Evaluating executive compensation packages.
Jarque, Arantxa ; Muth, John
Executive compensation is a topic that has received attention both
in the media and the academic literature. This article discusses issues
relevant to the construction and interpretation of compensation figures
typically reported in both sources. First, it is not clear what
precisely should be included within a measure of the chief executive
officer's (CEO's) income tied to his firm. Second, the study
of executive compensation remains constrained by the availability of
data. We discuss the main source of data used in most studies on the
topic: Execucomp. We highlight where the lack of data requires a
deviation between a theoretical "ideal" measure of
compensation and that which the researcher must use as an approximation.
In this way, we hope our article will be a useful first introduction for
those looking to do further research on the topic.
We propose a measure of realized annual pay, compare it to other
measures used in the literature, and illustrate the difficulties in
calculating it. Using data in Execucomp, we provide our pay measure for
CEOs of large U.S. firms in the period 1993-2012 and use it to estimate
sensitivity of pay to firm performance. The main diff culties in this
exercise lie in the fact that compensation packages of most executives
include stock and option grants on their own firm's shares, which
typically come with requirements that they be held by the executive for
at least three or four years. (1) This implies two important things.
First, the compensation figures that are reported by firms (and are
readily available to the press and researchers) are a combination of
both expected value of compensation (for deferred compensation in the
form of restricted stock and option grants that are not convertible into
cash right away) and realized value (salaries, bonus payments, and
perks). Second, a given year's compensation package provides income
for several years to follow, since the CEO will be able to realize gains
from selling and exercising stock and option grants once their vesting
restrictions expire. That is, an important part of the annual realized
pay of a CEO in any given year comes from his net gains from trading
stock that he received in a past grant. Due to the fact that stock price
realizations may differ from ex-ante expectations of those prices, the
ex-post realized gains from those trades will typically differ from the
valuation made at the time of the grant.
A measure of what is sometimes called direct compensation (the sum
of salary, bonus, other compensation such as pension plans or perks, and
the value of new stock and option grants during the year) is readily
available in Execucomp (variable TDC1). (2) As we just discussed, grants
included in this measure are valued in expectation. Our objective in
this article is to provide a measure of realized pay instead. We define
realized pay as the sum of salaries, bonuses, and other compensation,
plus the gains from trades that the CEO realizes in a given year. We
will argue that this measure is close to the one first proposed by Antle
and Smith (1985) and used later by important contributions such as Hall
and Liebman (1998) and Gayle and Miller (2009). Total yearly
compensation is defined in these studies as the change in the wealth of
the CEO that is tied to his employment in the firm, and it is calculated
in practice as direct compensation plus the year-on-year change in the
market value of stock and option holdings of the CEO from past grants.
This measure is, hence, still a measure of expected pay, although more
sophisticated than TDC1. The main departure of our measure of realized
pay with respect to this total yearly compensation is that it does not
attribute changes in the value of grants that are not yet exercised to
the realized pay in the year when they occur; rather, the final realized
value is captured in gains from trades and attributed to the period of
exercise of the grants. This simplification is useful in terms of the
calculation of the measure--we need to rely less heavily on assumptions
about the unavailable details of grants.
[FIGURE 1 OMITTED]
Still, only part of the information that we need for our measure
(about trades or vesting restrictions and exercise prices of past
grants) is available in Execucomp. When approximating the gains from
trades, in particular, we follow closely the algorithm used in Clementi
and Cooley (2009) to recover the executive's holdings of stocks and
options of his firm. (3) In the Appendix, we walk the reader through the
step-by-step construction of the portfolio, discussing the shortcomings
of the available data in Execucomp and how different assumptions about
the unknowns may affect the compensation numbers.
We use our measure of realized pay to provide an updated account of
CEO compensation through the year 2012. Figure 1 presents a comparison
of our measure of realized pay versus two measures of expected pay used
in the literature: "direct compensation," the variable TDC1 in
Execucomp, and "total yearly compensation," as calculated by
us following the implementation in Clementi and Cooley (2009) of the
concept introduced by Antle and Smith (1985). Median realized pay is
mostly below median direct compensation. The main difference observable
with total yearly compensation is that it is a lot more variable than
either of the other two measures. This figure suggests that different
measures of pay present different pictures of CEO compensation, and it
is important to understand what is behind the measurements before using
them to evaluate pay practices.
We use our realized pay measure to perform a sensitivity analysis
of annual realized pay to performance, with a special focus on the
finance sector throughout the recent crisis in 2008. We simplify some of
the difficulties of the analysis by assuming that the choice of selling
and buying stock is invariant to the stock price movements in our
counterfactual exercises; i.e., only the profits from the trades change,
not the quantities. We find that in the aftermath of the crisis the
realized pay of CEOs of finance firms has decreased in level relative to
other industries. Moreover, the sensitivity exercise suggests that,
during the whole sample period, mean realized pay for CEOs in finance
firms changes with the performance of the firm in similar magnitudes
than that of the average CEO.
We proceed as follows. In Section 1, we introduce compensation
instruments included in most CEO pay packages and discuss data
availability and measurement challenges. In Section 2, we present a
simplified model of compensation accounting to illustrate the
differences between three different measurement alternatives: the
measure of realized pay that we construct in this article, and two
measures of expected pay--the simple measure of expected pay readily
available in Execucomp, direct compensation, and the one based on the
concept of total yearly compensation introduced by Antle and Smith
(1985). Section 3 presents the results on the implied measure of
realized pay over time, with a special focus on pay sensitivity, as well
as a detailed look at the financial sector before and after the recent
financial crises. Section 4 concludes. The Appendix provides the
technical details on how we construct our realized pay measure from the
data available.
1. UNDERSTANDING COMPENSATION PACKAGES
Nowadays, companies pay their top executives mainly through
different combinations of the following instruments: a salary, a bonus
program, a signing bonus, stock grants (also referred to as
"restricted stock," since they are usually granted with
restrictions on the ability to sell them), grants of options on the
stock of the firm, and perks and long-term incentive plans that specify
severance payments, as well as pension plans.
The publicly available information on CEO compensation comes from
the compensation tables included by firms in their annual reports, as
mandated by the Securities and Exchange Commission (SEC). This is the
same data that Execucomp has compiled since 1992 and has been used in
numerous empirical studies of CEO compensation, including this article.
When the press publicizes information on CEO pay, it usually reports a
summary measure of total or "direct compensation," which is
also readily available in Execucomp as the variable TDC1. Direct
compensation is the sum of cash compensation (wage, bonus, and incentive
compensation), pension contribution and other perks, plus the expected
value of new stock and option grants given to the CEO within a given
year. Execucomp also reports separately the different components of
total compensation, and it includes some limited information on stock
ownership and the portfolio of unvested restricted stock and option
grants of the executives. A brief description of each of the instruments
and further details on the information available about them in Execucomp
follows. Table 1 presents statistics for their relative importance as a
share of total pay using data from 1993 to 2010 and summarizes the
information on availability.
Salaries are the simplest compensation instrument: They are not
contingent on performance and information on their level is readily
available on the proxy statements of firms. (4) Bonus plans and
incentive pay typically depend on yearly accounting results. Information
is available mainly on payouts and more recently on some limited details
of the bonus plans. Information on perks and other compensation is also
available, although not to a great level of detail. Grants of restricted
stock of the firm make pay depend on the results of the firm over a
longer time horizon, since the CEO is restricted from selling them until
their vesting period expires. Execucomp compiles information on their
expected value at the time of the grant (number of shares times market
price of stock), but it does not have separate information on the number
of shares granted. Grants of stock options allow the executive to
purchase stock of the firm at a pre-established price (the
"exercise price") and are also typically granted with
restrictions as to how soon they can be exercised. These also provide
incentives for longer-term performance, but they only pay off if the
stock price of the firm is above the exercise price. For option grants,
Execucomp has information on both the number and the Black and Scholes
value of the total grants during the year. Typically, both stock and
option grants come with a clause that forces the executive to forfeit
them in the event of employment termination. Information on the vesting
periods is not generally available in Execucomp for either stock or
option grants. (5)
It should be apparent that compensation instruments can be
classified according to two criteria: whether or not they are contingent
on the performance of the firm, and whether or not they are deferred.
(6) Table 2 summarizes this classification of the main compensation
instruments.
Given that executives are risk averse, paying them with contingent
instruments, such as bonuses, stocks, and options, comes at a cost,
since they will demand higher expected payments to compensate them for
the risk. The most accepted explanation for the inclusion of
compensation instruments that are contingent on the performance of the
firm is the existence of a moral hazard problem: The separation of
ownership and control of the firm implies the need to provide incentives
to the CEO that align his interests with those of the firm owners. (7,8)
Within this context of incentive provision, it is also commonly accepted
that expectations over future wages or jobs (career concerns), as well
as the threat of dismissal, are also important compensation
instruments--although less easy to study due to the lack of hard
information on them. (9)
Deferral of pay also comes at a cost if CEOs are more impatient
(i.e., they discount the future more) than the shareholders of the firms
they manage. Several reasons may explain the use of deferred
instruments. Perhaps the most accepted one is that, despite the cost of
waiting, deferral is valuable--in combination with commitment to
long-term contracts--because it allows to smooth incentives over time,
making (costly) exposure to risk less necessary. (10) Other reasons
include retention purposes in the face of lack of commitment to
long-term contracts or provision of incentives for hidden actions with
long-term effects. (11)
In most cases, instruments that are "cashed" within the
year (labeled "current" in the table) are straightforward to
value. In contrast, for contingent deferred instruments an expected
value needs to be calculated, which presents some challenges. For
example, the actual amount of compensation that the CEO will receive
from stock and options granted to him in a given fiscal year will depend
on the stock price of the firm at the moment he sells or exercises them.
Similarly, the value of future compensation will depend on the
performance of the firm during the tenure of the CEO. The value of
pension payments will be contingent on the firm being solvent once the
CEO retires. The value of severance payments is typically pre-set at the
time of contracting, but a full list of the contingencies that may lead
to termination is not written in the employment contract of the CEO.
Hence, in order to calculate the expected value of compensation, one
needs to know both the set of contingencies that trigger each payment
(for example, the circumstances that trigger firing of the CEO or the
performance targets for granting salary increases), as well as the
probability attached to each of these performance contingencies (for
example, the probability distribution over future stock prices of the
firm). These difficulties are important when choosing a measure of CEO
pay.
Measurement of Pay: Expected versus Realized Value
There are two main approaches to measuring CEO pay:
1. Expected value of pay: The expected value of compensation
granted in a given year, which includes the cash (realized value) he
receives in salary and bonus, plus the expected value of the deferred
contingent instruments such as stock and options;
2. Realized pay: The actual amount of money received in a given
year, which includes the cash he receives in salary and bonus, plus the
proceeds from selling past stock and option grants for which selling
restrictions have expired (all realized).
Any attempt at valuing contingent deferred compensation, either in
expectation or its realized value, will be constrained by the
availability of data. Table 3 summarizes the data available in proxy
statements and compiled by Execucomp about CEO holdings of stock and
options of his own firm, the evolution of which is key to measurements
in both categories. For stock holdings, we have the number of shares
held by the CEO at the end of the fiscal year, as well as the number and
value of both stock that remains restricted and of stock that vested
during the year. For option holdings, we know the number of options
exercised during the year, as well as their value. We also know the
number and value of options exercisable (but still unexercised) and
those whose vesting restrictions did not yet expire. These values,
however, are calculated using the "intrinsic" valuation (stock
price at the end of the year minus exercise price, times number of
options, if positive), hence ignoring the options that are currently out
of the money, and provide a simplistic evaluation (Black and Scholes
would be a more accurate choice).
We choose our measure of realized pay (presented in the next
section) in light of these data availability issues. Our choice tries to
minimize the sensitivity of our measurements to assumptions about the
unknown details of compensation packages, while still exploiting the
information we have available on the portfolio of stock and options of
the CEO.
Before we present our measure, it is important to note that we view
expected and realized measures of pay as complements rather than
substitutes when trying to understand incentives for CEOs. Expected pay
is a forward-looking measure, which gives important information about
the value of the current compensation package given to the CEO. However,
it is a difficult task to get a realistic valuation of stock or options
for the CEO, especially because of selling restrictions and risk
aversion considerations. In practice, the data in Execucomp reflects the
firm's estimate of that value for CEOs. For options, usually a
pricing model based on arbitrage conditions, such as Black and
Scholes' option valuation model, is used to provide a value in the
company's report with the SEC. Ad hoc modifications are often used
to accommodate the fact that CEOs are risk averse and there are selling
restrictions on the option grants. (12)
Realized pay, instead, is a backward-looking measure: Given past
performance, we can calculate how much payoff the CEO actually got in
the given period. In contract theory terms, we can view this measure as
a description of the contract payoffs on the equilibrium path. That is,
we observe what the CEO gets for the actual performance that
materialized, but we do not have information on what the payoffs would
have been for better or worse performances. For an estimate of these
off-the-equilibrium-path payoffs, in Section 3 we perform sensitivity
analyses that exploit the fact that we have some information on the
number of stocks and options the executive sold or exercised.
One advantage of our realized pay measure is that we do not need to
take expectations over the value of deferred contingent pay. Hence, we
will be able to use the publicly available information on compensation
packages without resorting to assumptions about the future value of
contingent compensation. Still, even for the purposes of measuring
realized pay, we are missing some important information on these
deferred contingent instruments. As reflected in Table 3, Execucomp
records the value of stock and the value and number of stock underlying
options at the time when they are granted to the CEO. The values are
approximations to the expected income that the CEO will realize in the
future, when their restrictions expire. However, we do not have explicit
information on the vesting schedules of these grants, or the exact date
when the vested stocks are sold or the options exercised, or the market
price of the stock at those times. This information is key to compute
the actual cash the CEO receives as a result of the original grant. Our
construction of a realized pay measure will necessarily involve
assumptions on these unknown characteristics of the compensation, which
we discuss in detail in the Appendix.
Larcker, McCall, and Tayan (2011) have a short and interesting
essay in which they also point out the differences in measuring expected
and realized pay. (13) The authors include illustrative examples of the
difference between expected and realized compensation based on data for
a handful of firms in the year 2010. In this article we will use a
larger number of firms and a longer period of time to illustrate
quantitatively the difference between the two measures.
2. CONSTRUCTING A MEASURE OF REALIZED PAY
In this section, we provide a framework for comparing different
measures of compensation. For this, we describe the types and timing of
the different components in a typical compensation package. Using this
framework, we introduce our proposed measure of realized contingent pay,
denoted [I.sub.t], which is defined as the sum of salary, bonus, and
gains from selling stock and exercising options in the current year. To
construct it, we use information on the several components of pay
packages that is publicly available, along with some assumptions. We
refer to the model to illustrate the need for these assumptions and to
justify our choices. Then we illustrate in the context of the model what
the differences are between our measure and two alternative ones: (1)
direct compensation, which is defined as the sum of salary, bonus,
perks, and other compensation, and the value of stock and options at the
time of grant, and (2) total yearly compensation, which is defined as
direct compensation plus dividends, plus the change in the value of
stock and options in the portfolio of the CEO.
Consider a CEO who lives for T years. He starts his tenure with a
firm at year t = 1. He receives compensation for all the years he is
working, and after he retires he consumes out of his accumulated wealth
and pension payments. We assume he has no sources of income other than
what he receives as payments for his job as CEO, which we denote as
[I.sub.t]. The value he attaches to his employment at the beginning of
period 1, denoted V0, is equal to the expected stream of income that he
expects to receive in exchange for his work in each of the periods of
his life: (14)
[V.sub.0]([e.sup.*]) = E[[T.summation over
(t=1)[I.sub.t]([p.sub.1], ..., [p.sub.t])/[(1 + r).sup.t-1]|[e.sup.*]],
(1)
where the expectation is with respect to stock price realizations
(which summarize the performance of the firm in this simple model),
conditional on the sequence of effort choices by the CEO (denoted
[e.sup.*]) given the optimal contract. We denote the market interest as
r.
In this article, we want to measure the realized value of
[I.sub.t]. A more ambitious objective, which would relate more directly
to theoretical models of CEO compensation based on repeated moral hazard
models (Wang 1997), would be to try to measure [V.sub.t] ([e.sup.*]). We
discuss some of the added difficulties of this measurement at the end of
this section.
Realized pay [I.sub.t] will not all be delivered directly in cash.
Rather, the executive will receive an annual compensation, [C.sub.t],
that will consist of two elements: a cash-based portion, or current
liquid payment, denoted [L.sub.t], and a grant-based portion, denoted
[G.sub.t]. We assume compensation is received only once per year, at the
end of the fiscal year. We have that
[C.sub.t] = [L.sub.t] + [G.sub.t] [for all]t, (2)
where
[L.sub.t] = [W.sub.t] + [B.sub.t] + [D.sub.t] + [K.sub.t] [for
all]t.
That is, [L.sub.t] is the sum of annual salary [W.sub.t], bonus
payment [B.sub.t], which usually will depend on the annual results of
the firm, dividends [D.sub.t], and perks and contributions to pension
plans [K.sub.t]. (15) Grants consist of both restricted stock of the
firm, [s.sup.r.sub.t], and options to buy stock, [o.sup.r.sub.t], and
are valued at any t' [greater than or equal to] t as (16)
[G.sup.t'.sub.t] = EV ([s.sup.r.sub.t]; [p.sub.t']) + EV
([o.sup.r.sub.t]; [x.sub.t], [p.sub.t']) =
[s.sup.r.sub.t][p.sub.t'] + EV ([o.sup.r.sub.t]; [x.sub.t],
[p.sub.t'])
In this expression, EV ([s.sup.r.sub.t]; [p.sub.t']) is the
estimated value of restricted stock, i.e., the amount of stock,
[s.sup.r.sub.t], valued at the stock price at the time of valuation,
[p.sub.t']. The estimated value of options, EV ([o.sup.r.sub.t];
[x.sub.t], [p.sub.t']), stands for some version of the Black and
Scholes (1973) option valuation formula and depends both on the market
price at the time of valuation, [p.sub.t'], and the exercise price,
[x.sub.t].
Our Measure of Realized Pay
The stream of realized pay [I.sub.t] that the CEO will receive from
the firm while working will be equal to the cash part of his
compensation, Lt, plus whatever net gains from trade he gets from buying
and selling unrestricted stock (or vested exercising options). To
compute these gains from trade, it will be important to keep track of
the accumulated number of stock and option grants that have vested, what
we will refer to as the "portfolio" of the CEO. (17) Let
[S.sub.t-1] denote his holdings of unrestricted stock at the beginning
of period t, and [O.sub.t-1] denote his holdings of vested options. Let
[T.sub.t] ([S.sub.t-1], [O.sub.t-1]) denote the gains from the sales of
stock and exercises of options at period t: Then, we can write realized
pay as
[I.sub.t] = [L.sub.t] + [T.sub.t] ([S.sub.t-1], [O.sub.t-1])
Tracking the holdings [S.sub.t] and [O.sub.t] involves
understanding the law of motion of the quantities of vested stock and
options available to the CEO. Under the assumption that the CEO did not
own any stock or options of the firm before his employment as CEO
started, we have that his holdings in the beginning of year 1 are equal
to zero:
[S.sub.0] = 0;
[O.sub.0] = 0.
Any subsequent year, the quantities available to trade will change
for two main reasons:
1. some of the past grants will have vested, or the CEO may choose
to buy unrestricted stock; these actions will increase his holdings;
2. some of the past grants in his holdings will be sold or
exercised, decreasing his holdings.
It is worth noting here that accurately evaluating the evolution of
the holdings of the CEO would necessitate a large amount of information.
For example, the CEO may choose to buy or sell stock, or exercise
options, at different times during the year--with different market
prices for each transaction. Also, he may choose to exercise options and
hold on to the stock that he obtains with this transaction. Moreover, he
may inherit or donate stock at any time. Unfortunately, the only data we
have for the holdings of stock and options is their quantities and value
at the end of each fiscal year (see Table 3), and we are lacking the
details on the specific transactions that determine their evolution.
Hence, we make the following important simplifying assumptions. First,
we assume each of the possible trades happens only once in the fiscal
year. Note that this still accommodates for a given sale of options to
include options from different past grants, which implies different
exercise prices. Second, we assume that the executive never purchases
options, and that he exercises options only if he plans to sell the
stock immediately. Third, we ignore any inheritances or donations.
We can summarize the above discussion in a formal law of motion for
the holdings of stock and options by introducing some notation. The
vesting restrictions on the stock and option grants determine the
available [S.sub.t] and [O.sub.t] in each period. Typically, only a
portion of the previous years' restricted stock vests every t.
Denoting the vested shares in year t by [s.sup.v.sub.t] and vested
options in year t by [o.sup.v.sub.t], the accumulated number of shares
and options available for selling in year t is
[S.sub.t] = [S.sub.t-1] - ([s.sup.s.sub.t] - [s.sup.b.sub.t]) +
[s.sup.v.sub.t], [O.sub.t] = [summation over [O.sub.g][member
of][O.sub.t-1]][O.sub.g] - [o.sup.e.sub.g,t] + [o.sup.v.sub.t], (3)
where we are denoting the three types of trades that can happen at
time t as follows:
1. selling stock [s.sup.s.sub.t] of the unrestricted stock
available at period t, [S.sub.t-1], at price [MATHEMATICAL EXPRESSION
NOT REPRODUCIBLE IN ASCII],
2. buying an amount [s.sup.b.sub.t] of stock from the market, at
price [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],
3. buying stock through the exercise of [o.sup.e.sub.g,t] of any
vested option grant g (with corresponding exercise [x.sub.g]) at price
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].
With this notation, we can write an expression for the gains from
trade:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (4)
This completes the description of our measure of realized pay,
[I.sub.t]. Next, before moving on to the estimates of [I.sub.t] using
data, we use the model in this section to compare our measure of
realized pay with alternative measures used in the literature.
Alternative Measures: Expected Pay
As we discussed in Section 1, the literature has used compensation
measures based on the expected value of pay. The theoretical measure of
expected pay is described by (1). The employment value, [V.sub.t], is
the sum of the expected stream of realized pay. For the measurement of
[V.sub.t] ([e.sup.*]) in the data, however, one would have to make
assumptions about the terms of the contract offered to the CEO regarding
compensation in future periods (i.e., what would trigger a wage
increase, or what is the schedule of future grants contingent on
realized performance). One would also need to understand the CEO's
expectations about stock prices in the future, which will determine his
future realized gains from trade. One would also need to understand his
expectations regarding his transitions to other firms and their
consequences for his realized pay. Moreover, one would need to model how
performance during the CEO's working life will affect his pension
payments. To the best of our knowledge, no study has provided a reliable
measure of [V.sub.t]. Instead, two different approximations to [V.sub.t]
have been widely used: "direct compensation" (TDC1) and
"total yearly compensation" (TYC). We define each of these
using our notation, in turn, and compare them to our measure of realized
pay.
The Execucomp variable TDC1 can be written in terms of our notation
as
[TDC1.sub.t] = [W.sub.t] + [B.sub.t] + [K.sub.t] + [G.sup.t.sub.t].
This measure of expected pay does not closely correspond to the
theoretical [V.sub.t], since it does not include any estimation of
future wages, bonuses, and new grants. It includes an estimate of the
expected future value of the grants given to the CEO in the current
year, [G.sup.t.sub.t] = [s.sup.r.sub.t][p.sub.t] + EV ([o.sup.r.sub.t];
[x.sub.t], [p.sub.t]), but it ignores the changes in the value of past
grants, or the realized gains from exercising them once they are vested,
as well as the dividends that correspond to the CEO from holding stock.
The main difference between our I measure and TDC1 is that we do not
include the value of grants, [G.sub.t], but rather the realized net
gains from trade, [T.sub.t]. Also, dividends are included in [I.sub.t]
but not in [TDC1.sub.t]
A second alternative measure of expected pay, TYC, has been used in
the literature since Antle and Smith (1985) proposed it. The idea behind
it is to calculate the expected value that the CEO attaches to working
in his firm, every period, as the current expected value of stock and
option holdings plus the expected future compensation; then one can
interpret the annual change in this expected value from one period to
the next as the TYC of the executive. (18) Because the expected value of
grants is updated every year, this measure presents a more accurate
picture of the incentive value of the CEO's contract. However, the
measure is not without problems. For example, a common simplifying
assumption when computing this measure is to assume that salary and
bonus payments remain constant in future years and that the expected
value of future grants is zero. (19)
We follow the description in the Appendix of Clementi and Cooley
(2009) to replicate their measure of TYC, assuming wages, bonuses, and
perks remain constant throughout the work life of the CEO, and no
turnover. We graph it for comparison purposes in Figures 1, 2, and 5. In
terms of our notation, TYC can be written as
[TYC.sub.t] = [W.sub.t] + [B.sub.t] + [K.sub.t] + [D.sub.t] +
[t.summation over ([tau]-1)]([G.sup.t.sub.[tau]] -
[G.sup.t-1.sub.[tau]]),
where [G.sup.t.sub.[tau]] in this case denotes the updated expected
value during period t of stock and (unexpired) option grants that were
given at period [tau] [less than or equal to] t and are still
unexercised. (20)
The measure TYC attributes initial grants as compensation in the
year when they are granted, and then subsequent appreciations and
depreciations of the grants to the periods when they happen--even if
they do not translate into realized pay in that particular period. In
comparison, our measure I of realized pay records only the realized
value of grants when they get exercised, and it attributes the gains
from trade to the particular period when they happen. It is easy to see
that the simple sum of [[summation].sup.T.sub.t=1] [I.sub.t] =
[[summation].sup.T.sub.t=1][TYC.sub.t]; however, the individual year
entries will differ, and hence the properly discounted sum will differ
as well.
3. MEASUREMENTS
In this section, we present the empirical measurement of pay
according to the methodology described above. In the Appendix, we
provide the details on how to map the elements of pay described in the
previous section to the data available in Execucomp.
In this article, we work with the August 2013 release of Execucomp,
which includes annual observations through the fiscal year 2012. We drop
CEOs who own 50 percent or more of the shares of their company, since we
want to focus on measuring incentives in relationships for which there
is an agency problem. Our final sample includes 45 different firms, for
a total 397 firm-CEO-year observations. (21, 22)
Figure 1 presented the median of our measure of realized pay from
1993 to 2012. We compare it to the two measures of expected pay
discussed earlier in this article: "total compensation"
reported in Execucomp as the variable TDC1 and our own calculation of
TYC following Clementi and Cooley (2009). (23)
Two features emerge from Figure 2. First, averages are much larger
than medians. This is well known for the measure TDC1, and it is
confirmed for our measure of realized pay, I. Second, average realized
pay is more volatile over time than average total compensation, and it
is typically above [TDC1.sub.t], while it was typically below it when we
looked at the medians in Figure 1. However, [TYC.sub.t] is more volatile
than either of the other two measures. This is true both when looking at
medians, in Figure 1, or when looking at means, here. Our analysis of
the different components of pay shows that the estimated gains from
trading stock are causing the volatility in realized pay. Also, every
year there are a few CEOs who realize very large gains from trading
stock, making the averages of the two measures of compensation differ
more than the medians. Moreover, the large revaluations of the portfolio
of the CEOs with changes in the stock price do not seem to translate
into gains from trades, causing the large deviation of the measure TYC
from the measure I. One potential explanation would be that CEOs have in
their portfolios a large fraction of restricted stock and options, so
even if their value increases they are not able to realize those gains.
However, the information available in Execucomp about restricted stock
and options does not seem to support this hypothesis (the restricted
grants are a small part of the portfolio of the CEO at any point in
time). However, it is still plausible that implicit selling restrictions
are in place even after the explicit vesting period expires, presumably
with the objective of strengthening the market perception about the
confidence of the CEO in the performance of his own firm.
[FIGURE 2 OMITTED]
[FIGURE 3 OMITTED]
In Figure 3, we display the liquid portion of compensation for mean
realized pay, [I.sub.t], and for mean total expected pay as measured in
[TDC1.sub.t]. We see that the higher volatility of mean [I.sub.t]
compared to that of mean [TDC1.sub.t] is mainly driven by the volatility
of trades. Figure 4 plots separately the medians of the different
components of realized pay, [L.sub.t] and [T.sub.t], and the median of
[I.sub.t]. (Figure 4 plots also these statistics for finance firms,
which we will discuss in the next subsection.) Both components, as well
as the total [I.sub.t], are increasing over time. For comparison, the
median value of grants, [G.sub.t], is included as well. The value of
grants is also increasing over time.
As a robustness check, we replicate Figure 2 in Figure 5 for a
subsample of the firms including only the CEOs that own less than 1
percent of the shares of their company. (24) The level of [TYC.sub.t] is
much lower, and mean realized pay is sometimes above [TDC1.sub.t]. The
main difference for this sample continues to be the higher volatility of
[TYC.sub.t].
[FIGURE 4 OMITTED]
Finance Firms
In Figure 4, we include statistics for firms in the finance sector
with the statistics for firms in all sectors. (25) Note that firms in
the finance sector are, on average, larger (in the sample, the average
size in finance is between five and six times larger than the average
size for all firms, year by year, with a decreasing trend between 2004
and 2009). Because the level of total compensation (TDC1) has been shown
to be positively correlated with size, we expect a higher realized pay
for CEOs in finance. This is confirmed in the data up to the financial
crisis of 2008. Figure 4 shows that the composition of realized pay is
slightly different among finance firms, with higher liquid compensation
and higher value of trades (which are also more volatile, although this
could be due to the smaller number of firms).
[FIGURE 5 OMITTED]
When looking in detail at the period since the 2008 financial
crisis, it is apparent in the graphs that there has been a steeper
decline in median realized pay--both for liquid compensation and
trades--for firms in finance than for the full sample of firms. It is
worth noting that the median value of grants is, for both groups of
firms, well above the median value of trades. The adjustment pattern of
median grants during the crisis is similar to that of realized pay,
i.e., we see a steeper decline for firms in finance.
Sensitivity of Realized Pay to Performance
Hall and Liebman (1998) provide a measure of sensitivity of pay to
performance by using information on stock holdings to construct
counterfactuals. (26) First, they construct a measure of the portfolio
of the CEOs, similar to our [S.sub.t] and [O.sub.t] holdings of stock
and options. Then, using the realized distribution of performances
(stock returns), they evaluate the holdings of each CEO in the data for
different performance scenarios corresponding to different percentiles
of the distribution of returns. We follow this methodology and provide a
similar counterfactual for our measure of annual realized pay. An
important caveat of this measure is that the quantities of stock traded
and of options exercised are assumed to remain constant when stock
prices vary in the counterfactual. A model of how these trades would
vary in a more realistic setup is beyond the scope of this article.
[FIGURE 6 OMITTED]
For our performance counterfactuals, we need to propose the support
and distribution of stock returns. For this, we use the observed
distribution of stock returns in each given year. We denote the annual
stock price return as
[r.sub.t] = [p.sub.t] - [p.sub.t-1]/[p.sub.t-1] (5)
This measure has the advantage of being comparable across firms, as
opposed to the stock price itself. In Figure 6, we summarize the
evolution of these distributions of returns [r.sub.t] of the 00 largest
firms in our sample over time by plotting the return value for the
median, and the 5th and 95th percentiles.
Each realization of returns in the support of the distribution can
be translated into a stock price for each individual firm using (5).
That is, when calculating the counterfactual value of [T.sub.t] for an
individual executive working for firm j, we will construct a
counterfactual stock price for various percentiles of the return
distribution. We use a hat to denote a variable's counterfactual
value, and a superscript nth to indicate the percentile to which we are
setting the performance of the firm. For the nth percentile, the
counterfactual price for firm j at time t is
[[??].sup.nth.sub.j,t] = (1 + [r.sup.nth.sub.t]) [p.sub.j,t-1].
With this price [[??].sup.nth.sub.j,t], a new valuation of
[T.sub.j,t] can be produced, assuming the return of the firm was equal
to the nth percentile return, [r.sup.nth.sub.t]. Recall that we
approximate the gains from trade coming from stock purchases and sales
as max [0, [[bar.p].sub.t][q.sub.t]], where [[bar.p].sub.t] is the
average price within the year. We will set the counterfactual for this
average price to
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (6)
that is, we assume that the proportionality between the average
price and the end-of-the-year price is maintained in the counterfactual.
For the portion of the gains from trade that comes from exercising
options, we will need several pieces of information. First, in order to
compute the net benefit per option exercised, ([MATHEMATICAL EXPRESSION
NOT REPRODUCIBLE IN ASCII]), we would need to construct the
counterfactual for the stock price at the time of exercise,
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], possibly using
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], and we would need
to know the exercise price, [x.sub.g], corresponding to each option
exercised. Unfortunately, as discussed earlier, we do not know
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] or [x.sub.g] (we do
not know which particular past grant g was used to purchase the shares).
The value of exercised options is recorded in Execucomp:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
We also have the number of options exercised within the year:
[o.sup.e.sub.t] = [summation over [o.sub.g][member of][O.sub.0]]
[o.sup.e.sub.g,t] [equivalent to] OPT_EXER_[NUM.sub.t] [for all]t.
To produce an estimate for the counterfactual value of exercising
options, we assume [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],
the average price during the year, and we solve for an
"effective" exercise price x using
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
Finally, we also assume that CEOs do not exercise options in the
counterfactual if they are "out of the money" (that is, if
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]). With these
assumptions, we have that our counterfactual for gains of trade is
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
This, together with the actual liquid compensation for the
executive in the data, [L.sub.t], which is not contingent on stock price
realizations, amounts to a calculation of a counterfactual
[[??].sup.nth.sub.j,t].
The numerical results are listed in Tables 4 (levels) and 5
(percentage changes). We display the percentage changes for the 5th,
median, and 95th percentile counterfactuals graphically in Figure 7.
Keeping in mind that percentage changes are bounded below by -100
percent, we see that there is an obvious asymmetry in changes when the
firm performs better rather than worse. This responds to the
uncontingent nature of the wage and the bonus in our calculations. Also,
we see in Figure 7 that the gains for the 95th percentile (i.e.,
outstanding stock return performance) is very extreme in particular
years. Two things can lead to high net gains from trade: particularly
good stock returns in the given year (i.e., the 95th percentile stock
return is an outlier when compared to the other 95th percentile returns
in other years) or particularly generous past grants that imply a large
number of stock or options are available for trade. We can use the
distribution of stock returns, plotted in Figure 5, to track which of
the two explanations seems more plausible. The years 2000, 2003, and
2009 represent examples of outlier stock return performance in the 95th
percentile; however, only in the year 2000 does this translate into a
very large counterfactual mean realized pay in the 95th percentile. The
spikes in income for the years 2005 and, to a lesser extent, 2008-09 may
correspond instead to particularly large net quantities traded, as
computed by us from the portfolios of the CEOs.
Sensitivity for Finance Firms
We observed a sharper decrease in median realized pay for firms in
finance during the recent financial crisis (see Figure 4). However, this
does not seem to correspond to a very different sensitivity of realized
pay to performance for financial firms during the crisis. Tables 5 and 7
replicate the sensitivity analysis of Tables 4 and 6 for firms in
finance. That is, using the stock and option holdings of financial
firms, we feed in the same percentile stock returns used in Tables 4 and
6 (i.e., those from the distribution of stock for the overall population
of firms) to calculate their counterfactual realized pays. We find that
the sensitivity estimates align with those of the general sample for the
whole sample period. (27) It is worth referring back to Figure 4 and
noting that the median liquid (uncontingent) compensation of CEOs in
finance is particularly large compared to the entire sample, up until
the recent crisis. This, together with the fact that sensitivity
estimates are similar to those of the overall sample, suggests that the
quantities of stock and options held by finance CEOs are larger than
those in other industries, hence implementing a similar risk in their
realized pay in spite of larger uncontingent compensation levels.
4. CONCLUSION
Information on CEO pay is typically obtained from the mandatory
disclosure of compensation required by the SEC for large public firms. A
good measure of realized pay for CEOs, which includes the actual gains
from trading stock rather than their expected value at the time when the
firm awards them to the CEO, is not readily in this source. This article
discusses how to construct an approximation to the value of realized pay
using the partial information compiled in the database Execucomp on the
stock owned, bought, and sold by CEOs each year. We present our
estimates for the period 1993-2012 and compare them to two alternative
measures of expected annual total compensation that are frequently used
in the media and the academic literature: direct compensation (the sum
of salary, bonus, other compensation, and the market value of new
grants) and total yearly compensation (which includes the year-on-year
change in the value of the stock holdings of the CEO). Our measure of
realized pay tends to be more volatile over time than direct
compensation, mainly due to the volatility of the gains that CEOs
realize from trading stock. However, total yearly compensation is
markedly more volatile than the other two measures. We find that, while
the average realized pay level has historically been at or above that of
direct compensation, its median has consistently been lower. We provide
descriptive statistics of realized pay for firms in the finance sector.
In the aftermath of the crisis the realized pay of CEOs of finance firms
seems to have decreased in level relative to the realized pay of CEOs in
all industries. Our calculations suggest, however, that realized pay of
finance CEOs changes with the performance of their firm in similar
magnitudes to that of the average CEO for the whole 1993-2012 period.
APPENDIX
In this Appendix, we show how to map the variables defined in
Section 2 to the Execucomp database. We discuss the elements of our
ideal measure of compensation that are missing in the data, and what
assumptions we make to go around these difficulties.
As we list the objects needed to calculate [I.sub.t], we will note
how the change in reporting requirements of the SEC in 2006 changes the
availability of data (or, sometimes, simply the name of the Execucomp
variable that corresponds to a given concept). For this purpose, we will
refer to the reporting period before 2006 as [P.sub.1], and the one
after as [P.sub.2].
Measuring Liquid Compensation, [L.sub.t]
Our measure of liquid or cash-based compensation, [L.sub.t], is the
sum of the executives' annual salary, bonus, dividends, and any
perks received within the year, such as contributions to pension plans.
Data on annual salary [W.sub.t] is directly available in Execucomp:
[W.sub.t] [equivalent to] [SALARY.sub.t], [for all]t.
Our measure of bonus, [B.sub.t], is the sum of the Execucomp
variable BONUS and two variables that capture payments received from
hitting "objective" performance targets such as sales growth
or stock price performance: (28)
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
We also have information in the data about the dividend yield
(dividends per share, divided by [p.sub.t], times 100) that the
executive receives from his stock ownership of the company. We back out
the total dividend payments as follows:
[D.sub.t] [equivalent to] [DIV_YIELD.sub.t]/100 x [PRCCF.sub.t] x
SHROWN_EXCL_[OPTS.sub.t] [for all]t,
where [PRCCF.sub.t] is Execucomp's record of the stock price
at the closing of the fiscal year:
[p.sub.t] = [PRCCF.sub.t] [for all]t.
Finally, our measure of perks and pension payments [K.sub.t] is the
sum of Execucomp variables related to "other compensation":
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
Tracking Grants, [G.sub.t]
Our measure of grant-based compensation [G.sub.t] is the sum of the
value of restricted stock grants and options in the period. We have data
on the value of the stock component of that sum, EV([s.sup.r.sub.t];
[p.sub.t]), with the following variables: (29)
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
In reality, there may be N grants within the year, each with a
quantity [s.sub.t,n] and a market price at the time of granting of
[p.sub.t,n], for n = 1 : N: The variables above that we observe in
Execucomp will not have the disaggregated information grant by grant,
but rather they correspond to
EV([s.sup.r.sub.t]; [p.sub.t]) = [N.summation over (n=1)]
[s.sub.t,n][p.sub.t,n].
The value of options awarded in the period is recorded in the data
as follows: (30)
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
Again, these variables aggregate all grants within a year, so
effectively we will set
EV ([o.sup.r.sub.t]; [x.sub.t], [p.sub.t]) = [M.summation over
(g=1)] EV ([o.sup.r.sub.g,t]; [x.sub.tg], [p.sub.g,t]),
where M is the total number of option grants in the year. There is
some partial information in Execucomp about the date and exercise price
of the different grants for an executive in a given year. However, we do
not have their vesting schedule or the date of their exercise (that is,
we do not know what the stock market price was at the time when the
executive exercised the options). See the related discussion in the
realized pay sensitivity analysis in Section 3.
Computing Net Gains from Trading Stock, [T.sub.t]
We will now define the components of our net gains from trade
measure, [T.sub.t]. To begin, recall that we assume each of these trades
happens only once in the fiscal year, and if the executive exercises
options, he sells the acquired shares immediately.
The portion of [T.sub.t] that comes from exercising options is
captured by the Execucomp variable OPT_EXER_VAL: (31)
[summation over [o.sup.e.sub.g,t][member
of][O.sub.t-1]][o.sup.e.sub.g,t]([p.sub.e,t] - [x.sub.g]) [equivalent
to] OPT_EXER_[VAL.sub.t], [for all]t.
The portion of [T.sub.t] that comes instead from buying and selling
stock on the open market, [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN
ASCII], must be estimated, because we cannot observe in the data the
quantities [s.sup.s.sub.t] or [s.sup.b.sub.t] (and, correspondingly, the
prices [p.sup.s.sub.t] or [p.sup.b.sub.t]). We use an algorithm similar
to Clementi and Cooley (2009) to estimate this difference, with slightly
different assumptions that we discuss later in this section. From the
law of motion for vested stock in (3), we have that the difference
between last year's unrestricted stock holdings and this
year's is either coming from the newly vested stock this year,
[s.sup.v.sub.t], or net purchases. We denote the net quantity of shares
sold in t as [q.sub.t] [equivalent to] [s.sup.s.sub.t] -
[s.sup.b.sub.t]. Rearranging (3) and substituting [q.sub.t], we have
[q.sub.t] = [S.sub.t-1] - [S.sub.t] + [s.sup.v.sub.t], [for all]t.
(7)
Typically, [q.sub.t] will be positive in the data, i.e., the CEO
will sell more shares than he buys in a given year. Occasionally,
however, [q.sub.t] calculated as in (7) will be negative. This could be
due to violations of our assumption that the CEO immediately sells stock
acquired through the exercise of options. (32) Because we would rather
bias our measure of realized pay upward, we set [q.sub.t] in our
calculations equal to the maximum of [q.sub.t] from (7) and 0.
To calculate [q.sub.t] using (7) we need [S.sub.t-1] and [S.sub.t],
which correspond to the CEO's holdings of unrestricted stock. We
observe this variable directly in Execucomp: (33)
[S.sub.t] [equivalent to] SHROWN_EXCL_[OPTS.sub.t], [for all]t.
We also need the variable [s.sup.v.sub.t], the stock vested within
the year. This variable maps directly into Execucomp's
SHRS_VEST_NUM in the reporting period [P.sub.2]. For observations in
[P.sub.1], when it is missing, we estimate it by examining annual
changes in aggregate restricted stock holdings and annual grants.
Specifically:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where our measurement of the number of stocks granted within the
year, [s.sup.r.sub.t], is an approximation to the real total number of
stock (unavailable in the data) that we recover from EV([s.sup.r.sub.t])
by assuming all grants are valued at the average price within the year,
denoted [bar.[p.sub.t]]: (34)
[s.sup.r.sub.t] = EV([s.sup.r.sub.t])/[bar.[p.sub.t]].
Note that [bar.[p.sub.t]] is not in Execucomp. We match the firms
in Execucomp to a different database from the Center for Research in
Security Prices (CRSP) containing daily stock prices, and we construct
the average price ourselves. For this, we take the 12-month window of
each firm's fiscal year. To summarize, in our notation, our
estimate for the amount of stock vested within t is
[s.sup.v.sub.t] = [S.sup.r.sub.t-1] - [S.sup.r.sub.t] +
[s.sup.r.sub.t].
Once we get [q.sub.t] from (7), we estimate the value [MATHEMATICAL
EXPRESSION NOT REPRODUCIBLE IN ASCII] by assuming the [q.sub.t] shares
were traded at the average market price over the year, i.e.,
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. Given our
assumption of non-negative net quantities traded, this amounts to
stating
[s.sup.s.sub.t][p.sub.s,t] - [s.sup.b.sub.t][p.sub.b,t] [equivalent
to] max[0, [[bar.p].sub.t][q.sub.t]].
Thus, adding the stock and option portions of [T.sub.t], we get
[T.sub.t] ([S.sub.t-1], [O.sub.t-1]) [equivalent to] max[0,
[[bar.p].sub.t][q.sub.t]] + OPT_EXER_V [AL.sub.t], [for all]t.
Note that there are two differences between our estimation of net
revenue from trade and the calculations in Clementi and Cooley (2009).
First, we use average instead of end-of-year prices to recover the
quantity of shares granted in a given year, [s.sup.r.sub.t], from the
value of the grants; this influences our estimate of the net quantities
traded, [q.sub.t]. Second, we use OPT_EXER_VAL directly to account for
the proceeds of options sales during the year: This variable is the true
value of option exercises collected in Execucomp and hence uses actual
exercise prices and actual stock prices on date of exercise. Clementi
and Cooley (2009) instead choose to lump the stock purchases resulting
from option exercises in with other stock sales, and they assume that
they are acquired at the average price.
REFERENCES
Aggarwal, Rajesh K., and Andrew A. Samwick. 1999. "The Other
Side of the Trade-off: The Impact of Risk on Executive
Compensation." Journal of Political Economy 107 (February): 65-105.
Antle, Rick, and Abbie Smith. 1985. "Measuring Executive
Compensation: Methods and an Application." Journal of Accounting
Research 23 (Spring): 296-325.
Baker, George P., and Brian J. Hall. 2004. "CEO Incentives and
Firm Size." Journal of Labor Economics 22 (October): 767-98.
Black, Fischer, and Myron S. Scholes. 1973. "The Pricing of
Options and Corporate Liabilities." Journal of Political Economy 81
(May/June): 637-54.
Bebchuk, Lucian, and Jesse Fried. 2004. Pay without Performance:
The Unfulfilled Promise of Executive Compensation. Cambridge, Mass.:
Harvard University Press.
Bolton, Patrick, Jose Sheinkman, and Wei Xiong. 2006.
"Executive Compensation and Short-termist Behaviour in Speculative
Markets." Review of Economic Studies 73 (July): 577-610.
Clementi, Gian Luca, and Thomas F. Cooley. 2010. "Executive
Compensation: Facts." Working Paper 15426. Cambridge, Mass.:
National Bureau of Economic Research (October).
Clementi, Gian Luca, Thomas F. Cooley, and Cheng Wang. 2006.
"Stock Grants as a Commitment Device." Journal or Economic
Dynamics and Control 30 (November): 91-216.
Edmans, Alex, and Qi Liu. 2011. "Inside Debt." Review of
Finance 15 (1): 75-102.
Edmans, Alex, Xavier Gabaix, and Augustin Landier. 2009. "A
Multiplicative Model of Optimal CEO Incentives in Market
Equilibrium." Review of Financial Studies 22 (December): 81-917.
Frydman, Carola, and Raven E. Saks. 2010. "Executive
Compensation: A New View from a Long-Term Perspective, 1936-2005."
Review of Financial Studies 23 (May): 99-138.
Gabaix, Xavier, and Augustin Landier. 2008. "Why Has CEO Pay
Increased So Much?" The Quarterly Journal of Economics 123 (1):
49-100.
Garen, John E. 1994. "Executive Compensation and
Principal-Agent Theory." Journal of Political Economy 102
(December): 75-99.
Gayle, George-Levi, and Robert A. Miller. 2009. "Has Moral
Hazard Become a More Important Factor in Managerial Compensation?"
American Economic Review 99 (December): 40-69.
Gibbons, Robert, and Kevin J. Murphy. 1992. "Optimal Incentive
Contracts in the Presence of Career Concerns: Theory and Evidence."
Journal of Political Economy 100 (June): 468-505.
Grossman, Sanford J., and Oliver D. Hart. 1983. "An Analysis
of the Principal-Agent Problem." Econometrica 51 (January): 7-45.
Hall, Brian J., and Jeffrey B. Liebman. 1998. "Are CEOs Really
Paid Like Bureaucrats?" The Quarterly Journal of Economics 113
(August): 653-91.
Hall, Brian J., and Kevin J. Murphy. "Stock Options for
Undiversified Executives." Journal of Accounting & Economics 33
(February): 3-42.
Haubrich, Joseph G. 1994. "Risk Aversion, Performance Pay, and
the Principal-Agent Problem." Journal of Political Economy 102
(April): 258-76.
Haubrich, Joseph G., and Ivilina Popova. 1998. "Executive
Compensation: A Calibration Approach." Economic Theory 12 (3):
561-81.
Jarque, Arantxa. 2010. "Hidden Effort, Learning by Doing, and
Wage Dynamics." Federal Reserve Bank of Richmond Economic Quarterly
96 (4): 339-72.
Jarque, Arantxa, and Brian Gaines. 2012. "Regulation and the
Composition of CEO Pay." Federal Reserve Bank of Richmond Economic
Quarterly 98 (4): 309-48.
Jensen, Michael C., and Kevin J. Murphy. 1990. "Performance
Pay and Top-Management Incentives." Journal of Political Economy 98
(April): 225-64.
Jenter, Dirk, and Fadi Kanaan. Forthcoming. "CEO Turnover and
Relative Performance Evaluation." Journal of Finance.
Jin, Li, and S. P. Kothari. 2008. "Effect of Personal Taxes on
Managers' Decisions to Sell Their Stock." Journal of
Accounting and Economics 46 (September): 23-46.
Larcker, David F., Allan L. McCall, and Brian Tayan. 2011.
"What Does it Mean for an Executive to 'Make' $1
Million?" Stanford Closer Look Series No. CGRP-22 (December 14).
Prescott, Edward S. 1999. "A Primer on Moral-Hazard
Models." Federal Reserve Bank of Richmond Economic Quarterly 85
(Winter): 47-77.
Schaefer, Scott. 1998. "The Dependence of Pay-Performance
Sensitivity on the Size of the Firm." The Review of Economics and
Statistics 80 (August): 436-43.
Spear, Stephen E., and Sanjay Srivastava. 1987. "On Repeated
Moral Hazard with Discounting." Review of Economic Studies 54
(October): 599-617.
Wang, Cheng. 1997. "Incentives, CEO Compensation, and
Shareholder Wealth in a Dynamic Agency Model." Journal of Economic
Theory 76 (September): 72-105.
We thank the editor, Ned Prescott, and the referees, Kartik
Athreya, Zhu Wang, and Peter Debbaut, as well as Huberto Ennis and Todd
Keister, for helpful comments. The views expressed in this article are
those of the authors and do not necessarily represent the views of the
Federal Reserve Bank of Richmond or those of the Federal Reserve System.
E-mail: arantxa.jarque@rich.frb.org.
(1) Moreover, it is a fact that most CEOs hold on to stock for
which selling restrictions have expired, or to options that are
exercisable and in the money. The reasons for these
"voluntary" holdings are not entirely clear, since CEOs are
risk averse and standard economic theory would suggest that they would
value a diversified portfolio of assets more. Overconfidence, privileged
information, or personal tax considerations have been proposed in the
literature as potential explanations (Jin and Kothari 2008).
(2) This measure has been studied, for example, in Gabaix and
Landier (2008) and Frydman and Saks (2010).
(3) For another recent application of the algorithm first developed
in Antle and Smith (1985), see Gayle and Miller (2009).
(4) The source for the shares of compensation that are reported
come from Jarque and Gaines (2012). See the article for details on
sample selection.
(5) A commonly cited length of this restriction period is four
years, with vesting taking place proportionally over this period--see
Hall and Liebman (1998).
(6) Firm performance is typically proxied by accounting measures
such as return on equity, sales, and profit, or on market-based measures
such as the stock price.
(7) See Prescott (1999) and Jarque (2010) for an introduction to
static and dynamic moral hazard problems, respectively. Classical
references in the literature include Grossman and Hart (1983), as well
as Spear and Srivastava (1987).
(8) Bebchuck and Fried (2004) argue that captive boards may use
stock and option grants as a less obvious instrument to transfer
excessive amounts of pay to their CEOs.
(9) See Jensen and Murphy (1990); Gibbons and Murphy (1992); and
Jenter and Kanaan (forthcoming).
(10) Wang (1997) fleshes out this explanation using a repeated
moral hazard model.
(11) See Bolton, Sheinkman, and Xiong (2006); Clementi, Cooley, and
Wang (2006); and Edmans and Liu (2011).
(12) See Hall and Murphy (2002) for a quantitative evaluation of
the difference between the executive's value of options and the
cost to the firm in providing them.
(13) Larcker, McCall, and Tayan (2011) also present a third measure
that they call earned pay (the value of pay at the moment when all
selling restrictions are lifted, which does not necessarily coincide
with the value at the time the CEO decides to sell). We do not have
enough information in Execucomp to calculate this measure.
(14) Note that the utility the CEO may get from a given value of
employment will also depend on his wealth from sources other than the
executive's employment. There is typically no information on this
outside wealth to be used in empirical studies of CEO compensation.
(15) Note that dividends are not included in Execucomp's TDC1
(which we will compare later to our own proposed measure of income). We
include them because they are attached to the grants given to the CEO,
and hence they are income that he receives because of his association
with the firm.
(16) Here and in the rest of the model description, we use capital
letters to denote values and lowercase letters to denote quantities.
(17) Note that option grants also come with expiration dates; we
are abstracting from those in this discussion, since the information we
have on expirations is limited.
(18) Examples of different implementations of this concept of
expected pay include Jensen and Murphy (1990); Garen (1994); Haubrich
(1994); Hall and Liebman (1998); Haubrich and Popova (1998); Schaefer
(1998); Aggarwal and Samwick (1999); Baker and Hall (2004); Clementi and
Cooley (2009); Edmans, Gabaix, and Landier (2009); and Gayle and Miller
(2009).
(19) See, for example, Clementi and Cooley (2009; 2, 29).
(20) Note that [G.sup.t-1.sub.[tau]] = 0 whenever [tau] > t.
Also, note that this re-evaluation of grants coincides conceptually with
our measure of gains from trade, for the portion of the vested portfolio
that is converted to cash in period t. That is, if, for example, only
grants given at t - 4 are exercised at t, then [T.sub.t] ([S.sub.t-1],
[O.sub.t-1]) = [G.sup.t.sub.t-4].
(21) The database includes up to five executives of a firm per
year, but we restrict our sample to those designated as the CEO by the
Execucomp variable CEOANN.
(22) We also exclude from our analysis Warren Buffett, the CEO of
Berkshire Hathaway, and Larry Ellison, the CEO of Oracle Corporation,
because their values of trades are extreme outliers.
(23) We replicate Clementi and Cooley's simpler calculation of
TYC, which uses intrinsic valuations for options when their value is
updated with new stock prices at the end of the fiscal year. Clementi
and Cooley report in their manuscript that their results do not change
substantially when they use Black and Scholes to produce those
revaluations.
(24) This subsample includes 69 out of our 45 firms, and 102 out of
our 34,497 observations.
(25) Firms in the finance sector are those with SIC classification
in the 6,000-6,300 range. There are 144 firms per year, on average, in
our subsample of finance. We performed the same analysis with a broader
category including real estate firms as well as insurance, and the plots
looked qualitatively similar.
(26) Given the limited quantitative importance of bonuses in total
compensation, we will ignore changes in bonus payments in our
sensitivity analyses.
(27) Given the way we construct the counterfactuals, any
differences in level between Tables 1 and 3 is due to the original
differences in the level of actual compensation between the average
finance firm and the average firm in the sample.
(28) Specifically, after 2005 Execucomp's BONUS variable was
modified to only include discretionary or guaranteed bonuses. So to
include payments from objective targets, we sum BONUS with NONEQ_INCENT,
the amount of income received in the year pursuant to non-equity
incentive plans being satisfied. Whenever NONEQ_INCENT is missing (i.e.,
prior to 2006), we add BONUS with LTIP, the amount of income received in
the year pursuant to long-term incentive plans that measure performance
over more than one year.
(29) Both variables measure the value of stock awards as of the
grant date. RSTKGRNT was reported by the companies themselves in the
Summary Compensation Table, while STOCK_AWARDS_FV is calculated by
Execucomp. Strictly speaking, each also contains restricted stock units
and phantom stocks.
(30) OPTION_AWARDS_BLK_VALUE is calculated by Compustat, during
that period of time when--prior to FAS 123R--companies typically
expensed options using the "instrinsic value" method, i.e.,
the difference between grant date stock price and exercise price of the
option, which nearly always led to no expensing of options.
OPTION_AWARDS_FV is the grant date fair value of option awards in the
year, reported by the company per FAS 123R using some version of Black
and Scholes (1973) or a similarly accepted calculation.
(31) OPT_EXER_VAL is the total value realized from option exercises
in the year, and is measured (for each g award, in our notation) as the
difference between the exercise price and stock price on the date of
exercise.
(32) In addition to what we have described, there are two other
types of transactions that will change CEO holdings: stock inheritances
and stock donations. We abstract from them, as these transactions will
typically be small, if non-zero. However, these could also be behind
some of the negative qt in the data.
(33) SHROWN_EXCL_OPTS reports shares of the firm owned by the CEO,
excluding options that are exercisable or will become so within 60 days.
This amount is reported as of some date between the fiscal year-end and
proxy publication.
(34) Clementi and Cooley (2009) use the end-of-the-fiscal-year
price for this calculation. We choose average price hoping to avoid some
of the idiosyncrasy of [p.sub.t] due to volatility of stocks.
Table 1 Summary of Annual Compensation Information Available in
Execucomp
Instrument (Average % of TDC1) Information in Execucomp
Salary (32%) Value
Bonus and Incentive Value, some details on
Compensation (23%) targets (after 2006)
Perks and Other Compensation (6%) Value
Restricted Stock Grants (11%) Value (stock price times
number of shares)
Stock Option Grants (28%) Value (Black and Scholes),
number of shares underlying
options
Notes: Information available in Execucomp about the components of
CEO compensation packages. For the percent calculations, the
sample includes the CEOs of the largest 1,500 public firms in the
United States in the period 1993-2010.
Table 2 Classification of Compensation Instruments
Current (within year) Deferred
Non-Contingent Salary, perks, Pension plan
signing bonus
Contingent Bonus plan Options, stock, severance,
future pay
Table 3 Summary of Information Available in Execucomp about Stock
and Option Holdings
Information in Execucomp
Stock Holdings Number of unrestricted
Number of restricted
Value of restricted
Number vested during the year
Value of vested during the year
Option Holdings Number exercised during the year
Value of exercised during the year
Number of all unexercised vested
Value of in-the-money unexercised
vested (intrinsic)
Number of all restricted
Value of restricted in-the-money
(intrinsic)
Table 4 Counterfactual Income: Mean Level of Income if Certain
Percentile Stock Return Had Been Achieved--All Firms
Year 5th 25th Median 75th 95th Actual
1994 4,301 5,630 6,378 7,203 8,821 6,182
1995 3,438 4,296 4,926 5,562 6,832 4,659
1996 3,694 4,923 5,734 6,554 8,157 5,286
1997 5,291 7,231 9,083 10,809 14,289 8,243
1998 5,246 9,150 12,397 15,911 24,021 9,364
1999 4,775 7,060 8,668 11,217 20,479 8,460
2000 10,618 27,698 41,383 55,636 87,588 11,268
2001 6,859 11,909 14,968 18,089 27,074 11,022
2002 4,439 6,956 8,910 10,472 13,455 7,448
2003 9,840 12,994 14,718 17,314 26,321 14,156
2004 7,917 11,463 13,211 15,242 20,491 13,023
2005 10,266 14,414 17,072 20,330 26,911 16,382
2006 8,883 11,849 13,802 15,503 19,754 13,200
2007 6,991 10,090 12,326 14,769 19,817 12,531
2008 5,076 8,934 12,177 15,324 20,812 9,636
2009 8,078 11,230 13,450 17,288 28,866 14,474
2010 6,899 8,447 9,459 10,749 13,825 9,640
2011 6,443 8,809 10,325 11,913 15,215 11,270
2012 7,996 10,718 12,080 13,623 18,149 12,747
Table 5 Counterfactual Income: Mean Level of Income if Certain
Percentile Stock Return Had Been Achieved-Finance Firms Only
Year 5th 25th Median 75th 95th Actual
1994 3,687 3,987 4,190 4,414 4,851 4,228
1995 5,256 5,700 6,072 6,488 7,679 7,047
1996 6,515 9,468 11,628 13,792 18,017 9,881
1997 6,656 8,323 11,956 15,389 22,304 12,473
1998 5,664 8,173 10,442 12,902 18,538 9,452
1999 6,341 8,800 10,453 12,996 22,127 9,485
2000 6,189 10,105 16,523 24,242 41,568 12,604
2001 8,507 13,528 16,610 19,757 28,852 14,333
2002 6,345 9,544 12,237 14,405 18,536 10,947
2003 8,898 11,051 12,293 14,159 20,626 12,467
2004 8,933 13,099 15,118 17,452 23,474 13,744
2005 11,591 15,828 18,226 21,159 27,077 17,473
2006 9,153 12,596 14,916 16,925 21,937 13,284
2007 10,455 14,958 18,105 21,476 28,427 14,184
2008 4,957 8,523 11,570 14,561 19,803 7,918
2009 5,600 7,378 8,645 10,847 17,502 8,275
2010 5,291 6,322 7,025 7,910 10,125 6,621
2011 4,314 5,350 6,024 6,719 8,152 8,980
2012 5,535 7,174 8,023 8,983 11,791 7,594
Table 6 Counterfactual Income: Mean Percent Change in Income if
Certain Percentile Stock Return Had Been Achieved-All Firms
Year 5th 25th Median 75th 95th
1994 -9.2 -2.2 2.1 6.8 16.4
1995 -11.5 -3.5 2.7 9.0 21.4
1996 -12.1 -1.6 5.4 12.5 26.5
1997 -15.0 -3.7 6.6 16.3 35.9
1998 -16.4 -1.7 11.2 25.5 62.6
1999 -16.6 -5.0 3.4 16.8 66.1
2000 -16.5 13.7 38.8 65.3 125.3
2001 -17.6 1.4 13.7 26.5 63.6
2002 -17.2 -1.5 11.5 22.1 42.4
2003 -13.8 -4.3 1.1 9.4 38.3
2004 -18.6 -3.7 4.2 13.4 37.5
2005 -18.7 -4.3 31.8 65.6 133.9
2006 -18.4 -6.4 1.7 8.9 26.9
2007 -20.8 -8.2 1.3 11.8 33.5
2008 -19.7 -1.9 14.0 29.9 57.8
2009 -15.3 -6.1 0.8 13.1 50.8
2010 -14.8 -6.2 -0.4 7.1 25.2
2011 -18.6 -5.8 2.8 12.0 31.3
2012 -18.2 -2.8 5.5 14.8 42.1
Table 7 Counterfactual Income: Mean Percent Change in Income if
Certain Percentile Stock Return Had Been Achieved--Finance Firms
Only
Year 5th 25th Median 75th 95th
1994 -7.7 -2.6 0.9 4.8 12.5
1995 -13.3 -8.9 -5.2 -1.1 7.5
1996 -17.8 -8.2 -1.2 6.3 21.0
1997 -22.1 -11.9 -2.2 7.2 26.1
1998 -21.9 -1.6 15.6 34.2 76.9
1999 -16.0 -0.5 10.1 26.6 85.6
2000 -21.0 -3.2 13.5 31.6 72.8
2001 -20.4 0.6 14.4 29.0 71.9
2002 -21.8 -8.1 5.7 17.0 38.5
2003 -13.7 -5.0 0.1 7.9 34.8
2004 -18.3 -1.1 7.9 18.3 45.1
2005 -13.6 -1.9 5.4 14.5 32.9
2006 -18.9 -7.8 -0.2 6.4 22.8
2007 -13.7 -0.2 10.0 20.9 43.4
2008 -12.7 6.8 24.3 42.5 75.4
2009 -7.8 2.6 10.4 24.1 65.5
2010 -10.2 -3.4 1.2 7.1 23.8
2011 -13.5 -4.9 1.1 7.3 20.2
2012 -11.1 0.9 7.6 15.2 37.5
