Relative Performan
ce Evaluation and Long-Term Incentives
Stephan Kramer
Erasmus University Rotterdam
Michal Matějka
Arizona State University
February 202
4
The p
aper has benefited from helpful comments of Melissa Martin, Adrianne Rhodes, and workshop participants at the UIC
Business Accounting Research Conference, Cornell University, and the AAA MAS 2024 meeting.
Relative Performance Evaluation and Long-Term Incentives
ABSTRACT: Many prior studies test the theoretical prediction that relative performance evaluation
(RPE) improves contracting by filtering out uncontrollable shocks to performance. However, the theory
also holds that such noise filtering can have adverse long-term effects because it undermines incentives to
invest and to optimize firm risk exposure. We argue that this trade-off between long-term incentives and
noise filtering affects earnings relatively more than stock returns because earnings reflect managerial
actions with a greater delay. In our empirical analysis, we find that incentive contracts use peer earnings
differently than peer stock returns and that CEO compensation is positively associated with peer earnings,
particularly when the CEO has greater control over the firm’s strategic direction and when peer
performance reflects persistent favorable shocks. Our findings imply that earnings-based RPE protects
executives from bad luck but rewards them for good luck and that this asymmetry may be driven by
optimal contracting as well as rent seeking.
Keywords: Relative performance evaluation, CEO compensation, incentives, peer group choice.
Data Availability: Data used in this study is publicly available.
1
I. INTRODUCTION
A large stream of RPE literature is motivated by single-period agency models where it is always
optimal to protect a risk-averse manager from uncontrollable shocks to performance (Holmström 1982;
Feltham and Xie 1994). However, it is also well known that in multi-period settings the anticipation of
future noise filtering undermines ex ante incentives to take long-term actions (Meyer and Vickers 1997;
Casas-Arce and Martinez-Jerez 2009). This trade-off between ex ante incentives to make long-term
investments and ex post demand for noise-filtering has motivated prior work on performance target
ratcheting (Indjejikian, Matějka, and Schloetzer 2014b), stock option repricing (Saly 1994; Acharya,
John, and Sundaram 2000), and the choice of performance measures (Hemmer 1996; Dikolli 2001; Dutta
and Reichelstein 2003), but there is hardly any empirical evidence on how it affects the use of RPE.
Relatedly, most prior RPE studies examine the use of peer stock returns in incentive contracting. It is
now well established that CEO compensation is associated positively with the firm’s own stock returns
but negatively with peer stock returns (Albuquerque 2009), which is consistent with the role of RPE in
filtering out common shocks reflected in own as well as peer performance. In contrast, there is little
evidence on accounting-based RPE and it remains unclear when and how peer earnings are used in CEO
incentive contracts (Lobo, Neel, and Rhodes 2018; Nam 2020).
We predict that peer earnings improve incentive contracting not only by filtering out noise but also by
incentivizing long-term investments.
1
The association between CEO compensation and peer earnings then
depends on whether the contractual demand for noise filtering exceeds the demand for long-term
investments. The tension between the two RPE roles is more pronounced for earnings than for stock
returns because earnings reflect information about CEO actions with a greater delay. Specifically, if a
performance measure immediately reflects all benefits of long-term investments, there is no ex ante
incentive conflict and the primary role of RPE is to ex post filter out noise. In contrast, if a performance
1
We use the term long-term investments or actions in a broad sense to refer not only to investments in new products,
technologies, facilities, or distribution channels but also to other costly choices with long-term benefits such as maintaining a
cash cushion or attracting new capital. Long-term actions also encompass strategic planning and increasing firm exposure to
high-growth industries or firm resilience to economic downturns (Gopalan, Milbourn, and Song 2010).
2
measure reflects investments with a delay, there is a contractual demand for long-term incentives in the
form of expected compensation contingent on future performance (Dutta and Reichelstein 2003). Yet, the
anticipation of future noise-filtering weakens the link between expected compensation and future
performance and thus undermines long-term incentives to invest (Meyer 1995; Meyer and Vickers 1997).
We therefore expect that peer earnings are used for noise filtering relatively less than peer stock returns.
The trade-off between long-term incentives and noise filtering should be even more pronounced for
firms that can benefit from strategic repositioning closer to a new and more profitable set of peers. This
makes the anticipation of future noise filtering even more damaging to long-term incentives because
investments in strategic repositioning can reduce current earnings and at the same time increase future
performance expectations (Indjejikian, Matějka, Merchant, and Van der Stede 2014a). In such settings,
firms may use RPE not so much to filter out noise but to reward the CEO for moving the firm in a
direction followed by successful peers. Empirically, this would manifest as a positive association between
CEO compensation and peer earnings.
Our empirical analysis relies on a comprehensive peer choice model that allows for changes in peer
group composition over time. This contrasts with the most common approach in prior literature which
identifies peers based on largely time-invariant characteristics such as industry and size (Albuquerque
2009). Although several recent studies allow for changing peer groups, it typically comes at the cost of
using only limited information about relevant peer characteristics. For example, some studies take
advantage of enhanced proxy statement disclosures, which now include a list of peers used by boards to
determine CEO compensation (Gong, Li, and Shin 2011; Hung and Shi 2023). The downside of this
approach is that data on self-disclosed peers are available only since 2006 and the choice of publicly
disclosed peers can also reflect various biases (Bizjak, Lemmon, and Nguyen 2011; Vaan, Elbers, and
DiPrete 2019). Other recent studies use time-varying characteristics such as stock-return correlations
(Bloomfield, Guay, and Timmermans 2022) or product similarity (Jayaraman, Milbourn, Peters, and Seo
2021) to identify peers outside of traditional industry categories. The downside is that relying on a small
3
number of such firm-peer characteristics disregards other relevant determinants of exposure to common
shocks.
We construct peer groups as follows. First, we estimate a model of the choice of self-disclosed peers
using Incentive Lab data. We include firm-peer economic characteristics such as relative sales, different
types of industry classification (Fama-French 48 and two- and three- digit SIC codes), diversification
strategy (single- or multi-segment firms), proximity of corporate headquarters, product and life-cycle
stage similarity, talent flows, stock-return correlations (daily and weekly), and quarterly earnings
correlations. Second, we use the coefficient estimates to calculate predicted values for all sample firm-
years, i.e., even when data on actual peer choice is unavailable. These values aggregate all relevant
economic determinants from the peer choice model. Using these predicted values rather than the disclosed
choices to identify peers also alleviates biases due to strategic disclosure. Third, for each firm-year, we
select peers with the highest 20 predicted values and use them to calculate peer stock returns and peer
earnings. In our validation tests, we show that our measures of peer performance are strongly associated
with firm own performance, even after controlling for measures of peer performance commonly used in
prior work.
We use our new peer performance measures to test several predictions motivated by the trade-off
between long-term incentives and noise filtering. First, we replicate findings from prior literature that
CEO compensation is positively associated with own stock returns and earnings but negatively associated
with peer stock returns. Second, we show that the association between CEO compensation and peer
earnings is significantly more positive than the association with peer stock returns, which is consistent
with our argument that peer earnings are used to provide long-term incentives because firm own earnings
reflect information about CEO effort with a delay.
Third, we compare the use of RPE in single-segment firms versus conglomerates. We expect that long-
term incentives are particularly important in conglomerates where the CEO can more easily reallocate
capital and thus has a greater control over firm strategy and peer group composition (Gopalan et al. 2010).
We find that CEO compensation in conglomerates is decreasing in peer stock returns but increasing in
4
peer earnings.
2
In contrast, the significantly positive association between CEO compensation and peer
earnings is not apparent in single-segment firms. This provides further evidence consistent with our
theory that peer earnings are informative about CEO long-term actions not yet reflected in current firm
earnings.
Next, as in Gopalan et al. (2010), we predict that the use of RPE is asymmetric because noise filtering
is relatively more important than long-term incentives during economic downturns, whereas the opposite
holds during upturns. We further predict that this RPE asymmetry is more pronounced for peer earnings
than for peer stock returns because conservative financial reporting makes earnings relatively slower in
incorporating good news than bad news (LaFond and Roychowdhury 2008). We find strong support for
both predictions in that CEO compensation is decreasing in both peer stock returns and peer earnings
during economic downturns but increasing in peer earnings during upturns.
To provide further evidence that our main findings are driven by the information content of peer
performance, we identify a cross-section of firms where peer earnings are highly informative about
current and future firm own earnings. Specifically, we measure peer information content as the average
predicted value from our peer choice model for the top 20 peers selected in each firm-year. A high peer
information content implies the focal firm has many peers that are economically similar, at least in terms
of the main determinants of peer group choice such as daily and weekly stock return correlations, relative
size, and talent flows. We show that our main findings are significantly stronger in the subsample of firms
with an above-median peer information content.
Finally, we note the strong RPE asymmetry documented in our study implies that CEOs are protected
from bad luck but rewarded for good luck (Bertrand and Mullainathan 2001; Garvey and Milbourn 2006).
Although we argue that this asymmetry arises as a result of optimal contracting, we also test whether it is
2
We do not assume that compensation contracts include explicit provisions that would make incentive awards contingent on
favorable performance of an ex ante defined peer group. Rather, we assume that performance of peers selected by our peer choice
model is correlated with private information corporate boards use to evaluate CEO long-term strategic actions such as expanding
new markets, acquisitions, and divestitures. Whereas firm stock returns often quickly incorporate such value-enhancing strategic
actions, firm earnings do so with a delay and may initially decrease even for strategic actions that boost long-term profitability. In
that sense, we view peer performance as an empirical proxy for unobservable leading indicators of future firm performance
(Hayes and Schaefer 2000; Said, HassabElnaby, and Wier 2003; Dikolli and Sedatole 2007).
5
related to governance strength and CEO rent seeking (Bebchuk, Fried, and Walker 2002). We measure
weak governance with an indicator variable for a large number of board members serving on many other
boards (Bloomfield et al. 2022). We find some evidence that the RPE asymmetry and its compensation-
increasing effects are more pronounced when governance is weak.
Combined, our findings make two major contributions to prior literature. First, most prior studies
examine the standard prediction of single-period agency models that CEO compensation is negatively
associated with peer performance. We argue that a simple extension of the theoretical arguments to multi-
period settings motivates a different prediction. If earnings reflect managerial effort with a substantial
delay, the resulting dynamic incentive conflicts can give rise to a positive association between CEO
compensation and peer earnings. Using a large sample of 41,958 firm-year observations from 1992–2021,
we show that this association is indeed significantly positive on average. We further show that the
association between CEO compensation and peer earnings is even more positive when the CEO has
greater control over the firm’s strategic direction and when peer earnings likely reflect persistent
favorable shocks, i.e., in settings where contractual demand for long-term incentives likely dominates the
demand for noise filtering.
Second, we validate a new approach to measuring peer performance and constructing peer groups,
which allows us to incorporate more information about relevant firm-peer characteristics than has been
feasible so far. A unique feature of our approach is that it leverages the proxy statement disclosures since
2006 to learn about peer groups with the greatest potential to facilitate RPE even in the pre-2006 period.
We show that this approach considerably increases the power of RPE tests. Moreover, it greatly reduces
the cost of RPE data collection because the coefficient estimates from our peer choice model allow
researchers to incorporate information about self-disclosed peers without purchasing or hand-collecting
such data.
6
II. THEORY AND HYPOTHESES
RPE and Noise Filtering
The theoretical underpinning of the RPE literature is the single-period moral hazard model where a
risk-neutral firm contracts with a risk-averse manager to take a costly unobservable action using a noisy
signal about firm performance and managerial effort (Holmström 1982; Feltham and Xie 1994). The firm
can incentivize effort only by making the manager’s compensation contingent on the noisy signal. This
creates contractual demand for noise filtering to make compensation less volatile, which reduces the costs
of incentive provision (Holmström 1979). As long as the firm and its peers are exposed to common
random shocks, peer performance should be used for noise-filtering purposes and make incentive
contracts more efficient.
Early empirical studies on RPE examine whether incentive contracts filter out noise by putting a
negative weight on peer stock returns but find only limited support for the theoretical prediction (Jensen
and Murphy 1990; Janakiraman, Lambert, and Larcker 1992; Aggarwal and Samwick 1999b; Garvey and
Milbourn 2003). One explanation for the mixed results is that researchers cannot accurately identify peer
groups used in incentive contracts and consequently peer performance is measured with error (Dikolli,
Hofmann, and Pfeiffer 2013). Albuquerque (2009) argues that peer performance measures derived from
market or industry indices reduce the power of RPE tests because they disregard other important
determinants of common shock exposure such as relative size. She finds strong evidence that incentive
contracts put a negative weight on stock returns of industry-size matched peers (further referred to as
SIC2-size peers). Several recent studies provide similar evidence that firms use peer stock returns to filter
out noise from CEO compensation. To measure peer stock returns, Gong et al. (2011) use self-disclosed
peers from firms’ proxy statements and Jayaraman et al. (2021) define peers based on size, book-to-
market ratio, and product similarity scores (Hoberg and Phillips 2016).
Another explanation for the inconclusive early findings is that the benefits of noise filtering may be
offset by its cost (Gibbons and Murphy 1990). In particular, RPE is costly or less effective when the
number of suitable peers is small (Jayaraman et al. 2021; Tice 2023). By putting a negative weight on
7
peer performance, RPE contracts incentivize aggressive competitive behavior (Bloomfield et al. 2022;
Feichter, Moers, and Timmermans 2022), peer-harming disclosures (Bloomfield, Heinle, and
Timmermans 2023), or a lack of cooperation (Holzhacker, Kramer, Matějka, and Hoffmeister 2019).
Noise filtering can also conflict with the retention objectives of compensation contracts if CEO labor
market opportunities are positively associated with peer performance (Oyer 2004; Rajgopal, Shevlin, and
Zamora 2006).
In addition, RPE can facilitate rent extraction if powerful CEOs influence the choice of peers,
performance measures, or other features of RPE contracts (Dikolli, Diser, Hofmann, and Pfeiffer 2018).
There is some evidence that RPE is asymmetric in the sense that CEOs are protected from bad luck but
rewarded for good luck, particularly when corporate governance is weak (Bertrand and Mullainathan
2001; Garvey and Milbourn 2006). There is also evidence that CEOs select peers strategically to justify
higher compensation (Bizjak et al. 2011; Faulkender and Yang 2013; Vaan et al. 2019; Bakke, Mahmudi,
and Newton 2020). However, both the RPE asymmetry and the selection of peers to justify higher pay can
also be explained as an optimal response to a greater demand for CEO talent or changes in labor market
opportunities (Bizjak, Lemmon, and Naveen 2008; Albuquerque, De Franco, and Verdi 2012).
Although there is now ample evidence on the costs and benefits of using peer stock returns in
incentive contracts, our understanding of the use of peer earnings is still limited. Several studies detect a
negative weight on peer stock returns but at the same time no weight or even a positive weight on peer
earnings in CEO incentive contracts (Gibbons and Murphy 1990; Janakiraman et al. 1992; Albuquerque
2009; Lobo et al. 2018; Nam 2020). These results cannot be explained by the costs of RPE discussed
above because those would affect both stock returns and earnings equally. Lobo et al. (2018) and Nam
(2020) show that selecting peers on accounting comparability, in addition to size and industry, yields a
negative association between peer earnings and CEO cash compensation but no association with equity
compensation.
8
RPE and Long-Term Incentives
Most prior RPE studies are motivated by the insights from the standard single-period moral hazard
model, even though there is also a large stream of analytical work on muti-period contracting examining a
closely related incentive design issue that can be characterized as follows (Acharya et al. 2000). Suppose
the firm and the manager agree on a compensation contract for two or more periods. What are the welfare
consequences of incorporating new information that becomes available after the first period? The simple
single-period intuition that more information is always better does not carry over to multi-period settings.
If the firm cannot commit to ignoring future information about its own or peer performance, it can be
considerably more costly to incentivize effort in the first period (Laffont and Tirole 1993).
From the perspective of the multi-period framework, RPE can be viewed as a special case of using
future information. The use of information about future peer performance has noise-filtering benefits but
the anticipation of it undermines first-period incentives, which means that better information about peer
performance can actually reduce rather than increase efficiency of incentive contracts (Meyer and Vickers
1997). Conversely, commitment not to use future information exposes the manager to more risk but also
strengthens long-term incentives to invest in profitable opportunities or to increase resilience to economic
downturns. This trade-off affects RPE as well as various other incentive design choices (Demski and
Frimor 1999; Indjejikian and Nanda 1999). For example, Acharya et al. (2000) discuss why resetting
strike prices on underwater stock options (using future information about noise realizations) is often
beneficial but too much resetting can be harmful. Indjejikian et al. (2014b) argue that it is efficient to
underuse information about past performance when revising annual performance targets.
Dutta and Reichelstein (2003) analyze a multi-period model where the manager exerts effort
increasing current profits but also makes investments increasing future profits. They examine whether
incentive contracts use leading indicators of performance (defined as noisy forecasts of future investment
returns) or delay rewards by making them contingent on realized future profits. The incentive problem is
that the firm rationally adjusts performance expectations for increased profits due to past investments and,
anticipating such adjustments, the manager has no incentives to invest. Commitment to long-term
9
contracts (by both the manager and the firm) makes rewards contingent on future profits more effective
but does not fully eliminate the demand for leading indicators. Similar to the use of peer performance for
noise filtering, the leading indicator is negatively associated with future compensation because future
performance benefited from first-period investment. However, the association between the leading
indicator and first-period compensation may be positive if uncertainty about the future would otherwise
lead to underinvestment (Dutta and Reichelstein 2003: 847). In other words, exactly because future RPE
filters out the effect of past investments on profitability, it may be important to put a positive weight on
peer performance early on when it acts as a leading indicator of future performance.
Several other RPE studies emphasize that the manager can take actions affecting the firm’s strategic
direction and long-term profitability. It is now well understood that the strong demand for noise filtering
and the resulting negative weight on peer performance in the single-period RPE model are driven by the
assumption that noise exposure is independent of effort (Ball, Bonham, and Hemmer 2020). If this
assumption does not hold and managerial effort affects firm exposure to random shocks, then noise
filtering can become very costly. Schäfer (2023) illustrates this point in a model where the manager
supplies costly effort but also decides on strategic differentiation that can increase firm profits by
reducing resemblance to peers. The cost of noise filtering in this setting is the manager’s reluctance to
engage in strategic differentiation. Similarly, some studies argue that noise filtering encourages costly
long-term competition and that a positive weight on peer performance motivates strategic actions that
soften competitive behavior (Aggarwal and Samwick 1999a; Vrettos 2013).
Ball et al. (2020) and Hemmer (2023) relax the assumption that exposure to common shocks is
independent of managers’ actions. RPE then provides incentives to increase effort as well as resemblance
to more profitable (aspirational) peers. Putting a negative weight on peer performance filters out noise but
undermines incentives to mimic aspirational peers. The optimal incentive contract puts a positive weight
on the correlation between firm and aspirational peer performance and the weight on peer performance
need not be negative. Several other studies provide a similar insight. If the correlation between own and
peer performance is increasing in own and peer actions (rather than independent of those actions, as
10
assumed in most single-period models), then RPE can manifest as a positive weight on peer performance
(Celentani and Loveira 2006; Magill and Quinzii 2006; Fleckinger 2012).
Gopalan et al. (2010) go a step further by abstracting away from managerial effort increasing firm
performance, which eliminates the well-understood demand for noise filtering. The key incentive design
issue is then how to increase exposure to a favorable common shock, or reduce exposure to an
unfavorable one, rather than how to supply more effort. The optimal incentive contract motivates the
CEO to direct resources to a business sector with high expected profitability, which necessitates a positive
weight on peer performance, particularly if the CEO has greater control over the firm’s strategic direction.
Hypotheses
The theory predicts that the sign of the association between CEO compensation and peer performance
depends on the relative importance of noise filtering versus long-term incentives. Strong contractual
demand for noise filtering should manifest as CEO compensation increasing in firm’s own performance
but decreasing in peer performance. In practice, this can be implemented with incentive grants that have
explicitly defined performance measures, peer groups, and payouts contingent on outperforming peers
(Carter, Ittner, and Zechman 2009; Bettis, Bizjak, Coles, and Kalpathy 2018; Pawliczek 2021).
3
Alternatively, firms can wait until peer performance is observed and then filter out noise from CEO
compensation by adjusting incentive awards upward for unfavorable shocks reflected in poor peer
performance and downward for good peer performance (Albuquerque 2009).
As the demand for long-term incentives increases, the association between CEO compensation and
peer performance should at first become less negative and then turn positive as peer performance
increasingly acts a leading indicator of firm’s future performance (Dikolli 2001). In practice, this does not
necessarily mean that firms explicitly contract on peer performance. Rather, peer performance is likely to
3
Incentive awards with explicitly defined peer groups and vesting conditions based on relative performance are not very common
in our sample. As discussed later, our main findings remain unchanged if we exclude or separately examine firm-year
observations with such incentive awards (see Table 10).
11
be correlated with boards’ private information about progress on strategic objectives and long-term goals
that is not yet reflected in current performance (Hayes and Schaefer 2000).
We make several predictions about the relative importance of noise filtering versus long-term
incentives and the implications for the association between CEO compensation and peer performance.
First, we expect that incentive contracts use peer earnings differently than peer stock returns. Given that
stock returns incorporate both the costs and long-term benefits of strategic investments relatively quickly
(Gopalan et al. 2010), the need to use peer stock returns as a leading indicator of firm’s own stock returns
is low and incentive contracts may use peer stock returns primarily for noise filtering. In contrast,
earnings incorporate investment benefits much slower than its costs (Penman and Zhang 2002;
Albuquerque 2009), which increases the demand for forward-looking information contained in peer
earnings (Dutta and Reichelstein 2003). Firms where strategic repositioning and other long-term
investments are critical may use peer earnings primarily as a leading indicator of their own future
earnings and, consequently, CEO compensation may even be increasing in peer earnings.
H1: The association between CEO compensation and peer performance is higher for peer earnings than
for peer stock returns.
Second, we test the prediction of Gopalan et al. (2010) that the weight on peer performance is higher
in conglomerates where the CEO can more easily reallocate capital and thus has greater control over the
firm’s strategic direction. As discussed in the theory section, incentive contracts that put a highly negative
weight on peer performance preserve the status quo because, besides noise, they also filter out the benefits
of strategic repositioning closer to more profitable peers. This may be less of a concern in single-segment
firms but a key role of conglomerate CEOs is to identify and invest in high-growth business segments,
which makes long-term incentives relatively more important than noise filtering. By H1, this effect should
manifest primarily in the weight on peer earnings because peer stock returns are used relatively more for
noise filtering than for long-term incentives.
H2a: The association between CEO compensation and peer performance is higher in conglomerates than
in single-segment firms.
12
H2b: The effect described in H2a is more pronounced for peer earnings than for peer stock returns.
Third, Gopalan et al. (2010) also predict that the weight on peer performance is higher for favorable
than for unfavorable shocks. The intuition is that risk-averse CEOs have relatively strong incentives to
avoid bad states of the world and, consequently, the need for noise filtering likely dominates the need for
long-term incentives to reduce exposure to unfavorable shocks. In contrast, when experiencing favorable
shocks and anticipating high compensation, CEOs may be complacent with good firm performance and
may underinvest in increasing the firm’s upside risk exposure. This strengthens contractual demand for
long-term incentives, which manifests empirically as a positive weight on peer performance.
Asymmetric RPE could arise as a result of optimal contracting even where there is no demand for
long-term incentives. For example, Celentani and Loveira (2006) assume that marginal product of effort
is greater in good states of the world, which makes it likely that managerial effort and peer performance
are high at the same time. In such environments, managerial compensation is decreasing in peer
performance in bad times but increasing in good times. A similar asymmetry also arises in the model of
Ball et al. (2020). If high peer performance is very informative about high managerial effort, then the use
of peer performance information boosts managerial compensation both for favorable and unfavorable
shocks.
Thus, we expect asymmetric RPE for all measures of peer performance. Nevertheless, given that it can
at least partly be explained by contractual demand for long-term incentives, which affects the use of
earnings more than the use of stock returns, we also predict that the RPE asymmetry is more pronounced
for peer earnings.
H3a: The association between CEO compensation and peer performance is higher during economic
upturns than during downturns.
H3b: The RPE asymmetry described in H3a is more pronounced for peer earnings than for peer stock
returns.
13
III. DATA AND RESEARCH DESIGN
Data Sources and Sample
Our main source of data is the intersection of Execucomp and Compustat for the years 1992–2021. We
obtain stock price and inflation data from the Center for Research in Security Prices (CRSP), self-
disclosed peers from Incentive Lab, and data on similarity in firms’ 10-K product descriptions from the
library of Hoberg and Phillips.
4
As in Albuquerque (2009), we drop all non-CEO observations, those with
multiple CEOs in a firm-year, and those with a new CEO on the job for less than a year. The resulting
dataset has 46,524 firm-CEO-year observations. We further drop 2,675 observations with missing data on
CEO compensation, stock returns, or return on assets (ROA) and obtain the main sample of up to 43,849
firm-year observations from 3,817 unique firms. The samples available for our tests are smaller due to
missing data on peer performance or one of the moderating variables, as shown in Table 1. All variables
denominated in dollars are adjusted for inflation and presented in 1992 dollars. All continuous variables
are winsorized at the 1st and 99th percentiles.
Variable Measurement
We use two sets of variables in our empirical analysis. The first set below largely follows the variable
definitions from Albuquerque (2009). The second set is derived from the peer selection model discussed
in the next subsection.
CEO compensation is the natural logarithm of one plus total annual flow compensation, including
salary, bonus, other incentive payouts, restricted stock, and stock options (tdc1 in Execucomp).
Firm stock return is the natural logarithm of [(1 + ret/100) / (1 + cpi)], where ret is the annual
(compounded) stock return obtained from monthly data and cpi is the annual rate of inflation.
Firm ROA is the natural logarithm of one plus annual income before extraordinary items (ib in
Compustat) divided by beginning-of-year total assets (at), both adjusted for inflation.
4
https://hobergphillips.tuck.dartmouth.edu/industryconcen.htm.
14
Peer stock return (SIC2-size) is calculated in the same way as Firm stock return and averaged for all
peers in the same two-digit SIC code and size quartile, excluding the own-firm stock return. Size quartiles
are based on beginning-of-year market value. When the number of peers in an industry-size group is two
or less, we use the average of all peer returns in the industry regardless of size. Favorable shock (peer
stock) is an indicator variable equal to one if Peer stock return > 0.
Peer ROA (SIC2-size) is calculated in the same way as Firm ROA and averaged for all peers in the
same two-digit SIC code and size quartile, using the same procedure as Peer stock return (SIC2-size).
Favorable shock (peer ROA) equals to one if Peer ROA > 0.
Assets are a proxy for firm size, measured as the natural logarithm of total assets (at in Compustat).
We use total assets rather than sales or market value because the latter two are more closely correlated
with firm stock returns and ROA.
CEO chair is an indicator for CEO being also the board chair.
Tenure is the natural logarithm of the number of years since the CEO took office. The number of years
is calculated as one twelfth of the number of months between the current fiscal year and month and the
month the CEO took office (becameceo in Execucomp).
Ownership is an indicator for an above-median CEO ownership, calculated as the percentage of shares
(excluding options) owned by the CEO divided by the number of common shares outstanding at the end
of the fiscal year.
Conglomerate is an indicator for firms reporting positive sales and assets in more than one three-digit
SIC code as in Gopalan et al. (2010). We drop observations for which the sum of reported sales in the
Compustat segment files does not fall within 25 percent of total firm sales in the annual files as in Ozbas
and Scharfstein (2010). Table 1 shows that these data requirements considerably reduce the number of
non-missing observations. Conglomerates comprise 50 percent of the sample and the remaining 50
percent are single-segment firms.
15
New Measures of Peer Performance
Peer stock returns and ROA, as defined in the previous section, are the most commonly used measures
of peer performance. Their main downside is that they are defined in terms of only two largely time-
invariant characteristics (industry and size). Our theoretical arguments call for measures of peer
performance that allow for changes in firm strategy and thus also peer group composition over time. To
construct such measures, we proceed as follows.
First, we use Incentive Lab data on self-disclosed peers. This includes peer groups used to benchmark
general compensation for the CEO (referred to as Peer Data for Benchmark Compensation Comparisons)
as well as peer groups for relative performance awards (Peer Data for Relative Performance Goals).
Ninety nine percent of the observations are from the 2006–2021 sample period following enhanced proxy
statement disclosure requirements. The remaining one percent are observations from voluntary
disclosures during 1998–2005.
Second, we measure several firm-peer characteristics expected to be major determinants of firms’
choices of peers for RPE purposes. As discussed below, this includes multiple measures based on
correlations between firm and peer performance, similarity in diversification strategy, geographical
proximity, lifecycle stage, product similarity, talent flows, industry classification, and relative size.
CorrD, CorrDP are indicator variables based on firm-peer correlations in daily stock returns over a
calendar year. Specifically, we calculate firm and peer returns as the natural logarithm of 1 + ret − spret,
where ret is the daily (firm or peer) stock return and spret is the daily return of the S&P500 index. For
each firm-year in our sample, we calculate correlations for all potential peers from the population of
CRSP firms with non-missing daily returns in a year (4,000–5,000 correlations depending on the year).
5
CorrD equals one if a peer is among the top 20 peers with the highest correlations in a given firm-year.
5
We exclude non-firm entities with daily returns on CRSP such as funds and trusts.
16
CorrDP equals one if a peer is among the top 20 peers in at least one other year during the 1992–2021
sample period.
6
CorrW, CorrWP are based on firm-peer correlations in weekly stock returns over a period of five
calendar years. We use the same daily returns adjusted for S&P500 as above. We aggregate them into
weekly returns with (typically) five trading day periods ending on Wednesday. For each firm-year in our
sample, we calculate correlations for all potential peers from the population of CRSP firms with non-
missing weekly returns over the current and four prior years (3,000–4,000 correlations depending on the
year). CorrW equals one if a peer is among the top 20 peers in a given firm-year. CorrWP equals one if a
peer is among the top 20 peers in at least one other year during the sample period.
CorrQ, CorrQP are indicator variables based on firm-peer correlations in quarterly sales changes,
calculated as (saleq
q
- saleq
q-1
)/ saleq
q-1
, over a period of ten calendar years. Given that 1986 is the first
year with quarterly data available, we calculate the quarterly correlations starting in 1995, which is the
first fiscal year with 40 observations available, including quarterly sales from 1986–1995. For each firm-
year in our sample, we calculate correlations for all potential peers from the population of Compustat
firms with non-missing quarterly sales over the ten-year period ending with the current year. CorrQ
equals one if a peer is among the top 20 peers for a given firm-year. CorrQP equals one if a peer is
among the top 20 peers in at least one other year during the sample period.
FF48, SIC2, SIC3 are indicators for a firm-year-peer match in terms of industry classification. FF48
equals one if both the firm and the peer have the same Fama French 48 industry code in a given fiscal
year. SIC3 is defined similarly for the three-digit SIC classification. SIC2 equals one if both the firm and
the peer have the same SIC2 code but a different SIC3 code.
Segments is an indicator variable for a firm-year-peer match in terms of diversification strategy. It
equals one if both the firm and the peer are single-segment firms in a given year or if they are both
conglomerates, as defined earlier.
6
For example, if a peer appeared among the top 20 for a given firm in years 1995 and 2000, then CorrD = 1 in those two years
and CorrD = 0 in all other years, whereas CorrDP = 1 for all years during the sample period.
17
Proximity is an indicator variable for proximity in terms of geographical location. It equals one if firm
and peer headquarters are less 100 miles apart in a given year (it changes over time only in the rare cases
where the firm or the peer move their corporate headquarters).
TNIC is the product similarity score of Hoberg and Phillips (2016) calculated for each firm-year-peer
combination. In particular, we use the TNIC3TSIMM variable from the TNIC3HHI data file.
TalentFlows is an indicator variable equal to one if at least one of the named executive officers in
Execucomp moved between the firm and the peer in the last five years (Albuquerque et al. 2012).
LifeCycle is an indicator variable equal to one if the firm and the peer are in the same life cycle stage,
calculated as in Drake and Martin (2020).
RelSize is an indicator for firm-year-peer similarity in terms of annual sales. For this calculation, we
consider all Compustat firms with sales, total assets, and market value of at least ten million. For each
firm-year, we calculate peer relative size ratios as the absolute value of (psale – sale)/sale, where psale
stands for annual peer sales and sale stands for firm sales. We use the ratio to split the Compustat
population into quartiles. RelSize equals one for all peers in the lowest quartile, i.e., for all peers with
sales that are relatively close to the sales of the focal firm in a fiscal year.
Third, we use the sample of 21,953,349 firm-year-peer observations with non-missing data on all the
variables defined above. This subsample includes 11,793 firm-years and 4,353 unique (potential) peers.
We use it to estimate the following peer choice model, where the dependent variable, Rpeer, is an
indicator that equals one for a peer disclosed by a firm as a compensation or performance benchmarking
peer in a given year:
7
Rpeer = β
1
+ β
2
CorrD + β
3
CorrDP + β
4
CorrW + β
5
CorrWP + β
6
CorrQ + β
7
CorrQP + β
8
FF48 +
+ β
9
SIC2 + β
10
SIC3 + β
11
Segments + β
12
Proximity + β
13
TNIC + β
14
TalentFlows +
+ β
15
LifeCycle + β
16
RelSize + ε
.
(1)
7
We have considered other potential determinants of peer choice, including firm-peer similarity in terms of leverage, market-to-
book ratio, R&D expenses and advertising expenses as a percentage of total assets, fiscal year end, and financial reporting
comparability (De Franco, Kothari, and Verdi 2011). We do not include them in model (1) either because their coefficient
estimates are not significantly positive or because increasing similarity does not have a monotonic effect.
18
Table 2 presents descriptive statistics for all variables in the peer choice model. Table 3 presents Logit
estimates of (1).
8
We find that the most important determinants of peer choice are stock return
correlations, relative size, and talent flows. Quarterly sales correlations are statistically significant in our
model but add less incremental explanatory power. As for industry, the Fama French classification has the
most explanatory power, followed by the SIC2 classification. The combined explanatory power of FF48,
SIC2, and SIC3 is comparable to TNIC, which suggests that product similarity scores are important in
explaining firms’ peer choices. Segments, Proximity, and LifeCycle are also highly statistically significant
in the peer choice model. The predicted values from model (1), Ppeer, represent the ex ante likelihood of
being selected as a peer for a given firm-year. The correlation between Ppeer and Rpeer, the actual peer
choice, is 0.408 (p < 0.001).
Fourth, we calculate Ppeer for all firm-year-peer observations during 1992–2021 that have nonmissing
data on at least one of the firm-peer correlations (CorrD, CorrW, or CorrQ). This is a much larger sample
of 237,425,299 firm-year-peer observations, including 43,078 firm-years and 18,665 unique peers. For
each firm-year, we then select Predicted peers as those with the highest 20 Ppeer values.
9
Importantly,
this (as well as the next) step no longer uses Incentive Lab data on self-disclosed peers. The coefficient
estimates from Table 3, combined with Compustat, CRSP, Execucomp, and TNIC data on the right-hand-
side variables in model (1), are sufficient to obtain the set of Predicted peers for any firm-year.
Finally, we use the analysis above to construct the following variables with descriptive statistics at the
bottom of Table 1. Predicted is the firm-year average of Ppeer for all 20 Predicted peers. Predicted peer
stock return is calculated the same way as Firm stock return (based on annual inflation-adjusted stock
returns) and averaged for all Predicted peers. Predicted peer ROA is calculated as Firm ROA and
averaged for all Predicted peers. As a simplification, we use the same labels as in the case of SIC2-size
8
Our main findings remain qualitatively unchanged results when we estimate the peer choice model in (1) using OLS or Logit
adjusted for rare events (King and Zeng 2001).
9
Missing values for the right-hand side variables are set to zero, which reduces Ppeer and the likelihood of being selected as a
predicted peer but does not prevent potential peers with some missing values from being among the top 20.
19
peers to refer to favorable shocks: Favorable shock (Predicted peer stock) equals one if Predicted peer
stock return > 0 and Favorable shock (Predicted peer ROA) equals one if Predicted peer ROA > 0.
Validation Analysis
Panel A of Table 4 compares the predictive power of Ppeer to several benchmarks. As in Table 1, the
unconditional mean of Rpeer is only 0.8 percent. However, conditional on being in the same SIC2
industry as the focal firm, the likelihood of being chosen as a peer increases to 8.7 percent. Being in the
same SIC2 industry and having the same relative size further increases the likelihood to 20.8 percent. The
likelihood is even higher, at 31.3 percent, in the subsample with the highest Ppeer values, holding the
number of observations constant. Thus, the overlap between self-disclosed peers and Predicted peers
(based on Ppeer) is higher than for SIC2-size peers because the former uses more information about firm-
peer characteristics, not just relative size and SIC industry classification.
Panel B of Table 4 examines the extent to which firm own performance can be explained by peer
performance measures defined in terms of SIC2-size peers and Predicted peers, respectively. To facilitate
the comparison, we present standardized coefficients in all regressions and include firm and year fixed
effects. Column (1) shows that Firm stock return is positively associated with SIC2-size peer stock return
(0.230, p < 0.001) but even more strongly associated with Predicted peer stock return (0.409, p < 0.001).
The within-R
2
of 0.217 is also larger than the within-R
2
from a regression including only SIC2-size peer
stock return as a predictor (0.144, untabulated).
10
Column (2) estimates the same regression as column (1)
but uses a subsample of observations with high peer information content (Predicted low = 0), defined as
above-median values of Predicted. As expected, the standardized coefficient estimate of Predicted peer
stock return (0.441, p < 0.001) is higher than in column (1) and the within-R
2
increases to 0.333.
The increase in explanatory power that comes from using Predicted peers is even more pronounced
for peer earnings. In an untabulated test, we find that Firm ROA is positively associated with SIC2-size
peer ROA but the within-R
2
is only 0.014. Adding Predicted peer ROA in column (3) yields a coefficient
10
We rely on the within-R
2
to measure explanatory power because our main analysis includes firm and year fixed effects and
thus removes all between-variation.
20
estimate that is more than ten times larger and the within-R
2
increases from 0.014 to 0.130. This suggests
that SIC2-size peers are largely ineffective in capturing within-firm common shocks to ROA. Using
Predicted peers is a considerable improvement but it is still the case that common shocks to ROA are
more difficult to capture than shocks to stock returns. Column (4) uses the subsample of observations
with a high peer information content. The within-R
2
further increases from 0.130 to 0.215 and the
coefficient on Predicted peer ROA increases from 0.464 to 0.569 (both p < 0.001), whereas the coefficient
on SIC2-size peer ROA increases only slightly from 0.033 to 0.057 (both p = 0.001).
Panel C of Table 4 examines the extent to which Predicted peer performance is a leading indicator of
future firm performance. As in Panel B, columns (1) and (3) use the full sample of firm-year observations
and columns (2) and (4) use subsamples where Predicted peer performance has a high information
content. Columns (1) and (2) show that peer stock returns do not predict next year’s Firm stock return
t+1
because the within-R
2
is not meaningfully larger than zero. In contrast, Predicted peer ROA does seem to
be a leading indicator of next year’s Firm ROA
t+1
. Specifically, in column (3), the standardized coefficient
on Predicted peer ROA (0.230, p < 0.001) is much larger than the coefficient on SIC2-size peer ROA
(0.029, p = 0.010) and the within-R
2
is 0.035. Both coefficients as well as the within-R
2
are higher in
column (4), which suggests that Predicted peer earnings with a high information content can better
explain contemporaneous firm earnings as well as better predict future firm earnings.
In summary, Table 4 compares Predicted peer performance to the commonly used measure of peer
performance from Albuquerque (2009). In all our tests, we find that our new measures of Predicted peer
performance have considerably more explanatory power. Appendix B presents a comparison with two
other measures of peer performance proposed in recent work. We find that our measures of Predicted
peer performance have more explanatory power than performance of peers based on financial reporting
comparability (Lobo et al. 2018; Nam 2020). Peers based on product similarity (Jayaraman et al. 2021)
perform slightly better than Predicted peers when explaining contemporaneous Firm stock return but
much worse than Predicted peers when explaining Firm ROA.
21
IV. MAIN FINDINGS
Peer ROA versus Peer Stock Returns in RPE
To test H1, we estimate a model of CEO compensation as a function of firm performance, peer
performance, a limited amount of control variables, firm and year fixed effects:
11
CEO Compensation = β
1
+ β
2
Firm stock return + β
3
Firm ROA + β
4
Peer stock return +
+ β
5
Peer ROA + β
6
Assets + β
7
CEO chair + β
8
Tenure + β
9
Ownership +
+ Firm FE + Year FE + ε.
(2)
The well-established finding from the RPE literature is that CEO compensation is negatively
associated with Peer stock return, β
4
< 0, as a way of filtering out noise and reducing compensation
fluctuations due to common shocks to firm and peer performance. H1 predicts that β
5
> β
4
because firm
earnings reflect CEO actions with a delay and consequently Peer ROA is used not only to filter out noise
but also to provide long-term incentives.
Column (1) of Table 5 estimates the RPE model in (2) using SIC2-size peers as in Albuquerque
(2009). We use a larger sample with 16 additional years but find similar results. Holding firm own
performance constant, CEO compensation is decreasing in peer stock returns but increasing in peer
earnings. Column (2) estimates the same regression but presents standardized coefficient estimates for all
performance variables to allow for a comparison of the relative magnitudes of their effects. Column (3)
uses Predicted peers to measure performance and shows largely similar coefficient estimates and R
2
.
Consistent with H1, we find that the difference between the weight on Predicted peer ROA and the weight
on Predicted peer stock return is significantly positive (β
5
− β
4
= 0.042, p < 0.001). The difference is even
more pronounced in column (2), which means that H1 is supported regardless of whether we use SIC2-
size peers or Predicted peers for our tests.
11
Some prior studies use additional control variables such as firm sales, market value or market-to-book ratio, dividends, and
volatility in stock returns. We do not include them in (2) because they are closely related either to firm earnings or to stock
returns, which could introduce a bias when estimating the relative importance of stock returns versus ROA.
22
Table 5 further shows several significant effects of our control variables. Not surprisingly,
compensation is higher in larger firms (as measured by total assets) and for CEOs who are also board
chairs or have longer tenures. We include these control variables in all our regressions, discussed next,
but we do not tabulate them because their effects remain qualitatively unchanged.
RPE in Conglomerates and Single-Segment Firms
The theoretical argument motivating H1 is that peer earnings can be used not only for noise filtering
but also to strengthen long-term incentives, which calls for a positive or at least a less negative association
with CEO compensation than in the case of peer stock returns. H2a predicts that the association between
CEO compensation and peer performance is less negative in conglomerates, where long-term incentives
are more important than in single-segment firms, because conglomerate CEOs are responsible for the
choice of industries they operate in and should therefore be more exposed to random fluctuations or
trends in industry performance. H2b predicts that this conglomerate effect will be more pronounced for
peer earnings because, as predicted by H1 and corroborated in Table 5, peer stock returns are used
relatively more for noise filtering than for long-term incentives.
In the first column of Table 6, we estimate model (2) after adding the main effect of Conglomerate and
its interactions with Predicted peer stock returns and ROA. We find that β
5
− β
4
> 0 both in single-
segment firms (0.026, p = 0.032) and in conglomerates (0.072, p < 0.001), which provides further support
for H1 and our theory that peer earnings are used relatively less for noise filtering and more for long-term
incentives than peer stock returns. Consistent with H2a and H2b, the first column of Table 6 also shows
that the weakly positive weight on Predicted peer ROA documented in Table 5 is largely driven by
conglomerates (0.040, p = 0.007). In single-segment firms, the weight on Predicted peer ROA is not
significantly different from zero (0.005, p = 0.554). As predicted by H2b, the conglomerate effect is more
pronounced for Predicted peer ROA than for Predicted peer stock return (0.045, p = 0.009).
Importantly, the second column of Table 6 shows that using only SIC2-size peers would yield no
support for H2a or H2b because none of the interactions between Conglomerate and peer performance is
significantly different from zero (0.005, p = 0.413, and 0.009, p = 0.436, respectively). This suggests that
23
using more information than just size and industry classification to select peers allows for more refined
tests of the theory.
An alternative way to estimate the RPE models in Table 6 is to use CEO cash compensation rather
than total compensation as the dependent variable. Table C1 in Appendix C shows that the results remain
largely unchanged. The only notable difference is that the interaction term of Conglomerate with
SIC2-size peer ROA is significantly positive (0.015, p = 0.044), although still much smaller in magnitude
than the interaction effect with Predicted peer ROA (0.062, p < 0.001).
The RPE Asymmetry
H3a predicts that long-term incentives are relatively more important in favorable economic
environments, whereas the demand for noise-filtering dominates when firms experience unfavorable
shocks. H3b predicts that this RPE asymmetry is more pronounced for peer earnings than for peer stock
returns because peer earnings can be used as a leading indicator of future firm performance and should
have more predictive power during economic expansions. To provide additional evidence consistent with
this theoretical motivation, we first extend our validation tests in Panel C of Table 4 by adding an
interaction term between Favorable shock (ROA) and Predicted peer ROA. In untabulated tests, we find
that peer earnings strongly predict future firm earnings in favorable environments but have significantly
weaker predictive power in unfavorable environments.
The first column of Table 7 presents our tests of H3a and H3b based on estimations that extend
model (2) by adding interactions between peer performance and Favorable shock. Consistent with H3a,
both interactions are highly significant. Specifically, during unfavorable times, the CEO compensation is
associated negatively both with Predicted peer stock return (–0.053, p < 0.001) and Predicted peer ROA
(–0.023, p = 0.038). These associations are significantly higher during favorable times, by 0.049
(p < 0.001) for peer stock returns and by 0.154 (p < 0.001) for peer ROA. Consistent with H3b, the
difference in this asymmetry is also significant (0.154 – 0.049 = 0.105, p < 0.001).
The estimates in the first column of Table 7 imply that RPE substantially increases expected CEO
compensation. In unfavorable states of the world, Predicted peer stock return filters out noise by
24
increasing CEO compensation—if a negative Predicted peer stock return decreases by one standard
deviation, CEO compensation increases by 5.2 percent. However, the noise-filtering effect is negligible in
favorable states of the world—if a positive Predicted peer stock return increases by one standard
deviation, CEO compensation decreases by only 0.4 percent. This RPE asymmetry is even more
pronounced for Predicted peer ROA. If a negative Predicted peer ROA decreases by one standard
deviation, CEO compensation increases by 2.2 percent. If a positive Predicted peer ROA increases by one
standard deviation, CEO compensation increases by 14.0 percent.
The second column of Table 7 estimates the same model using SIC2-size peers. The results for peer
stock returns are similar to those in the first column—CEO compensation is negatively associated with
SIC2-size peer stock return (–0.050, p < 0.001) during unfavorable times and this association is
significantly higher during favorable times (by 0.035, p = 0.010). However, there is no significant
asymmetry in the use of SIC2-size peer ROA and the association with CEO compensation is positive both
for favorable and unfavorable shocks.
Table C2 in Appendix C shows that the results are slightly different if we use CEO cash compensation
as the dependent variable. The main finding of a strong RPE asymmetry in the use of Predicted peer ROA
is qualitatively unchanged. However, we also find a significant asymmetry in in the use of SIC2-size peer
ROA and no asymmetry in the use of peer stock returns (regardless of the peer group definition).
V. ADDITIONAL EVIDENCE
We extend our main tests in three ways. First, we provide evidence that the finding of a strong RPE
asymmetry in Table 7 is driven by the information content of peer performance rather than by some
unobserved confounders. Second, we examine whether the RPE asymmetry, which increases CEO
expected compensation, could at least be partly driven by weak corporate governance. Finally, to reduce
heterogeneity in how RPE is implemented, we re-estimate our results separately for the subsamples of
firms that use explicit RPE incentive grants with pre-specified peer groups and those that do not.
25
RPE and Peer Information Content
Theory predicts that the reliance on peer performance in incentive contracts should be greater when
peer performance is more informative about managerial effort and shocks to firm own performance
(Holmström 1979). Some firms have many similar peers exposed to the same economic shocks, whereas
other firms have few peers and are therefore less likely to rely on RPE (Bloomfield et al. 2022). Our
approach to constructing peer groups allows us to measure the availability of peers exposed to similar
shocks, which we refer to as peer information content. Specifically, the firm-year average of the predicted
values from model (1), Predicted, measures the extent to which a firm has many peers in a given fiscal
year that are similar in terms of stock return correlations, relative size, talent flows, and other important
determinants of peer choice. Predicted low is an indicator variable for below-median values of Predicted
and represents a low peer information content.
Table 8 provides evidence that our main findings are stronger in the subsample of observations with a
high information content (Predicted low = 0). The first column re-estimates the model from Table 7 in
this subsample only. The second column presents similar results in the full sample after including the
main and interaction effects of Predicted low. The coefficient estimates on Predicted peer stock return
remain largely unchanged and its interaction effects are insignificant. In contrast, we find that the effect of
Predicted peer ROA depends on peer information content. When peer performance is highly informative,
we find a negative association with CEO compensation in unfavorable times (–0.051, p < 0.001) and a
positive association in favorable times (–0.051 + 0.195
≐ 0.143, p < 0.001). When peer information
content is low, the effect of Predicted peer ROA in unfavorable times is less negative (higher by 0.045,
p = 0.006). Similarly, the increase in the association during favorable times (0.195, p < 0.001) is less
pronounced when peer information content is low (–0.073, p = 0.013). These findings provide at least
some reassurance that our main results are driven by differences in the information content of peer
performance rather than by some confounders.
26
RPE and Corporate Governance
We further examine whether the RPE asymmetry in Table 7 can at least partly be attributed to weak
governance or rent seeking by entrenched CEOs (Bebchuk et al. 2002). RPE could be used as
“camouflage” or “stealth compensation” that makes total expected compensation less transparent. The
RPE asymmetry and its “failure to filter out windfalls” could be a scheme “designed to benefit executives
without being perceived as clearly unreasonable” (Bebchuk and Fried 2003). The role of the board and its
compensation committee is to prevent such abuses, so CEO rent seeking is unlikely to fully explain the
RPE asymmetry (Albuquerque et al. 2012). However, RPE involves numerous incentive design choices,
including the choices of peer group (Bizjak et al. 2008), performance-vesting provisions (Core and
Packard 2022), or adjustments to performance measures (Bloomfield, Gipper, Kepler, and Tsui 2021;
Curtis, Li, and Patrick 2021), and entrenched CEOs may be able to influence at least some of the choices
to their advantage.
To operationalize weak governance, we assume that board monitoring is less effective when boards
are larger and board members busier, as in Bloomfield et al. (2022). We measure board busyness as the
percentage of board members who serve on at least three other boards. We create indicator variables for
observations with (i) the number of board members at or above the sample median and (ii) board
busyness at or above the sample median. The data is available for 28,297 firm-year observations and 37.4
percent of them have Weak governance with both (i) and (ii) equal to one. The remaining 62.6 percent
observations have Strong governance, i.e., either a below-median number of board members or below-
median board busyness.
Table 9 provides evidence that our main findings are stronger when governance is weak. The first
column re-estimates the model from Table 7 in the subsample with weak governance only
(Strong governance = 0). The second column presents similar results in the full sample after including the
main and interaction effects of Strong governance. The coefficient estimates on Predicted peer stock
return remain largely unchanged and its interaction effects are insignificant. In contrast, the effect of
Predicted peer ROA depends on corporate governance. When governance is weak, the association
27
between CEO compensation and Predicted peer ROA is negative in unfavorable times (–0.066, p = 0.012)
and positive in favorable times (–0.066 + 0.220
0.154, p < 0.001). When governance is strong, the
effect of Predicted peer ROA in unfavorable times is less negative (higher by 0.050, p = 0.049). Similarly,
the increase in the association during favorable times (0.220, p < 0.001) is less pronounced when
governance is strong (–0.099, p = 0.008). These findings suggest that the RPE asymmetry exists even in
firms with strong governance, although it is more pronounced when governance is weak and thus may at
least partly be related to rent seeking by entrenched CEOs.
Explicit RPE Incentive Grants
In our last set of analyses, we examine the extent to which firms use RPE ex post after both firm and
peer performance is realized versus rely on explicit RPE incentive grants with ex ante specified peer
groups and performance measures. Incentive Lab data on the latter is available for 37.6 percent for our
sample. We identify explicit RPE incentive grants as follows. First, we start with all incentive grants in
the file “Grants of Plan-Based Awards” that have at least some vesting conditions based on performance
relative to peers (labelled as “Rel” or “Abs/Rel”) or non-zero relative performance goals
(numRelative > 0). Second, we retain only incentive grants with an ex ante specified peer group in the file
“Relative Performance Goals” and exclude grants where an index (most commonly S&P500) is used
instead of disclosing specific peers. Third, we calculate RPE grant % as the sum of the value of all
explicit RPE grants to the CEO in a given firm-year divided by the total value of all CEO grants in that
year. We find that explicit RPE incentive grants are relatively small on average but increasing over time.
Specifically, the average RPE grant % is 4.4 percent in the pre-2006 period, 11.3 percent in the post-2006
period, and around 13.3 percent since 2012.
Table 10 examines whether the results in Table 7 depend on how firms implement RPE. The first
column excludes all observations with RPE grant % > 0. The second column uses these excluded
observations to separately estimate a similar model, except that peer performance is calculated based on
the peers disclosed in the explicit RPE incentive grants. The third column estimates the results in the
subsample of observations from 1992 to 2005, a time period with little or no explicit RPE incentive
28
grants. The results are very similar to those in Table 7, particularly in the first and third columns of
Table 10. The second column also shows a significant RPE asymmetry, even though it uses only a small
subsample of 2,411 observations with RPE grant % > 0 and non-missing data on self-disclosed peer
groups. These findings suggest that our main results are unaffected by the heterogeneity in how firms
implement RPE.
VI. DISCUSSION AND SUMMARY
We introduce a new method of identifying peers and measuring peer performance for RPE purposes.
In contrast to prior work, we select peers based on multiple firm-peer characteristics including self-
disclosed peers, stock-return correlations, quarterly earnings correlations, several types of industry
classifications, relative sales, geographical proximity, diversification strategy, talent flows, product and
life cycle stage similarity. We find that relative firm size, daily and weekly stock return correlations, and
talent flows are the most important determinants of peer choice. Our new measures of peer performance
substantially increase the explained within-firm variance in performance relative to measures of SIC2-size
peer performance most commonly used in prior work. This is particularly important for measures of peer
earnings because the within-firm variance in firm ROA explained by SIC2-size peer earnings is close to
zero.
The theoretical motivation of our empirical analysis also contrasts with much of prior work on RPE
which derives its predictions from the single-period moral hazard model. We draw on analytical studies of
dynamic incentive contracts to argue that multi-period contracts use peer performance not only to filter
out noise but also to provide long-term incentives. While prior work shows that noise filtering calls for a
negative incentive weight on peer performance, we argue that the incentive weight need not be negative
and could even be positive in settings where peer performance is used primarily to incentivize long-term
investments. Our empirical analysis tests several hypotheses about the relative importance of noise
filtering and long-term incentives.
29
Specifically, the theory predicts that peer performance measures can facilitate intertemporal matching
between investments and their returns and consequently incentivize long-term managerial actions.
Contractual demand for long-term incentives or intertemporal matching is greater for performance
measures such as earnings that reflect managerial actions with a greater delay (Dutta and Reichelstein
2003). Consistent with the theory, we find that the association between CEO compensation and peer
earnings is more positive than the association with peer stock returns. We also show that peer earnings are
a leading indicator of future firm performance but peer stock returns are not.
The theory also predicts that the association between CEO compensation and peer performance should
be more positive in conglomerates where CEOs have greater control over long-term strategic choices,
particularly for peer earnings that are informative about future firm performance. Consistent with the
theory, we find that CEO compensation is negatively associated with peer stock returns both in
conglomerates and single-segment firms. In contrast, we find a significantly positive association between
CEO compensation and peer earnings in conglomerates and no association in single-segment firms, which
provides further evidence consistent with the theory that the incentive weight on peer performance
reflects a trade-off between noise filtering and long-term incentives.
Finally, the theory predicts that noise filtering is the primary purpose of RPE when firms experience
unfavorable shocks to their performance. In contrast, in favorable economic environments, the primary
purpose of RPE may be to provide long-term incentives and reward strategic repositioning closer to
successful peers (Gopalan et al. 2010; Schäfer 2023). This implies that optimal long-term contracts may
often feature asymmetric RPE. Using new measures of peer performance, our study provides evidence of
an RPE asymmetry much stronger in magnitude than documented in prior work. Our study is also the first
to show that the RPE asymmetry is stronger for peer earnings than for peer stock returns, particularly
when peer earnings have a high information content.
Our finding of a strong RPE asymmetry relates to the debate on optimal contracting versus rent
seeking motives behind CEO incentive design choices (Bebchuk and Fried 2003; Albuquerque et al.
2012). Adjusting CEO compensation for bad luck and at the same time rewarding good luck, as implied
30
by the RPE asymmetry, could be viewed as a difficult-to-detect scheme to enrich entrenched executives.
Although we cannot rule out that the asymmetry is at least partly driven by rent seeking, we provide
evidence that optimal contracting motives such as the demand for long-term incentives is an equally or
more important driver of the RPE asymmetry.
31
REFERENCES
Acharya V. V., K. John, and R. K. Sundaram. 2000. On the optimality of resetting executive
stock options. Journal of Financial Economics 57 (1): 65–101.
Aggarwal R. K., and A. A. Samwick. 1999a. Executive compensation, strategic competition, and
relative performance evaluation: Theory and evidence. Journal of Finance 54 (6): 1999–
2043.
———. 1999b. The other side of the trade-off: The impact of risk on executive compensation.
Journal of Political Economy 107: 65–105.
Albuquerque A. 2009. Peer firms in relative performance evaluation. Journal of Accounting and
Economics 48 (1): 69–89.
Albuquerque A., G. De Franco, and R. Verdi. 2012. Peer choice in CEO compensation. Journal
of Financial Economics 180 (1): 160-181.
Bakke T. E., H. Mahmudi, and A. Newton. 2020. Performance peer groups in CEO
compensation contracts. Financial Management 49 (4): 997–1027.
Ball R. T., J. Bonham, and T. Hemmer. 2020. Does it pay to ‘be like Mike’? Aspiratonal peer
firms and relative performance evaluation. Review of Accounting Studies 25 (4): 1507–
1541.
Bebchuk L. A., and J. M. Fried. 2003. Executive compensation as an agency problem. Journal of
Economic Perspectives 17 (3): 71–92.
Bebchuk L. A., J. M. Fried, and D. I. Walker. 2002. Managerial power and rent extraction in the
design of executive compensation. University of Chicago Law Review 69 (3): 751–846.
Bertrand M., and S. Mullainathan. 2001. Are CEOs rewarded for luck? The ones without
principals are. Quarterly Journal of Economics 116: 901–932.
Bettis J. C., J. Bizjak, J. L. Coles, and S. Kalpathy. 2018. Performance-vesting provisions in
executive compensation. Journal of Accounting and Economics 66 (1): 194–221.
Bizjak J., M. Lemmon, and T. Nguyen. 2011. Are all CEOs above average? An empirical
analysis of compensation peer groups and pay design. Journal of Financial Economics
100: 538–555.
Bizjak J. M., M. L. Lemmon, and L. Naveen. 2008. Does the use of peer groups contribute to
higher pay and less efficient compensation? Journal of Financial Economics 90 (2): 152–
168.
Bloomfield M., B. Gipper, J. D. Kepler, and D. Tsui. 2021. Cost shielding in executive bonus
plans. Journal of Accounting and Economics 72 (2–3): 1–24.
Bloomfield M. J., W. R. Guay, and O. Timmermans. 2022. Relative performance evaluation and
the peer group opportunity set. Working paper, The Wharton School of the University of
Pennsylvania.
Bloomfield M. J., M. Heinle, and O. Timmermans. 2023. Relative performance evaluation and
strategic peer-harming disclosures. Working paper, The Wharton School of the
University of Pennsylvania.
32
Carter M. E., C. D. Ittner, and S. L. Zechman. 2009. Explicit relative performance evaluation in
performance-vested equity grants. Review of Accounting Studies 14: 269–306.
Casas-Arce P., and F. A. Martinez-Jerez. 2009. Relative performance compensation, contests,
and dynamic incentives. Management Science 55: 1306–1320.
Celentani M., and R. Loveira. 2006. A simple explanation of the relative performance evaluation
puzzle. Review of Economic Dynamics 9 (3): 525–540.
Core J. E., and H. A. Packard. 2022. Non-price and price performance vesting provisions and
CEO incentives. The Accounting Review 97 (7): 109–134.
Curtis A., V. Li, and P. H. Patrick. 2021. The use of adjusted earnings in performance
evaluation. Review of Accounting Studies 26 (4): 1290–1322.
De Franco G., S. P. Kothari, and R. S. Verdi. 2011. The benefits of financial statement
comparability. Journal of Accounting Research 49 (4): 895–931.
Demski J. S., and H. Frimor. 1999. Performance measure garbling under renegotiation in
multiperiod agencies. Journal of Accounting Research 37: 187–214.
Dikolli S., C. Hofmann, and T. Pfeiffer. 2013. Relative performance evaluation and peer-
performance summarization errors. Review of Accounting Studies 18 (1): 34–65.
Dikolli S. S. 2001. Agent employment horizons and contracting demand for forward-looking
performance measures. Journal of Accounting Research 39: 481–494.
Dikolli S. S., V. Diser, C. Hofmann, and T. Pfeiffer. 2018. CEO power and relative performance
evaluation. Contemporary Accounting Research 35 (3): 1279–1296.
Dikolli S. S., and K. L. Sedatole. 2007. Improvements in the information content of nonfinancial
forward‐looking performance measures: A taxonomy and empirical application. Journal
of Management Accounting Research 19 (1): 71-104.
Drake K. D., and M. A. Martin. 2020. Implementing relative performance evaluation: The role of
life cycle peers. Journal of Management Accounting Research 32 (2): 107-135.
Dutta S., and S. Reichelstein. 2003. Leading indicator variables, performance measurement, and
long-term versus short-term contracts. Journal of Accounting Research 41: 837–866.
Faulkender M., and J. Yang. 2013. Is disclosure an effective cleansing mechanism? The
dynamics of compensation peer benchmarking. The Review of Financial Studies 26 (3):
806–839.
Feichter C., F. Moers, and O. Timmermans. 2022. Relative performance evaluation and
competitive aggressiveness. Journal of Accounting Research 60 (5): 1859–1913.
Feltham G. A., and J. Xie. 1994. Performance measure congruity and diversity in multi-task
principal-agent relations. The Accounting Review 69 (3): 429–453.
Fleckinger P. 2012. Correlation and relative performance evaluation. Journal of Economic
Theory 147 (1): 93–117.
Garvey G., and T. Milbourn. 2003. Incentive compensation when executives can hedge the
market: Evidence of relative performance evaluation in the cross section. Journal of
Finance 58: 1557–1581.
33
Garvey G. T., and T. T. Milbourn. 2006. Asymmetric benchmarking in compensation:
Executives are rewarded for good luck but not penalized for bad. Journal of Financial
Economics 82: 197–225.
Gibbons R., and K. J. Murphy. 1990. Relative performance evaluation for chief executive
officers. Industrial & Labor Relations Review 43: 30–51.
Gong G., L. Y. Li, and J. Y. Shin. 2011. Relative performance evaluation and related peer groups
in executive compensation contracts. The Accounting Review 86: 1007–1043.
Gopalan R., T. Milbourn, and F. Song. 2010. Strategic flexibility and the optimality of pay for
sector performance. The Review of Financial Studies 23 (5): 2060–2098.
Hayes R. M., and S. Schaefer. 2000. Implicit contracts and the explanatory power of top
executive compensation for future performance. Rand Journal of Economics 31 (2): 273–
293.
Hemmer T. 1996. On the design and choice of "modern" management accounting measures.
Journal of Management Accounting Research 8: 87.
———. 2023. Optimal dynamic relative performance evaluation. Journal of Management
Accounting Research: 1-15.
Hoberg G., and G. Phillips. 2016. Text-based network industries and endogenous product
differentiation. Journal of Political Economy 124 (5): 1423–1465.
Holmström B. 1979. Moral hazard and observability. Bell Journal of Economics 10 (1): 74–91.
Holmström B. 1982. Moral hazard in teams. Bell Journal of Economics 13 (2): 324–340.
Holzhacker M., S. Kramer, M. Matějka, and N. Hoffmeister. 2019. Relative target setting and
cooperation. Journal of Accounting Research 57 (1): 211–239.
Hung C.-Y., and Z. Shi. 2023. Peer-specific knowledge and peer group properties in relative
performance evaluation. Journal of Management Accounting Research: 1-29.
Indjejikian R., and D. Nanda. 1999. Dynamic incentives and responsibility accounting. Journal
of Accounting and Economics 27 (2): 177–201.
Indjejikian R. J., M. Matějka, K. A. Merchant, and W. A. Van der Stede. 2014a. Earnings targets
and annual bonus incentives. The Accounting Review 89 (4): 1227–1258.
Indjejikian R. J., M. Matějka, and J. Schloetzer. 2014b. Target ratcheting and incentives: Theory,
evidence, and new opportunities. The Accounting Review 89 (4): 1259–1267.
Janakiraman S. N., R. A. Lambert, and D. F. Larcker. 1992. An empirical investigation of the
relative performance evaluation hypothesis. Journal of Accounting Research 30 (1): 53–
69.
Jayaraman S., T. Milbourn, F. Peters, and H. Seo. 2021. Product market peers and relative
performance evaluation. The Accounting Review 96 (4): 341–366.
Jensen M. C., and K. J. Murphy. 1990. Performance pay and top-management incentives.
Journal of Political Economy 98: 225–264.
Laffont J. J., and J. Tirole. 1993. A theory of incentives in procurement and regulation.
Cambridge, Massachusetts: The MIT Press.
34
LaFond R., and S. Roychowdhury. 2008. Managerial ownership and accounting conservatism.
Journal of Accounting Research 46 (1): 101-135.
Lobo G. J., M. Neel, and A. Rhodes. 2018. Accounting comparability and relative performance
evaluation in CEO compensation. Review of Accounting Studies 23 (3): 1137–1176.
Magill M., and M. Quinzii. 2006. Common shocks and relative compensation. Annals of Finance
2 (4): 407–420.
Meyer M. A. 1995. Cooperation and competition in organizations - a dynamic perspective.
European Economic Review 39: 709–722.
Meyer M. A., and J. Vickers. 1997. Performance comparisons and dynamic incentives. Journal
of Political Economy 105 (3): 547–581.
Nam J. 2020. Financial reporting comparability and accounting-based relative performance
evaluation in the design of CEO cash compensation contracts. The Accounting Review 95
(3): 343–370.
Oyer P. 2004. Why do firms use incentives that have no incentive effects? Journal of Finance
59: 1619–1649.
Ozbas O., and D. S. Scharfstein. 2010. Evidence on the dark side of internal capital markets.
Review of Financial Studies 23: 581–599.
Pawliczek A. 2021. Performance-vesting share award outcomes and CEO incentives. The
Accounting Review 96 (5): 337–364.
Penman S. H., and X. J. Zhang. 2002. Accounting conservatism, the quality of earnings, and
stock returns. The Accounting Review 77 (2): 237–264.
Rajgopal S., T. Shevlin, and V. Zamora. 2006. CEOs' outside employment opportunities and the
lack of relative performance evaluation in compensation contracts. Journal of Finance
61: 1813–1844.
Said A. A., A. A. HassabElnaby, and B. Wier. 2003. An empirical investigation of the
performance consequences of nonfinancial measures. Journal of Management
Accounting Research 15: 193.
Saly P. J. 1994. Repricing executive stock-options in a down market. Journal of Accounting and
Economics 18 (3): 325–356.
Schäfer P. 2023. Relative performance evaluation and strategic differentiation. The Accounting
Review 98 (2): 419–453.
Tice F. M. 2023. The role of common risk in the effectiveness of explicit relative performance
evaluation. Management Science: forthcoming.
Vaan M. d., B. Elbers, and T. A. DiPrete. 2019. Obscured transparency? Compensation
benchmarking and the biasing of executive pay. Management Science 65 (9): 4299–4317.
Vrettos D. 2013. Are relative performance measures in CEO incentive contracts used for risk
reduction and/or for strategic interaction? The Accounting Review 88 (6): 2179–2212.
35
TABLE 1. Descriptive Statistics for the Sample Used in the Main Analysis
See Appendix A for variable definitions. Statistics for CEO Compensation, Assets, and Tenure are presented prior to taking the log
transformation. The main sample includes up to 43,849 firm-year observations with 1992–2021 data available in Execucomp and
Compustat. Lower sample size for some of the variables reflects missing values in additional data sources used.
Variable Obs. Mean SD Q1 Median Q3
CEO compensation 43,849 3,230 3,564 973 2,039 4,076
Firm stock return 43,849 0.056 0.407 -0.137 0.089 0.283
Firm ROA 43,849 0.042 0.095 0.012 0.043 0.085
SIC2-size peer stock return 43,797 0.074 0.259 -0.052 0.094 0.224
Favorable shock (SIC2-size peer stock )
43,797 0.676 0.468 0.000 1.000 1.000
SIC2-size peer ROA 41,757 0.011 0.106 0.007 0.033 0.062
Favorable shock (SIC2-size peer ROA )
41,757 0.780 0.414 1.000 1.000 1.000
Assets
43,848 7,271 20,290 416 1,295 4,546
CEO chair
43,849 0.542 0.498 0.000 1.000 1.000
Tenure
43,849 8.440 7.127 3.333 6.167 11.167
Ownership
42,874 0.500 0.500 0.000 0.500 1.000
Conglomerate
35,397 0.501 0.500 0.000 1.000 1.000
Predicted 43,077 0.230 0.204 0.071 0.154 0.341
Predicted peer stock return 43,077 0.037 0.261 -0.088 0.070 0.201
Favorable shock (Predicted peer stock )
43,077 0.625 0.484 0.000 1.000 1.000
Predicted peer ROA 43,077 0.026 0.071 0.013 0.037 0.063
Favorable shock (Predicted peer ROA )
43,077 0.843 0.364 1.000 1.000 1.000
36
TABLE 2. Descriptive Statistics for the Sample used for the Peer Choice Model
Table 2 presents descriptive statistics for the sample of all firm-year-peer observations with non-missing
1998–2021 data on the variables used in the estimation of the peer choice model. It includes 11,793 firm-
years and 4,353 unique (potential) peers. See Appendix A for variable definitions.
Variable Obs. Mean SD Q1 Median Q3
Rpeer
21,953,349 0.008 0.087 0.000 0.000 0.000
Ppeer
21,953,349 0.008 0.042 0.001 0.002 0.005
CorrD
21,953,349 0.007 0.084 0.000 0.000 0.000
CorrDP
21,953,349 0.039 0.193 0.000 0.000 0.000
CorrW
21,953,349 0.008 0.089 0.000 0.000 0.000
CorrWP
21,953,349 0.031 0.173 0.000 0.000 0.000
CorrQ
21,953,349 0.007 0.086 0.000 0.000 0.000
CorrQP
21,953,349 0.025 0.155 0.000 0.000 0.000
FF48
21,953,349 0.040 0.197 0.000 0.000 0.000
SIC2
21,953,349 0.023 0.149 0.000 0.000 0.000
SIC3
21,953,349 0.017 0.129 0.000 0.000 0.000
Segments
21,953,349 0.195 0.396 0.000 0.000 0.000
Proximity
21,953,349 0.032 0.177 0.000 0.000 0.000
TNIC
21,953,349 0.017 0.033 0.000 0.000 0.022
TalentFlow
21,953,349 0.000 0.011 0.000 0.000 0.000
LifeCycle
21,953,349 0.403 0.490 0.000 0.000 1.000
RelSize
21,953,349 0.250 0.433 0.000 0.000 1.000
37
TABLE 3. Peer Choice Model
Table 3 presents Logit model estimates from the sample described in Table 2. The
dependent variable, Rpeer, equals one if a potential peer is listed by a firm as a
compensation or performance benchmarking peer in a given year (firm-years
without Incentive Lab data on peers are dropped from the sample). See
Appendix A for variable definitions.
Intercept -6.857 140.900
CorrD
0.750 32.570
CorrDP
1.542 33.660
CorrW
0.480 17.460
CorrWP
0.758 19.970
CorrQ
0.295 7.720
CorrQP
0.392 8.720
FF48
1.014 9.570
SIC2
0.631 6.000
SIC3
0.275 2.400
Segments
0.699 15.320
Proximity
0.702 14.320
TNIC
8.802 16.830
TalentFlow
1.519 8.700
LifeCycle
0.301 14.740
RelSize
1.626 40.090
Pseudo - R
2
0.333
Observations 21,953,349
Rpeer
Coefficient t -statistic
38
TABLE 4. Validation Analysis
Panel A. The Likelihood of Selecting Peers with Different Characteristics
Panel A reports the unconditional mean of Rpeer (which equals one when a potential peer is listed by a firm as a
compensation or performance benchmarking peer in a given year) in the sample used in Tables 2 and 3 as well as
conditional means for various subsamples from the full sample of 21,953,349 firm-year-peer observations.
Panel B. Contemporaneous Associations between Firm and Peer Performance
***
,
**
represent significance at the 0.001 and 0.010 levels using standard errors clustered at the firm level. Panel B
examines the extent to which firm performance is contemporaneously correlated with alternative measures of peer
performance. Columns (1) and (3) use the sample of firm-year observations described in Table 1. Columns (2) and (4) use
a subsample with a high peer information content (above-median values of Predicted). Within-R
2
is the percentage of
variance explained that is not due to firm or year fixed effects. All variables are standardized to have zero mean and
variance of one to facilitate comparisons. See Appendix A for variable definitions.
Variable - statistic reported Obs. Mean
Rpeer - mean for all potential peers
21,953,349 0.008
Rpeer - mean for peers with the same SIC2 code
868,927 0.087
Rpeer - mean for peers with the same SIC2 code and RelSize quartile
229,088 0.208
Rpeer - mean for Predicted peers with the highest Ppeer values
229,089 0.313
(1) (2) (3) (4)
SIC2-size peer stock return
t
0.230
***
0.244
***
(0.000) (0.000)
Predicted peer stock return
t
0.409
***
0.441
***
(0.000) (0.000)
SIC2-size peer ROA
t
0.033
**
0.057
***
(0.001) (0.001)
Predicted peer ROA
t
0.464
***
0.569
***
(0.000) (0.000)
Standardized coefficients Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes
Firm fixed effects Yes Yes Yes Yes
Within - R
2
0.217 0.333 0.130 0.215
R
2
0.411 0.516 0.514 0.562
Observations 42,883 21,403 40,838 20,651
F
irm stock return
t
F
irm ROA
t
39
TABLE 4. Validation Analysis (Cont’d)
Panel C. Associations between Peer Performance and Future Firm Performance
***
,
*
represent significance at the 0.001 and 0.050 levels using standard errors clustered at the firm level. Panel C examines the
extent to which alternative measures of peer performance predict next year’s firm performance. Columns (1) and (3) use the
sample of firm-year observations described in Table 1. Columns (2) and (4) use a subsample with a high peer information content
(above-median values of Predicted). Within-R
2
is the percentage of variance explained that is not due to firm or year fixed
effects. All variables are standardized to have zero mean and variance of one to facilitate comparisons. See Appendix A for
variable definitions.
(1) (2) (3) (4)
SIC2-size peer stock return
t
-0.073
***
-0.085
***
(0.000) (0.000)
Predicted peer stock return
t
0.003 0.003
(0.819) (0.841)
SIC2-size peer ROA
t
0.029
*
0.081
***
(0.010) (0.000)
Predicted peer ROA
t
0.230
***
0.274
***
(0.000) (0.000)
Standardized coefficients Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes
Firm fixed effects Yes Yes Yes Yes
Within - R
2
0.004 0.005 0.035 0.061
R
2
0.268 0.293 0.484 0.493
Observations 34,768 17,820 33,144 17,184
F
irm stock return
t+1
F
irm ROA
t+1
40
TABLE 5. Peer ROA versus Peer Stock Returns in RPE
***
,
**
represent significance at the 0.001 and 0.010 levels, respectively, using standard errors clustered at the firm level.
Table 5 estimates a model of CEO compensation as a function of firm performance, peer performance, and control variables.
Column (1) presents coefficient estimates using SIC2-size peer stock return and SIC2-size peer ROA. Column (2) estimates
the same model as column (1) but presents standardized coefficient estimates for all performance variables to allow for a
comparison of the relative magnitudes of their effects. Column (3) uses Predicted peer stock return and Predicted peer ROA
and presents their standardized coefficients. Within-R
2
is the percentage of variance explained that is not due to firm or year
fixed effects. See Appendix A for variable definitions.
(1) (2) (3)
Firm stock return 0.156
***
0.064
***
0.063
***
(0.000) (0.000) (0.000)
Firm ROA 0.745
***
0.071
***
0.072
***
(0.000) (0.000) (0.000)
Peer stock return -0.082
***
-0.021
***
-0.019
***
(0.000) (0.000) (0.000)
Peer ROA 0.288
***
0.030
***
0.023
**
(0.000) (0.000) (0.002)
Assets
0.364
***
0.364
***
0.365
***
(0.000) (0.000) (0.000)
CEO chair
0.044
***
0.044
***
0.040
**
(0.001) (0.001) (0.002)
Tenure
0.033
***
0.033
***
0.036
***
(0.000) (0.000) (0.000)
Ownership
0.020 0.020 0.023
(0.136) (0.136) (0.091)
Standardized coefficients No Yes Yes
Year fixed effects Yes Yes Yes
Firm fixed effects Yes Yes Yes
Within - R
2
0.119 0.119 0.118
R
2
0.741 0.741 0.738
Observations 40,654 40,654 41,958
CEO compensation
SIC2-size peers SIC2-size peers Predicted peers
41
TABLE 6. RPE in Conglomerates and Single-Segment Firms
***
,
**
represent significance at the 0.001 and 0.010 levels, respectively, using standard errors clustered at the firm
level. We estimate an extended model of CEO compensation using the same control variables as in Table 5. Peer
stock return (ROA) represent Predicted peer stock return (ROA) in the first column and SIC2-size peer stock return
(ROA) in the second column. Within-R
2
is the percentage of variance explained that is not due to firm or year fixed
effects. See Appendix A for variable definitions.
Firm stock return 0.064
***
0.064
***
(0.000) (0.000)
Firm ROA 0.075
***
0.072
***
(0.000) (0.000)
Conglomerate 0.007 0.017
(0.706) (0.355)
Peer stock return -0.021
**
-0.028
***
(0.001) (0.000)
Peer stock return ꞏ Conglomerate -0.006 0.005
(0.422) (0.413)
Peer ROA 0.005 0.027
**
(0.554) (0.002)
Peer ROA ꞏ Conglomerate 0.040
**
0.009
(0.007) (0.436)
Control variables Yes Yes
Year fixed effects Yes Yes
Firm fixed effects Yes Yes
Within - R
2
0.123 0.124
R
2
0.740 0.744
Observations 33,729 32,609
Predicted peers SIC2-size peers
CEO compensation
42
TABLE 7. The RPE Asymmetry
***
,
**
,
*
represent significance at the 0.001, 0.010, and 0.050 levels, respectively, using standard errors clustered at the
firm level. We estimate an extended model of CEO compensation using the same control variables as in Table 5.
Peer stock return (ROA) represent Predicted peer stock return (ROA) in the first column and SIC2-size peer stock
return (ROA) in the second column. Within-R
2
is the percentage of variance explained that is not due to firm or year
fixed effects. See Appendix A for variable definitions.
Firm stock return 0.063
***
0.064
***
(0.000) (0.000)
Firm ROA 0.071
***
0.070
***
(0.000) (0.000)
Peer stock return -0.053
***
-0.050
***
(0.000) (0.000)
Favorable shock (stock) 0.025
*
0.029
*
(0.010) (0.010)
Peer stock return ꞏ Favorable shock (stock) 0.049
***
0.035
**
(0.000) (0.010)
Peer ROA -0.023
*
0.020
*
(0.038) (0.013)
Favorable shock (ROA) 0.035
*
0.027
*
(0.042) (0.030)
Peer ROA ꞏ Favorable shock (ROA) 0.154
***
0.029
(0.000) (0.117)
Control variables Yes Yes
Year fixed effects Yes Yes
Firm fixed effects Yes Yes
Within - R
2
0.123 0.119
R
2
0.740 0.741
Observations 41,958 40,654
Predicted peers SIC2-size peers
CEO compensation
43
TABLE 8. The RPE Asymmetry and Peer Information Content
***
,
**
,
*
represent significance at the 0.001, 0.010, and 0.050 levels, respectively, using standard errors
clustered at the firm level. The first column estimates the same model as Table 7 in the subsample of
observations with a high peer information content (Predicted low = 0). The second column presents similar
results in the full sample after including the main and interaction effects of Predicted low.
Predicted low is an indicator for below-median predicted values from model (1), which capture the ex ante
likelihood of being listed as a peer in the proxy statement. See Appendix A for other variable definitions.
Firm stock return 0.070
***
0.063
***
(0.000) (0.000)
Firm ROA 0.088
***
0.071
***
(0.000) (0.000)
Peer stock return -0.066
***
-0.058
***
(0.000) (0.000)
Favorable shock (stock) 0.035
**
0.023
(0.009) (0.067)
Peer stock return ꞏ Favorable shock 0.050
**
0.050
**
(0.006) (0.003)
Peer ROA -0.058
**
-0.051
***
(0.001) (0.001)
Favorable shock (ROA) 0.082
**
0.073
**
(0.002) (0.005)
Peer ROA ꞏ Favorable shock (ROA) 0.179
***
0.195
***
(0.000) (0.000)
Predicted low 0.021
(0.594)
Peer stock return ꞏ Predicted low 0.013
(0.376)
Favorable shock (stock) ꞏ Predicted low 0.002
(0.901)
Peer stock return ꞏ Favorable shock (stock) ꞏ Predicted Low -0.007
(0.754)
Peer ROA ꞏ Predicted low 0.045
**
(0.006)
Favorable shock (ROA) ꞏ Predicted low -0.059
(0.070)
Peer ROA ꞏ Favorable shock (ROA) ꞏ Predicted low -0.073
*
(0.013)
Control variables Yes Yes
Year fixed effects Yes Yes
Firm fixed effects Yes Yes
Within - R
2
0.106 0.124
R
2
0.741 0.740
Observations 21,022 41,958
CEO compensation
Predicted peers Predicted peers
44
TABLE 9. The RPE Asymmetry and Corporate Governance
***
,
**
,
*
represent significance at the 0.001, 0.010, and 0.050 levels, respectively, using standard errors
clustered at the firm level. The first column estimates the same model as Table 7 in the subsample of
observations with weak governance (Strong governance = 0), as reflected in compensation committees having
a large (above-median) number of members and many (above-median percentage) busy members. The second
column presents similar results in the full sample after including the main and interaction effects of Strong
governance.
Firm stock return 0.054
***
0.057
***
(0.000) (0.000)
Firm ROA 0.082
***
0.056
***
(0.000) (0.000)
Peer stock return -0.048
*
-0.050
**
(0.041) (0.002)
Favorable shock (stock) 0.016 0.011
(0.403) (0.510)
Peer stock return ꞏ Favorable shock 0.051 0.056
*
(0.078) (0.013)
Peer ROA -0.082
*
-0.066
*
(0.017) (0.012)
Favorable shock (ROA) 0.105
*
0.086
*
(0.012) (0.020)
Peer ROA ꞏ Favorable shock (ROA) 0.191
***
0.220
***
(0.000) (0.000)
Strong governance 0.036
(0.417)
Peer stock return ꞏ Strong governance -0.007
(0.670)
Favorable shock (stock) ꞏ Strong governance 0.021
(0.328)
Peer stock return ꞏ Favorable shock (stock) ꞏ Strong governance 0.007
(0.783)
Peer ROA ꞏ Strong governance 0.050
*
(0.049)
Favorable shock (ROA) ꞏ Strong governance -0.044
(0.279)
Peer ROA ꞏ Favorable shock (ROA) ꞏ Strong governance -0.099
**
(0.008)
Control variables Yes Yes
Year fixed effects Yes Yes
Firm fixed effects Yes Yes
Within - R
2
0.075 0.104
R
2
0.734 0.760
Observations 10,354 28,297
CEO compensation
Predicted peers Predicted peers
45
TABLE 10. The RPE Asymmetry and Explicit RPE Incentive Grants
***
,
**
,
*
represent significance at the 0.001, 0.010, and 0.050 levels, respectively, using standard errors clustered at the firm level.
The first column estimates the same model as Table 7 after excluding observations with explicit RPE incentive grants listed in
Incentive Lab. The second presents the estimation results for the observations excluded from the first column. Peer performance
is calculated using the peers disclosed in the explicit RPE incentive grants. The third column uses the subsample of observations
from 1992 to 2005, a time period with little or no explicit RPE incentive grants.
Firm stock return 0.063
***
0.049
**
0.069
***
(0.000) (0.002) (0.000)
Firm ROA 0.070
***
0.063
**
0.091
***
(0.000) (0.002) (0.000)
Peer stock return -0.051
***
-0.020 -0.058
***
(0.000) (0.391) (0.000)
Favorable shock (stock) 0.024
*
-0.009 0.020
(0.021) (0.746) (0.196)
Peer stock return ꞏ Favorable shock (stock) 0.048
***
-0.011 0.034
(0.001) (0.773) (0.087)
Peer ROA -0.019 -0.050
*
-0.017
(0.098) (0.031) (0.351)
Favorable shock (ROA) 0.035 0.114
*
-0.008
(0.053) (0.041) (0.792)
Peer ROA ꞏ Favorable shock (ROA) 0.154
***
0.113
**
0.192
***
(0.000) (0.004) (0.000)
Control variables Yes Yes Yes
Year fixed effects Yes Yes Yes
Firm fixed effects Yes Yes Yes
Within -
R
2
0.121 0.082 0.098
R
2
0.731 0.772 0.745
Observations 38,481 2,411 17,796
CEO compensation
Disclosed peers Predicted peersPredicted peers
46
APPENDIX A
Variable Definition
CEO compensation natural logarithm of 1 + tdc1, where tdc1 is total CEO compensation
(in $ thousands).
Firm stock return natural logarithm of [(1 + ret / 100) / (1 + cpi)], where ret is the compounded
annual stock return obtained from monthly data and cpi is the rate of inflation.
Firm ROA natural logarithm of (1 + ib / at), where ib is income before extraordinary items
and at are beginning-of-year total assets, both adjusted for inflation.
Peer stock return
(SIC2-size)
peer returns are calculated as in Firm stock return and averaged for all peers
with the same SIC2 code and size quartile.
Favorable shock
(peer stock)
an indicator for Peer stock return > 0.
Peer ROA
(SIC2-size)
peer ROA is calculated as in Firm ROA and averaged for all peers with the
same SIC2 code and size quartile.
Favorable shock
(peer ROA)
an indicator for Peer ROA > 0.
Assets natural logarithm of total assets (in $ millions).
CEO chair an indicator for CEO being also the board chai
r
.
Tenure natural logarithm of the number of years since the CEO took office.
Ownership an indicator for an above-median percentage of shares owned by the CEO.
Conglomerate an indicator for firms with positive sales and assets in more than one SIC3 code.
Predicted low an indicator for below-median value of Predicted (defined below).
Strong governance an indicator for compensation committees with below-median number of
members or below-median percentage of busy members.
Variables used in the peer choice model in Table 3
Rpeer an indicator for a peer disclosed by a firm as a compensation or performance
b
enchmarking peer in a given year.
Ppeer predicted value from the peer choice model in (1), i.e., the ex ante likelihood of
b
eing listed as a peer in the proxy statement of a firm-yea
r
.
CorrD an indicator for a peer with one of the 20 highest correlations between firm and
peer daily stock returns in a given firm-yea
r
.
CorrDP an indicator for a peer with CorrD = 1 in at least one other year during the
1992–2021 sample period.
CorrW an indicator for a peer with one of the 20 highest correlations between firm and
peer weekly stock returns over a period of five calendar years ending with the
current yea
r
.
CorrWP an indicator for a peer with CorrW = 1 in at least one other year during the
1992–2021 sample period.
47
CorrQ an indicator for a peer with one of the 20 highest correlations between firm and
peer correlations in quarterly sales changes over a period of ten calendar years
ending with the current yea
r
.
CorrQP an indicator for a peer with CorrQ = 1 in at least one other year during the
1992–2021 sample period.
FF48 an indicato
r
for a firm-yea
r
-peer match in the Fama French 48 code.
SIC2 an indicato
r
for a firm-yea
r
-peer match in SIC2 and a different SIC3.
SIC3 an indicato
r
for a firm-yea
r
-peer match in SIC3.
Segments an indicator for a firm-year-peer match in diversification strategy, i.e., in both
b
eing single-segment firms or both being conglomerates in a given year.
Proximity an indicator for firm-yea
r
-peer headquarters distance less 100 miles.
TNIC firm-yea
r
-peer product similarity score from Hoberg and Phillips (2016).
TalentFlows an indicator equal to one if at least one of the named executive officers in
Execucomp moved between the firm and the peer in the last five years.
LifeCycle an indicator equal to one if the firm and the peer are in the same life cycle stage.
RelSize an indicator for firm-year-peer similarity in terms of annual sales, i.e., the
lowest quartile of the absolute value of (psale – sale)/sale, where psale stands
for annual peer sales and sale stands for firm sales.
Variables defined based on the peer choice model
Predicted firm-year average of Ppeer for the 20 highest predicted values from model (1).
Peer stock return
(Predicted)
peer performance calculated as in Firm stock return and averaged for the 20
peers with the highest Ppeer for a given firm-year.
Peer ROA
(Predicted)
peer performance calculated as in Firm ROA and averaged for the 20 peers with
the highest Ppeer for a given firm-year.
48
APPENDIX B
Alternative Measures of Peer Performance
The main analysis compares our measures of Predicted peer performance to the most commonly used
measures of peer performance defined in terms of SIC2-size. Several recent studies show that using other
information to define peers can increase the power of empirical tests, even though the additional data
requirements reduce the sample size available for estimation. In what follows, we discuss two such
alternative measures of peer performance and present validation results similar to those in Table 4.
We define ALT1 peer stock return and ALT1 peer ROA as in Nam (2020) using the measure of
financial reporting comparability (FRC) from De Franco et al. (2011). Specifically, from each SIC2-size
peer group, we select those that have the highest quartile of FRC and from this subset further select the
top ten peers (minimum of five) that are closest to the focal firm in terms of market value.
We define ALT2 peer stock return and as in Jayaraman et al. (2021) using the measure of product
similarity from Hoberg and Phillips (2016). Specifically, we start from the peer groups defined by the
text-based network industry classifications of Hoberg and Phillips (2016). We then select one quarter of
the peer group (minimum of two peers) that is closest to the focal firm in terms of Mahalanobis distance
calculated from market value and book-to-market ratio. ALT2 peer ROA is calculated as all other peer
ROA measures using the same peer groups as ALT2 peer stock return.
Validation Results
Table B1 shows that Predicted peer stock return performs much better than ALT1 peer stock return
when explaining Firm stock return, particularly in column (2), which uses the subsample of Predicted
peers with a high information content. Predicted peer ROA also has slightly higher standardized
coefficients than ALT1 peer ROA when explaining Firm ROA in columns (3) and (4). Both measures of
peer earnings are significant predictors of next year’s firm earnings but Predicted peer ROA performs
better, particularly in the high information content subsample in column (6).
Table B2 shows that ALT2 peer stock return performs slightly better than Predicted peer stock return
in columns (1) and (2) but much worse in all other tests. Specifically, in column (1), the standardized
49
coefficient estimates for the contemporaneous association between Firm stock return and ALT2 peer stock
return is 0.390 (p < 0.001) as compared to 0.312 (p < 0.001) for Predicted peer stock return. The
difference is less pronounced in column (2). However, the standardized coefficient estimates for
Predicted peer ROA are several times larger than those for ALT2 peer ROA in columns (3)–(6), which
suggests that ALT2 peer ROA has much less information content about current and future firm earnings.
50
TABLE B1. Validation Analysis with Peers Based on Financial Reporting Comparability
***
represent significance at the 0.001 level using standard errors clustered at the firm level. The validation tests presented are similar to those in Table 4. ALT1 peer
stock return is defined as in Nam (2020) based on financial reporting comparability. See Appendix A for other variable definitions. Columns (1), (3), and (5) use all
available observations from 1992–2021. Columns (2), (4), and (6) use subsamples with a high peer information content (above-median values of Predicted).
(1) (2) (3) (4) (5) (6)
ALT1 peer stock return
t
0.079
***
0.055
***
(0.000) (0.000)
Predicted peer stock return
t
0.498
***
0.556
***
(0.000) (0.000)
ALT1 peer ROA
t
0.275
***
0.248
***
0.114
***
0.115
***
(0.000) (0.000) (0.000) (0.000)
Predicted peer ROA
t
0.313
***
0.394
***
0.146
***
0.207
***
(0.000) (0.000) (0.000) (0.000)
Standardized coefficients Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
Firm fixed effects Yes Yes Yes Yes Yes Yes
Within - R
2
0.222 0.328 0.201 0.278 0.038 0.069
R
2
0.441 0.535 0.555 0.607 0.486 0.513
Observations 21,611 12,814 21,611 12,814 17,492 10,579
F
irm stock return
t
F
irm ROA
t
F
irm ROA
t+1
51
TABLE B2. Validation Analysis with Peers Based on Product Similarity
***
represent significance at the 0.001 level using standard errors clustered at the firm level. The validation tests presented are similar to those in Table 4. ALT2 peer
stock return is defined as in Jayaraman et al. (2021) based on product similarity. See Appendix A for other variable definitions. Columns (1), (3), and (5) use all
available observations from 1992–2021. Columns (2), (4), and (6) use subsamples with a high peer information content (above-median values of Predicted).
(1) (2) (3) (4) (5) (6)
ALT2 peer stock return
t
0.390
***
0.384
***
(0.000) (0.000)
Predicted peer stock return
t
0.312
***
0.349
***
(0.000) (0.000)
ALT2 size peer ROA
t
0.119
***
0.159
***
0.063
***
0.109
***
(0.000) (0.000) (0.000) (0.000)
Predicted peer ROA
t
0.459
***
0.555
***
0.219
***
0.266
***
(0.000) (0.000) (0.000) (0.000)
Standardized coefficients Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
Firm fixed effects Yes Yes Yes Yes Yes Yes
Within - R
2
0.290 0.371 0.150 0.230 0.037 0.060
R
2
0.472 0.546 0.536 0.570 0.502 0.491
Observations 28,696 16,401 28,688 16,400 23,350 13,672
F
irm stock return
t
F
irm ROA
t
F
irm ROA
t+1
52
APPENDIX C
TABLE C1. CEO Cash Compensation and RPE in Conglomerates
***
,
**
,
*
represent significance at the 0.001, 0.010, and 0.050 levels, respectively, using standard errors clustered at
the firm level. We estimate the same model as in Table 6 except that the dependent variable is CEO cash
compensation, defined as salary, bonus, long-term incentive payouts (before 2007), and non-equity incentive plan
payouts (after 2006). See Appendix A for other variable definitions.
Firm stock return 0.098
***
0.097
***
(0.000) (0.000)
Firm ROA 0.076
***
0.082
***
(0.000) (0.000)
Conglomerate 0.015 0.022
(0.336) (0.175)
Peer stock return -0.019
***
-0.018
***
(0.000) (0.000)
Peer stock return ꞏ Conglomerate 0.003 0.006
(0.607) (0.161)
Peer ROA 0.020
**
0.011
(0.002) (0.060)
Peer ROA ꞏ Conglomerate 0.067
***
0.017
*
(0.000) (0.029)
Control variables Yes Yes
Year fixed effects Yes Yes
Firm fixed effects Yes Yes
Within - R
2
0.171 0.167
R
2
0.777 0.779
Observations 33,607 32,450
CEO cash compensation
Predicted peers SIC2-size peers
53
TABLE C2. CEO Cash Compensation and the RPE Asymmetry
***
,
**
,
*
represent significance at the 0.001, 0.010, and 0.050 levels, respectively, using standard errors clustered at
the firm level. We estimate the same model as in Table 7 except that the dependent variable is CEO cash
compensation, defined as salary, bonus, long-term incentive payouts (before 2007), and non-equity incentive plan
payouts (after 2006). See Appendix A for other variable definitions.
Firm stock return 0.099
***
0.099
***
(0.000) (0.000)
Firm ROA 0.072
***
0.078
***
(0.000) (0.000)
Peer stock return -0.026
***
-0.028
***
(0.000) (0.000)
Favorable shock (stock) 0.015
*
0.021
**
(0.026) (0.008)
Peer stock return ꞏ Favorable shock (stock) -0.001 0.007
(0.897) (0.496)
Peer ROA -0.021
**
0.003
(0.002) (0.544)
Favorable shock (ROA) 0.084
***
0.022
*
(0.000) (0.018)
Peer ROA ꞏ Favorable shock (ROA) 0.158
***
0.061
***
(0.000) (0.000)
Control variables Yes Yes
Year fixed effects Yes Yes
Firm fixed effects Yes Yes
Within - R
2
0.169 0.161
R
2
0.769 0.767
Observations 41,809 40,472
CEO cash compensation
Predicted peers SIC2-size peers
