Vous êtes sur la page 1sur 20

MANAGEMENT SCIENCE informs ®

Vol. 56, No. 10, October 2010, pp. 1633–1652 doi 10.1287/mnsc.1100.1205
issn 0025-1909  eissn 1526-5501  10  5610  1633 © 2010 INFORMS

CEO Ability, Pay, and Firm Performance


Yuk Ying Chang
School of Economics and Finance, College of Business, Massey University, Wellington 6021, New Zealand,
y.chang@massey.ac.nz

Sudipto Dasgupta
Department of Finance, Hong Kong University of Science and Technology, Clear Water Bay,
Kowloon, Hong Kong, dasgupta@ust.hk

Gilles Hilary
Department of Accounting, Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong;
and Department of Accounting, HEC Paris, 78350 Jouy-en-Josas Cedex, France, acgh@ust.hk

D o chief executive officers (CEOs) really matter? Do cross-sectional differences in firm performance and CEO
pay reflect differences in CEO ability? Examining CEO departures over 1992–2002, we first find that the
stock price reaction upon departure is negatively related to the firm’s prior performance and to the CEO’s prior
pay. Second, the CEO’s subsequent labor market success is greater if the firm’s predeparture performance is
better, the prior pay is higher, and the stock market’s reaction is more negative. Finally, better prior performance,
higher prior pay, and a more negative stock market reaction are associated with worse postdeparture firm
performance. Collectively, these results reject the view that differences in firm performance stem entirely from
non-CEO factors such as the firms’ assets, other employees, or “luck,” and that CEO pay is unrelated to the
CEO’s contribution to firm value.
Key words: CEO ability; CEO pay; managerial labor market; firm performance
History: Received January 14, 2009; accepted May 11, 2010, by Wei Xiong, finance. Published online in Articles
in Advance August 20, 2010.

1. Introduction CEOs are in fact in a unique position to entrench


Do chief executive officers (CEOs) really matter? themselves and extract “rents.” Shleifer and Vishny
Does CEO ability have any effect on firm value, (1989) argue that managers can entrench themselves
performance, and CEO pay? A large literature takes by making manager-specific investments that make
the relevance of CEO ability as given; for exam- it difficult for shareholders to replace them. Bebchuk
ple, the distribution of CEO talent in the economy and Fried (2004) argue that CEOs can easily capture
provides the theoretical underpinnings for the tal- their boards and essentially set their own pay. Pay,
ent assignment models of Rosen (1982) or Gabaix according to this view, is determined more by man-
and Landier (2008). Yet, a plausible alternative view agerial power than by ability.
is that the firm’s performance is largely determined The recent financial crisis and the storm over the
by the nature of its core competence, the quality of pay of executives in financial firms have brought the
its products, other employees and stakeholders, its question of whether CEOs meaningfully add value to
life cycle, and even possibly by luck. According to the companies they manage, and whether their pay
this view, CEO ability is relatively unimportant for reflects ability or rent extraction, into even sharper
firm value. Collingwood (2009, p. 3), for example, focus. However, although these questions are central
reports that “James March, a management professor ones, empirical resolution of these issues has been dif-
at Stanford, goes so far as to say that in any well- ficult. The main problem is that CEOs are not ran-
run company that’s conscientious about grooming its domly assigned to firms, and isolating CEO attributes
managers, candidates for the top job are so similar in from firm attributes is difficult, particularly in a
their education, skills, and psychology as to be virtu- purely cross-sectional study. One of the best oppor-
ally interchangeable. All that matters is that someone tunities to do so is to study the change in a firm’s
be in charge.” Collingwood (2009, pp. 1, 2) further market value and performance around a CEO change.
wonders, “How much difference can a CEO make, The financial market’s reaction to the news of a CEO’s
anyway? The answer doesn’t seem as obvious today departure is one way to gauge the contribution of
as it did a few years ago.” An even more cynical view that CEO relative to a potential replacement. If CEO
is that, far from meaningfully increasing firm value, pay and firm performance reflect CEO ability, then
1633
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
1634 Management Science 56(10), pp. 1633–1652, © 2010 INFORMS

the financial market’s reaction to the departure news finds another managerial appointment and whether
should be related to the pay and performance prior the CEO is better paid in the new job or ends
to departure. Prior pay and performance should also up in a bigger firm. Our results strongly support
be related to the external labor market’s perception the managerial ability hypothesis. A one-standard-
of managerial ability (as measured by the subsequent deviation increase in the prior three-year buy-and-
labor market progress of the manager), and the subse- hold returns, increase in our measure of relative pay,
quent operating performance of the firm that loses the and decrease in residual abnormal returns (i.e., the
CEO. To test these predictions, we study a sample of part of abnormal announcement returns unexplained
298 CEO departures (both voluntary and forced) over by prior performance and pay) increase the probabil-
the 1992–2002 period, and consider evidence from the ity of successful labor market progress by 10%, 8%,
financial and labor markets and directly from firm and 6%, respectively.
performance after the CEO departure. Examination of the stock market’s response is very
First, we hypothesize that if financial markets did informative about a CEO’s contribution to firm value
not attribute differences in prior firm performance to when there are no confounding contemporaneous
differences in CEO ability, there would be no associa- announcements, because the market reacts to the
tion between the firm’s performance under the CEO’s news of the departure alone. The labor market out-
management and the stock price reaction to the news comes are also especially informative because the pro-
of the CEO’s departure. In contrast, if the financial cess of selection and recruitment is likely to assign
market attributes performance differences to differ- significant weight to managerial attributes that are
ences in the CEO’s managerial ability, one should observable to the recruiter but not to the researcher.
observe a negative relationship. Similarly, if CEO pay However, as Huson et al. (2004, p. 239) note, “stock
is unrelated to either the CEO’s managerial ability price reactions around the time of management
or to the CEO’s ability to extract rents (the so-called turnover reflect investors’ expectations regarding these
“skimming” view), there should be no association
outcomes, but do not reveal the outcomes them-
between the pay and the stock market’s reaction to the
selves.” Perception, rather than actual ability, could
departure news. We focus on the “relative” pay of the
also be relevant for the managerial labor market.
CEO in relation to the pay of the other four highest-
To investigate whether actual or perceived ability
paid executives in the company to filter out the effect
explain the results from the stock and labor mar-
of potentially unobserved factors that affect the pay
kets, we also examine whether the firm’s post-CEO-
levels of top executives in a company. Thus, if higher
departure performance is related to the prior per-
relative pay is associated with a more negative stock
formance, relative pay, and abnormal returns around
price reaction, the null hypothesis is rejected in favor
the announcement date. Consistent with the manage-
of the managerial ability hypothesis; if the associa-
tion is positive, it is rejected in favor of the skimming rial ability hypothesis, the firm’s industry-adjusted
hypothesis. Our empirical results are consistent with operating performance one or three years after the
the financial market’s associating prior performance departure is worse if the prior stock performance is
and CEO pay with the CEO’s managerial ability. better, the CEO’s relative pay is higher, and the abnor-
A one-standard-deviation increase in prior three-year mal stock returns around the departure announce-
industry-adjusted buy-and-hold stock returns leads ment date are lower. We find similar results for the
to a 1.4% decrease in abnormal stock return (from firm’s stock market performance over one to three
one day before to seven days after the departure years after the departure. The results are economically
announcement). A one-standard-deviation increase in significant. A one-standard-deviation change in prior
our relative pay measure leads to a 2.7% decrease in three-year buy-and-hold returns, our measure of prior
the abnormal stock return. relative pay, and residual abnormal returns is associ-
Next, we examine whether prior firm performance ated with an absolute change in the industry-adjusted
and CEO pay are related to the CEO’s success in return on assets, from one year before to one year
the managerial labor market. Given that the stock after the departure, by 0.5%, 0.6%, and 0.8%, respec-
market reaction is consistent with the recognition tively. Given that the mean and median industry-
of CEO ability, we hypothesize that more negative adjusted return on assets for the calendar year prior
abnormal returns are associated with better labor to the departure are 2% and 0.6%, respectively, these
market progress. In addition, if differences in prior magnitudes are substantial.
performance and relative pay reflect variations in An important advantage of our three-pronged
CEO ability, the CEO’s labor market progress should approach is that even if alternative explanations could
be better when the prior performance is stronger, be plausibly offered for any of the individual results,
and the prior relative pay higher. We classify labor it is difficult to offer one alternative explanation con-
market progress on the basis of whether the CEO sistent with all our results. For example, it might be
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
Management Science 56(10), pp. 1633–1652, © 2010 INFORMS 1635

argued that CEOs “jump ship” and quit voluntar- suggest that governance does matter for the pay-
ily when they expect future prospects to deteriorate performance relationships.
and are forced to quit (for example, as a result of a The rest of this paper is organized as follows.
governance change) when the prior performance is Section 2 reviews the prior literature. Section 3 dis-
poor. This may explain why prior performance is neg- cusses the data and our sample. Section 4 develops
atively related to announcement returns and future the testable hypotheses. Section 5 reports the empiri-
operating performance but not why managers’ labor cal results. Section 6 concludes the paper.
market progress should be positively related to past
pay. Such an explanation requires prior pay to be low
when performance is poor, quite the opposite of a 2. Related Literature
governance-change explanation. Two additional find- Our study is related to several lines of research.
ings further rule out this alternative explanation. First, One of our main objectives is to design tests that can
the departure announcement abnormal return of the identify cross-sectional differences in CEO ability rele-
firm the CEO leaves and the appointment announce- vant for firm value and performance. To identify such
ment abnormal return of the new firm the CEO joins evidence, we rely on a sample of CEO turnovers and
are negatively and significantly correlated. This is focus on three variables that potentially correlate with
consistent with the notion that the market value of CEO ability: one is the abnormal stock price reaction
the old firm will decrease more and that of the new around the CEO departure announcement, the second
firm will increase more if the CEO’s ability is higher. is the firm performance prior to the CEO departure,
Second, the operating performance of the firm the and the third is the relative pay of the CEO. We then
CEO joins improves more if the CEO’s relative pay in examine how the last two of these variables relate to
the previous job is higher and the stock market’s reac- the stock market reaction around the CEO turnover
tion to the departure news is more negative. For these announcement, and how all three relate to the success
two tests, we are necessarily restricted to CEO moves of the departing CEO in the managerial labor market,
to publicly traded firms, which significantly reduces and the subsequent performance of the firm losing
the sample size. However, our results still support the the CEO. We organize our discussion of the relevant
managerial ability hypothesis. literature around these key points.
Our paper makes several contributions to the lit-
erature. First, we provide evidence that CEO ability 2.1. Cross-Sectional Differences in CEO Ability
differences exist and affect firm value and perfor- The empirical literature on this topic is fairly limited,
mance. Second, our results show that CEOs of higher but there is support for the idea that these differences
ability also receive higher pay, a finding that con- exist and matter. For example, Lieberman et al. (1990)
trasts with those of Bebchuk et al. (2008).1 Bebchuk find significant CEO fixed effects on productivity in
et al. (2008) find that, in a sample of firms in which the U.S. and Japanese automobile industries. Bertrand
the managers are entrenched, the CEO’s pay rela- and Schoar (2003) find that CEOs have different man-
tive to the pay of the four other highest-paid exec- agerial styles that are carried as they go from one firm
utives in the company is negatively associated with to another. They provide evidence that these differ-
firm performance. They interpret this as evidence that ences matter for a wide range of corporate decisions.
high CEO relative pay represents rent extraction by They also suggest that management styles appear to
CEOs. In untabulated tests, we find that CEO rela- be related to performance and pay. Our tests, in con-
tive pay is no longer significant for a subsample of trast, are specifically tailored toward relating prior
departures with poor governance characteristics, as performance and pay to financial and labor market
measured by the G-Index (Gompers et al. 2003). How- outcomes and operating performance changes when a
ever, our results are stronger, both economically and CEO leaves the firm. Moreover, because our primary
statistically, in the complementary subsample of firms source of identification of CEO changes is the Execu-
with good governance characteristics. These results Comp database, which starts in the year 1992, we do
not have a sufficiently long panel to reliably identify
1
managerial fixed effects.
We recognize that because our results are based on a sample
of CEO turnovers, we need to be cautious in this interpretation.
It is possible that the CEOs who are strongly entrenched very sel- 2.2. Market Reaction to Managerial
dom quit. Thus, our sample may consist mostly of high-quality Departures or Sudden Deaths
CEOs and low-quality CEOs who are not entrenched. If true, the Several studies examine the stock market’s reaction to
number of entrenched low-quality CEOs may be lower in our sam- the news of managerial departures or sudden deaths.
ple than in the overall population. However, the median entrench-
ment index for firms in Bebchuk et al. (2009) sample is identical to
The earlier papers find mixed evidence. Most of these
that in ours over the same sample period, suggesting that we are results do not distinguish between voluntary and
not capturing turnovers of less-entrenched managers. involuntary turnovers. Reinganum (1985) considers
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
1636 Management Science 56(10), pp. 1633–1652, © 2010 INFORMS

changes in the position of president and/or chair- 2.3. Prior Performance and Labor
man and finds no systematic reaction around the Market Outcomes
announcements of the executive departures. Warner Fee and Hadlock (2003) study whether the prior per-
et al. (1988) consider a sample of CEO departures. formance of a firm affects the probability that a top
They do not find any significant market reaction at executive in that firm will be “raided” by another
the announcement either in their overall sample or in firm. The main finding is that firms that outperform
the sample of forced departures.2 In contrast, several their benchmarks are likely to lose their executives to
more recent studies find an effect on the stock price. CEO positions elsewhere. Fee and Hadlock (2004) find
Khanna and Poulsen (1995) find a negative average that executives leaving S&P 500 firms do not typically
announcement effect for managerial replacements in do well, and seem to move to inferior jobs, in terms
firms that eventually sink into financial distress. Denis of both relative firm size and relative pay.5 In con-
and Denis (1995) find a significant positive effect for trast to these studies, we rank the CEOs in terms
forced turnovers. Hayes and Schaefer (1999) look at of their subsequent labor market success and try to
returns around the announcement dates for a sample examine whether the prior performance, the CEO’s
of CEO-to-CEO moves and non-CEO-to-CEO moves, relative pay, and the stock price reaction to the depar-
and they compare the announcement returns with ture announcement predict the degree of success in
those for a control sample of CEO deaths. They find the managerial labor market.
that for the former sample, abnormal returns average
−151%, compared to 3.82% for the latter.3 2.4. Postdeparture Performance
Johnson et al. (1985) and Nguyen and Nielsen Firm performance change after CEO turnover is
(2010) study abnormal returns around unexpected potentially an important way to identify the value
deaths of top executives. A major advantage of this of a CEO to the firm. Denis and Denis (1995) find
approach is that, unlike forced CEO changes, there that forced resignations of top managers are preceded
are unlikely to be many other types of concurrent by large and significant declines in operating perfor-
events or announcements that would confound the mance and followed by large improvements in per-
announcement period returns, and unlike retirements, formance, whereas normal retirements are followed
the event is not likely to be anticipated.4 Johnson by smaller improvements in operating performance.
et al. (1985) find that there is no average effect over- Huson et al. (2004) show that operating performance
all, but the abnormal returns are lower for firms with deteriorates prior to a CEO turnover and improves
higher CEO pay and prior performance. Nguyen and subsequently, and that these effects are strongest for
Nielsen (2010) find that the pay of the executive is forced turnovers. Of particular relevance to our anal-
negatively related to the abnormal return around the ysis, the two-day announcement period abnormal
announcement of the executive’s death. Nguyen and returns around CEO turnover are positively related
Nielsen (2010) also determine that executives keep to the change in operating performance. These results
around 80% of the value they create. Our tests on the suggest that price reaction around the CEO departure
stock market’s reaction to departure news extend the announcement can be a proxy for the value of the
CEO. However, one concern with studies of firm per-
results of the above studies to a sample of CEO depar-
formance around CEO turnover is that, like studies of
tures that are not sudden-death events. One advan-
announcement effects, identification of a CEO effect
tage of studying CEOs who stay alive is that we can
can be confounded by the nature of circumstances
use their subsequent labor market performance as an
that caused the turnover. Bennedsen et al. (2009), like
additional measure of ability.
Johnson et al. (1985), study sudden deaths of CEOs
and their family members to avoid this identification
2
Reinganum (1985) finds a small positive reaction in the sample of problem. They find that CEO deaths are strongly cor-
firms in which the chairman is replaced by an outsider (10 cases).
Warner et al. (1988) find a small positive effect in the subsample of
related with declines in operating profitability, asset
outside successions. growth and sales growth. Unlike the above studies,
3
Hayes and Schaefer (1999) find that the more negative the our tests explicitly link postdeparture performance
announcement effects, the larger is the ratio of assets of the new to the three variables that potentially reflect CEO
firm to the old firm, consistent with the implication of assignment ability—the prior performance of the firm under the
models (Rosen 1982, Gabaix and Landier 2008) that suggest that CEO’s management, the relative pay of the CEO, and
more capable managers are likely to move to bigger firms. They the stock market’s reaction to the departure news—for
also find that abnormal returns for firms that lose top managers
and those that hire the same individuals as CEOs are weakly neg-
a sample of CEO departures that include both volun-
atively related, suggesting that the loss to one firm is related to the tary and forced turnovers and are not sudden-death
gain to the other. events.
4
See Warner et al. (1988) and Denis and Denis (1995) for discussions
5
of the difficulty of interpreting event study results around CEO Several papers examine what former CEOs do. See Brickley et al.
turnovers. (1999) and Fahlenbrach et al. (2010).
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
Management Science 56(10), pp. 1633–1652, © 2010 INFORMS 1637

2.5. Compensation, Skill, and Entrenchment the proxy statement for information on the incom-
The issue of whether differences in CEO pay reflect ing CEOs and identify the firms where they were
differences in ability or in power is one of the most previously employed. If an incoming CEO was the
contentious issues in academic and policy circles. CEO of another company not already included in our
Some authors have argued that the compensation of sample, this gives us an additional case of a CEO
U.S. executives is too high and too insensitive to departure. We then retrieve information from proxy
their performance (e.g., Bebchuk et al. 2002). Bebchuk statements on the length of the CEO’s tenure and
and Fried (2004) argue that many of the features of exclude cases in which the CEO tenure was less than
CEO pay reflect that corporate insiders essentially set 18 months. We collect compensation data from the
their own pay. Bebchuk and Fried (2004) conclude proxy statements.7 We eliminate observations where
that higher CEO pay reflects skimming by entrenched the CEO change was caused by death or by merger
CEOs rather than talent. Others, however, argue that and acquisition activity. Finally, we exclude all cases
most available evidence is consistent with the efficient in which the departing CEO continued to have any
contracting view (Edmans and Gabaix 2009 provide relationship with the company, such as chairman of
a survey of the theoretical literature on the efficient the board or consultancy agreements. As a result,
contracting view). Based on a model in which agents we identify 434 CEO departures. However, we lose
gradually learn about CEO ability, Taylor (2010) esti- 54 observations because we cannot compute cumu-
mates that CEOs capture only 20%–33% of the surplus lative abnormal returns from day −1 to +7 around
resulting from good news, and they bear 8%–20% of the departure announcement as the companies are not
the negative surplus resulting from bad news, and the publicly traded, 67 additional observations because
difference is not statistically significant. we cannot compute three-year industry-adjusted buy-
and-hold returns, 10 observations because the rela-
tive pay of the CEO cannot be reliably computed,
3. Sample and 5 observations because other variables used in
3.1. Departure Sample our regression specification are missing. This leaves
We compile a sample of CEOs who left their jobs us with 298 CEO moves.
before the age of 61 (to minimize the possibility that In addition, we identify 44 CEO death events for
the departure is a retirement) using two databases: the the period from 1992 to 2002. As discussed in §2,
Annual Compensation Surveys published in Forbes several studies examine unanticipated deaths of top
(covering the period from 1992 to 2000) and Compu- executives, because this is a very clean way to iden-
stat’s ExecuComp database (covering CEOs of com- tify the effect of a CEO’s departure on a firm’s future
panies in the S&P 500, the S&P MidCap 400, and prospects. Because we cannot include these cases in
the S&P SmallCap 600 from 1992 through 2002).6 all our tests (as some of our tests require us to fol-
In addition to the availability of proxy statements low the subsequent career progress of the CEO), we
around the time of the CEO departure, we also exclude these cases completely in all results reported
require that the announcement be recorded in the in our tables. However, inclusion of these cases where
LexisNexis database. We use ExecuComp to iden- feasible does not affect our conclusions.
tify changes in the identity of the CEO of a firm
3.2. Comparison with Other Samples
from one year to another. For firms included in the
One difficulty in putting together the results of pre-
Forbes surveys, we have information on the length
vious studies on CEO movements discussed in §2 is
of the CEO’s tenure, and we can identify firm years
that they are often based on different samples, and
where the CEO had been in that position for no
the extent to which findings are robust across all
more than one year, indicating a recent CEO turnover.
such samples is not clear. Our sample brings together
For each CEO turnover, we search LexisNexis or
within one framework several related hypotheses
about managerial ability and reputation formation
6
The ExecuComp database includes active and inactive firms in for managerial turnovers in a large cross-section of
the S&P 1500 index. The data is available from 1992. The S&P firms. There are some important differences in sample
1500 firms are operating U.S. companies whose stocks are widely
available to investors and are highly liquid. The firms included in
composition between our sample and those used in
the Forbes Annual Compensation Survey are the 800 biggest com- prior studies. For example, we identify a total of 131
panies in the United States (“as measured by a composite rank- cases of CEO-to-CEO moves—a larger number than
ing of sales, profit, assets, and market value” (Forbes 2001)). We the 24 cases identified by Hayes and Schaefer (1999),
find 27 cases in which the old firm in the sample of CEO depar-
tures is included in the Forbes surveys, but does not appear in the
7
ExecuComp database. For example, Hill Department Stores, DQE Because the CEOs identified by this process were not all taken
Holdings Inc., and Continental Airlines are included in the Forbes from the ExecuComp database, we rely on compensation data from
surveys, but do not appear in the ExecuComp database. proxy statements.
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
1638 Management Science 56(10), pp. 1633–1652, © 2010 INFORMS

the 43 cases in sample 1 of Fee and Hadlock (2003), new directorships or there is no information indicat-
the 5 cases in sample 2 of Fee and Hadlock (2003), ing that the CEO is appointed to any new managerial
and the 12 cases in Fee and Hadlock (2004). Huson position. Progress is considered to be “satisfactory”
et al. (2004) study a sample of voluntary turnovers for the rest of the CEOs, who end up with manage-
and find a positive relationship between announce- rial positions in other public or private companies. We
ment period abnormal returns and change in operat- drop five cases in which the departing CEOs became
ing performance. However, their sample of voluntary politicians or educators. The inclusion of these cases
turnovers includes retirements, and the announce- in any particular category has no material impact on
ment period abnormal returns are significantly posi- our results.
tive. Therefore, it is unlikely to be a sample of depart- In our four-way classification, we further parti-
ing CEOs with “above average” ability. tion and order the unsatisfactory group as follows.
A CEO is classified as having experienced very poor
4. Hypothesis Development progress and given rank 1 if there is no information
In this section, we develop our testable hypotheses. indicating that he is appointed to any new position
First, we discuss the three key dependent variables (223 cases, 51%). He is classified as having experi-
corresponding to our three main tests. Next, we dis- enced “poor progress” and given rank 2 if he takes up
cuss the three explanatory variables that are the focus only new directorships (33 cases, 8%). The satisfactory
of our main testable hypotheses. progress group is also partitioned into two groups.
A CEO is classified as having made “good progress”
4.1. Key Dependent Variables and given the highest rank of 4 if any of the follow-
4.1.1. Cumulative Abnormal Returns (CARs). In ing criteria hold: (i) the CEO ends up as the CEO of
our first test discussed below, we examine how the another public company and either has a new salary
stock market’s reaction to the departure news of that exceeds his old salary8 or joins a company that is
the CEO is related to CEO relative pay and prior larger in size (following Gabaix and Landier 2008 and
firm performance—variables that potentially corre- Edmans et al. 2009, we define size as the market value
late with CEO ability. The stock market’s reaction, of equity plus the book value of debt)9 than his old
however, may have incremental information content company (49 cases);10 (ii) the CEO ends up as a non-
about CEO ability, perhaps reflecting more recent CEO executive of another company and has a higher
news, not reflected in pay or performance. There- salary than in his old company (6 cases); or (iii) the
fore, in our subsequent tests, we use the residual CEO joins a private company as a CEO within one
from the cumulative abnormal return (CAR) regres- year (17 cases). This gives us 72 cases (17%) of good
sion of our first test (Equation (1) in §4.3.1) as an progress. The remaining CEOs are classified as hav-
additional measure of CEO ability. In the tabulated ing made “moderate progress” (106 cases, 24%) and
regressions results, CARs are based on the market assigned a rank of 3.11
model estimated over a period from 300 days before Panel A of Table 1 reports subsequent employment
to 30 days before the event date. All tabulated results details for the good progress and moderate progress
are based on CARs over the (−1 day, +7 days) win- groups. Not surprisingly, most of the good progress
dow around the announcement date. However, as
we discuss below, our results are similar for CARs
8
over the (−1 day, +1 day) window (CAR−1 +1) as New salary is the dollar value of the base salary (including both
cash and noncash elements) earned during the first full fiscal year
well as for CARs based on the Fama-French three under the new appointment. Old salary is the dollar value of the
factor model (Fama and French 1993) over both the base salary earned during the last full fiscal year under the old
−1 +1 and −1 +7 windows (FFCAR−1 +1 and appointment.
FFCAR−1 +7, respectively). 9
Specifically, we define size as Compustat item data199 × data25 +
data6 − data60 − data74.
4.1.2. Postdeparture Labor Market Progress. We
10
use the LexisNexis database to collect information Only in four of these cases was the salary criteria satisfied, but
not the size one. As an alternative, we completely ignored the size
on the career progress of the departing CEO. After and salary criteria, and good progress was defined on the basis
identifying a CEO departure that meets our crite- of higher pay at the new company, where pay includes salary,
ria mentioned in §3.1, we search for an announce- bonus, total value of restricted stock granted, total value of stock
ment on any subsequent appointment taken up by the options granted, and all other compensation. Our results did not
departing CEO within three years of his departure. change qualitatively, and all variables of interest remained statisti-
cally significant.
We first create a two-way classification and then cre- 11
Panel A of Table 1 reports the career progress of CEOs in our
ate a four-way classification to see whether our results
initial sample of 434 CEO moves. For our labor market progress
are robust to a finer classification scheme. regressions, we have 298 cases that satisfy all the data require-
The two-way classification considers labor market ments. For this sample, the distribution of “progress” is 17% good,
progress as “unsatisfactory” if the CEO only takes up 23% moderate, 7% poor, and 53% very poor.
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
Management Science 56(10), pp. 1633–1652, © 2010 INFORMS 1639

Table 1 CEO Career Progress cases are CEO appointments in the new company.
Panel B of Table 1 reports summary statistics on new
Panel A: Postdeparture career progress
salary in relation to old salary and the size of the new
Moderate Good company in relation to that of the old company for
progress progress the good progress group and the moderate progress
(Frequency) (Frequency) group when the new company is a publicly traded
company. For the good progress group, the salary in
Top manager at another public company
the new company is 47% (41%) higher if we consider
Chairman, CEO, and president 5 9
Chairman and CEO 14 16
the mean (median) salary ratio. The new company
CEO and president 17 23 is bigger—six times larger if we consider the mean
CEO 5 1 size ratio and 72% larger if we consider the median.
Chairman 10 0 For the moderate progress group, the salary is 34%
Vice chairman, chief accounting 0 1 (28%) lower if we consider the mean (median) salary
and financial officer ratio. Size is 60% larger (30% smaller) if we consider
Vice chairman 1 0
Chief operating officer and 0 1
the mean (median) size ratio.
executive vice president 4.1.3. Postdeparture Change in Performance. To
Chief operating officer 0 1 examine how the performance of the firm changes
Chief financial officer and 0 2
after the departure of the CEO, we consider an
executive vice president
President 3 1 operating performance measure.12 In particular, we
Group vice president 2 0 examine the change in industry-adjusted return on
Senior vice president 1 0 assets (ROA) from the year before to the first or the
Vice president 2 0 third year after the year of the managerial depar-
Senior advisor 1 0 ture (ROA1 or ROA3). ROA is adjusted by sub-
Top manager at a private company tracting the contemporaneous median measure for all
Chairman, CEO, and president 3 1 non-CEO-turnover firms with the same two-digit SIC
Chairman and CEO 8 5 code. To facilitate comparison between the regres-
CEO and president 7 6
sions of ROA1 and those of ROA3, we calculate the
CEO 6 5
Chairman 10 0 annualized change in industry-adjusted ROA.13
Vice chairman 2 0
President 2 0 4.2. Key Explanatory Variables
Vice president 1 0
4.2.1. CEO’s Relative Pay (COMP). The pay of
Advisor 2 0
Partner 4 0 the CEO relative to that of the other top executives
of the company could potentially measure the CEO’s
Panel B: New vs. old positions ability or CEO power. Bebchuk et al. (2007, p. 1)
Moderate progress Good progress term this variable CEO centrality and point out that
it “might reflect the relative significance of the CEO in
Median Mean Median Mean
terms of abilities, contributions, and power. A greater
New salary ÷ old salary 07243 06659 14100 14782 relative significance of the CEO might come from
New employer’s size 06978 16019 17226 64310 being a ‘star CEO’—the CEO having (or being per-
÷ old employer’s size ceived as having) superior talents or qualifications
Notes. Panel A reports the postdeparture career progress of the departing that enhance the CEO’s relative contribution to the
CEOs. Panel B compares the departing CEO’s salary in his new and old posi- firm, as well as outside opportunities.” However, as
tions and the size (data199 × data25 + data6 − data60 − data74 using Com- these authors point out, higher CEO relative pay
pustat data) of the new and old firms. The good progress category comprises
could also reflect skimming or CEO entrenchment.
CEO departures that meet any of the following three conditions: (i) the CEO
ends up as the CEO of another public company within three years of the We use three alternative measures of the CEO’s rel-
departure and either has a new salary that exceeds his old salary or joins ative pay (COMP). All three measures consider the
a company that is larger in size (market value of equity plus book value of CEO’s pay in relation to the pay of the other four
debt) than his old company; (ii) the CEO ends up as a non-CEO executive
of another company within three years of the departure and has a higher
12
salary than in his old company; or (iii) the CEO joins a private company as a Operating performance measures have been used in the previ-
CEO within one year. The moderate progress category includes the remaining ous literature to study CEO turnovers (e.g., Denis and Denis 1995,
departing CEOs who take up a managerial position in another private or pub- Hotchkiss 1995, Huson et al. 2004).
13
lic company within three years of the departure and meet none of the three Specifically, for ROA1, we divide the difference between the
good progress conditions. New salary is the dollar value of the base salary industry-adjusted ROA for the year t + 1 and that for the year t − 1
(including both cash and noncash elements) earned during the first full fiscal by two whereas for ROA3, we divide the difference between the
year under the new appointment. Old salary is the dollar value of the base industry-adjusted ROA for the year t + 3 and that for the year t − 1
salary earned during the last full fiscal year under the old appointment. by four.
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
1640 Management Science 56(10), pp. 1633–1652, © 2010 INFORMS

highest-paid executives in the company. It is impor- ROA. BHAR is the three-year industry-adjusted buy-
tant to benchmark the CEO’s pay against that of the and-hold stock return (BHAR). Industry adjustment is
other highest-paid executives in the company because at the two-digit SIC level. As robustness checks, we
executive pay in a company is likely to depend on also consider unadjusted buy-and-hold returns, buy-
unobservable factors, and this type of benchmarking and-hold returns adjusted by subtracting the size and
partially filters out the effect of such factors on the book-to-market matched portfolio returns, as well
CEO’s pay. For example, a company may have a high- as those adjusted by subtracting returns of indus-
pay policy for its top executives, and a higher abso- try, size, and book-to-market matched portfolios.16
lute pay for the CEO in such a company need not We also include the firm’s industry-adjusted return
necessarily identify an especially capable CEO. on assets, ROA, as an additional measure of prior per-
Our first pay measure, relative_total_pay, is sim- formance. If stock returns are noisy measures of per-
ply the ratio of the CEO’s total pay to the total formance, there may be incremental information in
pay of the four other highest-paid executives in the firms’ accounting performance measures.
company over the last three full fiscal years before 4.2.3. Residual CARs (RCARs). The stock mar-
the CEO’s departure, and is similar to the Bebchuk ket’s reaction to the departure news potentially could
et al. (2008) CEO-centrality measure.14 The second contain information about the value of the CEO to
measure, abnormal_CEO_pay, is the three-year prede- a company. As discussed above, this variable (CAR)
parture average of the residual from a regression in is the dependent variable in one of our tests to be
which the CEO’s total pay over a full fiscal year discussed in §4.3.1. However, to allow for the possi-
is the dependent variable, and the explanatory vari- bility that the market might have information about
ables include the sum of the total pay of the four CEO ability that is not already reflected in prior per-
other highest-paid executives in the same company, formance and CEO pay, we include the residual from
sales growth, total assets, year, and industry indi- a regression of CAR on other predeparture variables
cator variables. We include total assets and sales (including COMP, BHAR, and ROA), in Equation (1)
growth because the scale of a firm’s operation has in §4.3.1, as an additional explanatory variable in
been consistently found to be one of the most sig- other tests.
nificant determinants of CEO compensation in earlier The announcement period return could be noisy,
studies. There is also evidence that CEO pay is sub- because other types of announcements may cluster
ject to widespread benchmarking, so we also include around the CEO’s departure announcement. Noise in
industry and year indicator variables to capture more the announcement period returns that is orthogonal
aggregative effects on CEO pay.15 The third measure, to the impact of a CEO’s departure on firm value,
abnormal_pay_difference, is the three-year predeparture per se, is not a problem for our analysis and biases
average residual from a regression in which the dif- our tests against finding significant effects. It is a more
ference between the CEO’s total pay and the average serious problem if the information released is corre-
of the total pay of the four other highest-paid execu- lated with the type of departures in such a way that
tives in the same company is the dependent variable, it biases our tests in favor of evidence of managerial
and the explanatory variables include sales growth, ability.
total assets, year, and industry indicator variables. In Table 2, we report, for each type of labor market
4.2.2. Prior Performance (BHAR and ROA). We progress classification, the percentage of times a given
include two variables in our regressions to reflect type of news announcement occurs from one day
the performance of the firm under the CEO’s man- before to seven days after the departure announce-
agement prior to the CEO’s departure, BHAR and ment. The “good performance” category includes
news items reporting an improvement in firm perfor-
mance, such as a dividend increase or an improve-
14
Our pay measure is the same as TDC1 in the Execucomp ment in earnings, sales or operating performance. The
database. Total pay includes salary, bonus, total value of restricted
stock granted, total value of stock options granted (using S&P’s
“poor performance” category includes news reporting
modified Black-Scholes model), long-term incentive payouts, and deterioration in firm performance, such as a dividend
all other compensation. Because our sample of CEOs is not cut, a decrease in earnings or sales, or deterioration in
restricted to firms in the S&P 1500, we obtain the primary informa- operating performance. The “other good news” cate-
tion from the proxy statements. gory comprises news reporting other good news asso-
15
In the regressions for pay, the firm characteristics (sales growth ciated with the company, such as an upgrade in its
and total assets) are either marginally significant or not significant
at all—presumably, the pay of the other highest-paid executives
16
subsumes many firm characteristics that determine the pay level We use 10 size groups and 5 book-to-market groups. A number of
in the company and is itself highly significant. Year and industry studies use prior stock returns as a measure of performance under
indicator variables are also generally significant and explain a sub- the management of a CEO, e.g., Warner et al. (1988) and Fee and
stantial part of the variation. Hadlock (2003).
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
Management Science 56(10), pp. 1633–1652, © 2010 INFORMS 1641

Table 2 Frequency of Other News Announcements outcomes in the managerial labor market.17 As dis-
cussed below, this biases our tests against finding evi-
Progress
dence consistent with managerial ability.
All Very poor Poor Moderate Good In our regressions, we explicitly control for such
Type of news (Proportion)
announcements by incorporating indicator variables
Good performance (%) 966 1018 303 1143 845 corresponding to the different types of the other
Poor performance 851 1239 1212 381 141 announcements during the departure announcement
Other good 299 310 000 381 282 period. Although we do not tabulate the coefficients
Other bad 460 575 606 381 141 of these indicator variables (except in Table 4), the
Expansion 437 354 909 571 282
poor performance and the other bad news news indi-
Restructuring 345 398 303 286 282
Total (number of obs.) 435 226 33 105 71 cators are generally significant in our regressions.

Notes. This table reports the frequency of six other types of new announce- 4.3. Hypotheses and Tests
ments over the same window as that of CAR(−1, +7) around the date of the
announcement of a CEO departure. The good performance category includes 4.3.1. Stock Market Reaction to Departure News.
news reports of an improvement in the performance of the firm, such as We first examine whether financial markets relate
an increase in its dividend payment or an improvement in earnings, sales value creation to CEOs. To this end, we regress the
or operating performance. The poor performance category includes news abnormal returns around the departure news (CARs)
reports of deterioration in the performance of the firm, such as a dividend cut,
a decrease in earnings or sales, or deterioration in operating performance.
on the performance measures prior to the CEO depar-
The other good category includes news reports of other good news con- ture (BHAR and ROA) and his prior relative pay
cerning the company, such as an upgrade in its credit rating. The other bad (COMP):
category includes news reports of other bad news concerning the company,
such as a downgrade in its credit rating. The expansion category includes CAR =
0 +
1 BHAR +
2 ROA +
3 COMP
articles revealing an expansion in the company’s business, including the for-
+
4 SIZE +
5 (other news announcement
mation of partnerships and strategic alliances and the winning of new con-
tracts. The restructuring category includes news revealing a restructuring of indicators) +  (1)
the company, including a reduction in the number of company employees
and the sale of plant. The good progress category comprises CEO depar- Our first null hypothesis is that the market does
tures that meet any of the following three conditions: (i) the CEO ends up as not associate prior performance with CEO ability.
the CEO of another public company within three years of the departure and
This null hypothesis reflects the view that a firm’s
either has a new salary that exceeds his old salary or joins a company that
is larger in size (market value of equity plus book value of debt) than his old performance is primarily determined by non-CEO
company; (ii) the CEO ends up as a non-CEO executive of another company factors such as its core competencies, product mix,
within three years of the departure and has a higher salary than in his old the innovativeness of other employees, or simply fac-
company; or (iii) the CEO joins a private company as a CEO within one year. tors outside the control of the CEO, such as shifts
The very poor progress category includes those CEOs who have not secured in demand for its products, prices of its inputs, or
a new appointment within three years of their departure from their previous
exchange rate movements. If this null hypothesis is
CEO positions. The poor progress category includes those CEOs who have
taken up only new directorships within three years of their departure from true, the market would not associate prior perfor-
their previous CEO positions. The moderate progress category includes the mance with the CEO running the firm, and BHAR or
remaining departing CEOs who take up a managerial position in another pri- ROA would not have any effect on the market’s reac-
vate or public company within three years of the departure and meet none of tion to the departure news, i.e.,
1 and
2 will be zero.
the three good progress conditions. An alternative view is that the market perceives the
ability of the CEO to be an important determinant of
credit rating. Similarly, the “other bad news” category firm performance. The null hypothesis will be rejected
comprises other bad news associated with the com- against this alternative hypothesis if either
1 or
2 is
pany, such as a downgrade in its credit rating. The negative.
“expansion” category reflects news items revealing Our second null hypothesis is that the market does
an expansion in the company’s business. Finally, the not associate higher relative CEO pay with either
“restructuring” category includes news items reveal- higher CEO ability or more skimming by the CEO.
ing a restructuring of the company, including a reduc- A possible reason for cross-sectional variation in rel-
tion in the number of company employees and the ative CEO pay that does not rely on differences in
either the ability or “power” of the CEO is that pay
sale of company plant. Table 2 indicates a higher inci-
responds to “luck” (i.e., factors outside the control of
dence of announcements that indicate poor perfor-
the CEO). Another possible reason is that manage-
mance or convey other types of negative informa-
ment of certain types of organizations requires higher
tion surrounding departures that are associated with
poor labor market progress. Thus, these other types 17
The evidence in Table 3 is consistent with this intuition: the aver-
of announcements make it less likely that we can age announcement effects for poor and very poor progress classifi-
find a positive stock price reaction when CEO depar- cations are insignificantly different from zero, rather than positive
tures are due to poor performance, or result in poor as might be expected.
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
1642 Management Science 56(10), pp. 1633–1652, © 2010 INFORMS

compensation for more managerial effort or risk bear- a potential employer and thus have an independent
ing. If this null hypothesis is valid, the coefficient effect on his labor market progress, it could also affect
of COMP,
3 , should be zero. If, on the other hand, the other parameter estimates. For example, an older
CEO pay reflects CEO ability, then
3 should be neg- CEO who has managed a firm well may want to “quit
ative. Finally, the null hypothesis could be rejected while still at the top” and take early retirement or a
in favor of the view that variations in relative CEO less active position in another company. As a result,
pay reflect variations in CEO power across firms. we report all our estimates of Equation (2) not only
Bebchuk et al. (2007) argue that the relative pay of for the overall sample but also for the subsample for
a CEO (“CEO-centrality”) measures how entrenched which AGE is below median.
the CEO is within the firm. If CEO pay responds Our first null hypothesis for this test reflects the
to luck, but responds more to good luck than bad same arguments as that for specification (1). If the
when CEOs are entrenched, higher relative CEO pay managerial labor market does not associate the prior
reflects rent extraction. If this view is true, the depar- performance and the announcement period returns
ture of an overpaid manager may be good news for with CEO ability, the coefficients i , i = 1 2 3 will
the shareholders and the coefficient associated with all be zero. This null hypothesis reflects the view
COMP may be positive. that firms perform differently due to reasons unre-
4.3.2. Labor Market Outcomes. We next examine lated to CEO ability, such as the uniqueness of prod-
whether the labor market rewards managers for per- ucts, stakeholder relationships, the innovativeness of
ceived managerial ability. Specifically, we estimate the other employees, or exogenous shocks. On the other
following model: hand, if better prior performance and a more nega-
tive market reaction are regarded by the managerial
PROGRESS = 0 + 1 RCAR + 2 BHAR + 3 ROA
labor market as indications of higher managerial abil-
+ 4 COMP + 5 AGE + 6 SIZE ity, then 1 should be negative and i , i = 2 3 should
be positive.
+ 7 (other news announcement
Our second null hypothesis also reflects the same
indicators) +  (2) arguments as that for specification (1). If the labor
market does not associate higher relative pay for
The dependent variable is the progress rank of the
the CEO with higher CEO ability or more CEO
CEO in the managerial labor market. We first esti-
power, the coefficient 4 in specification (2) should
mate a probit model in which managerial progress is
be zero. However, if CEO ability is recognized both
assigned a dichotomous ranking—1 if the progress is
within the firm and by the managerial labor mar-
“satisfactory” (i.e., either good or moderate progress)
ket, and rewarded by means of higher pay relative to
and 0 otherwise. Subsequently, we estimate an
other executives inside the firm and better labor mar-
ordered logit model in which progress has a rank
ket progress, respectively, the coefficient 4 should
from 1 to 4, with 4 corresponding to “good progress”
be positive. Finally, if higher CEO relative pay is
and 1 corresponding to “very poor progress.” The
four-way classification has the limitation that it associated with more skimming, the labor market is
assumes a “linear” effect of the independent variables unlikely to reward CEOs with better positions, and 4
from one category of progress to another; however, it should be negative.
has the advantage that it takes account of more infor- 4.3.3. Postdeparture Firm Performance. The tests
mation. For our results to hold under the four-way discussed above relate to whether or not financial
classification, the effects would have to be monotonic and labor markets associate higher relative CEO pay
across the different ranks. and better prior performance with higher CEO ability.
Compared to the specification in Equation (1), we It is important, however, to examine whether the per-
add two more explanatory variables. One is RCAR, ceived ability differences are “real.” In other words,
the regression residual from Equation (1). If there is actual firm performance after the CEO’s departure
is information content about managerial ability in consistent with relative pay, prior performance, and
the market’s reaction to the departure announcement, the market’s reaction to departure being systemati-
RCAR is likely to capture the other relevant informa- cally related to CEO ability differences?
tion about managerial ability that is not contained in To address this question, we examine whether the
the prior performance (BHAR) and the CEO’s relative change in the industry-adjusted ROA of a firm that
pay (COMP).18 The other additional variable is CEO
age (AGE). AGE could not only affect the CEO’s util-
The main advantage of including RCAR is that the coefficients of
ity from another employment or his attractiveness to COMP and BHAR then reflect their direct effects on the dependent
variable as well as any indirect effect that “works through” CAR.
18
Note that the coefficient of RCAR in (2) is the same as that of CAR Because ROA is insignificant in Equation (1), we drop ROA in gen-
when all other explanatory variables in (1) are also present in (2). erating the residual for subsequent analysis.
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
Management Science 56(10), pp. 1633–1652, © 2010 INFORMS 1643

loses its CEO is associated with RCAR, BHAR, and longer horizon, they are likely to be somewhat affected
COMP—variables that are potentially correlated with by confounding events.
the market’s perception of CEO ability. Specifically, An important additional advantage of our ap-
we estimate the following model: proach is that collectively, the three sets of tests make
it less likely that alternative explanations could lead
CHANGE_ROA to a false rejection of the null hypotheses. For exam-
= 0 + 1 RCAR + 2 BHAR + 3 COMP ple, it might be argued that CEOs jump ship and
quit voluntarily when they expect future prospects
+ 4 PAST_ROA + 5 SIZE to deteriorate and are forced to quit when the prior
+ 6 (other news announcement indicators) +  (3) performance is poor. If this is the case, voluntary
departures could signal reversal of prospects in the
PAST_ROA is the industry-adjusted ROA for the year future for better performers, and forced departures
before the year of the CEO’s departure. could indicate that governance has strengthened to
As in the earlier tests, our first null hypothesis cor- force out previously entrenched managers in poorly
responds to the view that prior performance and the performing firms. Although this might explain why
market’s reaction to the CEO’s departure news are prior performance is negatively related to announce-
unrelated to CEO ability. Under this null hypothesis, ment returns and future operating performance, it
the coefficients i , i = 1 2 should be all zero. An alter- would not explain why manager’s labor market
native view is that CEO ability is positively related to progress should be positively related to past pay. Such
prior performance, and negatively related to RCAR. an explanation requires pay to be low when perfor-
Thus, the null hypothesis is rejected against this alter- mance is poor prior to the departure, quite the oppo-
native if 1 is positive and/or 2 is negative. site of a governance-change explanation.
Our second null hypothesis, as in specifications (1)
and (2), concerns the coefficient of COMP. If cross-
sectional variations in COMP are unrelated to CEO
5. Results
ability or power, the coefficient 3 should be zero. 5.1. Univariate Comparisons
A second view is that CEO pay reflects CEO ability. If Table 3 reports the CARs based on the market model
3 is negative, the null hypothesis is rejected in favor for two windows: from day −1 to +1 and from day −1
of this alternative. A third view of CEO relative pay is to +7 of the departure announcement. For the overall
that it is associated with CEO entrenchment or power. sample, the mean and median CARs are significantly
If 3 is positive, the null hypothesis is rejected in favor negative. For the good and moderate progress groups
of this alternative. combined, the mean and median CARs are more neg-
4.3.4. Discussion. Our approach to the question ative. These CARs are also statistically smaller than
of whether CEO ability matters for firm performance, those for the poor and very poor progress groups com-
and whether pay reflects CEO ability, utilizes three sets bined (p-value between 0.01 and 0.07). These results
of tests, as discussed above. Each of our three tests are consistent with the idea that managers that expe-
has a particular advantage. For example, tests of the rience better progress in the managerial labor market
stock market’s reaction to the departure news utilize are perceived to be more valuable and the market asso-
information about the event within a short window. ciates their departures with greater loss of firm value.
Although we do find some evidence of other contem- Departures for the combined poor and very poor
poraneous news, such announcements are likely to progress group are not associated with statistically
make it more difficult for the ability hypothesis to be significant CARs. Earlier studies have recognized that
accepted over the null. The tests related to labor mar- abnormal returns around CEO departure announce-
ket progress have the advantage that the labor mar- ments, especially for poorly performing firms or
ket is likely to generate and utilize more information forced turnovers, can be ambiguous. For example,
about an executive than is observable to a researcher, Warner et al. (1988, p. 466) suggest that “abnormal
or even to the financial markets when the departure stock return at announcement is the sum of two com-
is announced. To that extent, as long as the progress ponents. One is an information component that is
classification is reasonably unambiguous, any relation negative if the change signals worse management per-
between prior performance and pay and labor market formance than anticipated. The second is a real com-
outcomes is strong evidence of the ability hypothesis. ponent that is positive if the change is in shareholders’
Finally, the tests based on financial market’s reaction, interest. A positive net effect is expected only if the
and to a lesser extent those based on labor market out- real component is larger in absolute value than the
comes, rely on market perceptions of ability. The tests information component.”19
on postdeparture performance do not have this limi-
tation. However, because they are necessarily over a 19
See also Denis and Denis (1995) for a similar argument.
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
1644 Management Science 56(10), pp. 1633–1652, © 2010 INFORMS

Table 3 Departure Announcement Effects Table 4 Stock Market Reaction to Managerial Departures

CAR(−1, +7) in % CAR(−1, +1) in % Intercept 0007∗ 0007∗ 0020∗∗


183 184 254
Median Mean Median Mean
(p-value) (p-value) (p-value) (p-value) Three-year buy-and-hold return −0010∗∗ −0010∗∗ −0010∗∗
−264 −264 −246
CEO departures −154 −252 −112 −153
ROA 0033 0033 0020
000 001 001 004
048 048 028
Good and moderate −259 −411 −179 −270
(satisfactory) progress 000 000 000 000 Abnormal_CEO_pay −4.2e−9∗∗
−246
Poor and very poor −066 −080 −008 −060
(unsatisfactory) progress 043 058 055 059 Abnormal_pay_difference −4.2e−9∗∗
CAR for the satisfactory 003 004 001 007 −249
progress < CAR for the Relative_total_pay −0036∗∗
unsatisfactory progress −277
Notes. This table shows the announcement effects of the CEO departures. Firm’s size −1.5e−7∗∗ −1.4e−7∗∗ −3.5e−8∗
The good progress category comprises CEO departures that meet any of the −252 −252 −191
following three conditions: (i) the CEO ends up as the CEO of another public News indicator good performance 0007 0007 0007
company within three years of the departure and either has a new salary that 041 041 050
exceeds his old salary or joins a company that is larger in size (market value
of equity plus book value of debt) than his old company; (ii) the CEO ends up News indicator other good news 0028 0028 0019
as a non-CEO executive of another company within three years of the depar- 060 059 046
ture and has a higher salary than in his old company; or (iii) the CEO joins News indicator poor performance −0090∗ −0090∗ −0086∗
a private company as a CEO within one year. The very poor progress cate- −191 −190 −181
gory includes those CEOs who have not secured a new appointment within
three years of their departure from their previous CEO positions. The poor News indicator other bad news −0013 −0013 −0023
progress category includes those CEOs who have taken up only new director- −029 −028 −049
ships within three years of their departure from their previous CEO positions. News indicator restructuring −0126∗ −0127∗ −0131∗
The moderate progress category includes the remaining departing CEOs who −202 −203 −213
take up a managerial position in another private or public company within
News indicator business expansion −0046∗∗ −0046∗∗ −0043∗∗
three years of the departure and meet none of the three good progress con-
−266 −267 −254
ditions. CAR(−1, +1) and CAR(−1, +7) are the cumulative abnormal return,
based on the market model, for the windows (−1 day, +1 days) and (−1 day, Number of firm years 298 298 298
+7 days) around the announcement date of the CEO departures. R2 0127 0127 0095

5.2. Regression Results Notes. This table reports the results of the three weighted least square
regressions—the estimated coefficients, their respective t statistics (in
5.2.1. Main Tests. parentheses) based on robust standard errors and by-year clustering, R2 ,
5.2.1.1. Stock Market Reaction to Departure News. and number of firm years. The dependent variable is the cumulative abnor-
In Table 4, we present our results for Equation (1). mal return (CAR), based on the market model, for the window (−1 day,
+7 days) around the announcement day of a top managerial departure.
Because extreme CARs can be very noisy, we esti- Three-year buy-and-hold return is the buy-and-hold return for the three-
mate Equation (1) using the weighted least squares year period before the year of the managerial departure. ROA is measured
procedure.20 Results are reported for three regressions over the year before the year of the executive departure. Three-year buy-
using our three different pay measures. The relation and-hold return and ROA are adjusted by subtracting the contemporane-
between past performance (BHAR) and the CAR is ous median measure for all non-turnover firms with the same two-digit
SIC code Abnormal_CEO_pay is the three-year predeparture average of
significantly negative, which rejects the null hypoth- the residual from a regression where the CEO’s total compensation is the
esis that managerial ability is inconsequential for dependent variable, and the explanatory variables include the sum of the
firm performance. This result supports the alternative total compensation of the four other highest-paid executives in the same
hypothesis that the market associates superior stock company, total assets, sales growth, year, and industry indicator variables.
performance with higher CEO ability, and reacts more Abnormal_pay_difference is the three-year predeparture average annual
residual from a regression where the difference between the CEO’s total
negatively when a more capable manager who gener- compensation and the average of the total compensation of the four other
ated higher stock returns leaves a company. highest-paid executives in the same company is the dependent variable, and
The pay measures are all negatively related to the explanatory variables include total assets, sales growth, year, and industry
announcement effects. This supports the hypothe- indicator variables. Relative_total_pay is the ratio of the CEO’s total compen-
sation to the sum of the total compensation of the four other highest-paid
sis that more capable managers are rewarded with
executives in the company over the last three full fiscal years prior to the
higher pay and that their departures are associated CEO’s departure. Firm’s size is the book value of debt plus the market value of
with more negative market reaction. Not only is there equity of the company at the end of year before the year of the CEO’s depar-
no support for the null hypothesis that CEO pay is ture. The six indicator variables for different types of news announcements
are described in Table 2.
∗∗∗ ∗∗
, , and ∗ indicate the 1%, 5%, and 10% levels of significance,
20
We thank an anonymous referee for suggesting the weighted least respectively.
squares procedure for this regression.
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
Management Science 56(10), pp. 1633–1652, © 2010 INFORMS 1645

unrelated to CEO ability, the negative coefficient on inconsequential for firm value and support the man-
COMP is also inconsistent with the “skimming” view agerial ability hypothesis. RCAR has a negative coef-
(Bebchuk and Fried 2004, Bebchuk et al. 2008). How- ficient in all specifications that is significant at least
ever, we find that the relationship of pay to the stock at the 5% level. BHAR has a positive significant coef-
market’s reaction to the departure announcement, the ficient in all regressions (significant at the 5% level
CEO’s labor market progress, and subsequent firm in the probit model in four out of six regressions).
performance, depends on firm governance. Untabu- Although the simple relative pay measure has the
lated tests indicate that relative CEO pay is either right sign, it is not significant at conventional levels.
insignificant or marginally significant in specifica- However, the two other relative pay measures that
tion (1), and insignificant in specifications (2) and (3), are based on regression residuals are significant at the
for a subsample of departures with poor governance 5% level in the ordered logit regression for the overall
characteristics, as measured by the G-Index. However, sample and the 10% level in the other regressions.
our results remain for the complementary subsample The incremental effects of the dependent vari-
of firms with good governance characteristics. ables in Table 5 are economically significant. A one-
The effects of the variables of interest in Equa- standard-deviation decrease in the RCAR increases
tion (1) are economically significant. A one-standard- the probability of a successful labor market outcome
deviation increase in BHAR leads to a 1.4% decrease by about 6% for the overall sample and about 12%
in CAR (from one day before to seven days after the for the young CEO group. A one-standard-deviation
departure announcement). A one-standard-deviation increase in the relative pay increases the probability
increase in our relative pay measure COMP leads to by 7% to 8% when CEO excess relative pay is cal-
a 2.7% decrease in CAR. culated based on the regression residuals, although
As robustness checks, we redo all the tests by the magnitudes for the simple measure of relative
changing the dependent variable to CAR−1 +1, pay are smaller. A one-standard-deviation increase in
FFCAR−1 +1, and FFCAR−1 +7, respectively. BHAR increases the probability of successful labor
The (untabulated) results are similar. The results also market progress by 10% for the overall sample but
hold when we adjust the buy-and-hold return by by as much as 15% for the sample of young CEOs.
subtracting the corresponding return of the industry, Table 5 shows that the sensitivity of CEOs’ labor mar-
size, and book-to-market-matched portfolio. The coef- ket progress to RCAR and BHAR is stronger for the
ficient associated with the industry-adjusted buy-and- sample of young CEOs than for the overall sample.
hold return is virtually identical when we separately Consistent with Table 4, firm size continues to have
include the industrial buy-and-hold return. This sug- a significant positive effect on labor market progress
gests that financial markets mainly “credit” the man- in most specifications. Interestingly, the ordered logit
model in Table 6, for which our progress classification
agers for industry-adjusted performance. Inclusion of
incorporates information on the size of the new firm
CEO death events does not affect our conclusions
and salary of the new position, shows that the size
(untabulated results).
effect is stronger, both economically and statistically,
Turning to other explanatory variables, firm size
for younger CEOs.
has a negative significant effect in all regressions.
We replicate our results by calculating cumula-
Most of the other control variables are insignificant,
tive abnormal returns using the market model over
with the exception of POORperf (our proxy for the dis-
the period from a day before to a day after the
closure of news items concerning poor performance),
departure announcement. The results remain unaf-
which is, unsurprisingly, negative and significant.
fected. Adding the Fama-French factors to calculate
ROA is generally insignificant. After controlling for
abnormal returns (i.e., FFCARs instead of CARs) pro-
stock performance, ROA appears to have no incre-
duces very similar results over both windows.21 These
mental information about managerial ability. This results are not tabulated. We also check the robustness
may reflect the fact that past stock returns may have of our results by adjusting buy-and-hold returns with
already incorporated the relevant information con- respect to industry, size, and book-to-market portfo-
tained in the past accounting performance. lio returns.22 The results are qualitatively unchanged.
5.2.1.2. Labor Market Outcomes. In Tables 5 and 6, To ensure that our results on relative pay and RCAR
we report regression results corresponding to Equa- are not due to potential nonlinearities in the relation-
tion (2). Table 5 reports the marginal effects for a ship between BHAR and career progress, we replicate
one-standard-deviation increase in the explanatory
variable on the probability of satisfactory labor mar- 21
In fact, the coefficients for both measures of residual pay become
ket progress from the probit model, whereas Table 6 more significant (at 5%) when we use FFCARs or the (−1, +1) win-
reports the coefficient estimates from the ordered logit dow when we use the two-way classification.
regression based on our four-way classification. Both 22
For this BHAR measure, relative total pay becomes more signifi-
models reject the null hypothesis that CEO ability is cant (at 10%) for the two-way progress classification.
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
1646 Management Science 56(10), pp. 1633–1652, © 2010 INFORMS

Table 5 CEO Career Progress (Incremental Effects)

Overall sample Young CEO sample


∗∗∗ ∗∗∗ ∗∗∗ ∗∗
Residual CAR−1 +7 −0059 −0060 −0051 −0117 −0117∗∗ −0122∗∗
−265 −267 −269 −234 −234 −248
Three-year buy-and-hold return 0103∗∗ 0102∗∗ 0102∗ 0147∗∗ 0148∗∗ 0134∗
192 195 182 203 203 182
ROA −0034 −0035 0036 0003 0003 0007
−123 −127 −121 006 006 017
Abnormal_CEO_pay 0077∗ 0069∗
186 188
Abnormal_pay_difference 0086∗ 0069∗
194 188
Relative_total_pay 0038# 0012
136 041
CEO age −0091∗∗∗ −0087∗∗∗ −0092∗∗∗ −0107∗∗∗ −0107∗∗∗ −0117∗∗∗
−383 −366 −407 −391 −392 −418
Firm’s size 0083∗∗∗ 0084∗∗∗ 0059# 0041 0041 0061∗
298 316 151 108 107 169
Number of firm years 298 298 298 167 167 167
Pseudo R2 0.101 0.098 0.088 0.137 0.137 0.135

Notes. This table reports the results of the two probit regressions: the marginal effect corresponding to an increase in the predicting
variable by one standard deviation, their respective z statistics (in parentheses) based on robust standard errors and by-year clustering,
pseudo R2 , and number of firm years. Young CEO sample consists of those cases in which the age of the departing CEO is at or below
the median age. The dependent variable is progress that takes a value of 1 if the departing CEO obtains a new managerial position
within three years of his departure, and 0 otherwise. Residual CAR−1 +7 is the residual cumulative abnormal return, based on the
market model, for the window (−1 day, +7 days) around the announcement day of the CEO departures from the regression in Table 4.
Three-year buy-and-hold return is the buy-and-hold return for the three-year period before the year of the managerial departure. ROA is
measured over the year before the year of the executive departure. Three-year buy-and-hold return andROA are adjusted by subtracting
the contemporaneous median measure for all nonturnover firms with the same two-digit SIC code. Abnormal_CEO_pay is the three-
year predeparture average of the residual from a regression where the CEO’s total compensation is the dependent variable, and the
explanatory variables include the sum of the total compensation of the four other highest-paid executives in the same company, total
assets, sales growth, year, and industry indicator variables. Abnormal_pay_difference is the three-year pre-CEO-departure average
residual from a regression where the difference between the CEO’s total compensation and the average total compensation of the four
other highest-paid executives in the same company is the dependent variable, and the explanatory variables include total assets, sales
growth, year, and industry indicator variables. Relative_total_pay is the ratio of the CEO’s total compensation to the sum of the total
compensation of the four other highest-paid executives in the company over the last three full fiscal years prior to the CEO’s departure.
CEO age is the age of the departing CEO. Firm’s size is the book value of debt plus the market value of equity of the company at the end
of year before the year of the CEO’s departure. The regressions include six indicator variables for other types of news announcements
(described in Table 2) over the same window as that of CAR−1 +7, but their coefficients are not reported here for brevity.
∗∗∗ ∗∗
, , and ∗ indicate the 1%, 5%, and 10% levels of significance, respectively; # indicates significance at the 10% level by a
one-sided test.

our results in Tables 5 and 6 by incorporating higher related to the performance change for both samples,
order terms of BHAR. The results (which are not tab- and the effect is significant at the 1% level in the first
ulated) are mainly unchanged. Our results in Table 6 year. BHAR is negatively related to the performance
are also robust to alternative definitions of good and change, and the effect is significant at the 10% level
moderate progress, and to using CEO age of 58 and in the third year for the overall sample, and at the 1%
55 as alternative cutoffs instead of age 61. level in the first year and 5% level in the third year for
5.2.1.3. Postdeparture Firm Performance. Table 7 the sample of satisfactory progress. CEO pay relative
reports results corresponding to Equation (3). Our to the pay of other top executives, as measured by
results are consistent with the notion that departures abnormal_CEO_pay, has a negative effect, significant at
of more capable CEOs are associated with worse sub- the 5% level in the first year for the overall sample
sequent operating performance. We present results and the 10% level in the third year for the sample of
for both the overall sample and the sample of sat- satisfactory progress. Results for the other two mea-
isfactory labor market progress (good and moderate sures of COMP are similar but are not reported in
progress). The latter sample is useful to study sepa- the table. Untabulated tests indicate that our results
rately because prior studies have mostly documented remain whether or not we include a sample of CEO
performance improvements around involuntary CEO death events in the analysis. The results are econom-
changes. Table 7 shows that RCAR is positively ically significant. A one-standard-deviation change in
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
Management Science 56(10), pp. 1633–1652, © 2010 INFORMS 1647

Table 6 CEO Career Progress with a Finer Progress Classification (Ordered Logit)

Overall sample Young CEO sample


∗∗ ∗∗ ∗∗ ∗∗
Residual CAR−1 +7 −1441 −1441 −1220 −2650 −2652∗∗ −2725∗∗∗
−220 −220 −256 −258 −258 −262
Three-year buy-and-hold return 0162∗ 0162∗ 0163∗ 0169∗ 0169∗ 0158#
173 173 165 165 165 156
ROA −0800 −0802 −0758 0343 0343 0421
−089 −089 −078 027 027 033
Abnormal_CEO_pay 47e−8∗∗ 51e−8∗
207 179
Abnormal_pay_difference 48e−8∗∗ 51e−8∗
206 180
Relative_total_pay 0302 0273
074 048
CEO age −0076∗∗∗ −0076∗∗∗ −0078∗∗∗ −0096∗∗∗ −0096∗∗∗ −0102∗∗∗
−466 −465 −491 −393 −393 −497
Firm’s size 29e−6∗∗∗ 28e−6∗∗∗ 14e−6∗∗∗ 41e−6∗∗∗ 41e−6∗∗∗ 57e−6∗∗∗
287 289 307 331 329 537
Number of firm years 298 298 298 167 167 167
Pseudo R2 0.056 0.056 0.053 0.071 0.071 0.070

Notes. This table reports the results of the two ordered logistic regressions: the estimated coefficients, their respective z statistics (in parentheses) based
on robust standard errors and by-year clustering, pseudo R2 , and number of firm years. Young CEO sample consists of those cases in which the age of the
departing CEO is at or below the median age. The dependent variable is progress. The good progress category (progress = 4) comprises CEO departures that
meet any of the following three conditions: (i) the CEO ends up as the CEO of another public company within three years of the departure and either has a new
salary that exceeds his old salary or joins a company that is larger in size (market value of equity plus book value of debt) than his old company; (ii) the CEO
ends up as a non-CEO executive of another company within three years of the departure and has a higher salary than in his old company; or (iii) the CEO joins
a private company as a CEO within one year. The very poor progress category (progress = 1) includes those CEOs who have not secured a new appointment
within three years of their departure from their previous CEO positions. The poor progress category (progress = 2) includes those CEOs who have taken up
only new directorships within three years of their departure from their previous CEO positions. The moderate progress category (progress = 3) includes the
remaining departing CEOs who take up a managerial position in another private or public company within three years of the departure and meet none of the
three good progress conditions. Residual CAR−1 +7 is the residual cumulative abnormal return, based on the market model, for the window (−1 day,
+7 days) around the announcement day of the CEO departures from the regression in Table 4. Three-year buy-and-hold return is the buy-and-hold return
for the three-year period before the year of the managerial departure. ROA is measured over the year before the year of the executive departure. Three-year
buy-and-hold return and ROA are adjusted by subtracting the contemporaneous median measure for all nonturnover firms with the same two-digit SIC code.
Abnormal_CEO_pay is the three-year predeparture average of the residual from a regression where the CEO’s total compensation is the dependent variable, and
the explanatory variables include the sum of the total compensation of the four other highest-paid executives in the same company, total assets, sales growth,
year, and industry indicator variables. Abnormal_pay_difference is the three-year pre-CEO-departure average residual from a regression where the difference
between the CEO’s total compensation and the average total compensation of the four other highest-paid executives in the same company is the dependent
variable, and the explanatory variables include total assets, sales growth, year, and industry indicator variables. Relative_total_pay is the ratio of the CEO’s total
compensation to the sum of the total compensation of the four other highest-paid executives in the company over the last three full fiscal years prior to the
CEO’s departure. CEO age is the age of the departing CEO. Firm’s size is the book value of debt plus the market value of equity of the company at the end of
year before the year of the CEO’s departure. The regressions include six indicator variables for other types of news announcements (described in Table 2) over
the same window as that of CAR−1 +7, but their coefficients are not reported here for brevity.
∗∗∗ ∗∗
, , and ∗ indicate the 1%, 5%, and 10% levels of significance, respectively; # indicates significance at the 10% level by a one-sided test.

BHARs, COMP, and RCARs is associated with an abso- after the departure of the CEO and conclude “the pos-
lute change in the industry-adjusted return on assets, sibility of a profitable trading rule (ignoring transac-
from one year before to one year after the departure, tions costs) from going short at announcement cannot
by 0.5%, 0.6%, and 0.8%, respectively. Given that the be dismissed.” If indeed the financial markets under-
mean and median industry-adjusted return on assets react to the news of the CEO’s departure, we could
one year prior to the year of the departure are 2% and expect the stock price performance over the next one
0.6%, respectively, these magnitudes are substantial. to three years to be also related to RCAR, BHAR, and
5.2.2. Additional Tests COMP in the same way that the change in operat-
5.2.2.1. Postdeparture Stock Returns. Although Ta- ing performance is. However, we caveat that empir-
ble 4 shows that the stock market reacts in a direction ical analysis based on long-term stock performance
consistent with the effect on firm profitability (as evi- is challenging for econometric (e.g., dealing with the
denced in Table 7) to the news of the CEO’s depar- lack of independence between observations) and eco-
ture, it is possible that it still underreacts. For exam- nomic reasons (e.g., identifying the right performance
ple, Warner et al. (1988, p. 482) find a delayed reaction benchmark).
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
1648 Management Science 56(10), pp. 1633–1652, © 2010 INFORMS

Table 7 Postdeparture Change in ROA of the Old Employers over the same period.23 We report results from a boot-
strap procedure in which we resample with replace-
Good and moderate
Overall sample progress sample ment 50,000 times with by-year clustering.
The results in Table 8 are generally consistent with
Dependent variable: ROA1 ROA3 ROA1 ROA3 our hypothesis that RCAR, BHAR, and COMP are
Residual CAR−1 +7 0060 ∗∗∗
0044 #
0125 ∗∗∗
0043# associated with CEO ability differences. RCAR has a
356 158 427 146 positive coefficient that is statistically significant at
Three-year buy-and-hold −0003 −0005∗ −0006∗∗∗ −0005∗∗ the 1% level for the overall sample and from the
return −098 −186 −549 −206 1% to 10% level for the good and moderate progress
Abnormal_CEO_pay −8.0e−10∗∗ −7.3e−11 −9.5e−10# −4.7e−10∗ group. BHAR has a significant negative effect on stock
−242 −021 −164 −181
returns over the two- and three-year horizon. How-
Past ROA −0246∗∗∗ −0107∗ −0289∗∗∗ −0091∗∗
−678 −213 −414 −215
ever, it has a positive significant coefficient on returns
Firm’s size −3.8e−8∗∗ 6.1e−9 −6.0e−8∗∗ −2.9e−8 in the first year after the departure for both sam-
−257 −033 −243 −150 ples. This could be due to market underreaction or
Number of firm years 269 190 126 89 delayed reaction to good news associated with firms
R2 0322 0199 0456 0341 that have good prior performance, similar to that
Notes. This table reports the results of the four regressions: the estimated
found in the momentum literature (see Jegadeesh and
coefficients, their respective t statistics (in parentheses) based on robust Titman 1993). However, after the first year, the effect
standard errors and by-year clustering, R2 , and number of firm years. The of BHAR becomes negative. Finally, relative pay has a
dependent variable is the annualized change in industry-adjusted ROA from negative sign and is significant at the 5% level or less
the year before to the first or the third year after the year of the managerial in three of the six regressions.
departure (ROA1 or ROA3). ROA is adjusted by subtracting the contem-
poraneous median measure for all non-CEO-turnover firms with the same
5.2.2.2. Appointment News CARs and Postappoint-
two-digit SIC code. Good and moderate progress sample consists of those ment Operating Performance. The sample of departures
cases in which the departing CEO obtains a new managerial position within that results in new managerial appointments in pub-
three years of his CEO departure. Residual CAR−1 +7 is the residual licly traded firms provides us with a small subsam-
cumulative abnormal return, based on the market model, for the window ple on which we can implement some additional tests.
(−1 day, +7 days) around the announcement day of the CEO departures
from the regression in Table 4. Three-year buy-and-hold return is the buy-
This sample is even smaller than the good and mod-
and-hold return for the three-year period before the year of the managerial erate progress subsample in Table 7 because many of
departure. Past ROA is measured over the year before the year of the exec- the departures in the latter sample do not result in
utive departure. Three-year buy-and-hold return and Past ROA are adjusted new appointments in publicly traded firms. If the abil-
by subtracting the contemporaneous median measure for all nonturnover ity hypothesis is valid, a higher ability CEO should be
firms with the same two-digit SIC code. Abnormal_CEO_pay is the three-
year predeparture average of the residual from a regression where the CEO’s
associated with a more negative departure CAR at the
total compensation is the dependent variable, and the explanatory variables old firm and a more positive appointment CAR at the
include the sum of the total compensation of the four other highest-paid new firm. Thus, we first examine the correlation of
executives in the same company, total assets, sales growth, year, and indus- the departure and appointment CARs. Consistent with
try indicator variables. Firm’s size is the book value of debt plus the market Hayes and Schaefer (1999), panel A of Table 9 con-
value of equity of the company at the end of year before the year of the CEO’s
departure. The regressions include six indicator variables for other types of
firms that this is the case. Four sets of correlations are
news announcements (described in Table 2) over the same window as that reported, corresponding to the (−1 +1) and (−1 +7)
of CAR−1 +7, but their coefficients are not reported here for brevity. announcement windows. All four correlations are neg-
∗∗∗ ∗∗
, , and ∗ indicate the 1%, 5%, and 10% levels of significance, respec- ative, and three out of the four are significant at a 10%
tively; # indicates significance at the 10% level by a one-sided test. level or less. These results are inconsistent with the
jump ship explanation discussed at the end of §4.
Table 8 reports the results of the following regres-
Next, in specifications similar to Equations (1)
sion:
and (3), we examine whether the appointment CAR
and operating performance of the new firm that the
BHR+1 +T  = 0 + 1 RCAR + 2 BHAR + 3 COMP
CEO joins are related to the CEO’s relative pay in the
+ 4 BENCH_BHR+1 +T  + 5 SIZE previous company, the prior performance of that firm
+ 6 (other news announcement
23
The buy-and-hold return of the benchmark portfolio is con-
indicators) +  (4) structed as follows. One trading day before the event day, we group
all stocks into three groups based on market capitalization (size)
The dependent variable is buy-and-hold stock returns and the market-to-book of equity within each two-digit SIC indus-
of the company losing the CEO from one week to try (so in each two-digit SIC industry, we have nine groups). We
then identify the group to which the event firm belongs, and we
T weeks after the departure, where T = 53, 105, or 157. use the remaining nonevent firms in the same group to form the
The new explanatory variable BENCH_BHR+1 +T  is benchmark portfolio. We skip one week and then compute the one-,
the buy-and-hold return on a benchmark portfolio two-, or three-year buy-and-hold return for that portfolio.
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
Management Science 56(10), pp. 1633–1652, © 2010 INFORMS 1649

Table 8 Postdeparture Stock Return of the Old Employers

Overall sample Good and moderate progress sample

One-year Two-year Three-year One-year Two-year Three-year


Dependent variable: return return return return return return

Residual CAR−1 +7 1937∗∗∗ 1787∗∗∗ 1502∗∗∗ 1872∗∗∗ 1122∗∗ 1593∗


594 374 339 409 229 191
Three-year buy-and-hold return 0163∗ −0134∗∗∗ −0145∗∗∗ 0396∗∗ −0070∗ −0135∗
166 −334 −275 242 −180 −179
Abnormal_CEO_pay −1.3e−8# −4.9e−8∗∗∗ −3.5e−8∗∗ −1.4e−8 −4.9e−8∗∗∗ −5.1e−8#
−156 −341 −208 −061 −301 −141
Benchmark portfolio return 0859∗∗∗ 0399∗∗∗ 0271∗ 0390 0010 00263
448 278 182 116 005 094
Firm’s size −8.6e−7 −1.8e−6∗∗ −1.5e−6# −1.3e−6 −2.2e−6 −2.4e−6
−123 −205 −154 −044 −100 −073
Number of firm years 264 222 155 100 98 89

Notes. This table reports the results—the estimated coefficients and their respective z statistics (in parentheses)—of 50,000-repeat bootstrap regressions
based on resampling with replacement and by-year clustering from the actual sample. The dependent variable is the one-, two-, or three-year (skipping the
immediate week) buy-and-hold stock return of the firm losing the CEO after the managerial departure. Residual CAR−1 +7 is the residual cumulative
abnormal return, based on the market model, for the window (−1 day, +7 days) around the announcement day of the CEO departures from the regression
in Table 4. Three-year buy-and-hold return is the industry-adjusted buy-and-hold return for the three-year period before the year of the managerial departure.
Abnormal_CEO_pay is the three-year predeparture average of the residual from a regression where the CEO’s total compensation is the dependent variable, and
the explanatory variables include the sum of the total compensation of the four other highest-paid executives in the same company, total assets, sales growth,
year, and industry indicator variables. Benchmark portfolio return is computed as follows. One trading day before the event day, we group all stocks into three
groups independently by market capitalization (size) and market-to-book of equity within each two-digit SIC industry. So in each two-digit SIC industry, we
have nine groups. We then identify the group to which the event firm belongs, and use the remaining nonevent firms in the same group to form a portfolio.
We skip one week, and then compute the one-, two-, or three-year buy-and-hold return for that portfolio. Firm’s size is the book value of debt plus the market
value of equity of the company at the end of year before the year of the CEO’s departure. The regressions include six indicator variables for other types of news
announcements (described in Table 2) over the same window as that of CAR−1 +7, but their coefficients are not reported here for brevity.
∗∗∗ ∗∗
, , and ∗ indicate the 1%, 5%, and 10% levels of significance, respectively; # indicates significance at the 10% level by a one-sided test.

under the CEO’s management, and the stock mar- change in industry-adjusted ROA (ROA) from one
ket’s reaction to the departure news. For the tests, year before to three years after the appointment.
we include the explanatory variables of interest one If the annualized change cannot be calculated over
at a time, because we have severe sample size attri- the three-year period because of data unavailability,
tion due to the cumulative effects of missing obser-
we use data for the first two years, and if that is
vations from all explanatory variables when these are
also not available, we only use data for the year after
all included. In the first three columns in panel B of
Table 9, the dependent variable is the appointment the appointment. Industry adjustment is done by sub-
CAR over the (−1, +1) window.24 Consistent with tracting the contemporaneous median ROA for all
panel A in Table 9, there is a significant negative nonturnover firms with the same two-digit SIC code.
association of the departure CAR over the (−1, +7) A more negative departure CAR is significantly asso-
window and the appointment CAR. The results for ciated with greater operating performance improve-
relative pay are inconsistent with the “rent extraction” ment in the new firm. Higher CEO relative pay at
hypothesis. Although relative pay does not have any the old firm has a significant positive association with
significant association at conventional levels, the sign
performance improvement in the new firm. Although
of the coefficient is consistent with the ability hypoth-
esis. The economic magnitude is comparable to that better prior performance is also associated with per-
for the same variable in Table 4.25 Prior performance formance improvement, the effect is not statistically
has a significant positive association, consistent with significant at conventional levels. Note, however, that
the ability hypothesis. the economic magnitude of the effect is much larger
In the last three columns of panel B of Table 9, than those reported in Table 7 on the change in oper-
the dependent variable is the annualized average ating performance of the old firm.
Overall, these results support the ability hypothesis
24
We focus on the (−1, +1) window because the correlations in panel and, in addition, further rule out the possibility that
A of Table 9 suggest that the signal-to-noise ratio in the market reac-
tion to appointment news may be the highest over this window. our main results could be due to alternative explana-
25
We need to be cautious in making this comparison, because the tions unrelated to CEO ability, such as the jump ship
specification reported in Table 4 includes BHAR. explanation discussed in §4.
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
1650 Management Science 56(10), pp. 1633–1652, © 2010 INFORMS

Table 9 Cumulative Abnormal Returns Around Departures and Appointments and Firm Performance of the New Employers

Panel A: Correlation between departure CAR and appointment CAR

Appointment CAR

CAR−1 +1 CAR−1 +7

Departure CAR−1 +1 −0369∗∗∗ −0188∗


00005 00874
Departure CAR−1 +7 −0221∗∗ −0048
0042 06679

Panel B: Appointment CAR and postappointment change in ROA of the new firm

Appointment Appointment Appointment


Dependent variable: CAR−1 +1 CAR−1 +1 CAR−1 +1 ROA ROA ROA
∗ ∗∗
Departure CAR−1 +7 −0143 −0062
179 −214
Appointment CAR−1 +1 0039
037
Three-year buy-and-hold return 0032∗∗∗ 0048
381 127
Abnormal_CEO_pay 3.8e−9 2.8e−9∗∗
110 202
Past ROA −0333∗ −0316∗ −0270
−177 −186 −165
Firm’s size −3.2e−7∗∗∗ −1.5e−7 −3.4e−7∗∗ 2.5e−7∗ 3.1e−7# 1.1e−7
−339 −099 −257 167 132 107
R-squared 011 014 003 050 052 043
Number of firm years 75 57 71 59 56 67

Notes. Panel A reports the correlation coefficients between the departure CARs and the appointment CARs and the corresponding p-values (in parentheses).
The departure CAR−1 +1 and CAR−1 +7 are the cumulative abnormal return, based on the market model, for the windows (−1 day, +1 days) and
(−1 day, +7 days) around the announcement of the CEO departures. The appointment CAR−1 +1 and CAR−1 +7 are the cumulative abnormal return,
based on the market model, for the windows (−1 day, +1 days) and (−1 day, +7 days) around the announcement of the new appointment of the departing
CEO. Panel B reports results of six regressions: the estimated coefficients, their respective t statistics (in parentheses) based on robust standard errors and
by-year clustering, R2 , and number of firm years. The dependent variable is the appointment CAR−1 +1 for the first three regressions and the annualized
change in industry-adjusted ROA from the year before to the first, second, or third year after the managerial appointment (ROA) for the next three regressions.
Industry adjustment is done by subtracting the contemporaneous median ROA for all nonturnover firms with the same two-digit SIC code. Three-year buy-
and-hold return is the buy-and-hold return of the old employer for the three-year period before the year of the managerial departure. Past ROA is the ROA of
the new employer over the fiscal year before the new executive appointment. Three-year buy-and-hold return and Past ROA are adjusted by subtracting the
contemporaneous median measure for all nonturnover firms with the same two-digit SIC code. Abnormal_CEO_pay is the three-year predeparture average
of the residual from a regression where the CEO’s total compensation is the dependent variable, and the explanatory variables include the sum of the total
compensation of the four other highest-paid executives in the same company, total assets, sales growth, year, and industry indicator variables. Firm’s size is
the book value of debt plus the market value of equity of the new company at the end of year before the year of the appointment.
∗∗∗ ∗∗
, , and ∗ indicate 1%, 5%, and 10% levels of significance, respectively; # indicates significance at the 10% level by a one-sided test.

5.2.2.3. Additional Untabulated Robustness Checks.26 nine cases in which departing CEOs are from non-
In our main sample, we include cases in which the ExecuComp firms. The results from our main regres-
incoming CEO is previously the CEO of a non- sion analyses stand.
ExecuComp company. Therefore, our sample may Bebchuk et al. (2008) find lower relative pay for
have a higher proportion of “good performance” CEO founders and higher relative pay for CEOs with
longer tenure. Therefore, our results regarding rela-
departures specifically from smaller firms relative to
tive pay may depend on whether the founder status
the population of CEO departures in small firms. This
and tenure are taken into consideration. As a robust-
may introduce a negative correlation between firm ness check, we control for the founder status and
size and performance in the labor market. To mit- tenure and repeat all our major analyses. None of our
igate this issue, we control for firm size in all our conclusions is affected.
regressions. As a further robustness test, we exclude The stock market reacts to the CEO departure
announcement based on the expected difference in
ability between the departing and the incoming CEO.
26
We thank an anonymous referee for suggesting these checks. Therefore, the announcement of a specific replacement
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
Management Science 56(10), pp. 1633–1652, © 2010 INFORMS 1651

for the CEO may make a difference, particularly for Bebchuk, L. A., M. Cremers, U. Peyer. 2008. CEO centrality. Har-
the analyses of the departure announcement effects vard Law and Economics Discussion Paper 601, Harvard Uni-
versity, Cambridge, MA.
and the post departure operating performance of
Bebchuk, L. A., J. M. Fried, D. Walker. 2002. Managerial power and
firms. We control for whether or not the replacement is
rent extraction in the design of executive compensation. Univ.
announced when the CEO departure is announced by Chicago Law Rev. 69(3) 751–846.
means of an indicator variable and rerun the regres- Bennedsen, M., F. Pérez-González, D. Wolfenzon. 2009. Do CEOs
sions. Our results are qualitatively similar. matter? Working paper, Copenhagen Business School, Fred-
eriksberg, Denmark.
Bertrand, M., A. Schoar. 2003. Managing with style: The effect
6. Conclusion of managers on firm policies. Quart. J. Econom. 118(4)
The role of the CEO in the firm is a controversial topic 1169–1208.
that has become even more contentious in the context Brickley, J. A., J. S. Linck, J. L. Coles. 1999. What happens to
of the recent financial crisis, where top management CEOs after they retire? New evidence on career concerns, hori-
zon problems, and CEO incentives. J. Financial Econom. 52(3)
pay has been at the center of academic and policy 341–377.
debates. Despite its extreme importance, there is sur-
Collingwood, H. 2009. Do CEOs matter? Atlantic Magazine, Online Edi-
prisingly little direct evidence whether CEOs (or more tion (June), http://www.theatlantic.com/magazine/archive/
generally, top executives) contribute positively to firm 2009/06/do-ceos-matter/7437/.
value. In this paper, we test the proposition that CEO Denis, D. J., D. K. Denis. 1995. Performance changes following top
ability matters. In addition, if ability difference exists management dismissals. J. Finance 50(4) 1029–1057.
across CEOs, we examine whether these differences Edmans, A., X. Gabaix. 2009. Is CEO pay really inefficient? A sur-
vey of new optimal contracting theories. Eur. Financial Manage-
are reflected in CEO pay. To do so, we study a sample
ment 15(3) 486–496.
of CEO departures and examine how the past per-
Edmans, A., X. Gabaix, A. Landier. 2009. A multiplicative model of
formance of the firm under the CEO’s management optimal CEO incentives in market equilibrium. Rev. Financial
and the pay of the CEO relative to that of the other Stud. 22(12) 4881–4917.
highest-paid executives of the company are related Fama, E. F., K. R. French. 1993. Common risk factors in the returns
to (i) the stock market’s reaction to the departure on stocks and bonds. J. Financial Econom. 33(1) 3–56.
news, (ii) the subsequent success of the departing Fahlenbrach, R., B. A. Minton, C. H. Pan. 2010. Former CEO direc-
CEO in the managerial labor market, and (iii) the per- tors: Lingering CEOs or valuable resources? Swiss Finance
Institute Research Paper Series 10-11, University of Geneva,
formance of the firm after the CEO leaves. Our results Geneva.
are consistent with the view that cross-sectional dif- Fee, C. E., C. J. Hadlock. 2003. Raids, rewards, and reputations
ferences in firm value and performance are related to in the market for managerial talent. Rev. Financial Stud. 16(4)
ability differences across CEOs. We also find evidence 1315–1357.
against the dominance of the “rent extraction” view Fee, C. E., C. J. Hadlock. 2004. Management turnover across the
within our sample of CEO departures. corporate hierarchy. J. Accounting Econom. 37(1) 3–38.
Forbes. 2001. Forbes executive pay. Online Edition, http://www
.forbes.com/ lists/ home.jhtml?passListId=12&passYear=2001&
Acknowledgments passListType=Person.
The authors are especially grateful to Alex Edmans and
Gabaix, X., A. Landier. 2008. Why has CEO pay increased so much?
Rik Sen for detailed suggestions that significantly improved Quart. J. Econom. 123(1) 49–100.
the paper. The authors also thank Rajesh Aggarwal, Radha
Gompers, P., J. Ishii, A. Metrick. 2003. Corporate governance and
Gopalan, Michael Hertzel, Michael Lemmon, Clive Lennox, equity prices. Quart. J. Econom. 118(1) 107–155.
David Matsa, the two referees, David Yermack, Wei Xiong
Hayes, R. M., S. Schaefer. 1999. How much are differences in man-
(the department editor), and seminar participants at the agerial ability worth? J. Accounting Econom. 27(2) 125–148.
Summer Research Conference in Finance of the Indian
Hotchkiss, E. S. 1995. Postbankruptcy performance and manage-
School of Business (2007) and the Asian Finance Associ- ment turnover. J. Finance 50(1) 3–21.
ation meetings in Yokohama (2008) for helpful comments
Huson, M. R., P. H. Malatesta, R. Parrino. 2004. Managerial succes-
and discussions. All errors are the authors’ responsibility. sion and firm performance. J. Financial Econom. 74(2) 237–275.
Gilles Hilary’s current affiliation is Department of Account- Jegadeesh, N., S. Titman. 1993. Returns to buying winners and sell-
ing, INSEAD, 77000 Fontainebleau, France. ing losers: Implications for stock market efficiency. J. Finance
48(1) 65–91.
Johnson, W. B., R. P. Magee, N. J. Nagarajan, H. A. Newman.
References 1985. An analysis of the stock price reaction to sudden exec-
Bebchuk, L. A., J. M. Fried. 2004. Pay Without Performance: The utive deaths: Implications for the managerial labor market.
Unfulfilled Promise of Executive Compensation. Harvard Univer- J. Accounting Econom. 7(1–3) 151–174.
sity Press, Cambridge, MA. Khanna, N., A. B. Poulsen. 1995. Managers of financially distressed
Bebchuk, L. A., A. Cohen, A. Ferrell. 2009. What matters in corpo- firms: Villains or scapegoats? J. Finance 50(3) 919–940.
rate governance. Rev. Financial Stud. 22(2) 783–827. Lieberman, M. B., L. J. Lau, M. D. Williams. 1990. Firm-level pro-
Bebchuk, L. A., M. Cremers, U. Peyer. 2007. Pay distribution in the ductivity and management influence: A comparison of U.S.
top executive team. Harvard Law and Economics Discussion and Japanese automobile producers. Management Sci. 36(10)
Paper 574, Harvard University, Cambridge, MA. 1193–1215.
Chang, Dasgupta, and Hilary: CEO Ability, Pay, and Firm Performance
1652 Management Science 56(10), pp. 1633–1652, © 2010 INFORMS

Nguyen, B. D., K. M. Nielsen. 2010. What death can tell: Are execu- Shleifer, A., R. W. Vishny. 1989. Management entrenchment: The
tives paid for their contributions to firm value? Working paper, case of manager-specific investments. J. Financial Econom. 25(1)
Chinese University of Hong Kong, Shatin. 123–139.
Taylor, L. A. 2010. CEO pay and CEO power: Evidence from a
Reinganum, M. R. 1985. The effect of executive succession on stock-
dynamic learning model. Working paper, The Wharton School,
holder wealth. Admin. Sci. Quart. 30(1) 46–60. University of Pennsylvania, Philadelphia.
Rosen, S. 1982. Authority, control, and the distribution of earnings. Warner, J. B., R. L. Watts, K. H. Wruck. 1988. Stock prices and top
Bell J. Econom. 13(2) 311–323. management changes. J. Financial Econom. 20(1–2) 461–492.