Determinants of peer group selection for relative
performance evaluation and the effect of low performance
and entrenchment on the peer group selection
Master Thesis by Dennis Groot (10165142)
University of Amsterdam Faculty of Economics & Business
MSc Finance Corporate Finance
Supervised by dhr. dr. T. Jochem
2
Statement of originality
This document is written by Dennis 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 are 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
Acknowledgement
I would like to thank Torsten Jochem for being my supervisor for this thesis. Without his advice and excel sheet this thesis would not be able to be created. I would also like to thank my parents, brother and girlfriend for their support during the whole process of making this thesis.
4
Abstract
This paper studies what the determinants are for peer group selection for relative performance evaluation. Further it is also studies whether or not low performing and entrenched companies change their peers. Hence, the research question is: What are the determinants of peer group selection in relative performance evaluation and do low performing and entrenched companies change their peers? For the research on the determinants the efficient benchmarking hypothesis is formulated. For the research on whether or not low performing and entrenched companies change their peers the strategic selection hypothesis is formulated. Logit regression models are used to test the two
hypotheses. The peer companies are hand-collected and the rest of the data are collected using various databases. The sample contains 180 RPE companies and they have on average 11.52 peer companies. The sample period is from 1-1-2007 until 31-12-2011. The results of the regression done for the efficient benchmarking hypothesis shows that the determinants of peer group selection in relative performance evaluation are same industry membership, correlated stock performance, revenues, market capitalization and same stock index membership. For the strategic selection hypothesis the results show that low performing companies do not change their peers. However, entrenched companies do have a higher probability to change their peer companies and therefore strategic selection seems to occur when choosing the peer companies.
5
Table of contents
1. Introduction ... 6
2. Literature review ... 8
2.1 Relative performance evaluation versus compensation benchmarking... 8
2.2 Determinants of relative performance evaluation ... 9
2.3 Determinants of compensation benchmarking ... 11
2.4 Management entrenchment... 13
2.5 Strategic selection ... 14
2.6 Hypotheses ... 16
3. Methodology ... 18
3.1 Model to test the efficient benchmarking hypothesis ... 18
3.2 Explaining the variables of the model to test the efficient benchmarking hypothesis ... 18
3.3 Model to test the strategic selection hypothesis ... 19
3.4 Explaining the variables of the model to test the strategic selection hypothesis ... 19
3.5 Important variables ... 21
3.6 Control variables ... 22
4. Data ... 23
4.1 Finding RPE companies and their peers ... 23
4.2 Databases ... 25
4.3 Summary statistics for the efficient benchmarking hypothesis ... 25
4.4 Summary statistics for the strategic selection hypothesis ... 28
5. Results ... 29
5.1 Efficient benchmarking hypothesis ... 29
5.2 Strategic selection hypothesis ... 32
6. Robustness checks ... 35
7. Conclusion ... 38
References ... 42
6
1. Introduction
On December the 15th 2006 a new SEC disclosure rule requires companies to give a detailed description of the process used to set performance targets that are used in the compensation of their executives. This new SEC rule gives the opportunity to research the use of relative performance evaluation (RPE) by companies. So far little research have been done on this topic, while the use of RPE is substantial. Gong, Li and Shin (2011) find that 25.44 percent of the S&P1500 companies use RPE. The use of RPE entails that companies use peer companies performance to evaluate their own performance and with that to set their incentive compensation. Because of the little research done on this topic the first question of this paper is what the determinants of the peer group selection for RPE are. The second question of this paper is if low performing and entrenched companies change their peers. combining these two question results in the research question of this paper. Hence, the research question is: What are the determinants of peer group selection in relative performance evaluation and do low performing and entrenched companies change their peers? To answer the research question two hypotheses are formed. The first hypothesis is the efficient benchmarking hypothesis. Under the efficient benchmarking hypothesis it is stated that same industry membership, correlated stock performance, revenues, market-to-book ratio, market capitalization and same stock index membership are all determinants for being selected as a peer company. The second hypothesis is the strategic selection
hypothesis. Under the strategic selection hypothesis it is stated that low performing and entrenched companies change their peers. Companies using RPE are referred to as RPE companies and the peer companies of these RPE companies are referred to as peer companies.
As already mentioned, limited research has been done on RPE. This due to limited disclosure before December 2006 on RPE contracts. Because of the limited research done on RPE, this paper could contribute by understanding the basis on how a peer company is selected and if strategic selection occurs with the use of RPE.
To research the efficient benchmarking hypothesis a logit regression is used. The dependent variable is labelled as chosen as peer. The independent variables of interest are all the variables mentioned in the efficient benchmarking hypothesis. Therefore all the independent variables are of interest for the efficient benchmarking hypothesis. The construction of the regression model is done by combining existing literature. The data of
7 the peer companies is hand-collected and the sample period is from 1-1-2007 until 31-12-2011. The data is hand-collected because there is not a database available. All other data is collected by using different databases. The data is unique in comparison to existing
literature, because the sample period is larger and therefore can capture more effects on the long run. To research the strategic selection hypothesis a logit regression model is used. The dependent variable is labelled as deselected as peer. The independent variables of interest are the variables stock return of the RPE company, stock return of the peer company, the interaction variable of stock return of the RPE company and the peer
company, entrenchment, the interaction variable of entrenchment and the stock return of the RPE company and the interaction variable of entrenchment and the stock return of the peer company. The construction of the regression model is done by modifying a model used by Denis, Jochem and Rajamani (2017) and combining existing literature. The data for the strategic selection hypothesis makes use of the hand-collected data collected for the efficient benchmarking hypothesis. All other data is collected by using different databases. The methodology and data is unique, because this is not researched before in existing literature. All the RPE companies have a index membership in the S&P500. A total of 322 companies fulfilled the criteria mentioned in section 4.1. 180 companies of these 322 companies are labelled as RPE companies and these 180 RPE companies have on average 11.52 peer companies.
The research finds that for the efficient benchmarking hypothesis same industry membership, correlated stock performance, revenues, market capitalization and same stock index membership all have significant influence on being chosen as a peer company. Market-to-book ratio is significant at the 5% level, but the coefficient of the variable is positive. This means that the larger the difference between the RPE company and the possible peer company, the more likely it is that the possible peer company is chosen as a peer company. This is not in line with expectations and with previous research discussed in section 2.2 and 2.3. For the efficient benchmarking hypothesis, the research finds that entrenchment is a significant influence on being deselected as a peer company. If the RPE company has a entrenched board then there is a higher probability that a peer company is deselected as a peer company. The stock returns of both the RPE company and the peer company are not significant at the 5% level. Also the three interaction variables are not significant at the 5% level. This means that the two return variables and the three interaction variables have no
8 influence on the deselection of peer companies. Using the findings of the two hypotheses, the research question can be answered. The determinants of peer group selection in relative performance evaluation are same industry membership, correlated stock performance, revenues, market capitalization and same stock index membership. Looking at the second part of the research question, the results show that low performing companies do not change their peers. However, entrenched companies do have a higher probability to change their peer companies and therefore strategic selection seems to occur when choosing the peer companies. The conclusion on the link of entrenchment and strategic selection needs to be taken carefully, because it may be possible that omitted characteristics of staggered boards and CEO ownership have an influence in this link.
The remainder of this paper is organized as follows. In section 2 the literature review discusses the distinction between relative performance evaluation and compensation benchmarking, previous research on determinants of selecting peer companies for relative performance evaluation and compensation benchmarking, entrenchment, strategic selection and the two hypotheses. In section 3 the methodology for the two hypotheses is discussed. Section 4 discusses the data collecting and provides descriptive statistics. In section 5 the results are discussed. Section 6 discusses the robustness checks and the last section discusses the conclusion for this research.
2. Literature review
2.1 Relative performance evaluation versus compensation benchmarking
The use of peer groups in compensation contracts can be categorized in two different groups. Namely, relative performance evaluation peer groups and compensation
benchmarking peer groups (Gong, Li, & Shin, 2011). Bizjak, Lemmon and Nguyen (2011), and Gong, Li, and Shin (2011) both indicate that compensation benchmarking provides the boards of directors with useful information for determining the level of executive pay. This information can be used to set a competitive pay level at which it is possible to retain, motivate , and attract top executives in a competitive market for executive talent.
Compensation benchmarking is an effective method of comparing the salaries of executives of a company with those of other companies. This because the compensation peer group
9 companies are also the companies which a company competes with for executive talent. (Bizjak, Lemmon, & Nguyen, 2011).
However, the use of relative performance evaluation differs with compensation benchmarking. Several papers explain the function of relative performance evaluation in compensation contracts. Aggarwal and Samwick (1999), Albuquerque (2009), Holmstrom (1982), and Murphy (1999) all indicate that relative performance evaluation can be valuable when the agent that set the pay level has some sort of uncertainty about the company’s performance position in the market. Common industry shocks will be filtered out by the relative performance evaluation and that helps to exclude the component of the executive compensation that is driven by the exogenous shock. Relative performance evaluation positions the company’s performance relative to its competitors and with that helps the company to determine if the executives should be rewarded or punished accordingly. Murphy (1999) also indicates that relative performance has an advantage over absolute performance when looking at the executive pay. This because relative performance essentially has the same incentives for executives as absolute performance does, but it insulates executives from common shocks of which the executives have no control over.
2.2 Determinants of relative performance evaluation
Research done on the determinants of peer group selection for the use of relative performance evaluation is limited and this results in limited information available about this topic. Therefore only two papers will be discussed in this section. That is the paper by Bizjak, Kalpathy, Li and Young (2017) and the paper by Gong, Li and Shin (2011).
The study conducted by Bizjak, Kalpathy, Li and Young (2017) is still a working paper which has not yet been reviewed and published. So the results of this paper should be considered with caution. Bizjak, Kalpathy, Li and Young (2017) use data from ISS Incentive Lab between 2006 and 2015. This resulted in 2,070 relative performance evaluation
company-year awards. Using a logit analysis they find that peers are more likely to be chosen when they are from the same industry, have a higher correlation in stock return relative to potential candidate companies, the peer company shares the same membership in a market index, the peer company has greater institutional ownership and a higher credit score compared to other possible peer companies (Bizjak, Kalpathy, Li, & Young, 2017).
10 Gong, Li and Shin (2011) use a new SEC disclosure rule that is effective for fiscal years ending on or after December the 15th 2006, to get the peers names of the relative
performance evaluation companies. This new rule requires companies to give a detailed description of the process used to set performance targets that are used in the
compensation of their executives. The new disclosure rule also makes it mandatory for companies to describe how the performance targets are translated into the compensation determination. So, this means that companies which are using relative performance evaluation are required to disclose detailed information about the relative performance evaluation contracts. Prior to 2006, this was all voluntarily (Gong, Li, & Shin, 2011). The initial sample of Gong, Li and Shin (2011) consists of 1,419 S&P1500 companies. They look at the first proxy statements filed for the fiscal years 2006 and 2007 and find that in their sample, 361 companies use relative performance evaluation. Of that sample of 361 companies, 135 companies are found that use self-selected peer groups and that sample is used in their research on the determinants of peer selection. Gong, Li and Shin (2011) use a logistic regression method with a total of 1,668 selected peers and 1,668 unselected peers. They find that higher return comovement, same industry and companies that are included in the S&P1500 index are more likely to be chosen as relative performance evaluation peers. Also similar companies are more likely to be chosen as peers. Similar companies are companies that are close to each other regarding the size of the company. Size is measured as market value of the company. While high return comovement is a determinant for peer selection, Gong, Li and Shin (2011) find that similar stock performance does not increase significantly the change of being selected as a peer company. Of all the determinants that increase the likelihood of being selected as a peer, same industry has the greatest economic significance. Companies which have the same first two-digit SIC code as the relative performance
evaluation company, have a 41.6% more likelihood of being selected as a peer company. Having a S&P index membership also has a significant influence on being selected as a peer company. Size similarity and performance comovement do have a significant impact.
However, the economic impact is much smaller than for the other determinants. This shows that the influence of each determinant is different on the chances of being selected as a peer company (Gong, Li, & Shin, 2011).
11 2.3 Determinants of compensation benchmarking
As previous mentioned, The use of peer groups in compensation contracts can be categorized in two different groups. Theory about the determinants of peer companies with compensation benchmarking could then have a basis for the theory about the determinants of peer companies with relative performance evaluation. Gong, Li and Shin (2011) indicate that peer performance is the opposite for compensation benchmarking peers than that for relative performance evaluation peers. High performing companies tend to pay higher compensation and so are more likely to be chosen as a peer for compensation
benchmarking. However, high performing companies are more difficult to outperform so are less likely to be chosen as a peer for relative performance evaluation. This means that the determinants for selecting compensation benchmarking peers could have the opposite effect. However, they could also be a determinant for selecting the relative performance evaluating peers. Gong, Li and Shin (2011) find that in their sample only 17.31% of the companies use the same peer groups for both compensation benchmarking and relative performance evaluation. The other companies use partially different peers or completely different peers for the two categories of compensation contracts. This shows that the determinants of the peer selection of the two categories of compensation contracts are substantially different.
The research done by Porac, Wade and Pollock (1999) use a final sample of 280 companies. All these companies are selected out of the S&P500 in the beginning of 1993. Proxy statements are taken from only these companies and only for the year 1993. Using ordinary least squares regression, they find that boards select comparable companies primarily on the basis of being in the same primary industry. Even when a part of the selected peers are active in a different industry, the majority of peers that were selected were active in the same industry or had the same industry-based attributes as of the comparing company. Porac, Wade and Pollock (1999) also find that when companies used peer companies which were active in a different industry, they discussed these comparisons significant less than companies which selected peers that are active in the same industry. This make it seem that the companies are aware of the counter-normative nature of the selection of the peers that are from outside the same industry.
Bizjak, Lemmon and Nguyen (2011) conclude that companies tend to select peers that operate in the same industry and are of a similar size. They find that 62% of the peer
12 companies are from the same industry and 36% of the peer companies that are in the same industry have revenues that are in the range of 50% to 200% of the revenues of the
companies that uses these peers. While the two determinants are of a significant large influence, they are not the only factors that influence the likelihood of being selected as a peer. Similar Market-to-book ratios and similar accounting performance are also
determinants that increase the likelihood of being selected as a peer. Bizjak, Lemmon and Nguyen (2011) use a sample size which is draw out of a sample of 1,161 companies. These companies are taken out of the ExecuComp database. 800 companies from these 1,161 companies report the peer group that they use for the setting of the executive
compensation. So the final sample size is 800 companies for the fiscal year 2006. They use a multivariate logit regression to conduct their research.
Using a sample of 763 companies and 2,066 company-years over a period from December 2006 until November 2010, Faulkender and Yang (2013) find that industry overlap and similar size are two important factors which explain why a company is chosen as a peer. Additionally companies tend to chose peer companies which are in the same index. So if a company is in the Dow 30 than it is more likely that the peer company is also in the Dow 30. The same occurs for companies which are in the S&P500. Faulkender and Yang (2013) also find that differential stock performance has not a significant effect on the likelihood of being chosen as a peer company. A discrete choice regression technique is used to find the
determinants that are used in the peer selection process.
Another paper by Faulkender and Yang (2010) use a hand-collected sample of 83 companies in 2005 and 373 companies in 2006. All these companies were included in the S&P500 and named the peer companies in their SEC DEF-14A filings, which were available from the EDGAR database. The sharp incline of companies which name their peers between 2005 and 2006 comes from the requirement that since 2006 companies are required to provide detailed information about the peer group. Faulkender and Yang (2010) also find that peer companies are chosen on the basis of being in the same industry and having a similar size. They also find that the level of compensation also have a statistically significant influence in the likelihood of being chosen as a peer company. Companies with a higher executive compensation are more likely to be chosen as a peer company. These results are found using a probit regression model.
13 2.4 Management entrenchment
Management entrenchment means that managers can take actions which results in making it for shareholders costly to replace them. The advantage for managers is that they obtain more latitude in determining corporate strategy, extract higher wages from
shareholders , extract larger perquisites and reduce to probability of being laid off (Shleifer & Vishny, 1989).
Faleye (2007) researches the effect of having a classified board on the probability of a forced CEO turnover. Using a sample of 2,072 companies, Faleye (2007) looked at the proxy statements of these companies together with newswire and newspaper reports collected from Factiva. During a sample period of January 1995 and December 2002, Faleye (2007) found that of the sample companies 1,483 CEOs were replaced. Of these 1,438 CEO replacements, 219 replacements were involuntary. Of these 219 involuntary turnover, 84 companies had classified boards, while 135 companies did not had classified boards. Using a cross-sectional time-series logistic model, Faleye (2007) finds that having a classified board has a negative effect on forced CEO turnover. This result is significant at the 1% level. This means having a classified board lowers the probability of having a forced CEO turnover and so increases the entrenchment of the management.
Morck, Shleifer and Vishny (1988) use an initial sample of 456 companies that are listed as Fortune 500 companies in December 1980. They research the relationship between market valuation and management ownership. Market valuation is measured by Tobin’s Q. The data to measure Tobin’s Q were obtained from the Griliches R&D master file for 1980. However, for 85 companies the values needed to acquire Tobin’s Q were not available, making the final sample consist of 371 companies. Using a piecewise linear regression, Morck, Shleifer and Vishny (1988) find that Tobin’s Q rises when board ownership rises from 0% to 5%. Tobin’s Q then declined when board ownership rises from 5% to 25%. Tobin’s Q then rises again when board ownership rises beyond 25%, although the increase of Tobin’s Q much slower is compared to the increase with board ownership from 0% to 5%. The
interpretation that Morck, Shleifer and Vishny (1988) give is that the increase of Tobin’s Q reflects the convergence of interest between the shareholders and the board. The decline of Tobin’s Q reflects the entrenchment of the board. Morck, Shleifer and Vishny (1988) state that the entrenchment does not come directly from the boards control of voting rights. They state that the decline in Tobin’s Q comes from the positive correlation between managerial
14 and company attributes and board ownership. This positive correlation attributes to the possibility of boards to entrench themselves.
2.5 Strategic selection
Executives which have some sort of entrenchment could have influence on the pay setting of the same executives. This is shown in the research done by Garvey and Milbourn (2006). They use a sample period from 1992 to 2001 and use data from the CRSP and
ExecuComp database. Garvey and Milbourn (2006) have a sample size of 6,262 observations and use ordinary least squares, two-stage, and piecewise linear regressions to conduct their research. They find that there is substantially less pay-for-luck when the luck is down, in contrast to when the luck is up. This effect also appears to be more prominent in situations in which the CEO has greater influence over the pay setting process (Garvey & Milbourn, 2006). The CEO has greater influence over the pay setting process, when the CEO is more entrenched.
Jenter and Kanaan (2015) use a final sample size of 3,042 companies with 31,185 company-year observations. Of this sample there were 3,365 CEO turnovers, of which 875 are classified as forced. The Factiva news database is used to classified the CEO turnover as forced or voluntary. The sample period is from 1993 to 2009. All companies in the final sample are from the ExecuComp database. Jenter and Kanaan (2015) use a two-stage hazard regression model and find that CEOs who underperform their peers are more likely to be fired than outperforming CEOs. This means that it is in the best interest of the CEOs that the peer companies perform worse than they do, otherwise the probability of being fired will increase significant.
Gong, Li and Shin (2011) find that companies are more likely to be selected as a peer company when the company underperforms the industry. This finding on peer selection reflects more rent seeking behaviour when looking at underperforming companies, because executives are probably more concerned about lowered compensation and potential
reputation loss than companies which are outperforming their peers. This rent seeking behaviour provides executives more incentives to manipulate the selection process of the peer companies (Gong, Li, & Shin, 2011). The higher incentive for underperforming
15 The study of Bizjak, Lemmon and Nguyen (2011) shows that companies are more likely to choose better performing and larger peers in comparison to potential peer
companies that have a worse performance and are smaller in size. However, Bizjak, Lemmon and Nguyen (2011) research about the selection of compensation peers. As previously mentioned , compensation benchmarking theory could have the opposite effect for relative performance evaluation. This means that for relative performance evaluation, it could be more likely that companies choose worse performing and smaller peer companies. This because it appears out of the study from Bizjak, Lemmon and Nguyen (2011) that companies tend to choose peer companies in the favour of the executives. Porac, Wade and Pollock (1999) also find, in line with the research by Bizjak, Lemmon and Nguyen (2011), that boards define peers in a self-protective way. Peer companies are more frequently taken out of other industries when the company performs poorly and the industry, in which the company is active, performs well (Porac, Wade, & Pollock, 1999).
While previous mentioned papers find the use of some sort of strategic selection by the boards, Denis, Jochem and Rajamani (2017) have a different finding. They examine whether and how weak shareholder support for say on pay votes influences changes in CEO compensation in companies that use weak-vote companies as their compensation peers. To conduct this research a difference-in-differences method with data from 1,061 companies which are in the S&P1500 and which resulted in 5,955 company-year observations between 2010 and 2013 is used. Denis, Jochem and Rajamani (2017) find that primary companies experiences significant relative changes in CEO compensation following weak shareholders support for the say on pay votes of companies which are a compensation peer. The primary company CEOs total compensation declines by 8% relative to the total compensation of the control companies. They also find that the primary companies which CEOs compensation is relative high are the ones which are more likely to have a relatively reduction of CEO
compensation in response to a weak say on pay vote on a peer company. Also Denis, Jochem and Rajamani (2017) find that primary companies do not drop weak-vote peers
disproportional from their compensation peer group. All these findings suggest that boards use compensation benchmarking for informational purposes instead of for opportunistic purposes. Denis, Jochem and Rajamani (2017) do place a side note that they find evidence that a subset of companies do not exhibit changes in the CEO compensation following a
16 weak say on pay vote on a peer company. This subset of companies are companies which are seen as most likely to have chosen their compensation peers for opportunistic reasons.
The paper of Bizjak, Kalpathy, Li and Young (2017) also studies whether or not the peer groups are selected in an opportunistic way. As discussed previously in section 2.2, the paper of Bizjak, Kalpathy, Li and Young (2017) is a working paper so the results should be considered with caution. To study whether or not peer groups are selected in an
opportunistic way, they compare the award payout of the actual peer group with three alternative sets of peer groups. The first alternative peer group is that with the highest propensity score. The second alternative peer group is that with companies which are similar in size and industry and the third alternative peer group is the compensation benchmark peer group. Using Monte Carlo simulations to calculate the expected award payouts Bizjak, Kalpathy, Li and Young (2017) find no evidence that custom peer groups are chosen in an opportunistic way. They find that the simulated payouts of the actual peer groups are lower relative to the three alternative peer groups. This shows that the peer group is not selected in an opportunistic way and therefore there is not a strategic selection in determining the relative performance evaluation peer group (Bizjak, Kalpathy, Li, & Young, 2017).
2.6 Hypotheses
To answer the research question, two hypotheses are formed. The first hypothesis relates to the determinants of peer group selection in relative performance evaluation. The second hypothesis relates to the part of the research question that asks whether or not low performing and entrenched companies change their peers.
The first hypothesis is called the efficient benchmarking hypothesis. Under the efficient benchmarking hypothesis it is stated that same industry membership, correlated stock performance, revenues, market-to-book ratio, market capitalization and same stock index membership are all determinants for being selected as a peer company. This
hypothesis is formed after the evaluation of previous literature which are discussed in the section 2.2and 2.3. The paper of Gong, Li and Shin (2011) is leading for this hypothesis, because they research in part the same topic as the efficient benchmarking hypothesis discusses. As previously discussed, Gong, Li and Shin (2011) find that the determinants under the efficient benchmarking hypothesis have a significant effect on whether or not a company
17 is selected as a peer company. The findings of the papers discussed in section 2.3 are also taken into account when formulating the first hypothesis. Most of the papers find that being in the same industry and having the same size are determinants for compensation peer selection. Bizjak, Lemmon and Nguyen (2011) also find that similar market-to-book ratio increases the likelihood of being selected as a compensation peer. Being a S&P500 company also increases the likelihood of being chosen as a compensation peer as found by Faulkender and Yang (2013).
The second hypothesis is called the strategic selection hypothesis. Under the strategic selection hypothesis it is stated that low performing and entrenched companies change their peers. A reason for this could be to make the company competitive better. If the RPE company selects new peers which perform slightly worse than the peers which are dropped out, then it will appear that the company isn’t performing that poorly. It also enables the executives to receive a higher bonus if the bonus is determinant on relative performance compared to the peer companies. The second hypothesis is formed after reviewing previous literature described section 2.5. As described in this section, Gong, Li and Shin (2011) find rent seeking behaviour when looking at the selection process. Garvey and Milbourn (2006) find substantially less pay-for-luck when the luck is down, in contrast to when the luck is up. They show that the CEO has greater influence over the pay setting process, when the CEO is more entrenched. Jenter and Kanaan (2015) find that CEOs who underperform their peers are more likely to be fired than outperforming CEOs. This result would give an entrenched CEO more incentive to strategically select the peers used for relative performance benchmarking. If the CEO select worse performing peers it would be easier to outperform the peers and so reduce the change of being fired. Porac, Wade and Pollock (1999) find that boards define peers in a self-protective way. All of this research would indicate that CEOs which are entrenched and so have an influence on the peer selection process would replace peers in times in which the company is performing poorly relative to its peers.
18
3. Methodology
3.1 Model to test the efficient benchmarking hypothesis
To test the efficient benchmarking hypothesis a logit regression model is used. The regression is:
The i stands for the RPE company and the j stands for the possible peer company of the RPE company. The t stands for the specific date that is linked to the data variable. This model is comparable to the model used by Gong, Li and Shin (2011).
3.2 Explaining the variables of the model to test the efficient benchmarking hypothesis Chosen as peer is a dummy variable that equals 1 if a company is the peer company for the RPE company and 0 if not. To create the chosen as peer variable, the peer companies of each RPE company is collected. The collection process of the peer companies will be discussed in the data section. Then for each selected peer company, a random company is selected as a peer which is not chosen as a peer company. So this means that for each RPE company the number of chosen peer companies equals the number of random unselected peer companies. The random function in Stata is used to randomize the unselected peer companies. The method for constructing the chosen as peer variable is also used by Gong, Li and Shin (2011). Same SIC2 is a dummy variable that equals 1 when the RPE company and the possible peer company have the same first 2 SIC numbers and 0 if not. The first 2 SIC numbers specify the industry which a company is active in. Corr(RPEReturn,PeerReturn) is the correlation of the stock return of the RPE company and the possible peer company for the last 5 years. RPERevenue_PeerRevenue is the absolute value difference in revenue of the previous year between the RPE company and the possible peer company. RPEMtB_PeerMtB is the absolute value difference in market-to-book ratio of the previous year between the RPE company and the possible peer company. RPEMKTCAP_PeerMKTCAP is the absolute value difference in market capitalization of the previous year between the RPE company and
19 the possible peer company. The three variables with absolute differences between the RPE company and the possible peer company are all in billions of dollars and are winsorized at the 1% level to account for outliers. S&P500 is a dummy variable that equals 1 if the possible peer company is in the S&P500 index and is 0 if not.
The variables same SIC2, corr(RPEReturn,PeerReturn), RPERevenue_PeerRevenue and RPEMKTCAP_PeerMKTCAP are also used by Gong, Li and Shin (2011). The variable RPEMtB_PeerMtB is used by the paper of Bizjak, Lemmon and Nguyen (2011), but is
modified to be in line with the absolute value difference as used by Gong, Li and Shin (2011). The variable chosen as peer and S&P500 is used by the paper of Faulkender and Yang (2013).
3.3 Model to test the strategic selection hypothesis
A logit regression is also used to test the strategic selection hypothesis. The regression is:
The i stands for the RPE company and the j stands for the possible peer company of the RPE company. The t stands for the specific date that is linked to the data variable. This model is a modified version of a model used by Denis, Jochem and Rajamani (2017).
3.4 Explaining the variables of the model to test the strategic selection hypothesis
Deselected as peer is a dummy variable that equals 1 if the peer company is dropped the next period and 0 if not. RPEReturn is the yearly buy-and-hold return of the previous year of the RPE company. PeerReturn is the yearly buy-and-hold return of the previous year of the peer company. The variable RPEReturn x PeerReturn is an interaction variable that
20 measures the interaction of the yearly buy-and-hold return of the previous year of the RPE company and the peer company. Entrenchment is a dummy variable that equals 1 if the RPE company has a classified board or/and has a CEO ownership between 5% and 25% in the previous year. The dummy variable entrenchment is 0 when the RPE company has not got a classified board and has not got a CEO ownership between 5% and 25% in the previous year. The variable Entrenchment x RPEReturn is an interaction variable that measures if the management of the RPE company is entrenched and the yearly buy-and-hold return of the previous year of the RPE company. The variable Entrenchment x PeerReturn is an interaction variable that measures if the management of the RPE company is entrenched and the yearly buy-and-hold return of the previous year of the peer company. Same SIC2 is a dummy variable that equals 1 when the RPE company and the possible peer company have the same first 2 SIC numbers and 0 if not. S&P500 is a dummy variable that equals 1 if the peer
company is in the S&P500 index and 0 if not. The variable peer company delist is a dummy variable that equals 1 if the peer company is delisted due to liquidation, bankruptcy, privatization or acquisition in the next year and 0 if not. RPERevenue_PeerRevenue is the absolute change in value difference in revenue of the previous two years between the RPE company and the peer company. The variable RPEMtB_PeerMtB is the absolute change in value difference in market-to-book ratio of the previous two years between the RPE
company and the peer company. RPEMKTCAP_PeerMKTCAP is the absolute change in value difference in market capitalization of the previous two years between the RPE company and the peer company. The three variables with absolute change differences between the RPE company and the possible peer company are all in billions of dollars and are winsorized at the 1% level to account for outliers.
Because the strategic selection hypothesis has not yet been researched by looking at deselecting a peer company, parts of different papers are used to construct the regression model. The variables Same SIC2, S&P500, RPERevenue_PeerRevenue, RPEMtB_PeerMtB and RPEMKTCAP_PeerMKTCAP are placed in the model, because they could be of interest when choosing a peer company as discussed in section 2 and 3.2. RPERevenue_PeerRevenue, RPEMtB_PeerMtB and RPEMKTCAP_PeerMKTCAP are slightly modified using the paper by Denis, Jochem and Rajamani (2017). The variable peer company delists is used by Denis, Jochem and Rajamani (2017). The variable entrenchment is constructed using the papers of Faleye (2007) and Morck, Shleifer and Vishny (1988). The variables deselected as peer,
21 RPEReturn, PeerReturn, RPEReturn x PeerReturn , Entrenchment x RPEReturn and
Entrenchment x PeerReturn are not constructed in line with any paper. This because previous research on this topic has not yet been done.
3.5 Important variables
In the model that is used to test the efficient benchmarking hypothesis, all the variables are of interest. This means that all the significant variables are used to test the hypothesis. The statistical significants will be determinant on the 10%, 5% and 1% level. The sign of the coefficients of the variables same SIC2, corr(RPEReturn,PeerReturn) and S&P500 are expected to be positive. The sign of the coefficients of the variables
|RPERevenue_PeerRevenue|, |RPEMtB_PeerMtB| and |RPEMKTCAP_PeerMKTCAP| are expected to be negative. The efficient benchmark hypothesis indicates that all the variables in the model are determinants for being selected as a peer company. The variables same SIC2, corr(RPEReturn,PeerReturn) and S&P500 are expected to have a positive coefficient, because a positive coefficients means that the variable increases the probability for a peer company to be selected and so is a determinant for being selected as a peer. The coefficient of the variables |RPERevenue_PeerRevenue|, |RPEMtB_PeerMtB| and
|RPEMKTCAP_PeerMKTCAP| are expected to be negative, because the larger the difference between the RPE company and the peer company the less likely the peer company is
selected as a peer.
In the model that is used to test the strategic selection hypothesis only a few
variables are of interest. The variables of interest are: RPEReturn, PeerReturn, RPEReturn x PeerReturn, Entrenchment, Entrenchment x RPEReturn and Entrenchment x PeerReturn. The statistical significants of coefficients will again be determinant on the 10%, 5% and 1% level. For the strategic selection hypothesis to be true, it will be expected that the coefficient of RPEReturn to be negative. This because when the RPE company performs well there would be no incentive to drop a peer company. The RPE company would probably outperform the peer company. The coefficient of the variable PeerReturn will be expected to be positive. When the peer company is performing very well the RPE company has more incentive to drop the peer company because it makes the RPE company performing poorly. The interaction variable RPEReturn x PeerReturn has an inconclusive expectation of the
22 coefficient. As explained previously, the variable RPEReturn on itself is expected to be
negative. However the variable PeerReturn on itself is expected to be positive. Therefore it is inconclusive to say which variable would dominate in the interaction variable. The
coefficient of the variable Entrenchment will be expected to be positive. If the board is entrenched it has more power to deselect a peer company if they want to. The coefficient of the variable Entrenchment x RPEReturn is expected to be negative. Although the coefficient of the variable Entrenchment is positive, the coefficient of the variable RPEReturn is
expected to be negative and is expected to have a greater influence on the dependent variable than the variable Entrenchment. This because when the RPE company is performing well, the company doesn’t have any incentive to deselect a certain peer. The entrenchment of the board and the deselecting of a peer will in theory only play a part when the RPE company is performing poorly. The coefficient of the variable Entrenchment x PeerReturn is expected to be positive. When the peer company is performing well, meaning a high peer return, the RPE company has an incentive to deselect the peer company. This because if a peer company is performing well, the RPE company seems to perform relatively poor in contrast if the RPE company would choose a peer company that is performing poorly. The variable entrenchment is also expected to have a positive effect on the deselecting variable as previously discussed, so the interaction variable Entrenchment x PeerReturn is expected to have a positive coefficient.
All variables discussed in this section needs to have at least a significants level of 5% to accept that the independent variable has an influence on the dependent variable.
3.6 Control variables
The model to test the efficient benchmarking hypothesis only controls for fixed effects. The fixed effects that are taken into account are industry fixed effects and year fixed effects. The first 2 sic numbers are used for the industry fixed effects.
The model to test the strategic selection hypothesis has multiple control variables. The control variables are placed in the regression, because all these variables could have an influence on determining a peer company and so could have an influence on deselecting a peer company. The variable Peer company delists is added as a control variable, because a peer company could be deselected as a peer solely on the fact that the peer company does
23 not exist anymore. The coefficients of the variables same SIC2 and S&P500 are both
expected to be negative. As mentioned in the literature review both variables would
increase the chance of being selected as a peer company and so should decrease the chance of being deselected as a peer company. The coefficient of the variable Peer company delists is expected to be positive. When a peer company delists of the stock exchange for what kind of reason, it is expected that the RPE company deselects the peer company. When a
company delists it is more difficult to get the data of that company or the company is bankrupt and ceases to exist. That is why a RPE company is more likely to deselect a peer company if it delists. The coefficients of the variables RPERevenue_PeerRevenue,
RPEMtB_PeerMtB and RPEMKTCAP_PeerMKTCAP are expected to be positive. This because when the size differences rise between the RPE company and the peer company, the RPE company is more willing to deselect the peer company. The model to test the strategic selection hypothesis also controls for industry fixed effects and year fixed effects. Just like in the model for testing the efficient benchmarking hypothesis, the first two SIC numbers are used for the industry fixed effects.
4. Data
4.1 Finding RPE companies and their peers
The monthly updated index constituents database of Compustat is used to select the RPE companies. A sample period from 1-1-2007 until 31-12-2011 is used. Then the RPE companies need to fulfil several criteria to be marked as RPE companies. The criteria are: (1) company needs to have a continuous presence in the S&P500 index during the sample period, (2) have a CIK code, (3) company uses RPE, (4) clearly indicate the name of the companies which are peers. After criteria 2, 322 companies are left in the sample for RPE companies. The CIK codes and names of these companies can be found in Appendix table A1. To fulfil the third criteria an excel file provided by dr. Torsten Jochem is used. This excel file loads SEC DEF 14A filings, also known as “Proxy Statements”, which are from the EDGAR database. The CIK codes are used to show the proxy statement of that company. Then the same method used by Gong, Li and Shin (2011) is used to evaluate if a company uses relative performance evaluation and so is a RPE company. A company is a RPE company if the
24 compensation is partly or completely based on the company’s performance relative to that of its peers. If this is not the case, the company is not seen as a RPE company. Further also criteria 4 needs to be fulfilled to get the full RPE companies sample. How the company peer names are found will be discussed in the last paragraph of this section. The descriptive statistics after using the 4 criteria can be found in table 1.
Table 1. Descriptive statistics on RPE companies and their peers.
Panel A shows descriptive statistics of the number of RPE companies in the sample. The total companies of 322 companies are found after implementing the first two criteria, namely (1) company needs to have a continues presence in the S&P500 index during the sample period and (2) have a CIK code. Whether or not a company is a RPE company is defined using two other criteria, namely (3) company uses RPE, (4) clearly indicate the name of the companies which are peers. Panel B shows descriptive statistics of the sub-sample of RPE companies. Minimum, maximum and mean stands for the number of peers used by a RPE company. The sample period is from 1-1-2007 until 31-12-2011.
Panel A: Frequency of RPE companies Total companies RPE companies Non-RPE companies Percentage of RPE companies 322 180 142 55.90%
Panel B: Peers characteristics of sample of only RPE companies Number of RPE
companies Minimum Maximum Mean
180 1 91 11.52
As can be seen in table 1 panel A, 180 companies of the 322 companies use relative performance evaluation. So of the sample of 322 companies, which is constructed after using the first two criteria, 55.90% of the companies use relative performance evaluation. These 180 RPE companies use on average 11.52 peer companies with a minimum of 1 peer company and a maximum of 91 peer companies. This can be seen in table 1 panel B. The CIK codes and the names of the 180 RPE companies can be found in Appendix table A2.
To find the peers, that the RPE companies use to benchmark, the excel file provided by dr. Torsten Jochem is used. Each peer of each RPE company is hand-collected and saved in the excel file using the DEF 14A filings. Certain key words are used to scan the proxy statements to filter out the peers names and to evaluate if the companies uses relative performance evaluation. These key words are partly used by Black, Dikolli and Hofmann (2015) and partly found while evaluating the proxy statements. The key words used are: (1) Peer, (2) Benchmark, (3) Comparison, (4) Competitive, (5) Performance, (6) Relative and (7) Comparator. Key words 1 to 4 are used by Black, Dikolli and Hofmann (2015). The other key words are found while evaluating the proxy statements. The peers that are named as peers for relative performance evaluation are used in the dataset. However, there are companies
25 which do state that they use relative performance evaluation but don’t mention the peers explicitly. They state only the compensation peers. Then it is assumed that the RPE peers are the same as the compensation peers. This is in line with the method of Black, Dikolli and Hofmann (2015). The peers are collected using the proxy statements filled between the start of year 2007 until the end of 2011.
4.2 Databases
For the remaining variables that are used in the models, different databases are used to collect the data. The CRSP Compustat merged Fundamentals annual database is used to collect the data for the variables SIC, Revenue, Market-to-Book ratio, and market
capitalization. To calculate the Market-to-Book ratio, the market value variable of CRSP Compustat merged Fundamentals annual database is used. For the book value the total common equity variable of CRSP Compustat merged Fundamentals annual database is used. For the dummy variable S&P500 the monthly updated index constituents database of Compustat is used. The CRSP monthly stock database is used to collect the stock prices returns of the companies. These stock prices returns are used to calculate the yearly buy-and-hold return of the companies. The CRSP monthly stock database is also used to get the data needed for the dummy variable of the peer company delist. For the entrenchment variable two different databases are use. For the CEO ownership the ExecuComp database is used and the ISS formerly Riskmetrics database is used for the classified boards. To see if a peer company is deselected the excel file provided by dr. Torsten Jochem is used. The peer company is deselected if it is a peer in the current year but is not anymore the next year. The monthly updated index constituents database of Compustat is used to get all the companies which could be a random unselected peer company. All the companies in this index between 2007 and 2011 are used as possible random unselected peer companies.
4.3 Summary statistics for the efficient benchmarking hypothesis
For the efficient benchmarking hypothesis the summary statistics for the whole sample can be seen in table 2. Table 2 shows that on average the half of the companies are chosen as a peer company. This is to be expected, because the design of the research is so that for each company chosen as peer a random company is selected. On average a quarter
26 of all the companies which are possible peer companies in this sample are in the same
industry active as the RPE company. On average the correlation of the stock return between the RPE company and its possible peer company is 0.3499. This shows that the same stock return behaviour is a bit low in the sample. A reason for this is that the random unselected peer companies have a low stock return correlation and so lowers the average as will be shown in table 4. Table 2 also indicates that the variable S&P500 has an average of almost a half. This shows that in this sample on average the half of the possible peer companies are in the S&P 500 index. Table 2 also shows that for all the variables the standard deviation is high. This indicates that there is a large data range for each variable. This can also been seen in the large differences between the minimum and maximum values of the variables. The number of observations varies depending on data availability.
Table 2. Summary statistics for the efficient benchmarking hypothesis whole sample.
This table shows the summary statistics for the efficient benchmarking hypothesis for the whole sample. Chosen as peer is a dummy variable that equals 1 if a company is used as a peer company. same SIC2 is a dummy variable that equals 1 if the RPE company and the possible peer company are in active in the same industry. corr(RPEReturn, PeerReturn) is the correlation of the stock return of the RPE company and the possible peer company for the last 5 years. |RPERevenue_PeerRevenue| is the absolute value difference in revenue of the previous year between the RPE company and the possible peer company. |RPEMtB_PeerMtB| is the absolute value difference in market-to-book ratio of the previous year between the RPE company and the possible peer company. |RPEMKTCAP_PeerMKTCAP| is the absolute value difference in market capitalization of the previous year between the RPE company and the possible peer company. The three variables with absolute differences between the RPE company and the possible peer company are all in billions of dollars and are winsorized at the 1% level to account for outliers. S&P500 is a dummy variable that equals 1 if the possible peer company is in the S&P 500. Number indicates the number of observations for the specific variable. The number of observations varies depending on data availability. The sample period is from 1-1-2007 until 31-12-2011.
Mean Median Minimum Maximum
Standard Deviation Number Chosen as peer 0.5285 1 0 1 0.4992 24529 same SIC2 0.2766 0 0 1 0.4473 24529 corr(RPEReturn, PeerReturn) 0.3499 0.3431 -1 1 0.2307 19247 |RPERevenue_PeerRevenue| 17.4400 7.5830 0.1113 131.6000 25.3200 23224 |RPEMtB_PeerMtB| 3.3650 1.2940 0.0167 60.6700 7.9230 19064 |RPEMKTCAP_PeerMKTCAP| 25.5900 11.4900 0.1921 180.9000 36.5800 19064 S&P500 0.4971 0 0 1 0.5000 24529
27 Table 3 and 4 show the summary statistics for the efficient benchmarking hypothesis for the two sub-samples of the selected and the random unselected peer companies. Table 3 is for the sub-sample of the peers which are selected as a peer company and table 4 is for the sub-sample of the random unselected peer companies. As can be seen when comparing the means of these two tables that the variables same SIC2, corr(RPEReturn, PeerReturn) and S&P500 are substantially different between the two sub-samples. This could indicate that these three variables are determinants for selecting peer companies for relative performance evaluation. The standard deviations are still high in both sub-samples. This again indicates that in the sub-samples the data range is large for each variable. This can also been seen in the large difference between the minimum and maximum value of each
variable. The number of observations varies depending on data availability.
Table 3. Summary statistics for the efficient benchmarking hypothesis for selected peers.
This table shows the summary statistics for the efficient benchmarking hypothesis for the sub-sample of only the possible peer companies which are selected as a peer. Chosen as peer is a dummy variable that equals 1 if a company is used as a peer company. same SIC2 is a dummy variable that equals 1 if the RPE company and the possible peer company are in active in the same industry.
corr(RPEReturn, PeerReturn) is the correlation of the stock return of the RPE company and the possible peer company for the last 5 years. |RPERevenue_PeerRevenue| is the absolute value difference in revenue of the previous year between the RPE company and the possible peer company. |RPEMtB_PeerMtB| is the absolute value difference in market-to-book ratio of the previous year between the RPE company and the possible peer company. |RPEMKTCAP_PeerMKTCAP| is the absolute value difference in market capitalization of the previous year between the RPE company and the possible peer company. The three variables with absolute differences between the RPE company and the possible peer company are all in billions of dollars and are winsorized at the 1% level to account for outliers. S&P500 is a dummy variable that equals 1 if the possible peer company is in the S&P 500. Number indicates the number of observations for the specific variable. The number of observations varies depending on data availability. The sample period is from 1-1-2007 until 31-12-2011.
Mean Median Minimum Maximum
Standard Deviation Number Chosen as peer 1 1 1 1 0 12963 same SIC2 0.4885 0 0 1 0.4999 12963 corr(RPEReturn, PeerReturn) 0.4501 0.4558 -0.4301 1 0.2076 11031 |RPERevenue_PeerRevenue| 15.2200 5.4410 0.1113 131.6000 24.8100 12030 |RPEMtB_PeerMtB| 3.1780 1.0910 0.0167 60.6700 7.9070 10759 |RPEMKTCAP_PeerMKTCAP| 24.3600 9.6590 0.1921 180.9000 36.6700 10759 S&P500 0.7505 1 0 1 0.4327 12963
28 Table 4. Summary statistics for the efficient benchmarking hypothesis for random unselected peers.
This table shows the summary statistics for the efficient benchmarking hypothesis for the sub-sample of only the random unselected peer companies which are not selected as a peer. Chosen as peer is a dummy variable that equals 1 if a company is used as a peer company. same SIC2 is a dummy variable that equals 1 if the RPE company and the possible peer company are in active in the same industry. corr(RPEReturn, PeerReturn) is the correlation of the stock return of the RPE company and the possible peer company for the last 5 years. |RPERevenue_PeerRevenue| is the absolute value difference in revenue of the previous year between the RPE company and the possible peer company. |RPEMtB_PeerMtB| is the absolute value difference in market-to-book ratio of the previous year between the RPE company and the possible peer company. |RPEMKTCAP_PeerMKTCAP| is the absolute value difference in market capitalization of the previous year between the RPE company and the possible peer company. The three variables with absolute differences between the RPE company and the possible peer company are all in billions of dollars and are winsorized at the 1% level to account for outliers. S&P500 is a dummy variable that equals 1 if the possible peer company is in the S&P 500. Number indicates the number of observations for the specific variable. The number of observations varies depending on data availability. The sample period is from 1-1-2007 until 31-12-2011.
Mean Median Minimum Maximum
Standard Deviation Number Chosen as peer 0 0 0 0 0 11566 same SIC2 0.0392 0 0 1 0.1940 11566 corr(RPEReturn, PeerReturn) 0.2153 0.2097 -1 1 0.1876 8216 |RPERevenue_PeerRevenue| 19.8400 10.0800 0.1113 131.600 25.6400 11194 |RPEMtB_PeerMtB| 3.6080 1.5910 0.0167 60.6700 7.9380 8305 |RPEMKTCAP_PeerMKTCAP| 27.1900 13.7600 0.1921 180.9000 36.4100 8305 S&P500 0.2131 0 0 1 0.4095 11566
4.4 Summary statistics for the strategic selection hypothesis
For the strategic selection hypothesis the summary statistics can be seen in table 5. As can be seen in table 5, only a small portion of peer companies are on average deselected as a peer company. On average 8% of the peers is deselected as a peer in the sample. Table 5 also shows that on average 3% of the peer companies delist of the stock exchange due to liquidation, bankruptcy, privatization or acquisition. Also on average around 33% of the RPE companies are entrenched. The average percentages can be stated, because these variables are all dummy variables and so only have a value of 1 or 0. The average RPE return and peer return are around 8.9% yearly which is close to each other. The means of the variables same SIC2 and S&P500 match the means of those variables in table 3. This is to be expected, because when using the data of the strategic selection hypothesis only the selected peer companies data is used. Table 3 also shows the data of only the selected peer companies. Just like with the data of the efficient benchmarking hypothesis the standard deviation reported in table 5 are high. This means that for the data for the strategic selection
hypothesis the data range for each variable is large. This can also been seen in the difference between the maximum and minimum value of each variable. The number of observations varies depending on data availability.
29 Table 5. Summary statistics for the strategic selection hypothesis.
This table shows the summary statistics for the strategic selection hypothesis. Deselected as peer is a dummy variable that equals 1 if the peer company is dropped the next period. RPEReturn is the yearly buy-and-hold return of the previous year of the RPE company. PeerReturn is the yearly buy-and-hold return of the previous year of the peer company. The variable RPEReturn x PeerReturn is an interaction variable that measures the interaction of the yearly buy-and-hold return of the previous year of the RPE company and the peer company. Entrenchment is a dummy variable that equals 1 if the RPE company has a classified board or/and has a CEO ownership between 5% and 25% in the previous year. The variable Entrenchment x RPEReturn is an interaction variable that measures if the management of the RPE company is entrenched and the yearly buy-and-hold return of the previous year of the RPE company. The variable Entrenchment x PeerReturn is an interaction variable that measures if the management of the RPE company is entrenched and the yearly buy-and-hold return of the previous year of the peer company. same SIC2 is a dummy variable that equals 1 when the RPE company and the possible peer company have the same first 2 SIC numbers. S&P500 is a dummy variable that equals 1 if the peer company is in the S&P500 index. The variable peer company delist is a dummy variable that equals 1 if the peer company is delisted due to liquidation, bankruptcy, privatization or acquisition in the next year. RPERevenue_PeerRevenue is the absolute change in value difference in revenue of the previous two years between the RPE company and the peer company. The variable RPEMtB_PeerMtB is the absolute change in value difference in market-to-book ratio of the previous two years between the RPE company and the peer company. RPEMKTCAP_PeerMKTCAP is the absolute change in value difference in market capitalization of the previous two years between the RPE company and the peer company. The three variables with absolute change differences between the RPE company and the possible peer company are all in billions of dollars and are winsorized at the 1% level to account for outliers. Number indicates the number of observations for the specific variable. The number of observations varies depending on data availability. The sample period is from 1-1-2007 until 31-12-2011.
Mean Median Minimum Maximum
Standard Deviation Number Deselected as peer 0.0801 0 0 1 0.2714 13027 RPEReturn 0.0890 0.0969 -0.9194 4.146 0.3746 13027 PeerReturn 0.0880 0.0705 -0.9828 13.65 0.4242 13027 RPEReturn x PeerReturn 0.0842 0.0254 -0.6330 7.572 0.2288 13027 Entrenchment 0.3256 0 0 1 0.4686 13027 Entrenchment x RPEReturn 0.0228 0 -0.7989 1.401 0.2212 13027 Entrenchment x PeerReturn 0.0211 0 -0.9780 13.649 0.2751 13027 same SIC2 0.4880 0 0 1 0.4999 13027 S&P500 0.7514 1 0 1 0.4322 13027
Peer company delists 0.0299 0 0 1 0.1704 13027
∆|RPERevenue_PeerRevenue| 0.5691 0.1431 -25.31 28.73 5.7370 8533
∆|RPEMtB_PeerMtB| 0.0345 0 -23.32 25.59 4.4900 7538
∆|RPEMKTCAP_PeerMKTCAP| -0.0492 0.0933 -55.70 45.16 12.6500 7538
5. Results
5.1 Efficient benchmarking hypothesis
Table 6 shows the results of multiple regressions done to test the efficient
benchmark hypothesis. The method used to perform the regressions is described in section 3.1.Each regression uses the same variables, however the fixed effects differ between each regression. Yearly fixed effects are used to capture yearly effects that might occur and the RPE industry fixed effect are used to capture industry effects for the RPE companies that also might occur.
As mentioned in section 3.5 all variables are of interest and the coefficients are expected to be positive. In table 6 the fourth model is of most interest, because in that model both fixed effects are taken into account. The result of this model indicates that the
30 variables same SIC2, corr(RPEReturn,PeerReturn) |RPEMtB_PeerMtB| and S&P500 have a significance level of 1% and the coefficients are positive. This indicates that all these variables have an influence for being chosen as a peer company. This result is partly in line with the research done by Gong, Li and Shin (2011), Bizjak, Lemmon and Nguyen (2011) and Faulkender and Yang (2013). All the three papers find that the variables used in the model to test the efficient benchmark hypothesis have an influence for being chosen as a peer
company. This means that the findings for the variables |RPERevenue_PeerRevenue| and |RPEMKTCAP_PeerMKTCAP| are not in line with these three papers.
Looking at the coefficients in table 6 model 4, it can been seen that the variable same SIC2 has a positive coefficient. This means that companies which are active in the same industry as the RPE company, have a higher probability to be chosen as a peer company for relative performance evaluation. The variable same SIC2 is a dummy variable that equals 1 if the RPE company and the possible peer company are active in the same industry. The
variable corr(RPEReturn,PeerReturn) has also a positive coefficient. So a higher correlation of stock return in the last 5 years between the RPE company and a possible peer company increases the likelihood of being selected as a peer company. The variable
|RPEMtB_PeerMtB| is constructed by taking the absolute difference in market-to-book ratio of the previous year between the RPE company and the possible peer company. The variable has a positive coefficient as can been seen in table 6 model 4. This means that the larger the difference between the revenues of the RPE company and the possible peer company, the larger the probability is that the possible peer company is chosen as a peer company for the RPE company. The coefficient of the variable S&P500 is positive. This means that if a possible peer company is in the same index as the RPE company, namely the S&P500, the probability of being selected as a peer company increases. The variable S&P500 is a dummy variable that equals 1 if the RPE company and the possible peer company are both in the S&P500. The findings of both significance and sign of coefficient of the variables of same SIC2, corr(RPEReturn,PeerReturn) and S&P500 are in line with the results of Gong, Li and Shin (2011), Porac, Wade and Pollock (1999), Bizjak, Lemmon and Nguyen (2011), Faulkender and Yang (2013) and Faulkender and Yang (2010). The positive coefficient of the variable
|RPEMtB_PeerMtB| is not in line with these three papers.
In table 6 model 4 it appears that the variables |RPERevenue_PeerRevenue| and |RPEMKTCAP_PeerMKTCAP| are not significant on a 10% level and therefore the sign of the
31 coefficient cannot be interpreted. The variable |RPERevenue_PeerRevenue| is constructed by taking the absolute difference between the revenue of the RPE company and the possible peer company. The variable |RPEMKTCAP_PeerMKTCAP| is constructed by taking the
absolute difference in market capitalization of the previous year between the RPE company and the possible peer company. The findings of the insignificance of both variables would contradict the findings of Gong, Li and Shin (2011) and Bizjak, Lemmon and Nguyen (2011). However the findings of the variables |RPERevenue_PeerRevenue|, |RPEMtB_PeerMtB| and |RPEMKTCAP_PeerMKTCAP| have different findings when repeating the regressions using different random unselected peer companies. This means that a conclusion for these three variables cannot be drawn using only the findings of table 6. In section 6there will be an elaborate discussion about the findings of repeating the regressions using multiple different random unselected peer companies.
Overall, it can be stated that the findings illustrated in table 6 are in line with the efficient benchmark hypothesis for three determinants. This shows that RPE companies chose peer companies which share similar company characteristics. It should however be noted that three variables have yet inconclusive results.
