizjak, Lemmon and Nguyen (2011) who found that firms, holding all else equal, are more likely to choose larger and better performing peers compared to potential peer firms that are smaller and have worse relative performance. They also found that this effect is larger for relative smaller firms. This might be caused by the ability of smaller firms to opportunistically select peers . Smaller firms have a greater number of potential peer firms that are larger relative to the larger firms and are therefore able to choose peer more opportunisticaly. Larger firms (S&P 500 firms) are more limited in their opportunism. Consequently, there might be a difference in how peer groups are set between these groups. These results are consistent with the findings of Puffer and Weintrop
9
(1991) and Farrell and Whidbee (2003) who found an increased likelihood of CEO turnover when realized earnings fall short of analysts’ expectations.
Misallocating exogenous performance components due to not optimally setting peer groups is not the only reason for CEO turnover. Farrel and Whidbee (2003) found that the chance of a turnover is higher if the CEO deviates a lot from the expected performance. They also found an increased chance of CEO turnover if expected long term performance is bad. Furthermore, they found that CEOs are often blamed for poor firm performance even when their decisions are similar to decisions made by the CEOs of comparable firms. This is in line with the findings of Jenter and Lewellen (2010) who found that a large fraction of CEO turnover is performance induced. Additionally, they find that the frequency of performance induced turnovers is roughly twice as high as the number of forced turnover in studies done prior to their paper. In addition to these findings, Eisfeldt and Kuhnen (2013) found that the overall industry performance is critical to the decision that leads to forced CEO turnover. However, the probability of a turnover when firm performance is mediocre in comparison with the competitors is higher.
Fishman, Khurana and Rhodes-Kropf (2014) found that increased pressure from shareholders may compel boards to act against CEOs when stock prices are down. They found that there is no difference in pressure when the CEO is the cause of this bad firm performance or when the CEO is not the cause of this bad firm performance. This is in line with earlier literature that found that despite the fact that the board of directors have access to inside information, the board of directors relies on the stock market for additional information to make turnover decisions (Huson, Parrino, and Starks, 2001; Bushman, Dai, and Wang, 2010; Kaplan and Minton, 2012; Jenter and Kanaan, 2015).
The first to make the link between the ability of the CEO and the CEO turnover where Hermalin and Weisbach (1998). They argued that owners’ beliefs about the CEO ability are revised over time based on periodically observing firm performance. Consequently, their beliefs of CEO ability became more precise during the employment relationship. They theoretically show how this increased precision reduces the importance of firm performance in affecting CEO dismissal decisions. They also argue that this reduces the demand of the owners for monitoring the CEO. In continuation, Dikolli, Mayew and Nanda (2014) found that the likelihood of firing CEO due to performance based arguments declines with tenure. This indicates that learning about the skill of the CEO is an argument
10
for CEO turnover. Brav et al. (2008) found that shareholder activism increases CEO turnover. They argue that increased shareholder activism increases boards control on CEO which accelerates the learning about the skill of the CEO. Furthermore, Laux (2008) found that increased board independence leads to increased CEO turnover. The intuition behind this is that when the board is independent, they monitor the CEO more closely and therefore learn about the skill of the CEO sooner. Jenter and Kanaan (2006) also argue that skill is an important factor for CEO turnover. They argue that increased CEO turnover during bad industry performance is due to the difference in explanatory power of the skill of the CEO between good and bad times. They state that CEO performance in bad times is more revealing about CEO skill than CEO performance in good times.
Some literature argues that the trend of increased turnover can be explained by a shift in managerial skills needed as a CEO. Frydman (2005) and Murphy and Zabojnik (2007) used the hypothesis that general managerial skills have increased in importance relative to specific skills. This makes CEOs more interchangeable since the costs (loss of firm-specific skills) of turnover decreased.
Previous research excluded some possible explanations for CEO turnover. Huson, Parrino and Starks (2001) found that while CEO turnover is negatively related to accounting performance and industry adjusted returns, the relations did not change significantly over time. Consequently, this cannot be the cause of the increased CEO turnover during their investigated time period. Bhagat, Bolton and Subramanian (2010) show in their research that the education of the CEO does not affect the decision by a firm to replace its current CEO. However, education does have an effect on the selection of the replacement of the CEO. Furthermore, they found that education does not have a significant effect on the long term performance of the firm.
2.3 Advantages CEO turnover
There are also some advantages that are related to CEO turnover. As mentioned before, firing a CEO that does not have the right skill set or is not capable enough should have an effect on firm performance. Several papers have proposed that the relative paucity of forced CEO turnovers after bad performance is due to CEO entrenchment and weak corporate governance (Weisbach (1988), Hermalin and Weisbach (1998), Taylor (2010)).
11
These papers document a significant price increase as a reaction to CEO turnover when a large part of the board consists of independent directors.
In line with these findings Denis and Denis (1995) find that stock price increased by 2.25% on the date of CEO dismissal announcements for companies underperforming in the prior three-year period. In addition, Huson, Parrino and Starks (2001) report that stock prices react positively to CEO turnover announcements during the 1971-1994 period. Demerjian, Lev and McVay (2012) measure the ability of a manager by measuring the managers’ efficiency in generating revenues. Based on this measure they find that stock price is negatively affected when a low ability CEO leaves. Furthermore, they find that the stock price is positively affected when a high ability CEO leaves. This implicates that CEO turnovers are indeed effective and might indicate that other existing measurements for ‘correct’ turnover are not sufficient.
2.4 CEO pay
There are also some trends in CEO pay that are worth mentioning. The reason for this is that the people who are responsible for setting the CEO pay are also responsible for CEO turnover in most cases. Therefore, it is relevant to see how the board of directors sets peer groups in CEO pay and what kind of trends can been seen over the last few decades.
Bertrand and Mullainathan (2001) find that CEO’s are paid for luck. However, CEO’s from poorly governed firms are rewarded more for luck than CEO’s from better governed firms. This is in line with previous findings that firms are unable to filter out exogenous shocks from firm performance (Jenter and Kanaan (2015)). Furthermore, Bizjak, Lemmon and Nguyen (2011) found that the level of CEO pay relative to that of the peer group median firm has a significant effect on the subsequent changes in pay. Consequently, their finding that peer groups are used to in a manner that biases compensation upwards might explain the increase in CEO pay over the past few decades. This is in line with the findings of Faulkender and Yang (2010) who also found that there is some evidence that peer groups are chosen in an way that inflates CEO pay. The results of Albuquerque, De Franco and Verdi (2009) and Cadman, Carter and Semida (2009) contradict these findings. They argue that peer groups are chosen largely based on labor market characteristics. They argue that there is only limited evidence that firms choose peer firms strategically in order to inflate CEO pay.
12 2.5 Relation to previous research and contribution
Previous literature has shown that the board of directors use peer groups to determine how the CEO is performing (Jenter and Kanaan(2015)) and how the CEO should be rewarded (Bizjak, Lemmon and Nguyen (2011)). These papers use a form of benchmarking that is related to the market or industry. This thesis adds to the previous literature by using a different benchmark than that is used in previous literature. This different benchmark is created by using the Hoberg and Phillips (2016) data library. This data library contains a similarity score that makes it is possible to create a peer group of the 10 most closely related firms relative to the firm of interest. Therefore, this peer group should contain more information of the firm of interest than the market and industry peer groups. It seems reasonable to assume that this benchmark is a better tool for filtering out observable exogenous shocks than using the market or the industry as a benchmark. This can be explained by the reasoning that the firms in the peer group should be a better fit for the firm than the industry or the market. The intuition behind this is that hand collected firms that are chosen by the similarity of the firms. Therefore, it should contain more information of the firm of interest than an average of all the firms in the industry. Consequently, these hand collected peer groups should be more able to capture the difference between exogenous shocks and idiosyncratic risk.
Additionally, creating a peer group by using the Hoberg and Phillips similarity score is also applicable to research related to CEO pay. Bizjak et al. (2011) found that firms choose firms in their peer groups that biases compensation upwards . They do so by looking how the pay of the CEO is related to the peers and industry. It might be valuable to use the Hoberg and Phillips database to see if the rise of CEO pay that is documented over the past few years can be explained by the CEO pay of the 10 most closely related peers. Finally, previous literature has shown that peer groups are an important measure for CEO’s performance. Consequently, adding a tool to look at the CEO’s performance that can better estimate the CEO’s performance is valuable. Furthermore, using the Hoberg and Phillips similarity score also makes it possible to see if firms choose their peers rationally. In an ideal world the board of directors choose peers based on the similarity of the peer firm and the firm of interest. Bizjak et al. (2011) found that this is not always the case.
13 3. Methodology
The hypothesis that is going to be tested in this thesis is: Is the Hoberg and Phillips peer return benchmark a more precise predictor for forced CEO turnover than industry peer return and market peer return? This will be tested with a regression that tests for weak-form relative perweak-formance evaluation. Weak-weak-form relative perweak-formance evaluation states that the likelihood of CEO dismissals should be negatively related to firm performance and positively related to the performance of the reference group. This hypothesis does not predict complete filtering of peer performance but assumes that only some filtering is done by corporate boards. This will be tested with the following logit regression:
𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 (𝐹𝑜𝑟𝑐𝑒𝑑 𝐶𝐸𝑂 𝑡𝑢𝑟𝑛𝑜𝑣𝑒𝑟𝑖,𝑡) = 𝛽0 + 𝛽1 (𝑟𝑒𝑡𝑖𝑡− 𝑟𝑒𝑡𝑗𝑡) + 𝛽2𝑟𝑒𝑡𝑗𝑡+ 𝜀𝑖𝑡 (1)
Where 𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦(𝐹𝑜𝑟𝑐𝑒𝑑 𝐶𝐸𝑂 𝑡𝑢𝑟𝑛𝑜𝑣𝑒𝑟𝑖,𝑡) is the probability that there has been a
forced turnover for firm i at time t. 𝛽0 is a constant and 𝛽1 (𝑟𝑒𝑡𝑖𝑡− 𝑟𝑒𝑡𝑗𝑡) is a relative performance measure. In this performance measure 𝑟𝑒𝑡𝑖𝑡 is the realized yearly return of the firm and 𝑟𝑒𝑡𝑗𝑡 is the realized yearly return of the relevant benchmark. If (𝑟𝑒𝑡𝑖𝑡− 𝑟𝑒𝑡𝑗𝑡) > 0
, then the firm outperformed the benchmark. The variable 𝑟𝑒𝑡𝑗𝑡 has different possible
characteristics. To test the difference in explanatory power of different benchmarks, the realized yearly return of the firm (𝑟𝑒𝑡𝑖𝑡) will be tested against the realized yearly return of
the market, industry or the peer group return that is created with the Hoberg and Phillips database. The 𝛽1 (𝑟𝑒𝑡𝑖𝑡− 𝑟𝑒𝑡𝑗𝑡) performance measure shows us how strong the
outperformance of each benchmark is in predicting a forced turnover and should in theory be the weakest for market return, more precise for the industry return and the strongest for the self-generated peer group. This can be explained by the difference in ability of these benchmarks to filter out market shocks that affect the firm. If this coefficient is insignificant this indicates that boards do not look at peer performance when they evaluate CEOs. If this coefficient is high this indicates that firms look at peer performance to evaluate CEOs. The stronger the coefficient is, the more exogenous shocks are filtered out when making a CEO turnover decision. The intuition behind this is that when a negative exogenous shocks happens, it happens to the peer group as well as to the firm of interest. Consequently, if the
14
exogenous shock influences the firm in a negative way. The exogenous shock should roughly affect the peer group firms in the same way. Therefore, the difference between firm performance and peer group performance is due to idiosyncratic factors. Making this a good estimator for the effect of idiosyncratic performance on CEO turnover.
Additionally, the 𝛽2𝑟𝑒𝑡𝑗𝑡 coefficient is a control variable that explains how well the
benchmark return on itself predicts turnover. According to theory this should be zero. The reason for this is that when making a turnover decision, the board should look at relative performance. Consequently, they should not be influenced by factors that are outside of the control of the CEO. The prediction that peer performance is completely filtered from the evaluation of the CEO is called the strong-form relative performance evaluation hypothesis (Alberquerque (2009)). However, past research (Jenter and Kanaan (2015)) has shown that this hypothesis does not always hold with the real world data. Jenter and Kanaan (2015) and Kaplan and Minton (2012) found that CEO’s are significantly more likely to be dismissed after bad industry performance and to a lesser extent to bad market performance. Consequently, for the variable 𝛽2𝑟𝑒𝑡𝑗𝑡 it is expected that it has a negative effect on forced
turnovers. Furthermore, it is expected that this effect is larger for the industry benchmark than for the market benchmark. Additionally, for the Hoberg and Phillips benchmark it is expected that the 𝛽2𝑟𝑒𝑡𝑗𝑡 variable has a negative effect on forced turnovers which is larger
than the effect that the industry and the market have on forced turnovers. The intuition behind this is that the Hoberg and Phillips peer groups should be a more precise predictor of forced turnovers than the industry and market. Conclusively, 𝜀𝑖𝑡 is the error variable, which
represent the random components in the regression.
The realized yearly returns are calculated monthly and in the following way:
𝑅𝑒𝑡𝑢𝑟𝑛𝑖𝑡 = (1 + 𝑟𝑒𝑡𝑖𝑡 −1) ∗ (1 + 𝑟𝑒𝑡𝑖𝑡 −2) ∗ … … … ∗ (1 + 𝑟𝑒𝑡𝑖𝑡 −11) ∗ (1 + 𝑟𝑒𝑡𝑖𝑡 −12) − 1 (2)
where 𝑟𝑒𝑡𝑖𝑡−1 are the returns of the firm 1 month before time period t and 𝑟𝑒𝑡𝑖𝑡−2
are the returns of the firm 2 months before time period t and so on. Since the returns are calculated like this it is possible to use rolling returns. Rolling returns are used in this research because CEO turnovers happens at any period in time. Since the returns should be as good as a representation of the work of the CEO this should match as precise as possible. Consequently, the returns when a forced turnover occurred are assigned to three months
15
before the forced turnover occurred. This can be explained by the tendency of the market to react to the news of a forced turnover. The reason for this is that the market reacts to the news of a forced turnover in a positive way when the CEO is bad and in a negative way when the CEO is good (Demerjian, Lev and McVay (2012)). This might lead to biased returns since the expectations of the market for the post-turnover period are included in the return of the pre-turnover period. This is accounted for by taking the yearly returns 3 months prior to the forced turnover. However, a consequence of this measure is that it still might be possible that so post-turnover period is include in the pre-turnover period returns and that valuable information three months prior to the CEO turnover is not included in the regression.
To test if the performance sensitivity is stronger in the domain of underperformance than in the domain of outperformance, regression (1) is extended in the following way:
𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 (𝐹𝑜𝑟𝑐𝑒𝑑 𝐶𝐸𝑂 𝑡𝑢𝑟𝑛𝑜𝑣𝑒𝑟𝑖 ,𝑡= Ф(𝛽0 + 𝛽1 (𝑟𝑒𝑡𝑖𝑡− 𝑟𝑒𝑡𝑗𝑡) + 𝛽2𝑟𝑒𝑡𝑗𝑡
+ 𝛽3𝑛𝑒𝑔𝑡 + 𝛽4 (𝑛𝑒𝑔𝑡∗ (𝑟𝑒𝑡𝑖𝑡− 𝑟𝑒𝑡𝑗𝑡) + 𝜀𝑖𝑡 (3)
Where 𝛽0, 𝛽1, 𝛽2 and 𝜀𝑖𝑡 have the same interpretation as in regression (1).
Furthermore, in this regression 𝛽3𝑛𝑒𝑔𝑡 is a dummy variable with value 1 if (𝑟𝑒𝑡𝑖𝑡−
𝑟𝑒𝑡𝑗𝑡) < 0 and value 0 if (𝑟𝑒𝑡𝑖𝑡− 𝑟𝑒𝑡𝑗𝑡) ≥ 0. Consequently, 𝛽4 (𝑛𝑒𝑔𝑡∗ (𝑟𝑒𝑡𝑖𝑡− 𝑟𝑒𝑡𝑗𝑡) is
the interaction variable between this dummy and the performance of the firm. For the 𝛽3𝑛𝑒𝑔 coefficient it is expected that this has a positive effect on forced turnovers for the market, industry and peer group . The reason for this is that in earlier literature (Eisfeldt and Kuhnen, (2013)) has been found that underperformance of the market and the industry influences the turnover decision of the board of directors. Additionally, underperforming the Hoberg and Phillips peer group benchmark should have a positive effect on a forced turnover since using this benchmark should allow the firm to filter out more of the idiosyncratic risk. Consequently, underperforming these firms is more likely because of bad management than bad luck. Furthermore, the coefficient 𝛽4 (𝑛𝑒𝑔𝑡∗ (𝑟𝑒𝑡𝑖𝑡− 𝑟𝑒𝑡𝑗𝑡) should
have a negative effect on forced turnover since it is zero if the firm outperforms it peers. Therefore, it only looks how the turnover changes if the underperformance changes. If the underperformance increases CEO turnover probability should increase. Furthermore, this coefficient should have a stronger effect than the dummy variable since this interaction coefficient also allows for difference in size of underperformance.
16
Regression (1) is also used to see the effect of different peer groups on the explanatory power of the Hoberg and Phillips peer group benchmark. Firms differ in belief about how many peer groups are optimal. Consequently, different firms use different sizes in peer groups. Therefore, explanatory power can differ when the size of the peer group changes.
Furthermore, regression (1) is used to see the effect of s imilarity score on explanatory power of the Hoberg and Phillips peer group benchmarks. In theory, when the similarity score is higher the peer group firms should be able to filter out more of the exogenous shock. This should increase the explanatory power of the Hoberg and Phillips peer group benchmark.
Additionally, regression (2) is used to test if the skill of the CEO is the reason CEO turnover. To test this the explanatory power of the relative performance benchmark tested in the crisis period (2007-2009) and the non-crisis period (1996-2006 and 2010-2014). Jenter and Kanaan (2015) argued that CEO turnover is higher during a crisis period. They argued that firm performance in bad times may be more revealing about CEO skill than performance in good times. To test this statement, in this thesis is tested if the explanatory power of the relative benchmarks are higher during the crisis period.
Finally, regression (1) is used to see how extreme under or outperformance influences the probability of a forced turnover. Jenter and Kanaan (2015) found that when looking at the 10th and 90th percentile of the relative performance the probability of a forced turnover doubled. Consequently, in this thesis an increased explanatory power of the relative performance measures is expected.
4. Data
For this thesis the data library of Hoberg and Phillips (2016) is a core dataset. Their dataset contains a measure of relatedness between companies for the period 1996-2014. This dataset makes it possible to add another measure of a firm performance that is not available yet. Their industry classifications are based on the notion that firms in the same industry use many of the same words to identify and describe their products. This measure differs from other industry related measures in the following way. The traditional SIC or
17
NAICS industry classifications place firms in predefined industry categories based on production processes, not the products they offer to customers. The Hoberg and Philips similarity measure does look at the similarity of the products offered. Furthermore SIC and NAICS industries also impose transitivity among group members and provide no measure of similarity between firms within an industry, or between firms in neighbouring industries according to Hoberg and Phillips (2016).
Another core dataset of this thesis is the forced turnover dataset provided by Peters and Wagner. This dataset contains 1,141 forced turnovers during the period of 1993-2014. However, the Hoberg and phillips database only contains data from 1996 therefore observations before 1996 are dropped. Consequently, 69 forced turnovers are dropped from this database. Furthermore, 8 observations are dropped because they were included twice for the exact same date and year. Therefore, the dataser used in this thesis contains before merging 1,064 forced turnovers during the period of 1996-2014. The forced turnover dataset of Peters and Wagner is based on nearly all S&P 1500 companies. To describe a turnover as “forced” they use the popular algorithm of Parrino (1997). They extent this basic algorithm by adding an age-based turnover classification since this is robust to the biases resulting from the extent of press coverage. An important consequence is that this is likely to add false “forced” turnover. However, if the age-based turnover classification is not added this will likely lead to an underestimation of the forced turnovers.
Furthermore, the Execucomp dataset contains the id’s for all the CEOs in the Compustat database. This makes it possible to place the extracted forced turnover obtained from Peters and Wagner dataset into perspective. The CRSP/Compustat merged database is used as a link between the different databases because it contains both the gvkey identifier as the lpermno identifier. Furthermore, the firm characteristics of the sample are imported from this database.
The database starts with the Execucomp dataset. The CEO id’s are extracted from this dataset for the 1996-2014 period. The forced turnover dataset from Peters and Wagner is merging into the Execucomp dataset. The year variable, at this point, is replaced by a month variable where the date is set at 31 December of that year. After this the CRSP/Compustat merged database is merged in to make it possible to merge in the stock returns. From this point on this dataset will be called the master dataset.
18
Next, the stock returns are downloaded from the CRSP database. First the yearly rolling returns for the firms and the market are computed and after this the SIC code is used to compute the yearly rolling returns for the industry. At this point, the master dataset can be merged in to link the returns to the dataset that contains the CEO id’s and forced turnovers.
The database obtained from the website from Hoberg and Phillips contains four coefficients: year, gvkey1, gvkey2 and score. In this database the score gives the Hoberg and Phillips similarity score between gvkey1 and gvkey2. In this database gvkey1 is the firm of interest and gvkey2 is the possible peer. According to Hansen et al. (2015) the amount of peer group S&P 1500 firms that measure the quality of the work of the CEO lies between 10-20 in 72,9% of the cases. For this reason the amount of peers is set on 10 in this thesis. Consequently, all other competitors are dropped for the firm. To test for other peer sizes this is also done for the peer size 5,15,20 and 25. At this point the CRSP/compustat database is merged in. This is done on the gvkey code of the peers (gvkey2) since the returns for the peers are desired here to create peer group return. At this point the rolling returns for the average peer group returns can be calculated and merged into the master dataset.
All variable definitions and sources are summarized in Table I.
Table 2 reports descriptive statistics. In Panel A the total observations are displayed. The final sample has 36,066 observations for time period 1996-2014. This final sample contains 938 forced turnovers and 35,128 non-forced turnovers. Panel B shows the firm characteristics for forced and non-forced observations. Firms where a forced turnover occurred at year t are larger in terms of book assets, revenue and number of employees relative to firms where no forced turnover occurred. Firms where no forced turnover occurred tend to have a larger market value of equity. This partly contradicts the results from Jenter and Kanaan (2015) who found that firms with a forced CEO turnover were on average smaller than firms with no forced CEO turnover. They argued the smaller firms are likely due to the fact that CEO dismissals are preceded by bad performance which is associated declines in firm size. Another possible explanation for the difference in firm characteristics is the range of the dataset used in this thesis. The dataset used in this thesis differs from the one used in Jenter and Kanaan (2015). Their dataset contains data from 1993-2009. Panel D shows the difference in firm characteristics for the time periods 1996-2007 and from 2008-2014. These results are added to see if the difference in time period is
19
the reason for the deviating results from the Jenter and Kanaan (2015) paper. The results show larger book assets, revenues and number of employees when a forced turnover occurred for both time periods. Furthermore, both time periods have a s maller market value of equity for forced turnovers. Consequently, the difference of firm characteristics is not the result of the different time period.
Panel B displays that the Hoberg and Phillips similarity score is 0.079 for firms where a forced turnover occurred in year t where this is 0.082 for firms where no forced turnover occurred in year t. Additionally, Panel B displays that the 3 months lagged, 12-month average stock return before a forced turnover is -10,91%. While the 12-month average stock return when no forced turnover occurred is 18,45%. Furthermore, market return before a forced turnover is also lower than when no forced turnover occurred. The market returns before a forced turnover are 10.15% while the market returns when no turnover occurs are 10.35%. Additionally, the industry returns are also lower before a forced turnover (0.04%) than when no turnover occurs (2.5%). The average industry returns are much lower than the average industry returns from Jenter and Kanaan (2015). This might indicate that the way the industry returns are calculated in this thesis are differ from how they calculated their industry performance. Another possible explanation can be that the dataset is flawed for the industry returns which could bias the results. Finally, the peer group returns show the same trend as the market and the industry returns. When a forced turnover occurs, the peer group returns are lower (10.34%) than when no forced turnover occurs (15.54%). This indicates that a forced turnover is more likely when the market, industry and peer group firms underperform. This is in line with the findings of Jenter and Kanaan (2015) who found that a forced turnover is more likely to occur when the market or the industry underperform.
Panel C reports the peer group performance when the number of peers change. This panel shows that when the number of peers increase, the average Hoberg and Phillips similarity score decreases. This is a logical result from how the peer groups are formed since the firms with the highest Hoberg and Phillips similarity score are chosen as peers. Consequently, adding more peer decreases the average similarity score of these peers. Furthermore, the peer group returns do not show large deviations from the peer group returns when the size of the peer group is 10. All peer group returns are lower for the forced turnover and larger for the no forced turnover data. Additionally, the difference in peer
20
group return relative to the chosen peer group is small. The absolute largest difference is 0.16% for the forced turnover data and 0.20% for the no forced turnover data.
5. Results
5.1 Explanatory power of different benchmarks
Table III presents the results of the logistic regression estimating the probability of a forced turnover. The results show that market returns have no significant effect on the CEO forced turnover decision. This indicates that market returns are completely filtered out when CEO performance is determined. This contradicts the results of Kaplan and Minton (2012) and Jenter and Kanaan (2015) who found that market performance is important for forced CEO turnover decisions. The results in table III show that in this sample CEOs from firms in poorly performing industries also are significant more likely to be fired at a 1% level. This indicates that industry performance is not completely filtered out when CEO performance is determined. This is in line with the findings of both Jenter and Kanaan (2015) and Kaplan and Minton (2012). Finally, table III indicates that peer group performance is not completely filtered out when CEO performance is determined. This result is significant on a 1% level. However, the coefficient for the effect of peer group return of CEO turnover is smaller than the effect of industry peer group return on CEO turnover. This indicates that peer group performance is better filtered out when determining CEO performance. Conclusively, the board of directors do a better job at filtering out peer group performance than they do at filtering out industry performance.
Furthermore, table 3 III contains relative performance measure coefficients. All relative performance measures are significant at a 1% level. The sensitivity of forced CEO turnover to idiosyncratic performance is the smallest when using industry returns as a benchmark. Furthermore, table III shows that the sensitivity of forced CEO turnover to idiosyncratic performance is larger for when the market return is used as a benchmark, relative to when the industry returns as a benchmark. This contradicts the results of Jenter and Kanaan (2015) and Kaplan and Mintion (2014) who found a stronger effect of relative industry performance on forced CEO turnover than form relative market performance. Finally, the results show that the sensitivity of forced CEO turnover to idiosyncratic
21
performance is the greatest when the Hoberg and Phillips peers are used as a benchmark. This indicates that the Hoberg and Phillips relative peer group performance measure is a more precise measure of CEO turnover than the industry or market relative performance measures.
A remark on these results is that table III shows that the Wald-test does not significantly differ for the difference between the relative market performance and the relative peer group performance. This indicates that these coefficients are not significantly different. Consequently, when making conclusions about these coefficients this should be taken into account.
5.2 the effect of underperformance
The results of table III shows the effect of the relative performance measures introduced in this paper on the probability of a forced CEO turnover. Table IV adds a dummy for underperformance and an interaction variable to test for the effect of relative performance when there is underperformance.
The variables of main interest in table IV are the dummy variable and the interaction variable. The market, industry and peer group underperformance dummy and interaction variables are significant at a 1% level. The results show that the effect of idiosyncratic performance is largest in the domain of underperformance for the relative Hoberg and Phillips peer group benchmark, relative industry benchmark and relative market benchmark. The results indicate that underperforming the market relative benchmark has the largest effect on the probability of a forced CEO turnover. Furthermore, Table IV shows that underperforming the relative industry benchmark has the smallest effect on the probability of a forced CEO turnover. This is in contradiction with the findings of Barro and Barro (1990), Gibbons and Murphy (1990) and Warner, Watts and Wruck (1988) who found that CEO turnover is more likely if firm performance is poor relative to industry.
Table IV shows that the relative performance measures for the market and the Hoberg and Phillips peer group benchmark are no longer significant when the dummy and the interaction variable are added to the regression. This indicates that the results found in table III for the relative performance benchmark coefficients are mainly caused by the sensitivity of forced CEO turnover to idiosyncratic performance when the firm
22
underperforms the benchmark. However, the industry relative benchmark is still significant at a 5% level. Despite that the explanatory power of this coefficient decreased, this indicates that the sensitivity of forced CEO turnover to idiosyncratic performance not only caused by underperformance when the industry is used as a benchmark.
5.3. the effect of the peer group size
The results in table V show the difference in relative performance for the Hoberg and Phillips peer group when the size of the peer group differs from 10. These results are relevant since firms do have different beliefs about the optimal size of the peer groups. Consequently, peer group size between firms does differ. This table makes it possible to account for this.
Table V shows that for all sizes of peer group the relative peer group returns are significant at a 1% level. The explanatory power of the Hoberg and Phillips peer group is the strongest when the size of the peer group is 15. The difference between the peer group (size=10) and possible other peer group sizes is tested with a Wald-test to see if a different peer group size would lead to different results. Table V shows that for peer group size 15 the results are significantly different at a significance level of 10%. This indicates weak statistical significance. Additionally, the difference in explanatory power is small and the difference between these coefficients are only significant at a 10% level. This indicates only weak statistical significance. Consequently, the results from table V show that the size of the peer group is not of significant interest.
5.4 similarity score
The results in table VI show the difference in relative performance for the peer group when the similarity score used to calculated the peer group returns differs. These results are relevant since a higher similarity score should in theory be a better peer. Consequently, a higher similarity score should increase the explanatory power of the relative peer group coefficient.
Table VI shows us that at the 10th, 20th and 50th percentile level the relative peer group returns are significant at a 1% level. At the 50th percentile the difference between the
23
highest and the lowest percentile is significant at a 1% level. Furthermore, the effect on forced CEO turnover is larger when the similarity score is high relative to low. Conclusively, table VI shows that the Hoberg and Phillips peer group ability to link idiosyncratic performance to forced CEO turnover increases when the similarity score is higher. This indicates that the Hoberg and Philips similarity score is a good indicator of which firms are ‘good’ peers. The intuition behind this is that better peer firms can filter out more exogenous shocks and therefore increase the effect idiosyncratic performance has on forced CEO turnover.
5.5 Forced turnover during crisis period
Table VII presents the results of the logistic regression estimating the probability of a forced turnover when there is a distinction between crisis and non-crisis period. This a relevant regression because Jenter and Kanaan (2015) found that during a crisis period more skill induced forced CEO turnovers occurred. They argued that this could be case, for exampl e, if bad industry performance correlates with changing skills requirements for CEOs.
Table VII shows that for both crisis as non-crisis period the market returns are filtered out when determining CEO performance since these coefficients are not significant. Furthermore, table VII shows that during the crisis period for both industry as peer group returns are better filtered out when determining CEO performance than during the non-crisis period. This supports the reasoning of Jenter and Kanaan (2015) that there is a difference in explanatory power of the skill of the CEO during good or bad times is correct. The intuition behind this is that when during the crisis period the exogenous shocks are filtered out, the increased CEO turnover cannot be attributed to misallocation of credit. Consequently, there must be another reason why CEO turnover is higher during bad industry performance and this could be the increased learning of skill of the CEO.
Table VII shows that forced CEO turnover is more sensitive to idiosyncratic performance during the non-crisis period than during the crisis period for the relative market, industry and peer group benchmarks. These results are all significant at the 1% level. This difference is only significantly different for the relative industry benchmark. Furthermore, the proportions of these benchmark relative to each other are for the crisis and non-crisis period the same as in table III.
24
5.6 Forced turnover during under- and outperformance
The results that are presented in table VIII are relevant since Jenter and Kanaan (2015) found that a decrease in firm performance from the 90th to the 10th percentile doubles the probability of a forced turnover. Consequently, it is relevant to look how under or outperformance of the market does influence the effect of the idiosyncratic performance on forced CEO turnover. Furthermore, it is relevant to check how well the market, industry or peer group returns are filtered out during underperformance or outperformance.
Table VIII shows that the dummy variables for market, industry and peer group returns at the 90th percentile are only significant for the market returns. This indicates that when the firm outperforms the market, the market returns are not filtered out while table III shows that for the overall sample this is the case. Furthermore, the results show that industry and the Hoberg and Phillips peer group performance are filtered out when the firm outperforms their peers.
Table VIII shows that the dummy variables for the market , industry and peer group returns at the 10th percent are all significant. Additionally, these dummy’s all have a positive effect on the probability of a forced CEO turnover. Furthermore, the dummy variables for relative market, industry and peer group returns are all significant at a 1% level and have a positive effect on the probability of a forced CEO turnover. These results confirm the results found at table IV that turnover-performance sensitivity is much greater in the domain of underperformance.
Furthermore, the results show that the relative performance coefficients for the market, industry and Hoberg and Phillips peer group benchmark are significant at a 1% level. Furthermore, they all have a negative effect on the probability of a forced turnover. The effect of outperforming the market is the largest. These results contradict the findings of Jenter and Kanaan (2015) that there is only a small effect that outperforming the benchmark has on the forced CEO turnover probability. Conclusively, outperforming the benchmark has a negative effect on the probability of a forced CEO turnover.
25
In this section the limitations and the robustness checks are discussed. As previously discussed, the number of peer groups may affect the results. However, table V shows that this is not the case. Furthermore, the effect of the similarity score is tested. Table VI shows that there is a significant difference between ‘high’ and ‘low’ similarity scores at the 20th percentile and the 50th percentile. The results indicate that higher similarity scores lead to an increase explanatory power of model. These results suggest that the Hoberg and Phillips similarity score is a good way of determining peers.
Moreover, the returns when a forced turnover occurred are assigned to three months before the forced turnover occurred. This can be explained by the tendency of the market to react to the news of a forced turnover. The reason for this is that the market reacts to the news of a forced turnover in a positive way when the CEO is bad and in a negative way when the CEO is good (Demerjian, Lev and McVay (2012)). This might lead to biased returns since the expectations of the market for the post-turnover period are included in the return of the pre-turnover period. This is accounted for by taking the yearly returns 3 months prior to the forced turnover. However, a consequence of this measure is that it still might be possible that so post-turnover period is include in the pre-turnover period returns and that valuable information three months prior to the CEO turnover is not included in the regression.
Furthermore, to prevent the returns to be influenced by ‘extreme’ returns all returns are winsorized at a 0.5% level. Additionally, for all regressions robust standard errors are used. To test if the variables between regressions are significantly different from each other the Wald-test is used.
6. Conclusion
In this thesis, the hypothesis whether or not the Hoberg and Phillips peer return benchmark is a more precise predictor for forced CEO turnover than industry peer return and market peer return is tested. This is tested with a logit regression where the relative firm performance against the market, industry and peer return benchmark is regressed on the forced turnover data. The results indicate that board of directors do a better job at filtering
26
out peer group performance than they do at filtering out industry performance and market performance. Furthermore, the sensitivity of forced CEO turnover to idiosyncratic performance is the greatest when the Hoberg and Phillips peers are used as a benchmark. Consequently, the Hoberg and Phillips peer return benchmarks seems to be a more precise indicator for forced CEO turnover than industry peer return and market peer return.
Furthermore, in this thesis the effect of underperforming the relative benchmark is tested. The results show that turnover-performance sensitivity is much greater in the domain of negative performance for all benchmarks. However, underperforming the relative market performance has the greatest effect on the forced CEO turnover probability. The relative industry benchmark has the smallest effect on the probability of a forced CEO turnover. This is in contradiction with the findings of Barro and Barro (1990), Gibbons and Murphy (1990) and Warner, Watts and Wruck (1988) who found that CEO turnover is more likely if firm performance is poor relative to industry.
Additionally, the effect of the peer group performance on forced CEO turnover when the number of peers differ is tested with the Wald-test. the results show that the size of the peer group is not of significant interest.
The effect of the similarity score on the explanatory power of the relative peer group performance is tested in this thesis. The results indicate that there is a significant difference between the explanatory power of the relative peer group performance. When the Hoberg and Phillips similarity score is ‘high’ then the explanatory power of the relative peer group performance is higher than when the Hoberg and Phillips similarity score is ‘low’. This indicates that the Hoberg and Philips similarity score is a good indicator of which firms are ‘good’ peers.
The effect of being in a crisis period or a non-crisis period has been tested in this thesis. The results indicate that during the crisis period for both industry and peer group returns are better filtered out when determining CEO performance than during the non-crisis period. This finding supports that reasoning of Jenter and Kanaan (2015) that there is a difference in explanatory power of the skill of the CEO during good or bad times. Furthermore, the results suggest that forced CEO turnover is more sensitive to idiosyncratic performance during the non-crisis period than during the crisis period for the relative market, industry and peer group benchmarks.
27
The effect of extreme underperformance and outperformance on forced CEO probability has been tested. The results show that when a firm outperforms the market, the market returns are not filtered out. Furthermore, when a firm outperforms the market, industry and peer group returns are filtered out. The dummy’s for underperformance confirm the results that already have been established of the effect of underperformance on the probability of a forced CEO turnover. Finally, outperforming the benchmark has a negative effect on the probability of a forced CEO turnover. This contradicts the results of Jenter and Kanaan (2015) who found that outperforming the market only has a small effect on the probability of a forced CEO turnover.
This thesis has the following limitations. First, in this thesis the crisis and the non-crisis period are calculated. The 1997-2000 crisis is not included in the crisis period while there actually was a crisis during this period. This might have biased the results. Second, the industry returns found in this thesis are low relative to previous research. This might indicate that something went wrong during the calculation of the industry returns which might lead to biased results. Finally, some of the forced turnover data had to be dropped since the database of Hoberg and Phillips only contained data from 1996. Furthermore, only forced turnovers are tested in this thesis. By using the forced turnover dataset from Peters and Wagner the forced turnover definition is as broad as acceptable. However, it is impossible to filter out all the forced turnovers so it is likely that the data sample used also contains unforced turnovers. Additionally, the data sample used probably misses some forced or performance induced turnovers that are labelled as voluntary turnovers by Peters and Wagner (2014) and Jenter and Kanaan (2015). This might bias the results.
Despite the limitations of this research, the results show that the Hoberg and Phillips peer return benchmark is a more precise predictor than the market and industry peer return benchmarks. Consequently, this benchmark is an interesting option to measure CEO performance in different research areas. However, since the results of this thesis are likely biased for the industry returns more research is needed before someone can consider to use this benchmark as a measure for CEO performance.
Consequently, more research to confirm the power of the Hoberg and Philips peer return benchmark is needed. Furthermore, the Hoberg and Phillips peer return benchmark might be a better indicator of CEO performance, but it still fails to completely filter out
28
exogenous shocks. Consequently, further research for a better indicator of CEO performance is needed.
References
Albuquerque, A. M., De Franco, G., & Verdi, R. S. (2013). Peer choice in CEO compensation. Journal of Financial Economics, 108(1), 160-181.
Barro, J. R., & Barro, R. J. (1990). Pay, performance, and turnover of bank CEOs. Journal of
Labor Economics, 8(4), 448-481.
Bertrand, M., & Mullainathan, S. (2001). Are CEOs rewarded for luck? The ones without principals are. The Quarterly Journal of Economics, 116(3), 901-932.
Bhagat, S., Bolton, B. J., & Subramanian, A. (2010). CEO education, CEO turnover, and firm performance.
Bizjak, J. M., Lemmon, M. L., & Naveen, L. (2008). Does the use of peer groups contribute to higher pay and less efficient compensation?. Journal of Financial Economics, 90(2), 152-168.
Bizjak, J., Lemmon, M., & Nguyen, T. (2011). Are all CEOs above average? An empirical analysis of compensation peer groups and pay design. Journal of Financial
Economics, 100(3), 538-555.
Bonnier, K. A., & Bruner, R. F. (1989). An analysis of stock price reaction to management change in distressed firms. Journal of Accounting and Economics, 11(1), 95-106. Brav, A., Jiang, W., Partnoy, F., & Thomas, R. (2008). Hedge fund activism, corporate
governance, and firm performance. The Journal of Finance, 63(4), 1729-1775.
Coughlan, A. T., & Schmidt, R. M. (1985). Executive compensation, management turnover, and firm performance: An empirical investigation. Journal of Accounting and
Economics, 7(1-3), 43-66.
Diamond, D. W., & Verrecchia, R. E. (1982). Optimal managerial contracts and equilibrium security prices. The Journal of Finance, 37(2), 275-287.
Demerjian, P., Lev, B., & McVay, S. (2012). Quantifying managerial ability: A new measure and validity tests. Management Science, 58(7), 1229-1248.
Denis, D. J. and Denis, D. K. (1995) Performance changes following management dismissals, Journal of Finance 50, 1029–1057.
29
Dikolli, S. S., Mayew, W. J., & Nanda, D. (2014). CEO tenure and the performance-turnover relation. Review of accounting studies, 19(1), 281-327.
Eisfeldt, A. L., & Kuhnen, C. M. (2013). CEO turnover in a competitive assignment framework. Journal of Financial Economics, 109(2), 351-372.
Farrell, K. A., & Whidbee, D. A. (2003). Impact of firm performance expectations on CEO turnover and replacement decisions. Journal of Accounting and Economics, 36(1), 165-196.
Faulkender, M., & Yang, J. (2010). Inside the black box: The role and composition of compensation peer groups. Journal of Financial Economics, 96(2), 257-270.
Fisman, R. J., Khurana, R., Rhodes-Kropf, M., & Yim, S. (2013). Governance and CEO turnover: Do something or do the right thing?. Management Science, 60(2), 319-337. Gibbons, R., & Murphy, K. J. (1990). Relative performance evaluation for chief executive
officers. ILR Review, 43(3), 30-S.
Hansen, E., Ma, C., Tankiang, F., & Wang, E. (2015) 2015 Peer Group Benchmarking: An
Analysis of S&P 1500 Companies. Retrieved from Equilar:
http://paygovernance.com/wp-content/uploads/2015/10/Equilar-2015-Peer-Group-Report.pdf
Hermalin, B. E., & Weisbach, M. S. (1998). Endogenously chosen boards of directors and their monitoring of the CEO. American Economic Review, 96-118.
Hoberg, G., & Phillips, G. (2010) Review of Financial Studies (October 2010), 23 (10), 3773-3811.
Hoberg, G., & Phillips, G. (2016), Journal of Political Economy (October 2016), 124 (5) 1423-1465.
Hölmstrom, B. (1979). Moral hazard and observability. The Bell journal of economics, 74-91. Hölmstrom, B. (1982). Moral hazard in teams. The Bell Journal of Economics, 324-340. Hölmstrom, B., & Kaplan, S. N. (2003). The state of US corporate governance: What's right
and what's wrong?. Journal of Applied Corporate Finance, 15(3), 8-20.
Huson, M. R., Parrino, R., & Starks, L. T. (2001). Internal monitoring mechanisms and CEO turnover: A long‐term perspective. The Journal of Finance, 56(6), 2265-2297.
Jenter, D., & Kanaan, F. (2006). CEO turnover and relative performance evaluation (No. w12068). National Bureau of Economic Research.
