An Empirical Analysis of CEO Risk Aversion and The Propensity To Smooth Earnings Volatility

An Empirical Analysis of CEO Risk
Aversion and the Propensity to Smooth

Earnings Volatility
A. R. ABDEL-KHALIK*
Volatility of firm performance is one measure of firm-specific risk that

is priced in the marketplace. Because CEOs are typically undiversified
investors in their companies’ stocks, the negative valuation effects of
this risk are asserted to provide them the incentive to reduce earnings
volatility. Furthermore, such an incentive is expected to increase in risk
aversion. Two alternative measures of risk aversion are generated: one
derives from wealth, and the other derives from CEOs’ choice of com-
pensation structure as to the risky and safe components. The results
consistently show a negative relationship between risk aversion and the
volatility of each performance measure: earnings and operating cash
flows. The results survived several tests of robustness. However, industry
analysis shows that these results do not hold for certain industries such
as public utilities.
Keywords: earnings volatility, firm-specific risk, management expense pref-

erence, income smoothing, Mayshar asset pricing model, utility
over wealth
1. Introduction
This study reports evidence on the relationship between chief executive offi-
cers’ (CEOs’) risk aversion and the volatility of performance measures (earnings
and operating cash flows). It is common knowledge that CEOs could not
*College of Business, University of Illinois at Urbana-Champaign

Early drafts ofthis paper were presented, and greatly benefited from comments, at workshops at
the University of Illinois; Pennsylvania State University; Nanyang Technological University; Univer-
sity of New South Wales; Katholic University of Leuven; Washington University mini conference;
and Athens University of Business and Economics workshop and the European Accounting Associa-
tion Congress in Gothenberg in 2005; City University of Hong Kong and The Chinese University of
Hong Kong. I benefited by the comments received at those workshop presentations and from Paul
Beck, Ying Cao, Dan Collins; John Core, Rajib Doogar, Steven Huddart, Joel Pike, Joshua Ronen,
Rachel Schwartz, Theodore Sougiannis, Stephen Taylor, Ken Trotman, T. J. Wong; and Martin Wu.
I appreciate the research assistance of Ying Cao, Sam Han, and Keejae Hong and the financial sup-
port of the Department of Accountancy at the University of Illinois at Urbana-Champaign.
201
202 JOURNAL OF ACCOUNTING, AUDITING & FINANCE
diversify their human capital as it is tied to the firms they manage because of
the exclusivity of their employment contracts, but it is less known that CEOs
invest heavily in the equity of their firms and therefore are ‘‘unable to fully
diversify their exposure to firm-specific risk’’ (Jin [2002], 31).1 In addition,
unlike other investors who could diversify at low cost and care only about sys-
tematic risk, nonsystematic (i.e., firm-specific) risk does matter to holders of
undiversified portfolios, such as CEOs. Finally, CEOs are in a unique position
because they, unlike other investors, are able to influence the level of volatility
of reported income or operating cash flows (smoothness) by strategically manag-
ing their choices of investments and accounting policies (Stein [1996]. More to
the point, Perrino, Poteshman, and Weisbach find that ‘‘risk-reducing projects
become more attractive and risk-increasing projects become less attractive’’ as
the manager becomes more risk averse (2002, 28).2
CEOs have traditionally been skeptical of market efficiency and, to the
extent that they do not believe they are price takers, they could act to reduce
the negative effects of earnings or cash flows volatility on their financial hold-
ings. Indeed, the public records of lobbying the regulators and a plethora of
reports in the financial press easily document managers’ preferences for
accounting rules that smooth earnings. This study hypothesizes that this prefer-
ence for smooth earnings is not uniformly distributed across CEOs and,
instead, it depends on CEOs’ propensity to take risk. Assuming that they have
similar utility over wealth, they are expected to exhibit different degrees of
aversion to risk, depending on their wealth endowments and compensation
structure. Using wealth endowments alone, risk aversion is measured by
Arrow-Pratt Decreasing Absolute Risk Aversion (DARA). Using wealth endow-
ment and compensation structure, risk aversion is measured by the extent to
which compensation choice is made up of stock-based awards (such as stock
options or stock appreciation rights). Given these measures, the hypothesis
being tested is that aversion to risk implies an aversion to earnings volatility.
1. This study is consistent with the study by Jin (2002) in assuming that CEOs ‘‘hold dis-
proportionally large amounts of stock and options in their firms and, consequently, to care about
both systematic and nonsystematic risk’’ and that ‘‘outside shareholders to be well-diversified
investors who value the firm as the market does’’ (Jin [2002], 32). In addition, earlier studies
(Jensen & Murphy [1990]; Hall & Liebman [1998]) show that the change in the market value of
CEOs’ managerial wealth is significantly larger than any annual performance-based compensa-
tion.
2. Caballero (1991) also argues that risk aversion has a role in investment decision. This
behavior could be a representation of different costing of the real option to defer investing
(McDonald & Siegel [1986]); Pindyck, [1988]; Dixit & Pindyck [1994]). A highly (in ordinal rank-
ing) risk-averse executive could, for example, defer investing in projects with relatively high
income volatility, while the same projects would appeal to executives with relatively lower risk
aversion. Diamond and Verrecchia (1982) model the relevance of nonsystematic risk in project
choice. ‘‘The nonsystematic risks are important to the stockholders because a risk averse manager
must bear some of them directly for incentive purposes and requires increased compensation for
bearing them’’ ([1982], 284).
AN EMPIRICAL ANALYSIS OF CEO RISK AVERSION 203
If it is true that relatively high-risk-averse CEOs act to reduce volatility, we

should observe a negative association between risk aversion and the volatility
of both earnings and operating cash flows.3
The sample used for empirical analysis is drawn from Compustat and Execu-
Comp databases for forty-four quarters covering the period 1992 to 2002. These
sources were supplemented by searching CompactD disclosure, especially for
missing data on CEOs’ age. Because the measurement of some variables requires
using lag and lead observations for up to twelve quarters, the empirical analysis
is carried out for the period 1993 to 2000 for 6,337 observations for analysis of
earnings volatility and for 5,334 observations for analysis of operating cash flow
volatility, with a different number of usable observations for different years,
depending on availability of data on all variables.
The empirical analysis is carried out in four stages. Alternative measures
of risk aversion are generated in the first two stages. Hypothesis testing is car-
ried out in the third stage. The fourth and final stage consists of several tests
of robustness. In the first stage, the assumption of power utility functions4
facilitated measuring the Arrow-Pratt DARA. The second stage of analysis
makes no assumptions about the CEOs’ utility over wealth, but it assumes that
CEOs self-select into compensation schemes that are consistent with their own
risk preferences (Lazear [1998]; Abdel-khalik [2003]). Managers with rela-
tively low risk aversion are expected to accept larger proportions of their com-
pensation to be made contingent on performance (such as stock options) as
compared with guaranteed pay (such as salary). Furthermore, as contingent
compensation increases, managers will tend to be more of risk takers because
the value of their stock options increases with increased volatility. Because
observed compensation is a function of exogenous labor market conditions, the
component that could be attributed to the individual’s risk aversion is a latent
variable and is estimated using CEOs’ demographic characteristics: age, gen-
der, tenure, and wealth. The choice of these variables is predicated on findings
in the finance and economics literature (e.g., Cohn, Lewellen, Lease, & Shlar-
baum [1975]); Friend & Blume [1975]; Donkers, Melenberg, & Van Söest
[2001]; Bajtelsmit & Bernasek [2001]), which examined determinants of risky
versus safe investment choices. For our purposes, age, tenure, percentage of
ownership, and managerial wealth are used as predictors of the latent risk
aversion metrics.5
3. A negative exponential utility function is often assumed, especially in analytical modeling,

for tractability. This form of utility function is concave, but assumed a Constant Absolute Risk Aver-
sion (CARA) in which all individuals are assumed to have the same degree of risk aversion over all
wealth conditions. CARA is, therefore, not acceptable for empirical analysis.
4. Other than constant exponential.
5. The variable ‘‘gender’’ is not used in the empirical analysis because of very low variability
since most CEOs in the data set are male.
Subject to the added assumption that the shapes of the estimated risk aver-
sion metrics are homogeneous across CEOs (i.e., two CEOs with similar wealth
endowments and characteristics are assumed to have the same degree of risk
aversion), the empirical findings consistently show negative association between
the degree of risk aversion and the volatility of each performance measure: earn-
ings and operating cash flows (conditional on other firm characteristics, size,
profitability, and growth). These results are obtained whether risk aversion is
measured as an Arrow-Pratt DARA or as a predicted latent index implicit in the
observed compensation structure.
In this type of analysis, selection bias may exist to the extent that omitted
selection variables affect the estimated results of volatility regressions. Analysis
of necessary conditions for selection bias shows lack of significance in affect-
ing the obtained results; that is, the error terms of the selection and structural
equations are not significantly correlated. Because the Inverse Mills Ratios
(IMR) is a function of the error terms, by construction, insignificant association
between these error terms implies the lack of significant selection bias, and the
need to estimate the coefficient on IMR does not arise (see, Heckman [1976];
Maddala [1983, 1991]; Heckman & Sedlacek [1990]; Greene [2003]). The final
stage of empirical analysis consists of robustness checks. The results are vali-
dated by a series of robustness checks for industry differences, firm size, and
CEO turnover.
The remainder of this paper consists of the following segments. Section 2
discusses the management’s motivation to reduce earnings volatility and presents
the statement of hypothesis. A discussion of measures of risk aversion is in Sec-
tion 3. Measurement of variables and specification of the regression models are
discussed in Section 4. Data selection and descriptive statistics are presented in
Section 5. Section 6 provides empirical estimates of the latent risk-aversion met-
rics. The results of estimating the regressions on volatility and of hypothesis test-
ing are in Section 7. Analysis of the significance of the selection bias using the
Heckman method is presented in Section 8. A summary of several robustness
and validation tests is presented in Section 9. Finally, a brief conclusion is in
Section 10.
2. Do Managers Attempt to Avoid Volatility?

2.1 Managerial Wealth as a Basis for Concern about Volatility
Assertions about the motives of managers in seeking to smooth volatility of
earnings have varied over the years from meeting investors demands to satisfying
executive compensation goals. Nothing on this issue, however, could have pre-
pared the Securities Exchange Commission (SEC) and the Office of Federal
Housing Enterprise Oversight (OFHEO) for the magnitude of the compensation-
induced income smoothing that took place at Fannie Mae and Freddie Mac. In
its extensive report of May 2006, OFHEO confirmed the SEC’s investigation of
manipulating the timing of reporting nearly $11 billion dollars of net income at
Fannie Mae alone. The report concludes that this manipulation of accounting
methods was motivated by income smoothing for considerations purely related to
executive compensation.6 Similar problems have plagued Freddie Mac to the
extent that the financial press referred to the firm for some time as ‘‘steady Fred-
die’’ (Wood [2003]). In this study, I add risk aversion and the wealth effect as
hypothesized motives for CEOs’ self-interest to smoothing volatility of earnings
and operating cash flows.
Because income smoothing or volatility is a measure of firm-specific (non-
systematic) risk, a question arises as to the relevance of this phenomenon to
investors who could not diversify nonsystematic risk. Mayshar (1981) offers an
equilibrium asset pricing model that provides the intuition for the relationship
between security prices and performance volatility, with the latter being a proxy
for nonsystematic risk. The Mayshar model is described as follows:
where Pj is the price of stock j; Rf is the risk free rate; E is the expectation oper-
ator; CFj is the jth firm performance (operating cash flows or earnings); CFm is
the cash flow (or earnings) for the market portfolio; p is the market price for
risk; is the volatility of the jth firm performance (earnings or operating
cash flows); and d is the ratio of diversification (that is bounded between 1
and 0 because short sale is not allowed). In this model, the degree of diversifi-
cation d is inversely related to transaction costs such that market prices are lin-
ear combinations of systematic and nonsystematic risk, with the latter being
presented by firm performance volatility . A high cost of diversification
(1d) implies that small performance volatility will have a negative effect on
market prices.
Empirical tests of this model are offered in Conine, Jensen, and Tamakrin.
(1989). Although the applicability of the Mayshar model to our setting is limited,
two issues arise. The first is the extent to which CEOs do not hold diversified
financial portfolios, and the second concerns the question of whether markets do
indeed price nonsystematic risk.
Concerning the first question, there is no known way to estimate the wealth
of CEOs short of administering a large survey. Given the interest of individuals
in their privacy, inquiring about their wealth raises sensitivity issues that reduces
the response rate and casts doubt on response validity. Only in cases in which a
government office or institutional support is made available can the collected in-
formation be considered credible. For example, Friend and Blume (1975) used
data collected by a Federal Reserve survey in the 1960s in which participants self-
reported estimates of their wealth. The same approach has been used in a recent
6. The report imposes $400 million fine; freezes the size of the loan portfolio of the company;
seeks restitution from the former CEO and other executives, and is considering criminal indictment.
study sponsored by the American Association of Retired Persons (Bajtelsmit

& Bernasek [2001]).7 This problem has led researchers (e.g., Himmelberg,
Hubbard, & Palia [1999]; Himmelberg & Hubbard [2000]; Jin [2002]) to use
managerial wealth (ownership in the firm) as a proxy for total financial
wealth.8 More specifically, I adopt the assumption invoked in these studies that
CEOs are usually heavily invested in the equity of the firms they manage and
that they face high transaction cost in altering the composition of their portfo-
lios because of constraints of their employment contracts, their desire to have
voting rights, and regulatory restrictions on their trading activities (e.g., prohib-
iting short sale, hedging, transferring stock options, or acting on inside informa-
tion).
The second issue is whether, in fact, capital markets price nonsystematic
risk. Some recent studies (e.g., Campbell, Lettau, Malkiel, & Su [2001]; Goyal &
Santa-Clara [2003]) show that idiosyncratic risk matters. The question for this
study, however, concentrates more on CEOs’ beliefs in regard to their ability to
influence market prices. As will be noted in the next section, managers, in general,
appear to resist earnings volatility because of their perception of adverse market
effects and their distrust of investors’ ability to decipher garbled information.
In other words, CEOs behave as if they live in a Mayshar world, a world in which
performance volatility negatively affects security prices. In fact, the RiskMetrics
Group asserts that ‘‘the corporate community is becoming increasingly aware of
the fact that earnings volatility can affect stock price valuation and shareholders
value’’ (1999, 3).9
2.2 Events Documenting Management Preference for Smooth Earnings

The noted concern with volatility is not a recent phenomenon, however. Sev-
eral decades earlier, accounting researchers tested hypotheses about management
efforts to dampen the temporal fluctuations of reported earnings (see Ronen &
Sadan [1981] for a review of the early literature). Furthermore, this line of
research continues in an effort to understand the manipulation motivated by goals
extending beyond smoothing of earnings (for literature review, see Healy &
Wahlen [1999]; Dechow & Skinner [2000]). In addition, management preference
7. The former study by Bajtelsmit and Bernasek (2001) was funded by the American Associa-
tion of Retired Persons, which allowed them to survey 11,596 individuals.
8. More important, managerial wealth is of direct relevance in this study because it is the com-
ponent of CEOs’ investment portfolios that entails high transaction cost (imposed by restricting trade)
and whose values will be affected by the levels of volatility of firm performance that CEOs
choose to report.
9. RiskMetrics Group separated from JPMorgan and became one of the most important advisors
in managing risk. www.riskmetrics.com
for smooth earnings could easily be discerned from the public documents of cor-
porate lobbying activities.10 Corporate lobbying of the regulators typically aims
to rescind or modify proposed accounting rules or provisions that have the poten-
tial of inducing variations in earnings. Historically, this intensive lobbying effort
has succeeded in changing the accounting standards being contested and has
allowed managers to attain their lobbying objectives, even if it comes back to
haunt them as in the case of Statement of Financial Accounting Standards
(SFAS) No. 87 regarding pension accounting. The interest in smoothing has
taken over management function to the extent that the reinsurance company Aon
Re, Inc. offers an annual award for the insurance company with least volatile
earnings.11
Additionally, and of more significance, is the fact that firms do utilize eco-
nomic resources to smooth earnings as witnessed by the rapid unexpected growth
in financial derivatives to hedge price and market risks with the main intent of
reducing earnings volatility (Stulz [1996]; Smithson [1998]; Barnett & Meul-
broek [2000]; Marrison [2002]). In fact, both SFAS No. 133 (1998) and Interna-
tional Financial Accounting Standard (IFRS) No. 39 measure hedge effectiveness
10. These documents are in the form of letters of comments that the Financial Accounting
Standards Board (FASB) has invited since 1973 and are made available at nominal fees. The letters
of comments on SFAS No. 13, for example, consist of five volumes, while the letters of comments
on SFAS No. 123 consist of nineteen volumes. Several cases document the impact of lobbying on
accounting changes that reduce volatility. Under intense pressure from the corporate sector, the FASB
replaced SFAS No. 8 (1975), concerning accounting for foreign currency translation, with SFAS No.
52 (1981), which shifted a component of the translation gains or losses to the equity section on the
balance sheet. At about the same time, the FASB was forced to replace SFAS No. 19 (1977), which
called for using the successful efforts method in accounting for the oil and gas industry, with SFAS
No. 39 (1980), which allowed companies in that industry to make their own choice of method and
flexibility to switch methods back and forth (Chen & Lee [1993]. Another well-publicized case is
SFAS No. 87 (1985) on accounting for pensions, Arthur Wyatt (then a new member of the FASB)
stated in his dissent that the provisions of the standard ‘‘because it is inherently a practical mecha-
nism to mitigate volatility’’ (SFAS No. 87 [1985], 27, emphasis added). More recently, upon issuance
of the Exposure Draft on accounting for stock options, intense lobbying forced the FASB to note that
‘‘the nature of the debate threatened the future of accounting standards setting in the private sector’’
(SFAS No. 123 [1995], par. 60, 25, emphasis added). Even after giving the management the choice
of method, corporate lobbying continued in an effort to reduce variation in the pro forma earnings,
which again forced the FASB to switch the choice of commencing amortization from vesting date (as
was suggested in the Exposure Draft) to options’ grant date because ‘‘the respondents’ overwhelming
opposition to vesting date and the potential resulting volatility in reported income’’(par. 158, 50, em-
phasis added). Following the discovery of malfeasance in granting stock options at Enron, World-
Com, and others in 2001 and 2002, the FASB has come full circle and required, in 2005, expensing
stock options (SFAS No. 123R [2005]).
11. The large reinsurance division of Aon, Corp. holds an annual conference to award recogni-
tion of property and liability insurance companies having the least volatile earnings. For example, the
award for 2002 was made on May 7, 2003, about which the Aon Corporation news release states this
policy ‘‘Once a year we want to recognize the companies that led their respective sectors in generat-
ing the least volatile earnings,’ said Michael Bungert, president and CEO of Aon Re, Inc.’’
by the potential for success in offsetting changes in prices.12 However, investing

in derivatives is a double-edged sword when it comes to earnings volatility,
because accounting for (trading or speculative) derivatives at their fair values
offers greater transparency and greater exposure of the underlying volatility. On
this issue, The Economist notes that European banks ‘‘have long ÔsmoothedÕ their
profits, to make themselves look stable. Booking financial instruments at fair
value would expose this practice for the lie it is—a prospect that sends shivers
through banks across Europe’’ (2004, 65).13 Nevertheless, the prospect of this
exposure has plunged the International Accounting Standards Board (IASB) into
a political crisis in Europe because
European banks and insurers, which use derivatives mainly to hedge interest
rate fluctuations, fear the [accounting] rule will unnecessarily make them
report huge swings in earnings. ‘‘This would create a false volatility’’ says
Sylvie Grillet-Brossler of the French Banking Federation. No one is protest-
ing more than the French Banks, which have enlisted President Jacques
Chirac and European Central Bank chief Jean-Claude Trichet to their cause.
The European Commission has said that unless the IASB gives in, it will
exclude derivatives accounting when it requires local companies to follow
the new standards (Business Week [2004], 54).
Not only have executives in European companies followed their counterparts
in the United States by emphatically revealing their preference for reporting
smooth earnings, but in light of this opposition, the European Union has also
backed down and modified its requirement regarding IFRS No. 39.
2.3 Hypotheses
The above discussion suggests that executives seek to smooth earnings and
operating cash flows because of the belief that volatility of performance mea-
sures negatively affects security prices. The research question addressed in this
study is the extent to which such belief depends on CEOs’ aversions to risk.
Thus, the hypothesis being tested in this study posits that CEOs’ aversions to
earnings volatility are consistent with their aversions to risk. More specifically,
12. Hedge effectiveness is measured by the derivative’s delta and is approximated by the
change in values (the 80/120 range, or by the squared correlation between the changes in values of
the hedge derivative and the hedged item). There is debate about the success of derivatives in achiev-
ing its smoothing goal. Stated differently, ‘‘the greater the risk retained by the firm, the lower the in-
surance premium and the greater the firm’s exposure to volatility in earnings and firm value’’
(Barnett & Meulbroek [2000], 5). The empirical evidence on effectiveness of hedging is mixed, how-
ever. Although Barton (2001) and Pincus and Rajgopal (2002) find that management uses a strategic
mix of hedging and accruals to smooth earnings volatility, Guay and Kothari (2003) find a relatively
small valuation effect.
13. This problem arises for the derivatives that do not qualify for hedge accounting and are
treated as speculative by posting changes in market values to earnings.
H1a (Null): Risk-aversion metrics of CEOs are not correlated with the vol-
atility of performance (earnings or operating cash flows) of
the firms they mange.
H1b (Alternative): Risk aversion metrics of CEOs are negatively related to the
volatility of performance (earnings or operating cash flows) of
the firms they mange.
3. Measurement of Risk Aversion

The usual assumption in agency and contracting theories is that managers (agents)
are risk averse, but empirical tests have thus far relied on numerical examples or simu-
lated data. For example, in modeling compensation portfolios, Lambert, Larcker, and
Verrecchia (1991) use constant absolute risk aversion of 4.5, which is the average index
reported in Friend and Blume (1975). Huddart (1994) assumes a constant relative risk
aversion (of 0.50), while Haubrich (1994) simulates numerical examples of DARA. As
in Friend and Blume (1975), Guiso and Paiella (2002) and Hartog, Ferrer-i-Carbonell,
and Jonker (2000, 2002) use the degree to which individuals allocate their capital invest-
ments between safe and risky projects to estimate risk aversion relative to the reference
group being studied. In this study, I use two approaches to measure risk aversion: (1) I
follow the assumption that CEOs have von-Neumann-Morgenstern-utility functions over
wealth and (2) I use specific demographics as instrumental variables under the assump-
tion that CEOs self-sort into compensation schemes that reflect their risk preferences.
3.1 Arrow-Pratt DARA

If CEOs aim to maximize utility over wealth instead of maximizing pure
wealth, the valuation effect of changes in performance volatility on their wealth
will depend on the shapes of their utility functions and the sizes of their wealth
endowments. One could assume any of the known power utility functions over
wealth from which risk-aversion measures could be derived. Two simple func-
tions are often used:14
(a) u1 ¼ ln mw
(b) u2 ¼ (mw)1/2
where mw is for managerial wealth. Risk aversion from either function is eval-
uated at , resulting in Arrow-Pratt DARA, APg(u1) ¼ 1/mw for the
1/2 3/2
first utility function, and ¼ (2(mw) / (4 mw ) for the second function.15
14. The utility function adopted by Diamond and Verrecchia (1982) is U(mw) ¼ 2 a2,
where a stands for effort. The negative exponential utility function often assumed in analytical work is
concave in wealth but assumes a constant risk aversion parameter.
15. The magnitudes of DARA are sensitive to the unit of measure (Arrow [1971]; Gollier [2001])
and tests are also performed based on the rank ordering of each, Rk(APg(u1) or Rk(APg(u2), because
ranking random variables is scale neutral but preserves order. In fact, the study by Jin (2002) follows
others in using the rank order of the variable.
Using either of these two functions for a cross-section of individuals assumes (1)
sufficiency—wealth is a sufficient indicator of risk preference (as is the case in
any Arrow-Pratt utility function), and (2) homogeneity—CEOs derive the same
level of utility from a given level of wealth.16 The hypothesis being tested sug-
gests that the wealth incentive for management to reduce volatility of reported
performance is likely to be more intense for the relatively high-risk-averse
CEOs.
3.2 Risk Aversion Implicit in Making Safe and Risky Choices

An alternative to specifying a particular utility function is to use the primi-
tive determinants of risk aversion that have been consistently documented in the
recent empirical literature: age, personal income, wealth, education, and gender.
The empirical evidence leading to these proxies has been gathered by administer-
ing large-scale cross-sectional economic surveys. One of the earliest studies in
this genre is by Friend and Blume (1975) in which they use the cross-section
data of the Federal Reserve Board Survey of 1962–1963 to estimate the propor-
tions of households’ personal portfolios invested in risky assets.17 Their results
show a pattern of increasing proportions of investment in risky assets as wealth
increases. But this effect is mitigated by the individual’s age because ‘‘age’’ has
the opposite effect: decreasing proportions of investment in relatively more risky
assets as investors get older. Friend and Blume conclude that either DARA or
Decreasing Relative Risk Aversion (DRRA) describes households’ attitudes to-
ward risk, depending on whether or not wealth includes human capital and the
value of residential homes.18
Contemporaneously, Cohn, Lewellen, Lease, and Shlarbaum (1975) pub-
lished findings similar to those of Friend and Blume using the compositions of
portfolios of 588 investors in a national brokerage firm.19 These findings have
been subsequently replicated by Morin and Suarez (1983) using Canadian data
and by Riley and Chow (1992). More recently, Hartog, Ferrer-i-Carbonell, and
Jonker (2000; 2002) utilize data collected about household investment preferen-
ces from more than 24,000 individuals to deduce Arrow-Pratt DARA; Donkers
16. Absolute risk aversion metrics of these two functions have similar properties.
17. Blume and Friend obtain results of DRRA when wealth excludes houses, cars, and human
capital. But when estimates of these assets are included in the measurement of wealth, they obtain
Constant Relative Risk Aversion (CRRA). Because the results are based on cross-section data, the
change from DRRA to CRRA could be an artifact of inflating the denominator, while keeping the nu-
merator unchanged.
18. Strictly speaking, Blume and Fiend (1975) and Cohn, Lewellen, Lease, and Shlarbaum
(1975) did not find DRRA. They use the term DRRA while in fact they found DARA for proportions
of investing in safe assets, which Lambert, Larcker, and Verrecchia (1991, 142) interpret as decreas-
ing proportional risk aversion.
19. Both articles actually analyze households’ ‘‘proportional’’ risk aversion as the fraction of
risky investments, but they referred to it as ‘‘relative’’ risk aversion. Cohn, Lewellen, Lease, and
Shlarbaum (1975) recognize the weaknesses associated with empirical estimates such as their own:
(1) human capital is ignored, (2) the estimate is cross-sectional rather than longitudinal, (3) tax and
debt incentives are ignored, and (4) there could be self-selection or self sorting of investors.
and van Söest (1999) and Donkers, Melenberg, and A. Van Söest (2001) sur-
veyed Dutch households seeking information about the importance of investing in
safe assets and about their expectation of high returns. Similarly, Guiso and Paiella
(2002) utilize data collected by the Bank of Italy in its 1995 Annual Survey of
Household Income and Wealth to evaluate risk aversion on the basis of the choice
between safe and risky investments. Two recent U.S. studies report similar findings
(Bajtelsmit & Bernasek [2001]; Bellante & Green [2004]).20 Collectively, these
studies conclude that risk aversion increases in age, and decreases in income,
wealth, and education, and that women are more risk averse than men.21
More recently, Aggarwal and Samwick (1999) examine the effect of equity
risk on executive compensation. Although risk aversion is parameterized in that
study, the empirical analysis focused on the systematic risk of the firms that
CEOs manage. The study by Moers and Peek (2000) is one of the first studies in
accounting that empirically examine risk aversion. They measure risk aversion
by the volatility of CEOs’ cash compensation (salary plus bonus) over a five-
year period, implying a self-sorting process in which high-risk-averse CEOs
could be attracted to firms that offer compensation contracts with relatively less
dependence on relative performance measures. Their empirical results are con-
sistent with this expectation, although the study could benefit further by perform-
ing analysis of the sensitivity of the findings to CEOs’ selectivity.
All these studies have two concepts in common that are of relevance in this
study: (1) that risk aversion is implicit in the choice of combinations of safe and
risky opportunities; and (2) that the risk aversion implicit in these choices could
be estimated as a latent variable using the observed choices and the individual’s
own demographic characteristics. These two concepts are applied in this study,
except that the combinations of relatively ‘‘safe’’ and relatively ‘‘risky’’ choices
are described by the observed compensation schemes. A fixed salary is the
‘‘safe’’ or sure component, while the contingent pay is the at-risk or ‘‘risky’’
component. Thus, the ratio of salary to total pay is the observed combination
variable reflecting risk aversion and other determinants (such as labor market
conditions). CEOs’ demographics are the predictors that will be used to estimate
the latent variable (+) capturing this implicit risk aversion.22
20. The former study by Bajtelsmit and Bernasek (2001) was funded by the American Associa-
tion of Retired Persons, which allowed them to survey 11,596 individuals.
21. Other psychological factors are introduced as determinants of risk aversion. For example,
Kowert and Hermann (1997) include traits such as ‘‘extraversion’’ and ‘‘impulsivity’’ as determinants
of risk aversion in international conflicts. These factors are not considered in this study. Halek and
Eisenhauer (2001) examine these variables in addition to race, willingness to migrate, intensity of
drinking alcohol, and the like. These factors are often obtained through interviews and are not used
in this study.
22. As noted before, ‘‘gender’’ and ‘‘education’’ are not feasible determinants in this study. Fur-
thermore, personal income should not be used as determinant because the dependent variable is a ra-
tio based on income components.
To estimate +, the following regression is estimated for each year:

0 1 2 3 4
ᏸ
where +, is the latent risk aversion metric, RCj,y is the ratio of salary to total
pay, A is for CEO age, T is for tenure on the job, PctOw is percentage of shares
owned, mw is managerial wealth, e is a residual error term, the indices j refer to
the jth CEO and t refers to the year.23 Risk-aversion metrics derived from this
analysis are compared against the Arrow-Pratt DARA both directly and indirectly
as will be reported below.
4. Measurement of Variables and Specification

of Regression Models
4.1 Measurement of Test Variables
The Arrow-Pratt metrics of risk aversion APg(u) are measured as
24
. For each year of analysis, CEO age (A) is measured at the begin-
ning of the year. CEOs’ managerial wealth (mw) is estimated by the sum of mar-
ket values of stock holdings, restricted stocks, exercisable options, and the
Black-Scholes value of granted options. Alternative measures are also used.25
CEOs’ personal income (PI) is measured by the sum of salary, bonus, long-term
incentive plan payment, and the compensation earned by exercising stock
options.26
The two measures used for firm performance volatility are as follows:
(a) ¼ the standard deviation of quarterly earnings; and
(b) ¼ the standard deviation of quarterly operating cash flows.
Each of these measures is estimated over twelve rolling quarters. For each
measure, two different temporal metrics are calculated: (1) prospective or for-
ward , which is the index of volatility measured over twelve quarters ahead
of any given year of analysis, and (2) historical index of volatility , which
23. Further validation of this analysis replaced the year by a random sample for estimation and
the results are used to predict + for all observations.
24. The single argument basis of the Arrow-Pratt risk aversion is a common criticism (Levy &
Levy [1991]).
25. A possible error arises from omitting other assets as well as from using Black-Scholes
values for stock options. For example, Huddart (1994) and Hall and Murphy (2002) highlight the
problem with the overstatement of the values of executive stock options to the executive in compari-
son to the value of a comparable traded option. This is due to restrictions on trade as well as to risk
aversion.
26. The noted measure total pay ¼ salary þ bonus þ long-term incentive plan payment þ com-
pensation earned by exercising stock options is replaced by other measures that also resulted in simi-
lar findings. In another version, the compensation from stock options is replaced by the Black-
Scholes value of granted options, but the results are unaltered in any significant way.
measures the volatility during the twelve quarters before the year of analysis.27
For any given period t, is the endogenous (dependent) decision variable,
whereas is assumed to be a given, exogenous control variable.
4.2 Measurement of Control Variables and Regression Models

In addition to that controls for historical and inherent patterns of per-
formance volatility, other (explanatory) control variables relate to the firms’ char-
acteristics are firm size, past profitability (ROE), and growth rates, both historical
as estimated by growth in sales sg and anticipated investment opportunities as
measured by market-to-book ratio m/b (DeFond & Park [1997]; Minton, Schrand,
& Walther [2002]; Jin [2002]). Thus, the linear models used in estimation take
two forms: the first model in eq. (3) uses Arrow-Pratt metrics, while the second
model in eq. (4) uses the latent metric of risk aversion + generated by estimating
the model in eq. (2). These two (volatility) regressions are expressed as follows:
where is prospective (forward) performance volatility measured over the

twelve quarters preceding the year of analysis; is historical earnings vola-
tility also measured over the twelve quarters following the year of analysis;
APg(u) is Arrow-Pratt risk aversion; size is log sales; ROE is the ratio of
accounting income to equity; sg is sales growth rate; m/b is the market to book
ratio; + is latent risk aversion metric measured in eq. (2) as an alternative to
APg(u); and e1 and e2 are residual error terms.
In these models, the estimated coefficients r1 in eq. (3) and a1 in
eq. (4) capture the component of (forward) performance volatility inherent in the
earnings series, which means that the remaining exogenous variables explain the
variation in the nonserial component of performance volatility.
5. Data and Summary Statistics

5.1 Sample
The data used in this study cover forty-four quarters from 1992 through
2002 and are collected from ExecuComp, Compustat, and CRSP databases.
27. Earnings or cash flow variation for most observations is based on twelve quarters, except
for year 2001 for which only eight quarters are used for the prospective volatility that required using
the eight quarters 2002 and 2003. The eight quarters are adopted for the first year (1993) when data
were not available for the entire past twelve quarters, and for the last year (2001) in calculating for-
ward variability because the data set ended in 2003.
CompactD database is used to complete and verify the accuracy of the observa-
tions retrieved from the three databases.28 Data editing and verification led to
omitting (for each year individually) firms that had missing observations on vari-
ables required for the analysis as well as excluding very small firms (assets
below $100 million or market value of equity below $30 million). Furthermore,
measurement of variables with the leads and lags required for constructing the
volatility variables trimmed the ending dates and reduced the usable sample to
the period from 1993 to 2000. The final number of firms in the sample that is
used in hypothesis testing ranges between 461 in 1993 and 913 in 2000, giving a
total of 6,337 firm-year observations for earnings volatility and 5,334 observa-
tions for volatility of operating cash flows. Finally, the analysis is carried out for
individual years and for randomly selected subsamples, because randomly
selected subsamples do not preserve the spurious relationship present in a pooled
sample.
5.2 Descriptive Statistics

Table 1 presents summary descriptive statistics for the variables used in the
study for 1999 as an illustrative example of the data structure. As these data
reveal, firm sizes vary widely as measured by sales volume ranging between $44
million and $32 billion, but the log transformation of sales (for firm size) with a
mean of 7.1 (median ¼ 7.0) and a standard deviation of 1.07. This transformed
variable has statistically insignificant skewness and kurtosis coefficients. CEOs’
managerial wealth averages $13.5 million with a standard deviation of 24 and a
range between $0.32 million and $168 million.29 However, the log transforma-
tion of managerial wealth is symmetrically distributed (insignificant skewness
and kurtosis coefficients) with a mean of 8.5 (median is 8.4) and a standard devi-
ation of 1.6. The market-to-book ratio ranges between 0.6 and 26 with an
average of 3.4 and a standard deviation of 2.28.
By definition, the risk aversion metric APg(u) is convex in managerial
wealth (Figure 1). Because of the similarity of the convexity of u1 ¼ ln mw, and
u2 ¼ , the results will be reported for u1 only. The mean of Arrow-Pratt
risk-aversion (DARA)30 metrics is about 0.46 and the standard deviation is 0.61.
This variable ranges between 0.005 and 3.1, although it should be emphasized
that the size of the index is a function of the scale used for wealth.
Performance volatility indexes, historical and forward , have
similar distributional properties. For earnings volatility , the mean (standard
deviation) is 18 (21) for historical index and is 25 (26) for the forward index.
The standard deviation is about 21 for both variables. The means (standard
28. Although ExecuComp commenced in 1992, data for that year are excluded from the analy-
sis because 1992 has more errors typical in initiating a new data set. Furthermore, the 1993 data form
the first year of differencing CEO compensation so that the empirical analysis begins with 1994.
29. These figures include the value of granted options and exclude very large outliers.
30. The measurement is scale sensitive.
FIGURE 1
Arrow-Pratt Decreasing Absolute Risk Aversion Metrics for u ¼ ln(w)
in Relationship to Wealth (in thousand dollars)
deviations) of cash flow volatility are 77 (95) for historical measures and
103 (114) for forward volatility.
Table 2 presents the matrix of Pearson correlation coefficients between rele-
vant variables. Not all of these variables are used in a single regression, but some
correlation coefficients attract interest. Firm size is measured by ln sales and is
used as an explanatory variable along with the historical volatility of performance
(earnings or cash flows ) in both regression models (3) and (4). But ln sales
and earnings volatility have a correlation of 0.51, while ln sales and have a
correlation coefficient of about 0.75. Similarly, for the same models, the correlation
coefficient between profitability (ROE) and market-to-book (m/b) ratio is 0.40.
These are potential sources of colinearity and will require further evaluation. Other
correlation coefficients of similar magnitudes are not relevant for the analysis
because these coefficients are between variables not included as explanatory varia-
bles in the same regression models. Of interest, however, are the correlation coeffi-
cients between historic and forward measures of volatility—they are 0.63 for
earnings volatility, and 0.87 for cash flow volatility.
TABLE 1
Descriptive Statistics for Basic Variables for a Sample Year
Variable Mean Median Std. Dev. Min. Max.

Volatility of earnings (forward) 25 16 26 0.52 141
Volatility of earnings (historical) 18 10 21 0.44 149
Volatility of operating cash flow (forward) 103 54 114 4.5 591
Volatility of operating cash flow (historical) 77 40 95 1.9 552
Sales (in million dollars) 2,675 1,114 4,204 44 32164
Log sales 7.1 7 1.07 3.8 10.4
ROE 0.16 0.146 0.17 0.82 1.64
Sales Growth 0.15 0.10 0.27 0.56 2.9
Market-to-Book 3.4 2.28 3.4 0.6 26
Salary-to-Total Pay 0.46 0.45 0.27 0.0007 1
CEO Age 55 55 0.72 36 83
CEO Wealth (in million dollars) 1,3520 4,666 24,378 332 193,663
Log CEO Wealth 8.5 8.4 1.4 5.8 12.3
Arrow-Pratt DARA 0.46 0.21 0.61 0.005 3.1
Latent Risk Aversion 0.49 0.52 0.12 0.0001 0.79
Notes: Forward volatility of performance (earnings or operating cash flows) is measured by the
standard deviation for the twelve quarters following the year of analysis. Historical volatility of perform-
ance (earnings or operating cash flows) is measured by the standard deviation for the twelve quarters
before the year of analysis. Arrow-Pratt DARA is measured as the negative of the ratio of the second
derivative to the first derivative of the utility function u (mw) ¼ log (mw) where mw stands for manage-
rial wealth. Latent Risk Aversion is measured as the predicted value of the ratio of salary to total pay
with age, tenure, and ownership as explanatory variables. All other terms are defined in the table.
6. Estimating Latent Risk Aversion Metrics

To develop measures of the latent risk-aversion index +, the linear model in
eq. (2) is estimated under several partitions of data: a year-by-year as well as for
several randomly selected samples. Additionally, both in-sample and out-of-
sample predictions are generated. The estimation results of the former analysis
are presented in Table 3.31 The fit of the regression models ranges between 0.13
(for 1993) and 0.27 (for 1999) and all have significant F-statistics (at p < 0.01)
and provide consistent results in that only wealth variables are statistically signif-
icant at p < 0.01. The latent risk aversion measure + is the predicted values of
the dependent variable, which is a function of pay structure (RC ¼ salary/total
pay). This prediction is carried out in two different ways: in sample and out of
sample. For the in-sample prediction, + is simply the expected values of the
regression function. For the out-of-sample prediction, the estimated regression
31. In this estimation, I used all observations for which the data are available on the variables
of eq. (2) irrespective of whether these observations have data on the other variables required for test-
ing the hypothesis.
TABLE 2
Pearson Correlation Coefficients
1 2 3 4 5 6 7 8 9 10 11 12 13 14
1. Volatility of earnings 1.00
(forward)
2. Volatility of earnings 0.60 1.00
(historical)
3. Volatility of operating 0.68 0.65 1.00
cash flow (forward)
4. Volatility of operating 0.63 0.71 0.87 1.00
cash flow (historical)
5. Sales 0.51 0.51 0.72 0.74 1.00
6. Log sales 0.54 0.55 0.73 0.75 0.77 1.00
7. ROE 0.02 0.009 0.10 0.06 0.03 0.09 1.00
8. Sales Growth 0.05 0.10 0.02 0.09 0.001 0.05 0.00 1.00
9. Market-to-Book 0.09 0.030 0.03 0.2 0.046 0.12 0.40 0.23 1.00
10. Salary-to-Total Pay 0.18 0.12 0.21 0.17 0.12 0.15 0.27 0.15 0.24 1.00
11. CEO Age 0.06 0.064 0.11 0.11 0.09 0.14 0.01 0.07 0.134 0.002 1.00
12. CEO Wealth 0.33 0.21 0.30 0.23 0.21 0.16 0.13 0.16 0.43 0.32 0.09 1.00
13. Log CEO Wealth 0.34 0.28 0.34 0.28 0.24 0.19 0.21 0.17 0.44 0.47 0.10 0.74 1.00
15. Arrow-Pratt DARA 0.25 0.21 0.25 0.22 0.17 0.14 0.19 0.13 0.28 0.39 0.07 0.37 0.80 1.00
16. Latent Risk Aversion 0.41 0.30 0.47 0.41 0.40 0.46 0.44 0.12 0.45 0.38 0.12 0.86 0.71 0.41
Notes: Forward volatility of performance (earnings or operating cash flows) is measured by the standard deviation for the twelve quarters following
the year of analysis. Historical volatility of performance (earnings or operating cash flows) is measured by the standard deviation for the twelve quarters
before the year of analysis. Arrow-Pratt DARA is measured as the negative of the ratio of the second derivative to the first derivative of the utility function
u (mw) ¼ log (mw) where mw stands for managerial wealth. Latent Risk Aversion is measured as the predicted value of the ratio of salary to total pay with
age, tenure, and ownership as explanatory variables. All other terms are defined in the table.
TABLE 3
OLS Estimation of Latent Risk Aversion (L) Implicit in CEO Compensation Structure
1993 1994 1995 1996 1997 1998 1999 2000 2001

Number of Observations fi 756 757 684 1157 1471 1571 1569 1536 1369
a0 Intercept 0.21 0.73 0.86 0.68 0.70 0.78 0.78 0.70 0.67
(t) (1.0) (9.9)a (10.0)a (12.8)a (13.2)a (14.6)a (15.0)a (13.1)a (10.1)a
a1 on Age 0.003 0.0006 0.0002 0.0008 0.0003 0.0005 0.0005 0.0009 0.0022
(t) (0.2) (0.5) (0.2) (0.9) (0.34) (0.6) (0.6) (0.9) (2.0)b
a1 on Tenure 0.005 0.002 0.002 0.0008 0.0003 0.0003 0.0007 0.001 0.001
(t) (1.46) (1.6) (1.6) (0.8) (0.3) (0.3) (0.07) (1.0) (1.1)
a1 on Percent Owned 0.0011 0.0005 0.0004 0.0005 0.0006 0.0006 0.0056 0.0006 0.0007
(t) (2.92)a (3.35)a (2.54)a (4.4)a (5.5)a (5.8)a (4.9)a (5.5)a (5.1)a
a1 on ln Managerial Wealth 0.09 0.038 0.051 0.03 0.04 0.041 0.04 0.042 0.04
(t) (9.2)a (11.6)a (9.6)a (13.6)a (15.1)a (19.4)a (20)a (19.1)a (12.7)a
F-statistics 25a 39a 28a 56a 73a 107a 106a 103a 56a
Adjusted R2 0.13 0.17 0.15 0.17 0.21 0.23 0.22 0.27 0.17
Notes: The dependent variable is RC ¼ Salary/ Total Pay.

The t-statistics between parentheses are corrected for heteroskedasticity
a b
, , and c connote statistical significance at p < 0.01, p < 0.05, and p < 0.10.
FIGURE 2
Out-of-Sample Latent Risk Aversion as Out-of-Sample Predictions,
1998–2001 Based on Coefficients Estimated for 1996 Data Only
coefficients in any year, or for any random sample, are used to predict + for the
entire sample. The correlation between the in-sample and out-of-sample predic-
tions ranges between 0.80 and 0.98. Therefore, for the purpose of this study,
using either prediction of + produces essentially the same results.
To appreciate the usefulness of this metric, two issues remain. The first
relates to the behavior of + given that the dependent variable is a ratio of pay
structure, and the second relates to the relationship between + and APg(u), that
is, the DARA metrics measured based on wealth alone. With respect to the first
issue, Figure 2 displays the behavior of + in relationship to CEO managerial
wealth. The pattern of the scatter plot is consistent with that obtained for DARA
and presented in Figure 1—both figures show that the derived risk aversion
measure is convex with respect to wealth. Given that the predicted values + for
the latent index is the ratio of fixed to total compensation RC, the similarity of
patterns is a partial validation of what + connotes. The two metrics are not per-
fect substitutes, however; the correlation between + and APg(u) for the plotted
data is 0.73.
7. Estimation and Testing Hypothesis

7.1 Hypothesis Testing Using Arrow-Pratt DARA
Testing the hypothesis aims to examine the extent to which CEOs’ risk aver-
sion affects the degree of the firm’s future performance volatility, which CEOs
will report beyond the variation inherent in the earnings-generating process.
Table 4 presents the results of estimating model (3), using Ordinary Least
Squares (OLS) with robust standard errors32 for the full sample (n ¼ 6,337) and
for each year separately.
The results in Table 4 show that adjusted R2 values for the estimated models
range between 0.43 for 1997 and 0.52 for 1993. The F-statistics obtained for
each year’s model estimate are statistically significant at p < 0.001. As would be
expected, forward volatility contains a persistent component of historical
or inherent volatility of earnings ; the annual coefficients r1 average about
0.39—slightly more than one-third of the expected earnings volatility is persis-
tent or inherent in the nature of business. The coefficient r3 on firm size is con-
sistently positive and significant at p < 0.01—larger firms have higher income
volatility. In all cases, the coefficients r5 and r6 on the growth variables sg and
m/b, respectively, are positive, but statistical significance is not consistent
throughout the years. Although these variables are used as controls, the signs
obtained for the coefficients are consistent with intuition: growth is a source of
variation in any series. Finally, the variance inflation factor (VIF) ranges between
1.26 and 1.49, suggesting absence of colinearity.
Of direct relevance in testing the hypothesis are signs and significance of
r2, the coefficient on risk aversion. Estimates of r2 are statistically significant at
p < 0.01 in every year and the obtained signs are consistently negative. The
importance of the risk-aversion variable relative to other determinants of earnings
volatility should be evaluated. As judged by the size of obtained t-statistics,
much of the model R2 values are derived from these three variables: (1) the
historical measure reflecting the persistent component of volatility; (2) firm size;
and (3) risk aversion. Ignoring the full sample, estimating partial R2 for each
variable in each year as [t2 / (t2 þ degrees of freedom)] would provide a measure
of the relative contribution of each of these variables to the model explanatory
power is estimated as follows:
Partial R2 Historical Volatility Firm Size Risk Aversion

Highest 0.083 0.125 0.033
Lowest 0.052 0.072 0.012
32. The corrections are made using the Huber-White test-corrected standard errors provided in
Stata.
TABLE 4
OLS Estimation of Quarterly Earnings Volatility (using DARA for CEO Risk Aversion)
Total
1993 1994 1995 1996 1997 1998 1999 2000 Sample
Number of Observations fi 461 682 693 756 848 907 905 913 6,337
r0 Intercept 24 20 25 29 20.2 19.7 24.4 26.0 25.8
(t) (5.7)a (5.60)a (7.11)a (3.47)a (6.3)a (6.4)a (7.72)a (8.85)a (16.5)a
r1 on : Historical Volatility 0.43 0.43 0.40 0.40 0.41 0.43 0.36 0.31 0.68
(t) (6.6)a (6.0)a (6.90)a (7.6)a (6.9)a (9.0)a (7.50)a (7.8)a (25.3)a
r2 on APg (u):Arrow-Pratt Risk 1.70 1.74 2.06 1.80 1.76 2.25 3.2 2.66 2.26
Aversion
(t) (2.35)a (2.84)a (2.72)a (2.5)a (2.3)a (4.12)a (5.60)a (3.5)a (7.78)a
r3 on s: Firm Size (ln sales) 5.22 4.49 5.47 6.2 4.8 4.9 5.8 5.53 5.3
(t) (7.70)a (7.30)a (9.31)a (10.9)a (9.5)a (9.9)a (11.3) (12.0)a (20.9)a
r4 on p : Profitability (ROE) 10.4 4.57 11.06 8.17 10.9 11.5 9.5 10.4 10.6
(t) (4.44)a (1.0) (3.3)a (1.84)c (3.4)a (3.4)a (3.4)a (4.1)a (6.9)a
r5 on g : Sales Growth 1.67 2.83 2.70 3.53 3.62 1.25 2.72 4.57 4.53
(t) (1.62) (1.54) (1.38) (1.65)c (2.2)a (0.9) (1.99)b (3.5)a (5.3)a
r6 on m/b: Market-to-Book 0.34 0.57 0.96 0.56 0.60 0.50 0.45 0.54 0.70
(t) (1.10) (1.56) (4.0)a (2.3)a (2.8)a (3.2)a (3.22)a (3.74)a (7.20)a
F-statistics 71a 69a 72a 88a 69a 106a 98a 84a 542a
Adjusted R2 0.52 0.49 0.50 0.49 0.43 0.47 0.46 0.45 0.49
Average VIF 1.46 1.41 1.41 1.41 1.33 1.37 1.34 1.26 1.39
Notes: The model estimated is where is prospective earnings volatil-

ity index measured by the standard deviation of quarterly earnings for twelve quarters ahead (in relationship to the year of analysis); is historical earnings
volatility index measured by the standard deviation of quarterly earnings for the preceding twelve quarters in relationship to the year of analysis; APg (u), is
the Arrow-Pratt metric of risk aversion measured for the utility function u1 ¼ ln mw with managerial wealth (mw) being measured in millions of dollars as
the sum of the market value of owned equity shares, of restricted stocks, of vested options, and the Black-Scholes value of granted options; s is for firm size
measured by ln sales; p is the firm’s rate of return on equity; g is the sales growth rate; m/b is the ratio of market-to-book values; and leverage is the ratio of
liabilities to total assets. The t-statistics are between parentheses.
a b
Thus, the contribution of CEOs’ risk aversion in explaining earnings volatil-

ity is, on average, about one-third of that of firm size, suggesting a greater
impact than one would have anticipated a priori.
Similar results are obtained in Table 5 in which performance volatility is
measured by the standard deviation of operating cash flows. The obtained R2 val-
ues for the estimated regression models are higher than those obtained for earn-
ings volatility, but as is the case with earnings, the three variables of historical
volatility, firm size, and risk aversion contribute large proportions of that explan-
atory power. In 1993, for example, historical volatility has a partial R2 value of
about 0.47. The coefficient a2 on the risk aversion variable is negative and sig-
nificant at p < 0.01 in every year except in 1993. Otherwise, the results using
operating cash flow volatility are consistent with those obtained for volatility of
earnings. Based on the findings of both types of analysis, it is likely the null
hypothesis will be rejected in favor of the alternative—that is, relatively lower
metrics of risk aversion are associated with relatively higher earnings volatility
and vice versa.
7.2 Hypothesis Testing Using Latent Risk Aversion

The results reported above are based on using Arrow-Pratt risk aversion
derived only from managerial wealth. The need to assume a particular form of
utility over wealth raised questions as to whether the findings are unique to the
assumed form of utility function and risk aversion (DARA). In this section, I use
the latent risk aversion metric + generated from estimating eq. (2) to repeat the
estimation and tests performed with DARA metrics. This is accomplished by esti-
mating regression model (4). The results are reported in Table 6.
The results behave in a consistent manner to those reported above: For each
year of estimation, the results show (1) a good fit (adjusted R2 values range
between 0.35 and 0.78); (2) significant F-statistics at p < 0.01; (3) significantly
positive coefficient on firm size; (4) significantly negative (at p < 0.01) coeffi-
cient on the latent risk aversion metric + for four of the eight years and at
p < 0.10 for one year; (5) partial R2 values for firm size range between 0.15 and
0.36, while the corresponding values for the latent risk aversion variable range
between 0.008 and 0.077; and (6) the variance inflation index (VIF) averages
below 2.7, suggesting absence of serious colinearity.
The results of the significance of the coefficients on latent risk aversion + are
not as strong as the results obtained using Arrow-Pratt metrics.33 These differences
disappear, however, when the volatility regressions are estimated after excluding
the collinear variables. It was noted above that firm size, ln sales, and the measure
of historical volatility ( ) are highly correlated. Additionally, as shown in
Table 2, the correlation coefficient between profitability (ROE) and market-
to-book (m/b) ratio is 0.545. In spite of low VIF statistics for colinearity, the
33. These results are not reported here but can be made available on request.
TABLE 5
OLS Estimation of Operating Cash Flow Volatility (using DARA for CEO Risk Aversion)
1993 1994 1995 1996 1997 1998 1999 2000 Total Sample
Number of Observations fi 383 571 576 530 709 766 776 723 5,334
a0 Intercept 86 70a 84 123 128 106 98 95 104
(f) (5.3)a (4.4)a (4.9)a (7.1)a (6.8)a (6.1)a (5.1)a (6.6)a (15.9)a
a1 on Historical Volatility 0.98 1.06 0.94 0.84 0.83 0.84 0.90 0.94 0.91
(f) (19.8)a (20.4)a (16.4)a (13.7)a (10.8)a (13.3)a (17.9)a (17.8)a (45.8)a
a2 on APg(u): Arrow-Pratt Risk Aversion 3.6 8.6 11.2 9.3 6.5 9.0 8.0 5.07 8.4
(f) (1.0) (3.63)a (4.7)a (2.9)a (3.11)a (5.0)a (3.1)a (2.1)a (8.8)a
a3 on s: Firm Size (In sales) 15.1 14.0 17.0 22.9 21.5 20.3 18.3 16.5 19.3
(f) (5.8)a (5.3)a (6.0)a (8.3)a (7.0)a (6.84)a (5.9)a (6.9)a (18.3)a
a4 on p: Profitability (ROE) 9.6 4.8 13.4 2.05 6.3 15.6 2.08 3.17 4.8
(f) (0.61) (0.34) (0.9) (0.2) (0.5) (1.1) (0.2) (0.5) (0.9)
a5 on Sales Growth 43.5 25.6 35.0 27.0 33.0 15.0 22.6 17.1 27.0
(f) (3.3)a (2.9)a (3.93)a (2.2)a (4.7)a (2.4)a (4.0)a (3.4)a (7.4)a
a6 on m/b: Market-to-Book 1.50 1.03 0.49 2.1 4.2 3.0 1.47 1.81 2.08
(f) (1.13) (0.96) (0.53) (2.3)a (4.2)a (4.85)a (2.8)a (3.2)a (6.63)a
F-statistics 194a 233a 201a 171a 136a 193a 246a 199a 1533
Adjusted R2 0.84 0.81 0.81 0.77 0.74 0.75 0.77 0.79 0.77
Adjusted R2 without the Historical Variable 0.57 0.56 0.59 0.54 0.55 0.57 0.59 0.57 0.56
Notes: The model estimated is where is prospective earnings volatil-

ity index measured by the standard deviation of quarterly operating cash flows for the twelve quarters ahead in relationship to the year of analysis; is
historical earnings volatility index measured by the standard deviation of quarterly operating cash flows for the preceding twelve quarters in relationship to
the year of analysis; APg(u), is the Arrow-Pratt metric of risk aversion measured for the utility function u1 ¼ lnw with CEO managerial wealth (w) measured
in millions of dollars as the sum of the market value of owned equity shares, of restricted stocks, of vested options, and the Black-Scholes value of granted
options; s is for firm size measured by ln sales; p is the firm’s rate of return on equity; g is the sales growth rate; m/b is the ratio of market-to-book values;
and leverage is the ratio of liabilities to total assets. The t-statistics are between parentheses.
a b
TABLE 6
OLS Estimation of Earnings Volatility on CEOs’ Latent Risk Aversion (+) Implicit in CEOs’ Compensation
(Dependent Variable: standard deviation of quarterly earnings based on twelve moving quarters)
1993 1994 1995 1996 1997 1998 1999 2000

Number of Observations fi 546 425 677 717 1003 783 782 768
g1 on hs Historical Volatility 0.78 0.67 0.87 0.71 1.03 0.89 0.52 0.46
(t) (6.5)a (5.0)a (8.8)a (9.7)a (9.4)a (16.4)a (8.2)a (6.4)a
g2 on + Latent Risk Aversion 2.7 20.3 38.6 31.5 1.09 4.8 44 34.1
(t) (0.4) (1.7)c (2.9)a (4.6)a (0.1) (0.7) (7.8)a (8.0)a
g3 on s: Size (ln sales) 4.3 3.8 3.1 6.5 6.2 4.6 5.7 6.9
(t) (3.8)a (3.1)a (3.0)a (6.6)a (4.9)a (4.3)a (7.6)a (7.5)a
g4 on p : Profitability (ROE) 15.3 14.6 40.3 28.0 22.9 10.9 4.9 5.1
(t) (2.3)a (2.0)a (3.7)a (3.3)a (1.6) (2.2)a (1.2) (1.2)
g5 on g : Sales Growth 2.7 11.4 6.0 19.1 8.9 1.07 6.1 7.47
(t) (1.95)a (1.6) (0.9) (4.5)a (2.5)a (0.4) (2.1)a (2.6)a
g6 on m/b: Market-to-Book 1.73 0.96 4.9 2.51 7.2 1.25 0.15 0.08
(t) (4.6)a (1.4) (6.4)a (4.5)a (9.0)a (3.2)a (0.6) (0.34)
F-statistics 593a 34a 2024a 2907a 1497a 2752a 207a 424a
Adjusted R2 0.76 0.49 0.89 0.89 0.88 0.89 0.61 0.64
Notes: The model estimated is where is prospective earnings volatility index

measured by the standard deviation of quarterly earnings for twelve quarters ahead (in relationship to the year of analysis); is the historical earnings vol-
atility index measured by the standard deviation of quarterly earnings for the preceding twelve quarters in relationship to the year of analysis; + is the latent
metric of risk aversion measured as predicted values generated from estimates of model eq. (2) and reported in Table 3; s is for firm size measured by ln
sales; p is the firm’s rate of return on equity; g is the sales growth rate; m/b is the ratio of market-to-book values; and leverage is the ratio of liabilities to
total assets. The t-statistics are between parentheses.
a b
existence of these relatively high correlation coefficients between these variables

raises questions about the effect on the estimated results. To examine this issue, the
linear regressions models (2) and (4) are reduced for reestimation after excluding
ln sales and m/b (market-to-book); thus, the explanatory variables used in this
reduced model are as follows: historical volatility (either or ), risk aver-
sion variable (either Arrow-Pratt APg(u) or latent index +), ROE, and sg (sales
growth rate). The reduced form models are estimated for each year separately.
A summary of the results of estimating four models (two for each volatility
measures and two for each risk aversion index) are summarized in Table 7 in
which the coefficients on the risk-aversion variables and the related t-statistics
are produced. These results are consistent with earlier findings using Arrow-Pratt
metrics—the coefficients on risk aversion are statistically significant (at p <
0.01) and negative in 31 of the 32 estimated regression coefficients.
7.3 Sensitivity of Findings to Measurement of Variables

In the above analysis, risk-aversion metrics are measured by some transforma-
tion of other variables. In the case of Arrow-Pratt measures, risk aversion (the sec-
ond derivative of utility with respect to wealth divided by the first derivative) ended
up to be the inverse of CEOs’ wealth, while in the second case, the Latent measure
of risk aversion is a ratio falling in the range [0, 1]. However, earnings and cash
flow volatility are measured by their standard deviations over moving twelve-
quarter periods. To assess the extent to which the findings are affected by the mone-
tary denomination of these variables, the above analysis is repeated using ranks.
Table 8 presents the results of estimating eq. (4) using rank for both, the
dependent (forward) variable, , and the explanatory, historical variable,
, as well as the rank of risk aversion (RAPg(u)). The results reported in
this table are consistent with earlier findings and inferences. In fact, the statistics
for model fit (adjusted R2 and F-statistics) have higher values than earlier models
and the t-statistics for levels of significance of individual variables are also
larger. For example, the t-statistics for the coefficient on historical volatility
range between 9.7 and 14.5 compared with the range of 6.6 to 9.0 in
Table 4. Similarly, the t-statistics on the rank of risk aversion (RAPg(u)) range
between 6.5 and 4.1 compared with the range of 5.6 and 2.3 shown in
Table 4. These results suggest that the findings are not sensitive to the scale
choice.34 Further analysis using ranks is discussed below.
34. Prior drafts of this study used the coefficient of variation of volatility variables instead of
the standard deviation. Further analysis shows that the coefficient of variation suffers from firm size
as a spurious variable in that both the standard deviation and the mean of quarterly earnings are func-
tions of firm size. Although the results using coefficient of variation measures are weaker, the find-
ings are not altered.
TABLE 7
Dealing with Colinearity Results of the Estimated Coefficients on the Risk Aversion Variable for Reduced
Models—Models That Exclude Firm Size (log sales) and Market-to-Book Values (m/b)
1993 1994 1995 1996 1997 1998 1999 2000

Models for Earnings Volatility
(A) r2, m on APg(u):Arrow-Pratt risk aversion 3.5 2.1 4.6 4.1 4.6 1.6 7.8 9.5
(t) (3.23)a (2.02)a (3.31)a (3.45 )a (3.64)a (1.42) (8.8)a (7.36)a
(B) a2, m on +, Latent risk aversion 29.1 12.0 41.4 32.0 49.7 22.8 38.8 50.6
(t) (2.86)a (1.52) (3.75)a (5.1)a (5.75)a (8.1)a (8.1)a (10.7)a
Models for Cash Flow Volatility

(C) a2, m on APg(u):Arrow-Pratt risk aversion 26.04 12.2 26.4 11.9 15.7 10.2 145 4.8
(t) (6.05)a (3.6)a (5.6)a (2.22)a (5.9)a (3.3)a (5.3)a (7.7)a
(D) a2, m on +, Latent risk aversion 154 103 208 121 80 73 90 36.6
(t) (6.1)a (3.9)a (6.9)a (6.2)a (3.6)a (3.8)a (5.4)a (2.5)a
Notes: Reduced models used for estimation are as models (3) and (4) after excluding firm size (log sales) and market-to-book values (m/b). The coeffi-
cient obtained from reduced regression models excludes two collinear variables: log sales and market to book.
(A) From regression on earnings volatility using Arrow-Pratt risk aversion.
(B) From regression on earnings volatility using the latent metric of risk aversion (+).
(C) From regression on operating cash flow volatility using Arrow-Pratt risk aversion.
(D) From regression on operating cash flow volatility using latent metric of risk aversion (+).
TABLE 8
OLS Estimation of Ranks of Quarterly Earnings Volatility (using Ranks of DARA for CEO Risk Aversion)
1993 1994 1995 1996 1997 1998 1999 2000

Number of Observations fi 478 706 724 800 909 960 961 940
Intercept 3408 1,075 1,635 1,496 508 643 1,246 1007
(t) (3.2)a (1.1) (1.7)c (1.8)c (0.7) (0.9) (14.5)a (4.5)a
: Rank of Historical Volatility 0.627 0.69 0.68 0.69 0.75 0.72 0.66 0.67
(t) (9.7)a (11.6)a (11.1)a (11.9)a (14.5)a (14.3)a (14.5)a (14.3)a
RAPg(u): Rank of Arrow-Pratt Risk 0.21 0.30 0.23 0.21 0.18 0.15 0.164 0.165
Aversion
(t) (4.1)a (6.5)a (5.6)a (5.3)a (5.0)a (4.3)a (4.9)a (4.6)a
s: Firm Size (In sales) 1,254 951 963 974 657 801 949 1,052
(t) (8.7)a (6.9)a (6.9)a (7.7)a (6.0)a (7.4)a (9.9)a (10.7)a
p : Profitability (ROE) 2,148 1,285 2,250 1,578 2,945 2,673 2,479 2,230
(t) (2.9)a (1.5) (3.6)a (1.33) (4.2)c (4.4)a (4.7)a (3.9)a
g : Sales Growth 567 1,351 930 583 911 704 909 1,300
(t) (2.18)a (2.2)a (2.3)a (1.40) (2.76)a (2.1)a (3.6)a (4.4)a
m/b: Market-to-Book 18.8 32 173 60 53.5 88 86 69
(t) (0.3) (0.5) (4.0)a (1.02) (1.2) (3.3)a (2.85)a (2.25)a
F-statistics 176a 242a 264a 278a 220a 242a 254a 241a
Adjusted R2 0.62 0.60 0.61 0.59 0.55 0.55 0.58 0.57
Notes: The model estimated is where is the rank of forward or pro-

spective earnings volatility index measured by the standard deviation of quarterly earnings for twelve quarters ahead (in relationship to the year of analysis);
is the rank of historical earnings volatility index measured by the standard deviation of quarterly earnings for the preceding twelve quarters in rela-
tionship to the year of analysis; RAPg (u), is the rank of Arrow-Pratt metric of risk aversion measured for the utility function u1 ¼ ln mw with managerial
wealth (mw) being measured in millions of dollars as the sum of the market value of owned equity shares, of restricted stocks, of vested options, and the
Black-Scholes value of granted options; s is for firm size measured by ln sales; p is the firm’s rate of return on equity; g is the sales growth rate; m/b is the
ratio of market to book values; and leverage is the ratio of liabilities to total assets. The t-statistics are between parentheses.
a b
7.4 Summary of Test Results

The above analysis is consistent with the alternative hypothesis that risk
aversion is likely to imply aversion to volatility of reported performance (earn-
ings or operating cash flows). The test results document a significant negative
association between earnings volatility and CEOs’ risk aversion. This result is
obtained when risk-aversion metrics are measured as a derivative of CEOs’
wealth alone, or as a latent index based on the personal choices CEOs make, free
of any assumption about the shapes of utility over wealth. To substantiate this
finding, each measure of risk aversion, the Arrow-Pratt DARA and the latent risk
aversion metric, is partitioned into deciles based on rank order from low to high.
Figures 3 and 4 illustrate the patterns of behavior of mean earnings volatility per
decile over the entire range. These figures provide two useful pieces of informa-
tion: (1) unconditional on other variables, earnings volatility is inversely related
to risk aversion, and (2) the observed inverse relationships have similar patterns
for the two risk measures.
8. Selection Bias
A concern for the validity of the research design adopted above is the possi-
ble presence of self-selection bias. It is natural to expect that CEOs self-select
into firms that fit their attitudes toward risk. Self-selection bias arises, however,
when the selection criteria consist of unrecognized random variables other than
those accounted for by the analysis. To examine the significance of such bias,
the Heckman method (Heckman [1976]; Maddala [1983]; Heckman & Sedlacek
[1990]; Greene [2003]) is applied. First, CEOs in the sample are partitioned by
managerial wealth (in relationship to the mean) into two groups of H (high) and
L (low), which is denoted DumLS. Next, two models are estimated jointly: (1) a
selection function that sorts CEOs into H or L, and (2) a linear function that
evaluates the determinants of performance volatility.
These models may be expressed as follows:
Selection Criterion
0 1 2 3 4 5
Volatility Regression
Without correction for selectivity:
With correction for selectivity

FIGURE 3
Earnings Volatility in Deciles of Risk Aversion, Using Arrow-Pratt DARA
FIGURE 4
Earnings Volatility per Deciles of Latent Risk Aversion
where IMR is Inverse Rills Ratio (i.e., the hazard rate) and all other variables are
as defined above.35 It should be noted that eq. (6b) is the functional form also
discussed in eq. (3), except for using IMR in lieu of risk aversion metrics and
CEOs’ demographics.
In the Heckman method, the selection bias coefficient l is a product of
the variance-covariance matrix of the error terms e and v1 such that l ¼
r(e, v1)s(e) for the truncated distribution. Therefore, selection bias would be
relevant only if each standard deviation of the selection model s(e) and the
correlation coefficient r(e, v1) is significantly different from 0 (see Maddala
[1983]; Greene [2003]). The results of the two-step estimation of the Heckman
method for each measure of performance volatility show that the obtained
r(e, v1) is not significantly different from 0 at conventional levels.36
9. Robustness Checks
9.1 Industry Analysis
The sample used for estimation and hypothesis testing come from all types
of industries. Different industries, however, face different competitive markets
and operate with different technologies, which could influence the extent to
which CEOs would allow their preferences to influence variation in earnings. To
examine the sensitivity of the findings to the industry factor, the regression
model (3) is estimated for each of ten industries, grouped by the two-digit Stand-
ard & Poor’s Industrial Classification Index (SPENDIX). Industry-by-industry
analysis shows statistically significant coefficients on risk aversion variables for
six of the ten industries (coded as two-digit SPENDIX). These are as follows: 15
(Industry Specialties); 20 (Machinery and Diversified Commercial); 25 (Con-
sumer Products and Services); 30 (Retailing); 45 (Networks, Computers, and
Software Industries); and 50 (Telecom and Wireless Service Companies). It is
surprising that industry 35 (Biotech, Health Care, and Health Services) falls out-
side this category. It is not surprising, however, that industry 55 (Electric and
Public Utilities) shows no results of any type. These results are consistent with
the a priori expectations that technology firms differ from public utilities. In the
latter industry, operating costs drive the revenues and competition is lower
because of their ability to operate under an umbrella of regulatory protection.
Including these public utility firms with other industries biases the evidence in
favor of the null hypothesis.
For further analysis for different industry grouping, Group A includes indus-
tries in SPINDEX 35 (Pharmaceuticals), 45 (Networks, Computers, and Software
Industries), and 50 (Telecom and Wireless Service Companies); Group B consists
35. In the selection function, log assets are used for firm size.
36. Given this result, reporting the detailed estimation of the Heckman model would be super-
fluous.
FIGURE 5
Rank of Earnings Volatility per Deciles of Latent Risk Aversion
of firms in Group A plus industries in SPINDEX 20 (Machinery and Diversified

Commercial); Group C includes all firms; Group D includes all firms with R&D
(research and development)] expenditures less than 4 percent of sales (chosen
arbitrarily); and Group E is for firms with R&D expenditures greater than 4 per-
cent of sales. Except for firms in industry Group D (R&D < 0.4 of sales), the
coefficients on the rank of risk aversion, Rk APg (u), are negative and signifi-
cant (at p < 0.01).
9.2 Extent of Options Holding

As noted in the use of latent risk-aversion metrics that aversion is partially
induced—CEOs with large holdings of stock options are likely to have greater
preference for volatility, because higher volatility implies higher options values.
A question arises as to whether the findings reported above would hold for dif-
ferent levels of risk aversion. Different tests have been performed to examine
this issue. The findings varied, however, depending on the measurement of
options holdings. When ‘‘granted options’’ are used for classification, the results
did not hold for the lower quintile. When the sum (SuOp) of all options granted,
held and exercised, is used for classification, the results did not hold for the low-
est decile, but they were consistent with the above findings for the remainder of
the sample. To provide evidence on this result, Table 9 presents the results of
regressing forward volatility of earnings ( ) on historical volatility ( ) and

Latent risk aversion (+) for each quartile of the sum of options (SuOp) for year
1999. The coefficient on risk aversion for each quartile is negative and statisti-
cally significant at p < 0.01.
9.3 Other Factors and Replication

Additional robustness checks are carried out to evaluate the sensitivity of the
findings to the duration of CEO tenure and the extent of holding stock options.
Finally, these results are replicated with similar conclusions and findings for the
second-highest paid executive who is also a board member.37
10. Conclusion
CEOs invest disproportionately large shares of their wealth in the equities of
the firms they manage. They face restrictions on transferring or reallocating these
investments to other assets. They are therefore unable to fully diversify their
financial assets (Jin [2002]), which exposes them to the valuation effects of firm-
specific risk. This study tests the hypothesis that CEO aversion to performance
volatility (earnings or cash flow) is consistent with their aversion to risk.
Risk aversion is measured in two ways: (1) the Arrow-Pratt DARA, which
derives from assumptions about utility over wealth; and (2) the latent risk aver-
sion metric, which derives from CEOs’ preferences for relatively safe and rela-
tively risky choices. The mix of salary and at-risk pay is used to capture these
choices. CEOs’ demographics are adopted as explanatory variables to predict the
latent risk aversion that is implicit in the choices of pay mix. The two types of
risk-aversion measures take similar shapes in relationship to CEOs’ managerial
wealth and to the test variables (volatility of earnings and operating cash flows).
Consistently, the empirical findings show that volatility of performance meas-
ures are negatively related to CEOs’ risk aversion for each year of analysis during
the period 1993 to 2000. The results are sustained whether unconditional on other
factors, conditional on control variables (firm size, growth, and profitability), or on
the historic patterns of volatility inherent in the earnings or cash flow series. Anal-
ysis of colinearity and selection bias did not alter the findings. Industry analysis,
however, shows that the results hold for most, but not all, industries.
The findings suggest the possible existence of a wealth effect as a self-inter-
est factor motivating CEOs to seek smoothing earnings over time. Although the
results are reasonably strong and consistent, no causal inferences could be made.
Attempts to sequence the event analysis and volatility as an outcome by using
the lead and lag volatility measures do not allow for satisfying the required tem-
poral sequence of events and outcomes. Furthermore, more work needs to be
done to exclude other possible causes. A stronger case could be made for
37. These results can be obtained from the authors on request.

TABLE 9
OLS Regression Results for Different Partitions
of Options Using SuOp for Year 1999
(SuOp is the sum of the number of granted options, options held, and options exercised during the
year)
First Quartile Second Quartile Third Quartile Fourth Quartile
Sample Size 414 259 109 175
Intercept 19.45 28.4 32.4 35.7
(t) (3.7)a (5.5)a (4.65)a (5.9)a
Historical Volatility 0.88 0.81 0.57 0.72
(t) 7.0)a 8.4a (5.5)a (9.3)a
Latent Risk Aversion 20.6 40.1 42.0 54.8
(t) (2.2)a (3.7)a (2.45)a (3.6)a
F-statistic 24.5a 40a 17a 56a
Adjusted R2 0.36 0.50 0.32 0.48
Notes: The t-statistics are between parentheses.

a b
, , and c connote statistical significance at p < 0.01, p < 0.05, and p < 0.10
causality if future work would show that CEOs do, in fact, live in a world of the
Mayshar model. This might call for controlled experimentation, if such experi-
ments could be designed to entice experienced executives to participate. For the
time being, this study should be viewed as an analysis of association between
risk aversion and firm-specific risk and contributes to the literature by showing
wealth effects as a possible motive to smooth performance volatility.
REFERENCES
Abdel-khalik, A.R. 2003. ‘‘Self Sorting, Incentive Compensation and Human Capital Assets.’’ The
European Accounting Review 12(4): 661–697.
Aggarwal, R.K., and A. Samwick. 1999. ‘‘The Other Side of the Tradeoff: The impact of Risk on
Executive Compensation.’’ Journal of Political Economy 107: 65–105.
Arrow, K.J. 1971. Essays in the Theory of Risk Bearing. Chicago: Markham.
Bajtelsmit, Vickie L., and Alexandra Bernasek. 2001. ‘‘Risk Preferences and the Investment
Decisions of Older Americans.’’ Research Report No. 2001-11. Washington, D.C.: American
Association of Retired Persons.
Barnett, Jonathan, and Lisa Meulbroek. 2000. ‘‘Honeywell, Inc. and Integrated Risk Management.’’
Case No. 9-200-036, Harvard Business School.
Barton, Jan. 2001. ‘‘Does the use of Financial Derivatives Affect Earnings Management Decisions?’’
The Accounting Review 76: 1–26.
Beatty, A., B. Ke, and K. Petroni. 2002. ‘‘Earnings Management to Avoid Earnings Declines across Pub-
licly and Privately Held Banks.’’ The Accounting Review 77 (July): 547–570.
Bellante, Don, and Carole A. Green. 2004. ‘‘Relative Risk Aversion among the Elderly.’’ Review of
Financial Economics 13: 269–281.
Bowen, Robert M., Shivaram Rajgopal, and Mohan Venkatachalam. 2005. ‘‘Accounting Discretion, Cor-
porate Governance and Firm Performance.’’ Working paper, Duke University.
Capell, K. & David Henry. ‘‘When Bankers Keep Saying Non; The French are Thwarting the Adop-
tion of Global Accounting Standards.’’ Business Week. Issue 3872, March 1, 2004, pg. 54–55
Caballero, Ricardo J. 1991. ‘‘On the Sign of the Investment-Uncertainty Relationship.’’ American
Economic Review 81: 163–183.
Campbell, J., M. Lettau, B. Malkiel, and Y. Su. 2001. ‘‘Have Individual Stocks Become More Vola-
tile? An Empirical Exploration of Idiosyncratic Risk.’’ The Journal of Finance 56: 1–43.
Chen, Kevin S. W. and C. W. Jevons Lee. 1993 ‘‘Financial Ratios and Corporate Indurance: A Case
of the Oil and Gas Industry.’’ Contemporary Accounting Research. Vol 9: 667–695
Cohn, Richard A., Wilbur G. Lewellen, Ronald C. Lease, and Gary G. Schlarbaum. 1975. ‘‘Individual
Investor Risk Aversion and Investment Portfolio Composition.’’ The Journal of Finance 30:
605–620.
Conine, Thomas E., Oscar W. Jensen, and Maurry Tamakrin. 1989. ‘‘On Optimal Production and the
Market to Book Ratio Given Limited Shareholder Diversification.’’ Management Science 35:
1004–1013.
Dechow, Patricia M., and D.J. Skinner. 2000. ‘‘Earnings Management: Reconciling the Views of
Accounting Academics, Practitioners, and Regulators.’’ Accounting Horizons 14: 235–250.
DeFond, Mark L., and Chul W. Park. 1997. ‘‘Smoothing Income in Anticipation of Future Earnings.’’
Journal of Accounting and Economics 23: 115–139.
Diamond, Douglas W., and Robert E. Verrecchia. 1982. ‘‘Optimal Managerial Contracts and Equilib-
rium Security Prices.’’ The Journal of Finance 37: 275–287.
Dixit, Avinash, and Robert S. Pindyck. 1994. Investment Under Uncertainty. Princeton, N.J.:
Princeton University Press.
Donkers, Bas, Bertrand Melenberg, and A. Van Söest. 2001. ‘‘Estimating Risk Attitudes Using Lotteries:
A Large Sample Approach.’’ The Journal of Risk and Uncertainty 22: 165–195.
Donkers, Bas, and A. Van Söest. 1999. ‘‘Subjective Measures of Household Preferences and Finan-
cial Decisions.’’ Journal of Economic Psychology 20: 613–642.
The Economist. 2004. ‘‘Fair’s Fair.’’ 8365 (370): 65, March 6.
Financial Accounting Standards Board. 1973–2006. Statements of Financial Accounting Standards.
Available at www.fasb.org.
Friend, Irwin, and Marshall E. Blume. 1975. ‘‘The Demand for Risky Assets.’’ The American
Economic Review 65: 900–922.
Fundenberg, D., and Tirole, J. 1995. ‘‘A Theory of Income and Dividends Smoothing Based on
Incumbency Rents.’’ Journal of Political Economy 13: 75–93.
Gollier, Christian. 2001. The Economics of Risk and Time. Cambridge, MA: MIT Press.
Goyal, Amit, and Pedro Santa-Clara 2003. ‘‘Idiosyncratic Risk Matters!’’ The Journal of Finance 58:
975–1007.
Greene, William. 2003. Econometric Analysis, 5th ed. New York: Macmillan.
Guay, Wayne, and S.P. Kothari. 2003. ‘‘How Much Do Firms Hedge with Derivatives?’’ Journal of
Financial Economics 70: 423–461.
Guiso, Luigi, and Monica Paiella. 2002. ‘‘Risk Aversion, Wealth and Background Risk.’’ Working
paper, Bank of Italy and University of Sassari, Italy.
Halek, Mark, and Joseph G. Eisenhauer. 2001. ‘‘Demography of Risk Aversion.’’ The Journal of Risk
and Insurance 68: 1–24.
Hall, Brian J., and Kevin J. Liebman. 1998. ‘‘Are CEOS Really Paid Like Bureaucrats?’’ Quarterly
Journal of Economics 113: 653–691.
Hall, Brian J., and Kevin J. Murphy. 2002. ‘‘Stock Options for Undiversified Executives.’’ Journal of
Accounting and Economics 33: 3–42.
Hartog, Joop, Ada Ferrer-i-Carbonell, and Nicole Jonker. 2000. ‘‘On a Simple Measure of Individual
Risk Aversion.’’ Working Paper Series 363, CESifo, Munich, Germany.
Hartog, Joop, Ada Ferrer-i-Carbonell, and Nicole Jonker. 2002. ‘‘Linking Measured Risk Aversion to
Individual Characteristics.’’ Kyklos 55: 3–26.
Haubrich, Joseph G. 1994. ‘‘Risk Aversion, Performance Pay, and the Principal-Agent Problem.’’ The
Journal of Political Economy 102: 258–276.
Healy, P.M., and J.M. Wahlen. 1999. ‘‘A Review of the Earnings Management Literature and its
Implications for Standard Setting.’’ Accounting Horizons 13: 365–383.
Heckman, James. 1976. ‘‘The Common Structure of Statistical Models of Truncation, Sample Select-
ing and Limited Dependent Variables and a Sample Estimation of Such Models.’’ Annals of
Economics and Social Measurements 5: 475–492.
Heckman, James J., and Guilherme L. Sedlacek. 1990. ‘‘Self-Selection and the Distribution of Hourly
Wages.’’ Journal of Labor Economics 8: S329–S363.
Himmelberg, Charles P., and Robert Glenn Hubbard. 2000. ‘‘Incentive Pay and the Market for CEOs:
An Analysis of Pay-for-Performance Sensitivity.’’ Working paper, Columbia University.
Himmelberg, C.P., R.G. Hubbard, and D. Palia. 1999. ‘‘Understanding the Determinants of Manage-
rial Ownership and the Link between Ownership and Performance.’’ Journal of Financial
Economics 53: 353–383.
Huddart, Steven. 1994. ‘‘Employee Stock Options.’’ Journal of Accounting and Economics 18: 207–231.
Jensen, Michael C., and Kevin J. Murphy. 1990. ‘‘Performance Pay and Top-Management Incen-
tives.’’ Journal of Political Economy 98: 225–264.
Jin, Li. 2002. ‘‘CEO Compensation, Diversification, and Incentives.’’ Journal of Financial Economics
66: 29–63.
Kowert, Paul, and Margaret Hermann. 1997. ‘‘Who Takes Risks? Daring and Caution in Foreign Pol-
icy Making.’’ The Journal of Conflict Resolution 41: 611–637.
Lambert, Richard A., David F. Larcker, and Robert E. Verrecchia. 1991. ‘‘Portfolio Considerations in
Valuing Executive Compensation.’’ Journal of Accounting Research 29: 129–149.
Lazear, E.P. 1998. Personnel Economics for Managers. New York: Wiley.
Leuz, C., D. Nanda, and P. Wysocki. 2003. ‘‘Investor Protection and Earnings Management: An
International Comparison.’’ Journal of Financial Economics 69: 505–527.
Levy, Haim and Ariel Levy. ‘‘Arrow-Pratt measures of Risk Aversion: The Multivariate Case.’’ Inter-
national Economic Review 33 (1991): 891–898.
Maddala, G.S. 1983. Limited Dependent and Qualitative Variables in Econometrics. New York: Cam-
bridge University Press.
Maddala, G.S. 1991. ‘‘A Perspective on the Use of Limited-Dependent and Qualitative Variables
Models in Accounting Research.’’ The Accounting Review 66: 788–807.
Marrison, Chris. 2002. The Fundamentals of Risk Measurement. New York: McGraw Hill.
Mayshar, Joram. 1981. ‘‘Transaction Costs and the Pricing of Assets.’’ The Journal of Finance 36:
583–597.
McDonald, Robert, and Daniel Siegel. 1986. ‘‘The Value of Waiting to Invest.’’ Quarterly Journal of
Economics 101: 707–728.
Minton, Bernadette A., Catherine M. Schrand, and Beverly Walther. 2002. ‘‘The Role of Volatility in
Forecasting.’’ Review of Accounting Studies 7: 195–215.
Moers, Frank, and Peek, Erik. 2000. ‘‘An Empirical Analysis of the Role of Risk Aversion in Execu-
tive Contracts.’’ Working paper, Maastricht University, The Netherlands.
Morin, R.A., and A.F. Suarez. 1983. ‘‘Risk Aversion Revisited.’’ The Journal of Finance 38: 1201–1216.
OFHEO. 2006. Report of the Special Examination of Fannie Mae, i–328. Washington, D.C.: Office
of Federal Housing Enterprise Oversight.
Perrino, Robert, Allen M. Poteshman, and Michael S. Weisbach. 2002. ‘‘Measuring Investment Dis-
tortions When Risk-Averse Managers Decide Whether to Undertake Risky Projects.’’ NBER
Working Paper 8763, National Bureau of Economic Research, Cambridge, MA.
Pincus, Morton, and Shivaram Rajgopal. 2002. ‘‘The Interaction between Accrual Management and
Hedging: Evidence from Oil and Gas Firms.’’ The Accounting Review 77: 127–160.
Pindyck, Robert S. 1988. ‘‘Irreversible Investment, Capacity Choice, and the Value of the Firm.’’
American Economic Review 78: 969–985.
Pratt, J.W. 1964. ‘‘Risk Aversion in the Small and in the Large.’’ Econometrica 32: 122–136.
Riley, W.B., and K.V. Chow. 1992. ‘‘Asset Allocation and Individual Risk Aversion.’’ Financial
Analysts Journal 48: 32–37.
RiskMetrics Group. 1999. Corporate Metrics: The Benchmark for Corporate Risk Management. New
York: RMG.
Ronen, Joshua, and Simcha Sadan. 1981. Income Smoothing, Objectives, Means and Implications.
Reading, MA: Addison-Wesley.
Sharpe, William F. 1964. ‘‘Capital Asset Prices: A Theory of Market Equilibrium under Conditions
of Risk.’’ The Journal of Finance 19: 425–442.
Smithson, Charles W. 1998. Managing Financial Risk: A Guide to Derivative Products, Financial
Engineering, and Value Maximization, 3rd ed. New York: Irwin.
Stein, Jeremy C. 1996. ‘‘Rational Capital Budgeting in an Irrational World.’’ The Journal of Business
69: 429–455.
Stulz, Rene. 1996. ‘‘Rethinking Risk Management.’’ Journal of Applied Corporate Finance 9: 8–24.
Trueman, Brent, and Titman, Sheridan. 1988. ‘‘An Explanation for Accounting Income Smoothing.’’
Journal of Accounting Research 26: 127–139.
Wood, Duncan. 2003. ‘‘Steady Freddie and the Fear of Earnings Volatility.’’ Risk News. Available at
www.erisk.com (accessed August 1, 2003).

An Empirical Analysis of CEO Risk Aversion and The Propensity To Smooth Earnings Volatility

Uploaded by

Copyright:

Available Formats

An Empirical Analysis of CEO Risk Aversion and The Propensity To Smooth Earnings Volatility

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

An Empirical Analysis of CEO Risk Aversion and The Propensity To Smooth Earnings Volatility

Uploaded by

Copyright:

Available Formats

An Empirical Analysis of CEO Risk

Aversion and the Propensity to Smooth

Volatility of firm performance is one measure of firm-specific risk that

Keywords: earnings volatility, firm-specific risk, management expense pref-

*College of Business, University of Illinois at Urbana-Champaign

If it is true that relatively high-risk-averse CEOs act to reduce volatility, we

3. A negative exponential utility function is often assumed, especially in analytical modeling,

2. Do Managers Attempt to Avoid Volatility?

study sponsored by the American Association of Retired Persons (Bajtelsmit

2.2 Events Documenting Management Preference for Smooth Earnings

by the potential for success in offsetting changes in prices.12 However, investing

3. Measurement of Risk Aversion

3.1 Arrow-Pratt DARA

3.2 Risk Aversion Implicit in Making Safe and Risky Choices

To estimate +, the following regression is estimated for each year:

4. Measurement of Variables and Specification

4.2 Measurement of Control Variables and Regression Models

where is prospective (forward) performance volatility measured over the

5. Data and Summary Statistics

5.2 Descriptive Statistics

Variable Mean Median Std. Dev. Min. Max.

6. Estimating Latent Risk Aversion Metrics

1993 1994 1995 1996 1997 1998 1999 2000 2001

Notes: The dependent variable is RC ¼ Salary/ Total Pay.

7. Estimation and Testing Hypothesis

Partial R2 Historical Volatility Firm Size Risk Aversion

Notes: The model estimated is where is prospective earnings volatil-

Thus, the contribution of CEOs’ risk aversion in explaining earnings volatil-

7.2 Hypothesis Testing Using Latent Risk Aversion

Notes: The model estimated is where is prospective earnings volatil-

1993 1994 1995 1996 1997 1998 1999 2000

Notes: The model estimated is where is prospective earnings volatility index

existence of these relatively high correlation coefficients between these variables

7.3 Sensitivity of Findings to Measurement of Variables

1993 1994 1995 1996 1997 1998 1999 2000

Models for Cash Flow Volatility

1993 1994 1995 1996 1997 1998 1999 2000

Notes: The model estimated is where is the rank of forward or pro-

7.4 Summary of Test Results

With correction for selectivity

of firms in Group A plus industries in SPINDEX 20 (Machinery and Diversified

9.2 Extent of Options Holding

regressing forward volatility of earnings ( ) on historical volatility ( ) and

9.3 Other Factors and Replication

37. These results can be obtained from the authors on request.

Notes: The t-statistics are between parentheses.

You might also like