Regulation and the composition of CEO pay.
Jarque, Arantxa ; Gaines, Brian
It is well known that the use of stock options for compensating
executives in large U.S. companies was widespread during the last 15
years. But were all firms using them with equal intensity? We are
interested in the answer to this question because option grants are
different from other compensation instruments in the type of incentives
they provide, how transparent they are to investors, and the level of
insider trading that they allow. In this article, we provide an
empirical examination of the trends in the last two decades of the use
of different compensation instruments, mainly focusing on restricted
stock grants and option grants. We find that there have been important
changes, and that they coincide in time with two changes in regulation:
the modifications to reporting requirements for option grants introduced
by the passage of the Sarbanes-Oxley Act in 2002, and the 2006 adoption
of revised accounting standards from the Financial Accounting Standards
Board (FASB) included in statement no. 123R (FAS 123R), which mandated
the expensing of option grants.
Today, companies pay their top executives through some or all of
the following instruments: a salary, a bonus program, stock grants
(usually 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 retirement and severance payments, as well as pension
plans and deferred benefits. 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 chief executive officer (CEO) that align his
interests with those of the firm owners.
In the presence of moral hazard, the optimal contract prescribes
that the pay of the executive should vary with the results of the firm.
However, in spite of the need for incentives, limited funds on the part
of CEOs or risk aversion considerations imply that exposing the CEO to
the same risk as shareholders is typically either an unfeasible or an
inefficient arrangement. The optimal contract should balance incentives
and insurance. Some part of the compensation should not be subject to
risk, like the annual salary, providing some insurance to the CEO
against bad performance of the firm over which he does not have control.
However, some part of the compensation should be variable and tied to
some measure of performance of the firm. The main variable pay
instruments can be classified in three categories. First, bonus plans,
which make annual pay dependent on yearly accounting results. Second,
grants of stock of the firm (often referred to as "restricted
stock," since the executive cannot sell them for some time after
they are granted, typically about three or four years); these make pay
in the longer term dependent on the results of the firm over a longer
time horizon. Third, grants of stock options, which allow the executive
to purchase stock of the firm at a pre-established price (the
"exercise price") and also typically are granted with
restrictions as to how soon they can be exercised; these also provide
incentives for longer-term performance, but they only pay off for the
executive if the stock price of the firm is above the exercise price.
(1)
These different compensation instruments differ in how transparent
they make compensation to shareholders or outside investors. For
example, bonus schemes that are based on both objective and subjective
performance targets may be more difficult for an outside investor to
evaluate than a plain restricted stock grant. These instruments also
differ on how robust they are to insider trading and other opportunistic
behavior; the exercise of stock options or the sales of vested stock can
potentially be timed by the CEO to the disclosure of particularly good
or bad news on the prospects of the firm, for example, and bonuses may
be sensitive to creative accounting practices where some annual results
are made to look better by using the degree of freedom present in
accounting standards, or by fraudulent misrepresentation of financial
results.
Another important factor is that various compensation instruments
are treated differently for taxation purposes and are subject to
different disclosure requirements and accounting standards. As an
example of heterogeneity in tax treatment, non-qualified option grants,
which have been the most popular type of option grant in the last two
decades, trigger a tax deduction for the company when they are exercised
by an employee; salaries or any other compensation that is not
performance-based (like plain restricted stock awards) in excess of one
million dollars, instead, do not qualify for a deduction. (2)
As an example of heterogeneity in disclosure requirements,
compensation that is given to executives in the form of perks does not
need to be detailed in the compensation disclosure tables of proxy
statements if its value is less than $10,000; when the value exceeds
that sum, the disclosure is only in a footnote. Salary, bonus, stock,
and option grants are disclosed in the mandatory compensation table
instead. (3)
These differences have historical origins, and are likely subject
to political pressures. One cannot ignore, however, the distortion that
the tax, disclosure, or accounting treatment may potentially have on the
choice of instruments, and through that--as we just argued--on the
efficiency of incentives and the transparency of compensation practices
to shareholders. Hence, in this article we ask the following questions:
Are firms in certain industries or of larger size more likely to use
option grants? Are firms that use options more likely to pay higher
compensation to their CEOs? Have restricted stock grants replaced option
grants after expensing and reporting rules were changed in 2002 with
Sarbanes-Oxley, and then again in 2006 with the adoption of Statement of
Financial Accounting Standards 123R (SFAS 123R) accounting standards?
Has the relative importance of salary and bonuses decreased in recent
years in favor of option or stock grants?
Regulatory Changes
In this article we consider two major changes in the U.S.
regulation of compensation practices. The first change is the
Sarbanes-Oxley Act of 2002, which aimed to improve corporate governance
after several earnings management scandals surfaced in the early 2000s.
The second change is a revision of accounting rules introduced by the
SFAS 123R in 2006, which for the first time mandated a positive expense
for options awarded "at the money" (with exercise price equal
to the stock market price at the date of the grant).
We use available data on executive compensation from 1993 to 2010
to evaluate the effect of these two regulation changes in the choice of
compensation instruments of large public U.S. firms. Note that the
Dodd-Frank Act, which was motivated by the financial crises of 2008, was
passed in 2010 and it affected financial firms only. It would certainly
be interesting to know how the increased scrutiny of incentive schemes
that the Act mandates (both for executives and lower level employees) is
affecting pay practices at large financial institutions. However, we do
not have enough observations in our sample to deal with that regulatory
change in this article.
The Sarbanes-Oxley Act, as part of its effort to improve
transparency to shareholders, decreased the time window allowed for the
disclosure of insider trades to two business days. (4) Before the Act,
firms had until the end of the fiscal year to report any type of insider
trading, including the grants of options to employees of the firm. As it
became apparent after some investigations, a number of firms were able
to exploit the lax reporting requirements to engage in
"backdating," the (illegal) practice of artificially changing
the grant date of options to the day with the lowest stock price in the
time window allowed for reporting. (5), (6) The benefits of this
practice were twofold, and hinged on both the accounting standards and
the tax treatment of options. First, it allowed the firm to report
higher earnings. At the time, accounting standards under SFAS 123
allowed firms to expense grant options according to their
"intrinsic value," which is zero for options granted at the
money. Instead, the intrinsic value of an option in the money (which is
what was being effectively granted without the backdating) would have
been positive, and hence a compensation expense would have been deducted
from the firm's income, resulting in lower reported earnings.
Second, it allowed a larger tax deduction for the firm at the time that
the option was exercised. Under Internal Revenue Code (IRC) section
162(m), firms can deduct from their tax liability any compensation costs
that originate in incentive pay. In contrast, there is a limit of one
million dollars for deducting compensation that is not tied to
incentives. Hence (provided the employee was already receiving one
million dollars in non-incentive compensation), the tax deduction would
have been lower for an option in the money, since the difference between
the stock price at the time of grant and the exercise price would not
have been considered incentive pay. (7) Backdating options without
proper disclosure, then, implied both misreporting to investors the
amount of incentive pay given to employees, and engaging in fraudulent
accounting to save on taxes. (8) The Act, by decreasing the time window
allowed to report the granting of options to two business days after the
trade takes place, constrained the firms' ability to misreport the
actual date of the grant, and hence made options a less attractive
compensation instrument for firms that were backdating, or for those
that were considering the possibility of doing it at some point.
The second piece of regulation that we consider is SFAS 123R, a
revision to accounting standards SFAS 123, which was adopted by the
Security and Exchange Commission (SEC) in 2006. The main change
introduced by the revision was a homogenized method of valuation of
options to "fair value" calculations, such as Black and
Scholes. Previously, the "intrinsic valuation" method was
allowed, which attributed a zero value to options granted at the money.
Because option grants are accounted for as expenses in the income
statement of the firm, this change in valuation method effectively
eliminated the possibility of not charging any expense of compensation
for options granted at the money. (9) The general view on this piece of
regulation is that, after its adoption, companies were no longer able to
"hide" the dent of option grants on their accounting profit.
This view is supported by the numerous complaints by large U.S.
corporations when the measure was first proposed, arguing that lower
earnings per share would hurt, for example, their ability to borrow and
grow, hindering innovation and job creation. However, under the
disclosure requirements in SFAS 123 before 2006, firms were already
required to report (in a footnote in their proxy statement) enough
information about their grants of employee options for any interested
shareholder to compute the cost of these (using, for example, the Black
and Scholes valuation). Hence, the economic impact of this change in
regulation remains unclear, and it somehow hinges on the assumption that
the information disclosed in the footnotes was somewhat less available
to the public than after it was officially included as an expense in the
income statement. (10)
It is important to note that the first proposal for the expensing
of options was drafted as far back as 1993. Due to strong opposition
from the corporate sector and other political forces, the final
recommendations in FASB 123 issued in 1995 merely recommended the
expense, but did not mandate it. The public debate about the pros and
cons of expensing, which involved senators, congressmen, the SEC, and
lobbyists from the corporate sector, was ongoing for more than a decade.
Finally, in 2006, the SEC endorsed the revision SFAS 123R, which
mandates expensing. It is worth noting that many large public firms
started the expensing on a volunteer basis as early as 2002; some
commentators have noted that this voluntary adhesion, and the final
political push that lead to the mandatory requirement, were rooted in
the Enron and other accounting scandals in 2002. (11) Hence, the effect
of SFAS 123R is potentially present as early as the passage of the
Sarbanes-Oxley Act, preventing the separate identification in the data
of the effect of the two regulations. Nevertheless, in our analysis we
find significant changes in the patterns of usage of stock and option
grants coinciding with both changes in regulation.
Outline
In this article, we start by describing the data. We provide a
motivating example that illustrates the primary difficulties in using
the currently available data on CEO compensation to answer the main
questions of interest to us. In Section 2 we briefly review some
previous attempts in the academic literature to shed light on similar
issues, and the differences with the approach we take here. We proceed
with our main analysis in two parts: First, in Section 3, we document
facts related to the extensive margin (i.e., when and by which firms are
stock and option grants used), and second, in Section 4, we discuss
facts related to the intensive margin (i.e., what is the relative
importance of stock, options, and other forms of pay for the firms that
use them). We document the change in compensation practices across the
different regulatory regimes. We also explore the correlation of other
firm characteristics, like size, industry classification, and executive
characteristics, like age, tenure, and gender, with the choice and
importance of the different available compensation instruments. We also
examine the relationship of usage of stock and option grants with the
level of pay. We conclude in Section 5.
1. SAMPLE DESCRIPTION AND DATA INTERPRETATION ISSUES
Thanks to disclosure requirements by the SEC, we have data
available on pay to the top executives of public U.S. companies starting
in 1992. This data is collected systematically by Compustat into a
database called Execucomp. Many academic studies have used Execucomp and
other available data to document the regularities in the level of pay
and its sensitivity to firm performance, across time and also for firm
characteristics like size and industry. (12)
The Execucomp data set is published by Compustat four times per
year. Each release includes the new information for companies that filed
their proxy statements with the SEC in that period (companies can decide
when their fiscal years start, and hence there is variation in when
annual proxies are filed). Execucomp tries to collect data on the firms
that are listed in the S&P 1500 index, which roughly corresponds to
the 1,500 largest U.S. firms by market capitalization. This article uses
the information on CEO pay of the October 2011 edition of the Execucomp
data, which covers 1992 to 2010, for a total of 19 complete fiscal
years. We exclude observations in year 1992, since there are very few
and they may not be representative. We exclude CEOs who own a large
fraction of the firm's stock, since presumably pay is not set to
provide incentives for these owner-CEOs. Next, we elaborate on the
issues in choosing the threshold value for this selection.
Ownership, Incentives, and Steve Jobs
In this article we are interested in the decisions of firms to use
or not use a given compensation instrument. One potential concern with
this analysis is that the choice of a firm of not using stock or options
may be explained by the fact that its CEO is a founder of the company,
or that he or she is very vested in the firm already. An example of this
would be Steve Jobs, who is, in our sample from 1997 to 2010, listed as
the CEO of Apple, Inc.
Jobs's history of compensation over 12 years is easily
summarized. In 1997, the year he took the CEO position, Jobs received,
as a director of the company, 30,000 stock options with an exercise
price of $23, to be vested proportionally over a three-year period.13
The salary of Jobs was $1 for all the years we observe him in the
sample. He received sporadic bonus and "other compensation"
payments, stock in 2003, and options in 1997, 2000, and 2002. (14), (15)
In the company's own words:
"In 2010, Mr. Jobs's compensation consisted of a $1 annual salary.
Mr. Jobs owns approximately 5.5 million shares of the Company's
common stock. Since rejoining the Company in 1997, Mr. Jobs has not
sold any of his shares of the Company's stock. Mr. Jobs holds no
unvested equity awards. The Company recognizes that Mr. Jobs's level
of stock ownership significantly aligns his interests with
shareholders' interests. From time to time, the Compensation
Committee may consider additional compensation arrangements for Mr.
Jobs given his continuing contributions and leadership." (16)
Jobs's ownership shares are only reported in Execucomp,
combined with option holdings, for four of the years, and they never
exceed 1.35 percent of the total shares outstanding, which is about the
67th percentile ownership in the original sample of CEOs. Even if one
may be tempted to think that Jobs was not an "agent" for the
shareholders of Apple due to the great value of the stock that he owned
(especially after the 2003 grant, valued at more than $80 billion at the
grant date), a closer look at the evolution of his ownership shows that
he went from owning one share in 1997 to owning 5.5 million shares
mainly as a result of his compensation packages. Moreover, Jobs had a
considerable amount of wealth from his investment in Pixar, and one
could argue that stakes had to be necessarily high in order to provide
him with adequate incentives. Finally, when Jobs's illness was made
public, markets reacted, providing proof that the value that Jobs was
bringing to the company was real.
The case of Steve Jobs is easy to check and understand, but in
general the data on ownership in Execucomp shows some inconsistencies,
and there are many missing values, since ownership is recorded only if
it is over 1 percent. Hence, a back-of-the-envelope calculation of the
value of the stock held by the CEO is not always available, and even if
it were it would be hard to determine when the CEO is subject to a moral
hazard problem based on those numbers. (17) From our analysis of the
Jobs case, however, we conclude that we cannot rule out that ownership,
in our sample, is a result of dynamic incentives provided by the firm.
Hence, we are most comfortable adopting a conservative criteria of only
dropping CEOs from our sample if their ownership reaches 50 percent in
any of the years that they worked for a given firm, as opposed to more
restrictive selection criteria in the literature. (18) Our final sample
includes information on 6,146 different executives, and 3,248 firms,
which amounts to 6,416 unique executive-firm pairs. In the year 1993, we
observe 1,147 firms, and every year after that the number is at least
1,500, with a maximum of 2,010 firms in the year 2007.
Compensation Measures
Our focus in this article is on the choice of compensation
instruments by the firm, and we use the information readily available in
Execucomp about each of the components of total compensation: salary,
bonus and incentive compensation, stock and option grants, and
"other compensation" such as pension plans, life insurance
premiums, or perks. Note that to avoid discontinuity issues with the
"bonus" and "incentive compensation" variables due
to changes in reporting requirements in 2006, we sum these two to
construct a single series, which we refer to as BIC throughout the
article. Also, in spite of the different accounting standards during the
sample period, Execucomp contains the Black and Scholes valuation of
option grants for the whole period: Companies that used alternative
valuations prior to SFAS 123R were required to provide the parameters
necessary to calculate the Black and Scholes value. Whenever we need a
measure of total compensation, we use the sum of these components (the
variable TDC1 in Execucomp). (19)
Figure 1 presents the evolution of the mean total compensation in
our sample over time, and its components. All the amounts here and in
the rest of the article are normalized to thousands of 2010 dollars
using the consumer price index. The year 2000 stands out as the peak in
our measure of compensation, with an average of $8,553,690 and a median
of $3,107,580. The year 2009 seems to be the last one of a decreasing
compensation trend coinciding with the financial crisis: Mean
compensation for this year was $4,637,950, while the median was
$3,030,940.
[FIGURE 1 OMITTED]
The most salient fact about the composition of pay in Figure 1 is
that the variation of pay with the business cycle is implemented through
the grants of stock and options, rather than through salary, bonus, or
other compensation. For example, the graph shows that the decline in
average total compensation between 2000 and 2003 is driven by a decline
in the value of stock options. However, after 2002, the category BIC
becomes somewhat cyclical as well. It is important to keep in mind that,
of these components of total compensation, only bonus and incentive
payments are mechanically related to the results of the firm. For
example, the amount used to construct Figure 1 is the expected value of
the grant at the time when it was awarded. Hence, the fact that
compensation was the highest in the year 2000 is not due to a high value
of past grants driven by a stock market boom, but rather to a conscious
decision by the firms to increase the value of compensation for their
CEOs. (20)
2. PREVIOUS LITERATURE
Before we start our analysis of the data, we review the relevant
literature and explain our contribution.
In an influential chapter of the Handbook of Labor Economics,
Murphy (1999) provides some suggestive evidence for a sample of firms
between 1992 and 1996 that the importance of the different compensation
instruments in pay packages (salary, bonus, stock, and option grants)
varies across firms according to their size and the industry to which
they belong. (21)
In his graphical analysis for the effect of size, Murphy compares
S&P 500 industrials, mid-cap industrials, and small-cap industrials.
We replicate and extend his analysis (including data up to 2010) in
Figure 2, where we classify firms in our sample, year by year, into four
quantiles according to their volume of sales.
[FIGURE 2 OMITTED]
The most striking fact that emerges from Figure 2 is that firms
with larger sales figures have higher levels of pay. The variation in
the relative importance of the different compensation instruments is
difficult to evaluate in a systematic manner, although it is clear that
larger firms have a larger portion of their pay given in stock and
options. Also, the increase in the relative importance of options in the
late 1990s that has been frequently commented on both in the academic
and the popular press seems to have been disproportionately concentrated
in the quantile of the largest firms.
Murphy's graphical analysis for the regularities across
different industries is limited to S&P 500 firms, and it uses a
classification of SIC industries in four groups: mining and
manufacturing firms, financial services firms, utilities firms, and
other industries. In Figure 3 we replicate this evidence, again
extending the sample to include data from 1992 to 2010, as well as all
firms in the S&P 1500. Figure 3 does not allow us to draw any clear
conclusions. If anything, it seems to suggest that firms in utilities
seem to rely more on restricted stock than option grants. In a related
study, and for the period 1992-2001, Murphy (2003) classifies firms into
"new economy" versus "old economy" according to the
industry sector they belong to, and he finds that new economy firms
(those competing in the computer, software, inter-net,
telecommunications, or networking fields) use stock-based compensation
(both restricted stock and options) more often and to a larger extent.
[FIGURE 3 OMITTED]
One important shortcoming of the simple facts reported in Murphy
(1999, 2003) is that they do not inform us about the relationship
between combinations of individual characteristics (industry and size
together, for example) and usage of instruments. Also, the information
about the variation in the cross-section is lost in the graphs. Our
contribution in this article consists of analyzing the data according to
firm characteristics by running some simple regressions. Our analysis is
still partial, since we are not exploiting the panel component in the
data, but we are able to provide a more accurate description of the
facts by controlling for several individual firm characteristics. We
also split our analysis into the extensive margin (which compensation
instruments are used) and the intensive margin (given a set of
instruments that is being used, what is their individual share of total
compensation) .
In addition to answering the questions posed above about the trends
in the usage of different compensation instruments, we explore whether
factors other than firm size or industry classification may be
associated with the usage of certain instruments. For example, given the
limits on tax deductions imposed on salaries, firms that--for reasons
other than their industry and size--choose to compensate their CEO with
a larger sum of money may benefit more from issuing non-qualified option
grants or restricted stock grants. As another example, executives who
have longer tenures may need fewer restricted stock grants if they
already hold a large number of shares of the firm from previous grants.
This last point, which is an interesting one, refers to the dynamic
nature of incentives for CEOs. There have been important efforts in the
literature of CEO compensation that track the evolution of the portfolio
of grants of the executives, so that at each point in time we have a
better understanding of how the executive's wealth would vary with
a particular realization of the firm's results. Some important
examples are Hall and Liebman (1998), Core and Guay (1999), and, more
recently, Clementi and Cooley (2010). These measures of incentives are a
way of controlling for outstanding past issues of stock and option
grants. The focus of these studies, however, has not generally been the
trends in the usage of compensation instruments. An important exception
is Core and Guay (1999), who study this in detail for a shorter time
period than the one we are analyzing here. They construct a model of the
optimal level of stock holdings of the CEO, for incentives purposes.
They find evidence that new grants (combining stock and options) are
aimed at maintaining that level of incentives, as old grants expire or
go out of the money. However, as far as we know, none of the studies
that construct the portfolio measures address the potential effects of
regulation on the trends in the usage of individual compensation
instruments.
One important shortcoming of our data set is that it starts in
1993. Regulations on tax deductibility of CEO pay had just changed at
the time (see IRC section 162(m)). Data on compensation practices prior
to 1993 would be useful to the understanding of the distortions that
162(m), and other tax advantages introduced earlier, may have induced on
pay practices. (22) Detailed compensation data for a broad
representative set of firms going further back in time is not available;
however, Frydman and Saks (2010) provide a historical analysis of a
limited set of firms. (23)
As part of their analysis, Frydman and Saks (2010) plot the median
of the partial sums of salary and bonus payments, successively adding
the value of stock and option grants. They find that, even though the
usage of options picks up considerably after taxation advantages are
introduced in 1950, their relative importance in total compensation, as
well as that of stock grants, does not become significant until the
1980s. (24) Since their sample of firms is necessarily limited (because
of the long historical scope), and for comparison purposes, we replicate
their graphical analysis for our sample in Figure 4. (25)
[FIGURE 4 OMITTED]
For the overlapping period from 1992 to 2005, and again with the
caveat of not controlling for individual characteristics in this simple
graphical analysis, we confirm their findings: Option grants have been
an increasingly important component of the median pay of CEOs for the
whole period, while the importance of stock awards started to pick up
around 2002. With respect to the mean compensation that we plotted in
Figure 1, we see that the importance of options was not as marked for
median pay in the 1999-2001 period as it was for mean pay. Other than
that, the main patterns seem to align between the two figures.
3. THE COMPOSITION OF PAY PACKAGES: THE EXTENSIVE MARGIN
We start this section by documenting the usage of the different
compensation instruments over time. Then we proceed to analyze more
formally which firm characteristics may be relevant for the choice of
instruments of compensation. We find that variables like size and
industry classification have some explanatory power over whether firms
decide to include options or stock in their compensation packages.
Changes in regulation during the period we are studying exhibit the
highest correlation with changes in usage patterns.
The Use of Different Compensation Instruments: A First Look
For all the firms in our sample, we check year by year which ones
use each instrument (for example, a firm "uses" stock if it
reports a positive stock grant to their CEO, regardless of the amount of
the grant). This is plotted in Figure 5.
[FIGURE 5 OMITTED]
As is apparent from the graph, the use of both salary and other
compensation is fairly universal and fairly constant over time (with a
slight trend up for other compensation in the last five years). The use
of bonus and incentive compensation is volatile around 85 percent, with
no obvious trends. But the most striking feature in Figure 5 is the
run-up in the use of restricted stock grants starting around 2003, which
coincides with an important decrease in the use of option grants.
Given the strong variation over time in the usage of stock and
options, it is worth thinking about the factors that could potentially
be determining the decision of a firm to include either type of grant in
the compensation package to its CEO. Here we point to three main
factors: (1) differences in tax advantages and accounting standards, (2)
differences in sensitivity to firm performance, and (3) fixed costs of
adoption of each instrument.
First we turn to tax and expensing differences. As we discuss at
length in the introduction, both the passage of the Sarbanes-Oxley Act
in 2002, and, especially, the change in expensing requirements and
valuation of options in SFAS 123R approved in 2006 (and expected and
voluntarily adopted by many firms as early as 2002), seem to have
decreased the relative attractiveness of option grants over stock
grants. We can summarize the comparison between the two instruments as
follows. Restricted stock grants do not qualify for a tax deduction,
they have to be accounted for as compensation expenses, and, before
2002, they had to be reported as insider transactions within 10 days of
the grant. Options were more advantageous than stock before 2002 because
they only had to be reported as insider transactions by the end of the
fiscal year of the company; after Sarbanes-Oxley, both types of grants
have to be reported within two business days of the transaction. Options
were more advantageous than stock before 2006 because (i) they could be
deducted for tax purposes, and (ii) they did not need to be expensed;
after 2006, advantage (i) is still present, but (ii) is no longer there.
Second, stock and options may implement different incentives for
the CEO. That is, in principle, without any accounting or tax
differential treatment, stock and options could be substitutes in a
compensation package: One could transfer a given amount of resources to
the CEO either with a stock grant or with an option grant of equal
expected value. However, the value of each of these two grants could
change differently with changes in the value of the firm, i.e., the
sensitivity of the compensation may be different depending on whether it
includes only options or only stock (or both). Hence, idiosyncratic
characteristics of the firm, like industry, size, or financial health,
may determine the optimal sensitivity of pay to performance, and hence
instrument choice.
Third, there may be a fixed cost of including an extra instrument
in a compensation package (perhaps related to communication of new or
more complex compensation practices to shareholders and creditors); this
would imply that larger firms decide to include a different set of
compensation instruments than their smaller counterparts.
To shed some light on these and other potential hypotheses, we will
formally analyze the correlation of different firm characteristics on
the choice of compensation instruments. We start our analysis of the
data by classifying firms into four mutually exclusive groups, I,
according
to which set of compensation instruments they use:
T = {S, O, B, N},
with typical element I. That is,
* a firm with I = S includes restricted stock grants (but no
options) in its compensation package to the CEO,
* a firm with I =0 includes options (but no stock),
* a firm with I = B includes both restricted stock and options, and
* a firm with I = N includes none of the two.
Figure 6 presents the evolution of the proportion of firms in each
of these four groups over our sample period. The evidence is consistent
with the changes in regulation prompting firms to switch from using
options only (0) to using stock only (S); but it is also apparent that a
higher portion of firms use both instruments (B), suggesting that some
firms may have chosen to add stock to the use of options, rather than
completely substituting options with stock. Next, we formally evaluate
the role of changes in regulation in these variations in usage patterns,
after also considering other potential determinant factors for these
patterns, such as the size of the firm and the industry to which it
belongs.
[FIGURE 6 OMITTED]
The Determinants of the Composition of Pay Packages
Table 1 presents the breakup of firms in the compensation groups in
I according to the regulatory periods, the industry group, and several
firm and CEO characteristics available in Execucomp. It also includes
statistics that describe the cross relations between these variables. We
have established the existence of three different subperiods in our
sample determined by important changes in regulation. Table 1 reports,
in its first three rows, the fraction of firms that choose each
instrument in the sample, and the changes in these fractions after the
two changes in regulation. We denote as period I observations those from
1993 to 2001, as period II those from 2002 to 2005, and as period III
those from 2006 to 2010. We see in the table that the fraction of firms
in S increases in both subperiods, but especially in the later one. The
fraction of firms in 0 options decreases, again more sharply in the
later subperiod. The fraction of firms in N remains fairly constant over
time around its overall mean of 22 percent. As we explained in the
previous subsection, both regulations had the effect of decreasing the
relative attractiveness of options over stock grants. Given this, it is
useful to trace the changes in the fraction of firms that use options at
all, i.e., 0 [union] B. In period I, we see that 73 percent of firms had
options in their compensation packages. In period II this fraction
remains fairly constant, at 71 percent: The decrease in 0 is almost
exactly offset by the increase in B. That is, in the second subperiod
firms were more likely to use options together with stock, rather than
alone, but still as likely as before to use options at all. However, in
the last subperiod the fraction drastically decreases to 54 percent:
Although the fraction of firms in B increases, the decrease in 0 is
three times as large. This is consistent with the annual evidence
presented in Figures 5 and 6, which show that the main adjustment in the
usage of options was gradual and took place mainly over the course of
period II.
Table 1 Descriptive Statistics of the Sample
S 0 B N
Lev. Diff. Lev. Diff. Lev. Diff. Lev. Diff.
Overall .12 .43 .24 .22
Per. I .04 base .57 base .16 base .24 base
Per. II .09 .05 .45 -.12 .26 .10 .20 -.04
Per. III .26 .17 .19 -.26 .35 .09 .20 .00
Other .11 base .42 base .21 base .26 base
Min/Man .10 -.01 .47 .05 .25 .04 .19 -.07
FIRE .17 .06 .36 -.06 .27 .06 .20 -.06
Utilities .17 .06 .36 -.06 .23 .02 .25 -.01
size [Q.sub.1] .11 base .48 base .20 base .21 base
size [Q.sub.2] .12 .01 ,45 -.03 .27 .07 .17 -.04
size [Q.sub.3] .12 .01 .41 -.07 .32 .12 .15 -.06
size [Q.sub.4] .11 .00 .40 -.08 .37 .17 .12 -.09
age [Q.sub.1] .12 base .45 base .23 base .21 base
age [Q.sub.2] .11 -.01 .45 .00 .26 .03 .17 -.04
age [Q.sub.3] .12 .01 .42 -.03 .28 .05 .18 -.03
age [Q.sub.4] .12 .01 .39 -.06 .20 -.03 .28 .07
tenure [Q.sub.1] .11 base. .41 base .28 base .19 base
tenure [Q.sub.2] .12 .01 .42 .01 .27 -.01 .18 -.01
tenure [Q.sub.3] .13 .02 .43 .02 .24 -.04 .20 ,01
tenure [Q.sub.4] .12 .01 .44 .03 .18 -.08 .27 .08
Male .12 base .43 base .24 base .22 base
Female .18 .06 .34 -.09 .27 .03 .20 -.02
Asset val (mil. $) Age (y) Tenure female
(y)
(ave) (ave) (ave) (%)
Overall 14,924 55 7 .02
Per. I 11,926 56 7 .01
Per. II 16,893 56 7 .02
Per. III 18,155 55 7 .03
Other 5,545 54 7 .03
Min/Man 6,030 56 7 .02
FIRE 61,872 56 8 .01
Utilities 13,833 56 6 .01
size [Q.sub.1] 324 54 8 .03
size [Q.sub.2] 1,150 55 8 .02
size [Q.sub.3] 3,675 56 7 .02
size [Q.sub.4] 54,604 57 7 .01
age [Q.sub.1] 8,460 47 5 .03
age [Q.sub.2] 17,100 54 6 .02
age [Q.sub.3] 17,769 58 7 .01
age [Q.sub.4] 16,583 65 12 .00
tenure [Q.sub.1] 18,103 53 1 .03
tenure [Q.sub.2] 17,516 54 4 .02
tenure [Q.sub.3] 14,912 55 7 .02
tenure [Q.sub.4] 11,481 59 17 .01
Male 15,095 56 7 0
Female 5,594 52 5 1
For industry classification, we use the simple four groups of firms
proposed by Murphy (1999). Firms are classified into: 1) mining and
manufacturing, 2) finance and real estate (FIRE), 3) utilities, and 4) a
mixed group containing any other firm. Table 1 reports that the choice
of S is relatively more likely in FIRE and utilities, while that of 0 is
more likely in mining and manufacturing and other. The choice of B is
relatively more likely in FIRE, and less in other. Finally, the
proportion of firms choosing N is much lower in mining and manufacturing
and FIRE.
Next we report the breakout into compensation groups according to
size. The literature has established that size is an important factor in
the determination of pay levels. We use total assets as a measure of
size. (26) Year by year, we classify the firms in our sample according
to which of the four quantiles of the distribution of asset value they
belong. Table 1 reports the fraction of firms that choose each
instrument in the sample, and the differences from the fractions for the
control group, which is the quantile of smallest firms. The patterns of
usage of S seem to be independent of size, while 0 and N are relatively
more popular in smaller firms. On the other hand, using stock and
options together (B) is more frequent in larger firms.
Other potentially important characteristics are the tenure, age,
and gender of the CEO. We briefly discuss each of these in turn.
Younger executives may have different career concerns than older
ones, less experience, or different attitudes toward risk. More tenured
executives may be more vested in the firm by means of historical grants,
or firm-specific human capital. In Table 1 we see that the choice of S
seems to be fairly independent of both age and tenure. The choice of 0,
instead, is more frequent for younger executives, while, interestingly,
given the natural correlation of these two variables, it is less
frequent for shorter tenured ones. The frequencies of choice of B are
hump-shaped with respect to age, and decreasing for tenure. Firms seem
more likely to use none of the instruments more frequently for
long-tenured executives, and less frequently for middle-aged ones.
Some have argued that women are more risk averse than men (see
Schubert et al. [1999] for a discussion of the evidence); this could
influence the choice of compensation instrument. We only have 548
firm-executive-year observations that correspond to a female CEO, versus
30,032 for males. Despite this, we report the average use of instruments
by gender in Table 1 to point to one apparently significant difference:
Female CEOs are about 10 percent less likely than male CEOs to receive
options exclusively.
We now proceed to validate these raw statistics by performing a
formal check on the effect of firm characteristics on the choice of
instrument. We model the value to a given firm i of choosing a set of
instruments I at time t as
V [(I).sub.it] = [alpha] + [3.summation over (k = 1)]
[[beta].sub.k] [industry.sub.i] + [2.summation over (k = 1)]
[[beta].sub.3 + k] [period.sub.t] + [[beta].sub.6] [tenure.sub.it] +
[[beta].sub.7] [age.sub.it] + [[beta].sub.8] ln [(assets).sub.it] +
[[beta].sub.9] [female.sub.it] + [[epsilon].sub.it]. (1)
In words, the value V [(I).sub.it] is assumed to depend linearly on
a constant, a, dummy variables for the three distinct regulatory periods
in the sample, an indicator variable for the industry group to which
that firm i belongs, and the characteristics of firm i in year t that we
selected based on our sample analysis. We do not observe directly the
value V [(I).sub.it] but rather the discrete choice of firms for I
[member of] {S, 0, B, N}. Hence, our statistical model is
Pr (V [(I).sub.it] > v [(I').sub.it]) [for all]I' [not
equal to] I,
where the probability of observing the choice of a given I depends
on whether V [(I).sub.it] the value derived by a firm i from using
instrument I at time t, is higher than the value of the other
instruments. We assume the noise term[[epsilon].sub.t] has a type I
extreme value distribution, so our discrete choice regression is a
multinomial logit.
One concern with the interpretation of the results of the
regression is the potential for colinearity. Table 1 reports, starting
in the column labeled "Period," the averages of each variable
in the subgroups defined in the different rows of the table. When
analyzing those, the most salient fact is the uneven average size across
periods (average size is increasing) and across industry groups (FIRE
contains firms that are, on average, 10 times the size of firms in
mining and manufacturing). However, when we plot the actual size
distribution across periods it is not significantly different, thanks to
the high variation in the size of firms within periods that we get by
using the cross section. The difference in the distribution of size
across industry groups is more apparent. However, we perform a
robustness check of the main qualitative features of our regression
results by running our regression in four different samples according to
industry group, and we confirm them all. (27)
The results of the regression using the benchmark specification for
V (I) in (1) are reported in Table 2, under the column labeled as
regression specification (1). In the first row of Table 2, we report the
regression sample averages, or, equivalently, the average predicted
probability in the model. The numbers differ slightly from those
reported in Table 1 because some of the observations have missing values
for some of the regressors, and hence they are dropped from the sample.
Table 2 Regression Results Instrument
Instrument S O
Regression (1) (2) (3) (1) (2) (3)
spec.
Mean(Pr(I|X)) 0.122 0.123 0.122 0.427 0.427 0.427
*** *** *** *** *** ***
(0.002) (0.002) (0.002) (0.003) (0.003) (0.003)
sic4: ALE 0.110 0.108 0.110 0.416 0.408 0.411
("Other") *** *** *** *** *** ***
(0.004) (0.003) (0.004) (0.006) (0.005) (0.006)
AME(Mm/Man) -0.006 -0.006 -0.006 0.037 0.032 0.041
(0.004) (0.004) (0.004) *** *** ***
(0.007) (0.007) (0.007)
AME(FIRE) 0.038 0.049 0.038 -0.015 0.034 0.006
*** *** *** (0.010) *** (0.011)
(0.007) (0.007) (0.007) (0.010)
AME(Utilities) 0.080 0.089 0.082 -0.076 -0.018 -0.053
*** *** *** *** (0.010) ***
(0.008) (0.008) (0.008) (0.010) (0.010)
Period: ALE(I) 0.039 0.040 0.039 0.579 0.596 0.577
*** *** *** *** *** ***
(0.002) (0.002) (0.002) (0.005) (0.004) (0.004)
AME(II) 0.049 0.048 0.049 -0.121 -0.145 -0.119
*** *** *** *** *** ***
(0.004) (0.004) (0.004) (0.008) (0.007) (0.008)
AME(III) 0.227 0.222 0.227 -0.390 -0.410 -0.387
*** *** *** *** *** ***
(0.005) (0.005) (0.005) (0.006) (0.006) (0.006)
AME -0.001 -0.005 -0.000 -0.008 -0.048 -0.020
(In(assets)) (0.001) ** (0.002) *** *** ***
(mean = 7.71, SD (0.002) (0.002) (0.002) (0.002)
= 1.78)
4ME(In(TDC1)) 0.010 0.102
(mean = 8.00, SD *** ***
= 1.16) (0.002) (0.003)
AME (Ave(In 0.000 0.034
(TDC1))) (0.003) ***
(mean = 8.21, SD (0.004)
= 0.91)
N 26736 26686 26736 26736 26686 26736
Pseudo 0.109 0.215 0.120 0.100 0.215 0.120
[R.sup.2]
Instrument B N
Regression (1) (2) (3) (1) (2)
spec.
Mean(Pr(I|X)) 0.2.18 0.248 0.248 0.203 0.202
*** *** *** *** ***
(0.003) (0.002) (0.003) (0.002) (0.002)
sic4: ALE 0.245 0.222 0.233 0.229 0.263
("Other") *** *** *** *** ***
(0.005) (0.005) (0.005) (0.005) (0.004)
AME(Mm/Man) 0.036 0.038 0.040 -0.067 -0.064
*** *** *** *** ***
(0.006) (0.006) (0.006) (0.006) (0.005)
AME(FIRE) -0.049 0.034 -0.013 0.026 ** -0.117
*** *** (0.009) (0.010) ***
(0.008) (0.009) (0.007)
AME(Utilities) -0.043 0.025 ** -0.013 0.039 -0.095
*** (0.009) (0.009) *** ***
(0.009) (0.010) (0.007)
Period: ALE(I) 0.165 0.173 0.164 0.216 0.192
*** *** *** *** ***
(0.003) (0.003) (0.003) (0.004) (0.003)
AME(II) 0.096 0.079 0.097 -0.025 0.019
*** *** *** *** ***
(0.006) (0.006) (0.006) (0.006) (0.005)
AME(III) 0.186 0.169 0.192 -0.023 0.020
*** *** *** *** ***
(0.006) (0.006) (0.006) (0.006) (0.005)
AME 0.051 0.001 0.030 -0.042 0.052
(In(assets)) *** (0.002) *** *** ***
(mean = 7.71, SD (0.002) (0.002) (0.002) (0.002)
= 1.78)
4ME(In(TDC1)) 0.117 -0.228
(mean = 8.00, SD *** ***
= 1.16) (0.003) (0.003)
AME (Ave(In 0.054 ***
(TDC1))) (0.004)
(mean = 8.21, SD
= 0.91)
N 267.16 26686 26736 26736 26686
Pseudo 0.109 0.215 0.120 0.109 0.215
[R.sup.2]
Instrument
Regression (3)
spec.
Mean(Pr(I|X)) 0.203
***
(0.002)
sic4: ALE 0.245
("Other") ***
(0.005)
AME(Mm/Man) -0.075
***
(0.006)
AME(FIRE) -0.031
***
(0.009)
AME(Utilities) -0.016
(0.009)
Period: ALE(I) 0.220
***
(0.004)
AME(II) -0.027
***
(0.006)
AME(III) -0.031
***
(0.005)
AME -0.009
(In(assets)) ***
(mean = 7.71, SD (0.002)
= 1.78)
4ME(In(TDC1))
(mean = 8.00, SD
= 1.16)
AME (Ave(In -0.088
(TDC1))) ***
(mean = 8.21, SD (0.003)
= 0.91)
N 26736
Pseudo 0.120
[R.sup.2]
Standard errors in parentheses. AMEs are not reported
for variables that were not statistically significant.
* p < 0.05, ** p< 0.01, *** p < 0.001
In order to provide an intuitive sense of the estimated relative
importance of each regressor, the rest of the rows in the table report
the average of the partial derivatives of the probability of usage, or
the average marginal effect of each explanatory variable [x.sub.ij] in
the vector of all explanatory variables [X.sub.i], defined as
AME(j) [equivalent to] [Mean.sub.i] ([partial
derivative]Pr(I/[X.sub.i]))/[delta][x.sub.ij].
That is, using the estimated coefficients, we calculate how much
the probability of using each instrument changes for each of the firms
in the sample when we marginally increase the value of a given
explanatory variable [x.sub.ij], evaluated at the true value of the
vector of regressors [X.sub.i] for firm i; then we take the average of
those marginal changes over i. Note that the marginal effects are
calculated in a slightly different way for discrete variables. The
marginal effects with respect to the variable "Period," for
example, represent the average change induced by hypothetically
switching a firm from the base period, I, to each of the remaining
periods. (28) Formally,
AME(j) [equivalent to]
[Mean.sub.i][Pr(I/[[X.sub.i].sup.[x.sub.ij]], [x.sub.ij] = n) -
Pr(I/[[X.sub.i].sup.[x.sub.ij]], [x.sub.ij] = base)] ,
for all n different than base, where base denotes the value of the
regressor [x.sub.j], in this case period I, and n [not equal to] base
represents period II and period III. The notation
[[X.sub.i].sup.[x.sub.ij]] represents the vector of regressors [X.sub.i]
excluding regressor [x.sub.j]. In order to provide a benchmark to
evaluate these discrete changes in probability, we also report, for
discrete regressors, the level of the average predicted probability in
the sample when setting [x.sub.ij] = base (we denote this by ALE (j) in
the table).
Some of the strongest economic effects are associated with the
three regulatory subperiods. Size and industry are also statistically
significant for all groups. Despite the differences reported in Table 1,
and possibly due to high standard deviations, our controls for gender,
age, and tenure of the CEO often do not have a statistically significant
effect on the choice of compensation instruments, so we choose to not
report the AMEs for these variables.29 We now summarize the findings
regarding period, industry, and size.
Regulatory periods Firms move away from compensation packages that
include only options after both regulatory changes, either to use only
stock or to use options together with stock. They mainly add stock to
their compensation packages during period II, and they mainly substitute
stock for options in period III.
We find that if the same firm went from living in period I, before
Sarbanes-Oxley, to period II, the probability of it choosing 0 would
decrease by a substantial 12 percentage points (pp), while that of
choosing S would increase 5 pp and that of choosing B would increase by
almost 10 pp. In period III, after FAS 123R went into effect, the
probability of using options would be 39 pp lower than in the initial
period, leaving it at about 20 percent. The most favored category in
that switch would be stock, with a 23 pp increase, followed by both,
with a 19 pp increase. The use of none decreases at a modest 2-3 pp in
each of the two periods.
Industry classification Firms in FIRE and utilities favor packages
that include stock exclusively, or no grants at all, more frequently
than the average firm. Both these industries make less use of packages
that include options exclusively, or stock and options together. Firms
in mining and manufacturing, in contrast, use options exclusively, or
together with stock, slightly more than the average firm, and they are
less likely to compensate without using any grants at all.
The control industry, "other," aligns with the average
usage probabilities in the overall sample. We see that switching from
"other" to "mining and manufacturing" is associated
with a shift away from using N into 0, or B, in comparable magnitude.
Switching to FIRE is associated with an important shift away from 0 and
B (by 2 and 5 pp, respectively), into S and N (by 4 and 3 pp,
respectively). Switching to utilities, which includes transportation,
communications, electric, gas, and sanitary services, presents, perhaps
surprisingly, a similar pattern than FIRE. For utilities, however, the
effects are even stronger: The decrease in B and 0 is by 4 and 8 pp,
respectively, and the increase in N and S is by 4 and 8 pp,
respectively.
Size Larger firms are less likely to use compensation packages that
include options exclusively, and more likely to use those that include
both stock and options. They are also less likely to compensate without
using any type of grants at all.
Firm size is a continuous variable (the log of the value of assets
measured in thousands of 2010$). An increase in firm size significantly
increases the probability of choosing B, to the detriment of choosing 0
or N. It is difficult to compare the economic importance of size with
respect to the discrete variables that we just commented on, since the
numbers in the table represent the effect of infinitessimal increases in
size, not changes in industry or period as above. For comparison, we can
calculate the implied average change in the probabilities for an
increase in log size equal to one standard deviation. This
back-of-the-envelope calculation implies that the probability of
choosing B increases by approximately 9 pp, while that of 0 decreases by
1 and that of N by 7. This suggests that the magnitude of changes
associated with size is similar to that of changes in the industry
classification, but smaller than that of the regulatory subperiods.
The Role of Individual Firm Characteristics
In Figure 7 we provide a simple graphical evaluation of the fit of
the model in equation (1). We have plotted the sample percentage of
users of each instrument by year in each of the subplots (blue line),
the predicted probability of usage by the full model in (1) (red line),
and a limited model that includes as regressors only the period dummy
indicators (green line). (30) The main result that emerges from this
comparison is that the explanatory power of the variables that indicate
the different subperiods is high. Although the econometric model is not
able to fit the smooth decline in the use of options, we have already
pointed out earlier in this article that the new standards in FAS 123R
were already recommended by the FASB in 1993, and some companies started
adopting them earlier than 2006. This may explain most of the
discrepancies between the green line and the true data.
[FIGURE 7 OMITTED]
In contrast with the good tit of the model over time, a low
pseudo-R (2) seems to suggest that, even including the individual
characteristics, our model does not do a good job in explaining
individual cross-sectional variation in the use of instruments. As is
apparent from the figure, the individual characteristics of the firm
that we include in the full regression do not add much to the
explanatory power of the model over time. In fact, the pseudo-R (2) of
the regression using only the regulatory subperiods as explanatory
variables is already 7 percent, compared to the 11 percent of the full
model in specification (1). More work is needed to understand which
individual characteristics of firms determine their choice of
compensation instruments.
The Role of the Level of Pay
One potential explanatory variable of the choice of instrument that
we left out of our analysis in regression specification (1) is the level
of pay itself. It may be the case that tax advantages, or transparency
concerns of the firm, make it more convenient for the firm to pay large
sums to its CEO in the form of stock or options, rather than through a
salary or a bonus program. To explore this possibility, we present in
Table 3 the average level of total compensation (first row of all
periods, labeled as TDC1) by group and subperiod. We can see that firms
in N have a remarkably lower level of compensation, across all
subperiods. Moreover, firms choosing B have the highest average
compensation in all subperiods. The statistics for the group using only
stock or only options are interesting: The level of pay is higher for 0
in period I, when options were more likely to be used on their own than
stock (see Table 1). During period II, average pay is equal across the
two groups of users. As we discussed in Section 3, this is a period when
Sarbanes-Oxley had just been passed, making the choice of options more
costly--at least in terms of opportunities for backdating and maybe in
terms of public image. In period III, after the new accounting standards
that made the valuation of options less arbitrary became compulsory, the
ranking of average pay reverses: CEOs of firms that are users of stock
are paid, on average, more than those that are users of options.
Table 3 Average Total Compensation and Average Mean
Compensation by Firm
Period S O B N
I [TDC1.sub.it] 5,683 6,211 8,809 1,668
(37,035) (16,821) (15,811) (3,360)
mean[(TDC1).sub. 5,752 5,707 7,457 4,227
i] (7,804) (7,707) (7,707) (6,211)
II [TDC1.sub.it] 5,672 5,677 8,854 2,185
(8,023) (7,399) (10,304) (3,627)
mean[(TDC1).sub. 5,422 5,380 7,206 4,380
i] (6,447) (5,876) (7,510) (7,398)
III [TDC1.sub.it] 5,539 4,754 7,621 1,762
(7,861) (6,645) (7,827) (2,866)
mean[(TDC1).sub. 5,236 4,616 6,688 3,625
i] (6,163) (5,196) (6,421) (5,605)
Across [TDC1.sub.it] 5,585 5,893 8,299 1,804
Periods (16,417) (14,020) (11,448) (3,299)
mean[(TDC1).sub. 5,350 5,486 7,056 4,095
i] (6,499) (6,409) (7,125) (6,335)
Period Across
Groups
I [TDC1.sub.it] 5,519
(16,176)
mean[(TDC1).sub. 5,630
i] (7,002)
II [TDC1.sub.it] 5,808
(8,106)
mean[(TDC1).sub. 5,660
i] (6,771)
III [TDC1.sub.it] 5,372
(7,218)
mean[(TDC1).sub. 5,310
i] (6,090)
Across [TDC1.sub.it] 5,542
Periods (12,412)
mean[(TDC1).sub. 5,542
i] (6,691)
Notes: Standard deviations in parenthesis.
In order to explore formally the explanatory power of the level of
pay after controlling for firm characteristics, we replicate the
regression in equation (1), but add the level of total compensation (the
log of the variable TDC1 in Execucomp) as a regressor. The results are
reported in Table 2, under the column labeled as regression
specification (2).
The meaningfulness of these estimated effects needs to be evaluated
in the context of the mechanics of compensation, since there is an
obvious relation between the level of compensation and the use of
grants. This mechanical relationship exists unless we think that
sometimes firms issue grants that are very small in value. In the
sample, the minimum value for stock and option grants is in the order of
$3; the 1st percentile value is $18,667 for stock grants and $38,355 for
options; the 10th percentiles values are about $200,000 and $250,000,
respectively. Given that the 10th percentile of salary payments in the
sample is $364,000, and that of total compensation is in the order of
$750,000, these statistics suggest that fairly low values of the grants
are possible and not that uncommon.
Another concern with the results of regression specification (2) is
endogeneity: Since stock and options are risky assets, CEOs receiving
their compensation in the form of grants (as opposed to salary or other
less risky instruments, such as bonuses) may need to be compensated for
their risk aversion with higher levels of pay. See Hall and Murphy
(2002) for a formal explanation and quantification of the effect of risk
aversion on the value of grants to executives, and the comparison of
that value to the cost for the firm.
Total compensation is indeed a significant variable according to
the results of the multinomial logit. Also, the pseudo-[R.sup.2] doubles
with respect to specification (1). We find that a marginal increase in
the level of pay leaves the probability of using stock almost unchanged,
while it increases the one for choosing 0 by 10 percent and of B by 12
percent; it decreases the probability of using N by 23 percent.
The effects of size change significantly in specification (2). An
increase in size now has a small but significant positive effect on the
probability of S. The effects on 0 remain negative but increase
significantly in magnitude. Moreover, the positive relation between size
and choosing B becomes negligible (and insignificant) when controlling
for the level of pay. Finally, the negative effect of size on the
probability of choosing N changes to positive when controlling for the
level of pay, suggesting that if a given firm is granting a relatively
high level of total compensation, the fact that it is a larger firm
actually makes it less likely to include stock or options in its
compensation package.
We can also consider the changes in the marginal effects of the
rest of the regressors with respect to those reported in specification
(1). As for the period variable, the new model has similar implications
both for period II and III when it regards the choice of S. However, the
negative effect on the probability of choosing 0 is even stronger, while
that of choosing B is still positive but weaker. The effect on choosing
N changes sign in both periods with respect to specification (1):
Although magnitudes are small, firms are more likely to choose N in
later periods than in the initial one. Note in Table 3 that there is no
clear trend of average compensation over time across groups; however,
compensation is lower in later periods for firms that include options in
their packages.
As for the industry dummy, while the results for mining and
manufacturing are very robust, for FIRE and utilities we observe some
important changes. While firms in FIRE were more likely to choose S or N
in specification (1), when including the level of total compensation as
a control they are more likely to choose S, and 0 or B are now favored
(in similar magnitudes), while N is now less likely to be chosen.
Utilities is also more likely to choose S or N in specification (1); in
specification (2) it becomes a likely user of B and a less likely user
of N. A possible explanation of these reversals in the sign of the
coefficients is that, given their size, firms in FIRE and utilities tend
to have lower levels of pay, which are associated with a lower
probability of choosing B and a higher probability of choosing N; when
the level of pay is not a control, that effect is assigned by the model
to the industry dummy.
One may suspect that the covariance of the regressors with the
level of pay is a potential cause of these changes in the estimated
coefficients. However, the covariance is not perfect, and both size and
pay remain significant in the robustness check, suggesting that
specification (1) may have an omitted variable problem. Numbers need to
be taken with caution.
As a final robustness check, we replicate the regression in
equation (1) but add as a regressor the average level of total
compensation (the log of the average of the variable TDC1 in Execucomp)
of a firm across the years that it stays in the sample, rather than the
actual level of TDC1 in each year. The results are reported in Table 2,
under the column labeled as regression specification (3). Table 3
reports the average and standard deviation of this measure of pay in the
sample (labeled mean[(TDC1).sub.i]). The most striking feature is the
much higher pay for firms choosing N when compared to the average of
contemporary level of pay. This reflects the fact that many of the firms
in N are in one of the other compensation groups in some of the years.
The hope in including average pay as a regressor is that this may
break slightly the mechanical link between the level of pay and the
presence of grants, and rather pick up some firm characteristics that
are correlated with, for example, the outside opportunity of the CEO, or
any other characteristic that determines his average pay across the
years but not necessarily the timing of the grants. We see in Table 3
that the pseudo-[R.sup.2] is higher than in specification (1), but much
lower than in specification (2). The average level of pay is a
significant explanatory variable for O, B, and N, and the sign of the
coefficients is aligned with that of the contemporaneous level of pay,
but its economic importance is much smaller, confirming that some of the
effects of the level of pay on the choice of instruments are purely
mechanical.
Choosing N and the Timing of Grants
There is some anecdotal evidence that companies tend to have fixed
timing rules when it comes to giving stock or option grants to their
executives. Hence, when we observe a firm choosing N in our sample it
may just mean that the firm is in a "non-granting" year, but
that it will grant again the following year, or in a couple of years,
depending on its timing rule. For example, taking the compensation of
Steve Jobs over his tenure as Apple CEO (see Section 1), according to
our classification, the company chose N in 10 of the years, 0 in 3, and
S in 1. Why firms may not want to smooth out grants is, to our
knowledge, an open question, and beyond the scope of this article.
However, the common practice of having the selling restrictions of both
stock and option grants vest progressively over time does provide some
smoothing. Unfortunately, there is no good data readily available on
these vesting periods. Nonetheless, we should keep in mind that if the
practice of timing grants on a regular basis is really prevalent, then
the statistics about usage presented here should be understood as
informative about the timing of grants, and changes in usage patterns
would be informative about changes in this timing.
A thorough analysis of the reincidence patterns in the usage of
instruments is beyond the scope of this article, and is left for future
research. However, in order to provide a sense of how much of the
variation in instrument choice in the data is not coming from timing of
grants, we now report on a measure of the frequency of instrument use at
the individual firm level: We calculate the fraction of years that a
given firm is in each of the groups, or the "firm's time share
of I" Denoting by [T.sub.iI] the number of years that a firm i is
in compensation group I, and by [T.sub.i] the total number of years that
firm i is in the sample, firm i's time share of I is defined tit
defined
[T.sub.iI] [equivalent to] [t.sub.iI]/[T.sub.i]
for each I in I. To give more meaning to the extreme values
[T.sub.iI] = 0 and [T.sub.iI] = 1, we construct a balanced subset of the
sample that includes only firms that we observe for at least six years
([T.sub.i] [greater than or equal to] > 6, a total of 489 firms out
of the original full sample of 3,248 firms). From the fact that there
are mass points at 0 (and, to a lesser extent, at 1) in the frequencies
for this subsample, we conclude that, provided compensation cycles are
shorter than six years, not all the firms are following alternating
times for the inclusion of options or stock grants in their compensation
packages. This means that at least some of the variation that we see in
the data comes from meaningful choices about the usage of the different
compensation instruments.
Because the timing choices themselves may be influenced by the
regulation period, in Tables 4 and 5 we report statistics of [T.sub.iI]
by regulatory subperiod. We report this for the balanced subsample
(denoted "Bal."), as well as for our original full sample
(denoted "Full").
Table 4 reports the fraction of firms with [T.sub.iI] = 0, i.e.,
they are never in compensation group I. It shows a pattern consistent
with the evidence in our previous regression results: The fraction of
firms that never were in group S or B decreases over time, while that of
firms never choosing 0 increases. Interestingly, the increase in the
fraction of firms with [T.sub.iN] = 0 over time suggests that, if
anything, timing decisions have changed toward using grants more
frequently.
Table 4 Percentage of Firms that are Never in a Given
Compensation Group
Percent S 0 B N
of Firms
with
[T.sub.iI]
= 0
Period Bal. Full Bal. Full Bal. Full Bal. Full
I. .82 .86 .08 .12 .49 .61 .40 .42
II. .80 .80 .26 .29 .39 .53 .65 .60
III. .59 .55 .65 .64 .29 .42 .72 .59
Table 5 reports the average value of [T.sub.iI] contingent on it
being positive; that is, the average time share [T.sub.iI] for firms
that are in compensation group I for at least one year. To report both
the averages and their significances, we run an ordinary least squares
regression of [T.sub.iI] on period dummies, for each I. The first column
under each I reports, for the balanced sample, the coefficients for the
constant (the level in the control period, I) and the included dummies
(the change in the average [T.sub.iI] in each subsequent period with
respect to period I), while the second column reports the same
coefficients for the larger sample of firms that are in the data for at
least six years. The patterns and significances are remarkably similar
across the two samples of firms. There is no evidence of a significant
decrease in the fraction of years that firms choose to grant options
only ([T.sub.iO]) in period II, while it is significant both
statistically and economically in period III. There is an important
upward trend for both [T.sub.iS] and [T.sub.iB], and a lot less markedly
for [T.sub.iN]. (31) That is, (1) firms that choose 0 do so less
frequently in period III, (2) firms that choose S, or B, do so more
frequently in the later periods than in the initial one, and (3) firms
that choose N do so only slightly more often in the last two periods
than in the first one. Since these changes in grant timing patterns
align with the trends in the usage of instruments that we have reported
in Table 2, we conclude that our results could be due, at least partly,
to a change in the frequency of usage of stock and options, rather than
a change in the number of different firms that use them.
Table 5 Mean Time Shares
Mean S O D N
([T.sub.iI]
|[t.sub.iI]
> 0)
Period Bal Full Bal. Full Bal. Full Bal. Full
I: Level .26 .28 .62 .65 .36 .40 .34 .41
(.03) (.02) (.01) (.01) (.02) (.01) (.01) (.01)
II: .15 .17 [-.02] -.02] .19 .15 .07 .10
Change (.04) (.02) (.02) (.01) (.02) (.01) (.02) (.01)
from
Period I
III: .31 .27 -.14 -.14 .33 .24 [.05] .08
Change (.04) (.02) (.03) (.02) (.02) (-01) (.03) (.01)
from
Period I
Adjusted .17 .13 .03 .03 .18 .10 .01 .03
[R.sup.2]
N 387 1,345 979 3,532 894 2,531 602 2,489
Notes: Time share [T.sub.iI] represents the fraction of years
that firm i belongs to group I, out of the total number of
years that firm [T.sub.iI] is in the sample. This table
reports mean time shares for each I, for firms with positive
[T.sub.iI]. Square brackets indicateinsignificance at the 5
percent con&lence level.
It is important to keep in mind that the evidence on the timing of
grants that we have provided in this section is partial, since it does
not control for the amount of past grants and it only exploits the panel
aspect of the data in a limited way. It would be interesting to perform
the analysis of usage that we do here with a comprehensive measure of
the wealth of the CEO vested in the firm at each point in time (as in
Clementi and Cooley [2010]), as a way of controlling for outstanding
incentives. This is left for future research.
4. THE IMPORTANCE OF DIFFERENT COMPENSATION INSTRUMENTS: THE
INTENSIVE MARGIN
In the previous section we asked what determines the choice of
compensation instruments. A natural complementary question to that is
what is the relative importance of each instrument in the total
compensation of the CEO. In this section, we provide some simple
statistics about the share of total compensation that salary, bonus and
incentive compensation (BIC), stock grants, option grants, and
"other compensation" represent.
Table 6 documents the average of these shares in our sample,
disaggregated by groups of users. The most salient feature of those
statistics is the difference in the shares of grants across firms in S,
0, and B: Firms in B have a combined share of grants of 57 percent,
higher than the shares of grants for firms using stock exclusively (45
percent) or options exclusively (49 percent). The share of BIC is
similar for firms in S, 0, and B, around 20 percent. In contrast, firms
in N, who do not use stock or options, use both BIC and "other
compensation" more intensely than the rest of firms, but the share
of the only incentive instrument, BIC, is 30 percent, well below the
combined shares of incentive instruments (BIC + stock + option) of the
rest of the firms. In other words, BIC, stock, and options do not appear
to be perfect substitutes for each other. This evidence complements what
we presented in Table 3 about the relationship between the level of
compensation and the usage choices, suggesting that the relative
importance of different instruments may be related to the choice of
instruments through the level of pay. We saw in Table 3 that firms in N
have levels of total compensation between one-third and one-fourth of
the rest of firms. In spite of this, the relative importance of the
salary is much higher for them. Hence, there seems to be a fixed
component in the determinant of the salary, or a "cap," which
is somewhat independent of whether the firms choose to also award grants
or not. The most obvious explanation is the limits to tax deductions for
salaries above a certain leve1.32 However, other factors may be
important, like the need to provide incentives through variable pay.
This also possibly plays a role in explaining the difference in the
shares of salary across the firms in S, 0, and B. The share of salary is
the lowest (19 percent) for firms in B, which are the ones that have the
highest total compensation according to Table 3. However, the share of
salary is equal for firms in S than for firms in 0, in spite of the
average total compensation in S being 90 percent of that in 0.
Table 6 Shares of Total Compensation, by Instrument
Salary BIC Stock Option Other
(%) (%) (%) (%) (%)
All .32 .23 .11 .28 .06
S .27 .23 .45 0 .05
0 .27 .20 0 .49 .04
B .19 .20 .26 .31 .04
N .59 .30 0 0 .10
Our previous analysis has shown that the use of instruments differs
importantly across subperiods, and to some extent across industry
groups. Hence, we now look at the average shares controlling for these
two variables.
Table 7 presents evidence on the changes in the relative importance
of the instruments over the three different regulation subsamples. For
convenience, we replicate the sample frequencies of each group of
compensation, within a period, that we already discussed following Table
1.
Table 7 Shares of Total Compensation, by Instrument
Salary BIC Stock Option Other
Freq. (%) (%) (%) (%) (%)
Period I .35 .22 .04 .33 .05
.04 S .34 .26 .33 0 .07
.57 0 .28 .20 0 .48 .04
.16 B .22 .18 .20 .36 .04
.24 N .62 .28 0 0 .09
Period II .29 .23 .10 .32 .05
.09 S .27 .26 .42 0 .06
.45 0 .25 .20 0 .51 .04
.26 B .18 .20 .25 .33 .04
.20 N .56 .32 0 0 .11
Period III .29 .23 .24 .18 .06
.26 S .25 .22 .49 0 .05
.19 0 .27 .21 0 .48 .04
.35 B .18 .21 .31 .27 .04
.20 N .55 .32 0 0 .12
When we look at the shares for all the users together, we see that
while the shares of BIC and "other compensation" remained
fairly constant at about 23 percent and 5 percent, respectively, the
share of salary was higher in period I (35 percent as opposed to 29
percent post-2002). The share granted in the form of options also
experienced a sharp decline, but only in period III, when it went from
32 percent to 18 percent. The share of compensation that is no longer
granted through salary after 2002 and no longer granted through options
after 2006 is granted through stock: There is an increase in the share
of stock of 6 pp in period II, and then of 14 extra pp in period III.
These changes in the share of stock over time (intensive margin) are in
line with the changes in the choice of S reported in Table 1 (extensive
margin), where we saw that firms tended to "add" stock to
their compensation package in period II, rather than completely
substitute options for stock. Note, however, that these numbers for the
share of total compensation that are given in the form of stock are
representative both of firms in S and B in Table 1. We discuss the data
in each compensation group next.
When we look at the statistics disaggregated by user groups, we see
slightly different changes over the regulatory periods for each of them.
The most striking fact may be the increase in the share of stock, which
happens both for firms that are in S and in B. For firms in S, the share
of stock increases by 9 pp in period II (compensated mainly by a
decrease in the share of salary of 7 pp), and then by 7 pp in period III
(compensated mainly by a decrease in the share of BIC by 4 pp). For
firms in B, the share of stock increases by about 5 pp each period,
while the share of options decreases (3 pp in period II, 6 extra pp in
period III). In addition, for firms in 0 the share of options stays
constant overall (and it even increases by 3 pp in period II). In other
words, for the firms that continue to rely exclusively on option grants
in spite of the regulatory hurdles, the relative importance of options
with respect to salary, BIC, and "other compensation" does not
decrease. That is, if what we observe is a response to the regulatory
changes, it seems to have taken place through the extensive margin (with
firms in 0 going from 57 percent of the sample to 19 percent), rather
than the intensive one. This suggests that there might be some fixed
cost to adopting a new instrument of compensation, maybe related to
accounting costs or perhaps to communication to shareholders.
Table 8 presents the shares of each compensation instrument by
industry group. The variation in the shares across instrument users,
within a given industry group, is fairly in line with the patterns by
users that we described in Table 6, so we do not report the
disaggregated numbers here.(33) One main conclusion stands out from
Table 8--mining and manufacturing and other use options and salary more
intensely than do FIRE and utilities, which rely more on BIC and stock.
FIRE, which includes financial firms, has in fact the lowest share for
salary. It is important to keep in mind that, as reported in Table 1,
the proportion of firms in each user group is not constant across
industry groups; this, together with the (omitted) evidence that shares
for user groups within industry align with those reported in Table 6,
implies that most of the variation across industries is due to
composition effects, without important industry-specific patterns for
the shares of each compensation instrument.
Table 8 Shares of Total Compensation, by Industry Group
Salary BIC Stock Option Other
(%) (%) (%) (%) (%)
Min/Man .32 .22 .11 .31 .05
FIRE .29 .27 .15 .23 .06
Utilities .34 .25 .14 .21 .06
Other .33 .20 .11 .30 .06
5. CONCLUSION
In the last decade several regulatory changes took place in the
United States regarding the reporting and expensing of stock option
grants. This article provides an empirical analysis of the impact of
these changes in the composition of pay packages for CEOs at the largest
U.S. firms from 1993 to 2010. Both the passage of the Sarbanes-Oxley Act
in 2002 and the changes in accounting standards in SFAS 123R mandated by
the SEC in 2006 erased some advantage of granting options versus stock
as part of the compensation of CEOs. We find evidence indicating that
firms may have responded to this by shifting away from options and into
stock. Even though, after the two regulatory changes, there is still a
significant portion of firms in the sample that choose to grant options
to their CEO (about 55 percent of firms in the 2006-2010 period,
compared to 67 percent before 2002), alone or combined with stock, the
fraction of firms that are awarding options but not stock in a given
year decreases (from 57 percent before 2002 to 19 percent after 2006).
However, while only 4 percent of firms used exclusively stock grants
before 2002, this percentage increases over the period we analyze to
reach 26 percent after 2006.
How firms decide whether to include options, stock, both, or none
of the two types of grants in their pay packages remains to be
understood, but we find some regularities. Firms in finance and in
utilities are more likely to use stock or neither, while firms in mining
and manufacturing are more likely to use options, or stock and options
together. Larger firms tend to use stock and options together, although
this effect disappears if we control for the level of pay, which is
higher at larger firms. A higher level of pay is associated with a
higher probability of using stock and options together, or only options.
We also find that different compensation instruments do not appear
to be perfect substitutes within compensation packages. The relative
importance of bonuses in overall compensation has not decreased over
time, while that of the salary has, in favor of stock and option grants.
Perhaps surprisingly given the decrease in the popularity of option
grants starting in the early 2000s, the relative importance of options
in relation to the total amount of compensation has not decreased over
time for firms that still include options in their compensation
packages.
The views expressed here do not necessarily reflect those of the
Federal Reserve Bank of Richmond or the Federal Reserve System. E-mail:
arantxa.jarque@rich.frb.org.
REFERENCES
Baker, George P., and Brian J. Hall. 2004. "CEO Incentives and
Firm Size." Journal of Labor Economics 22 (October): 767-98.
Bickley, James M. 2012. "Employee Stock Options: Tax Treatment
and Tax Issues." Congressional Research Service Report for Congress
(June 15).
Brown, Lawrence D., and Yen-Jung Lee. 2011. "Changes in
Option-Based Compensation Around the Issuance of SFAS 123R."
Journal of Business Finance e4 Accounting 38 (November/December):
1,053-95.
Cheng, Ing-Haw, Harrison G. Hong, and Jose A. Scheinkman. 2012.
"Yesterday's Heroes: Compensation and Creative
Risk-Taking." ECGI--Finance Working Paper No. 285/2010; AFA 2011
Denver Meetings Paper (June 24).
Clementi, Gian Luca, and Thomas F. Cooley. 2010. "Executive
Compensation: Facts." Fondazione Eni Enrico Mattei Working Paper
2010.89.
Core, John, and Wayne Guay. 1999. "The Use of Equity Grants to
Manage Optimal Equity Incentive Levels." Journal of Accounting and
Economics 28 (December): 151-84.
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): 2,099-138.
Gabaix, Xavier, and Augustin Landier. 2008. "Why Has CEO Pay
Increased So Much?" The Quarterly Journal of Economics 123 (1):
49-100.
Guay, Wayne, David Larcker, and John Core. 2005. "Equity
Incentives." In Top Pay and Performance: International and
Strategic Approach, edited by Shaun Tyson and Frank Bournois.
Burlington, Mass.: Elsevier Limited, Chapter 8.
Guay, Wayne, S. P. Kothari, and Richard Sloan. 2003.
"Accounting for Employee Stock Options. "American Economic
Review 93 (May): 405-9.
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 Jeffrey B. Liebman. 2000. "The Taxation of
Executive Compensation." In Tax Policy and the Economy, Volume 14,
edited by James M. Poterba. Cambridge, Mass.: MIT Press, 1-44.
Hall, Brian J., and Kevin J. Murphy. 2002. "Stock Options for
Undiversified Executives." Journal of Accounting and Economics 33
(February): 3-42.
Heron, Randall A., and Erik Lie. 2007. "Does Backdating
Explain the Stock Price Pattern around Executive Stock Option
Grants?" Journal of Financial Economics 83 (February): 271-95.
Jarque, Arantxa. 2008. "CEO Compensation: Recent Trends and
Regulation. "Federal Reserve Bank of Richmond Economic Quarterly 94
(Summer): 265-300.
Jensen, Michael C., and Kevin J. Murphy. 1990. "Performance
Pay and Top-Management Incentives."
Journal of Political Economy 98 (April): 225-64.
Lewellen, Wilbur G. 1968. Executive Compensation in Large
Industrial Corporations. New York: National Bureau of Economic Research.
Meyers, Arthur S. 2012. "Preserving Your Company's Tax
Deduction for Stock Awards." StockSense: Quarterly Newsletter from
Fidelity Stock Plan Services (December).
Murphy, Kevin J. 1999. "Executive Compensation." In
Handbook of Labor Economics, Vol. 3b, edited by Orley C. Ashenfelter and
David Card. Amsterdam: Elsevier Science North Holland; 2,485-563.
Murphy, Kevin J. 2003. "Stock-Based Pay in New Economy
Firms." Journal of Accounting and Economics 34 (January): 129-47.
Rosen, Sherwin. 1992. "Contracts and the Market for
Executives." In Contract Economics, edited by Lars Werin and Hans
Wijkander. Cambridge, Mass.: Blackwell Publishers.
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.
Schubert, Renate, Martin Brown, Matthias Gysler, and Hans Wolfgang
Brachinger. 1999. "Financial Decision-Making: Are Women Really More
Risk-Averse?" The American Economic Review 89 (May): 381-5.
Securities and Exchange Commission. 2006. "Executive
Compensation and Related Person Disclosure." Available at
www.sec.gov/rules/fina1/2006/33-8732a.pdf.
(1.) When options are granted with the exercise price equal to the
stock market price at the date of grants, they are called "at the
money" this is the most popular practice, although some options are
occasionally also granted "in the money" (with exercise price
below market price) or "out of the money" (with exercise price
above market price).
(2.) This differential tax treatment was introduced in 1993, IRC
section 162(m). See Hall and Liebman (2000) for an analysis of the
taxation of executive compensation. See Meyers (2012) for a recent
explanation of requirements on stock awards that are considered
performance-based and qualify for a tax deduction.
(3.) See Securities and Exchange Commission (2006).
(4.) "Ownership Reports and Trading by Officers, Directors and
Principal Security Holders," Release No. 34-46421 (Aug. 27, 2002)
[56 FR 56461] at Section II.B.
(5.) Investigations pointing to the existence of backdating became
well-known only in 2005. Since Sarbanes-Oxley was passed in 2002, it may
be the case that the change in reporting requirements was not directly
aimed at preventing backdating. In other words, in trying to improve
corporate governance in general, the Act inadvertedly limited the
possibility of backdating.
(6.) For a discussion of the issues and anecdotal evidence, see the
Wall Street Journal article "The Perfect Payday" (March 18,
2006). For an academic evaluation of the backdating practice, see Heron
and Lie (2007) and references therein.
(7.) This tax treatment applies to "non-qualified" option
grants, which are the most common in executive compensation packages
during the time period that we study. Firms are also allowed to grant
"qualified" options, or "incentive stock options,"
to their employees, which are limited to a maximum value of $100,000,
and hence are not usually granted to executives. See Bickley (2012) for
details on the taxation of employee stock grants.
(8.) Because backdating implies a violation of the SEC's
disclosure rules, a violation of accounting rules, and a violation of
tax laws, the SEC has sued a number of companies suspected to have
engaged in this practice. See, for example, the testimony of Christopher
Cox as Chairman of the SEC on September 6, 2006 (available at
www.sec.gov), where he states that charges related to this matter were
made as early as 2003.
(9.) Accounting standards and a detailed description of accepted
"intrinsic value" calculations can be found in APB 25, from
the FASB.
(10.) See Guay, Kothari, and Sloan (2003) and clear exposition of
these issues.
(11.) See, for example, Brown and Lee Stock Opt ion Expenses: Is
the Debate (www.nysscpa.org/epajourna1/2005/1105/essentia1s/p38.htm).
Guay, Larcker, and Core (2005) for a (2011), or "Reporting Employee
Over?" by Paulette A. Ratliff
(12.) For the analysis of sensitivity of pay to performance, see
the seminal contributions of Jensen and Murphy (1990), Rosen (1992), and
Hall and Liebman (1998). For the relationship of pay level and
sensitivity to firm size in the cross section, see Schaefer (1998) and
Baker and Hall (2004). A more recent study of the variation of the level
of pay over time and its potential relationship to firm size is Gabaix
and Landier (2008). Frydrnan and Saks (2010) provides a comprehensive
historical overview of both level and sensitivity of pay facts using a
small sample of firms over an unusually long period, from 1936 to 2005.
(13.) See Apple's Definitive Proxy statement on March 16,
1998.
(14.) According to Execucomp, Jobs received a bonus payment in 2001
and 2002, and two big sums as "other compensation" in 2001 and
2002.
(15.) Given the compensation pattern of Jobs, it is interesting to
note that Apple stated in April 2003 its decision to voluntarily expense
option grants to its employees according to FASB recommendations
(16.) See Apple's Definitive Proxy statement on January 7,
2011.
(17.) In spite of the sparse availability, we did construct a value
of shares owned for the CEOs for which we had data: The average value
for those that we classified as non-owners was $1,928,000, compared to a
mean total compensation of $2,507,000.
(18.) Clementi and Cooley (2010) used the more restrictive
threshold of 1 percent ownership. We conducted a robustness check of our
main analysis by dropping all CEOs who owned 3 percent or more shares on
average over their tenure and results did not change qualitatively.
(19.) For recent studies that use this same measure of total
compensation, see Gabaix and Landier (2008); Frydman and Saks (2010);
and Cheng, Hong, and Scheinkman (2012).
(20.) Note that firms amortize the expense from these grants over
their vesting period, and hence compensation expenses are actually
smoothed out over time by the firms.
(21.) See Figures 2 and 3 in Murphy (1999).
(22.)See Jarque (2008) for a review.
(23.) See also Lewellen (1968).
(24.) See Frydman and Saks (2010, Figure 2, p. 2,108).
(25.) See Frydman and Saks (2010, Figure 1, p. 2,107).
(26.) For recent estimates, see Gabaix and Landier (2008, Table I,
p. 66). Other size measures used in the literature are the number of
employees and sales value. We confirm that in our data set the size
measure with the highest [R.sup.2] for the level of pay is asset value.
Details are available upon request.
(27.) Details are available upon request.
(28.) To calculate the marginal effect for "2002-2005,"
Stata calculates the predicted probabilities by setting to 1 the dummy
for the baseline period of "1993-2001" into each observation
while leaving all other regressors at their true sample values. Then it
calculates this predicted probability again by substituting
"2002-2005" instead. The average of this difference is the
reported marginal effect.
(29.) Details are available upon request.
(30.) The SEC adopted FAS 123R reporting rules for firms filing
their proxy statements after December 15, 2006. Hence, we classify firms
as being in period III if the month in which their fiscal year ends
falls after November 2006. This means that in the year 2006, the firms
in the sample are split across two subperiods. This explains the extra
kink in the predictions using the restricted model.
(31.) Note that Table 5 is providing evidence for firms that have
[T.sub.iI] > 0, and these firms differ across Is; hence, the
percentages across rows do not typically sum up to 1.
(32.) The Omnibus Budget Reconciliation Act Resolution 162(m) of
1992 imposed a $1 million cap on the amount of the CEO's
non-performance-based compensation that qualifies for a tax deduction.
See Jarque (2008) for a review of the academic literature that studied
the effects of that change of regulation on pay practices.
(33.) A more detailed table with shares across industries and
compensation groups is available upon request.
