inside the black box: the role and composition of

8/14/2019 Inside the Black Box: The Role and Composition Of

1/28

Inside the Black Box: The Role and Composition of

Compensation Peer Groups *

Michael Faulkender Olin School of Business

Washington University in St. Louis

Jun Yang Kelley School of Business

Indiana University

* We would like to thank Richard Mahoney (retired CEO from Monsanto Co.) and seminar participants atWashington University in St. Louis. We also thank Cassandra Marshall and Raj Mistry for their researchassistance. All errors are ours. Email: [email protected] Email: [email protected]


2/28

Inside the Black Box: The Role and Composition of

Compensation Peer Groups

Abstract

This paper documents the features of compensation peer groups and demonstratesthat they play a significant role in determining CEO compensation. Anecdotally, weknow that compensation peer groups have had a growing role in determining executivecompensation but only recently have firms begun voluntarily disclosing the members of these peer groups. To empirically test their role, we hand-collect a sample of 83 of theS&P 500 firms that provided explicit lists of compensation peer firms in their fiscal 2005disclosures. Results show that inclusion of the groups median compensation more thantriples the portion of the variation in CEO cash compensation that can be explained,dominating measures such as size and firm performance. The average peer group has

more than eleven firms in it with just over half of them coming from the same 3-digit SICas the firm. Univariate analysis suggests that firms forego lower paid potential peers intheir same industry in favor of higher paid peers outside of their industry whenconstructing the peer groups. In multivariate regression analysis, this result carriesthrough as we find that even after controlling for industry and relative size, peer groupcomposition is significantly affected by the level of compensation of the potential peers.Firms appear to select high paid peers as a mechanism to increase CEO compensationand this effect is strongest in firms with low GIM index values, low E-scores, and low

blockholder ownership. We conclude that in firms with weak internal governance, CEOsare most able to create benchmarks (compensation peer group compositions) that helpgenerate higher compensation for themselves. Given that disclosure of peer group

composition had until recently been voluntary, our results are likely to underestimate theextent to which peers are selected by characteristics seemingly unrelated to managerial performance or their reservation wage.


3/28

1. Introduction

Recent growth in CEO compensation, especially the dramatic increases for top

paid CEOs, have led many to question whether CEOs have too much influence over their

own compensation. The pay package of $187 million for former New York Stock

Exchange Chairman Richard Grasso, and the $210 million golden parachute for the

ousted, and arguably mediocre performing, former Home Depot CEO generated notable

press coverage and have led us to wonder who sets CEO pay. The academic literature on

this issue is exploding, but has not reached a consensus. Many view the pay increases as

a sign of CEOs' abuse of power,1

but others argue that the compensation simply reflectsmarket equilibrium where the board optimally sets up CEO pay. 2

Theoretically, the pay setting process is quite transparent in the US. For a publicly

traded company, the initial pay recommendations typically come from the companys

human resources department, often working in conjunction with outside compensation

consultants. If accepted by the compensation committee, the recommendations are then

passed to the full board of directors (BOD) for approval. This process seems at least to

provide the management with opportunities to influence CEO pay via the initial

recommendations. 3 To address this potential influence, the NYSE instituted a new rule

that took effect in 2003 that bans CEOs of NYSE-listed firms from sitting on

compensation committees as well as from choosing external consultants. However, even

with this new rule in place and independent directors who approach their jobs with

1 See, for example, Bebchuk and Fried (2003, 2004), Bertrand and Mullanaithan (2001).2 See, for example, Murphy and Zabojnik (2004), Oyer (2004), Baranchuk, MacDonald and Yang (2006),and Gabaix and Landier (2006).3 The empirical evidence on whether CEOs influence their pay setting is somewhat mixed. Focusing on therole of compensation committees, OReilly, Main, and Crystal (1988), Main, OReilly, and Wade (1995),and Newman and Mozes (1997) suggest the existence of the influence; while Anderson and Bizjak (2003)suggest the opposite.

1


4/28

diligence, intelligence, and integrity, we are still likely to observe board actions that tend

to favor the CEO given a range of market data on competitive pay levels (see Murphy

1999).

To further enable compensation transparency, the SEC has issued a new

requirement that comes into effect for fiscal years ending on or after December 31, 2006:

The Compensation Committee report should disclose the nature of thegroup with which the Committee is comparing the registrant'scompensation (e.g., Fortune 100 companies), the extent to which it differsfrom the peer group and where in the range established by that comparisonthe issuer targets its compensation (e.g., high, median, or low end of therange). Where different competitive standards are used for different

components of the pay package, that should be made clear.

In this study, we seek to further our understanding of executive compensation by

examining the role and composition of these compensation peer groups. Using hand-

collected data from SEC filings of the list of these peer groups for S&P 500 firms in

2005, we begin by documenting the extent to which the level of compensation of the peer

group members does indeed explain observed compensation amounts. 4 When added to

our traditional measures of compensation, how much incremental explanatory power

does peer group compensation have? The second goal is to document the characteristics

of the peer group. We would expect that firms in the same industry and of similar size

would be obvious peer group members and this paper examines whether indeed that is the

case. Or, are there factors aside from relative size and industrial focus that explain peer

group membership, and therefore influence overall managerial compensation?

4 Some firms began voluntarily disclosing their compensation peer groups as early as 2004 but not until2005 were there a sufficient number of them reporting such that an empirical examination of these groupscan be conducted. Because such disclosure is voluntary, we will necessarily need to address potentialsample selection issues. We elaborate in the Data Description section.

2


5/28

We find that the median and 75 th percentile of compensation for the peer group do

generate significant incremental explanatory power in understanding cross-sectional

variation in observed CEO pay. The inclusion of this measure in our regression of total

cash compensation (salary and bonus payment) eliminates the statistical significance of

the estimated coefficient on our measure of firm size and the adjusted R-squared more

than triples after adding these variables to the specification. Examining the composition

of these peer groups in light of this economically significant role that these peer groups

play, we see that firms in the same industry and generally larger in size are more likely to

be included in the compensation peer group. However, we also find that even after controlling for industry and relative size, the level of compensation of the potential peer

is also statistically significant, suggesting that factors seemingly unrelated to the CEOs

reservation wage appear to also affect peer group composition. 5 In other words,

compensation committees seem to be endorsing compensation peer groups that include

unrelated firms because such firms would potentially ratchet up the level of pay for the

CEOs. This result complements the results of Bizjak, Lemmon and Naveen (2007) that

CEOs whose pay is below the median pay level of their counterparts in firms of similar

size and industry receive raises that are larger in both percentage and dollar terms.

Moreover, we examine whether corporate governance affects the make-up of the

compensation peer group. There is an extensive literature on measures of corporate

governance (notably Gompers, Ishii, and Metrick (2003) and Bebchuk, Cohen, and Ferell

(2004)) and the influence of governance on firm valuation and stock performance.

However, people tend to use the GIM-index and E-index in a more general context as

5 Possibly to justify selection of such firms as members of the compensation peer group, firms often statethat they choose firms with which they compete for managerial talents.

3


6/28

proxies for corporate governance. We find that as the external governance worsens (when

the GIM-index and E-index get higher), firms tend to select compensation peers with

lower pay, after controlling for the effects of industry and size. This is in line with the

notion that a weaker market for corporate control may be associated with stronger boards,

and it is the compensation committee of the board which ultimately sets CEO pay. 6 Using

large block holdings in place of these measures of external governance, we find that in

the presence of greater incentives to monitor, resulting from larger block holdings, there

is less sensitivity of peer group formation to the compensation of the CEOs at the

potential peer firms.The rest of the paper is organized as follows. Section 2 details the empirical

strategy that we will follow in exploring the role of compensation peer groups. The data

that we use is outlined in Section 3. The role of compensation peer groups in explaining

observed CEO pay is covered in Section 4 while the factors determining the composition

of these peer groups are discussed in Section 5. Section 6 concludes.

2. Empirical Strategy

We proceed in three steps. The first is to provide basic statistics on peer group

composition and to analyze the potential self-selection bias inherent in the fact that

compensation peer group disclosure was voluntary in 2005. Because only 83 of the 498

firms in the S&P 500 disclose their compensation peer group, how representative are the

firms that disclose the compensation peer group relative to those that choose not to?

6 Cyert, Kang, and Kumar (2002) examine, both theoretically and empirically, the strategic role of theBOD in setting up CEO pay and the impact of potential takeovers. In equilibrium, internal governance bythe BOD and the external takeover threat act as substitutes in constraining the management's profligacy of awarding equity-based compensation to itself. In a more recent empirical study, Gillan, Hartzell, and Starks(2007) present evidence that internal corporate governance (via board monitoring) and external corporategovernance (via takeovers) appear to serve as substitutes in disciplining managers.

4


7/28

Recognize though that because disclosure is voluntary, we would expect that firms that

select compensation peer groups that would be more difficult to explain to shareholders

would be the firms less likely to disclose the members of the peer group. As a result, we

likely underestimate the extent to which factors seemingly unrelated to CEO performance

actually impact CEO compensation.

The second step is to examine the incremental power generated from including

peer compensation in a regression of CEO compensation on firm characteristics

previously documented in the literature as explaining observed compensation. We begin

by running a baseline specification on the 83 CEOs compensation for which we have thenames of the compensation peer group. Specifically, we follow Bizjak, Lemmon and

Naveen (2007) and run the following regression:

CEO Compensation t = + 1*log(sales t) + 2*ROA t + 3*ROA t-1

+ 4*StockRet t + 5*StockRet t-1 + 6*Volatility +

where our measures of compensation will separately be total compensation (TDC1) and

total cash compensation (TCC). As a measure of the size of the firm, we take the natural

log of the firms sales (COMPUSTAT Item #12) in the corresponding fiscal year and

ROA is defined as EBIT (item #13) divided by book assets (item #6). Also included is

the performance of the firms stock over the fiscal year (ExecuComp item TRS1YR) and

the volatility of the firms stock over the previous 60 months (ExecuComp item

BS_VOLAT).

Once we have the initial values estimated, we then add a variable containing

either the median or 75 th percentile compensation of the peer group in the previous fiscal

year (consistent with practice and to ensure availability at the time compensation is

5


8/28

determined) into the specification. Our interest in the results extends beyond merely

interpreting the coefficient, but also to looking at the increase in R-squared that results

from this measures inclusion as well as the impact it has on the other coefficients in the

regression.

Given the large impact the peer compensation has, and therefore recognizing that

the important economic question is the composition of the compensation peer group, our

focus turns to an examination of the determinants of membership in that group. What are

the factors that determine whether or not a firm is included in the compensation peer

group? To conduct such an examination, one has to not only have the list of firmsselected for peer group membership but also those not chosen. While there are more than

5000 firms listed on COMPUSTAT in 2005 that are arguably potential peers, we limit

ourselves to the firms in the S&P 500 during 2005, as these are the potential firms that

are of similar size and visibility, as well as the fact that we have compensation data for

this subset of potential peer group members.

We then run a probit regression of whether the potential peer is indeed included in

the corresponding firms compensation peer group on a baseline set of controls that have

been previously documented to explain cross-sectional variation in compensation.

Specifically, we include whether the potential peer is in the same industry as the

underlying firm and the relative size differences between the potential peer and the firm. 7

We then add variables that are unlikely related to the extent to which the potential firms

are particularly comparable but that are related to potential malfeasance. If the firm

wanted to raise the CEOs compensation but still justify that the CEO is making the

7 In estimating the standard errors, we follow Petersen (2006) and cluster them at the firm level, arguingthat errors in estimating peer group inclusion are likely to be correlated for a particular firm.

6


9/28

median of his peers, the solution is to select peers that have relatively high compensation

themselves. Therefore, we add a variable capturing the potential peer firms previous

year compensation. If the coefficient corresponding to such a variable were found to be

significantly positive, it would suggest that even after controlling for size and industry,

some peers are chosen because they would raise the median for the group, justifying

higher CEO pay. Finally, we interact this sensitivity to potential peer pay with measures

of governance and CEO power to see whether or not such peer group selection is more

acute in exactly the firms where the CEO is more likely able to successfully extract rents

for himself.

3. Data Description

Our primary dataset was generated by hand-collecting the names of the

compensation peer groups for the members of the S&P 500 in 2005 off of their SEC

DEF-14A filings that are available on EDGAR. Of those 498 firms (Fannie Mae does not

have an available CIK number and Apollo Group Inc. does not provide a DEF-14A

statement), 76 of them gave a detailed list of the firms that are in the compensation peer

group. For example, Dynegy Inc. stated:

We believe that these surveys, together with our independentcompensation consultant's analysis of the proxy data for our peer companies, provide a comprehensive compensation competitivenessevaluation. Our peer group for the fiscal year ended December 31, 2005,which we refer to as the 2005 Peer Group, comprises AES Corporation;Calpine Corporation; Duke Energy Corporation; El Paso Corporation;

NRG Energy, Inc.; and Reliant Energy, Inc.

Another 7 stated that their peer group was comprised of exactly the firms that made up a

particular index. For example, Quest Diagnostics, Inc. stated:

7


10/28

In 2005, the Committee evaluated the competitiveness of senior management's total compensation relative to the pay of executives at a

peer group comprising of the Standard & Poors' 500 HealthcareEquipment & Services Index, the same peer group used for totalshareholder return comparison purposes in the performance graph shown

on page 35.

In other words, 83 of the 498 S&P 500 firms provided an explanation that enabled us to

determine exactly which firms were in the peer group, and that list of the 83 firms is

provided in Table 1.

The disclosures for the other 415 ranged, for example, from providing no

information to stating that they are firms of similar size to stating that some of them

belonged to a particular index but provided no information on the rest of the firms, to

giving an incomplete list. Due to the variation in disclosures, we need to estimate the

extent to which sample selection may bias the results of our examination. We therefore

merge our hand-collected data with both ExecuComp and COMPUSTAT to retrieve

information on executive compensation as well as various accounting variables for both

the firms themselves and the peers.

There are some significant differences between those that provide full disclosure

and those that do not, as can be seen by the summary statistics provided in Table 2.

Disclosing firms are larger in size and earned higher stock returns during fiscal year 2005

so not surprisingly, the CEOs of these firms were paid an average of $1.33 million more

in salary and bonus and $3.81 million more in total compensation (including option

grants) than non-disclosers. There does not appear to be significant differences in the

GIM-Index and E-index measures of governance for reporting versus non-reporting

firms, although non-disclosing firms do appear to have greater block-holder ownership

than firms that report their compensation peer groups. In terms of industry representation,

8


11/28

commercial banks (SIC 6020), pharmaceutical preparations (SIC 2834), and petroleum

refining (SIC 2911) have the highest numbers of disclosing firms.

To determine whether there is a significant difference in compensation between

those disclosing peer group representation and those not disclosing, we ran a regression

of CEO compensation on size, the Fama-French 12 industry classifications, and the

firms stock return plus a dummy variable for whether or not full disclosure occurred.

The economic interpretation of the coefficient corresponding to the full disclosure

dummy variable is the incremental compensation associated with having provided

complete information on the members of the compensation peer group. In unreportedresults, we find that the estimated coefficient is positive but statistically insignificant,

indicating that the disclosing group may be reasonably representative of the entire S&P

500. Our prior is that if anything, those fully disclosing are the ones least likely to be

manipulating the compensation process since it is easier to question the process of

determining CEO pay when more information is provided. The firms most likely to be

excessively paying their chief executives would likely be the firms least willing to

provide the names of the compensation peer group and then be in a position to have to

justify such a group. We therefore believe that our results likely extend beyond the 83

firms for which we have complete data and if anything, under-estimate the extent to

which compensation practices may be manipulated through the selection or compensation

peer groups.

Our evaluations of the effects of governance on peer group composition utilize

three common metrics of corporate governance, the GIM index, the E-index, and the

percentage of shares owned by blockholders. Gompers, Ishii, and Metrick (2003)

9


12/28

construct the GIM index that incorporates 24 provisions of takeover defenses followed by

IRRC whereas Bebchuk, Cohen and Farrell (2004) limit their entrenchment index ( the E-

index ) to six provisions that seem to be relevant for valuation and stock returns.

Blockholder data is acquired from Compact Disclosure and a blockholder is defined as a

shareholder who owns at least five percent of the outstanding shares. The ownership

shares are then aggregated for all of the blockholders.

4. Role of Peer Compensation on CEO Pay

Our primary objective is to understand the role of compensation peer groups onthe observed level of CEO pay so we begin with a baseline estimation of the determinants

of the level of compensation for CEOs among the firms for which we have peer group

information. As can be seen by the results located in the first column of Table 3, larger

firms that generate high stock returns have higher total compensation, consistent with

results previously documented in the literature. Approximately 24% of the cross-

sectional variation in pay is explained by the baseline model.

The results from adding the median level of compensation for the members of the

peer group to the regression specification, located in column 2, show that indeed peer

compensation is an important consideration in understanding the level of CEO pay. The

coefficient itself is highly significant, statistically at better than the one percent level, and

economically it suggests that the CEO of the corresponding firm earns an extra $1.14 for

each dollar increase in the median compensation among the peers, all else equal. In

addition, notice that the adjusted R-squared of the regression nearly doubles to 42.8% and

that the estimated coefficient on size is no longer statistically significant and has fallen in

10


13/28

magnitude by 84.4%. Obviously that does not mean that size is unimportant in CEO pay

since it will certainly be a factor in choosing the peers. However, what the results to

indicate is that size does not play a role once its effect on choosing members of the

compensation peer group has been controlled for. In column 3, we repeat the analysis

using the 75 th percentile of CEO compensation for the peer group and find results similar

to for the median. The adjusted R-squared is a little smaller and the coefficient suggests

that for a $1 increase in the 75 th percentile of peer compensation, the corresponding

CEOs pay increases by $0.72.

We repeat our analysis looking instead at just salary and bonus (total cashcompensation) instead of total compensation and report these results in Table 4. The

baseline specification indicates that size seems to play the dominant role with stock

return having a statistically insignificant effect on salary and bonus. Given the lack of

significance of the variables included in the specification, it is not surprising to see that

only 8.3% of the cross-sectional variation is explained. However, when we add the

median salary and bonus from the previous fiscal year for the compensation peer group,

we see that variable again being strongly significant, both statistically and economically.

The coefficient indicates that for an additional $1 of median salary and bonus in the peer

group, the corresponding CEOs pay increases by nearly $0.90. Statistically, the

coefficient is significant at better than one percent and the adjusted R-squared of the

regression more than triples, with 27.6% of the cross-sectional variation now explained.

In addition, the one variable, size, that was significant in the baseline specification has

declined in magnitude by more than 77% and is no longer statistically significant. Using

the 75 th percentile of lagged salary and bonus for the peer group, we find very similar

11


14/28

results statistically and a coefficient consistent with CEO salary and bonus rising by

$0.78 for a $1 increase in the 75 th percentile of total cash compensation for the peer

group.

Overall, the results indicate that firms are indeed following a compensation policy

in which a set of peers are chosen and then the board determines CEO compensation

based upon the observed salary and bonus of that peer group. When included in

regressions explaining compensation, these variables dominate all of the other controls

that have previously been used to explain CEO pay and the explanatory power of the

model increases dramatically. Therefore, explaining CEO compensation requiresuncovering the factors that determine the selection of the compensation peer group. That

is what we explore next.

5. Selection of Compensation Peer Groups

Given the large role that these peer groups have in understanding observed

executive compensation, we begin with some summary statistics on the composition of

these peer groups. As provided in Table 5, the average peer group is comprised of more

than eleven firms, just over half of them in the same 3-digit SIC as the firm itself. On

average, firms choose peers that are larger than themselves, though the data actually has a

negative skew as seen by the median being larger than the mean. 8 Since size and industry

have previously been documented to predict compensation, as well as the theoretical

argument that the outside opportunity for a CEO would likely be a senior position in a

firm of similar size and probably in the same industry, it is not surprising to see that these

8 General Electric is one of the 83 firms in the sample and the largest contributor in our data to the negativeskew. There are few firms that have sales larger than GE, meaning that its peers will necessarily be smaller and some of them are particularly small, hence the large negative value for that measure.

12


15/28

are important elements to examine when evaluating the make-up of these groups. The

median total compensation for the average peer group is $10.77 million, with $3.765

million of that in the form of cash compensation.

Looking at some univariate results for characteristics of firms chosen to be in the

compensation peer group, we break up the potential peers into four categories based upon

two measures: whether or not they are selected for the peer compensation group and

whether or not the potential peer is in the same 3-digit SIC as the firm. As demonstrated

by the results in panel B, we examine 41,549 potential firm-peer pairs. Since there are

nearly 500 potential peers in the sample and the average peer group has eleven firms,most of the potential peers will not end up being peers. Consistent with the earlier

results, a large fraction of the firms chosen (44%) are in the same industry and 37% of the

time, another S&P 500 firm in the same industry is chosen to be part of the compensation

peer group.

Aside from the industry break-down though, there are some interesting patterns

that emerge with regard to the compensation at the potential peers in the previous year.

The table provides mean and median total compensation for each of the four categories in

panel B with the same statistics for cash compensation in panel C. If we look at the

potential peers outside of the firms industry that are selected as peers (upper right

quadrant), these firms have the highest compensation when measured at both the mean

and median values. In contrast, firms in the same industry that were not chosen as peers

(lower right quadrant) have the lowest total compensation of the four categories and

second lowest cash compensation. In other words, at least based upon univariate analysis,

the selection of potential firms for the compensation peer group seems to favor higher

13


16/28

paid firms outside of their industry over lower paid potential peers that belong to the

same industry.

To determine whether these univariate results are robust to other controls, we

proceed by conducting multivariate regression analysis, starting with a baseline

specification containing just size and whether the firm and potential peer are in the same

industry. As shown by the results in Table 6, firms with the same three-digit SIC code

and larger firms are the ones most likely to be chosen for the peer compensation group.

We also examined industry classification at the two- and four- digit levels, which were

also statistically significant at better than one percent, but found that three-digit industrymeasurement had the strongest statistical results. This specification also includes the

differences in the natural log of the size of the potential peer and the firm, but the results

hold when we instead use the raw difference (without logs) as well as just the overall

level of peer size. So, even though firm size does not seem to have a large effect on

compensation once we control for the pay level of the peer group, the size of the potential

peer relative to the size of the firm does play an important role in the choice of which

firms are chosen for the peer group.

We now turn to an examination of another factor that may influence observed

CEO compensation that is arguably more related to rent-seeking rather than the

reservation wage of the CEO. Because compensation is so related to the level for the

peer group, the way to increase CEO pay is to select firms for the peer group based upon

the level of compensation those firms pay their CEOs. Therefore, we add to the baseline

specification a variable measuring the total compensation paid in the previous fiscal year

to the CEO at the potential peer.

14


17/28

The results from adding that additional variable, contained in column 2 of Table

6, indicate that the level of CEO compensation at the potential peer company does indeed

have a significantly positive effect on the likelihood of including that firm in the peer

group. Interestingly, the results are stronger when we use the raw dollar amount of the

potential peer compensation rather than its natural log (regressions unreported).

Normally, the benefit of taking logs is a reduction in the outliers in the data, which often

generates better fit. In this case, the firm would find it optimal to choose outliers as they

are the ones that would move up the peer measure of compensation the most (assuming

that enough of them were chosen). Therefore, it does appear that allowing for greater skew to remain in the data actually better captures the economic outcome that being in

the upper tail of the distribution increases the likelihood of being selected for the peer

group.

Considering that even after controlling for size and industry that CEO pay at the

potential peer firm is significant, we now turn to the characteristics of the firms in which

such sensitivity is higher by interacting the effect of potential peer compensation with

measures of governance. The idea is that if indeed CEOs are using peer group

membership to boost his own pay, we would expect that to most be the case in firms in

which the CEO is most entrenched or where the CEO has the most power. We begin

with the GIM index and again find that potential peer compensation is statistically

significant but that the sensitivity to potential peer compensation declines as the firms

GIM index increases. So, we see that the potential peers CEO compensation is more

likely to influence membership in the peer group at the democracies rather than the

dictatorships. This result initially appears to be contrary to the role we might have

15


18/28

anticipated that governance would take. However, there is some recent work such as

Cyert, Kang, and Kumar (2002) and Gillan, Hartzell, and Starks (2007) arguing that some

firms use external governance whereas other firms choose internal governance to monitor

and discipline managers. If that is the case, we might actually see firms with stronger

internal governance (and therefore those with higher GIM values) being the ones that are

less beneficial for CEOs when it comes to constructing the compensation peer group,

consistent with our results. Similar results emerge when we use the E-Index

(Bebchuck, Cohen, and Ferrell (2004)) or block ownership in that CEO pay at the

potential peer has a significant role in constructing the peer group but the effect declinesas the incentive to monitor increases.

We also use total cash compensation rather than total compensation for the

potential peers and also generate similar results. As indicated by the results in Table 7,

higher cash compensation at the potential peer makes it more likely that the firm will be

chosen to be part of the reference group, but the sensitivity to the potential peers

compensation declines as internal governance improves. It is not clear a priori from the

earlier results whether total compensation or just total cash compensation has a larger

influence on which firms are chosen for the peer group but the results appear to be robust

to using either measure.

6. Conclusion

Numerous firms have stated that they follow a process of basing CEO

compensation on an analysis of similar companies, but only recently has this process

become more transparent with greater disclosure of peer group membership. We believe

16


19/28

our work is the first to document that compensation peer groups do indeed play an

important role in determining CEO compensation. Inclusion of measures of the median

or 75 th percentile of compensation for the group dominates other characteristics that have

traditionally been used to explain cross-sectional variation in executive pay. Additionally,

we document a number of summary statistics regarding compensation peer groups and

analyze the determinants of group composition. We find that while industry and size are

important in explaining the composition of these compensation peer groups, the level of

compensation at the potential peer firms also plays a significant role. This effect is

particularly strong for firms that are likely to have weak internal governance.Until recently, disclosure of this information was voluntary leading us to believe

that we have likely underestimated the effects that we have documented. However,

beginning just recently, firms are now required to provide this information in their annual

SEC filings. Such increased transparency should lead to greater analysis by shareholders,

as well as other firm stakeholders, of how potential firms are selected as members of the

compensation peer group. It will be interesting to observe whether this additional scrutiny

will alter the patterns that we have documented here.

17


20/28

References

[1] Anderson, R. and J. Bizjak, 2003, An Empirical Examination of the Role of the CEO

and the Compensation Committee in Structuring Executive Pay, Journal of Banking and Finance , 27(7), pp. 1323-1348.

[2] Baranchuk, N., G. MacDonald, and J. Yang, 2006, The Economics and Super Managers,

working paper, Washington University in St. Louis.

[3] Bebchuk, L. and J. Fried, 2003, Executive Compensation as an Agency Problem, Jour-

nal of Economic Perspectives , 17(3), pp. 71-92.

[4] Bebchuk, L. and J. Fried, 2004, Pay without Performance: The Unfullled Promise of Executive Compensation, Cambridge, MA: Harvard University Press.

[5] Bebchuk, L., A. Cohen, and A. Ferrell, 2004,What Matters in Corporate Governance,

Harvard Law School John M. Olin Center Discussion Paper, No. 491.

[6] Bertrand, M. and S. Mullainathan, 2001, Are CEOs Rewarded for Luck? The Ones

Without Principals Are. Quarterly Journal of Economics, 116 (3), pp. 901-932.

[7] Bizjak, J., M. Lemmon, and L. Naveen, 2007, Has the Use of Peer Groups Contributed

to Higher Levels of Executive Compensation? revise and resubmit Journal of Financial

Economics .

[8] Cyert, R. M., S. Kang, and P. Kumar, 2002, Corporate Governance, Takeovers, and

Top-Management Compensation: Theory and Evidence, Management Science , 48(4),

pp. 453-469.

[9] Gabaix, X. and A. Landier, 2006, Why Has CEO Pay Increased So Much? working

paper, Princeton and NYU.

[10] Gillan, S., J. Hartzell, and L. Starks, 2007, Explaining Corporate Governance: Boards,

Bylaws, and Charter Provisions, the AFA Chicago meeting paper.

[11] Gompers, P., J. Ishii, and A. Metrick, 2003, Corporate Governance and Equity Prices,

Quarterly Journal of Economics , 118, pp. 107-155.

18


21/28

[12] Main, G., C. OReilly, and J. Wade, 1995, The CEO, the Board of Directors, and Execu-

tive Compensation: Economic and Psychological Perspectives, Industrial and Corporate

Change, 11, pp. 606-628.

[13] Murphy, K., 1999, Executive Compensation, Orley Ashenfelter and David Card (eds.),

Handbook of Labor Economics , Vol. 3, North Holland.

[14] Murphy, K. and J. Zabojnik, 2004, CEO Pay and Appointments: A Market-Based Ex-

planation for Recent Trends, American Economic Review Papers and Proceedings .

[15] Newman, H. and H. Mozes, 1999, Compensation Committee Composition and its Inu-

ence on CEO Compensation Practices, Financial Management , Autumn, 1999.

[16] OReilly, C., B. Main, and G. Crystal, 1988, CEO Compensation as Tournament and

Social Comparison: A Tale of Two Theories, Administrative Science Quarterly, 33, pp.

257-274.

[17] Oyer, P., 2004, Why Do Firms Use Incentives that Have No Incentive Effects? Journal

of Finance, 59, pp. 1619-1649.

19


22/28

Table 1: Disclosing Firms

SIC Gvkey Name SIC Gvkey NAME

1311 8068 OCCIDENTAL PETROLEUM CORP 1311 11923 ANADARKO PETROLEUM CORP1311 15084 BURLINGTON RESOURCES INC 1311 16478 EOG RESOURCES INC1389 22794 BJ SERVICES CO 1531 2845 CENTEX CORP1531 8823 PULTE HOMES INC 2030 5568 HEINZ (H J) CO2085 2435 BROWN-FORMAN -CL B 2086 12756 COCA-COLA ENTERPRISES INC2111 8543 ALTRIA GROUP INC 2600 6104 INTL PAPER CO

2631 10426 TEMPLE-INLAND INC 2711 6475 KNIGHT-RIDDER INC2820 4087 DU PONT (E I) DE NEMOURS 2834 1478 WYETH2834 2403 BRISTOL-MYERS SQUIBB CO 2834 6266 JOHNSON & JOHNSON2834 6730 LILLY (ELI) & CO 2834 7257 MERCK & CO2834 8530 PFIZER INC 2834 9459 SCHERING-PLOUGH2836 9699 SIGMA-ALDRICH CORP 2840 8762 PROCTER & GAMBLE CO2844 1920 AVON PRODUCTS 2911 2991 CHEVRON CORP2911 7017 MARATHON OIL CORP 2911 8549 CONOCOPHILLIPS2911 10156 SUNOCO INC 2911 15247 VALERO ENERGY CORP3011 5234 GOODYEAR TIRE & RUBBER CO 3420 10016 STANLEY WORKS3531 2817 CATERPILLAR INC 3570 5606 HEWLETT-PACKARD CO3577 11636 XEROX CORP 3640 3497 COOPER INDUSTRIES LTD3663 7585 MOTOROLA INC 3663 24800 QUALCOMM INC3674 157858 FREESCALE SEMICONDUCTOR INC 3711 8253 PACCAR INC3760 6774 LOCKHEED MARTIN CORP 3812 7985 NORTHROP GRUMMAN CORP3841 2111 BECTON DICKINSON & CO 4011 7923 NORFOLK SOUTHERN CORP4813 2146 BELLSOUTH CORP 4841 3226 COMCAST CORP4911 9846 EDISON INTERNATIONAL 4922 4242 EL PASO CORP4924 8470 PEOPLES ENERGY CORP 4931 5742 CENTERPOINT ENERGY INC5331 4016 DOLLAR GENERAL CORP 5399 29028 COSTCO WHOLESALE CORP5411 6502 KROGER CO 5651 7922 NORDSTROM INC5731 2184 BEST BUY CO INC 5812 11366 WENDYS INTERNATIONAL INC5940 14624 OFFICE DEPOT INC 6020 2019 BANK OF NEW YORK CO INC6020 4737 FIRST HORIZON NATIONAL CORP 6020 7238 MELLON FINANCIAL CORP6020 7647 BANK OF AMERICA CORP 6020 7711 NATIONAL CITY CORP6020 7982 NORTHERN TRUST CORP 6020 8007 WELLS FARGO & CO6020 8245 PNC FINANCIAL SVCS GROUP INC 6020 10187 SUNTRUST BANKS INC6020 11856 BB&T CORP 6035 5216 GOLDEN WEST FINANCIAL CORP6111 10121 SLM CORP 6111 15208 FEDERAL HOME LOAN MORTG CORP6211 7267 MERRILL LYNCH & CO INC 6211 12124 MORGAN STANLEY & CO. Inc6211 30128 LEHMAN BROTHERS HOLDINGS INC 6211 114628 GOLDMAN SACHS GROUP INC6311 1487 AMERICAN INTERNATIONAL GROUP 6311 6742 LINCOLN NATIONAL CORP6311 143356 PRUDENTIAL FINANCIAL INC 6331 9351 SAFECO CORP6351 24287 AMBAC FINANCIAL GP 7320 4423 EQUIFAX INC8011 23877 COVENTRY HEALTH CARE INC 8071 64166 QUEST DIAGNOSTICS INC9997 5047 GENERAL ELECTRIC CO

20


23/28

Table 2: Summary Statistics

The descriptive statistics are based on 490 of S&P500 rms that ExecuComp provides CEO compensa-tion data in 2005. T CC is the cash compensation (salary and bonus payment). T DC 1 is the CEO totalcompensation including option grants during the year (in $million). Sales is rm sales (in $million);ROA is the return on assets; Return is one year stock return ( T rs 1yr in ExecuComp ). Volatility is the60-month volatility used for option valuation ( BS V OLAT in ExecuComp ). The 5% ownership is thetotal ownership by block holders (with ownership 5%), provided by Compact Disclosure . Disclosingrms are rms that disclose explicit lists of compensation peers in their SEC lings: DEF-14A in 2005.*, ** and *** indicate the difference of variable between the two groups of rms is signicant at the10%, 5% and 1% levels, respectively.

Disclosing Firms (A) Non-dscl. Firms (B) Difference (A-B)

TCC ($million) Mean 4.205 2.872 1.334***Median 2.950 2.352Std Dev 4.224 2.868

TDC1($million) Mean 13.207 9.401 3.806***Median 8.975 6.852Std Dev 11.604 8.723

Sales($million) Mean 27,054 14,083 12,970***Median 14,057 6,523.7Std Dev 35,414 12,970

ROA Mean 6.647% 6.279% 0.368%

Median 6.063% 5.259%Std Dev 5.243% 6.814%

Return Mean 18.09% 9.941% 8.151%***Median 6.063% 5.259%Std Dev 10.11% 5.981%

Volatility Mean 0.2838 0.3440 -0.0600***Median 0.2560 0.2890Std Dev 0.1216 0.1932

GIM index Mean 9.901 9.649 -0.252Median 10 10Std Dev 2.370 0.1932

E index Mean 2.432 2.441 0.0087Median 2 3Std Dev 1.322 1.314

Block ownership Mean 21.86% 26.85% -4.99%**Median 19.35% 24.92%Std Dev 15.57% 18.43%

# of Obs. 83 407

21


24/28

Table 3: Effect of Peer Total CompensationThe dependent variable is CEO total compensation (including option grants): T DC 1/ 1000 inthe ExecuComp database. The 50% peer pay is the median total compensation of a rms peergroup, and the 75% peer pay is the 75th percentile of the total compensation of a rms peers.Other variables are dened in Table . The P-value of a coefficient is given in the parenthesesunder the coefficient, and *, ** and *** indicate the coefficient is signicant at the 10%, 5%and 1% levels, respectively.

Dependent Variable: Total Compensation T DC 1 ($million)

Independent Variables Benchmark 50% Peer Pay 75% Peer Pay

Intercept -21.26** -4.533 -7.400(0.04) (0.64) (0.47)

log(sales) 3.776*** 0.590 1.041(0.00) (0.58) (0.37)

ROA 0.401 0.686 0.618(0.42) (0.12) (0.18)

Lagged ROA -0.598 -0.752* -0.791*

(0.24) (0.09) (0.09)Stock return 0.115*** 0.133*** 0.128***

(0.01) (0.00) (0.00)

Lagged stock return 0.035 0.029 0.010(0.54) (0.56) (0.85)

Volatility -12.47 -11.89 -10.05(0.21) (0.17) (0.27)

50% peer pay 1.143***(0.00)

75% peer pay 0.721***(0.00)

# of Obs 82 82 82

Adj. R2 0.2356 0.4278 0.3579

22


25/28

Table 4: Effect of Peer Cash CompensationThe dependent variable is CEO total cash compensation: TCC/ 1000 in the ExecuComp data-base. The 50% peer pay is the median cash compensation of a rms peer group, and the 75%peer pay is the 75th percentile of the cash compensation of a rms peers. Other variablesare dened in Table . The P-value of a coefficient is given in the parentheses under the coef-cient, and *, ** and *** indicate the coefficient is signicant at the 10%, 5% and 1% levels,respectively.

Dependent Variable: Cash Compensation T CC ($million)

Independent Variables Benchmark 50% Peer Pay 75% Peer Pay

Intercept -3.597 -0.273 -1.543(0.39) (0.94) (0.67)

log(sales) 0.976** 0.219 0.256(0.02) (0.58) (0.50)

ROA -0.171 -0.089 -0.102(0.39) (0.61) (0.56)

Lagged ROA -0.014 -0.052 -0.016

(0.95) (0.77) (0.93)Stock return 0.011 0.015 0.013

(0.53) (0.33) (0.39)

Lagged stock return -0.014 -0.012 -0.013(0.53) (0.54) (0.50)

Volatility -0.705 -0.435 0.779(0.86) (0.90) (0.82)

50% peer pay 0.897***(0.00)

75% peer pay 0.784***(0.00)

# of Obs 82 82 82

Adj. R2 0.0837 0.2758 0.2966

23


26/28

Table 5: Summary Statistics on Peer and Potential Peer Firms

Same industry is determined using the 3-digit SIC code. Mean sales (peer-rm) is the the average

sales of the rms in the peer group less the sales of the rm, both in the previous year. Total pay isT DC 1/ 1000 and cash pay is TCC/ 1000 in the Execucomp.

Panel A: Descriptive Statistics of Chosen Peer FirmsMean Median

Number of peers 11.57 10% same industry 0.502 0.500Mean sales(peer-rm) ($million) 1,044 3,106Mean sales(peer-rm) (%) 0.732 0.39550% peer total pay ($million) 10.77 9.94275% peer total pay ($million) 15.92 14.1450% peer cash pay ($million) 3.765 3.21375% peer cash pay ($million) 4.943 4.252

Panel B: Descriptive Statistics on Total Pay of Non-Peer vs. Peer Firms

Non-Peer (A) Peer (B) Difference (B-A)

Diff industry (C) Mean 9.585 12.20 2.613***Median 7.146 10.10# of Obs. 39,842 554

Same industry (D) Mean 9.380 10.48 1.100**Median 9.166 8.839# of Obs. 732 431

Difference (C-D) 1.714***

Panel C: Descriptive Statistics on Cash Pay of Non-Peer vs. Peer Firms

Non-Peer(A) Peer (B) Difference (B-A)

Diff industry (C) Mean 2.896 3.828 0.932***Median 2.377 3.075# of Obs. 39,850 554

Same industry (D) Mean 3.280 3.476 0.196Median 2.438 2.735# of Obs. 732 431

Difference (C-D) 0.352**

24


27/28

Table 6: Peer Group Selection Peer Total Compensation

The dependent variable is 1 if a S&P500 rm is a compensation peer of the 83 disclosing

rms under consideration. Same industry is 1 if the rm and peer share the same 3-digit SICcode. Diff log (sales ) is log(sales ) of the peer minus log(sales ) of the rm in the previousyear. Peer pay is the CEO total compensation of the peer in the previous year ( T DC 1/ 1000in ExecuComp ). GIM index is the Gompers, Ishii, and Metrick (2003) corporate governanceindex, E index is the entrenchment index used by Bebchuk, Cohen, and Ferrell (2004) andBlock ownership is the total ownership of block holders (with ownership 5%). The P-valueof a coefficient is given in the parentheses under the coefficient, and *, ** and *** indicatethe coefficient is signicant at the 10%, 5% and 1% levels, respectively. Standard errors areclustered at the rm level.

Dependent Variable: Firm is Peer

Independent Variables Baseline Peer Pay GIM-Index E-index Block Own.

Intercept -2.185*** -2.277*** -2.541*** -2.321*** -2.362***(0.00) (0.00) (0.00) (0.00) (0.00)

Same industry 2.277*** 2.281*** 2.226*** 2.228*** 2.267***(0.00) (0.00) (0.00) (0.00) (0.00)

Diff log(sales) 0.113*** 0.102*** 0.104*** 0.109*** 0 .110***(0.00) (0.00) (0.00) (0.00) (0.00)

Peer pay 0.008*** 0.042*** 0.022*** 0.021***

(0.00) (0.00) (0.00) (0.00)GIM index 0.027

(0.18)

GIM index*Peer pay -0.003***(0.00)

E index 0.022(0.52)

E index*Peer pay -0.006***(0.00)

Block ownership 0.004(0.25)

Block own.*Peer pay -0.001***(0.00)

# of Obs. 40,419 40,411 39,412 39,412 36,961

# of Clusters 82 82 80 80 75

Pseudo R 2 0.2651 0.2674 0.2507 0.2518 0.2733

25


28/28

Table 7: Peer Group Selection Peer Cash Compensation

The dependent variable is 1 if a S&P500 rm is a compensation peer of the 83 disclosing

rms under consideration. Same industry is 1 if the rm and peer share the same 3-digit SICcode. Diff log (sales ) is log(sales ) of the peer minus log(sales ) of the rm in the previousyear. Peer pay is the CEO cash compensation of the peer in the previous year ( TCC/ 1000in ExecuComp .) GIM index is the Gompers, Ishii, and Metrick (2003) corporate governanceindex, E index is the entrenchment index used by Bebchuk, Cohen, and Ferrell (2004) andBlock ownership is the total ownership of block holders (with ownership 5%). The P-valueof a coefficient is given in the parentheses under the coefficient, and *, ** and *** indicatethe coefficient is signicant at the 10%, 5% and 1% levels, respectively. Standard errors areclustered at the rm level.

Dependent Variable: Firm is Peer

Independent Variables Baseline Peer Pay GIM Index E index Block Own.

Intercept -2.185*** -2.307*** -2.484*** -2.348*** -2.364***(0.00) (0.00) (0.00) (0.00) (0.00)

Same industry 2.277*** 2.278*** 2.220*** 2.223*** 2.257***(0.00) (0.00) (0.00) (0.00) (0.00)

Diff log(sales) 0.113*** 0.095*** 0.097*** 0.102*** 0.103***(0.00) (0.00) (0.00) (0.00) (0.00)

Peer pay 0.036*** 0.042*** 0.077*** 0.067***(0.00) (0.00) (0.00) (0.00)

GIM index 0.018(0.41)

GIM index*Peer pay -0.008***(0.00)

E index 0.020(0.56)

E index*Peer pay -0.018***(0.01)

Block ownership 0.003(0.46)

Block own.*Peer pay -0.002**(0.03)

# of Obs. 40,419 40,419 39,420 39,420 36,969

# of Clusters 82 82 80 80 75

Pseudo R 2 0.2651 0.2690 0.2514 0.2532 0.2737

inside the black box: the role and composition of

Documents