Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium IVIember Libraries
http://www.archive.org/details/relativeperformaOOgibb
of economics
RELATIVE PERFORMANCE EVALUATION
FOR CHIEF EXECUTIVE OFFICERS
Robert Gibbons
Kevin J Murphy
.
No.
532
August 1989
massachusetts
institute of
technology
50 memorial drive
Cambridge, mass. 02139
RELATIVE PERFORMANCE EVALUATION
FOR CHIEF EXECUTIVE OFFICERS
Robert Gibbons
Kevin J Murphy
.
No.
532
August 1989
RELATIVE PERFORMANCE EVALUATION
FOR CHIEF EXECUTIVE OFFICERS*
Robert Gibbons
MIT and NDER
Kevin
J.
Murphy
University of Rochester
August. 1989
(forthcoming, Industrial
and Labor Relations Rcx'icw, 1990)
Abslract
Measured individual performance often depends on random factors which also
same firm, industn,', or market. In tiiese
cases, relative performance evaluation (RPE) can provide incentives while partially
insulating workers from the common uncenainty. Basing pay on relative performance,
however, generates incentives to sabotage the measured performance of co-workers, to
affect the performances of other workers in the
collude with co-workers and shirk, and to apply for jobs with inept co-workers. RPE
contracts also are less desirable when the output of co-workers is expensive to measure
or in the presence of production externalities, as in the case of team production.
The purpose of this paper is to review tlie benefits and costs of RPE and to test for
the presence of RPE in one occupation where the benefits plausibly exceed the costs: toplevel management. Rewarding chief executive officers (CEOs) based on performance
measured
relative to the industry or
market creates incentives
to lake actions increasing
shareholder wealth while insuring executives against the vagaries of the stock and
product markets that are beyond their control. We expect RPE to be a common feature of
implicit CEO compensation and dismissal contracts because the potential benefit of
filtering out common uncertainty is high, the cost of measuring the performance of other
firms is small, and opportunities for sabotage and collusive shirking are limited.
bur empirical evidence strongly supports the RPE
pay revisions and retention probabilities are positively and
to firm performance, but are negatively and significantly related to
In contrast to previous research,
hypothesis
— CEO
significantly related
industry and market performance, ceteris paribus. Our results also suggest that CEO
performance is more likely to be evaluated relative to aggregate market movements than
relative to industry
We are grateful
many
movements.
comments from John Abowd, Joshua Angrist, James Bricklcy, David Card,
Lawrence Katz, AJan Krueger, Edward Lazear, Meg Meyer,
Sherwin Rosen, Sharon Smith, Ross Wans, and Michael Weisbach as well as from participants in seminars
for
helpful
Jeffrey Coles, Ronald Ehrcnbcrg, Henry Farbcr,
NBER, NYU,
and the Princeton Labor Lunch. This research is supported by the Sloan Foundation (both
(grant SES 88-09200) and the Industrial Relations Section at Princeton University
(Gibbons); and the John M. Olin Foundation and the Managerial Economics Research Center, University of
Rochester (Niurphy).
at
authors); the
NSF
RELATIVE PERFORMANCE EVALUATION
FOR CHIEF EXECUTIVE OFFICERS
Robert Gibbons and Kevin
It is
for their
common
for workers to be
Murphy
rewarded not simply for
performance measured relative
explicit contests
J.
to the
their
own performance
perfonnances of co-workers.
and tournaments, the allocation of
a fixed
bonus pool
a
are
awarded
to
one
group of subordinates.
Economic
risk-sharing.
Examples
rhcrcorporate or
at
divisional level, forced-distribution performance appraisals, and promotions
worker from
but rather
theorj' provides a rationale for relative
performance evaluation (RPE) based on
Compensation based solely on observed individual performance provides
incentives for a worker to be productive, but measured performance (and therefore
almost always affected by random factors beyond the worker's control. The
compensation)
is
random
which
factors
other workers in the
affect
same
one worker's perfomiance often
division, firm, industr}', or market.
incentives while panially insulating the worker from the
Although
uncertainty,
it
RPE
affect the
offers benefits
also entails costs.
when
observed performances of
In these cases,
common
RPE
can provide
uncenainty.
there exists an important source of
common
risk or
Basing pay on relative performance generates incentives
to
sabotage the measured performance of co-workers (or any other reference group), to collude
with co-workers and shirk, and to apply for jobs v/ith inept co-workers. Relative-performance
contracts also are less desirable
when
the output of co-workers
is
expensive to measure or when
there are production externalities, as in the case of team production.
One occupation
for
which the risk-sharing advantages of
counterproductive side-effects
is
top-level corporate
RPE
likely
exceed
its
management. Executive compensation
contracts based only on firm performance subject executives to vagaries of the stock and product
markets that are clearly beyond management control. Measuring managerial performance relative
Robert Gibbons AND
to the
Kevi>:
J.
performance of other firms
same
in the
from such industry and market shocks.
(CEOs)
officers
are inexpensive to administer, since the stock performance of rival firms
CEOs
in rival finns,
The purpose of
this
paper
In addition,
RPE
in the
test for the
is to
presence of
who developed
who
RPE
in
We begin
CEOs
the
economic theory of
tested for the existence of
Our debt
are
with a review of the benefits and costs
We
than one thousand publicly held corporations.
in
relative
RPE
agency theory, especially Holmstrom (1982),
performance evaluation, and Antle and Smith
executive compensation contracts.
in
We
extend
as well as the benefits of relative-performance
to other related research is discussed
The Theory
A
which
executive compensation by analyzing a longitudinal sample of
Holmstrom's analysis by discussing the costs
contracts.
tend to have limited
investigate empirically the extent to
This paper builds on previous research
(1986),
CEOs
workplace, and then apply the analysis to the case of top-level management.
CEOs from more
over 1,600
because
is
sabotage and collusive shirking seem unlikely.
rewarded on the basis of relative performance.
of
industry or market partially insulates executives
Relative-performance contracts for chief executive
available on a daily basis at trivial cost.
interaction with
August. 1989
2
MuRPinr
below.
of Relative Performance Evaluation
Simple Model of the Benefits of RPE
This section presents an informal review of the economic theory of relative performance
evaluation.
Early papers on this subject include Baiman and
Rosen (1981); Wolfson (1985) applies
analysis in this section applies to
theory to top-level
Suppose
example,
management
all
Demski (1980) and Lazear and
the theory to shared-equity real estate contracts.
employees
in a
workplace;
we
The
defer our application of the
until the next section.
that output, y,
depends on the worker's
effort, a, but is stochastic.
For
ROBERTClBDONSANDKEVINJ.MURPin'
AUGUST, 1989
3
y = a +
(1)
where E
is
a noise
term with zero mean.
e,
Because the worker's
employer and worker cannot sign a contract specifying
of effort.
Instead, the
employer
offers the
that the
worker
a
effort is unobservable, the
employee supply
a particular level
compensation contract, w(y),
that
determines the worker's wage as a function of the realized output of the production process. The
worker then chooses whether
utility IJO.
to sign the contract or to take
up alternative employment yielding
After signing the contract, the worker chooses a level of effort, and production
"""
"
occurs.
Suppose
that
compensation, y-w.
compensation
risk.
the
at
is
The worker, on
risk-neutral:
full
efficiency (the 'Tirst-best")
must supply the productively
which the expected value of
he cares only about output net of
the other hand, dislikes effort and
Under these assumptions,
conditions: (a) the worker
of effort
employer
is
defined by two
efficient level of effort (that
the output resulting
averse to
is
from an additional
is,
the level
unit of effort
equals the disutility to the worker from this additional unit of effon); and (b) the worker must be
perfectly insured against compensation risk.
If the
worker were
risk-neutral, insurance
be achieved via a contract w(y) = y -
F
would be unnecessary and
that, in effect, sells the
the first-best could
worker the firm:
the
worker
receives the entire output, y, as compensation, but pays the employer an up-front fee, F, so that
the worker's expected utility just equals his reservation utility.
however, the optimal contract
income
risk
is
When
workers are risk-averse,
necessarily "second-best," both because the worker bears
and because the worker's
effort choice is smaller than is productively efficient.
The
optimal contract for a risk-averse worker reflects a trade-off between the goals of insuring the
worker and providing incentives for
The
efficient effort choice.
fact that the best possible contract of the
form
w(y') induces a risk-averse
supply less than the productively efficient amount of effort suggests that employers
to
y.
improve the contract by making pay contingent on a second variable,
Holmstrom (1979) showed
that the optimal contract has the
z, in
worker
may
to
be able
addition to output,
form w(y,z) whenever z contains
Robert Gibbons AND Kevin
J.
August. 1989
4
MuRPinr
incremental information valuable in assessing the worker's unobservable effort.' The intuition
for including z in the contract w(y,z) is either to provide
more insurance without reducing
incentives for effort or to provide greater incentives v-dthout increasing risk exposure.
As
suppose
a concrete example,
that the worker's output
idiosyncratic noise, E, but also on an additional noise term,
experienced by some
set
y = a + E +
6
clearly
where
is
a
common shock
e.
could reflect seasonal or random variations in industry demand, market-wide stock
movements, or the fixed-effect of
workers,
9,
among
other possibilities.
a single supervisor evaluating
If 6 is perfectly
each
member
of a group of
observable, contracts of the form w(y-G)
improve on contracts of the form w(y). The presence of 9 makes y a noisier signal about
effort than
it
would be
in the
absence of
Therefore, the employer can reduce the noise in his
9.
evaluation of the worker's effort by basing contracts on y-9 rather than on y alone. Even
common
shock
8 is not
perfecdy observable, the employer
form w(y) by incorporating an estimate of
the
and
effort
of workers in the same division, firm, industry, or market:
(2)
The shock
depends not only on
common
shock
9 with error,
9,
denoted by
9,
still
if
the
can improve on contracts of the
into the contract.
Since 9 measures
optimal contracts incorporating the estimate 9 will generally be
of the form w(y,9) rather than of the simpler form w(y-0), and the
common
risk will
be only
partially filtered out of the worker's compensation.
To
see
how
might be computed, consider the following elaboration of the example
above.2 Suppose that there are n identical workers whose outputs are given by
for i=l,...,n,
where
Then, assuming
estimate of the
subject to the
^
More
yi
=
ai
-f-
ej
-f-
9,
£i are
the i^ worker's effort and idiosyncratic noise, respectively.
that the Ej's are
independendy and identically distributed with zero mean, an
aj
and
common
common
shock
shock
is 9
0,
= y
and a
precisely, the optimal contract will
-
is
a,
where y
is
the average output of
all
other workers
the average anticipated effort level of these n-1 other
depend on z
if
y
is
not a sufficient
statistic for the pair
(y^) with
respect 10 ihe woricer's efforL
•^
AUemaiive methods of calculating
9 applicable to special situations include seasonal adjustments and
prorating output to adjust for unanticipated plant closings.
Robert Gibbons AND Kevin J. MuRTifY
August, 1989
5
workers conditional on their compensation packages. ^ Under the assumption
that the
normally distributed, Holmstrom (1982) showed that the optimal contract for the
WjCyi.y).''
Typically, the
wage
paid to the i^ worker will be positively related to his
but negatively related to the average output of
Holmstrom
shock.
also
very precise estimate of
showed
G),
i'^
that as the
all
other workers subject to the
number of workers becomes
worker
own
same common
ver)' large (so that
performance of the i^ worker measured relative to the performance of others
is
output
form Wi(yi-y), where yj-y
the optimal contract approaches the
are
Ei's
is a
is
the
in the reference
group.
Relative performance evaluation
may be
beneficial even in the absence of risk-aversion
and incentive problems. Alternative models based on learning about
also predict
RPE when
individual output,
yi,
workers
depend on
in a
ability
group are exposed
and not
(3)
yi
where
ri; is
distributed with
that ability is not
known
mean zero and variance
but uninformative.)
Suppose
o^'^.
common
shock.
=
Tli
+
ei
+
and
is
ihe unobser\'able
(Allowing for a non-zero mean
common shock
common
is
and noise terms are independent. Given these assumptions,
(see, for
example. Theorem
and Graybill, 1963)
straightforward
mean
zero and
and the n workers' individual
abilities
Mood
it is
that the
straightforward to
The
worker.
A
more
=y
efficient
If the joint distribution
likelihood estimate of 9.
9, is as
(4).
estimate in the text, 8
other workers.
show
expected ability of the i^
worker, given his output and the outputs of the n-1 co-workers subject to the shock
described by equation
let
precisely; rather, riiis believed to be normally
variances oe^ and c^}, respectively, and that the
of
Formally,
e.
further that 9 and £[ are normally distributed with
9. 11
example,
effort:
the i^ worker's ability, Eiis idiosyncratic noise,
Suppose
shock.
to a
ability rather, for
-
a,
OLS
is
the
OLS
estimate of 6 given the available information about the n-1
estimate of 9 would also include the information available about the
of the idiosyncratic noise terms
One can
is
knov/n, one can also compute the
i'*"
maximum
relax the assumption that the idiosyncratic noise terms are independently and
identically distributed.
Note
that y; plays
two roles
in this contract.
improves the estimate of 9 as described
in footnote 3.
First,
it
provides incentives
in the
usual way.
Second,
it
RobertGibbons andKevinJ. MuRPinr
Eh,ly,,...,yn]
(4)
August, 1989
6
= ,^2 I
^2
Vi
^^T^^^J^l^
'^0"
^J^
j=l
'
V
Suppose
that individual outputs are observable
Then
wage offered
the
negatively related to
to the i^
tlie
average output of
In both the incentive
risk
worker will
model and
from workers' compensation
all
that
tj'pically
wages are
a function of expected ability.
be positively related to his
other workers subject to the
the learning model,
contracts.
economic rationales behind) both models
distinguish between these
and
The
own
output but
same common shock.
RPE provides
a
way
to filler
common
qualitative predictions of (an'3" underlying
are similar
enough
that
we make no
attempt to
complementary theories of RPE.
Disadvantages of Rewarding Workers Based on Relative Performance
The
benefits of
RPE
be reduced by filtering out some or
relative
that
— the
risk
exposure of risk-averse workers can
common
risk.
This rationale for basing rewards on
are straightforward
all
of the
performance rather than on absolute performance accords with the general rule-of-thumb
workers should not be rewarded or punished for "things beyond their control." Before
embracing the theory completely, however, we must also consider the costs of compensation
systems based on relative performance.
An
obvious cost of
subject to the
is
same common shock.
whether or not a firm uses
in
RPE
RPE at all
the cost of observing the average output of the workers
If the cost
of measuring y
is
a fixed cost, C,
it
affects
but does not affect the irriportance of relative performance
compensation contracts that use RPE: firms
vrill
use relative performance
if
C is less
than the
increased value of the best contract wj(yi,y) relative to the best contract wi(yi). Aheraatively, the
size of the reference
additional workers
group
may be
a choice variable for
from the pool of workers subject
to the
incremental gain from improved contracts (associated with
the cost of measuring an addinonal worker's outpuL
employers
—employers
common
more
will
sample
shock as long as the
precise estimates of 6) exceeds
RobertGibbons andKevinJ. MuwiiY
There
yi-y, instead
of on absolute performance,
worker can lake actions
ways
RPE
another set of costs associated with
is
that
may
much more imponant
be
measurement costs described above. Paying workers based on
the simple
four
August, 1989
7
in
ifiat affect
y,-,
than
relative performance,
distorts the worker's incentives
whenever
the
the average output ofilie reference group, y. TTiere are at least
which a worker's action can have
tliis
kind of effect: sabotage
(in
which
the
worker
takes unproductive actions to diminish the measured performance of another worker in the
reference group); collusion (in which workers in the
same reference group collude and
shirk);
choice of reference group (in which a worker chooses which of several groups of other workers
to
which
worker
to
be compared); and production externalities
affect the outputs
(in
which the productive actions of one
of other workers).
— such
fellow employee—
Sabotage ranges from relatively subtle forms
from, or spreading false gossip about, a
as concealing relevant information
to outright theft or destruction of co-
worker output. Sabotage can be effective even
if
it
diminishes one's
own
output provided that
it
has a larger negative effect on the average ourput of other workers. Lazear (1987) argues that the
sabotage-related problems of
organizations.
According
RPE may
to Lazear,
reduces uncooperative behavior that
is
be
a driving force in the
pay equality observed
in
many
reducing pay differences based on relative performance
detrimental to the firm. Withholding valuable conmients
on a colleague's paper or deleting data from a colleague's hard disk are examples of potential
sabotage
in
academia.
Rewards based on
and supply
less effort than
and Dye 1984).
relative
performance also generate incentives for workers
would be supplied
in the
absence of collusion (see Mookherjee 1984
In academia, for example, colleagues could agree not to
weekends, or during summers. The general conditions for canel
(Stigler 1964) apply to a
potential colluders
is
workplace
cartel as well
—
collusion
small, for example, and collusion
observe the output and compensation of co-workers.
workplace
is
more
likely to
be stable than collusion
to collude
It
in the
is
is
work
stability in the
more
more
likely
seems plausible
on
product market
when
effective
at night,
the
number of
when workers can
that collusion in the
product market will be, both because
RobertGibdons akdKevinJ, MUKPiTY
workplace collusion
workers
who
is
not an antitrust violation and because
are close at
hand than
to
August, 1989
'
8
it
may be
easier to monitor co-
monitor the costs and outputs of geographically dispersed
rival firms.
Relative perfomiance evaluation also distorts incentives
determination of which of several possible reference groups
Workers
is
when agents can
used
influence the
in the evaluation process.
are often involved in interviewing, selecting, and recruiting potential colleagues.
Workers paid on the basis of relative performance
clearly have an incentive not to hire rookies of
For example, Carmichael
superior ability, and will instead desire incompetent co-workers.
(1988) argued that non-tenured faculty have incentives to recruit less-able colleagues. Tlie same
rationale also applies to job search
Dye
firms with inferior co-workers.
relative
— RPE generates incentives
for
workers
to self-select into
(1988) argued that paying top executives based on industry-
performance yields incentives to locate
in industries
with inept
so that the manager
rivals,
appears to be both talented and hard-working relative to other managers in the (carefully selected)
reference group.
Similarly,
declining industry, and
if
may
RPE may
adversely affect an executive's decision to leave a
cause a mediocre performer to leave a rapidly grov.'ing industry even
staying in the industry maximizes the value of the organization.
Finally,
complementary
RPE
when
will be less desirable
the effort levels of individual workers are
factors in the firm's production process, as in
Workers paid on the
basis of relative performance will not
that uniformly increase the productivity
of
all
team production, for example.
have incentives
to
develop innovations
other workers in the reference group.
Optimal
contracts in the presence of such production externalities typically involve rewarding individuals
for group output, 3wj(yi,y)/9y
relative to the
The
effective in
RPE
0, instead
of rewarding individual performance measured
performance of others, 3wi(yi,y)/3y <
costs and benefits of
some
subject to the
and
>
RPE
suggest that filtering out
situations than in others.
same common shock
will therefore
be
are
0.
common
risk will
be more
Incentives to sabotage or collude with other agents
more important when
less effective for small
the
number of co-workers
is
small,
groups than for large groups, ceteris paribus.
RobertGibbons andKevl-;
Also,
RPE
J.
more
will be
effective
group, and will be less effective
team production), or when
August. 1989
9
MLrRPiri"
when
when
the
worker
there are externalities in the production process (such as
output of co-workers
i)ie
unable to influence the choice of reference
is
is
expensive to measure.
Rewarding Executives on Relative Performance
The
relationship between the shareholders of a publicly
corporation's top
manager
diffuse shareholders,
owned
corporation and the
an archetypal example of a principal-agent relationship. Widely
is
assumed
to
be well diversified and therefore risk-neutral, want the manager
to lake actions that increase shareholder wealth.
takes actions that increase his expected utility.
The
imperfectly monitorable manager, however,
Managerial contracts tying pay and dismissal
decisions to changes in shareholder wealth help align the interests of shareholders and the
manager by giving
the
manager
Although the change
output or firm performance,
it
incentives to take actions that increase shareholder wealth.
in
is
shareholder wealth
makes decisions
changes
clearly
how hard an
many
factors
by managers.
(Managerial effort
in this
executive works but rather to whether or not he takes actions or
that increase shareholder wealth.)
in shareholder
an appropriate measure of managerial
an imperfect measure of managerial effort, since
affect the value of a firm besides the actions taken
context refers not to
is
Executives facing contracts based only on
wealth are subject to vagaries of the stock and product markets that are
beyond management
control.
optimal managerial contracts
may
The theory of relative performance
evaluation suggests that
base compensation and dismissal decisions on performance
relative to other firms in the industry or
market
in order to filter
common
risk
from the
compensation of risk-averse managers.
Top-executive compensation and dismissal policy seems a likely place to observe RPE.
First,
it
seems plausible
that industry or maiicet
shocks (such as the October 1987 stock-market
crash) are important and that filtering out these shocks can benefit risk-averse executives.
Second, although there are of course important problems related to defining an industry,
Robert Gibbons AND Kevin
J.
Murphy
August. 1989
10
measuring the stock-market performance of other firms
accomplished on
interaction
v,'ith
a daily basis at trivial cost.
CEOs
production externalities
reference group,
may be
choosing the industry
CEOs
there
in
collusion
defined industry
CEOs
easily
tend to have limited
sabotage, collusive shirking, and
important, however (to the extent that the
which
his firm competes): under
RPE,
a
CEO
plays a key role in
manager who
have the correct incentives
is
among
to leave the industry.
be production externalities across firms: for example,
more
is
probably small. Another source of distorted incentives, the choice of a
in a declining industry will not
may
Third, because
in rival firms, the potential for
is
a
in
RPE makes
ihe best
Finally,
oligopolistic
difficult to sustain.
Another way
to filter risks associated with industry
and market shocks
is to
pay (or
dismiss) executives in a given firm on the basis of iheir performance measured relative to the
f>erformance of other executives in the
that
same firm. Lazear and Rosen
top-management promotions are a form of RPE
measured output
the costs of
is
promoted
to
RPE may be large.
CEO.
It is
in
(1981), for example, argue
which the vice president with the highest
precisely this type of situation, however, in which
First, there are typically
only a small number of vice presidents
with a non-trivial chance of winning the promotion tournament, and therefore the opponunities
for sabotaging or colluding with rival vice presidents
seems
large. Also,
CEOs
t>'pically
play an
important role in selecting the team of vice presidents as well as their ultimate successor (Vancil
1987), so a
CEO paid
subordinates
presidents.
may
on the basis of his performance measured relative
believe he can hold his position longer
Finally, because firm output (in this case,
if
shareholder's objective of maximizing team production.
performance of
he chooses incompetent vice
changes
presumably jointly produced by the top management team,
to the
in shareholder wealth) is
RPE may
interfere with the
1
Robert Gibbons and Kevin
J.
Murpiiy
August. 19S9
1
Rclati>e Performance and
We
following
lest the
all
importance of
RPE
CEO
Compensation
executive compensation and dismissal decisions by
in
chief executive officers listed in the Executive Compensation Surveys published
Forbes from 1974
corporations, or a total of 10.400
dismissal rather than on his
surs'eys include 2,214 executives serving in
These
1986.
to
CEO-years of data.
total incentives
threat of dismissal but also by
changes
in
1,295
We focus on the CEO's compensation and
(which are created not only by compensation and
in the
value of his outstanding stock options and of his
inside stockholdings) because changes in the value of previous holdings of stock and stock
options are determined entirely by firm performance, independent of relative performance. Tlius,
performance
if relative
policy.
on
is
important,
We examine RPE in
relative
importance must arise
in the
measured using the population of
coverage, and historical research
digit
SIC
rate of
1 1
weighted portfolio of
files.
Stock returns for
("the market") and
fums
in the
Industry return
is
all
is
same
is
full
the
one-, two-, three-, and four-
defined as the continuously accrued
Compustat firms (excluding
the continuously accrued return
Compustat firms (excluding the sample
New York
firm's shareholders,
Several potential reference groups arc analyzed:
industry; market return
all
by the
Reference-group stock-market performance
reuim realized on a value- weighted portfolio of
same SIC
logarithm of shareholder wealth,
,000 firms listed on the Compustat expanded industrial,
industries as the sample firm.
firm) in the
in the
rate of return received
including price appreciation and dividends.
Compustat population
compensation and dismissal
next section.
measure firm performance as the change
which equals the continuously accrued
entire
in
executive compensation contracts in this section; dismissals based
performance are examined
We
its
the
sample
on the value-
firm).
Stock Exchange (N^'SE) firms arc calculated using
fiscal-
year stock-market data available from the Center for Research in Security Prices (CRSP). Fiscal-
year industry classifications for N^'SE and American Stock Exchange firms are also obtained
RodertGibbons ANuKE\aN
J.
MuRTHY
12
August. 1989
'
from CRSP. Compustat industry classifications and stock data are used for firms included
Compustat
files
but not on
CRSP
in
After matching these performance variables with the
files.^
Forbes compensation sample and eliminating observations with missing
data, our final
sample
contains 9,425 observations and includes 1,668 executives from 1,049 corporations. All rates of
return and
monetary variables are adjusted for inOation (using the consumer price index for the
month of tlie
closing
predicts that compensation will depend on absolute or relative
Agency theory
performance but gives
relation
1986-constanl dollars.
fiscal year) to represent
little
between pay and
guidance as
firm, industry,
to the
form of the
Rewarding
CEOs
related to the industry
salary
1
and market
return,
CEO pay revisions
and bonus
RPE
0.
should be negatively
Y<0.
to the rates of return realized
Column
significantly related to firm
(I)
shows
performance
that
as
by shareholders
CEO
salaries
measured by the
1.6% for each 10% return on the
that
we
common
CEO
in the firm, two-digit
SIC
and bonuses are positively and
rate of return
return coefficient of .1562 (i=19.5) implies, for example, that
Note
P>
reports estimated coefficients from regressions relating the growth rate of
industry, and market.
The
^^Indusiry and/or Market Return).
for taking actions that increase shareholder wealth implies that
implies that, holding the firm's rate of return constant,
Table
We estimate the following
and market performance:
AIn(CEO Pay) = a + P(Sharcholdcr Rctum)+
(5)
contract.
CEO
on
common
stock.
pay changes by about
stock.^
interpret our estimate as having identified a
compensation formula, rather
than a causal relation from the compensation formula to observed performance.
The
latter inter-.
Compustat industry designations arc appropriate only for the current or final year (generally 1987), whereas
data allow the fuTTi's industry to change from year to year. Compustat stock remms arc based on calendaryear and not fiscal-year data, but when we replicated our analysis (including the construction of industry returns)
for the subset of firms with December fiscal closings, our results were qualitatively unchanged.
CRSP
The estimated return coefficient in column (1) is highly significant, but its economic magnitude is small.
The sample medians for CEO pay and firm size (measured by market value) are S465,000 and S81 2,000 ,0(X),
TTius, a 10% return in a median-size firm is $81,200,000 and corresponds to a pay raise for the
median-pay CEO of S7,300, or less than nine cents for each SI, 000 of increased shareholder wealth. See Jensen
and Murphy (1990) for funher discussion of the magnitude of the empirical relation between pay and performance.
respectively.
Robert Gibbons and Kevin
J.
Murphy
AUGUST, 1989
13
TABLE
1
Coefficients of Ordinar)' Least Squares Regressions of
Aln(CEO
(l-statistics in
Salary
Common
Industry, and Market Rates of Return on
* Bonus) on Firm,
Stock
parentheses)
Independent
Intercept
Firm's Rate of Return
on
Common
2-Digit
Aln(CEO
Dcpcndi:nl Variable:
Variable
Slock
(2)
(3)
(4)
.046
.055
.068
.068
.1562
.1781
.1805
.1824
(19.5)
(19.4)
(21.0)
(19.9)
—
-.0750
—
SIC Industry
(-0.6)
—
—
Market Value-Weighted
Common Slock
r2
.0467
Sample size is 7,755 for all
reponcd in Forbes on 1,668
would be
.0496
ob.'icrvalions.
CEOs
incorrect, because
it
The sample
example) will induce more
adjust for this effect on the
observed. Subsequent
effort
in
and
industry trends
is,
although
rate is
it
is
shareholder return
is
in years
receives an increase of only
when
7.6%
profits, the stock price will
column
(2)
shareholder return
in years
when
is
1
negative
is
for
column
(2)
coefficients in
The same
20%
of Table
make some adjustment
receives an average pay increase of
zero in both the firm and the industry.
pay increase of 9.1%
true that a higher piece rate
in place.
when determining executive compensation. The
CEO
.0538
our independent variable, the firm's rate of return, are thus
suggesting that boards of directors
suggest, for example, that a
.0537
announced, well before future profits are
coefficient for industry rate of return in
statistically significant,
(-5.8)
construcicd from longitudinal data
and so improve expected future
independent of the compensation formula
The estimated
-.1407
(-7.6)
ignores the role of the capital market in bidding up the
day the piece
movements
is
-.1490
serving in 1,049 firms from 1974-1986.
stock price to reflect a firm's expected profit. That
(for
-.0112
-
(^.9)
Return on
pretation
+ Bonus)
(1)
Valuc-Weighicd Return
Nolc:
Salar)'
5.5%
in years
CEO receives an
and industry return
all firms in the industry realize
is
when
average
zero, but
20%
rates of
Robert Gibbons
ANT)
Kemn J.
return. This result
consistent with the hypothesis that uncertainties shared with other firms are
is
partially filtered out of the
Column
August, 1989
14
MuRTiiY
(3) of
CEO's compensation
Table
1
contract.^
CEO
reports coefficients from regressions of
shareholder return and the rate of return in the m:u"ket.
The negative and
pay growth on
significant market-
execuuve
return coefficient suggests that market risks are also partially filtered out of
compensation. The explanatory power of column (3) exceeds
CEO compensation
contracts are
The market-return
CEOs
coefficient
is
in
column
performance. The industry-return coefficient
(4) includes
to industry
industry.
Table
1
is
more
that
both industryjind market
is
no longer
common
significant after controlling for market-
is still
negative.
uncertainties will be filtered out of executive
of course silent on whether industry or market uncertainties are the most
is
common
important source of
I>erformance
risk.
The
results in
column
(4) of
likely to be evaluated relative to aggregate
movements, where the industry
We explore
is
defined as
all
Table
other firms in the
the robustness of this conclusion by re-estimating
Columns
from these regressions
(1), (2),
and
(3)
1
suggest that
CEO
market movements than relative
same two-digit SIC
columns
for three alternative industry definitions: one-digit, three-digit,
industries. Results
suggesting that
performance and not only on absolute
wide stock-price movements, although the point estimate
compensation, but
(2),
negative and statistically significant, indicating that
are rewarded on the basis of market-relative
The theory of RPE suggests
of column
clo.sely linked to market-relative returns than to two-digit
The regression
industry-relative returns.
returns.
more
that
(2)
and
(4) of
and four-digit SIC
are presented in Table 2.
of Table 2 show that
CEO
pay growth
is
positively related to
shareholder return and negatively related to the value-weighted returns for other firms in the
same
CEO pay is
more
one-digit, three-digit,
and four-digit SIC industry. These results suggest
that
closely linked to broad industry definitions than to narrow industry definitions.
In fact,
interpreting the market return as our broadest industry definition, the regressions in Tables
In keeping with the discussion in footnote 6, the
1
and
economic magnitude of the industry adjustment implied by
median-paid CEO in a median-size firm receives ten cents for each 51,000
of increased shareholder wealth when his firm's rcuim Ls 20% and the industry reium is zero, but receives only six
cents per SI, 000 when both his firm and the industrj' realize returns of 20%.
column
(2) is not large.
In particular, a
Robert Gibdons and Ke\t<
J.
Murphy
August. 1989
15
TABLE 2
Coefficients of Ordinar)' Least Squares Regressions of Aln(CEO Salary + Bonus) on
Shareholder Return, Industr)- Return, and Market Return for Thrc^ Alternative Industry' EHifinitions
(i-siatistics in
parentheses)
Rcgrcssiion Coefficients
Independent
Dependent Variable
Variable
Intercept
is
Aln(CEO
Salary + Bonus)
(1)
(2)
(3)
(4)
(5)
(6)
.064
.051
.050
.069
.068
.068
Firm's Rate
.1850
.1722
.1637
.1855
.1810
.1771
of Return
(20.6)
(18.7)
(18.4)
(20.7)
(19.5)
(19.5)
-.1266
—
—
-.0554
-^'
—
-.0018
—
1-Digit
SIClndusuy
Rate of Return
3-Digit
(-7.1)
(-2.0)
—
SlCIndusuy
—
—
-.0452
Rate of Return
(-0.1)
(-3.5)
—
—
4-Digit SIC Industry
Rate of Return
—
-.0303
—
-.0019
(-2.8)
Value- Weighted Market
—
(-0.2)
—
—
-.1027
-.1462
-.1444
(-3.4)
(-6.6)
(-6.9)
Rate of Return
Sample Size
r2
Note: TTic sample
is
7.757
7,687
7.429
7.757
7,687
7.429
.0528
.0483
.0479
.0542
.0536
.0539
constructed from longitudinal data reported in Forbes on 1.668
CEOs
scrying in 1.CM9 rirms from
1974-1986.
2 suggest that the magnitude and significance of the industry coefficient, and the explanatory
power of the
The
regression, decline monotonically as the industry
becomes more narrowly
defined.'
regressions in columns (4), (5), and (6) of Table 2 include market return as an
additional explanatory variable and replicate the regression in
alternative industr>' definitions.
The market-return
column
coefficient
is
(4)
of Table
1
for the three
negative and significant in
all
three regressions, indicating that market risks arc pardally filtered out of executive compensation
after controlling for industry returns.
growth
is
Holding market returns constant, however,
negatively and marginally significantly related to one-digit industry returns, but
significantly related to three-digit or four-digit industry returns.
that
CEO
CEO performance is more likely
relative to industry
movements. An
to
be evaluated relative
These
is
pay
not
results support the finding
to aggregate
market movements than
alternative interpretation of these results,
however,
is
that
RobertGibbons andKevin J. MuRPitY
August, 1989
16
firms are rarely engaged in a single industry, in which case our inability to delect an industry
movements could simply
effect after controlling for market
definitions based on a single
SIC code
reflect the fact that industry
for each firm are inappropriate for purposes of relative
performance evaluation. Other alternative interpretations based on a connection between the
performance of the stock market and conditions
in the labor
CEOs may
market for
also be
possible.
The RPE hypothesis
movements
portfolio.
predicts that
the higher the correlation
Using
CRSP
CEO
pay
between the
monthly stock returns for
will be
more strongly
firm's return
NYSE
and the return on- the market
firms,
we compute
the correlation
We
re-estimate the
between firm and market return over each preceding two-year period.
regression in column (3) of Table
1
(CEO pay changes on
related to market
finn and market return) for the
subsamples of observations above and below the median value of
this correlation statistic
The
find coefficients on the market return of -.1775 and -.0916, respectively.
between the coefficients
support for the
RPE
is
significant (t=-1.8).
difference
interpret this result as strong additional
hypothesis.
The regressions
in
Table
1
partially adjusted for industry- and
consistent with the
We
and
RPE
and Table 2 indicate that
CEO
salaries
market-wide stock price movements.
and bonuses are
This evidence
is
hypothesis.^ These results are based only on salaries and bonuses,
The evidence also is consistent with a subtly diffcrcni hypothesis: CEOs are rewarded for shareholder return
excess of the market return because this is ihc true objective of the shareholders. This laiier hypothesis is
incorrect, because it confuses the shareholders' ex ante and ex post objectives. To understand the distinction
between the shareholders' ex ante and ex post objectives, consider the following two-siage model of invesmient and
production. During ihc investment stage of the model there arc two kinds of agents, entrepreneurs and investors.
Entrepreneurs have ideas but no capital; investors have capital but no ideas. If an entrepreneur receives capital
from (at least) one investor, then the entrepreneur becomes a CEO, the investor becomes a shareholder, and the
production stage of the model begins. The production stage is identical to the model in which yj = ai + E; + 9,
°
in
where
is,
yj is
now
the
change
in
shareholder wealth (gross of
completely firm-specific) noise, and 6
is
wage payments)
at the
a market-wide shock. (Note that this
action to affect either the variance of the firm's returns or
its
i* firm,
e^ is
idiosyncratic (that
model docs not allow
the
CEO's
covariance with the market.)
According to the capital-asset pricing model, at the investment stage investors care about the covariancc
between the firm's return and the market (among other things). But at the production stage, shareholders at the i'^
firm face precisely the conu^cting problem described above (in the section entitled "Rewarding Executives on
Relative Performance"): they have invested their capital and now want to make the most of it. Thus, in this
model, the covariance between the firm's return and the market has ex ante relevance (because it helps determine
which entrepreneurs will receive capital) but not ex po5/ relevance.
RobertGibbons ANTsKEsaNJ-MuRPitY
however, whereas
CEOs
August. 1989
17
in fact receive
compensation in many additional forms
— including stock
options, profit-sharing arrangements, stock grants, savings plans, long-term performance plans,
and other fringe benefits. These other forms of compensation
and market movements, and
not only the cash
it
is
therefore useful to analyze the
also be adjusted for industr)'
CEO's
total
compensation and
component of compensation.
The annual Forbes surveys include
Unfortunately, although
bonus.
may
tlie
data on total compensation as well as on salary and
salary and
bonus data are easily verifiable and
definitions have remained constant throughout the
components of compensation
relate to the largest
sample period, the Forbes ^{a for other
are neither easily verifiable nor reliable.
non-cash component of compensation
The most
— stock options.
severe problems
Forbes surveys do
not include any stock-option data prior to 1978, and after 1978 the surveys report gains
exercising stock options but
their
do not repon stock options granted during the
year.
from
The gains from
exercising options are typically associated with options granted five to ten years previously, and
therefore using these data imposes serious timing problems.
Columns
(1)
and
(2) of
Table 3 report regressions based on the Forbes
total-
compensation data, excluding both stock-option grants and the gains from exercising stock
options.
The
definition of total compensation varies
somewhat from year
includes salary, bonus, value of restricted stock, savings and
The
one-digit industry coefficient in
CEO's
total
column
(1) is
The
plans, and other benefits.
negative and significant, indicating that the
compensation (exclusive of stock options)
absolute firm performance.
thrift
to year, but typically
is
based on industry-relative as well as
regression in colunin (2) suggests that market returns are a
more
important determinant of total compensation than arc industry returns, although the industry
coefficient is negative
and
statistically significant.
are qualitatively similar to the results in
earlier conclusions are not
columns
The
(1)
results in
and
columns
(4) of
Table
(1)
and
(2)
of Table 3
2, suggesting that
changed by analyzing broader measures of compensation.
our
Robert Gibdons and Kivin
J.
August, 1989
18
Murj'iiy
TABLE
Coefficients of Ordinal")'
Lca.<;i
3
Squares Regressions of Aln(Salary + Bonus) and AIn(Tota] Pay) on Firm and
Murphy (1985) Sample of Manufacturing Finns
Industry Return for Forbes Sample and for
(t-statistics in parcntlieses)
Independent
Regression Coefficients
Variable
1,049-Firm Forbes Sample
73-Firrn
Dependent Variable:
Aln [Total Pay]
Intercept
Firm's Rate of Return
on
Common
1-Digil
Slock
SIC Industry
Value-Weighted Return
Market Value- Weighted
Return on Common Slock
Aln[Salary
(1)
(2)
(3)
(4)
(5)
(6)
.074
.082
.024
.025
.031
.025
.2CM8
.2055
.1441
.1455
(17.9)
(18.0)
(7.1)
(7.2)
(5.1)
(4.9)
-.1859
-.0827
-.1533
-.1089
-.3282
-.5670
(-8.2)
(-2.4)
(4.7)
(-1.2)
(-5.2)
(-3.3)
—
-.1487
—
-.0522
—
.2807
r2
.1918
(1.5)
7,757
7.757
997
997
997
997
.0397
.0415
.0495
.0498
.0328
.0351
Note: Forbes sample consirucicd from longitudinal data on 1,668
Forbes
.t99i-'
(-0.5)
(-3.8)
Sample Size
-f-
Muqihy Sample
Dependent Variable:
Bonus]
AIn[Total Pay]
CEOs
serving in 1,{M9 firms from 1974-1986.
TTic
definition of total compensation typically includes salary, bonus, value of restricted stock, savings and
thrift plans, and other benefits but does r\ol include the value of stock options granted or the gains from
exercising stock options. The Murphy (1985) data arc based on 173 CEOs serving in 73 large manufacturing
firms from 1966 to 1983. Murphy's definition of total compensation includes salary, bonus, value of restricted
stock, savings and thrift plans, oilier benefits, and the estimated value of slock options granted.
The
total-pay variable used in
columns
(1)
and
(2)
excludes stock options, but a better
measure would include the value of stock options granted during the year. Although these data
Forbes sample, they are available
are not available for the
manufacturing firms
first
for a longitudinal
sample of 73
analyzed by Murphy (1985). This sample, constructed from proxy
statements, includes data on salaries, bonuses, deferred compensation, stock options, and fringe
benefits,
and covers 173
sum of
salaries,
CEOs
firom
1966 through 1983. Total compensation
is
defined as the
bonuses, fringe benefits, stock options, the face value of deferred
compensation, and restricted stock awarded during the year (valued
at the
year-end stock price).
Stock options are valued on the date of grant using a Black-Scholes (1973) valuation formula that
allows for continuously paid dividends as described in
Murphy
(1985).
RoBERTGronoNSAKD Kevin J. Murphy
Columns
(3)
19
through (6)
in
August. 1989
Table 3 report coefficients from relative-performance
regressions based on the 73-firm sample of manufacturing firms.
earlier results,
we
cash compensation
of Table
.01%
Table 3 shows
in
that the
growth
rate of
Neither the industry- nor market-return variables are
individually statistically significant in the regression in
(4)
check on our
positively related to firm performance and negatively related to
is
industry performance, ceteris paribus.
significant at the
a robustness
re-estimated the one-digit industry salary-and-bonus regressions reported in
Table 2 for the manufacturing finn sample. Column (3)
CEO
As
These
level.
2, indicating that
our
column
(4),
although they dJt jointly
results are qualitatively similar to those in
earlier results are not specific to the particular
columns
(1)
and
sample analyzed
there.
The
column
industry-return coefficient in
indicating that the
CEO's
total
Table 3
(5) of
is
negative and significant,
compensation (including stock option grants)
relative as well as absolute firm performance.
additional explanatory variable, and
shows
that
Column
CEO pay
(6) includes
growth
is
based on industr)'-
is
market returns as an
negatively and significantly
related to one-digit industry renjms, but positively (and insignificantly) related to
Thus, our earlier finding that
CEO
performance
is
more
likely to be evaluated relative to
aggregate market movements than relative to industry movements
for broader definitions of
finding
may
simply
is
compensation and industry performance.
refiect a
market returns.
not robust across samples
On
the other hand, this
high correlation between market performance and one-digit industry
performance for manufacturing firms. Indeed, when we re-estimaie the regressions
(3)
through (6) of Table 3 using two-digit industry definitions,
are a
more important determinant of CEO pay changes than
As an
additional robustness check,
we
use the Forbes sample of 1,668
In(CEO
where
(5).
Pay),i
=
^^
are
reponed
in
in
Table
4.
i'^
CEO, and
the
movements.
CEOs
Table
+ a(Age),j + P X(Sharcholder Rcium),^+ 7X(1-Digil
Xi is a separate intercept or "fixed-effect" for the
Our results
reponed
columns
movements
find that market
are two-digit industry
the levels equation analogous to the first-difference specification
(6)
we
in
2,
to estimate
namely:
Industry Retum)^,
first
difference of (6)
is
Robert Gibbons \sd Kevin
J.
20
MuRPinr
August. 1989
TABLE 4
CocfTicicnts of Ordinary Least Squares Regressions of
One-Digit
ln(CEO Salary + Bonus) on Firm and
Shareholder Return for Level and First-Difference Spccificaiions
Industry'
(l-suiiistics in
parentheses)
Dcixindcnt Variable
Ln
Change in
ln(CEO Pay)
Independent
Variable
Intercept
Current Slock Return (Rji)
in
(CEO
Pay)
1000s of 1986-Dollars
Change in
ln(CEO Pay)
Ln
(CEO Pay)
(1)
(2)
(3)
(4)
.064
(CEO- specific)
.005
(CEO-specific)
.1850
—
.1801
—
(20.6)
Current
1
-Digit Industry Return
(20.1)
—
-.1266
(Iji)
Time Trend
(-2.7)
—
(t)
—
-.0740
(-7.1)
—
—
.0343
(15.5)
X
Cumulative Stock Return
—
^ii
1^=1
—
Sii.
^T = l
—
-.0011
-.0449
(-0.1)
(-2.2)
J
Year Dummies
Sample Size
r2
Note: The sample
.1762
(24.3)
J
Cumulative 1-Digil
Industry Return
—
.1829
(25.0)
(no)
(no)
(yes)
G'cs)
7.757
9,361
7.757
9.361
.0528
.8926
.0758
.8964
consuiiclcd from longitudinal data reported in Forbes on 1,668 CEOs serving in
Columns (2) and (4) include 1,668 individual inlcrccpts.
is
1,049 firms from 1974-1986.
t
t
Regression Specification:
Level:
lr»(Wit)
=
Xj
-f-
at
-t-
P ]^ Rjx
T=l
Aln(wi[) =
First Difference:
To ease comparison, column
of Table
2.
Column
(2) of
(1)
should equal the intercept
industry stock returns in
industry return in
column
(2) of
Table 4
in
column
column
is
pRjt
T
S 'il + ^it
T=l
+ -fiii +
v;[
of Table 4 repeats the regression reported
in
column
(1)
Table 4 reports estimated coefficients from the fixed-effects levels
equation, (6). If the levels regression
(2)
a+
-^
is
correctly specified, the time-trend coefficient in
column
(2)
(1),
column
and the coefficients for the cumulative firm and
should equal the coefficients for shareholder return and
(1), respectively.
In fact, the cumulative-stock-retum coefficient in
very close to the stock-return coefficient
in
column
(1),
but the
RobertGibbons AND Renin J. Murphy
August, 1989
21
cumulative-industr}'-retum coefficient in column (2)
coefficient in
coefficient in
The
column
columns
is
(1) is significantly negative.
(1)
and
whereas the corresponding
insignificant
This difference
(2) suggests the possibihty that the
model
in the industry-return
is
columns
first-difference and fixed-effects specifications in
misspecified.
and
(1)
(2) of
control for observable and unobservable individual characteristics that are fixed for a
time, but
do not allow for economy- wide shocks
time-effects in
columns
(3)
and
(4)
by including year
columns
fiirst-difference specifications in
that vary
(1)
and
(3)
from year
dummy
shows
again similar, and the industrj'-reium coefficient in column
the coefficient reported in
effects specifications in
column
columns
(1),
(2)
column
(4), in contrast to its
A
comparison of the
that the rate-of-retum coefficients are
though smaller
(3),
magnitude than
in
remains negative and significant. Comparing the fixed-
and
(4), the
cumulative-stock-return coefficient
essentially unchanged, but the cumulative-industry-retum coefficient
in
CEO over
We allow for these
to year.
variables.
Table 4
insignificance in
column
(2).
is
is
again
negative and significant
Thus, including year dummies
appears to resolve the discrepancy between the first-difference and fixed-effects specifications.
Comparison with Previous Research
Antle and Smith (19S6) conducted a careful examination of the relation between
salary and
bonus and industry
filtered out
rates of return but
found no evidence
that two-digit industry risk is
of executive cash compensation contracts.^ In contrast,
we
partially filtered out of contracts (although the effect diminishes
movements
as well).
find that industry risk
when we
Ande-Smith methodology and our own.
is
include market
we
investigated
First,
whereas our
In an attempt to reconcile these conflicting findings,
several differences between the
CEO
regression specification relates changes in compensation to absolute and industry stock returns.
Anile and Smith (1986), Tabic 6B. Although Antic and Smith find no relation between salary and bonus and
industry returns, they do document a weak relation between industry returns and a broader definiiion of
compensation that includes the net returns on inside slockJioldings. They measure these net returns on inside
stockholdings as the change in the value of stockholdings minus the change
portfolio. Thus, their results
may be
in the
value of the S&.P 500
driven by correlations between industry returns and the S<&:P portfolio.
Robert Gibbons and Kevin J. Murphy
22
August, 1989
TABLE
5
Aln(CEO Salary + Bonus) on A(Sharcholdcr
A(Retum on Assets), and A(2-Digil Industry Return on Assets) for
1,668-CEO Sample, CEOs in Three Indusunes, and CEOs in 26 Firms
CocfficicnLs of Ordinary Least Squares Regressions of
Return),
A(Industry Return),
Full
(t-siatjstics in parcntJiescs)
Independent
E>epcndent Variable
Variable
Forbes
Sample
Change
in Firm's
Shareholder Return
Change
in
2-Digit
Industry Return
Change
in
Indusirj'
(3)
(4)
(5)
(6)
.065
.066
.061
.060
.051
.048
-.0431
.0583
.0646
.0436
(8.9)
(3.6)
(2.5)
(1.2)
(-0.3)
-.1071
-.0828
-.1494
-.0738
-.1370
-.0026
(-11.0)
(-8.4)
(-5.5)
(-2.6)
(-2.5)
(-0.0)
1.220
—
1.590
—
3.426
—
r2
Note:
(8.0)
—
.3730
Return on Assets
Sample Size
(7.1)
—
1.804
(3.3)
(2.0)
1.682
(2.2)
7,693
7,468
1,079
1,059
247
247
.0196
.0539
.0273
.1036
.0264
.2162
sample and subsamplcs consu^clcd from longitudinal dala rcporltd in Forbes on 1,668 CEOs serving in
1,049 firms from 1974-1986. The thrce-indusU'y subsample includes 218 CEOs serving in 135 chemical
(S1C=28), clccu-onics (S1C=36) and transporiaiion (SIC=37) firms. TTic 26-firm subsample includes 44 CEOs.
Full
the Antle-Smith specification relates
pay changes
to
changes
in absolute
and industry stock and
accounting returns. Also, whereas our sample pools data from over l.CXX) firms
SIC
-.0114
.0731
(11.2)
(15.2)
in 2-Digil
26 Antle-Smith
Firms
Industries
(2)
Return on Assets
Change
Salary + Bonus)
(1)
—
Firm's
Aln(CEO
Three Antle-Smith
Full
Intercept
is
industries, the
in
58 two-digit
Antle-Smith results are based on separate regressions for 39 firms
in three
industries.
Column
(1) of
Table 5 reports
results
from relative-performance regressions using our
data and the Antle-Smith specification.
AIn(CEO Pay) = a
(7)
The
coefficient
CEOs
on
the
change
in
-i-
pA(Shareholder
Return)-*- YA(2-Digil Industry Return).
shareholder return
is
positive and significant, indicating that
are rewar<led for taking actions that increase the value of their firm, ceteris paribus.
Moreover, the coefficient on the change
in industry return is
negative and significant, indicating
RobertGibbons AND Kevin
that
J.
Murphy
August. 1989
23
when
boards of directors adjust for industry shocks
evidence
is
consistent with the
RPE
revising the
CEO's compensation. This
hypothesis, and indicates that specification differences do
not explain the difference between our resuUs and Antle and Smith's.
Our
specification does,
however, have more explanatory power than the Antle-Smith specification, with R^ = .0496
column
(2) of
Our
Table
1,
compared
results in Tables
1
R2 = .0196
to
in
column
(1)
of Table 5.1°
through 4 are based on pooled cross-sectional time-series data,
whereas Anile and Smith report separate regressions for each firm.
replicate their results,
we
in
estimate the regression
firms in our sample that had six or
more usable
in
column
obscrx'ations.
(1)
In a further attempt to
of Table 5 separajelyfor the 690
The average and median
t-siaiistics
associated with the coefficients on the change in industr>' return in these firm-specific regressions
are -0.39 and -0.37, respectively.
Two-thirds (450) of the 690 estimated coefficients are less
than zero, and 89 (13%) of the associated t-statistics are less than -1.8.
coefficients are based on varying degrees of freedom and
significance of the individual coefficients.
provide fairly strong support for the
It
seems
RPE
likely that the use of
As
a
it
is
The estimated
therefore difficult to test the joint
whole, however, these firm-specific regressions
hypothesis.
RPE
varies across industries, and
particular firms in the three industries analyzed
it
is
possible that the
by Antle and Smith (electronics, aerospace, and
chemical companies) are for some reason characterized by a lack of relative-performance
evaluation.
changes
Column
in firm
(3)
of Table 5 reports coefficients from a regression of pay growth on
and industry
rates of return for
CEOs
in the three industries
and Smith: chemicals and pharmaceuticals (SIC code 28),
electric
examined by Antle
machinery and equipment (SIC
^°
Our specificaiion in (5) restricts ihc ccxtfficicnls on lagged shareholder and industry returns lo be zero,
whereas the AnUc-Smith specificaiion (7) implicitly restricts these cocfficienLS lo be equal in magnitude and
opposite in sign to the analogous coefricicnis on the contemporaneous performance variables. The Appendix
presents results of both Hrst-differcnce and fixed-effects (level) regressions of compensation on jxrfonnance and
lagged performance. The results in columns (1) and (3) suggest that changes in CEO pay arc positively related to
shareholder return and (to a much smaller degree) lagged shareholder return. In addition, the results
in column
show that pay changes arc negatively related to contemporaneous industry return, but are
(3), for example
—
—
The estimated coefficients in the associated pay-levels regression in
suggest that misspeciflcaiion is not the cause of this perplexing result. Column (4) docs, however,
provide an alternative statement of the result: although current and lagged returns are important determinants of
the level of pay, we find suppoa for \hc RPE hjpoihesis with respect to current but not lagged industry raums.
positively related to lagged industry return.
column
(4)
24
Robert Gibbons AKD Kevin J. MuRTiry
ccKde 36),
August. 1989
and transponation (SIC code 37).'' The industry-retum coefficient
in
column
(3) is
negative and significant (t=-5.5), suggesting that the specific industries analyzed by Anile and
Smith do not account for the difference
firms are included in our data, and
in
our
column
results.
(5) of
Twenty-six of the 39 Antle-Smith sample
Table 5 reports
RPE
regressions for this 26-
firm subsample. (Most of the 13 remaining firms were delisted or merged prior to the beginning
of our sample.)
The industry-retum
coefficient
is
again negative and significant; thus, the
difference between our results and Antle and Smith's does not appear to be due to differences
between the two samples of fimis.
In addition to stock
performance, the Anlle-Smiih specification also includes changes
firm and industry return on assets.
Columns
(2), (4),
and
(6) of
in
Table 5 replicate the Antle-
Smilh specification (including return on assets) for the Forbes sample, the three-industry sample,
and 26-firm sample, respectively.'^ After controlling for accounting returns, the change
industry-stock-retum coefficient
the three-industry
(6). '3
sample
in
is
negative and significant for the
column
(4),
but
is
full
sample
insignificant for the 26-firm
in
column
sample
in
in the
(2)
and
column
Therefore, a combination of differences in sample size and composition and in regression
specification
seems
to reconcile
our results with Antle and
Smith's.'''
Although Jensen and Murphy (1990) do not directly
analyze Forbes compensation data and conclude that changes in
related to
changes
in
test the
RPE
hypothesis, they
CEO pay are much more closely
shareholder wealth than to changes in shareholder wealth measured relative
^^
The Anile-Smilh sample is from the aircraft-manufacturing subset of the two-digit transponation industry
(which also includes automotive firms) and the chemical subset of the two-digit chemical and pharmaceuticals
industry, but their industry-relative returns are based on the more broadly defined two-digit industries.
Return on assets is conslnicicd from Compustat data and is defined as the sum of after-tax income (before
extraordinary items), interest, and interest tax shields, divided by total assets. Industry return on assets is defined
as the assct-weighted-averagc return on assets for all other firms in the same two-digit industry.
^^
The return on assets and industry return on assets variables are positive and significant in columns (2), (4),
and
(6) of
Table
5.
RPE hjpothesis with respect to
RPE with respect to accounting returns.
Therefore, although our results support the
objective (shareholder return),
'*
we do
not find evidence for
the principal's
There are other differences between the Antic-Smith methodology and ours that may have some bearing on
Antic and Smith average the after-tax salary and bonus of the three highest-paid
executives, for example, and their data cover an earlier lime period (1946-77). Also, whereas our measure of
industry returns is the value-weighted average return in the industry, Antle and Smith calculate weighted-average
industry returns in which the weight for each firm in the two-digit industry is the correlation coefficient between
the returns of that firm and the sample furo.
the difference in results.
Robert Gibbons AXD Kevin
to the
Murphy
J.
25
market or two-digit industry.
CEO
of the change in
August, 1989
In the spirit of their specification,
pay on the change
we
estimate a regression
shareholder wealth and the change in wealth the
in
shareholders would have received had they invested in an industry portfolio rather than in the
Shareholder-wealth changes are measured as RiV(.j, where R^
sample firm.
return and V(.]
measured
is
the firm value at the end of the previous year.
where V,
as IiV(.j,
portfolio of
all
j
is
as before and
Hie estimated
other firms in the two-digit industry.
shareholder
Industry-wealth changes are
the rate of return
I[ is
is
on
a value-weighted
coefficients
from
this
regression are
A(Salar>' and
(8)
Bonus) =32.1 + .0133CR,V,.,) + .0021
The
coefficient
on the change
specification therefore yields
from column
(2)
of Table
salary and bonus, as
opposed
growth-rate of pay) in Table
differ
from those
value.
two reasons.
RPE
First, the
to the cliange in the
in (8) is ix)sitive
hypothesis.
1,
The regression
dependent variable
although the
to
V,.!, the
Tliis
in (8) differs
in (8) is the
change
logarithm of salary and bonus (that
because the former are multiplied by
and Table
and insignificant.
Second, the independent variables (excluding the intercept)
1.
Both of these differences appear
results in (8)
related
in (5)
indusu^ wealth
no support for the
for
1
in
(1,V,.,).
(1=1.0)
(1=6.3)
is,
in
the
in (8)
beginning-of-period market
have some bearing on the difference between our
latter
seems
work (Gibbons and Murphy 1989) we show
to
be the more important of the two.^^
In
that the rate-of-retum specification of the
indef>endent variables reduces the impact of heterogeneity by firm size in pooled regressions.
'
^
Re-esdmaiing
(8) with the rate
of growth of CEO pay as the dependent variable yields
Aln(Silary and Bonus) = .061 + .00001 69(R,V,.,)
-
and rc-cstimating
(8) with rates of return instead of wealth changes as independent variables yields
A(Salary and Bonus) = 28.4
+ 110.7(RJ
(1=15.0)
is
.0OO0025a,V,.,).
(i=-1.0)
(1=6.4)
-
38.89at).
(l=-3.2)
Both of these regressions >ield negative point estimates for the industry variable, but the industry coefficient
where wealth changes arc replaced by rates of return.
significant only in the second case,
Robert Gibbons and Kevin
J.
Murpiiy
August.
26
Relative Performance and
CEO
directors (an arrangement that
under which the
maybe
CEO
CEO will be encotiraged or forced
The
to leave office.
"new blood"
CEO
succession process
into the helm, replacing tired or
unhealthy CEOs, and providing promotion opportunities for other executives
in the organization.
focus on one important aspect of the succession process that provides direct
incentives for the current
CEO:
the threat of dismissal for poor performance. Implicit contracts
linking dismissal to changes in shareholder wealth generate incentives for
that increase shareholder wealth.
absolute performance subjects
price
and his board of
either exphcit or implicit) is a specification of the conditions
serves a variety of purposes, including bringing
We
movements.
As
CEOs
performance yields incentives
CEOs
analyzed 286
in the
to increase
are rarely openly fired
CEOs
and market stock-
cannot take actions to affect the measured
market, basing dismissals on relative
shareholder wealth while filtering out exogenous risks.
from
their positions;
Weisbach (1988),
CEO changes over ten years and found only nine cases in
CEO
frequently: over half (1,000) of the 1,914
was replaced.
CEOs
in
CEO
our sample for
available left office during the 1974-86 sample period.
CEO dismissal
to take actions
case of compensation, however, basing dismissals on
in the industry or
performance as a reason why the
CEOs
to uncertainties associated with industry
Therefore, as long as
performance of the other firms
for example,
which boards mentioned
turnover, however, occurs
whom
performance data are
We estimate the importance of RPE in
decisions by estimating the effect of firm, industry, and market performance on
the probability that a
CEO will
soon leave
office. In particular,
we
estimate the following logistic
regression:
(9)
989
Turnover
important part of the contractual arrangement between a
An
1
\n(
^'^'^1
p^upl^
) =
a+ P(SharchoIderRcium)+-y<l-Digit Industry and/or Market Return).
Robert GrBBONS AND Kevln J. MuRPifY
27
August, 1989
TABLE 6
Esiimatcd Logistic Models Prcdiciing CEO Turnover Using Shareholder, Industry,
and Market Rates of Return
(asymptotic
parentheses)
l-statistics in
Independent
Coefficient Estimates
Variable
Intercept
(1)
(2)
(3)
(4)
-2.10
-2.19
-2.22
-2.22
-.4152
-.7130
-.7542
-.7341
Slock
(-4.1)
(-5.7)
(-6.2)
(-5.8)
1-Digil IndusU7 Value-Weighted
—
.8330
—
-.2613
Firm's Rale of Return
on
Common
Return on
Common
Slock
Sample
size
is
9,372
serving in his last
—
—
Market Value-Weighicd
Return on Common Stock
Nolc:
for all regressions.
The dcpcndcnl
rales of return are calculalcd for fiscal year.
llic
.01%
Tlic sample
if
the
many
(3.1)
is
equal lo
if
1
the
CEO
is
ollicrwisc (8.372 observations);
constructed from longitudinal data
CEO
is
All regressions are
serving in his last full fiscal year, and zero
predicts that
in the firm, but are less likely to
and market, ceteris paribus.
explicit dismissals are rare,
is
1.350
(4.9)
serving in 1,104 firms from 1974-1986.
The theory of relative performance evaluation
be dismissed following bad years
the industry
CEOs
1.114
level.
The dependent variable equals one
otherwise.^^
voiiablc
year (1.000 obscr^'alions) ajid
full fiscal
reported in Forbes on 1,915
significant at
(-0.6)
(3.9)
We interpret
P <
CEOs
are
more
likely to
leave following bad years in
and y >
as evidence that, although
instances of turnover are in fact dismissals and are based on
relative performance.
Table 6 repons coefficients from
logistic regressions predicting the probability
turnover as a function of shareholder, industry, and market rates of return.
that,
Column
unconditional on industry and market performance, the probability that a
his last full fiscal year is negatively
shareholders during that year.
and significantly related
to the rate
CEO
is
of
(1)
CEO
shows
serving in
of return realized by
Converting the regression coefficients into estimated turnover
The 1988 Forbes survey was examined to identify CEOs whose last fiscal year was 1986. The final CEOyear for firms leaving the Forbes survey is excluded, since we cannot determine whether or not this is the last year
CEO. A total of 582 firms were deleted from the Forbes surveys during the 1974-86 sample period. Of
293 are still 'going concerns" as of 1987, 214 were acquired by or merged with another firm, and 35
liquidated, went bankrupt, or went private. Current stauis data arc unavailable for 40 of the 582 firms.
for that
these,
Robert Gibbons AND Kevin
J.
Murphy
column
probabilities, the regression in
years
when
realizes a
August, 1989
28
(1) implies that a
CEO
has a .109 turnover probability in
the firm earns a zero return, but a .131 dismissal probability in years
-50%
when
the firm
return.'"'
Columns
and (3) of Table 6 show that turnover probabilities are negatively and
(2)
significantly related to shareholder return, but positively and significantly related to returns in the
The
one-digit industry and market, ceteris paribus.
example, that a
CEO
has a .098 turnover probability in years
earn a zero return. The
same
CEO
return while the market return
the firm and market earn a
Column
(4) in
in size.
The
is
-50%
in
(2)
definitions.
and
when both
has a .137 turnover probability
(3) implies, for
the firm and the market
when
the firm earns a
zero, but only a .083 turnover probability in years
-50%
when both
return.
column
(4) is positive
and
industry-return coefficient in
significant after controlling for
columns
column
Table 6 includes both industry and market returns. As
market-retum coefficient
reduced
regression in
statistically significant,
column
(4)
column
although
remains positive but
market-wide stock-price movements.
(4) using the two-digit, three-digit,
in
We
is
it
(3), the
is
much
no longer
also re-estimated
and four-digit alternative industry
Consistent with our earlier results on compensation,
the industry coefficient declines monotonically as the industry
we
find that the magnitude of
becomes more narrowly defined
(although the results for one-digit and two-digit industries are virtually identical). Thus, our
results suggest that
movements than
CEOs
CEO
turnover policy
to control for industry
is
more
likely to control for aggregate
movements.
leave office for a variety of reasons and usually leave their position only after
reaching normal retirement age; a third of the
CEOs
in
and more than half left between the ages of 62 and 66.
by re-estimating the
logit regressions in
our sample
left
when
they were 64 or 65,
We attempt to exclude normal retirements
Table 6 for the subset of CEOs younger than 64.
y
^'
market
Dismissal probabilities are computed as p=
where x = -2.10
- .4
152 (Return).
Our
Robert Gibbons AND Kevin
J.
MuRPinr
August, 1989
29
TABLE 7
Eslimaicd Logistic Models Predicting
1-Digii
SIC
CEO Turnover Using Current and Lagged Shareholder,
Industry',
and Market Rates of Return
(asymptotic t-statistics in parentheses)
Indcpcjidcnt
Coefficient Estimates
Variable
Intercept
Current Slock Return
Lagged Slock Return
Current
1
Industry'
Lagged
-Digit
(3)
(4)
-2.10
-2.21
-2.26
-2.25
-.3499
-.6531
-.6981
-.6747
(-3.4)
(-5.2)
(-5.7)
(-5.3)
-.2660
-.4114
-.4122
(-2.7)
(-3.3)
(-3.4)
—
.8253
—
—
siic
is
scrying in his
9.291 for
last
significant at the
full
regressions.
all
fiscal
The dependent
.01%
although the asymptotic
CEOs
The samp'c
is
CEO
(3.1)
.3835
.1924
(1.7)
(0.4)
is
equal lo
1
if
the
CEO
is
.
constructed from longitudinal data
serving in 1,092 firms from 1974-1986.
All regressions arc
to the results for the full sample,
and significance levels tend to be higher for the subsample.
plausibly require
to leave office.
CEO
1.379
(4.8)
level.
t-statistics
may
1.117
olhcrv,isc (8.299 observations);
factors can affect stock prices in addition to industry
the probability of
more than
and market movements, and
a single bad year before forcing or
Table 7 reports coefficients from
turnover as a function of current
market rates of return. The results
effects of
variable
year (992 obscrvalions) and
younger subsample are qualitatively similar
boards of directors
.2377
(0.6)
—
—
reported in Forbes on 1,896
encouraging a
—
—
—
rales of return arc calculated for fiscal year.
Many
-.2905
(1.6)
Lagged Market Reium
results for the
'"(-3.4)
(-0,7)
.3263
Return
Sample
-.4295
-
(3.9)
Current Market Return
Noic:
(2)
Reium
1-Digit
lndustr>'
(1)
in
and lagged
Table 7 are very similar
logit regressions predicting
shareholder, industry, and
to the results in
Table
6,
and the
contemporaneous performance on turnover probability are uniformly greater than are
the effects of lagged performance.
We
also re-estimated the regressions in Table 7 after
incorporating an additional lagged performance variable (namely, performance measured two
years before the CEO's final fiscal year)' and obtained similar results.
These
results are
Robert Gibbons and
Kjenin
J.
Murphy
consistent with the findings of
that decisions to replace
CEOs
August. 1989
30
Wamer, Watts, and Wruck (1988) and Weisbach, who concluded
are based
on relatively recent }>erformance
data.
Comparison with Previous Research
Wamer,
Watts, and
that ihe probability
Wruck
estimated logit regressions for 87
of CEO turnover
is
CEO
changes and found
negatively related to contemporaneous shareholder return
and positively related to contemporaneous market return, whereas the relation between turnover
probability and past performance
Watts, and
Wruck found
changes), the
generally insignificant. Consistent with our resuhs,
that turnover
market performance. Although
CEO
is
their
in
The
unrelated to industry performance after controlling for
sample
Wamer-Watts-Wruck
performance from the date when the
fiscal year.^^
is
CEO
data
much
smaller than ours (which
may be
is
based on
less noisy than ours since they
1,(XX)
measure
change was announced rather than over the preceding
is
that are consistent with
RPE
robust to these potentially important timing problems.
Several studies that did not directly
CEO
is
similarity of our results and theirs suggests that our evidence supporting
CEO turnover policy
between
Wamer,
test the
RPE
hypothesis nevertheless provide results
our findings. Coughlan and Schmidt (1985) found an inverse relation
turnover and abnormal stock-price performance (estimated as the cumulative
annual residual from the market model), and Weisbach (1988) obtained similar results by
measuring performance as shareholder return minus the return on a value-weighted market
portfolio.
entire
In addition,
ft
Shleifer,
top-management team leaves
abnormal performance
1
Morck,
is
and Vishny (1988) found that the probability
that the
negatively related to abnormal performance minus
in the industry.
In addition, Wamcr-Watls-Wruck also excluded cases in which the CEO title changed hands but the topmanagement team remained iniacl (for example, the chairman-CEO remained chairman but passed the CEO title to
the president). We have no way of idenuf>'ing these cases in our data.
Robert GiDBONS AND Kevin
J.
August, 1989
31
MuRPifY
Conclusion
Rewarding
performance
top-level executives based
in the industry or
on performance measured
market creates incentives to take actions increasing shareholder
and product markets
v^ealth while insuring executives against the vagaries of the stock
beyond
their control.
relative to aggregate
We expect RPE to be a common
feature of implicit
common
dismissal contracts because the potential benefit of filtering out
cost of measuring the performance of other firms
is
that are
CEO compensation
and
uncertainly -is high, the
small, and opportunities for sabotage and
collusive shirking are limited.
The empirical evidence presented above supports
CEO's pay and
the probability that a
CEO
remains
RPE
the
hypothesis: the revision in a
in his position for the
following year are
positively and significantly related to firm performance, but are negatively and significantly
Our
related to industry and market performance, ceteris paribus.
performance
to industry
is
more
likely to be evaluated relative to aggregate
movements, although our
market movements may
results also suggest that
CEO
market movements than relative
inability to detect an industry effect after controlling for
reflect the inappropriateness
of industry definitions based on SIC codes
for purposes of relative performance evaluation.
The
theoretical justification for rewarding top-level executives
performance
is strong,
and therefore the
fact that previous research
significant relative-performance effect has been troublesome.
The
more work remains
to provide completely convincing support for the
the
RPE
to
has failed to detect a
results in this paper begin to
resolve the puzzle by documenting a significant negative relation betv.'een
industry and market performance, but
on the basis of relative
CEO compensation and
be done before ihc data can be said
RPE hypothesis.
hypothesis with respect to current shareholder return,
with respect to changes in shareholder wealth (equation
Although our
we do
8),
results support
not find evidence for
RPE
lagged shareholder return
32
Robert Gibbons AND Kevin J. MuRPiiY
(Appendix), or accounting returns (Table
Finally,
5).
between the incentive and learning versions of the
RPE
we have made no
CEOs from
many
industry and market shocks, there are
are not protected from industry and market
attempt to distinguish
hypothesis.
Although the theory of relative p>erformance evaluation predicts
protect
August. 1989
...
movements
that optimal contracts will
aspects of a
CEO's wealth
The gains and
in stock prices.
that
losses
associated with stock options and inside stockholdings, for example, depend solely on the firm's
absolute stock-market performance and not on performance measured relative to the industry or
market. Therefore, although the CEO's compensation and probability of dismis.saLare panially
insured against industry and m:u"kct risk, other potentially
more important components of
his
wealth are not.
It
would be straightforward, however,
to design a
compensation plan
that duplicates the
rewards and losses from stock ownership or stock options after adjusting for industrj'- or
market-wide stock movements, and
compensation systems.
It
may be
it
is
puzzling that few firms have adopted such explicit
that current stock-option
and restricted-stock plans are easy
administer and offer tax advantages, and that the costs of suntching to
potential benefits. Alternatively, the distorted incentives induced
collusion, choice of reference group, or production externalities
top-level managers. Determining
future research.
why RPE
is
not
RPE
by
new systems outweigh
RPE
—may
more comprehensive
—
in
the
terms of sabotage,
in fact
is
to
be important for
an important goal for
Robert Gibbons
ANT)
KfviN
J.
August, 1989
33
MuRTin'
References
Anile, Rick, and Abbie Smith. 1986. "An Empirical Investigation of the Relative Performance
Evaluation of Corporate Executives." Journal of Accounting Research, Vol. 24, No. 1
(Spring), pp. 1-39.
Baiman, Stanley and Joel Demski. 1980. "Economically Optimal Performance Evaluation and
Control Systems." Journal of Accounting Research, Vol. 18 (Supplement), pp. 184-220.
1973. The Pricing of Options and Corporate Liabilities."
Journal of Political Economy, Vol. 81, No. 3 (May/June), pp. 637-54.
Black, Fischer and
M>Ton
Scholes.
1988. "Incentives in Academia: Why
Political Economy, Vol. 96, No. 3 (June), pp. 453-72.
Carmichael, H. Lome.
Is
There Tenure?" Journal of
Coughlan, Anne T. and Ronald M. Schmidt. 1985. "Executive Compensation, Management
Turnover, and Firm Performance: An Empirical Investigation." Journal of Accounting
and Economics, Vol. 7, Nos. 1-3 (April), pp. 43-66.
1984. "The Trouble with Tournaments."
(January), pp. 147-49.
Dye, Ronald A.
Economic Inquiry,
\'o\. 22,
No.
\
Dye, Ronald A. 1988. "Relative Performance Evaluation and Project Selection." Unpublished
paper,
Nonhwestem
University.
Gibbons, Robert, and Kevin J. Murphy. 1989. "Optimal Incentive Contracts in the Presence of
Career Concerns: Theory and Evidence." Unpublished paper, Massachusetts Institute of
Technology.
Holmstrom, Bengt. 1979. "Moral Hazard and Observability." Bell Journal of Economics, Vol.
10, No. 1 (Spring), pp. 74-91.
Holmstrom, Bengt. 1982. "Moral Hazard
2 (Autumn), pp. 324-40.
C, and Kevin
Jensen, Michael
J.
Murphy.
Incentives." Journal of Political
Lazear, Edward P.
1987.
in
Teams." Bell Journal of Economics,
1990. "Performance Pay and
{forthcoming).
\'o\. \3,
No.
Top-Management
Economy
"Pay Equality and Industrial
Politics."
Unpublished paper. Hoover
Institution.
Lazear,
Edward
P.,
and Sherwin Rosen.
Contracts." Journal of Political
Mood, Alexander, and Franklin
Edition.
New
Mookherjee, Dilip.
Economic
1981. "Rank-Order Tournaments
Economy, Vol. 89, No. 5 (October),
Graybill.
as
Optimum Labor
pp. 841-64.
1963. Introduction to the Theory of Statistics, 2nd
York:
McGraw-Hill.
1984.
"Optimal Incentive Schemes with
Many
Agents."
Review of
Studies, Vol. 51, No. 3 (July), pp. 433-46.
Morck, Randall, Andrei
Shleifer, and Robert W. Vishny.
Corporate Control." Unpublished paper, N^ER.
1988. "Alternative Mechanisms for
34
RobektGibbons andKj;\'ln J.MuRPifY
Murphy, Kevin
J.
1985. "Corporate Performance and Managerial Remuneration:
Analysis." Journal of Accounting
Stigler,
George
1
J.
August, 1989
1964.
(February).
and Economics, Wol.lyNos.
An
Empirical
1-3 (April), pp. 11-42.
Political Economy, Vol.
"A Theory of Oligopoly." Journal of--— "
,
12, No.
-
Vancil, Richard F. 1987. Passing the Baton: Managing the Process of
Harvard Business School Press.
Boston:
CEO
Succession.
Warner, Jerold B., Ross L. Watts, and Karen H. Wruck. 1988. "Stock Prices, Event
Prediction, and Event Studies: An Examination of Top Management Changes." Journal
of Financial Economics, Vol. 20, No. 1/2 (January/March), pp. 461-92.
Weisbach, Michael S. 1988. "Outside Directors and CEO Turnover." Journal of Financial
Economics, Vol. 20, No. 1/2 (January/March), pp. 431-60.
_..
_
Wolfson, Mark.
Rights:
1985.
"Tax, Incentive, and Risk-sharing Issues
in the
Allocation of Property
The Generalized Leasc-or-Buy Problem." Journal of Business, Vol. 58, No. 2
(April), pp. 159-72.
Robert Gibbons asd Ke\is
J.
Murphy
35
AUGUST, 1989
APPENDIX
ln(CEO Salary + Bonus) on Firm and
and Firsi-Diffcrcnce Spccificaiions
CocfTicicni^ of Ordinar)' Lcasi Squares Regressions of
Indusu^' Shareholder
Rcium
for Level
(t-slalisiics in parcniliescs)
Dcpcri dent Variable in 1000s of 1986-Dollars
Change in
In(CEO Pay)
Independent
Variable
Intercept
Rcium
Cunrcnl Slock
Lagged Slock Return
(Rii)
(R\[.\)
In
(CEO
Pay)
1
-Digit
lndusU7 Return
Lagged
1
(3)
(4)
.046
(CEO-spccific)
.043
(CEO-spccific)
.1792
.0166
.1746
(1.9)
(19.5)
.0284
—
.0275
.0222
.
--(2.5)
—
(3.2)
-.0893
-.1443
-.0855
-.0636
(-5.0)
(-9.5)
(-3.1)
(-2.3)
.0849
—
.0584
—
Industry Return Gii.])
Time Trend
Pay)
(2)
(20.2)
(IjO
-Digit
In
(CEO
(1)
(3.3)
Current
Change in
ln(CEO Pay)
(2.1)
(6.1)
—
(t)
.0256
—
—
—
.1676
(10.8)
/
Cumulative Stock Return
\
t
Z
^T=l
Cumulaiivc
1
-Digit
/^
y
—
'"^
—
.0830
-.0244
(-1.1)
(4.7)
Sample Size
r2
is
(20.8)
j
Year Dummies
Note: TTic sample
.1779
(22.0)
^
-^
Industry Return
—
Kit
(no)
(no)
(yes)
(yes)
7,696
9.361
7.696
9.361
.0652
.8941
.0769
.8965
dau reported in Forbes on 1,668 CEOs serving
and (4) include 1,668 individual inlcrcepls.
conslrucicd from longinadinal
1,049 firms from 1974-1986.
Columns
(2)
in
Regression Specificaiion:
t
t
Level:
.
In(wiL)
=
a;
-t-
at
-^
(Po-Pl)Rii
"^
Pi
Z
1=1
First Difference:
Aln(wii)
^a
+ PoRjt
-
Z lit
Kit + (Yo-Yl)Iit+ 7l
(Po-Pl)Rit-I + TO^it
t=l
-
(TO-Tl)Iit-l
"^
Vit
+
Eit
2385 005
Date Due
net 09
OCT.
1991
2 4 1996
,5-'^-v,
MIT
3
TDflO
UBRARIES DUPL
I
QDSTfiTT?
b