Managerial Optimism and Earnings Smoothing

Journal of Banking & Finance 41 (2014) 283–303
Contents lists available at ScienceDirect
Journal of Banking & Finance

journal homepage: www.elsevier.com/locate/jbf
Managerial optimism and earnings smoothing

Christa H.S. Bouwman ⇑
Case Western Reserve University, Cleveland, OH 44106, USA
Wharton Financial Institutions Center, Philadelphia, PA 19104, USA
a r t i c l e i n f o a b s t r a c t
Article history: This paper empirically examines how CEO optimism affects earnings smoothing and earnings surprises.
Received 13 August 2012 The main finding is that optimistic managers smooth earnings more than rational managers and are asso-
Accepted 23 December 2013 ciated with smaller (in absolute value) earnings surprises. A possible theoretical explanation is offered for
Available online 7 January 2014
these results based on a combination of the ‘‘torpedo effect,’’ the innate behavior of optimists, and the risk
of litigation/prosecution for over-reporting earnings.
JEL classification: Ó 2013 Elsevier B.V. All rights reserved.
M41
M43
D80
Keywords:
Smoothing
Earnings surprise
Earnings management
Behavioral
1. Introduction (Graham et al., 2005).1 However, the degree of earnings smoothing

varies in the cross-section of firms. This has led to research that has
A well-known stylized fact in the literature is that managers en- uncovered several factors that help explain cross-sectional varia-
gage in earnings smoothing: they report earnings that are some- tions in earnings smoothing.
times higher than economic earnings and sometimes lower (see, Numerous papers have tried to explain differences in the degree
e.g., Beidleman, 1973; Lev and Kunitzky, 1974; Ronen and Sadan, of earnings smoothing across firms. While earlier contributions
1981; Hand, 1989; Fudenberg and Tirole, 1995; Barth et al., tended to focus on firm-specific attributes, more recent papers
1999; Goel and Thakor, 2003; Leuz et al., 2003; Lang et al., 2006; have analyzed attributes of decision makers. For example, Healy
Myers et al., 2007). Recent survey evidence provides further confir- (1985) and Bergstresser and Philippon (2006) examine the impact
mation that managers actively smooth earnings, as evidenced by a of executive compensation on earnings smoothing. Klein (2002)
quote from an interviewed CFO: ‘‘businesses are much more and Bowen et al. (2008) focus on the characteristics of boards of
volatile than what their earnings numbers would suggest’’ directors to understand the issue. Ge et al. (2011) link CEO personal
1
The popular press tends to view accounting discretion, including earnings
smoothing, as a device used by self-interested rent-seeking managers to manipulate
earnings. See, for example, the following quote from Fortune (1997): ‘‘If Microsoft is
the archetype of a hugely successful company trying to tone its earnings down so
people don’t get their expectations too high, Boston Chicken bespeaks an altogether
different and more common phenomenon. It is a business that isn’t successful yet but
has used accounting to help convince investors that it already is, or at least will be
soon.’’ The academic literature is divided on the question whether managers use
accounting discretion, including earnings smoothing, to efficiently maximize share-
holder value (e.g., Ronen and Sadan, 1981; Chaney and Lewis, 1995) or to
opportunistically make themselves better off at the expense of shareholders (e.g.,
⇑ Address: Weatherhead School of Management, Case Western Reserve Univer- Warfield et al., 1995). Papers that attempt to disentangle whether efficiency or
sity, 10900 Euclid Avenue, 362 PBL, Cleveland, OH 44106, United States. Tel.: +1 216 managerial opportunism drives accounting discretion include Christie and Zimmer-
368 3688; fax: +1 216 368 6249. man (1994) and Bowen et al. (2008). See also Dechow and Skinner (2000) for the
E-mail address: christa.bouwman@case.edu practitioner and academic viewpoints on why firms smooth earnings.
0378-4266/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jbankfin.2013.12.019
284 C.H.S. Bouwman / Journal of Banking & Finance 41 (2014) 283–303
characteristics (e.g., age and education) to earnings smoothing. fixed effects are included to address potential concerns that opti-
Managerial beliefs could matter as well. For example, recent evi- mists may self-select to work in certain industries.
dence indicates that managerial optimism, where optimism is de- The optimism proxies used were developed by Malmendier and
fined as an upward bias in beliefs about future outcomes, does Tate (2005, 2008). Their proxies are based on the assumption that
affect a wide range of corporate and individual decisions.2 Notable an (over)optimistic manager systematically overestimates the out-
accounting applications include Hribar and Yang (2012), who find comes of her firm’s projects, thereby delaying the exercise of her
that overconfident/optimistic managers are more likely to issue options in the hope that the stock price will rise further. While they
overly optimistic earnings forecasts and are more likely to have refer to their measures as overconfidence measures, they also
earnings that miss such forecasts; and Schrand and Zechman acknowledge that the literature typically associates overconfi-
(2012), who find that CEOs of firms which misstated earnings tend dence with overestimation of a signal’s variance (e.g., Goel and Tha-
to be overly optimistic about their firms’ performance and may fol- kor, 2008), while overestimation of the mean of a signal (as their
low initially unintentional misstatements with intentional misstate- measures do) is referred to as (over)optimism (e.g., Manove and
ments if optimistic expectations are not realized. Padilla, 1999; Van den Steen, 2004; Coval and Thakor, 2005; Puri
Continuing in the tradition of this literature, in this paper, I and Robinson, 2007).5 Following Baker et al. (2007), Jin and Kothari
empirically address the question: how does managerial optimism (2008), Page (2008) and Hackbarth (2009), I therefore refer to their
affect earnings smoothing? It is natural to ask this question since proxies as measures of managerial optimism.6 The initial sample
smoothing is affected significantly by expectations about future used for my tests is the same sample of 477 large U.S. corporations
earnings, and optimism affects such expectations. Stated differ- used by Malmendier and Tate (2005, 2008).
ently, smoothing involves making tradeoffs between how much I have two main findings. First, firms with optimistic CEOs
to report in the current period and how much to report in the fu- smooth earnings more than firms with rational managers. Second,
ture, and such tradeoffs depend on managerial beliefs about future optimistic managers are just as likely as rational managers to show
events. Nonetheless, while it is intuitively straightforward that positive or negative earnings surprises, but the surprises of the
optimism would affect earnings smoothing, it is less apparent optimists are smaller in absolute value. Smaller earnings surprises
whether they would smooth more or less than rational managers. seem consistent with greater smoothing, but may seem at odds
The reason is as follows. An optimist attaches a higher probability with the finding by Hribar and Yang (2012) who find that optimis-
than a rational manager to achieving high earnings in the future tic managers are more likely to miss forecasts. Recall, however,
and hence is more willing to ‘‘borrow’’ from those earnings to in- that they measure earnings surprises relative to management’s
flate current earnings, which means that an optimist may be in- own forecasts, whereas I measure them relative to analyst
clined to report higher current earnings than a rational manager forecasts.
would. However, the ‘‘intertemporal adding-up’’ constraint on I perform several robustness tests and additional analyses to
earnings means that this will make it more likely that the optimist more deeply understand these results. First, the sample contains
will have less ‘‘elbow room’’ for reporting high earnings in the fu- firms that appeared at least four times in the Forbes 500 between
ture. In other words, given the same true earnings distribution, an 1984 and 1994, instead of the entire universe of Compustat firms.
optimistic manager cannot continually report higher earnings than One therefore has to be careful to check whether the results are
a rational manager, which makes the smoothing implication of spuriously driven by nonrandom selection. To address this poten-
optimism not quite as obvious as the observation that an optimist tial sample selection bias, I use Heckman’s two-step correction
has a greater propensity to inflate earnings relative to a rational method. I obtain results similar to my main results, providing reas-
manager, everything else equal. I therefore empirically address surance that nonrandom selection is not a problem.
the smoothing question but will also offer a potential theoretical Second, several studies find that managers use derivatives to
explanation for my findings. smooth earnings (Barton, 2001) and that analyst forecasts are more
The empirical tests use measures of earnings smoothing but accurate and have lower dispersion for derivative users (DaDalt
also of earnings surprises measured relative to analyst forecasts. et al., 2002). Given this evidence and the fact that I use earnings
The reason is that smoother earnings are more easily predictable variability and analyst forecasts as my main variables, it is impor-
for analysts (e.g., Skinner and Sloan, 2002) and may therefore be tant to control for firms’ use of derivatives in my analyses. When I
associated with fewer or smaller earnings surprises.3 Specifically, do this, I obtain results that are very similar to the main results.
the empirical tests regress measures of earnings smoothing and Third, one potential concern is that CEOs who are classified as
earnings surprises on managerial optimism proxies and a set of con- optimists – on the basis of delayed executive stock option exercise
trol variables which includes firm size, market-to-book, and lever- – are not truly optimists. They could simply be rational CEOs who
age, as well as variables that represent controls for operational have favorable inside information about future firm performance
differences, agency problems, asymmetric information, corporate that makes them delay their option exercises until this information
governance, CEO stock and option ownership, systematic risk, diver- becomes public. Additional tests rule this out.
sification, and year and industry fixed effects.4 CEO stock and option Fourth, another concern is that CEOs who exercise their options
ownership are controlled for because smoothing may be higher at late (and are therefore classified as optimists) just happen to work
firms with CEOs whose compensation is more sensitive to their at firms with lower volatility in stock returns. Given lack of perfect
firms’ stock prices (see Bergstresser and Philippon, 2006). Industry diversification of her own firm’s risk, higher risk can induce earlier
option exercise by the CEO, and lower risk may induce later
2
In a non-accounting context, these decisions include credit policies (Manove and
5
Padilla, 1999); financial intermediary existence (Coval and Thakor, 2005), investment Following the literature on self-serving attribution, Malmendier and Tate (2005,
choices and acquisition decisions (Malmendier and Tate, 2005, 2008), dividend policy 2008) use the term ‘‘overconfidence’’ to refer to an upward bias in the manager’s
and capital structure (Ben-David et al., 2007), portfolio holdings and the choice to assessment of future outcomes that are firm-specific and potentially attributable to
remarry after divorce (Puri and Robinson, 2007), and CEO succession within firms the manager’s own skill. They view this as being different from optimism related to a
(Goel and Thakor, 2008). general overestimation of all outcomes, including those outside the CEO’s control,
3
Note that the definition of earnings surprises used here is different from Hribar such as the level of the stock market.
6
and Yang’s (2012): they focus on surprises relative to earnings guidance by In contrast, Hribar and Yang (2012) use the term overconfident, indicating that:
management. They do not examine how such guidance affects expectations of ‘‘For consistency with prior research, we use the term ‘overconfident’ to describe our
analysts or earnings surprises relative to those expectations. construct of interest, despite the fact that the empirical measure we use is actually a
4
Robustness checks in Section 5.2 also control for a firm’s hedging activities. relative measure of confidence.’’
C.H.S. Bouwman / Journal of Banking & Finance 41 (2014) 283–303 285
exercise. However, additional analyses that consider idiosyncratic across firms. This section discusses these three strands of this liter-
stock return volatility and total stock return volatility, are able to ature and describes the marginal contribution of this paper relative
rule this out. to these strands.
Another issue to consider is that an optimistic CEO, knowing One strand of the literature has provided various reasons for
that she will delay exercise of her options relative to a rational earnings smoothing. Smoothing may reduce a firm’s perceived
CEO, may adopt policies that make the stock price more volatile, earnings volatility and risk, hence lowering the required rate of re-
since this will increase the value of her options. Since earnings turn (Trueman and Titman, 1988). Smoothing may lead to higher
smoothing is typically thought of as dampening stock price volatil- stock prices (Thomas and Zhang, 2002; Francis et al., 2004),
ity, such an incentive would work against the argument that opti- possibly because of the accompanying drop in the potential losses
mists smooth more. However, the reasoning that optimists might uninformed stockholders suffer when they trade for liquidity
wish to knowingly increase stock price volatility (relative to a ra- reasons (Goel and Thakor, 2003). Firms may engage in downward
tional CEO) is valid only if the optimist knows she is an optimist. earnings management prior to a stock repurchase to achieve high-
But we know that this is not possible because a behavioral bias er-than-expected post-repurchase earnings growth (Gong et al.,
(like optimism) cannot survive self-awareness, i.e., if the person 2008).8
with the bias recognizes her affliction, there can no longer be any A paper in this first strand that is closely related to my work is
bias because behavioral biases lead to choices that reduce the per- Fudenberg and Tirole (1995). That paper provides a theoretical
son’s expected utility relative to rational choices, so nobody would explanation for smoothing based on the idea that reporting bad
remain biased if they recognized they were biased (for a discus- earnings can be personally very costly for managers, and this
sion, see Goel and Thakor, 2008). Theoretically, therefore, there is causes them to report earnings in a way that effectively ‘‘transfers’’
no reason for an optimist to attempt to increase price volatility earnings from good to bad states through time. The authors as-
any more or less than a rational manager. sume that every manager is rational (i.e., managers in their paper
Fifth, to gain a deeper understanding of the earnings surprise do not have biased beliefs) but infinitely risk-averse. They also as-
results, I investigate whether they are driven by large negative, sume that every manager enjoys private benefits from keeping her
small negative, large positive and/or small positive surprises. job, and can be fired for poor performance (bad reported earnings).
Having established that optimistic managers smooth more and In this setup, the authors show that the manager will smooth earn-
are associated with smaller earnings surprises than rational man- ings: she boosts reported earnings in bad times to avoid dismissal
agers, I offer a possible theoretical explanation for these findings, and under-reports in good times in order to ‘‘save’’ and add to re-
based on a combination of three ingredients: the torpedo effect, ported earnings in future bad times. Evidence in support of the the-
the behavior of optimists relative to rational managers, and the risk ory is provided by DeFond and Park (1997).
of litigation/prosecution for earnings misreporting. Here, I briefly My marginal contribution relative to these papers is as follows.
summarize the argument. The torpedo effect induces all managers First, there are key differences between my theoretical argument
to smooth more – they over-report earnings in bad states (i.e., and that in Fudenberg and Tirole (1995). The manager in Fuden-
states in which true realized earnings are low) and under-report berg and Tirole (1995) is not behaviorally biased, whereas she is
in good states (i.e., states in which true realized earnings are high). in my paper. In Fudenberg and Tirole (1995), the manager may
Optimistic managers over-report more than rational managers in have high or low expectations about future earnings, depending
bad states, effectively ‘‘borrowing’’ more earnings from the future, on her information set, but an expectation of high earnings is ra-
because they are more bullish about having sufficiently high future tional, not inflated by optimism, and an expectation of low earn-
earnings to ‘‘pay’’ for this borrowing. This reduces the earnings the ings too is rational, not deflated by pessimism.9 Second, as
optimist can report in the good state in the future. Moreover, even discussed above, optimists do not realize that they are optimists,
ignoring this adding-up-constraint effect, the ability of optimists to or they would rationally undo the effects of this bias. So optimists
deliver positive earnings surprises in good states is lower than that do not have a stronger incentive to smooth than rational managers.10
of rational managers because the higher level of (expected) base- Rather, their behavioral bias causes them to smooth more. Third, gi-
line earnings leaves them with less room to report higher-than-ex- ven the strong support provided by DeFond and Park (1997) for the
pected earnings. This intuition delivers both greater smoothing of Fudenberg and Tirole (1995) theory, I verify that the effect of
reported earnings relative to true cash flows as well as smaller
earnings surprises relative to analyst expectations for optimists
than for rational managers.7
8
The remainder of the paper is organized as follows. Section 2 Fonseca and Gonzalez (2008) examine the determinants of income smoothing by
describes the related literature. Section 3 explains the empirical banks around the world. Asdrubali and Kim (2008) also examine international
smoothing.
approach, discusses the variables, describes the data and provides 9
To see this, suppose there are two firms in the economy, one led by an optimist
descriptive statistics. Empirical results are presented in Section 4. and one led by a rational CEO. Suppose that at date 0, the true probability of date 1
Section 5 addresses robustness issues and performs additional earnings being Good (Bad) is 0.7 (0.3) for both firms. The rational CEO correctly
tests. Section 6 provides a potential theoretical explanation for believes that the probabilities of these two states are 0.7 and 0.3, while the optimist
will believe that the probability of the Good state is higher and that of the Bad state is
the findings. Section 7 summarizes and concludes.
lower (for example 0.8 and 0.2). Similarly, at date 1, the true probability of date 2
earnings being Good (Bad) may be 0.4 (0.6) for every firm. The rational CEO again
correctly believes that the probabilities of these two states are 0.4 and 0.6, while the
2. Related literature
optimist will again believe that the probability of the Good state is higher and that of
the Bad state is lower (for example 0.5 and 0.5). Thus, regardless of whether the true
Most closely related to this paper is the literature on why man- probability of Good earnings next period is high (0.7) or low (0.4), the rational CEO
agers prefer to report smooth earnings, the empirical detection of always correctly assesses each probability while the optimist always overestimates it.
10
earnings smoothing, and factors that lead to smoothing differences All of my analyses focus on the effect of CEO optimism on smoothing and not on
the effect of CEO pessimism. This is consistent with the theories, which show that
optimists tend to beat out rational agents and pessimists to become CEOs (e.g., Coval
7
Admittedly, other explanations for why optimists smooth more and show smaller and Thakor, 2005; Goel and Thakor, 2008; Gervais et al., 2011). The Malmendier and
earnings surprises may exist. I leave that as an interesting topic for future research. Tate measures used in my paper also capture CEO optimism, not pessimism. In
Also, all my robustness checks notwithstanding, it should be recognized that any Section 5.5, however, I will present some evidence on managers with high versus low
empirical measure of earnings smoothing is imperfect, so the results should be optimism based on when CEOs exercise their options. This evidence reveals that very
viewed in light of that caveat. few CEOs in the sample have low optimism.
optimism on smoothing survives the effect DeFond and Park (1997) vector of twelve control variables (described in Section 3.5) plus
document. year and industry fixed effects. All regressions are estimated with
A second strand of the literature attempts to empirically de- robust standard errors, clustered by firm to control for heteroske-
tect earnings smoothing. Many studies focus on managers’ use dasticity as well as possible correlation between observations of
of discretionary accruals (e.g., Dechow et al., 1995, 2012). Recent the same firm in different years. Multicollinearity is tested for by
studies have examined deferred taxes and/or the provision for computing variance inflation factors for all regressions. Multicollin-
taxes (Phillips et al., 2003; Dhaliwal et al., 2004), or assessed earity does not seem to be a problem in the data since all variance
the variability of reported earnings relative to the variability of inflation factors are close to one.12
cash flows (Land and Lang, 2002; Leuz et al., 2003; Lang et al.,
2006; Myers et al., 2007). In this paper, I use the latter approach 3.2. Optimism measures
of computing relative earnings volatility to measure earnings
smoothing.11 Following Malmendier and Tate (2005, 2008), three optimism
A third strand of the literature explores the various factors that measures are constructed based on the timing of executive option
lead to smoothing differences across firms. They include: CEO bo- exercise. According to option pricing theory, investors should opti-
nuses (Healy, 1985); CEO option and stock ownership (Warfield mally hold their options until expiration (Black and Scholes, 1973;
et al., 1995; Bergstresser and Philippon, 2006); the percentage of Merton, 1973). This result is based on the premise that investors
outside directors and the number of independent auditors on the can engage continuously in dynamic trading to fully hedge their
audit committee (Klein, 2002); the proportion of top executives option positions at every point in time, and options can be priced
on the board (Bowen et al., 2008); and CEO personal characteristics (using the equivalent Martingale measure) as if investors were risk
such as age and education (Ge et al., 2011). Hribar and Yang (2012) neutral. While this is a reasonable assumption for the typical
focus on the effect of overconfidence/optimism on earnings fore- investor, it does not adequately describe the situation of many
casting and earnings management. top executives (Hall and Murphy, 2002). CEOs at large U.S. corpo-
The main intended contribution of this paper relative to the rations are typically quite underdiversified for at least a couple of
existing literature is to provide evidence that earnings smoothing reasons. First, their human capital is disproportionately invested
and earnings surprises are affected by managerial optimism. Thus, in their own firms. And second, they generally receive sizeable
it adds to the growing literature on the importance of managerial option grants that are non-tradable, can be exercised only after a
attributes in affecting corporate decisions, pioneered by Bertrand vesting period has elapsed, and come with short-selling
and Schoar (2003). Perhaps more importantly, consistent with restrictions.
the theme in the third strand of the smoothing literature dis- Hall and Murphy (2002) show that since CEOs cannot fully
cussed above, it seeks to add to our understanding of smoothing hedge their positions in dynamically complete markets, they
differences across firms and the factors that determine these should rationally exercise their options early.13 Unlike rational
differences. CEOs, optimistic CEOs overestimate their firm’s future earnings.
Optimistic CEOs therefore believe that their firm’s stock price will in-
3. Empirical approach, variables, and sample crease beyond what should be rationally anticipated, and conse-
quently exercise their options later than rational CEOs would. As
This section describes the tests used to examine whether opti- indicated in the introduction, the optimistic CEOs do not realize they
mistic managers exhibit different smoothing behavior than ra- are (irrationally) optimistic, and hence do not believe that they are
tional managers and whether earnings surprises associated with irrationally delaying option exercise. This means they will not know-
optimists are different from those associated with their rational ingly undertake actions that may increase price volatility relative to
counterparts. Also discussed are the variables and the sample. what rational CEOs do, just to reflect the higher relative option
values they would get from doing so. The three optimism measures
3.1. Empirical approach exploit the expected difference in the timing of option exercise be-
tween rational and optimistic CEOs (see Malmendier and Tate,
Univariate test statistics are calculated to obtain preliminary 2005, 2008).
evidence on whether optimistic and rational managers are associ-
ated with different smoothing behavior and different earnings sur- 3.2.1. Longholder
prises than rational managers. In this examination, t-tests are used The first measure focuses on the year before the options expire,
to establish whether the differences in means are statistically typically year ten in my sample. It classifies a CEO as optimistic
significant. (‘‘Longholder’’) for all of her years in the sample if she ever held
Multivariate regressions are used to control for other factors an option until the year of expiration, although the option is at
that may affect smoothing and earnings surprises. The following least 40% in the money at the beginning of that year. Given a typ-
models are estimated: ical four-year vesting period and a ten-year duration, a CEO who
holds options until the final year of its duration has postponed
SMOOTHi;t ¼ b0 þ b1 OPTIMIST i;t þ BX i;t ð1Þ
exercise by at least five years. Note that this measure treats
optimism as a managerial fixed effect since a CEO is classified as
SURPRISEi;t ¼ c0 þ c1 OPTIMIST i;t þ GX i;t ð2Þ
optimistic for all of her years in the sample.
SMOOTHi,t measures earnings smoothing at firm i in year t using one
of the smoothing variables described in Section 3.3. SURPRISEi,t is 3.2.2. Pre-/Post-Longholder
the earnings surprise at firm i in year t as defined in Section 3.4. The second measure splits the Longholder optimism measure
OPTIMISTi,t is a dummy variable that equals 1 if an optimistic CEO into two parts. It classifies a CEO as a ‘‘Post-Longholder’’ from the
(as defined in Section 3.2) heads the firm and 0 otherwise. Xi,t is a
12
Variance inflation factors, the diagonal elements of the inverse of the correlation
11
Some papers in this strand argue that a greater amount of earnings smoothing matrix, range from 1 to infinity. Multicollinearity is not considered a problem if the
should lead to a larger informational asymmetry between the firm and investors variance inflation factors are close to 1 (see, e.g., Chatterjee et al., 2000).
13
because smoothing is merely an obfuscation device that keeps the manager’s private The optimal timing depends on their wealth, degree of risk-aversion, and level of
information from reaching investors (Bhattacharya et al., 2003). diversification.
year after she holds an option until expiration for the first time, The second smoothing variable recognizes that net income is
even though the option is at least 40% in the money at the likely to be more volatile at firms with more volatile cash flows.
beginning of that year. A CEO is classified as a ‘‘Pre-Longholder’’ It therefore takes the first smoothing variable, calculated as
for all other years during which she was classified as optimistic explained above, and adjusts it for the variability of the firm’s cash
using the Longholder measure. If optimism is a managerial fixed flows divided by total assets, (DCF/TA), where the latter is
effect, the coefficients on both Pre- and Post-Longholder will be calculated in a similar fashion as (DNI/TA). That is, the second
significant, indicating that both Pre- and Post-Longholders smooth smoothing variable is defined as VARIABILITY (DNI/TA) over (DCF/
earnings more than rational CEOs. However, if a CEO is only opti- TA), and is calculated by dividing the variance of residuals from
mistic after she has first displayed signs of optimism by exercising regressions of (DNI/TA) on the six control variables by the variance
options late, only the coefficient on Post-Longholder will be of residuals from regressions of (DCF/TA) on those same control
significant. variables.
The third smoothing variable focuses directly on the smoothing
3.2.3. Holder 67 effect of accruals. It examines the correlation of accruals and cash
The third measure focuses on options that have recently be- flows, both normalized by total assets, i.e., CORR ((ACC/TA), (CF/
come fully vested. CEOs are rationally expected to exercise options TA)). This correlation should be more negative at firms that smooth
soon after the vesting period is over, provided the options are suf- earnings, because their managers respond to poor cash flows by
ficiently in the money. If there are two occurrences of a CEO failing increasing accruals (see, e.g., Land and Lang, 2002; Lang et al.,
to exercise an option with five years remaining that is at least 67% 2006; Myers et al., 2007). Thus, if optimistic managers smooth
in the money, then that CEO is classified as optimistic (‘‘Holder earnings more than rational managers, this correlation should be
67’’), starting the year after she fails to exercise this option for more negative at firms headed by optimistic CEOs since they use
the first time.14,15 accruals to smooth earnings. Accruals are calculated as in Dechow
et al. (1995) using quarterly data as:
3.3. Smoothing variables ACC it ¼ DCAit DCASHit DCLit þ DSTDit DEPit ; ð3Þ
Three smoothing variables are calculated in the spirit of Leuz where DCAit is the change in total current assets (COMP #40),
et al. (2003), Lang et al. (2006) and Myers et al. (2007). Each vari- DCASHit is the change in cash and cash equivalents (COMP
able is constructed using regression analysis in a way explained be- #36), DCLit is the change in total current liabilities (COMP #49),
low. Following Myers et al. (2007), quarterly data are used and DSTDit is the change in short-term debt included in current liabil-
seasonal effects are removed by taking fourth differences. In each ities (COMP #45), and DEPit is depreciation and amortization
regression, five years of quarterly data are used, if available, and (COMP #5) at firm i in year t. The CORR ((ACC/TA), (CF/TA)) is then
observations for which data are not available for at least twelve defined as the correlation between the regression residuals of
quarters are dropped. After having constructed the smoothing vari- (ACC/TA) and the regression residuals of (CF/TA), where the resid-
ables, only the fourth-quarter variables are kept to ensure that for uals have been calculated using the regression approach de-
each sample firm there is only one observation per year. scribed above.
The first smoothing variable is the variability of earnings, or
more precisely, the variability of the change in net income divided 3.4. Earnings surprise variables
by total assets, VARIABILITY (DNI/TA). If optimistic managers
smooth earnings more than rational managers, the variability of To calculate earnings surprises, actual reported earnings per
earnings should be lower, ceteris paribus, at firms led by optimistic share data and one-year-ahead earnings per share forecasts are
managers. The VARIABILITY (DNI/TA) is calculated using the two- collected from Thomson Financial’s Institutional Brokers Estimate
step procedure described in Lang et al. (2006) and Myers et al. System (IBES) database. Earnings surprises are then calculated as
(2007). First, the change in net income divided by total assets, the actual reported earnings per share minus the median analyst
DNI/TA, is regressed on a set of six control variables that may affect forecast based on the last one-year ahead earnings per share fore-
changes in earnings, including: leverage (total liabilities divided by cast available in IBES before earnings are reported.
total assets, COMP #44 minus COMP #60 divided by COMP #44);
sales growth (percentage annual growth in COMP #2); debt issu- 3.5. Control variables
ance (percentage change in total liabilities, COMP #44 minus
COMP #60); equity issuance (percentage change in shares out- The vector of control variables, X, includes twelve variables that
standing adjusted for splits, COMP #61 times COMP #17); annual may affect earnings smoothing and earnings surprises (firm size,
asset turnover (sales divided by total assets, COMP #2 divided by market-to-book ratio, book leverage, profitability, two controls
COMP #44); and size (logarithm of the market value of equity, for agency and asymmetric information, two governance variables,
COMP #61 times COMP #14). Second, VARIABILITY (DNI/TA) is then CEO stock and option ownership, systematic risk, and diversifica-
calculated as the variance of the residuals of these regressions. tion), and year and industry fixed effects.
Firm size, LNASSETS, is measured as the log of total assets (COMP
14
This is Malmendier and Tate’s (2005) definition. Their (2008) definition classifies
#6). The firm’s market-to-book ratio, M/B RATIO, is defined as the
a CEO as optimistic if she exercises options late at least once (rather than twice), and market value of assets divided by the book value of assets. The mar-
hence yields more optimistic CEOs. ket value of assets is the fiscal year-end stock price (COMP #199)
15
When Malmendier and Tate (2005, 2008) use this measure, they only include in times the number of shares outstanding (COMP #54), plus the cur-
the sample CEOs who have options that are at least 67% in the money during year five
rent portion of long-term debt (COMP #34), long-term debt (COMP
(and depending on whether the CEOs exercise these options on time, they are
classified as rational or optimistic). This sample restriction is designed to avoid #9), and preferred stock (COMP #10), minus deferred taxes (COMP
classifying a CEO as rational when she truly is optimistic but never had the #35). Book leverage, BOOKLEV, is defined as interest-bearing debt
opportunity to display such optimism. However, when this restriction is applied, so (COMP #9 plus COMP #34) divided by total assets (COMP #6).
many firms are lost that there is little cross-sectional variation left and virtually all Malmendier and Tate (2005, 2008) show that optimistic manag-
CEOs are classified as optimistic. I therefore do not impose this restriction. As a
consequence, this approach likely classifies some truly optimistic CEOs as rational,
ers overinvest and are more likely to engage in value-destroying
which biases the tests against finding the hypothesized results using the Holder 67 mergers. This suggests operational differences between firms led
optimism measure. by optimists and rational managers that may then lead to earnings
Table 1
Summary statistics.
Panel A: Number of firms and number of CEOs

Number of firms 210
Number of CEOs 374
Obs Mean Median Minimum Maximum

Panel B: Summary statistics of optimism measures and smoothing variables
Optimism measures
LONGHOLDER 1705 0.194 0.000 0.000 1.000
PRE-LONGHOLDER 1705 0.080 0.000 0.000 1.000
POST-LONGHOLDER 1705 0.114 0.000 0.000 1.000
HOLDER 67 1705 0.401 0.000 0.000 1.000
Smoothing variables
VARIABILITY (DNI/TA) 1705 0.000 0.000 0.000 0.058
VARIABILITY (DNI/TA) over (DCF/TA) 1705 0.157 0.042 0.000 3.068
CORR ((ACC/TA), (CF/TA)) 1705 0.887 0.950 1.000 0.359
Panel C: Summary statistics of control variables
Control variables
LNASSETS 1705 7.757 7.738 4.760 11.517
M/B RATIO 1705 1.069 0.797 0.138 4.271
BOOKLEV 1705 0.268 0.276 0.000 0.742
PROFITABILITY 1705 0.169 0.156 0.002 0.463
RETEARN_CS 1705 0.633 0.651 0.208 1.544
COLLATERAL 1705 0.652 0.678 0.004 0.939
BOARD 1705 12.141 12.000 5.000 24.000
CHAIRMAN 1705 0.179 0.000 0.000 1.000
PCTOWN 1705 0.016 0.001 0.000 0.518
PCTVESTOPT 1705 0.019 0.004 0.000 1.056
SYSTRISK 1705 0.002 0.002 0.000 0.012
HHI_ASSETS 1705 0.309 0.292 0.000 0.839
This table contains summary statistics. Panel A shows the number of firms and CEOs in the sample. Panel B reports summary statistics of
the optimism measures and smoothing variables. Panel C contains summary statistics of the control variables used to test the
hypothesis that optimistic CEOs smooth earnings more than their rational counterparts.
Optimism measures: A CEO is classified as a Longholder (for all of her years in the sample) if she ever held an option until the year of
expiration, although the option is at least 40% in the money at the beginning of that year. A CEO is classified as a Post-Longholder (Pre-
Longholder) for the years after (up until) she has held options that are at least 40% in the money until the year of expiration for the first
time. If a CEO twice fails to exercise an option with five years remaining duration that is at least 67% in the money, she is classified as a
Holder 67 from the year after she exercises such options late for the first time.
Smoothing variables: VARIABILITY (DNI/TA) is the variability of earnings, the change in net income divided by total assets. It is calculated
as the variance of residuals from regressions of (DNI/TA) on six control variables. The six control variables include: leverage (total
liabilities divided by total assets); sales growth (percentage annual growth); debt issuance (percentage change in total liabilities);
equity issuance (percentage change in shares outstanding adjusted for splits); annual asset turnover (sales divided by total assets); and
size (logarithm of the market value of equity). VARIABILITY (DNI/TA) over (DCF/TA) is the variability of earnings (as defined above)
divided by the variability of cash flows, the change in cash flows divided by total assets (calculated using the same approach as the
variability of earnings). CORR ((ACC/TA), (CF/TA)) is the correlation between the regression residuals of (ACC/TA) and the regression
residuals of (CF/TA), where the residuals have been calculated using the regression approach discussed above. Accruals (ACC) equal the
change in current assets minus the change in cash and cash equivalents minus the change in current liabilities plus the change in short-
term debt included in current liabilities minus depreciation and amortization.
Control variables: LNASSETS is the log of total assets. M/B RATIO is the firm’s market-to-book ratio, defined as the market value of assets
divided by the book value of assets. BOOKLEV is interest-bearing debt divided by total assets. PROFITABILITY is EBITDA divided by total
assets. RETEARN_CS is retained earnings as a fraction of common stock. COLLATERAL is tangible assets divided by total assets. BOARD is
the number of board members. CHAIRMAN is a dummy that equals 1 if the CEO serves as the chairman of the board, and 0 otherwise.
PCTOWN is the fraction of company stock owned by the CEO. PCTVESTOPT is the number of options exercisable within 60 days from the
start of the year (multiplied by 10) divided by the number of shares outstanding. SYSTRISK is systematic risk, measured as b2 times the
variance of the value-weighted market index, where b is estimated with monthly return data using a one-factor market model.
HHI_ASSETS is an asset-based Herfindahl index, calculated at the firm level as one minus the sum of squares of each segment’s assets
divided by total assets.
management differences. To capture the incremental effect of opti- and provide evidence that firms in which retained earnings are
mism on smoothing above and beyond these operational differ- high relative to common stock face potentially greater agency
ences, I add a measure of the firm’s operating profitability as a problems. To control for differences in asymmetric information,
control variable. PROFITABILITY is calculated as operating profit I add COLLATERAL, measured as tangible assets (net plant, prop-
(EBITDA) divided by total assets. The idea is that if optimistic man- erty, and equipment (COMP #8) plus inventory (COMP #3)) di-
agers have smoother earnings only because they are making profit- vided by total assets (COMP #6) (see, for example, Almeida and
depleting investment decisions, then controlling for profitability Campello, 2007). Asymmetric information problems are pre-
should make the smoothing difference between optimistic and ra- sumed to be larger at firms with low tangible assets (see, e.g.,
tional managers go away. Myers and Majluf, 1984; Kohers and Ang, 2001). To control
To control for agency problems, RETEARN_CS, retained for differences in corporate governance, I add BOARD, the num-
earnings as a fraction of common stock, is added. DeAngelo ber of board members, and CHAIRMAN, a dummy that equals
et al. (2006) observe that unlike contributed equity, retained one if the CEO is also the chairman of the board (see, e.g.,
earnings do not come with the benefit of additional monitoring, Malmendier and Tate, 2008).
Table 2
Main Result #1: optimistic CEOs smooth earnings more than rational CEOs.
Panel A: Univariate results

(i) VARIABILITY (DNI/TA) (ii) VARIABILITY (DNI/TA) over (iii) CORR ((ACC/TA), (CF/TA))
(DCF/TA)
N Mean N Mean N Mean
Panel I
LONGHOLDER 331 0.0002 331 0.1027 331 0.9154
RATIONAL 1374 0.0005 1374 0.1699 1374 0.8806
Difference 0.0003 0.0672 0.0348
p-Value (0.051)** (0.000)*** (0.000)***
Panel II
PRE-LONGHOLDER 137 0.0001 137 0.0920 137 0.9157
RATIONAL 1568 0.0005 1568 0.1625 1568 0.8848
Difference 0.0004 0.0705 0.0308
p-Value (0.099)* (0.005)*** (0.029)**
POST-LONGHOLDER 194 0.0003 194 0.1102 194 0.9152
RATIONAL 1511 0.0005 1511 0.1628 1511 0.8838
Difference 0.0002 0.0526 0.0314
p-Value (0.307) (0.015)** (0.009)***
Panel III
HOLDER 67 684 0.0003 684 0.1289 684 0.9137
RATIONAL 1021 0.0005 1021 0.1756 1021 0.8697
Difference 0.0002 0.0467 0.044
p-Value (0.070)* (0.001)*** (0.000)***
Panel B: Regression results

Panel I: Longholders Panel II: Pre-/Post-Longholders Panel III: Holders 67
(i) (ii) (i) (ii) (i) (ii)
LONGHOLDER 0.001* 0.046*
(1.89) (1.93)
PRE-LONGHOLDER 0.000* 0.031
(1.81) (1.15)
POST-LONGHOLDER 0.001* 0.058**
(1.78) (2.11)
HOLDER 67 0.000 0.006
(1.38) (0.26)
LNASSETS 0.000** 0.007 0.000** 0.007 0.000** 0.007
(2.25) (0.56) (2.26) (0.54) (2.21) (0.59)
M/B RATIO 0.001 0.017 0.001 0.017 0.001 0.020
(1.33) (0.83) (1.33) (0.83) (1.37) (1.02)
BOOKLEV 0.002* 0.096 0.002* 0.094 0.002* 0.093
(1.70) (0.85) (1.69) (0.83) (1.67) (0.81)
PROFITABILITY 0.006* 0.402** 0.006* 0.406** 0.007* 0.430**
(1.75) (2.34) (1.75) (2.37) (1.78) (2.50)
RETEARN_CS 0.001 0.141*** 0.001 0.142*** 0.001 0.141***
(1.23) (2.74) (1.24) (2.75) (1.24) (2.73)
COLLATERAL 0.000 0.014 0.001 0.013 0.001 0.003
(1.08) (0.20) (1.11) (0.18) (1.52) (0.04)
BOARD 0.000 0.001 0.000 0.001 0.000 0.001
(0.52) (0.31) (0.52) (0.31) (0.73) (0.17)
CHAIRMAN 0.000 0.036 0.000 0.034 0.000 0.039
(0.95) (0.93) (0.92) (0.87) (0.98) (1.01)
PCTOWN 0.000 0.260* 0.000 0.267* 0.000 0.247*
(0.26) (1.85) (0.20) (1.89) (0.26) (1.70)
PCTVESTOPT 0.003 0.214** 0.003 0.206** 0.003 0.269***
(1.49) (2.21) (1.49) (2.15) (1.65) (2.87)
SYSTRISK 0.093 1.129 0.094 1.102 0.094 1.613
(1.55) (0.19) (1.56) (0.19) (1.51) (0.28)
HHI_ASSETS 0.001* 0.072 0.001* 0.071 0.001* 0.072
(1.75) (1.56) (1.75) (1.53) (1.76) (1.56)
Year & Industry Dummies Yes Yes Yes Yes Yes Yes
Observations 1705 1705 1705 1705 1705 1705
Adjusted R2 0.11 0.16 0.11 0.16 0.11 0.15
This table contains results of univariate tests (Panel A) and OLS regressions (Panel B) that compare earnings smoothing by optimistic and rational managers. The evidence
suggests that optimistic CEOs smooth earnings more than rational CEOs.
Smoothing variables: Column (i) contains results using VARIABILITY (DNI/TA) which is the variability of earnings, the change in net income divided by total assets. It is
calculated as the variance of residuals from regressions of (DNI/TA) on six control variables. The six control variables include: leverage (total liabilities divided by total assets);
sales growth (percentage annual growth); debt issuance (percentage change in total liabilities); equity issuance (percentage change in shares outstanding adjusted for splits);
annual asset turnover (sales divided by total assets); and size (logarithm of the market value of equity). Column (ii) contains results using VARIABILITY (DNI/TA) over (DCF/TA),
which is the variability of earnings (as defined above) divided by the variability of cash flows, the change in cash flows divided by total assets (calculated using the same
approach as the variability of earnings). CORR ((ACC/TA), (CF/TA)) is the correlation between the regression residuals of (ACC/TA) and the regression residuals of (CF/TA), where
the residuals have been calculated using the regression approach discussed above. Accruals (ACC) equal the change in current assets minus the change in cash and cash
equivalents minus the change in current liabilities plus the change in short-term debt included in current liabilities minus depreciation and amortization.
Optimism measures: Panel I shows results for rational CEOs and Longholders. A CEO is classified as a Longholder (for all of her years in the sample) if she ever held an option
until the year of expiration, although the option is at least 40% in the money at the beginning of that year. CEOs not classified as Longholders are classified as rational. Panel II
shows results for rational CEOs and Pre-/Post-Longholders. A CEO is classified as a Post-Longholder (Pre-Longholder) for the years after (up until) she has held options that are
at least 40% in the money until the year of expiration for the first time. CEOs not classified as Pre-/Post-Longholders are classified as rational. Panel III contains results for
rational CEOs and Holders 67. If a CEO twice fails to exercise an option with five years remaining duration that is at least 67% in the money, she is classified as a Holder 67
from the year after she exercises such options late for the first time. CEOs not classified as Holders 67 are classified as rational.
Control variables used in regressions: LNASSETS is the log of total assets. M/B RATIO is the firm’s market-to-book ratio, defined as the market value of assets divided by the book
value of assets. BOOKLEV is interest-bearing debt divided by total assets. PROFITABILITY is EBITDA divided by total assets. RETEARN_CS is retained earnings as a fraction of
common stock. COLLATERAL is tangible assets divided by total assets. BOARD is the number of board members. CHAIRMAN is a dummy that equals 1 if the CEO serves as the
chairman of the board, and 0 otherwise. PCTOWN is the fraction of company stock owned by the CEO. PCTVESTOPT is the number of options exercisable within 60 days from
the start of the year (multiplied by 10) divided by the number of shares outstanding. SYSTRISK is systematic risk, measured as b2 times the variance of the value-weighted
market index, where b is estimated with monthly return data using a one-factor market model. HHI_ASSETS is an asset-based Herfindahl index, calculated at the firm level as
one minus the sum of squares of each segment’s assets divided by total assets.
All regressions include a constant, year and industry fixed effects. In Panel A, p-values are in parentheses. In Panel B, t-statistics based on robust standard errors clustered by
firm are in parentheses.
*
Significance at 10% level.
**
***
Bergstresser and Philippon (2006) show that earnings manage- more prevalent in certain industries, then self-selection may occur
ment may be higher at firms with CEOs whose compensation is via firms in those industries hiring optimistic managers and firms
more sensitive to their firms’ stock prices. This is consistent with in other industries hiring rational managers. If so, differences
smoothing theories, which argue that managers view earnings across industries rather than managerial optimism would drive
smoothing as contributing to a higher stock price on average, so the smoothing results presented in this paper. The industry fixed
managers with more stock and option holdings will tend to smooth effects in the regressions help deal with this issue.
more. This effect is opposite the direct effect of options which
encourages the manager to increase stock price volatility. That is, 3.6. Sample selection and data description
the direct effect assumes that higher stock price volatility increases
the value of the options without changing the average stock price, The analysis starts with an initial sample of 477 large, listed U.S.
whereas the smoothing effect is that earnings smoothing leads to a firms that have appeared on a Forbes 500 list at least four times be-
higher stock price on average. It is an empirical issue which of the tween 1984 and 1994. For those firms, I use the core dataset that
two effects is stronger, and I therefore control for both CEO stock has been used in Malmendier and Tate (2005, 2008) and is de-
and option ownership. Stock ownership is calculated as the frac- scribed in detail in Hall and Liebman (1998) and Yermack (1995).
tion of company stock held by the CEO and her family at the begin- This core dataset spans the years 1980–1994 and contains detailed
ning of the year (PCTOWN). CEO option ownership is the number of annual information that is needed to construct the CEO optimism
options exercisable within 60 days from the start of the year di- measures, including the number of options, the exercise price,
vided by the number of shares outstanding (PCTVESTOPT). The and the duration of options each CEO holds.17
number of options is multiplied by 10 to ensure that the mean is This dataset is complemented with Compustat, CRSP and IBES
comparable to mean stock ownership as in Malmendier and Tate data. For each firm in the sample, quarterly Compustat data are
(2005). collected to construct the earnings smoothing variables, and an-
Firms with higher systematic risk may also smooth earnings nual Compustat data to create most of the control variables.
more (see Lev and Kunitzky, 1974; Bange and De Bondt, 1998). I Monthly CRSP stock return data plus value-weighted index returns
therefore add systematic risk as a control variable. Systematic risk are obtained to construct risk measures. One-year-ahead earnings
is estimated using a one-factor market model: rit = ai + bi ⁄ rmt + eit, per share forecasts and actual earnings per share data are obtained
where rit is the return of firm i for month t and rmt is the return of from IBES to calculate earnings surprises. All variables are winsor-
the CRSP value-weighted index for month t. The model is esti- ized at the 1% and 99% level to reduce the impact of outliers.
mated using five years of monthly return data, and observations The final sample contains 210 firms and 374 CEOs (see Panel A
are dropped if fewer than 36 monthly returns are available. Sys- Table 1). Panel B reports summary statistics of the optimism mea-
tematic risk, SYSTRISK, is measured as b2 times the variance of sures and smoothing variables. Panel C contains summary statis-
the value-weighted market index (see Shin and Stulz, 2000). tics of the control variables.
Managers may use a diversification strategy as a key risk man-
agement tool (e.g., May, 1995): firms that are more diversified may
have smoother earnings. To account for this, I control for the de- 4. Main empirical results
gree of diversification of a firm’s activities. Following Lang and
Stulz (1994) and Comment and Jarrell (1995), I obtain each firm’s This section first tests whether optimistic managers show dif-
self-reported lines of business from Compustat’s Industry Segment ferent earnings smoothing behavior than rational managers. It then
files, and include HHI_ASSETS, an asset-based Herfindahl index cal- examines whether optimists are associated with different earnings
culated at the firm level for each fiscal year. The index is computed surprises than their rational counterparts.
as one minus the sum of squares of each segment’s assets over to-
tal assets, so indices closer to one indicate a higher level of 4.1. Main Result #1: Optimists smooth more
diversification.16
Industry fixed effects are included in all regressions to address Table 2 Panel A contains preliminary evidence based on univar-
potential self-selection concerns. In particular, if smoothing is iate statistics. Columns (i–iii) contain the results for the three
smoothing measures VARIABILITY (DNI/TA), VARIABILITY (DNI/TA)
over (DCF/TA), and CORR ((ACC/TA), (CF/TA)), respectively. For each
16
Similar results are obtained when a sales-based Herfindahl index or a dummy smoothing variable, the table shows the average amount of
that equals 1 if the firm reports more than one industrial segment in a fiscal year are
included instead (e.g., Berger and Ofek, 1995; Denis et al., 2002; Mansi and Reeb,
17
2002). I am grateful to Brian Hall for providing the CEO option holdings data.
Table 3
Main Result #2: optimistic CEOs are less likely to show big surprises and more likely to show small surprises.
Negative surprise Positive surprise Big surprise Small surprise

Panel A: Univariate statistics: fractions of optimistic and rational managers showing surprises of certain sizes and/or signs
Panel I
LONGHOLDER 46.30% 46.30% 40.50% 52.10%
RATIONAL 46.70% 47.90% 56.20% 38.50%
Difference 0.40% 1.60% 15.70% 13.60%
p-Value (0.905) (0.646) (0.000)*** (0.000)***
Panel II
PRE-LONGHOLDER 46.50% 44.90% 34.60% 56.70%
RATIONAL 46.70% 47.90% 54.80% 39.70%
Difference 0.20% 3.00% 20.20% 16.90%
p-Value (0.967) (0.522) (0.000)*** (0.000)***
POST-LONGHOLDER 46.20% 47.70% 46.20% 47.70%
RATIONAL 46.70% 47.50% 53.30% 40.90%
Difference 0.50% 0.20% 7.20% 6.80%
p-Value (0.908) (0.973) (0.126) (0.142)
Panel III
HOLDER 67 45.60% 47.20% 47.00% 45.80%
RATIONAL 47.50% 47.90% 57.40% 38.00%
Difference 1.90% 0.70% 10.40% 7.80%
p-Value (0.531) (0.817) (0.001)*** (0.009)***
Panel B: Regression results

Panel I
LONGHOLDER 0.065 0.049 0.331** 0.322**
(0.61) (0.44) (2.23) (2.18)
All Control Variables Yes Yes Yes Yes
Year & Industry Dummies Yes Yes Yes Yes
Panel II
PRE-LONGHOLDER 0.007 0.018 0.485*** 0.450**
(0.05) (0.11) (2.74) (2.41)
POST-LONGHOLDER 0.117 0.109 0.192 0.203
(0.81) (0.78) (1.09) (1.18)
Panel III
HOLDER 67 0.030 0.029 0.085 0.099
(0.28) (0.27) (0.65) (0.78)
This table examines whether optimistic CEOs are associated with different earnings surprises than rational CEOs. Panel A contains results of univariate tests which compare
the fractions of optimistic and rational managers that show earnings surprises of certain sizes and/or signs. Panel B shows results of regressions in which a particular type of
earnings surprise is regressed on CEO optimism plus controls, year and industry fixed effects. In both panels, percentages do not add up to 100% because zero surprises are not
reported. Only the coefficients on the optimism measures are shown for brevity. The findings suggest that optimistic CEOs and rational CEOs are equally likely to show
negative and positive surprises. However, optimistic CEOs are (significantly) less likely to show big surprises and (significantly) more likely to show small surprises.
Earnings surprise: Earnings surprises are calculated as actual earnings per share minus the median analyst forecast based on the last one-year ahead forecast before the
earnings announcement date. A surprise is negative (positive) if earnings fell short of (exceeded) the median analyst forecast. A surprise is big (small) if the absolute value of
earnings minus the median analyst forecast exceeded (was less than) 3 cents per share.
Optimism measures: Panel I shows results for Longholders: a CEO is classified as a Longholder (for all of her years in the sample) if she ever held an option until the year of
expiration, although the option is at least 40% in the money at the beginning of that year. Panel II shows results for Pre-/Post-Longholders: a CEO is classified as a Post-
Longholder (Pre-Longholder) for the years after (up until) she has held options that are at least 40% in the money until the year of expiration for the first time. Panel III
contains results for Holders 67: if a CEO twice fails to exercise an option with five years remaining duration that is at least 67% in the money, she is classified as a Holder 67
from the year after she exercises such options late for the first time.
All regressions include a constant, all the control variables (not shown for brevity), and year and industry fixed effects. In Panel A, p-values are in parentheses. In Panel B, t-
statistics based on robust standard errors clustered by firm are in parentheses.

**
***
smoothing by rational managers and optimistic managers, where The univariate results suggest that optimistic managers smooth
optimism is measured using the Longholder, Pre-/Post-Longholder, earnings more than rational managers. The variability of earnings
and Holder 67 measure in turn (see Subpanels I through III). Differ- is significantly lower for optimistic managers (see Column (i) in
ences in the average amounts of earnings smoothing across ra- Subpanels I through III), especially after controlling for differences
tional and optimistic managers are also presented. in the variability of cash flows (see Column (ii) in Subpanels I
Table 4
Main Results #1 and #2 revisited: addressing sample selection bias.
Panel A: Main regression Results #1 with Heckman correction

(i) (ii) (i) (ii) (i) (ii)
LONGHOLDER 0.001* 0.046*
(1.90) (1.94)
PRE-LONGHOLDER 0.000* 0.032
(1.81) (1.17)
POST-LONGHOLDER 0.001* 0.058**
(1.77) (2.08)
HOLDER 67 0.000 0.007
(1.45) (0.30)
LNASSETS 0.000 0.004 0.000 0.005 0.000* 0.005
(1.60) (0.30) (1.58) (0.31) (1.68) (0.30)
M/B RATIO 0.001 0.016 0.001 0.016 0.001 0.020
(1.29) (0.78) (1.30) (0.79) (1.34) (0.97)
BOOKLEV 0.002* 0.097 0.002* 0.095 0.002* 0.094
(1.68) (0.86) (1.67) (0.84) (1.66) (0.82)
PROFITABILITY 0.006* 0.408** 0.006* 0.411** 0.007* 0.437**
(1.81) (2.39) (1.81) (2.41) (1.83) (2.54)
RETEARN_CS 0.001 0.142*** 0.001 0.143*** 0.001 0.143***
(1.27) (2.76) (1.27) (2.77) (1.28) (2.76)
COLLATERAL 0.000 0.015 0.001 0.014 0.001 0.004
(1.07) (0.20) (1.11) (0.19) (1.51) (0.05)
BOARD 0.000 0.001 0.000 0.001 0.000 0.001
(0.52) (0.30) (0.52) (0.31) (0.75) (0.16)
CHAIRMAN 0.000 0.037 0.000 0.035 0.000 0.040
(0.98) (0.94) (0.95) (0.88) (1.01) (1.01)
PCTOWN 0.000 0.258* 0.000 0.265* 0.000 0.245*
(0.25) (1.85) (0.19) (1.88) (0.27) (1.70)
PCTVESTOPT 0.003 0.201* 0.003 0.196* 0.003 0.254**
(1.36) (1.91) (1.35) (1.88) (1.49) (2.43)
SYSTRISK 0.094 1.253 0.095 1.209 0.094 1.723
(1.49) (0.21) (1.50) (0.21) (1.46) (0.30)
HHI_ASSETS 0.001* 0.073 0.001* 0.071 0.001* 0.072
(1.73) (1.55) (1.73) (1.52) (1.72) (1.56)
MILLS 0.000 0.013 0.000 0.011 0.000 0.014
(0.19) (0.25) (0.24) (0.22) (0.02) (0.27)
Observations 1705 1705 1705 1705 1705 1705
Adjusted R2 0.11 0.16 0.11 0.16 0.11 0.15
Panel B: Main regression Results #2 with Heckman correction

Panel I
LONGHOLDER 0.065 0.048 0.331** 0.322**
(0.61) (0.44) (2.29) (2.23)
Panel II
PRE-LONGHOLDER 0.011 0.030 0.460*** 0.432**
(0.07) (0.19) (2.61) (2.30)
POST-LONGHOLDER 0.132 0.119 0.211 0.220
(0.90) (0.85) (1.25) (1.32)
Panel III
HOLDER 67 0.004 0.007 -0.037 0.066
(0.04) (0.06) (-0.28) (0.51)
This table uses Heckman’s two-step correction method to address possible sample selection bias. The first step estimates a selection equation: using the entire Compustat
universe, a selection dummy (=1 if the firm is included at least 4 in one of the Forbes 500 lists between 1984 and 1994) is regressed on the highest ranking a firm achieved
based on sales, profits, assets, and market capitalization between 1984 and 1994 (TOPRANK_SALES, TOPRANK_PROFITS, TOPRANK_ASSETS, and TOPRANK_MKTCAP), and the
number of times the firm appeared in Compustat over this time period (N_OBS). The estimates from this regression are used to compute the inverse mills ratio, which is
included as an explanatory variable in two second stage regressions (i.e., in the main regression models shown in Panels B of Tables 2 and 3). The second-stage results below
are similar to the main results: even after addressing sample selection issues, optimistic CEOs smooth earnings more than rational CEOs (Panel A); and while optimistic CEOs
and rational CEOs are equally likely to show negative and positive surprises, optimistic CEOs are (significantly) less likely to show big surprises and (significantly) more likely
to show small surprises (Panel B).
one minus the sum of squares of each segment’s assets divided by total assets. MILLS is the inverse mills ratio.
*
**
***
through III), and the correlation between accruals and cash flows is tions (see the first two columns in Panels A and B).20 The difference
significantly more negative (see Columns (iii) in Subpanels I between this result and Hribar and Yang (2012), who find that opti-
through III).18 mistic CEOs are more likely than rational CEOs to miss their own
Table 2 Panel B contains the multivariate regression results. earnings forecasts, is intriguing. It can potentially be explained by
Subpanels I through III show results for Longholders, Pre-/Post- the fact that management and analyst forecasts often differ widely
Longholders, and Holders 67, respectively. In each Subpanel, and that management forecasts tend to be more accurate than those
Columns (i) and (ii) present results for two smoothing measures: by analysts only 50% of the time (e.g., Hutton and Stocken, 2010).
VARIABILITY (DNI/TA) and VARIABILITY (DNI/TA) over (DCF/TA). Next, I analyze whether optimistic CEOs are equally likely to
The third smoothing variable is not used as a dependent variable show big and small surprises as rational CEOs. A ‘‘big (small) earn-
in the regressions since this variable is defined as a correlation. ings surprise’’ is defined as an earnings report that misses the med-
The coefficients on the optimism measures are negative and ian analyst report by at least (less than) 3 cents per share.21
significant in both specifications for Longholders and Post-Long- Table 3 shows that optimists are less likely to show big surprises
holders, negative and significant in one specification for Pre-Long- and more likely to show small surprises than rational managers.
holders and negative but not significant for Holders 67. Thus, the While the univariate statistics in Panel A show significance based
regression results generally confirm the univariate results and sug- on three of the four optimism measures, the regression results in
gest that optimistic CEOs smooth earnings (significantly) more Panel B show significance only for Longholders and Pre-Longholders
than their rational counterparts. (see third and fourth columns in Panels A and B).
Turning briefly to the coefficients on the control variables: they
generally have the expected sign. Specifically, earnings are 5. Robustness issues and additional tests
smoother (i.e., less volatile) at firms that are bigger, more profit-
able, and more diversified; they are also smoother at firms that The results above suggest that optimistic CEOs smooth more
have less leverage, greater agency problems, and higher CEO stock and show smaller earnings surprises than rational CEOs. This sec-
and option ownership. As discussed earlier, the direct effect of option establishes the robustness of these results and examines
tions on the manager is to induce greater earnings volatility, but whether optimists are truly optimistic. It first addresses sample
the (indirect) smoothing-induced effect appears to dominate in selection issues (Section 5.1) and the impact of firms’ use of deriv-
the data. This finding is in line with Bergstresser and Philippon atives to smooth earnings (Section 5.2). It then explores three pos-
(2006) who document that CEOs with greater vested option hold- sible reasons for (even rational) managers to delay their option
ings manipulate earnings more, which in turn is consistent with exercise: the manager may possess favorable private information
Burns and Kedia (2006), who find that these CEOs are more likely (Section 5.3), or she may work at a firm with lower stock price vol-
to restate earnings.19 atility (Section 5.4). Next, it examines whether the effects of opti-
mism on smoothing are inadvertently driven by CEOs who
rationally save for bad times (Section 5.5). Finally, to gain a deeper
4.2. Main Result #2: Optimists report smaller earnings surprises
understanding of the earnings surprise results, additional tests
examine whether the findings are driven by big negative, big posi-
Table 3 Panels A and B report the results of univariate tests and
tive, small negative and/or small positive surprises (Section 5.6).
regressions, respectively. As a first test, I examine whether
optimistic managers are more or less likely to report negative or
5.1. Dealing with sample selection
positive surprises than rational managers. Clearly, optimistic and
rational CEOs are equally likely to miss or beat analyst expecta-
The main results are based on a sample of firms that appeared
at least four times on one of the Forbes 500 lists between 1984
and 1994, instead of on the entire universe of Compustat firms. It
18
Correlation coefficients around 0.90 that are documented here are in line with is therefore possible that the main results are driven by non-
the existing literature. Leuz et al. (2003) find average correlations across the countries random selection. Heckman’s two-step correction method is used
in their study of 0.85. Myers et al. (2007) report average correlations of 0.96 for to correct for this potential bias.
their sample firms and 0.93 for their control firms.
19
Bergstresser and Philippon (2006) also provide supporting anecdotal evidence:
20
they mention that Xerox manipulated earnings during the 1990s, during which period In both panels, percentages do not add up to 100% because zero surprises are not
the CEO exercised a large number of stock options. The SEC sued Xerox in 2002 and reported.
21
forced it to reduce reported earnings for the previous five years. I obtain qualitatively similar results using a cutoff of 2 cents per share.
Table 5
Main Results #1 and #2 revisited: controlling for hedging.
Panel A: Main regression Results #1 controlling for hedging activities

(i) (ii) (i) (ii) (i) (ii)
LONGHOLDER 0.001** 0.060**
(2.31) (2.31)
PRE-LONGHOLDER 0.000** 0.044
(2.02) (1.48)
POST-LONGHOLDER 0.001** 0.074**
(2.23) (2.44)
HOLDER 67 0.000 0.017
(1.34) (0.74)
** **
LNASSETS 0.000 0.012 0.000 0.012 0.000** 0.012
(2.17) (0.89) (2.19) (0.86) (2.13) (0.92)
M/B RATIO 0.001 0.015 0.001 0.015 0.001 0.021
(1.25) (0.71) (1.25) (0.70) (1.33) (1.01)
BOOKLEV 0.003* 0.089 0.003* 0.087 0.003* 0.083
(1.79) (0.70) (1.79) (0.69) (1.74) (0.64)
PROFITABILITY 0.006 0.336* 0.006 0.341* 0.007* 0.375**
(1.61) (1.79) (1.62) (1.82) (1.67) (2.02)
RETEARN_CS 0.001 0.169*** 0.001* 0.170*** 0.001* 0.170***
(1.65) (3.08) (1.66) (3.09) (1.66) (3.05)
COLLATERAL 0.000 0.023 0.000 0.022 0.001 0.006
(0.85) (0.32) (0.88) (0.31) (1.29) (0.08)
BOARD 0.000 0.000 0.000 0.000 0.000 0.001
(0.45) (0.02) (0.44) (0.03) (0.71) (0.14)
CHAIRMAN 0.000 0.040 0.000 0.038 0.000 0.043
(0.85) (0.92) (0.78) (0.85) (0.90) (1.01)
PCTOWN 0.001 0.238 0.001 0.248 0.001 0.224
(0.62) (1.46) (0.55) (1.51) (0.64) (1.31)
PCTVESTOPT 0.003 0.198** 0.003 0.189** 0.003* 0.254***
(1.43) (2.14) (1.41) (2.06) (1.66) (2.69)
SYSTRISK 0.087 0.334 0.086 0.323 0.088 0.078
(1.49) (0.05) (1.49) (0.05) (1.44) (0.01)
HHI_ASSETS 0.001 0.091* 0.001 0.089* 0.001* 0.090*
(1.63) (1.83) (1.65) (1.80) (1.67) (1.83)
HEDGER 0.000 0.016 0.000 0.016 0.000 0.013
(0.23) (0.43) (0.16) (0.41) (0.14) (0.33)
Observations 1426 1426 1426 1426 1426 1426
Adjusted R2 0.12 0.19 0.12 0.19 0.12 0.18
Panel B: Main regression Results #2 controlling for hedging activities

Panel I
LONGHOLDER 0.079 0.073 0.327** 0.333**
(0.71) (0.63) (2.05) (2.12)
Panel II
PRE-LONGHOLDER 0.019 0.007 0.484** 0.471**
(0.12) (0.04) (2.56) (2.31)
POST-LONGHOLDER 0.133 0.134 0.178 0.204
(0.87) (0.90) (0.94) (1.13)
Panel III
HOLDER 67 0.028 0.028 0.046 0.073
(0.24) (0.25) (0.33) (0.55)
This table examines whether hedging activities affect the main results and concludes they do not. Controlling for hedging, optimistic CEOs smooth earnings more than
rational CEOs (Panel A); and while optimistic CEOs and rational CEOs are equally likely to show negative and positive surprises, optimistic CEOs are (significantly) less likely
to show big surprises and (significantly) more likely to show small surprises (Panel B).
one minus the sum of squares of each segment’s assets divided by total assets. HEDGER is a dummy that equals 1 if the firm used foreign currency or interest rate derivatives
in 1993 or 1994.
*
**
***
In the first step, I estimate a selection equation. Forbes con- Given this evidence and the fact that my main variables use
structs four lists of the very largest firms based on sales, profits, as- earnings variability and analyst forecasts, it is important to control
sets, and market capitalization, and firms have to appear at least for firms’ use of derivatives in my analyses. A key challenge, how-
four times to be selected. Therefore, using the entire Compustat ever, is that firms did not have to provide any insight into their
universe, I regress a selection dummy (=1 if the firm is included usage over most of my sample period. While firms have been re-
at least 4 in one of the Forbes 500 lists between 1984 and quired to disclose some information (through Financial Accounting
1994) on five variables that likely affect selection: the highest Standard (FAS) 105) from mid-June 1990 onward, it was only from
ranking a firm achieved based on sales, profits, assets, and market mid-December 1994 onward that they have been required
capitalization between 1984 and 1994 (TOPRANK_SALES, (through FAS 119) to provide more insight and distinguish be-
TOPRANK_PROFITS, TOPRANK_ASSETS, and TOPRANK_MKTCAP), and tween financial instruments held for trading versus hedging pur-
the number of times the firm appeared in Compustat over this time poses. This is the reason that older studies on derivatives usage
period (N_OBS).22 The higher a firm’s top rankings and the more generally rely on survey evidence (e.g., Nance et al., 1993). Studies
often it has data in Compustat, the more likely it should be to end that focus on the 1990s typically hand-collect data from 10Ks be-
up in the sample. Indeed, all of the top rankings variables (except cause hedging data are not included in Compustat (e.g., Mian,
TOPRANK_ASSETS) and N_OBS have the correct sign and are highly 1996; Geczy et al., 1997; Allayannis and Weston, 2001; Barton,
significant (not shown for brevity). The estimates from this regres- 2001).24 Since many firms do not report the notional amounts
sion are used to compute the inverse mills ratio (the non-selection hedged, these studies tend to create hedging dummies.
hazard), which is included as an explanatory variable in two I hand-collect hedging data from 10Ks from 1993 to 1998.25 An
second-stage regressions (i.e., in the main regression models shown earlier starting year is infeasible because 10Ks are not readily avail-
in Panels B of Tables 2 and 3). able for most sample firms before 1993. Ending the data collection a
Table 4 shows the second-stage results. Clearly, they are similar few years after 1994 (the last sample year) allows me to establish
to the main results: even after addressing sample selection issues, whether hedging behavior is persistent among my sample firms,
optimistic CEOs smooth earnings more than rational CEOs (Panel and I conclude it is: firms that hedge tend to hedge every year. Based
A); and while optimistic CEOs and rational CEOs are equally likely on this insight, I create HEDGER, a dummy that equals 1 if a firm dis-
to show negative and positive surprises, optimistic CEOs are (sig- closes that it hedges foreign currency and/or interest rate exposures
nificantly) less likely to show big surprises and (significantly) more in 1993 or 1994, and rerun the main regressions using data from the
likely to show small surprises (Panel B). entire sample period. Alternative specifications in which HEDGER is
based on hedging behavior from 1993 to 1998 (instead of 1993–
5.2. Controlling for hedging 1994), or running the regressions using data from 1993 to 1994 only
(instead of the entire sample period) yield results that are similar to
It is well-known that firms may use derivatives as an impor- the ones presented here.
tant risk management tool. However, extending the model of Table 5 Panels A and B show the results which are similar to
Holmstrom and Ricart i Costa (1986) and DeMarzo and Duffie those presented in Panels B of Tables 2 and 3. Thus, controlling
(1995) argue that hedging does not just reallocate risks, but also for firms’ hedging behavior leaves the main results unchanged.
has an informational effect: by reducing external shocks, hedging
increases the information content of reported earnings since it 5.3. Are optimists merely rational CEOs with favorable private
allows shareholders to learn about the quality of the firm’s information about their firms’ future performance?
management and its investment projects. Consistent with this
view, Barton (2001) finds that managers use derivatives to It is possible that CEOs exercise options late not because they
smooth earnings.23 DaDalt et al. (2002) document that derivatives are optimistic but because they have favorable private information
usage is associated with lower asymmetric information: analyst and delay exercise in the hope that the private information will be-
forecasts are more accurate and have lower dispersion for deriva- come public and move the stock price up before exercise. If so, then
tive users. such rational CEOS may be misclassified as optimists, and
differences in smoothing behavior may be driven by the CEO’s
22
Alternative specifications which include each firm’s average (instead of top)
24
rankings, or use the top ranking based on all four lists, or add average growth rates, or Some studies use the ‘‘Database of users of derivatives’’ (e.g., DaDalt et al., 2002).
use a log specification, or use the log of the actual dollar amounts yield second-stage This was published by Swaps Monitor Publications until 1997 and is not available
results that are similar to the ones presented here. online.
23 25
Huang et al. (2008) examine how income smoothing through abnormal accruals Search words used: hedg, swap, and derivative (as in Graham and Rogers,
and derivative usage affect firm value. 2002).
Table 6
Optimists are not rational CEOs with favorable private information.
Panel A: Regress smoothing on optimists who should have exercised (‘‘True Optimists’’), optimists who did OK and controls
Panel I: Smoothing by Panel II: Smoothing by Pre-/ Panel III: Smoothing by
Longholders Post-Longholders Holders 67
(i) (ii) (i) (ii) (i) (ii)
LONGHOLDER: should have exercised 0.001* 0.050*
(1.81) (1.72)
LONGHOLDER: did OK 0.000 0.033
(1.61) (1.47)
PRE-LONGHOLDER: should have exercised 0.000 0.028
(1.57) (0.71)
PRE-LONGHOLDER: did OK 0.000 0.038*
(1.39) (1.70)
POST-LONGHOLDER: should have exercised 0.001* 0.063**
(1.73) (2.10)
POST-LONGHOLDER: did OK 0.000 0.022
(1.40) (0.43)
HOLDER 67: should have exercised 0.001 0.016
(1.47) (0.60)
HOLDER 67: did OK 0.000 0.002
(1.18) (0.08)
All Control Variables Yes Yes Yes Yes Yes Yes
Observations 1705 1705 1705 1705 1705 1705
Adjusted R2 0.11 0.16 0.11 0.16 0.11 0.15
Panel B: CEOs classified as optimists split into those who should have exercised and those who did OK
CEOs classified as optimists Percentage of all CEOs Percentage of all optimists
Panel I
LONGHOLDER 19.4
Should have exercised 15.4 79.2
Did OK 4.0 20.8
Panel II
PRE-LONGHOLDER 8.0
Did OK 2.8 34.3
POST-LONGHOLDER 11.4
Did OK 1.3 11.3
Panel III
HOLDER 67 40.1
Did OK 29.1 72.7
Panel A contains results of OLS regressions of earnings smoothing on CEO optimism and control variables. Optimists are split into those who ‘‘should have exercised early’’
and those who ‘‘did OK’’ based on whether exercising their in-the-money options one year earlier and investing the funds in the S&P500 yields higher or lower returns than
holding onto their options one more year. CEOs who should have exercised early are likely ‘‘true’’ optimists; those who did OK may be rational CEOs with favorable private
information. The results support the main result: (‘‘true’’) optimists smooth earnings more than rational CEOs. Panel B shows summary statistics.
Smoothing variables: In each Panel, Column (i) contains results using VARIABILITY (DNI/TA) which is the variability of earnings, the change in net income divided by total assets.
It is calculated as the variance of residuals from regressions of (DNI/TA) on six control variables. The six control variables include: leverage (total liabilities divided by total
assets); sales growth (percentage annual growth); debt issuance (percentage change in total liabilities); equity issuance (percentage change in shares outstanding adjusted
for splits); annual asset turnover (sales divided by total assets); and size (logarithm of the market value of equity). In each Panel, Column (ii) contains results using
VARIABILITY (DNI/TA) over (DCF/TA), which is the variability of earnings (as defined above) divided by the variability of cash flows, the change in cash flows divided by total
assets (calculated using the same approach as the variability of earnings).
All regressions include a constant, all the control variables (not shown for brevity), and year and industry fixed effects. t-Statistics based on robust standard errors clustered
by firm are in parentheses.
*
**
⁄⁄⁄
inside information about the firm’s future performance rather than 5.4. Are optimists merely rational CEOs at firms with lower stock price
managerial optimism. volatility?
Malmendier and Tate (2005) recognize this possibility, but
note that such information is likely to be transitory. Hence, a Top executives are typically greatly underdiversified in terms of
CEO with favorable private information about her firm may exer- exposure to their own firms’ risks (see Section 3.2). As a result,
cise options late, but should not do this persistently or delay exer- risk-averse managers who work at firms with higher stock price
cise for many years. In contrast, the optimism measures are based risk may have a greater propensity to exercise options early and
on a habitual tendency to exercise options late. Longholders fail to thereby shed some risk, while those who work at firms with lower
exercise in the money options for at least five years. Similarly, risk may be more willing to hold on to their options longer. The
Holders 67 at least twice failed to exercise options that are well evidence in Bettis et al. (2005) is consistent with this conjecture.
in the money. This raises the possibility that managers who are classified as opti-
Nevertheless, I go beyond these observations and specifically mists are merely rational CEOs who work at firms with lower stock
confront the misclassification possibility. One way to empirically price volatility. I now examine this, by focusing first on the firm’s
distinguish between privately-informed rational CEOs and opti- idiosyncratic stock return volatility (unsystematic risk), and then
mistic CEOs is to examine the ex-post performances of their op- on its total stock return volatility.
tion holdings. Specifically, we could view CEOs who personally Each firm’s unsystematic risk is estimated using the same one-
profited from exercising in-the-money options late as those with factor market model that was also used to obtain its systematic
favorable private information, and those who did not profit as risk (see Section 3.5): rit = ai + bi ⁄ rmt + eit, where rit is the return
the true optimists. Following Malmendier and Tate (2005, 2008), of firm i for month t and rmt is the return of the CRSP value-
I therefore decompose CEOs who are classified as optimists into weighted index for month t. As before, the model is estimated
those who profited from exercising late and those who did not. using five years of monthly return data, and observations are
To compute the profitability of late exercise, I compare the Long- dropped if fewer than 36 monthly returns are available. A firm’s
holder’s return from exercising options in the year of option unsystematic risk is computed as the variance of the regression
maturity (which is the year of actual option exercise) with the residuals (Shin and Stulz, 2000). If this alternative explanation
hypothetical return from exercising those options one year earlier were correct, optimists should work at firms with significantly low-
and investing the proceeds in the S&P 500. Similarly, for Holders er unsystematic risk.
67, I compare the return from exercising options in year 6 with The (untabulated) results show that based on the Longholder,
the hypothetical return from exercising options in year 5 (the first Post-Longholder, and Holder 67 measures, optimists work at firms
year in which exercise was possible but the CEO decided against with significantly higher unsystematic risk than rational managers
this) and investing the proceeds in the S&P 500. Since I do not (p-values of 0.064, 0.008, and 0.001, respectively). Admittedly,
know the price at which the CEO exercised her options, I assume however, of critical importance are the results based on the Pre-
that she was able to perfectly time the market and exercised the Longholder measure, since that measure contains the observations
options at the maximum price during the fiscal year. This assump- before the manager has first displayed signs of optimism, i.e., they
tion is conservative in that it biases the outcome in favor of include only the classification phase observations. Based on the
documenting profitable late exercise. Longholders, Pre-/Post- Pre-Longholder measure, managers who are classified as optimists
Longholders and Holders 67 are classified as ‘‘did OK’’ if they do indeed tend to work at firms with lower unsystematic risk, but
earned positive abnormal returns by holding options to expiration importantly, the difference is not significant (p-value 0.344). Thus,
or year 6, respectively; otherwise they are classified as ‘‘should optimists do not seem to be rational CEOs who work at firms with
have exercised.’’ The main regressions are rerun using these two significantly lower unsystematic risk.
component variables. Next, I focus on total stock price volatility and obtain this for
Table 6 Panel A shows the results. The coefficients on the ‘‘did each firm in the sample. Stock price volatility is the annualized vol-
OK’’ variables are negative and (almost) significant in a few cases atility of stock returns, calculated as the standard deviation of five
(based on Longholders and Pre-/Post-Longholders),26 providing years of monthly stock returns multiplied by the square root of
some evidence that CEOs who may have had positive private infor- twelve (see, e.g., Bettis et al., 2005). As an additional check, three
mation did smooth more than rational CEOs. Importantly, however, (rather than five) years of monthly stock returns are used. If this
the coefficients on the ‘‘should have exercised’’ variables are nega- alternative explanation has merit, I should find that stock price vol-
tive and also significant (again limited to Longholders and Pre-/ atility is significantly lower at firms led by (supposedly) optimistic
Post-Longholders).27 This suggests that ‘‘true’’ optimists smooth CEOs.
earnings more than rational CEOs, and confirm the main smoothing Table 7 shows that, even though optimists smooth earnings
result of the paper. more, stock price volatility is significantly higher at firms led by
Table 6 Panel B shows the percentage of Longholders, Pre-/Post- optimists using all optimism definitions and regardless of whether
Longholders and Holders 67 that ‘‘did OK’’ and ‘‘should have stock price volatility is calculated using five years (Column (i)) or
exercised’’. A large percentage of the Longholders (79.2%), Pre- three years (Column (ii)) of monthly stock returns. Thus, optimistic
Longholders (65.7%) and Post-Longholders (88.7%) would have CEOs do not seem to be rational CEOs who work at firms with low-
benefited from exercising options earlier, suggesting that most of er stock price volatilities.
them did not have positive private information but were true opti-
mists. In contrast, among the Holders 67, only 27.3% should have 5.5. Rationally saving for bad times
exercised early, suggesting that many Holders 67 may truly be ra-
tional CEOs with favorable private information rather than opti- As highlighted above, the marginal contribution of my paper is
mists. This may explain why the results are generally weaker that biased beliefs about future performance affects smoothing
based on the Holder 67 measure. behavior of firms: CEOs who have upwardly biased beliefs about
future firm performance (optimistic CEOs) smooth more than CEOs
26
who have rational beliefs. Having upwardly biased beliefs is dis-
The coefficient on Longholder in Panel I Column (i) becomes significant after
controlling for hedging.
tinct from rationally expecting that future performance may be
27
The coefficient on Pre-Longholder in Panel II Column (i) becomes significant after good. Fudenberg and Tirole (1995) focus on the latter to show that
controlling for hedging. the manager will smooth earnings: she boosts earnings in bad
Table 7
Optimists are not rational CEOs working at firms with lower stock price volatility.
N Mean stock price volatility

(i) Calculated using 5 years of monthly returns (ii) Calculated using 3 years of monthly returns
Panel I
LONGHOLDER 331 0.3216 0.3163
RATIONAL 1373 0.2747 0.2695
Difference 0.0469 0.0469
p-Value (0.000)*** (0.000)***
Panel II
PRE-LONGHOLDER 137 0.3081 0.3090
RATIONAL 1567 0.2817 0.2759
p-Value (0.004)*** (0.000)***
POST-LONGHOLDER 194 0.3311 0.3215
RATIONAL 1510 0.2777 0.2730
p-Value (0.000)*** (0.000)***
Panel III
HOLDER 67 684 0.3158 0.3116
RATIONAL 1020 0.2623 0.2564
p-Value (0.000)*** (0.000)***
This table contains results of univariate tests that compare the stock price volatility at firms led by CEOs classified as optimists with the stock price volatility at firms led by
rational managers. Stock price volatility is the annualized volatility of stock returns calculated as the standard deviation of five years or three years of monthly stock returns
multiplied by the square root of twelve. The results show that CEOs classified as optimists work at firms with higher stock price volatility than the firms at which rational
CEOs work, mitigating concerns that CEOs classified as optimists are truly rational CEOs who happen to work at firms with lower stock price volatility and hence are more
willing to hold on to their options longer and also smooth earnings less.
p-Values are in parentheses.


***
times to avoid dismissal and saves for future bad times in good Following DeFond and Park (1997), I also include DISCR_ACCRUALS,
times. (lagged) discretionary accruals calculated as reported total accruals
DeFond and Park (1997) provide interesting supporting empir- minus the fitted values of estimated normal accruals, but I obtain
ical evidence for Fudenberg and Tirole’s theory. Specifically, they qualitatively similar results when I exclude this variable.29
show that discretionary accruals are income-increasing at firms Table 8 shows the results. The coefficients on the main variables
that have poor current performance (relative to the industry med- of interest continue to be significant: Longholders, Pre- and Post-
ian) and good expected future performance (relative to the indus- Longholders, and Holders 67 all smooth more than rational CEOs.
try median), consistent with CEOs at these firms borrowing from Thus, controlling for good current performance and good expected
the future to reduce the threat of dismissal in the current period. performance strengthens the main results of this paper.
They also show that discretionary accruals are income-reducing Biased beliefs about expected future earnings can, in principle,
in the opposite cases – when they have good current performance span the continuum from high pessimism to high optimism. It
and poor expected future performance, consistent with CEOs at would therefore be interesting to assess whether the degree of opti-
these firms saving earnings to possibly reduce the threat of dis- mism affects smoothing: i.e., do managers who are less optimistic
missal in the future. smooth less? The Malmendier and Tate measures do not allow
I now check whether the effect of optimism that I document one to look at CEO pessimism per se,30 but they do allow me to cre-
survives the smoothing effect that DeFond and Park (1997) docu- ate a measure of ‘‘low optimism.’’ Recall that the Holder 67 measure
ment. For this purpose, I create: GOOD CURRENT PERF, a dummy is based on the premise that CEOs are rationally expected to exercise
that equals 1 if the firm’s current EPS exceeds the EPS of the med- options soon after the vesting period is over, provided the options
ian firm in the same industry; GOOD EXPECTED PERF, a dummy that are sufficiently in the money. It classifies a CEO as optimistic if she
equals 1 if the median analyst forecast of next year’s EPS for the twice fails to exercise an option with five years remaining that is
firm exceeds the median analyst forecast of next year’s EPS for at least 67% in the money. One could then argue that a CEO who
the median firm in the same industry;28 GOOD CUR- exercises options with five years remaining that are less than 33%
RENT EXPECTED PERF, an interaction term of the two dummies. I
then rerun my main regressions while controlling for these variables.
29
As in DeFond and Park (1997), I estimate normal accruals using a variation of the
Jones model by regressing total accruals/lag total assets on 1/lag total assets, (the
28
As in DeFond and Park (1997), I use the earlier of the following two forecasts: (a) change in total revenues minus the change in accounts receivable)/lag total assets,
the median forecast in the fourth month after the firm’s fiscal year end; and (b) the and gross property plant and equipment/lag total assets. As before, total accruals are
second median forecast after management announces its current year earnings. The calculated using Eq. (3). Discretionary accruals are then obtained as the difference
former recognizes that some firms announce earnings rather late, and the latter between reported total accruals and the fitted values obtained from the regression.
30
recognizes that the first forecast may have been made before current earnings have Malmendier and Tate (2008) only differentiate between optimists and rational
been announced. CEOs.
Table 8
Controlling for good current and good expected relative performance.

(i) (ii) (i) (ii) (i) (ii)
LONGHOLDER 0.000*** 0.081**
(3.43) (2.41)
PRE-LONGHOLDER 0.000*** 0.076**
(2.90) (-2.03)
POST-LONGHOLDER 0.000*** 0.084**
(2.82) (2.09)
HOLDER 67 0.000 0.052*
(0.01) (1.75)
LNASSETS 0.000 0.047*** 0.000 0.047*** 0.000 0.044**
(0.90) (2.63) (0.90) (2.63) (1.25) (2.32)
M/B RATIO 0.000 0.032 0.000 0.032 0.000 0.032
(0.70) (1.27) (0.70) (1.28) (0.35) (1.18)
BOOKLEV 0.000 0.008 0.000 0.008 0.000 0.026
(0.03) (0.06) (0.04) (0.06) (0.11) (0.19)
PROFITABILITY 0.001 0.266 0.001 0.266 0.001 0.320
(0.78) (1.51) (0.79) (1.52) (0.73) (1.66)
RETEARN_CS 0.001*** 0.160*** 0.001*** 0.160*** 0.001*** 0.170***
(3.03) (2.96) (3.03) (2.96) (3.01) (3.04)
COLLATERAL 0.000 0.017 0.000 0.018 0.000 0.020
(0.37) (0.17) (0.35) (0.18) (0.27) (0.19)
BOARD 0.000 0.008 0.000 0.008 0.000 0.005
(1.08) (1.01) (1.08) (1.01) (0.47) (0.62)
CHAIRMAN 0.000 0.082 0.000 0.080 0.000 0.090
(0.86) (1.18) (0.90) (1.13) (0.82) (1.25)
PCTOWN 0.002** 0.761*** 0.002** 0.767*** 0.002** 0.759***
(2.49) (3.78) (2.46) (3.79) (2.24) (3.34)
PCTVESTOPT 0.001 0.237 0.001 0.236 0.001* 0.209
(1.20) (1.33) (1.18) (1.32) (1.79) (1.24)
SYSTRISK 0.005 17.147** 0.005 17.122** 0.013 12.931
(0.08) (2.10) (0.08) (2.10) (0.20) (1.43)
HHI_ASSETS 0.000 0.150* 0.000 0.149* 0.000 0.139*
(1.26) (1.89) (1.24) (1.88) (1.09) (1.81)
DISCRETIONARY ACCRUALS 0.001 0.226 0.001 0.222 0.002 0.247
(1.26) (1.43) (1.26) (1.41) (1.33) (1.48)
GOOD CURRENT PERF 0.000 0.046 0.000 0.046 0.000 0.025
(0.12) (1.12) (0.15) (1.12) (0.22) (0.58)
GOOD EXPECTED PERF 0.000 0.051 0.000 0.051 0.000 0.049
(1.12) (1.09) (1.13) (1.09) (1.32) (1.04)
GOOD CURRENT EXPECTED PERF 0.000 0.016 0.000 0.017 0.000 0.001
(0.39) (0.32) (0.38) (0.33) (0.86) (0.01)
Observations 656 656 656 656 696 696
Adjusted R2 0.19 0.21 0.19 0.21 0.19 0.21
This table contains results of OLS regressions of cash flow variability on optimism and control variables. It tests whether the main results survive the addition of good current
and good expected relative performance variables from DeFond and Park (1997), and shows they do.
Cash flow volatility measure: Each panel shows results using VARIABILITY (DCF/TA), the variability of cash flows, which is calculated as the variance of residuals from
regressions of (DCF/TA), the change in cash flows divided by total assets, on six control variables. The six control variables include: leverage (total liabilities divided by total
assets); sales growth (percentage annual growth); debt issuance (percentage change in total liabilities); equity issuance (percentage change in shares outstanding adjusted
for splits); annual asset turnover (sales divided by total assets); and size (logarithm of the market value of equity).
Control variables: LNASSETS is the log of total assets. M/B RATIO is the firm’s market-to-book ratio, defined as the market value of assets divided by the book value of assets.
BOOKLEV is interest-bearing debt divided by total assets. PROFITABILITY is EBITDA divided by total assets. RETEARN_CS is retained earnings as a fraction of common stock.
COLLATERAL is tangible assets divided by total assets. BOARD is the number of board members. CHAIRMAN is a dummy that equals 1 if the CEO serves as the chairman of the
board, and 0 otherwise. PCTOWN is the fraction of company stock owned by the CEO. PCTVESTOPT is the number of options exercisable within 60 days from the start of the
year (multiplied by 10) divided by the number of shares outstanding. SYSTRISK is systematic risk, measured as b2 times the variance of the value-weighted market index,
where b is estimated with monthly return data using a one-factor market model. HHI_ASSETS is an asset-based Herfindahl index, calculated at the firm level as one minus the
sum of squares of each segment’s assets divided by total assets. DISCR_ACCRUALS are discretionary accruals, calculated as reported total accruals minus the fitted values of
estimated normal accruals. GOOD CURRENT PERF is a dummy = 1 if actual EPS exceeds median actual industry EPS. GOOD EXPECTED PERF is a dummy = 1 if expected next
period’s EPS exceeds median expected industry EPS. GOOD CURRENT EXPECTED PERF is an interaction term of GOOD CURRENT PERF and GOOD EXPECTED PERF.
All regressions include a constant, year and industry fixed effects. t-Statistics based on robust standard errors clustered by firm are in parentheses.
*
**
***
in the money is a low optimism CEO. I find, however, that only 6 ing CEOs with low optimism. This, however, should not be surprising
CEOs have low optimism. This means that the sample is dominated in light of the theories discussed above. These theories show that
by CEOs who have relatively high optimism, precluding tests involv- CEOs are predictably more optimistic/overconfident on average than
others (e.g., Coval and Thakor, 2005; Goel and Thakor, 2008; Gervais 6. A possible theoretical explanation for the findings
et al., 2011).
Having established that optimistic managers smooth more
5.6. Main earnings surprise result revisited and are associated with smaller earnings surprises than rational
managers, I offer a possible theoretical explanation for these
A final analysis tries to understand the main earnings surprise findings, based on a combination of three ingredients: the
result more deeply. Specifically, I address whether optimistic CEOs torpedo effect, the behavior of optimists relative to rational
are less likely to show both big negative surprises (3 cents per managers, and the risk of litigation/prosecution for earnings
share or more) and big positive surprises (+3 cents per share or misreporting.
more), and are also more likely to show both small negative sur-
prises (between 3 cents and 0 cents per share) and small positive 6.1. Torpedo effect
surprises (between 0 cents and +3 cents per share). To put this
analysis in perspective, recall from Table 3 Panel B that while opti- Skinner and Sloan (2002) have noted empirically that there is a
mists are less (more) likely to show big (small) surprises based on ‘‘market-response asymmetry’’ or ‘‘torpedo effect’’ in that the cost
all optimism measures, significance was found based on the Long- (in terms of stock price reaction) of reporting earnings say a penny
holder and Pre-Longholder measures only. below expectations exceeds the benefit of reporting earnings one
Table 9 shows the regression results. As can be seen, the ‘‘big’’ penny above expectations (see also Burgstahler and Eames,
surprise results for Longholders and Pre-Longholders seems to be 2006). While subsequent empirical work has raised questions
driven by big negative surprises. In contrast, the ‘‘small’’ surprise about this finding (e.g., Payne and Thomas, 2003, 2011), it has also
results for Longholders and Pre-Longholders seems to be driven been documented in the survey evidence of Graham et al. (2005)
by both small negative and small positive surprises. That is, the dif- that managers may behave as if they perceive a torpedo effect.
ference between optimistic and rational CEOs seems to be most So, faced with earnings that exceed expectations, every manager
evident when it comes to big negative surprises and in small abso- prefers to under-report earnings now in order to ‘‘save’’ them for
lute (negative and positive) surprises. the future and be able to reduce or eliminate an earnings shortfall
Table 9
Main Result #2 (earnings surprise) revisited.
Big negative surprise Big positive surprise Small negative surprise Small positive surprise
Panel I
LONGHOLDER 0.238* 0.141 0.170 0.234*
(1.88) (1.00) (1.42) (1.65)
Panel II
PRE-LONGHOLDER 0.370** 0.232 0.367** 0.196
(2.34) (1.38) (2.54) (0.95)
POST-LONGHOLDER 0.129 0.061 0.030 0.268
(0.80) (0.36) (0.18) (1.60)
Panel III
HOLDER 67 0.090 0.005 0.082 0.063
(0.82) (0.04) (0.66) (0.47)
This table tries to understand Result #2 more deeply by splitting ‘‘big’’ surprises into big negative and big positive surprises, and ‘‘small’’ surprises into small negative and
small positive surprises. The findings suggest that optimistic CEOs are significantly less likely to show big negative surprises and significantly more likely to show small
(negative and positive) surprises.
Earnings surprise: Earnings surprises are calculated as actual earnings per share minus the median analyst forecast based on the last one-year ahead forecast before the
earnings announcement date. A surprise is negative (positive) earnings fell short of (exceeded) the median analyst forecast. A surprise is big (small) if the absolute value of
earnings minus the median analyst forecast exceeded (was less than) 3 cents per share. A surprise can be big negative (3 cents per share or more), big positive (+3 cents per
share or more), small negative (between 3 cents and 0 cents per share), or small positive (between 0 and 3 cents per share).
Control variables: LNASSETS is the log of total assets. M/B RATIO is the firm’s market-to-book ratio, defined as the market value of assets divided by the book value of assets.
BOOKLEV is interest-bearing debt divided by total assets. PROFITABILITY is EBITDA divided by total assets. RETEARN_CS is retained earnings as a fraction of common stock.
COLLATERAL is tangible assets divided by total assets. BOARD is the number of board members. CHAIRMAN is a dummy that equals 1 if the CEO serves as the chairman of the
board, and 0 otherwise. PCTOWN is the fraction of company stock owned by the CEO. PCTVESTOPT is the number of options exercisable within 60 days from the start of the
year (multiplied by 10) divided by the number of shares outstanding. SYSTRISK is systematic risk, measured as b2 times the variance of the value-weighted market index,
where b is estimated with monthly return data using a one-factor market model. HHI_ASSETS is an asset-based Herfindahl index, calculated at the firm level as one minus the
sum of squares of each segment’s assets divided by total assets.
All regressions include an intercept, all the control variables, and year and industry fixed effects (not shown for brevity). t-Statistics based on robust standard errors clustered
by firm are in parentheses.
*
**

in the event of a low earnings realization then; since reported and the intertemporal adding-up constraint precludes such behavior
economic earnings converge in the long run, reporting policy can continuing indefinitely, but it may be possible to do it in the short
only shift reported earnings through time without affecting the to- run. However, even in the short run, managers may avoid this be-
tal amount of reported earnings. Similarly, faced with earnings that cause misreporting in any period could invite the risk of being
are below expectations, every manager prefers to ‘‘borrow’’ earn- prosecuted by the SEC or being litigated by investors, especially
ings from the future and over-report earnings now. Thus arises if reported earnings are a lot lower than expected. Thus, the risk
earnings smoothing for all managers. Reported earnings will be of some sort of penalty for misreporting can attenuate to some ex-
less volatile over time than the underlying cash flows of the firm, tent the manager’s propensity to smooth excessively. However, the
and hence will display less intertemporal volatility than in the ab- optimist will be deterred less by this risk than the rational manager
sence of smoothing. Essentially, the market-response asymmetry because of her more rosy assessment of future outcomes. Hence,
makes every manager’s benefit of having a gap between reported there is some, but not maximal, smoothing, and the optimist
earnings and earnings expectations concave in the gap, inducing smooths more.
earnings smoothing by optimists as well as rational managers for
the same reasons that risk-averse individuals smooth consumption 6.4. Combining the three ingredients
over time.
To summarize, the torpedo effect induces all managers to
6.2. Behavior of optimists smooth more – they over-report in bad states and under-report
in good states. Optimistic managers over-report more than rational
Optimistic managers exhibit two forms of behavior relative to managers in bad states, effectively ‘‘borrowing’’ more earnings
rational managers that induce them to smooth more and cause from the future, because they are more bullish about having suffi-
them to end up with smaller earnings surprises relative to market ciently high future earnings to ‘‘pay’’ for this borrowing. This re-
expectations. duces earnings the optimist can report in the good state.
The first is that, because they are more bullish about the future, Moreover, even ignoring this adding-up-constraint effect, the abil-
optimists over-report earnings relative to cash flows or economic ity of optimists to deliver positive earnings surprises in good states
earnings by a greater amount than rational managers in bad states is lower than that of rational managers because the higher level of
(low cash flows). To see this, note that when true cash flows or eco- (expected) baseline earnings leaves them with less room to report
nomic earnings are low, so that an earnings report that accurately higher-than-expected earnings. This intuition delivers both greater
reflects this realization would be below market expectations, the smoothing of reported earnings relative to true cash flows as well
torpedo effect drives the manager to over-report earnings. The as smaller earnings surprises for optimists than for rational
manager realizes that doing this will create a ‘‘reported earnings managers.
deficit’’ for the future, but the greater the likelihood of high future
cash flows or economic earnings, the smaller is the manager’s 7. Conclusion
assessment of the expected cost of this future deficit. Since the
optimist believes that the likelihood of high future cash flows is The main goal of this paper has been to examine the effect of a
higher than what the rational manager believes, the optimist re- specific managerial behavioral bias, optimism, on earnings
ports higher earnings than the rational manager in this bad state. smoothing and earnings surprises. The intended contribution is
In the good state, when cash flows are high, the optimistic manager to add to the small but growing literature on how manager-specific
has more to ‘‘pay back’’ for past over-reporting, and hence ends up attributes can explain differences in earnings smoothing practices
reporting lower earnings than the rational manager. Thus, the opti- across firms.
mist smooths more. In line with the existing literature on behavioral biases, mana-
The second form of behavior that distinguishes the optimists is gerial optimism is defined as an upward bias in the assessment
that they provide higher earnings guidance (see Hribar and Yang, of firm-specific future outcomes. Existing optimism measures
2012). Even apart from formal earnings guidance, an optimistic based on the timing of exercise of executive stock options are used
CEO may informally communicate in a more bullish way about in the analyses.
the firm’s future prospects, and this may subtly introduce an up- The paper has two main findings. First, firms with optimistic
ward bias in analyst expectations of future earnings. When true CEOs smooth earnings more than firms with rational managers.
economic earnings are high, both types of managers under-report Second, while optimistic managers are equally likely to show posi-
earnings relative to this realization (torpedo effect). The optimist tive or negative earnings surprises as rational managers, they are
is prone to under-report by a smaller amount than the rational less likely to report large earnings surprises and are more likely
manager, but ends up under-reporting by a larger amount to report small earnings surprises than rational managers. These
because of greater past over-reporting in the bad state and results obtain even after controlling for other factors that may af-
the intertemporal adding-up constraint. Moreover, because fect earnings smoothing and earnings surprises, such as firm size,
analysts have higher earnings expectations for the optimists, even market-to-book, leverage, profitability, asymmetric information,
if the under-reporting by optimists in the good state was of a agency problems, corporate governance, CEO stock and option
smaller magnitude, the positive earnings surprise for the optimists ownership, systematic risk, diversification, and year and industry
is smaller because the baseline defined by analysts’ expectations fixed effects. A variety of tests are performed to examine whether
is higher. optimists are truly optimistic and to establish the robustness of the
results.
6.3. Litigation/prosecution risk I also offered a possible theoretical explanation for the
smoothing and earnings surprise findings, based on a combina-
The torpedo effect (which implies a concave payoff function tion of three ingredients: the torpedo effect, the behavior of opti-
that is similar to the preference of a risk averse individual) suggests mists relative to rational managers, and the risk of litigation/
that both rational and optimistic managers would like to take prosecution for earnings misreporting. While other explanations
smoothing to its natural limit and report constant earnings and for why optimists smooth more and show smaller earnings sur-
end up with zero earnings surprises, which would eliminate any prises may exist, I leave that as an interesting topic for future
differences between rational and optimistic managers. Of course, research.
Acknowledgement Fortune, 1997. Learn to Play the Earnings Game (and Wall Street Will Love You)
(March 31), pp. 77–80.
Francis, Jennifer, LaFond, Ryan, Olsson, Per M., Schipper, Katherine, 2004. Costs of
I thank an anonymous referee, Mo Khan, and Doug Skinner for equity and earnings attributes. Accounting Review 79 (4), 967–1010.
helpful comments, and Brian Hall for providing data. Fudenberg, Drew, Tirole, Jean, 1995. A theory of income and dividend smoothing
based on incumbency rents. Journal of Political Economy 103, 75–93.
Ge, Weili, Matsumoto, Dawn, Zhang, Jenny, 2011. Do CFOs have style? An empirical
References investigation of the effect of individual CFOs on accounting practices.
Contemporary Accounting Research 28, 1141–1179.
Geczy, Christopher, Minton, Bernadette A., Schrand, Catherine, 1997. Why firms use
Allayannis, George, Weston, James P., 2001. The use of foreign currency derivatives
currency derivatives. Journal of Finance 52 (4), 1323–1354.
and firm market value. Review of Financial Studies 14 (1), 243–276.
Gervais, Simon, Heaton, J.B., Odean, Terrance, 2011. Overconfidence, compensation
Almeida, Heitor, Campello, Murillo, 2007. Financial constraints, asset tangibility,
contracts, and capital budgeting. Journal of Finance 66, 1735–1777.
and corporate investment. Review of Financial Studies 20, 1429–1460.
Goel, Anand, Thakor, Anjan V., 2003. Why do firms smooth earnings? Journal of
Asdrubali, Pierfederico, Kim, Soyoung, 2008. On the empirics of international
Business 76 (1), 151–192.
smoothing. Journal of Banking and Finance 32, 374–381.
Goel, Anand, Thakor, Anjan V., 2008. Overconfidence, CEO selection and corporate
Baker, Malcolm, Ruback, Richard S., Wurgler, Jeffrey, 2007. Behavioral corporate
governance. Journal of Finance 63, 2737–2784.
finance. In: Espen Eckbo, B. (Ed.), The Handbook of Corporate Finance. Elsevier
Gong, Guojin, Louis, Henock, Sun, Amy X., 2008. Earnings management and firm
(Chapter 4).
performance following open-market repurchases. Journal of Finance 63, 947–
Bange, Mary M., De Bondt, Werner F.M., 1998. R&D budgets and corporate earnings
986.
targets. Journal of Corporate Finance 4, 153–184.
Graham, John R., Rogers, Daniel A., 2002. Do firms hedge in response to tax
Barth, Mary E., Elliott, John A., Finn, Mark W., 1999. Market rewards associated with
incentives? Journal of Finance 57, 815–839.
patterns of increasing earnings. Journal of Accounting Research 37 (2), 387–413.
Graham, John R., Harvey, Campbell R., Rajgopal, Shiva, 2005. The economic
Barton, Jan, 2001. Does the use of financial derivatives affect earnings management
implications of corporate financial reporting. Journal of Accounting and
decisions? Accounting Review 76, 1–26.
Economics 40, 3–73.
Beidleman, Carl R., 1973. Income smoothing: the role of management. Accounting
Hackbarth, Dirk, 2009. Determinants of corporate borrowing: a behavioral
Review 48, 653–667.
perspective. Journal of Corporate Finance 4, 389–411.
Ben-David, Itzhak, Graham, John R., Harvey, Campbell R., 2007. Managerial
Hall, Brian J., Liebman, Jeffrey B., 1998. Are CEOs really paid like bureaucrats?
Optimism and Corporate Policies. Duke University Working Paper.
Quarterly Journal of Economics 113, 653–691.
Berger, Philip G., Ofek, Eli, 1995. Diversification’s effect on firm value. Journal of
Hall, Brian J., Murphy, Kevin J., 2002. Stock options for undiversified executives.
Financial Economics 37, 39–65.
Journal of Accounting and Economics 33 (1), 3–42.
Bergstresser, Daniel B., Philippon, Thomas, 2006. CEO incentives and earnings
Hand, John R.M., 1989. Did firms undertake debt-equity swaps for an accounting
management. Journal of Financial Economics 80, 511–529.
paper profit or true financial gain? Accounting Review 64 (4), 587–623.
Bertrand, Marianne, Schoar, Antoinette, 2003. Managing with style: the effect of
Healy, Paul M., 1985. The effect of bonus schemes on accounting decisions. Journal
managers on firm policies. Quarterly Journal of Economics 118, 1169–1208.
of Accounting and Economics 7 (1–3), 85–107.
Bettis, J. Carr, Bizjak, John M., Lemmon, Michael L., 2005. Exercise behavior,
Holmstrom, Bengt, Ricart i Costa, Joan, 1986. Managerial issues and capital
valuation, and the incentive effects of employee stock options. Journal of
management. Quarterly Journal of Economics 101, 835–860.
Hribar, Paul, Yang, Holly, 2012. CEO Overconfidence and Management Forecasting.
Bhattacharya, Utpal, Daouk, Hazem, Welker, Michael, 2003. The world price of
Working Paper.
earnings opacity. Accounting Review 78, 641–678.
Huang, Pinghsun, Zhang, Yan, Deis, Donald R., Moffitt, Jacquelyn S., 2008. Do
Black, Fischer, Scholes, Myron, 1973. The pricing of options and corporate liabilities.
artificial income smoothing and real income smoothing contribute to firm value
Journal of Political Economy 81 (3), 637–654.
equivalently? Journal of Banking and Finance 33, 224–233.
Bowen, Robert M., Rajgopal, Shivaram, Venkatachalam, Mohan, 2008. Accounting
Hutton, Amy P., Stocken, Phillip C., 2010. Prior Forecasting Accuracy and Investor
discretion, corporate governance and firm performance. Contemporary
Reaction to Management Earnings Forecasts. Working Paper.
Accounting Research 25 (2), 351–405.
Jin, Li, Kothari, S.P., 2008. Effect of personal taxes on managers’ decisions to sell
Burgstahler, David, Eames, Michael, 2006. Management of earnings and analysts’
their stock. Journal of Accounting and Economics 46, 23–46.
forecasts to achieve zero and small positive earnings surprises. Journal of
Klein, April, 2002. Audit committee, board of director characteristics, and earnings
Business Finance and Accounting 33 (5–6), 633–652.
management. Journal of Accounting and Economics 33, 375–400.
Burns, Natasha, Kedia, Simi, 2006. The impact of performance-based compensation
Kohers, Ninon, Ang, James, 2001. Earnouts in mergers: agreeing to disagree and
on misreporting. Journal of Financial Economics 79, 35–67.
agreeing to stay. Journal of Business 73 (3), 445–476.
Chaney, Paul K., Lewis, Craig M., 1995. Earnings management and firm valuation
Land, Judy, Lang, Mark H., 2002. Empirical evidence on the evolution of
under asymmetric information. Journal of Corporate Finance 1, 319–345.
international earnings. Accounting Review 77, 115–133.
Chatterjee, Samprit, Hadi, Ali S., Price, Bertram, 2000. Regression Analysis by
Lang, Larry H.P., Stulz, Rene M., 1994. Tobin’s q, corporate diversification, and firm
Example, third ed. John Wiley & Sons.
performance. Journal of Political Economy 102, 1248–1280.
Christie, Andrew A., Zimmerman, Jerold L., 1994. Efficient and opportunistic choices
Lang, Mark, Raedy, Jana Smith, Wilson, Wendy, 2006. Earnings management and
of accounting procedures: corporate control contests. Accounting Review 69
cross listing: are reconciled earnings comparable to US earnings? Journal of
(4), 539–566.
Accounting and Economics 42, 255–283.
Comment, Robert, Jarrell, Gregg A., 1995. Corporate focus and stock returns. Journal
Leuz, Christian, Nanda, Dhananjay, Wysocki, Peter D., 2003. Earnings management
of Financial Economics 37, 67–87.
and investor protection: an international comparison. Journal of Financial
Coval, Joshua, Thakor, Anjan V., 2005. Financial intermediation as a beliefs-bridge
Economics 69, 505–527.
between optimists and pessimists. Journal of Financial Economics 75 (3), 535–
Lev, Baruch, Kunitzky, Sergius, 1974. On the association between smoothing
570.
measures and the risk of common stocks. Accounting Review 49 (2), 259–270.
DaDalt, Peter, Gay, Gerald D., Nam, Jouahn, 2002. Asymmetric information and
Malmendier, Ulrike, Tate, Geoffrey, 2005. CEO optimism and corporate investment.
corporate derivatives use. Journal of Futures Markets 22, 241–267.
Journal of Finance 60 (6), 2661–2700.
DeAngelo, Harry, DeAngelo, Linda, Stulz, Rene M., 2006. Dividend policy and the
Malmendier, Ulrike, Tate, Geoffrey, 2008. Who makes acquisitions? CEO optimism
earned/contributed capital mix: a test of the lifecycle theory. Journal of
and the market’s reaction. Journal of Financial Economics 88, 323–354.
Manove, Michael, Padilla, A. Jorge, 1999. Banking (conservatively) with optimists.
Dechow, Patricia M., Skinner, Douglas J., 2000. Earnings management: reconciling
Rand Journal of Economics 30 (2), 324–350.
the views of accounting academics, practitioners, and regulators. Accounting
Mansi, Sattar A., Reeb, David M., 2002. Corporate diversification: what gets
Horizons 14 (2), 235–250.
discounted? Journal of Finance 57, 2167–2183.
Dechow, Patricia M., Sloan, Richard G., Sweeney, Amy P., 1995. Detecting earnings
May, Don O., 1995. Do managerial motives influence firm risk reduction strategies?
management. Accounting Review 70 (2), 193–225.
Journal of Finance 50, 1291–1308.
Dechow, Patricia M., Hutton, Amy P., Kim, Jung Hoon, Sloan, Richard G., 2012.
Merton, Robert C., 1973. Theory of rational option pricing. Bell Journal of Economics
Detecting earnings management: a new approach. Journal of Accounting
4 (1), 141–183.
Research 50, 275–334.
Mian, Shehzad L., 1996. Evidence on corporate hedging policy. Journal of Financial
DeFond, Mark L., Park, Chul W., 1997. Smoothing income in anticipation of future
and Quantitative Analysis 31 (3), 419–439.
earnings. Journal of Accounting and Economics 23, 115–139.
Myers, Stewart, Majluf, Nicholas S., 1984. Corporate financing and investment
DeMarzo, Peter M., Duffie, Darrell, 1995. Corporate incentives for hedging and
decisions when firms have information that investors do not have. Journal of
hedge accounting. Review of Financial Studies 8, 743–771.
Financial Economics 13 (2), 187–221.
Denis, David J., Denis, Diane K., Yost, Keven, 2002. Global diversification, industrial
Myers, James N., Myers, Linda A., Skinner, Douglas J., 2007. Earnings momentum
diversification, and firm value. Journal of Finance 57, 1951–1979.
and earnings management. Journal of Accounting, Auditing and Finance 22 (2),
Dhaliwal, Dan S., Gleason, Christi A., Mills, Lillian F., 2004. Last-chance earnings
249–284.
management: using the tax-expense to meet analysts’ forecasts. Contemporary
Nance, Deana R., Smith Jr., Clifford W., Smithson, Charles W., 1993. On the
Accounting Research 21 (2), 431–459.
determinants of corporate hedging. Journal of Finance 48, 267–284.
Fonseca, Ana Rosa, Gonzalez, Francisco, 2008. Cross-country determinants of bank
Page, Lionel, 2008. Is there an optimistic bias on betting markets? Economics Letters
income smoothing by managing loan-loss provisions. Journal of Banking and
102, 70–72.
Finance 32, 217–228.
Payne, Jeff L., Thomas, Wayne B., 2003. The implications of using stock-split Skinner, Douglas J., Sloan, Richard G., 2002. Earnings surprises, growth expectations
adjusted I/B/E/S in empirical research. Accounting Review 78 (4), 1049–1068. and stock returns, or don’t let an earnings torpedo sink your portfolio. Review of
Payne, Jeff L., Thomas, Wayne B., 2011. The torpedo effect: myth or reality? Journal Accounting Studies 7 (2–3), 289–312.
of Accounting, Auditing and Finance 26, 255–278. Thomas, Jacob K., Zhang, Huai, 2002. Inventory changes and future returns. Review
Phillips, John, Pincus, Morton, Rego, Sonja O., 2003. Earnings management: new of Accounting Studies 7 (2–3), 163–187.
evidence based on deferred tax expense. Accounting Review 78 (2), 491–521. Trueman, Brett, Titman, Sheridan, 1988. An explanation for accounting income
Puri, Manju, Robinson, David, 2007. Optimism and economic choice. Journal of smoothing. Journal of Accounting Research 26 (Suppl.), 127–139.
Financial Economics 86 (1), 71–99. Van den Steen, Eric, 2004. Rational overoptimism (and other biases). American
Ronen, Joshua, Sadan, Simcha, 1981. Smoothing Income Numbers: Objectives, Economic Review 94, 1141–1151.
Means, and Implications. Addison-Wesley Publishing Company. Warfield, Terry D., Wild, John J., Wild, Kenneth L., 1995. Managerial ownership,
Schrand, Catherine M., Zechman, Sarah L.C., 2012. Executive overconfidence and the accounting choices, and the informativeness of earnings. Journal of Accounting
slippery slope to financial misreporting. Journal of Accounting and Economics and Economics 20 (1), 61–91.
53 (1), 311–329. Yermack, David, 1995. Do corporations award CEO stock options effectively?
Shin, Hyun-Han, Stulz, Rene, 2000. Firm Value, Risk, and Growth Opportunities. Journal of Financial Economics 39 (2–3), 237–269.
NBER Working Paper 7808.

Managerial Optimism and Earnings Smoothing - 104642

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

Managerial Optimism and Earnings Smoothing - 104642

Uploaded by

Copyright:

Available Formats

Journal of Banking & Finance 41 (2014) 283–303

Contents lists available at ScienceDirect

Journal of Banking & Finance

1. Introduction (Graham et al., 2005).1 However, the degree of earnings smoothing

Panel A: Number of ﬁrms and number of CEOs

Obs Mean Median Minimum Maximum

Panel A: Univariate results

Panel B: Regression results

Negative surprise Positive surprise Big surprise Small surprise

Panel B: Regression results

Panel A: Main regression Results #1 with Heckman correction

Panel B: Main regression Results #2 with Heckman correction

Panel A: Main regression Results #1 controlling for hedging activities

Panel B: Main regression Results #2 controlling for hedging activities

N Mean stock price volatility

Panel I: Longholders Panel II: Pre-/Post-Longholders Panel III: Holders 67

You might also like