Stock Options as Lotteries

BRIAN H. BOYER and KEITH VORKINK∗

ABSTRACT

We investigate the relationship between ex-ante total skewness and holding returns on
individual equity options. Recent theoretical developments predict a negative relationship
between total skewness and average returns, in contrast to the traditional view that only
coskewness is priced. We find, consistent with recent theory, that total skewness exhibits
a strong negative relationship with average option returns. Differences in average returns
for option portfolios sorted on ex-ante skewness range from 10% to 50% per week, even
after controlling for risk. Our findings suggest that these large premiums compensate
intermediaries for bearing unhedgable risk when accommodating investor demand for
lottery-like options.

∗ Boyer and Vorkink are with The Marriott School of Management, Brigham Young University. We acknowl-
edge financial support from the Harold F. and Madelyn Ruth Silver Fund, and Intel Corporation. Vorkink notes
support from a Ford research fellowship. We thank Turan Bali, Nick Barberis, Josh Coval, Jefferson Duarte,
Danling Jiang, Chris Jones, Don Lien, Francis Longstaff, Grant McQueen, Todd Mitton, Lasse Pedersen, Josh
Pollet, Tyler Shumway, Grigory Vilkov, seminar participants at Brigham Young University, Florida State, UC
Irvine, UCLA, and the University of Michigan, and conference participants at the 2012 Adam Smith Asset Pric-
ing Conference and the 2012 Western Finance Association Meetings for helpful comments and suggestions. We
also thank Greg Adams and Troy Carpenter for research support.
 Recent research shows that standard rational asset pricing models have difficulty explain-
ing many of the basic empirical facts about the aggregate stock market, the cross section of
average returns, and individual trading behavior. Furthermore, experimental economists find
that individuals deviate from standard utility theory when making choices in the face of uncer-
tainty (see, for example, Kahneman and Tversky (1979)). As a result, many researchers have
turned their attention to the asset pricing implications of models that depart from the standard
representative agent/expected utility framework.

One prominent departure considers investors who prefer skewness or lottery-like features
in asset return distributions, and how these preferences influence asset prices in equilibrium.1
These newer models posit that total skewness, including idiosyncratic skewness, is priced,
because investors optimally choose to underdiversify. Asset returns in these models have a
strong negative relationship with total skewness. In stark contrast, the prevailing view is that
only coskewness with the market is value relevant (Kraus and Litzenberger (1976), and Harvey
and Siddique (2000)).2

The individual equity options market offers an ideal arena to study these competing views
of skewness preference on asset prices for three reasons. First, the implicit leverage in an
option contract combined with a nonlinear payoff creates unusually dramatic lottery-like fea-
tures in option returns. The ex-ante return skewness of option contracts can easily be more
than 10 times higher than equity return skewness. Second, cross-sectional variation in ex-ante
skewness is relatively straightforward to identify among options, both for the investor and
the econometrician. Third, the skewness of individual equity options is largely idiosyncratic.
Since the moments of index options are predominantly systematic, index options are poorly
suited to differentiate between the effects of total skewness and coskewness on prices.3

Our findings strongly suggest that total skewness is priced. Portfolios of short-term op-
tions with high ex-ante skewness lose about 10% to 50% per week on average. Portfolios
of otherwise similar options with low ex-ante skewness earn average weekly returns close to
zero. We find a statistically significant and economically large effect of total skewness prefer-

2
ence on option prices in both call and put option markets and across maturities ranging from
one week to about three months after controlling for the influence of risk and other charac-
teristics following Duarte and Jones (2007) and Broadie, Chernov, and Johannes (2009). We
base these findings on a simple ex-ante measure of option skewness that we develop and use in
our empirical tests. The magnitude and robustness of the negative relationship between total
skewness and returns suggest that skewness preference may be of first-order importance in the
pricing of securities with extreme lottery-like features.

Recent papers find that total skewness is priced in stocks. This literature also concludes,
however, that estimating ex-ante skewness for stock returns is quite difficult because the cor-
rect set of predictive instruments is not known. Lagged skewness in particular is a poor in-
strument for ex-ante skewness since low probability events that largely determine estimates
of the third moment are not persistent.4 In contrast, our measure of ex-ante option return
skewness is simple to construct and demonstrates that most of the cross-sectional variation in
ex-ante skewness for individual option returns, holding maturity fixed, can be explained by
just two variables: moneyness, and underlying asset return volatility. Moneyness is directly
observable, and underlying asset return volatility is relatively straightforward to estimate. We
also show that option return skewness is largely unaffected by the ex-ante skewness of the
underlying stock return.

Our findings also contribute generally to the literature in the following three ways. First,
because we analyze the full cross section of both call and put options on individual equities,
our results extend the findings of Coval and Shumway (2001) and Ni (2009) to broaden our
understanding of average option returns. Second, our results support the case for demand-
based option pricing as in Garlaneau, Pedersen, and Poteshman (2009) by showing that one
important source of demand in option markets is skewness preference, and that this demand
has an economically significant effect on prices. Third, our results contribute to the vast liter-
ature on the relation between option prices and moneyness.5 A stylized fact of this literature
is that out-of-the-money options are overvalued relative to standard models. Ni (2009) adds to

3
this literature by showing that out-of-the-money call options earn low average returns. In this
paper, we show that ex-ante skewness unrelated to moneyness is priced in both call and put
securities. In contrast, after controlling for our measure of ex-ante skewness, we find that the
relationship between option returns and moneyness documented by Ni (2009) largely disap-
pears. Our results suggest that investor demand for lottery-like assets combined with limits-
to-arbitrage in the form of market-maker hedging costs may help explain the well-documented
relation between moneyness and option prices.

We test whether state variables commonly used in asset pricing can explain the incredibly
low average option portfolio returns we document, including the excess market return, excess
returns on a zero-delta index straddle, and the coskewness factor of Harvey and Siddique
(2000). We find that average returns of the underlying stocks can be largely explained by
return covariation with the excess market return. Average option returns, however, cannot be
explained by return covariation with any of the state variables we consider.

We confirm our findings in Fama-MacBeth (1973) regressions that allow us to jointly

control for the influence of a variety of risk factors and other characteristics, as well as in
double-sorted portfolio tests that enable us to control for characteristics while allowing for
nonlinearities in the pricing relationships. Our main results hold after controlling for two
different measures of coskewness, the bid-ask spread, volume, and even moneyness.6

Empirical work in option pricing typically relies on the estimation of fully specified para-
metric models. Option returns are more straightforward to interpret economically than the
pricing errors of such models because returns represent the actual gains or losses to an in-
vestor on purchased securities. Several others have also noted the advantages of analyzing
average (risk-adjusted) option returns.7 For example, Duarte and Jones (2007), who analyze
equity options using Fama-MacBeth (1973) regressions, argue that many parametric option
pricing models impose a highly rigid structure on both risk premia and the relative riskiness of
different option contracts in a manner that can lead to misleading conclusions. Further, the fea-

4
sibility of imposing realistic parametric assumptions on the entire cross section of underlying
assets is questionable.

Of course, there are obvious issues to consider when applying standard asset pricing met-
rics to option returns given their nonnormality and nonlinearity. Broadie, Chernov, and Jo-
hannes (2009) apply standard asset pricing methods to option returns but anchor hypothesis
tests at appropriate null values by comparing estimated parameters to those obtained using
artificial data generated under formal option pricing models. Following their procedure we
first simulate data under the usual Black-Scholes (1973) assumptions and then again under the
jump diffusion model of Merton (1976). In both simulations we impose the null that skewness
is not priced. We then apply the same standard asset pricing methods to the simulated data as
we do using the actual data. These simulations inform us regarding how often properties such
as the nonnormality and nonlinearity of option returns can lead to results as extreme as those
found in the actual data. We find the answer to be virtually never: the results we estimate
using the actual data look quite abnormal relative to results simulated under these two models.
In fact, our simulation exercise leads to the same conclusions we arrive at using standard em-
pirical methods, namely, that total skewness is priced in the cross-section of individual equity
option returns.

While investors appear willing to pay substantial premiums for the lottery-like character-
istics of individual equity options, we find that the losses of option buyers are not passed on
as gains to investors who write options. We find that investors who write lottery-like options
near the bid earn risk-adjusted returns that are generally insignificant from zero and do not
vary systematically with ex-ante skewness. This helps explain why the remarkably low av-
erage returns on options with high ex-ante skewness are not arbitraged away and can persist
in equilibrium. We conjecture that the ability of intermediaries to effectively hedge short po-
sitions in individual options deteriorates with the ex-ante skewness of the option.8 High ask
prices therefore compensate intermediaries for bearing the unhedgeable risk of writing op-
tions with high ex-ante skewness while bid prices are more closely aligned to the textbook

5
no-arbitrage relations with the stock. Garleanu, Pedersen, and Poteshman’s (2009) model of
option pricing illustrates that demand pressure on an option’s price is related to the variance
of the unhedgeable part of the option.9 Building on this work, we begin to uncover sources of
such demand pressure.

Our findings do not appear to be merely driven by restrictions in supply by intermediaries.

In a small data set of intraday transaction prices and quotes, we find that nearly half of all
trades in options with high ex-ante skewness occur near or at the ask price. These findings
corroborate our interpretation of earlier results, that a strong preference for total skewness
exists among option investors. Barberis and Huang (2008), for example, study the implications
of cumulative prospect theory for asset pricing but state “... the fact that people depart from
expected utility in experimental settings does not necessarily mean that they also do so in
financial markets.” Our study of the individual equity options market suggests that perhaps
people do, or at a mimimum that we should further consider the implications of models in
which total skewness is priced to help us understand other asset pricing phenomena.

The rest of the paper is organized as follows. Section I introduces our ex-ante skewness
measure. In Section II we form option portfolios and report their average returns. Section III
presents results on average portfolio returns after controlling for risk and other option charac-
teristics. Section IV investigates the returns from writing options at the bid price. Section V
offers concluding remarks.

I. Ex-Ante Skewness

To understand whether differences in the lottery-like characteristics of options help ex-

plain cross-sectional variation in their expected returns, we make the simplifying assumption
that skewness is a proxy for the lottery-like characteristics of options consistent with much of
the lottery-preference literature (e.g., Brunnermeier and Parker (2005), Barberis and Huang
(2008), and Mitton and Vorkink (2007)). We construct closed-form ex-ante skewness mea-

6
sures for the physical distribution of option returns by integrating the appropriate PDF under
the assumption that stock prices are lognormal.

The lognormal assumption does not perfectly characterize the distribution of the under-
lying stocks. We make this assumption because it allows for a simple approach to estimate
the physical ex-ante skewness of an option contract that uses only information available to
an investor at the time of purchase and because of its familiarity to others in the finance pro-
fession. Integrating the truncated lognormal PDF to obtain closed-form moments for options
is relatively simple (Lien (1985)). As such, our method to derive option moments should be
straightforward for readers to understand and less prone to accusations of “cherry picking”
assumptions.

In the Internet Appendix, we explore the limits of the lognormal assumption for our
analysis.10 In particular, we account for ex-ante underlying stock skewness (both positive
and negative) that the lognormal PDF is unable to represent. We derive ex-ante stock skew-
ness similar to Boyer, Mitton, and Vorkink (2010) and find that cross-sectional variation in
ex-ante underlying stock skewness explains little of the variation in realized option skewness.
We also find that accounting for ex-ante underlying stock skewness even further strengthens
our pricing results. In light of these findings, the lognormal assumption appears to be con-
servative. We report these results in an Internet Appendix, and further discuss the drivers of
cross-sectional variation in option return skewness below. The critical issue for our asset pric-
ing tests is whether our measure actually predicts realized skewness in option returns. Below
in Section II.A we demonstrate that it does.

A. Ex-Ante Skewness under Lognormality

We define our measure of ex-ante skewness for option i over horizon t to T as

Et [Ri,t:T − µi,t:T ]3
ski,t:T = , (1)
[σi,t:T ]3

7
where Ri,t:T denotes option i’s return, Et [·] denotes the expectation given information known
at time t, µi,t:T = Et [Ri,t:T ], and σi,t:T = (Et [R2i,t:T ] − µ2i,t:T )1/2 . By rewriting equation (1) in
terms of its raw moments,
h i h i
Et Ri,t:T − 3Et Ri,t:T µi,t:T + 2µ3i,t:T
3 2
ski,t:T = h h i i1.5 , (2)
2 2
Et Ri,t:T − µi,t:T

we see that only the first three raw moments of the option return are required to calculate
ski,t:T .

To understand how we calculate these raw moments, note that the return from holding a
call option to maturity, Rci,t , is simply

(Si,T − Xi )+
Rci,t:T = , (3)
Ci,t

where (·)+ is the max(0, ·) function, Si,T is the value of the underlying asset at maturity, Xi is
the exercise price, and Ci,t is the call premium at time t. Equation (3) indicates that the jth raw
moment for call option i can be written
" j #
h
ji Si,T − Xi
Et Rci,t:T = Et |Si,T > Xi Pt (Si,T > Xi ), (4)
Ci,t

where Pt (·) indicates the probability given information as of time t. Assuming that Si,T is dis-
tributed lognormally, equation (4) illustrates the raw moments for a call option are a function
of the raw moments of a truncated lognormal distribution. Lien (1985) derives the moments of
a truncated lognormal distribution that we use to construct ski,t:T for any option contract. We
further demonstrate how to construct our expected skewness measure, ski,t:T , in the Appendix
below.

8
B. Option Characteristics and Skewness

We provide plots to understand how different option characteristics influence our expected
skewness measure, ski,t:T . Figure 1 plots ski,t:T as a function of moneyness, Xi /Si,t , for both
call and put options and for a number of maturities. This figure illustrates that a strong rela-
tionship exists between moneyness and ex-ante skewness. Options trading out-of-the-money
offer substantially more skewness than in-the-money options, especially as maturity decreases.
The ex-ante skewness of short-term, out-of-the-money options is well over 10, several times
higher than the ex-ante skewness of equity returns (see Boyer, Mitton, and Vorkink (2010) and
Conrad, Dittmar and Ghysels (2012)). One other observation from Figure 1 is that put options
can offer skewness opportunities that are at least as large as their corresponding call options.

-FIGURE 1 ABOUT HERE-

In Figure 2 we plot the relationship between ski,t:T and the annualized volatility of the
underlying stock return (σsi ). Panels A and B of Figure 2 plot the relationship for options
trading at a moneyness level of 0.9, corresponding to in-the-money call options and out-of-
the-money put options. We see that the volatility of the underlying can have a strong impact
on skewness, but that the magnitude of the relationship is influenced by both maturity and
moneyness. In Panel A we see that higher underlying stock volatility results in slightly higher
skewness for in-the-money call options. However, in Panel B the relationship flips for out-
of-the-money put options: higher underlying stock volatility leads to much lower skewness.
Panels C and D of Figure 2 plot the relationship for a moneyness level of 1.1 corresponding to
out-of-the-money call options and in-the-money put options. Similar to out-of-the-money put
options, the relationship between volatility and skewness is strong and negative for out-of-the-
money call options: higher underlying stock volatility leads to much lower skewness in Panel
C. We observe essentially no relationship between volatility and skewness for in-the-money
put options in Panel D.

9
 -FIGURE 2 ABOUT HERE-

Figures 1 and 2 also shed light on the relationship between maturity and skewness. Skew-
ness generally decreases with maturity for out-of-the money options, but increases with matu-
rity for in-the-money options.

II. Empirical Results

A. Option Portfolio Formation

We obtain data for options written on common stock, including end-of-day closing bid and
ask quotes, underlying asset values, open interest, and trading volume, from the Ivy Option-
metrics database and create option portfolios on the first trading date of each month and on the
second Friday of each month, one week before options expire. Before creating our portfolios
we first screen out records that may contain errors and quotes that may not be tradable. This
procedure, detailed in the Appendix below, eliminates options from each portfolio using in-
formation observable on or before the corresponding formation date. For example, we screen
out options that do not trade on the formation date, options that have zero open interest on the
trading day immediately prior to the formation date, and options that have excessive bid-ask
spreads. Portfolio formation dates extend from February 1, 1996 through October 9, 2009.11
We also eliminate options that expire after December 2009.

For our analysis we also need the underlying asset value on each option’s expiration date.
We observe this value in Ivy for approximately 99.4% of our screened data. After filling in as
many missing values as possible using CRSP stock prices, we observe underlying asset values
on expiration dates for 99.9% of our observations. The other 0.01% are unobservable due to
events such as mergers and delistings. We also eliminate these few records from our data even
though this information is not directly observable on the portfolio formation dates.12

10
 Panel A of Table I presents the number of option quotes we observe on portfolio formation
dates for each year, as well as the number of quotes for which we can observe the underlying
asset value at expiration. In 2007, for example, we observe 194,822 option quotes across the
24 portfolio formation dates for the year, and the corresponding underlying asset values at
expiration for 194,402 of these quotes in the Ivy database. We are able to locate another 10
underlying asset values at expiration in CRSP, implying that we observe the corresponding
underlying asset values at expiration for 99.8% of the original quotes (194,412). In 2009 we
pull numerous underlying asset values from CRSP because our Ivy database ends October 30,
2009, implying that we must turn to CRSP to get underlying asset values at expiration for
options that expire at the end of 2009. The far right column in Panel A of Table I gives the
number of unique underlying assets in our data each year.

-TABLE I ABOUT HERE-

Each month we include options that fit into one of three expiration bins and then within
each expiration bin we rank options into ex-ante skewness quintiles. The first expiration bin
contains options that expire in one week. Since options expire on the third Friday of each
month, we form these bins on the second Friday of each month. We do not create portfolios
of any other maturity on these dates. The second and third expiration bins contain options that
on average expire in 18 and 48 days, respectively. We form these two bins on the first trading
date of the month of expiration and the month prior.13 We do not investigate the returns of
options with longer expirations due to low trading volume.

Next, on each portfolio formation date we sort options within each expiration bin into ex-
ante skewness quintiles. To estimate ex-ante skewness as illustrated by equations (1) through
(4), we need estimates of the expected return and volatility for every underlying asset and
formation date in our sample. We use six months of daily data immediately prior to each
formation date to estimate these moments.14 Other variables needed to compute the skewness
of the option include the underlying stock price, time to maturity, strike, and price of the option

11
on the formation date, all of which are readily obtained from the Ivy database. If on any given
date there are less than 10 options within an ex-ante skewness/expiration bin, we exclude this
bin from the analysis for this date.

Panel B of Table I illustrates how often we observe the same underlying asset across differ-
ent skewness bins. Among put options with seven days to maturity, for example, our sample
contains on average 230 options per ex-ante skewness bin and 590 unique underlying assets
per portfolio formation date. The same underlying asset is observed on average across 1.72
different skewness bins. The ex-ante screens described above prevent us, in many cases, from
observing options written on a broad cross section of strikes for the same underlying asset on
the same portfolio formation date. To verify that our results are not driven by differences in
stock characteristics, we create another set of option portfolios for which the set of underlying
assets is identical across ex-ante skewness bins and across contract type (call/put) for each
maturity bin and formation date. These results are reported in Section III.A.

Panel A of Table II reports the average, over time, of the median ex-ante skewness measure,
ski,t:T , across all options in each portfolio at each formation date. Within each expiration
group, skewness increases across the skewness quintiles by construction. The variation in
expected skewness across these quintiles is large, especially among short-term options. For
example, among call options that expire in seven days, expected skewness ranges from 0.40
to 24.94. In comparison, the typical skewness for a stock varies from around zero up to three
(see, for example, Boyer, Mitton, and Vorkink (2010)).

-TABLE II ABOUT HERE-

We next investigate the ability of our expected skewness measure to forecast subsequent
return skewness, and to do so, we adopt two approaches, a time-series test and a cross-sectional
test. For our first test we estimate the time-series skewness of each option portfolio. For
each ex-ante skewness/expiration bin, we calculate the equally weighted hold-to-expiration
returns for each formation date. This provides a time series of portfolio returns that we use to

12
estimate a skewness estimate. We report these estimates in Panel B of Table II. The bottom
two rows in this panel test for differences in the skewness estimates across the low and high
ex-ante skewness portfolios. We compute the standard error for this difference by GMM using
the approach of Newey and West (1987) to account for both cross-sectional and time-series
dependence. In this panel, we observe that portfolio skewness increases monotonically across
ex-ante skewness bins for a given maturity bin. This relationship is consistent for both call
and put options and supports our ex-ante skewness measure as a strong predictor of subsequent
return skewness. While the differences between the high and low ex-ante skewness bins are
statistically significant, the magnitudes are muted relative to the skewness results of Panel
A. This is likely the result of two effects. First, by forming portfolios, we diversify away
much of the idiosyncratic skewness of the individual options. In addition, our time series is
relatively short and given that skewness emphasizes small probability events, our estimates
may be inefficient, especially for out-of-the-money options.

For our second test, we estimate skewness in the cross section, following Zhang (2005).
Given that there are many more options than time periods, we should be better able to capture
small probability events that relate to skewness in the cross section than using time-series esti-
mates. This approach relies on the observation that the average cross-sectional skewness for a
given bin is positively related to the average time-series idiosyncratic skewness of the options
in that bin. In Panel C of Table II we report the time-series average of the cross-sectional
skewness estimates within each skewness/expiration bin. These results provide additional
evidence that our expected skewness measure, ski,t:T , does a good job as a forecast. The aver-
age cross-sectional skewness increases across the skewness quintiles for each maturity group.
Rows labeled “(t-stat)” in this panel test for a significant difference in average cross-sectional
skewness across the bottom and top skewness quintiles. We derive these t-statistics using the
approach of Newey and West (1987).

We examine the empirical relationship between each input and our ex-ante skewness mea-
sure in the Internet Appendix. In cross-sectional regressions, holding maturity fixed, we find

13
that about 92% of the variation in ex-ante skewness for our sample can be explained by mon-
eyness alone. After controlling for moneyness, the next most important input is underlying
asset volatility. Accounting for both moneyness and volatility enables us to explain 97% of
the cross-sectional variation in ex-ante skewness. While moneyness plays an important role
in characterizing an option’s ex-ante skewness, other option characteristics are also important
and we make use of variation in ex-ante skewness unrelated to moneyness in our tests later in
the paper.

B. Option Characteristics and Portfolio Returns

In Table III we report summary statistics on liquidity for each ex-ante skewness/expiration
bin. In Panel A of Table III, we report the average bid-ask spread, where we define the bid-ask
spread as the difference between the closing bid and ask prices on portfolio formation dates
scaled by their midpoint. We find average option bid-ask spreads to be large and monoton-
ically increasing with our measure of ex-ante skewness holding maturity fixed. For options
with the shortest maturity, the average bid-ask spread for the low skewness quintile is about
7%, and for the high skewness quintile about 75% to 100%. Such wide spreads make it very
difficult for outside investors to arbitrage away the overvaluations we document in this pa-
per. As is standard when computing option returns, our proxy for “true” option prices is the
midpoint of the bid-ask spread. Given the particularly large bid-ask spreads of option con-
tracts, this proxy may be subject to considerable measurement error, especially for options
that are highly skewed. However, as Blume and Stambaugh (1983) show, this measurement
error induces an upward bias in computed returns because of Jensen’s inequality. Hence, if
anything, our estimated expected returns are too high, especially for options that are highly
skewed. Moreover, our use of long-horizon returns (hold-to-expiration) rather than daily re-
turns should attenuate the effects of measurement error in our study.

-TABLE III ABOUT HERE-

14
 In Panel B of Table III, we report summary statistics on average daily volume per contract,
and in Panel C we report the average total dollar volume for each ex-ante skewness quintile. In
general, we observe that high ex-ante skewness options have more trading volume relative to
lower ex-ante skewness options. For example, among call options with seven days to maturity
in the low (high) skewness quintiles, we observe an average of 182 (319) contracts traded per
security per day, where each contract is for delivery of 100 shares of stock. Hence, options
with high skewness are actively traded, despite their comparatively high bid-ask spreads. Av-
erage daily total dollar volume, as reported in Panel C, across call options in the low (high)
skewness bins is on the order of $52 million ($2.8 million). Unlike contract volume, dollar
volume is decreasing in ex-ante skewness because options with high skewness are less expen-
sive. However, dollar volume for options with high ex-ante skewness is still relatively high
and comparable to that of the smallest size decile of stocks that trade on the NYSE. In the In-
ternet Appendix we investigate open interest and find patterns very similar to those of Panels
B and C of Table III.

We report time-series averages of portfolio returns for each ex-ante skewness/maturity bin
in Table IV. In each case returns are scaled to be weekly. This table provides initial evidence
that skewness preference influences prices in equilibrium. The returns decrease dramatically
across skewness bins for every maturity group, especially among short-term options where
we observe the most cross-sectional variation in ex-ante skewness (Panel A of Table II). For
example, among call options that expire in seven days, the average weekly return is monoton-
ically decreasing from 1.87% for the low skewness bin to -35.25% for the high skewness bin.
The paired t-statistic for the difference is 7.56. These results are within the range of average
returns reported by Ni (2009). We find even more dramatic results for put options. Among put
options that expire in seven days, returns are again monotonically decreasing, this time from
-5.38% for the low skewness bin to -59.98% for the high skewness bin. The paired t-statistic
for this difference is 14.13. Standard errors for Table IV are calculated using the approach of
Newey and West (1987), in part to account for overlapping observations of options with 48

15
days to maturity. While the average difference in spreads between the low and high skewness
portfolios decreases as maturity increases, it is positive and significant in all cases.

-TABLE IV ABOUT HERE-

Our returns ignore the possibility of early exercise. This simplification should have little
impact, however, on our relative results. Ignoring the possibility of early exercise biases
downward the returns of options that become optimal to exercise early. The likelihood of
optimal exercise increases with moneyness. But options that are in-the-money tend to be less
skewed as discussed in Section I. Therefore, ignoring early exercise should, if anything, tend
to bias downward the returns of in-the-money, less skewed options. The point of our paper
is to show that such options earn higher risk-adjusted returns than out-of-the-money skewed
options. In the Internet Appendix we adjust our returns for the possibility of early exercise
similar to Pool, Stoll, and Whaley (2008) and show that doing so has little impact on our
results.

In summary, the magnitude in the return differential between high and low skewed options
is remarkable for short-dated options. These results indicate that individual equity option
investors give up average returns on the order of 50% weekly for exposure to the lottery
opportunities that options with high ex-ante skewness offer. Below, we investigate what other
factors influence this relationship and determine that most of this return differential can be
attributed to skewness preference.

III. Controls for Risk and Other Characteristics

The traditional asset pricing framework maintains that assets earn low expected returns
only if they pay off high in states when the marginal utility of investors is high. In this section
we try to explain the incredibly low average returns of Table IV by the return comovement

16
with state variables commonly used in the asset pricing literature. In addition to a simula-
tion exercise, we also explore the ability of various firm- and option-level characteristics to
explain the low average returns. First, we estimate the pricing errors of linear factor models.
Second, we run two simulations, one under the Black-Scholes (1973) paradigm and another
under the jump diffusion model of Merton (1976) to test the robustness of our linear pricing
model results to assumptions underlying the empirics. Third, we estimate Fama-MacBeth
(1973) cross-sectional regressions that allow us to simultaneously control for a variety of fac-
tor loadings and option characteristics. Fourth, we use a double-sorting exercise that allows
for a non-linear relationship between returns and characteristics. All methods yield similar
results. We find a negative significant relationship (both economically and statistically) be-
tween ex-ante skewness and average option returns after controlling for both comovement and
characteristics.

We feature results that control for coskewness as a test of the prevailing view that only
coskewness is priced. Recent research suggests that total skewness is priced in equities (see
footnote 4). Option markets, however, offer a richer environment to explore the extent and
manner in which skewness preference influences asset prices given the enormous skewness
accessible through option markets and the ease, relative to stocks, of determining which op-
tions offer the greatest lottery-like payoffs.

We also feature results after controlling for moneyness to separate our findings from the
extensive literature on the relation between moneyness and option prices. The discussion in
Section I.B and Figures 1 and 2 show that out-of-the-money call and put options have partic-
ularly high ex-ante skewness. However, a well-known stylized fact already established in the
options pricing literature is that out-of-the-money options are overvalued relative to standard
theoretical models. More recently, Ni (2009) shows that out-of-the-money call options, in
particular, earn low average returns. In this paper we show that ex-ante skewness unrelated
to moneyness is priced. In contrast, after controlling for our measure of ex-ante skewness,
we find that the relationship between option returns and moneyness documented by Ni (2009)

17
largely disappears. Hence, investor demand for lottery-like assets may help explain the well-
documented relation between moneyness and prices.

A. Pricing Errors of Linear Factor Models

Can the low option returns documented in Table IV be explained by their comovement
with state variables? Here we explore this question by investigating the pricing errors of linear
factor models. Specifically, we regress the excess return of each ex-ante skewness portfolio
on the excess returns of zero-investment portfolios over the same time period,

r p,t:T − r f ,t:T = α p + ∑ βi,p fi,t:T + e p,t , (5)

i

where r p,t:T , r f ,t:T , and fi,t:T represent the net returns on the option portfolio, the risk-free
asset, and zero-investment portfolio i from t to T , respectively.

We begin by considering a single state variable, namely, the excess market return. We
examine the ability of this single state variable to explain both the returns of the option port-
folios of Table IV and the underlying assets of these options. We report the intercepts from
estimating this one-factor model for option portfolios in Panel A of Table V.15 Perhaps not
surprisingly, these intercepts, or CAPM alphas, look similar to the average returns in Table IV.
CAPM alphas are generally decreasing across skewness quintiles and spreads in CAPM alpha
across the low and high ex-ante skewness quintiles are quite significant, both economically
and statistically, with t-statistics in the range of four to 12. We calculate standard errors for
Table V via GMM using the approach of Newey and West (1987), in part to account for the
overlapping observations of options with 48 days to maturity. In the Internet Appendix we
report CAPM alphas using each portfolio’s instantaneous beta and find similar results.16

-TABLE V ABOUT HERE-

18
 In Panel B of Table V we report the CAPM alphas for the portfolios of underlying stocks
for the options in each ex-ante skewness bin. We compute returns for each underlying asset
across the same holding periods as our option returns, and take the equally weighted average
across stocks in a given maturity/ex-ante skewness bin to get stock portfolio returns for each
date. We then regress excess stock portfolio returns on excess market returns over time as
in equation (5) and report the intercepts of these regressions. Remarkably, in contrast to the
results given in Panel A, the CAPM alphas of the underlying stocks of short-dated options
are generally insignificantly different from zero, and do not vary systematically across the ex-
ante skewness bins. In particular, the CAPM alpha spread across high and low ex-ante option
skewness quintiles is insignificantly different from zero for the seven-day and 18-day maturity
bins. Short-maturity options written on individual stocks therefore appear to be nonredundant
securities from the simple perspective of the CAPM.

For options with 48 days to maturity the underlying CAPM alpha spreads are significant
in Table V. In the Internet Appendix we show that these significant spreads are caused by
momentum. After controlling for the past six-month return using a double-sort procedure,
virtually all CAPM alphas and all spreads in CAPM alpha for the underlying stock portfolios
become insignificantly different from zero. For short-term option portfolios, however, we still
find a significant cross-sectional relationship between returns and ex-ante skewness even after
controlling for momentum (see Section III.D). Hence, while underlying asset characteristics
appear capable of explaining the average returns for longer-term options, they cannot explain
the average returns of short-term options.

To further illustrate that the CAPM mispricing we document for short-term options in
Table V is unrelated to the characteristics of the underlying assets, we repeat the analysis
of Table V using option portfolios for which the set of underlying assets is identical across
ex-ante skewness bins and contract type (call/put) for options with the same maturity. To
maximize sample size for this investigation, we focus only on options in the high and low
ex-ante skewness quintiles.

19
 On a given formation date we classify options with a given maturity into one of four
groups: call in high quintile (CH), call in low quintile (CL), put in high quintile (PH), and
put in low quintile (PL). For each underlying asset, we count the number of option contracts
within each of these four groups and then identify the group with the least number of contracts.
For example, on May 1, 2001 our sample contains 14 options written on Applied Materials
Inc. that mature on May 18 of the same year: four calls in the low quintile, three calls in
the high quintile, two puts in the low quintile, and five puts in the high quintile. The group
with the least number of contracts, therefore, is PL, which has two contracts. We then force
the number of option contracts written on a given underlying asset to be the same across all
four groups. For the example using Applied Materials Inc. this involves eliminating options
from CH, CL, and PH until all groups have only two option contracts, the same as PL. We
select option contracts to eliminate from each group with independent uniform probability. By
conducting this exercise for every portfolio formation date we develop a sample in which the
set of underlying assets is identical across ex-ante skewness bins and contract type (call/put)
for options of the same maturity.17

We report the results of this exercise in Table VI. Here we see that the CAPM alpha spreads
for call and put options, given in Panels A and B, respectively, are quite similar to those
reported in Panel A of Table V. In Panel B of Table VI we report the CAPM alphas of the
underlying stocks for each bin. For every maturity, not only is the CAPM alpha spread across
the high and low ex-ante skewness bins zero by construction, but the CAPM alphas for each
stock portfolio are also all insignificantly different from zero. These results provide clear
evidence that the large differences in CAPM alpha documented in Table V are not driven by
differences in underlying stock characteristics, including differences in risk or distributional
properties.

-TABLE VI ABOUT HERE-

20
 While the results of Tables V and VI indicate that shorter-term individual stock options
are not redundant, these options may still earn low average returns because they covary with
other state variables orthogonal to the market return. Other work suggests that the relevant
state variable is nonlinear in the market return and that coskewness is the appropriate pricing
measure (see Kraus and Litzenberger (1976) and Harvey and Siddique (2000)). This view
stands in contrast to recent models, discussed in the introduction, predicting that total skew-
ness is priced. In addition, index options are often considered nonredundant because they
insure the holder against changes in systematic volatility and/or systematic jumps (see, for
example, Bakshi and Kapadia (2003a and 2003b), Pan (2002), and Duarte and Jones (2007)).
We analyze options on individual stocks for which much of the variation is idiosyncratic. Fur-
ther, we find strong results among call options, which do not insure against downward jumps.
Hence, models with volatility or jump risk premia are unlikely to explain our findings. Nev-
ertheless, we now examine whether coskewness or volatility risk can explain the low average
returns documented in Table IV.

To investigate the ability of coskewness and volatility risk to explain our results, we es-
timate the intercepts of a linear pricing model with three state variables: the excess market
return, the zero-investment coskewness portfolio of Harvey and Siddique (2000), and the ex-
cess return on a zero-delta S&P 500 index straddle. Harvey and Siddique (2000) find that
their zero-investment coskewness portfolio is priced in the cross section of equity returns and
considerably reduces the time-series pricing errors of standard linear factor models. Coval and
Shumway (2001) find that zero-delta straddles earn significantly negative returns and Ang et
al. (2006) find that excess straddle returns represent a priced risk factor in the cross section of
equities.

To construct returns on the zero-investment coskewness portfolio, we use five years of

monthly data to estimate coskewness for every common stock in the CRSP universe as defined
in equation (11) of Harvey and Siddique (2000). At the beginning of each month we then
rank stocks based on their past historical coskewness and use the 30% with the most negative

21
coskewness to create a value-weighted portfolio. We then compound the subsequent daily
returns of this portfolio over the appropriate option holding period (from t to T ) and subtract
the corresponding risk-free return to construct factor realizations.

To construct zero-delta straddle returns we choose the closest-to-the-money put and call
options on the S&P 500 each day in our sample and form long positions in each such that the
delta of the portfolio is zero. We then compound daily straddle returns over the appropriate
option holding period (from t to T ) and subtract the corresponding risk-free return to construct
factor realizations.

We report intercept estimates for the three-factor model in Table VII. While the pricing
errors reported in this table and their spreads are somewhat smaller in absolute terms relative
to the CAPM intercepts of Table V, the results indicate that the low average option returns
documented in Table IV cannot be explained by either coskewness or volatility risk. For
options that mature in seven days, the pricing error spreads in Table VII are still 35.31% for
call options and 49.88% for put options with t-statistics of 7.45 and 10.28. Total skewness
appears to be priced in the individual equity options market.

-TABLE VII ABOUT HERE-

In summary we do not find evidence that the low average option returns documented in
Table IV can be explained by their comovement with common state variables. We further
investigate the ability of market risk, volatility risk, and coskewness to explain the returns of
option portfolios sorted on ex-ante skewness in Fama-MacBeth regressions and double sorts
below. These methods also enable us to investigate the influence of various characteristics on
option returns.

22
B. Simulation Exercise

We next investigate whether more robust models of stock return dynamics can generate the
patterns in individual stock options we observe in the data. Broadie, Chernov, and Johannes
(2009) discuss some of the concerns regarding the empirical analysis of option returns and
allege that standard methods to compute pricing errors for options can be misleading. We ad-
dress four of the prominent concerns to using standard empirical analysis of option returns in
our simulation exercise: nonnormality, nonlinearity, nonadditivity, and the tendency of option
returns to suffer from peso problems in finite samples. First, option returns deviate substan-
tially from normality and the small-sample distributions of standard CAPM alpha estimates
may not conform with asymptotic inference. Second, the nonlinear relation between option
and stock returns is likely to cause e p,t in the regression given by equation (5) in the paper
to be correlated with the pricing factors ( fi,t:T ), thereby causing OLS estimates of alpha to be
biased and inconsistent. Third, because returns are non-additive, expectations and betas do
not scale linearly with time.18 This implies, for example, that if stock prices follow a geo-
metric Brownian motion and Merton’s (1971) continuous-time CAPM holds, then the CAPM
cannot hold over discrete horizons. Estimates of alpha may therefore be unduly influenced
by the particular horizon over which we choose to measure returns. Finally, finite samples
of option returns may lead to peso problems; they may lack important certain extreme rare
events correctly anticipated by option investors ex-ante but not reflected in our estimates of
alpha measured ex-post.

Following Broadie, Chernov, and Johannes (2009) we run two simulations, one under
the Black-Scholes (1973) paradigm and another under the jump diffusion model of Merton
(1976). We simulate under the Black-Scholes assumptions as a base case. We simulate under
the Merton (1976) jump-diffusion assumptions as this model can generate more flexible dis-
tributions in the underlying returns than the log-normal assumption. In addition, the Merton
jump diffusion model is a relatively tractable model that can be applied to a variety of options
across a large cross section of underlyings at each point in time.

23
 We calibrate both simulations to match the moments and size of our original data and in
both simulations impose the null that skewness is not priced. These simulations allow us to
determine how often we might expect to observe results as extreme as those we find in the
original data using the same empirical methods because of either statistical issues or peso
problems. We find that the answer is virtually never. The results we estimate using the actual
data look quite abnormal relative to both of the models we simulate. In fact, our simulation
exercise leads us to the same conclusions that we arrive at using standard empirical methods.
This consensus may be attributable to the fact that we simulate portfolios of options written
on a broad cross section of stocks rather than on a single underlying index. Likewise, Broadie,
Chernov, and Johannes (2009) find that concerns surrounding the extreme statistical nature
of option returns can be largely alleviated by analyzing option portfolios. Given that our
simulation exercise does not alter our interpretation of results, for sake of brevity we report all
details and results of our simulation exercise in the Internet Appendix.

C. Fama-MacBeth

To further assess the influence of ex-ante skewness on option returns, we conduct cross-
sectional regressions following the approach of Fama and MacBeth (1973). We find a strong
economic and statistical relation between average returns and ex-ante skewness in the cross
section that persists even after including in these regressions portfolio betas and other option
characteristics that may influence expected returns.

On each portfolio formation date we sort each option within a given maturity bin into one
of 100 bins based on ex-ante skewness and then calculate the equally weighted portfolio return
for each ex-ante skewness bin.19 For each date t we then estimate the following cross-sectional
regression across portfolios with the same maturity horizon:

skew
r p,t:T = γ0,t + γ1,t R p,t + φt0 Z p,t + ε p,t , (6)

24
where r p,t:T is the equally weighted net return for portfolio p observed over the horizon from
skew is the ex-ante skewness portfolio rank for portfolio p (from one to 100), and Z
t to T , R p,t p,t

is a vector of control characteristics and factor loadings for portfolio p.

skew instead of average portfolio skewness on the right side of our regressions
We use R p,t
because the cross-sectional distribution of average portfolio skewness is itself quite positively
skewed with an extremely high standard deviation that complicates the economic interpre-
tation of our results (see footnote 22). For consistency, we use average ranks of the other
characteristics as control variables in Z p,t , although our results are robust to the use of ranks
for our controls. We calculate these average characteristic ranks by independently sorting
each option on a given portfolio formation date with a given maturity into 100 bins based on
each control characteristic. We then average the control characteristic rank across all options
within each of the 100 skewness bins. The control characteristics we consider include money-
ness (Xi /Si,t ), volume (total contracts traded on the portfolio formation date), bid-ask spread
(scaled by the midpoint), and volatility smirk as in Xing, Zhang, and Zhao (2010).20 We also
include each portfolio’s market beta, volatility risk beta, momentum beta, and coskewness
beta. These factor loadings are all estimated by regression as in equation (5) using the market
excess return, the excess return on a zero-delta S&P 500 straddle, the standard momentum
factor obtained from Ken French’s website, and the squared market excess return as right-side
variables, respectively.21

We include moneyness in our regressions to separate our findings from those of Ni (2009).
We include volume and the bid-ask spread in our regressions to investigate the influence of
liquidity on our pricing results. We include the volatility smirk to control for asymmetries
in the return distribution of the underlying stock. The volatility smirk is unique among the
characteristics we consider in the Fama-MacBeth regressions in that it is the same for all
options with the same underlying and maturity. We include the momentum beta in our analysis
given that momentum helps to explain the underlying asset returns (see the Internet Appendix).
We include the volatility risk beta to investigate whether risk factors unique to index option

25
markets can explain our results. We include the coskewness beta to verify whether our results
can be explained by traditional models of coskewness or if they are more consistent with new
models that suggest total skewness is priced.

In Table VIII, we report the time-series averages of the γ and φ coefficients from equation
(6), along with Newey-West (1987) t-statistics. Panel A reports results for call options while
Panel B reports results for put options. In the interest of brevity, we report results only for
options that expire in 18 days. The top row in each panel of Table VIII reports the cross-
sectional pricing of ex-ante skewness. The coefficient on expected skewness is negative and
highly significant in all cases. For example, in column 10 of Panel A, we report results in-
cluding all controls we consider, and we see that the average coefficient on ex-ante skewness
is -0.411 with a t-statistic of -4.07. This implies that increasing the ex-ante skewness of a
portfolio of call options from the top of the bottom quintile to the bottom of the top quintile
is associated with a 25% decline in average weekly returns (−0.411 × 60), comparable to the
results of Table IV.22 In column 10 of Panel B the average coefficient on ex-ante skewness for
put options is -0.437 with a t-statistic of -3.46.

-TABLE VIII ABOUT HERE-

The results of Table VIII indicate that a relationship exists between ex-ante skewness and
expected option returns unrelated to loadings on the four risk factors and the other portfolio
characteristics we consider. Of particular note, a relationship between ex-ante skewness and
expected option returns exists that is unrelated to moneyness. Used in isolation without other
controls, the average coefficient on moneyness is negative for call options and positive for put
options and highly significant, indicating that out-of-the-money options earn low average re-
turns. When ex-ante skewness is included in the regressions, however, the sign on moneyness
flips, suggestive of the high positive correlation between these two variables. That the sign on
ex-ante skewness remains the same when moneyness is included indicates that the negative as-
sociation between average option returns and our ex-ante skewness measure is not subsumed

26
by any option characteristic related to moneyness. Skewness unrelated to moneyness appears
to be priced.

To further explore this issue, we conduct the same analysis using the average returns from
moneyness-sorted portfolios as left-side variables in our cross-sectional regressions. As right-
side variables we use the average ex-ante skewness rank across these portfolios and the actual
moneyness rank of each portfolio. In isolation, the average coefficients on each of these
variables are again highly significant and of the expected sign. However, when used together,
the average coefficients on both variables lose their significance. That is, in skewness-sorted
portfolios moneyness cannot trump our ex-ante skewness measure, but in moneyness-sorted
portfolios skewness causes moneyness to lose its significance. Again, this exercise illustrates
the ability of our ex-ante skewness measure to explain the cross section of option returns above
and beyond moneyness as documented by Ni (2009).

Table VIII also provides evidence that the low average option returns we document cannot
be easily explained by option coskewness. The results are more consistent with new total
skewness models. Used in isolation without other controls, the coskewness beta is negative
and significant for call options, consistent with theory, but insignificant for put options. When
ex-ante skewness is included in the regressions, the coskewness beta is negative for both call
and put options, but is significant at the 10% level only for put options. The coefficient on
total ex-ante skewness, however, remains negative and significant, again suggesting that total
skewness is priced.

D. Double Sorts

Fama-MacBeth (1973) regressions impose linearity on the structure between ex-ante skew-
ness, returns, and characteristics. In this section we control for the influence of characteristics
using double-sorted portfolios. Following the methodology in Ang et al. (2006, 2008), on
each portfolio formation date we first sort options of a given maturity into deciles based on

27
some characteristic. Then within each characteristic-sorted decile, we rank options into two
ex-ante skewness bins. We next average the returns across options with the same ex-ante
skewness rank across all characteristic deciles, thereby creating the returns for two (equally
weighted) portfolios similar in terms of the given characteristic but different in terms of their
ex-ante skewness.23 After creating two such portfolios for each formation date in our sample,
we estimate and compare their CAPM alphas. We choose to examine CAPM alphas given
the considerable discrepancy we find in such pricing errors for options versus their underly-
ing assets (see Tables V and VI). This approach also allows us to easily report results across
different option maturities.

We conduct this exercise using moneyness (strike price scaled by underlying asset price)
and a measure of ex-ante coskewess (similar to our measure of ex-ante skewness) as control
variables. We report results controlling for moneyness in Panel A of Table IX, where we see
that the difference in CAPM alphas remains large and statistically significant. For call options
with seven days to maturity, the CAPM alpha spread is 8.81% per week (t-statistic of 4.93)
while for the analagous put options, the spread is 9.70% per week (t-statistic of 4.05). In
Panel B of Table IX, we report the results of an exercise where we reverse the order of sorting.
We first sort into deciles based on our ex-ante skewness measure and then within each decile,
rank options into two moneyness bins. We then average the returns for options with the same
moneyness rank across all ex-ante skewness deciles, thereby creating two portfolios similar
in terms of ex-ante skewness but different in terms of their moneyness. After controlling for
ex-ante skewness we find the spread in CAPM alphas across out-of-the-money options and
in-the-money options to be quite small. In fact, for most maturity bins the differences are
not statistically significant.24 These findings again indicate that our results are not entirely
subsumed by an explanation related to option moneyness. In particular, after controlling for
our measure of ex-ante skewness, the relationship between moneyness and expected returns
documented by Ni (2009) virtually disappears.

-TABLE IX ABOUT HERE-

28
 We report results controlling for ex-ante coskewness in Panel A of Table X. We estimate
ex-ante coskewness at the individual option level under the assumption that stock returns are
lognormal. Using the results of Lien (1985) we obtain an ex-ante estimate of the coskewness
measure of Harvey and Siddique (2000) for each option, denoted as βskd , as follows:

Et [εi,t:T ε2M,t:T ]
βskd
i,t:T =r h i h i, (7)
2
Et εi,t:T Et εM,t:T 2

where t and T represent the portfolio formation date and expiration date of the option, εi,t:T
is the component of the option return (from t to T ) orthogonal to the market return, and εM,T
is the deviation of the market return (over t to T ) from its expected value. We estimate the
ex-ante moments of these residuals assuming the option’s underlying stock return and the
market return are jointly lognormal in the same manner that we estimate ex-ante skewness.
Our measure of ex-ante coskewness is a function of moneyness, maturity, option price, stock
price, and the first two moments of stock and market returns including the covariance. We
provide additional details in the Appendix below.

-TABLE X ABOUT HERE-

In Table X we see that controlling for ex-ante coskewness has a relatively small impact on
our results. In Panel A the CAPM alpha spread for seven-day call options is 20.14% per week
(t-statistic of 8.79) while for seven-day put options the spread is as high as 29.61% per week
(t-statistic of 10.68). Hence, CAPM alpha spreads are somewhat smaller than those in Table
V where we sort unconditionally on ex-ante skewness. The results of Table X, however, are
not consistent with models that suggest only coskewness is priced.

We again conduct a conditional sort exercise where we reverse the order of sorting and
report the CAPM alphas of these conditionally sorted portfolios in Panel B of Table X. After
controlling for ex-ante skewness we find that alpha spreads across ex-ante coskewness port-
folios are generally insignificant and small compared to those of Panel A. If anything, short

29
maturity call options with lower coskewness earn lower returns, holding total skewness fixed,
contrary to the traditional co-skewness pricing models.25 The results of Table X not only pro-
vide empirical support for new models that suggest total skewness is priced, but also further
highlight the unique ability of our ex-ante skewness measure to explain the cross section of
option returns since our measure of ex-ante coskewness is a function of the same inputs as our
measure of ex-ante skewness.

In the Internet Appendix we also conduct this double-sort exercise using as control vari-
ables each of the four characteristics considered in our Fama-MacBeth regressions (money-
ness, volume, bid-ask spread, and volatility smirk), the past six-month underlying stock return,
option vega, and each of the inputs into our measure of ex-ante skewness (stock price, stock
volatility, option price, and stock expected return). We control for the past six-month underly-
ing stock return as a control for momentum. We use vega as a proxy for an option’s sensitivity
to volatility risk. We also individually control for each input into our measure of ex-ante skew-
ness to verify that our pricing results are not unduly driven by a single input that may proxy
for some other priced risk or characteristic. Our findings remain after controlling for each of
these other variables. The low average returns we document cannot be easily explained by
liquidity effects, volatility smirk, momentum, volatility risk, or any single input to our ex-ante
skewness measure.

IV. CAPM Alphas From Writing Options

The returns we calculate to produce all results thus far use the midpoint of closing bid and
ask prices as the proxy for the “true” price. In Table III we show that bid-ask spreads are
monotonically increasing in skewness at all horizons. The return spreads we document across
ex-ante skewness quintiles, therefore, are even larger for investors who buy options near the
ask.

30
 In Table XI we report the CAPM alphas earned across our sample from writing options at
the bid price. We obtain these alphas from estimating the one-factor model given in equation
(5) using option portfolio returns computed at bid prices. Table XI shows that investors who
write options with high ex-ante skewness generally earn CAPM alphas that are insignificantly
different from zero and sometimes negative. The premiums investors pay to buy options with
high ex-ante skewness are not passed on to investors who write options. The only high ex-
ante skewness portfolio with positive CAPM alpha is put options that expire in seven days.
Further, differences in CAPM alphas across the high and low skewness quintiles are generally
insignificant. The only exceptions are for call options that expire in 18 days, in which case the
CAPM alpha in the high ex-ante skewness portfolio is significantly lower (Low minus High
equals 8.84), and for put options that expire in seven days (Low minus High equals -25.75).

-TABLE XI ABOUT HERE-

To the extent that bid-ask spreads represent the cost of holding inventory that cannot be
perfectly hedged, the results of Table XI suggest that intermediaries are better able to hedge
their long positions in lottery-like options than short positions, since estimated CAPM alphas
using bid prices for options with high ex-ante skewness are more closely aligned with the zero
CAPM alphas of the underlying assets. Further, the ability of dealers to hedge long positions
does not vary with the ex-ante skewness properties of the option. On the other hand, results
reported previously indicate that dealers cannot easily hedge the risk of incurring extreme
losses on their short option positions with high ex-ante skewness, and demand compensation
to bear such risk.

To further verify that our results indeed provide evidence for skewness preference, and to
investigate the extent to which investors actually transact at or near the ask price, we analyze a
small set of transaction data gathered from Bloomberg, which provides trade and quote data on
every option transaction on every major U.S. exchange in the recent past. Using our screened
data, we begin by choosing 30 underlying stocks at random among the top quintile of all stocks

31
ranked by total option trading volume in 2009. We then gather trade and quotes for options
with three different maturities (seven days, 18 days, and 48 days) during the months of April,
May, and June of 2011. Specifically, on the first trading date of these three months, we pull
all transaction data for all options that expire within the same month (options with 18 days to
expiration) or expire the following month (options with 48 days to expiration). We also pull
all transaction data on the second Friday of each month for all options that expire that month
(options with seven days to expiration). We gather the option price for each transaction, as
well as the most recent best bid and ask prices quoted at least one second prior to the option
transaction time stamp. We then limit our data to the top quintile of call options and the top
quintile of put options ranked by ex-ante skewness on each transaction date. This procedure
leaves us with data on 5,992 transactions.

We then calculate the position of each transaction price relative to the best bid and ask
price as
trade − bid
π= (8)
ask − bid
and plot histograms of π in Figure 3. Here we see that options in our sample generally trade
right at the ask or bid, with approximately half trading right at the ask. For example, among
call options that expire in seven days 50% of all trades occur at the ask, and among put options
that expire in seven days 47% of all trades occur at the ask. Hence, many investors appear
willing to buy options with high ex-ante skewness at or near the ask price despite the high
premiums they must pay to intermediaries.

-FIGURE 3 ABOUT HERE-

V. Conclusion

We find evidence suggesting that the impact of skewness preference on prices and subse-
quent returns of individual equity options is large. Our results also lend support to the view

32
that total skewness is more relevant to the pricing of options than coskewness. Investors are
willing to compromise as much as 50% per week in order to gain exposure to options with
the greatest lottery potential. In comparison, recent papers on the effect of skewness in equi-
ties markets find that lottery-preferring investors are willing to lose on average around 12%
per year to hold stocks offering the greatest lottery potential among equities. The differen-
tial impact of skewness preference on equity and options markets is likely driven by at least
two factors. First, option markets offer skewness opportunities that are multiples of those of-
fered in the equity market. Second, the precise lottery features that attract skewness-preferring
investors to these securities increase the risk that short-sellers, whose positions cannot be com-
pletely hedged, will be wiped out, and thus act as a limit-to-arbitrage. Our results suggest that
attention to the lottery prospects of securities may deepen our understanding of investor pref-
erences and asset prices in equilibrium. More research is needed to understand how lottery
prospects, skewness-preferring investors, and the incentives to arbitrage skewness (sell lottery
tickets) by smart or institutional money interact.

33
 Appendix

A. Ex-ante Skewness and Coskewness

In this appendix, we demonstrate how our ex-ante skewness and coskewness measures,
ski,t:T and βskd
i,t:T , are constructed assuming lognormal stock prices. We make use of Lien’s

(1985) theorem regarding truncated lognormal distributions. We begin by restating Lien’s

(1985) Theorem 1.

0 0
THEOREM  1: Let (u
1 , u2 ) be a normal random vector with mean (0, 0) and covariance
σ21 σ12
matrix =  . Then
2
σ12 σ2

 
h−a exp [−D/2Q]
E(exp(ru1 + su2 ) | u1 > a) = N ,
σ1 N( −a
σ1 )

where h = rσ21 + sσ12 , D = −Q r2 σ21 + 2rsσ12 + s2 σ22 , Q = σ22 σ21 − σ212 , and N(·) is the CDF

of the normal.

A.1. Ex-ante Skewness

In this section we show how we construct of ex-ante skewness measure ski,t:T . To begin,
we first note that Lien’s (1985) Theorem 1 can be used to derive closed-form solutions for the
raw moments of option returns given by equation (4). These raw moments can be substituted
into equation (2) to construct ski,t:T . We walk through the solution of equation (4) for the case
when j = 1. Solving the cases when j = 2 or j = 3 simply involves more algebra. For j = 1,
equation (4) can be written as
     
c St ST ST X X ST X
E [Rt:T ]= E | > − P > , (A1)
Ct St St St Ct St St

34
where St is the value of the underlying asset at time t < T and we suppress the subscript i for
clarity. Let r̃ = ln(ST /St ), the log stock return, and define A as A = ln(X/St ). Then equation
(A1) can be written as
 
c St r̃

X
E [Rt:T ]= E e |r̃ > A − P(r̃ > A). (A2)
Ct Ct

Now assume that r̃ is distributed N(µ̃, σ̃2 ), where in general µ̃ can be nonzero. Under this
assumption, the stock return, ST /St , is lognormal. Further, define z = r̃ − µ̃, so that z is dis-
tributed N(0, σ̃2 ). Then note that

E er̃ |r̃ > A = E ez+µ̃ |z > A − µ̃

(A3)

= eµ̃ E (ez |z > A − µ̃) . (A4)

A direct application of Lien’s (1985) theorem implies that equation (A4) can be written as
h i
σ̃2
exp µ̃ + 2 N(d¯1 )
E er̃ |r̃ > A =


, (A5)
N(d¯2 )

σ̃2 +ln( SXt )+µ̃

where d¯1 = σ̃ and d¯2 = d¯1 − σ̃. Note that P(r̃ > A) = N(d¯2 ). We can then plug
equation (A5) into equation (A2) to get the first moment of the call option return,

σ̃2
 
St X
c
E [Rt:T ]= exp µ̃ + N(d¯1 ) − N(d¯2 ). (A6)
Ct 2 Ct

Following this same approach, we calculate the raw holding-period call return moments
h i h i
c )2 and E (Rc )3 ,
E (Rt:T t:T

35
 h 2 i
St2 exp 2σ̃2 + 2µ N d¯3 − 2XSt exp σ̃2 + µ N d¯1
 


h i
c 2
E (Rt:T ) = + (A7)
Ct2
X 2 N d¯2

Ct2
3 exp 9 σ̃2 + 3µ̃ N d¯ − 3XS2 exp 2σ̃2 + 2µ̃ N d¯
 
 

h i S 4 3
c 3
E (Rt:T ) = t 2 t
+ (A8)
Ct3
h 2 i
3X 2 St exp σ̃2 + µ̃ N d¯1 − X 3 N d¯2



,
Ct3

where d¯3 = d¯1 + σ̃, and d¯4 = d¯1 + 2σ̃.

The corresponding raw moments for put options are

h 2 i
XN −d¯2 − St exp σ̃2 + µ̃ N −d¯1



 p 
E Rt:T = (A9)
Pt
h 2 i
X 2 N −d¯2 − 2XSt exp σ̃2 + µ̃ N −d¯1



h
2 i
p
E Rt:T = + (A10)
Pt2
St2 exp 2σ̃2 + 2µ̃ N −d¯3
 


Pt2
h i
3 N −d¯ − 3X 2 S exp σ̃2 + µ̃ N −d¯



h i X 2 t 2 1
p 3

E Rt:T = + (A11)
Pt3
3XSt2 exp 2σ̃2 + 2µ̃ N −d¯3 − St3 exp 92 σ̃2 + 3µ̃ N −d¯4
 
 

,
Pt3

where Pt is the put premium at time t. Equations (A6) through (A11) can be used to construct
ski,t:T for both call and put options for any level of moneyness, maturity, stock volatility, and
expected stock return.

36
 A.2. Ex-ante Coskewness

In this section we demonstrate how we construct estimates for βskd

i,t:T as defined in equation

(7). Our approach follows that of the prior section’s construction of the ex-ante skewness
measure, ski,t:T . To begin, we define εi,t:T as

εi,t:T = Ri,t:T − aLN LN

i,t − βi,t (RM,t:T ) (A12)

where Ri,t:T is the simple holding-period option return (either call or put) observed at maturity
as in equation (1), RM,t:T is the corresponding market return observed over the same period,
and aLN LN
i,t , βi,t are ex-ante coefficients defining the linear relationship under the assumption that

stock and market returns are jointly lognormal. Using equation (A12), we define the deviation
of the market return from its expected value, aLN
M,t , as

εM,t:T = RM,t:T − aLN

M,t ,

where LN superscripts denote the linear relationship under lognormality and time subscripts
indicate when variables are observed. Below we drop the LN superscripts for expositional
ease when the meaning is clear. The numerator of βskd
i,t:T defined by equation (7) can then be

written as

E[εi ε2M ] = E[(Ri − ai − βi RM ) (RM − aM )2 ] (A13)

= E[Ri R2M ] − 2aM E[Ri RM ] + a2M E[Ri ] + (2βi aM − ai )E[R2M ]

+ ai a2M − βi a3M − βi E[R3M ],

37
and the denominator of βskd
i,t:T defined by equation (7) can be written as

r h i
E ε2i,T E ε2M,T
 
(A14)

1/2
= E[R2i ] + a2i + β2i E[R2M ] − 2ai E[Ri ] − 2βi E[Ri RM ] + 2βi ai aM

+ E[R2M ] − a2M .

Now define βLN

i,t:T as
E[Ri RM ] − E[Ri ]E[RM ]
βLN
i,t:T = (A15)
E[R2M ] − E[RM ]2

and αLN
i,t:T as

αLN
i,t:T = E[Ri ] − βi E[RM ]. (A16)

The numerator and denominator of βskd

i,t:T given by equations (A13) and (A14) can be seen as
j j j
functions of E[Ri ], E[RM ], and E[Ri RM ] for j = {1, 2, 3}.

As noted in Section A.1 of the Appendix, the log stock return underlying option i is de-
noted as r̃i = ln(ST /St ), with mean µ̃i and variance σ̃2i , where subscript i references the op-
tion and again we note that superscript ∼ indicates association with the underlying as op-
j
posed to the option. Given this assumption we can derive option return moments, E[Ri ] for
j = {1, 2, 3}, from equations (A6) through (A11) for both calls and puts. Further, define
r̃M = ln(RM ), the log market return, and assume that r̃M is distributed N(µ̃M , σ̃2M ). We can also
j
calculate E[RM ] using the properties of the lognormal distribution,

j 1
E[RM ] = exp( jµ̃M + j2 σ̃2M ). (A17)
2

Again from Section A.1 of the Appendix, we define A = ln(X/St ) and let option i be a call
j
option with return Rci and price Ct . We can then write E[Rci RM ] as
 
j St r̃i + jr̃M

X  j
E[Rci RM ] = E e |r̃i > A − E RM |r̃i > A P (r̃i > A) . (A18)
Ct Ct

38
Let σ̃i,m denote the covariance between the stock underlying option i and the market. Applying
the results of Lien (1985), equation (A18) can be written as

j
E[Rci RM ] = (A19)
(σ̃2i + 2 jσ̃iM + j2 σ̃2M ) j2 σ̃2M
   
St X
N(q̄1 ) exp µ̃i + jµ̃M + − N(q̄2 ) exp jµ̃M + ,
Ct 2 Ct 2

σ̃2i + jσ̃iM −A+µ̃i p

where q̄1 = σ̃i and q̄2 = q̄1 − σ̃i . If option i is a put option with return Ri and price
Pt , then we have

p j
E[Ri RM ] = (A20)
j2 σ̃2M (σ̃2i + 2 jσ̃iM + j2 σ̃22 )
   
X St
N(−q̄2 ) exp jµ̃M + − N(−q̄1 ) exp µ̃i + jµ̃M + .
Pt 2 Pt 2

We estimate µ̃i , σ̃2i , µ̃M , σ̃2M , and σ̃iM using six months of daily data prior to t and use these
j j j
to estimate E[Ri ], E[RM ], and E[Rci RM ] for j = {1, 2, 3} as defined by equations (A6) through
(A11), (A17), (A19), and (A20). We then plug these estimated moments into equations (A13)
to (A16) to obtain estimates of ex-ante coskewness as defined in equation (7).

B. Option Database Screening Procedure

We create portfolios on the first trading date of each month. Let ti be the formation date for
portfolio i. We eliminate all options from portfolio i with any of the following characteristics
observable in the Ivy database on or before date ti :

1. Underlying Asset is an Index: Optionmetrics “index flag” is nonzero.

2. Underlying Asset is Not Common Stock: Optionmetrics “issue type” for underlying is
nonzero.

3. AM Settlement: The option expires at the market open of the last trading day, rather than
the close.

39
 4. Nonstandard Settlement: The number of shares to be delivered may be different from
100, additional securities and/or cash may be required, and/or the strike price and pre-
mium multipliers may be different than $100 per tick. Optionmetrics “special settlement
flag” is nonzero.

5. Missing Bid Price: The bid price on date ti is 998 or 999. Ivy uses these as missing
codes for some years.

6. Abnormal Bid-Ask Spread: The bid-ask spread on date ti is negative or greater than $5.

7. Abnormal Delta: The option delta on date ti , as calculated by Ivy, is below −1, above
+1, or missing.

8. Abnormal Implied Volatility: Implied volatility on date ti , as calculated by Ivy, is less

than zero or missing.26

9. Extreme Price: The midpoint of the bid and ask price is below 50% of intrinsic value or
$100 above intrinsic value.

10. Duplicates: Another record exists on date ti for an option of the same type (call or put),
on the same underlying asset, with the same time-to-maturity and same strike price.

11. Zero Open Interest: Open interest on the trading date immediately prior to date ti is zero.

12. No Trade: The Optionmetrics “last date” value is before ti .

13. Underlying Price History in CRSP is too Short: The underlying asset does not have at
least 100 nonmissing daily returns in CRSP over the six-month period prior to date ti .

14. Expiration Restrictions: The expiration month is greater than mi + 6, where mi is the
month in which portfolio i is formed, or the option expires after 2009.

Screens 1 and 2 allow us to focus on options written on common stock. We follow Duarte
and Jones (2007) in applying screens 3 through 11. Screen number 12 helps exclude stale
option quotes from the analysis. We apply screen 13 because we use six months of daily data
from CRSP prior to date ti to estimate moments of underlying assets, and we apply screen 14
because of data limitations.

40
 References

Ang, Andrew, Robert J. Hodrick, Yuhang Xing, and Xiaoyan Zhang, 2006, The cross-
section of volatility and expected returns, Journal of Finance 61, 259-299.

Ang, Andrew, Robert J. Hodrick, Yuhang Xing, and Xiaoyan Zhang, 2008, High idiosyn-
cratic volatility and low returns: International and further U.S. evidence, Journal of Financial
Economics 91, 1-23.

Arditti, Fred D., 1967, Risk and the required return on equity, Journal of Finance 22,
19-36.

Bakshi, Gurdip, and Nikunj Kapadia, 2003a, Delta hedged gains and the negative market
volatility risk premium, Review of Financial Studies 15, 527-566.

Bakshi, Gurdip, and Nikunj Kapadia, 2003b, Volatility risk premium embedded in indi-
vidual equity options: Some new insights, Journal of Derivatives Fall, 45-54.

Bali, Turan, Nusret Cakici, and Robert Whitelaw, 2010, Maxing out: Stocks as lotteries
and the cross-section of expected returns, Journal of Financial Economics 99, 427-446.

Barberis, Nicholas, and Ming Huang, 2008, Stocks as lotteries: The implications of prob-
ability weighting for security prices, American Economic Review 95, 2066-2100.

Bates, David 2003, Empirical option pricing: A retrospection, Journal of Econometrics

116, 387-404.

Black, Fischer, and Myron S. Scholes, 1973, The pricing of options and corporate liabili-
ties. Journal of Political Economy 3, 637-654.

Blume, Marshall E., and Robert Stambaugh, 1983. Biases in computed returns: An appli-
cation to the size effect, Journal of Financial Economics 12, 387-404

41
 Bollen, Nicholas P. B., and Robert E. Whaley, 2004, Does net buying pressure affect the
shape of implied volatility functions?, Journal of Finance 59, 711-753.

Bondarenko, Oleg, 2003, Why are put options so expensive? working paper, University of
Illinois at Chicago.

Boyer, Brian, Todd Mitton, and Keith Vorkink, 2010, Expected idiosyncratic skewness,
Review of Financial Studies 23, 169-202.

Brandt, Michael W., Alon Brav, John R. Graham, and Alok Kumar, 2008, The idiosyn-
cratic volatility puzzle: Time trend or speculative episodes?, Review of Financial Studies 23,
863-899.

Broadie, Mark, Mikhail Chernov, and Michael S. Johannes, 2009, Understanding expected
option returns. Review of Financial Studies 22, 4493-4529.

Brunnermeier, Markus K., Christian Gollier, and Jonathan A. Parker, 2007, Optimal be-
liefs, asset prices and the preference for skewed returns, American Economic Review Papers
and Proceedings 97, 159-165.

Brunnermeier, Markus K., and Jonathan A. Parker, 2005, Optimal expectations, American
Economic Review 95, 1092-1118.

Conine, Thomas E., and Maurry J. Tamarkin, 1981, On diversification given asymmetry
in returns, Journal of Finance 36 1143-1155.

Conrad, Jennifer, Robert F. Dittmar, and Eric Ghysels, 2013, Ex ante skewness and ex-
pected stock returns, Journal of Finance 68, 85-124.

Coval, Joshua D., and Tyler Shumway, 2001, Expected Option Returns, Journal of Fi-
nance, 56 983-1009.

Dierkes, Maik, 2009, Option-implied risk attitude under rank-dependent utility, Working
paper, University of Münster.

42
 Driessen, Joost, Pascal J. Maenhout, and Grigory Vilkov, 2009, The price of correlation
risk: Evidence from equity options. Journal of Finance 64, 1377-1406.

Duarte, Jefferson, and Chris Jones, 2007, The market price of volatility risk, Working
paper, USC.

Fama, Eugene F., and James D. MacBeth, 1973, Risk, return and equilibrium: Empirical
tests. Journal of Political Economy 81, 607-636.

Frazzini, Andrea, and Lasse H. Pedersen, 2011, Embedded leverage, Working paper, New
York University.

Garleanu, Nicolae, Lasse Heje Pedersen, and Alan M. Poteshman, 2009, Demand-based
option pricing, Review of Financial Studies 22 4259-4299.

Goyal, Amit, and Alessio Saretto, 2009, Cross-section of option returns and volatility,
Journal of Financial Economics, 94, 310-326.

Harvey, Campbell R., and Akhtar Siddique, 2000, Conditional skewness in asset pricing
tests, Journal of Finance 55, 1263-1295.

Kahneman, Daniel, and Amos Tversky, 1979, Prospect theory: An analysis of decision
under risk, Econometrica 47, 263-292.

Kraus, Alan, and Robert H. Litzenberger, 1976, Skewness preference and the valuation of
risky assets, Journal of Finance 31, 1085-1100.

Lien, Da-Hsiang Donald, 1985, Moments of truncated bivariate log-normal distributions,

Economic Letters 19, 243-247.

Merton, Robert C., 1971, Optimum consumption and portfolio rules in a continuous-time
model, Journal of Economic Theory, 3, 373-413.

Merton, Robert C., 1976, Option pricing when underlying stock returns are discontinuous,
Journal of Financial Economics 3, 125-144.

43
 Mitton, Todd, and Keith Vorkink, 2007, Equilibrium underdiversification and the prefer-
ence for skewness, Review of Financial Studies 20, 1255-1288.

Newey, Whitney K., and Kenneth D. West, 1987, A simple, positive definite, heteroskedas-
ticity and autocorrelation consistent covariance matrix, Econometrica 55, 703-708.

Ni, Sophie X., 2009, Stock option returns: A puzzle, Working paper, Hong Kong Univer-
sity of Science and Technology.

Pan, Jun, 2002, The jump-risk premia implicit in options: Evidence from an integrated
time-series study, Journal of Financial Economics, 63, 3-50.

Polkovnichenko, Valery, and Feng Zhao, 2013, Probability weighting functions implied
by options prices, Journal of Financial Economics, 107, 580-609.

Pool, Veronika Krepely, Hans R. Stoll, and Robert E. Whaley, 2008, Failure to exercise
call options: An anomaly and a trading game. Journal of Financial Markets, 11:1-35.

Rubinstein, Mark, 1973, The fundamental theorem of parameter-preference security valu-

ation, Journal of Financial and Quantitative Analysis 8, 61-69.

Scott, Robert C., and Phillip A. Horvath, 1980, On the direction of preference for moments
of higher order than the variance, Journal of Finance 35, 915-919.

Simkowitz, Michael A., and William Beedles, 1978. Diversification in a three-moment

world, Journal of Financial and Quantitative Analysis 13, 927-941.

Whaley, Robert E., 2003, Derivatives, in George Constantinides, Milton Harris, and Rene
Stulz, eds. Handbook of the Economics of Finance: Volume 1B Financial Markets and Asset
Pricing (Elsevier North-Holland Publishing).

Xing, Yuhang, Xiaoyan Zhang, and Rui Zhao, 2010, What does individual option volatility
smirk tell us about future equity returns? Journal of Financial and Quantitative Analysis 45,
641-662.

44
 Zhang, Yijie, 2005, Individual skewness and the cross-section of average stock returns,
Working paper, Yale University.

45
 Notes

1 This literatures includes the endogenous probabilities model of Brunnermeier

and Parker (2005) and Brunnermeier, Gollier, and Parker (2007), the heterogeneous skew-
ness preference model of Mitton and Vorkink (2007), and the cumulative prospect theory
model of Barberis and Huang (2008).

2 Work on skewness preferences pre-dates the articles cited above. Arditti (1967)
and Scott and Horvath (1980) show that well-behaved utility functions include a preference
for positive skewness. Rubinstein (1973) develops a model in which expected returns are
a linear function of higher comoments between returns and future wealth. Simkowitz and
Beedles (1978) and Conine and Tamarkin (1981) show that agents that have skewness
preference may prefer underdiversified portfolios in equilibrium in contrast to standard
representative agent equilibrium holdings.

3 Dierkes (2009) and Polkovnichenko and Zhao (2013) report empirical evidence

suggesting that skewness preference influences index option prices.

4 For further discussion see Boyer, Mitton, and Vorkink (2010) who use a set of
stock characteristics to forecast stock return skewness in a rolling cross-sectional regres-
sion framework. Zhang (2005) uses estimates of cross-sectional skewness within stock
peer groups to forecast future stock return skewness. Conrad, Dittmar, and Ghysels (2013)
obtain estimates of implied risk-neutral stock return skewness from the cross section of
equity options. Bali, Cakici, and Whitelaw (2010) use the maximum daily return as a
measure of lottery potential.

5 Whaley (2003) and Bates (2003) provide summaries of this literature. Whaley
(2003), in particular, argues that researchers should consider supply and demand effects in
option markets to understand the relationship between option prices and moneyness.

46
 6 The two measures of coskewness we control for include an ex-ante measure of
coskewness (similar to our ex-ante measure of total skewness) and return covariation with
the squared excess market return as in Harvey and Siddique (2000).

7 For example, see Coval and Shumway (2001), Bonderanko (2003), Driessen
Maenhout and Vilkov (2009), Duarte and Jones (2007), Broadie, Chernov, and Johannes
(2009), Goyal and Saretto (2009), and Frazinni and Pedersen (2011).

8 Intermediaries cannot perfectly hedge their option positions because of the im-
possibility of trading continuously, jumps in the underlying, stochastic volatility, and trans-
action costs. Hedging individual option positions can be especially difficult because of
greater short-sale constraints, lower liquidity, and the lack of a viable futures market.

9 Bollen and Whaley (2004) provide empirical evidence that net buying pressure
influences option prices.

10 The Internet Appendix is available in the online version of the article on The
Journal of Finance website.

11 The Ivy database currently begins January 4, 1996 and ends October 30, 2009.
Since we cannot observe open interest on the trading date immediately prior to the first
trading date of January 1996, we exclude this formation date from our sample.

12 This is the only screen that relies on information observable after the portfolio
formation dates.

13 Given irregular calendar intervals between the first trading date of each month
and option expiration dates, the exact maturities for portfolios in the second and third ex-
piration bins vary slightly across time. For expositional clarity below, however, we simply
refer to each maturity bin by the average number of days to expiration. For example, we
refer to options in the second expiration bin simply as “options that expire in 18 days.”

47
 14 We also estimate stock moments using a window of five years of daily data
prior to each formation date and use these as inputs for our measures of ex-ante skewness.
We find that this has little impact on our results. See the Internet Appendix for further
details.

15 We report betas for the various portfolios in the Internet Appendix.

16 The instantaneous beta for a call option is defined as ∆t CStt βS,t , where ∆t is the
option’s delta and βS,t is the underlying stock’s beta with respect to the market.

17 The average number of contracts per maturity/skewness bin in the revised sam-

ple is on the order of 45 to 65 with a minimum imposed of 10.

18 While log-returns are additive and are a satisfactory approximation for simple
stock returns, they may not be a satisfactory approximation for the returns of options held
to maturity as noted by Coval and Shumway (2001). The log-return of any option expiring
worthless is negative infinity, and all moments for the log-return of any option are either
positive or negative infinity.

19 For each period we exclude portfolios that do not have at least 10 options, as
we do in forming portfolios for our time-series results above. We also exclude time periods
that do not have at least 30 portfolios with sufficient options. Our results are robust to these
choices.

20 We calculate the volatility smirk for all options written on asset i for portfo-
lio formation date t as the volume-weighted average implied volatility across all puts for
which 0.80 < Xi /Si,t < 0.95 minus the volume-weighted average implied volatility across
all calls for which 0.95 < Xi /Si,t < 1.05.

21 Alternatively, we control for the influence of coskewness by including the av-

erage ex-ante coskewness rank for each option in Z p,t , where ex-ante coskewness is calcu-
lated as in the Appendix under the lognormal assumption. Doing so gives similar results.

48
 22 When average skewness is used as the sole explanatory variable the coefficient
is around -0.35 with t-statistics in the range of 2 to 3. This result is difficult to interpret
economically however, since the cross-sectional standard deviation of average skewness
across portfolios is about 22,000.

23 We find that the ten-two sorting combination is especially helpful in disentan-

gling the effects of variables that are highly correlated, such as moneyness and skewness,
because it helps eliminate cross-sectional variation in the control characteristic across the
two conditionally sorted skewness portfolios. Other sorting combinations give similar
pricing results, but also produce portfolios with greater cross-sectional variation in the
control characteristic, thereby complicating the interpretation.

24 We show in the Internet Appendix that CAPM alpha spreads across portfolios
unconditionally sorted on moneyness are similar to those of the ex-ante skewness portfo-
lios of Tables V and VII.

25 In the Internet Appendix we report results after unconditionally sorting on ex-

ante coskewness and find no significant relation between ex-ante coskewness and option
alphas.

26 Duarte and Jones (2007) argue that eliminating options that do not have a re-
ported delta or implied voltility in the Ivy Optionmetrics database induces a bias in mea-
suring average returns. We have estimated the alphas based on regression betas for Tables
VI and VII after including these observations, and find that our results don’t change.

49
 Figure 1. Option Retu urn Skewnesss Against Mo oneyness (sto ck return volaatility = 0.4, aannualized exxpected
he stock = 8%
return on th wness for a calll option, while
%, risk-free raate = 5%). Paanel A reportss return skew
Panel B reports skewness for a put option.
 

50
Figure 22: Option Return n Skewness Agaiinst Volatility Th his figure plots op urn skewness agaainst implied volaatility, σs, assuminng the annualized
ption holding retu d
expectedd return on the sto
ock is 8% and the risk-free rate is 5%.
5 Panel A repports return skewn
ness for in-the-m
money call optionss (X/S = 0.9). Pannel B reports retuurn
skewnesss for out-of-the-mmoney put optionss (X/S = 0.9). Paanel C reports retu urn skewness for out-of-the-moneyey call options (X//S = 1.1). Panel D reports return
skewnesss for in-the-moneey put options (X//S = 0.9). 
51
 

Figure 3: Frequency y of Trades at Bid and Ask. We reeport histogrrams of π = (Trade –

Bid)/(Askk – Bid)) for end of day option
o pricess on options taken from Bloomberg during Apriil,
May, and d June 2011. We limit our o data on each transactiion date to ththe top ex-annte skewnesss
(ski,t:T) qu
uintile for caalls and similarly for putts.

52
 Table I
Number of Option Quotes
This table reports summary statistics for individual equity options taken from the Ivy Database that survive our
data filter as described in Section B of the Appendix. Panel A displays the total number of option quotes each
year, the source for the closing prices at expiration (ST), and the number of unique underlying assets. On each
portfolio formation date we then sort options into ex-ante skewness quintiles where ex-ante skewness is defined
by equations (1) through (4). Panel B reports for each maturity the average number of options in each ex-ante
skewness quintile, the average number of unique underlying assets across skewness quintiles (“Unique
Underlying”), and the average number of skewness quintiles spanned by a single underlying asset (“Underlying
Span”).

Panel A. Number of Option Quotes

Screened ST ST ST Unique
Year Data from Ivy from CRSP Observable Underlying
1996 37,695 37,695 0 37,695 1,094
1997 54,806 54,805 1 54,806 1,440
1998 67,527 67,515 4 67,519 1,679
1999 91,415 91,386 0 91,386 1,817
2000 152,511 152,509 0 152,509 1,914
2001 105,854 105,838 0 105,838 1,864
2002 94,856 94,835 0 94,835 1,954
2003 94,203 94,165 0 94,165 1,917
2004 108,917 108,825 0 108,825 2,147
2005 125,908 125,691 7 125,698 2,250
2006 160,236 160,017 14 160,031 2,430
2007 194,822 194,402 10 194,412 2,613
2008 221,945 221,616 0 221,616 2,543
2009 169,600 161,586 7,880 169,466 2,347

Total 1,680,295 1,670,885 7,916 1,678,801

% of Total 99.44% 0.47% 99.91%
 

53
 
 Table I (continued)

Panel B. Average # of Securities

Skew Days to Expiration
Quintile 7 18 48
Calls
Low 276 438 449
2 277 439 450
3 277 439 450
4 277 439 450
High 276 439 450
Unique Underlying 729.59 962.87 934.42
Underlying Span 1.69 1.93 1.89

Puts
Low 230 342 306
2 230 342 307
3 230 342 307
4 230 342 307
High 230 342 306
Unique Underlying 589.51 743.26 637.56
Underlying Span 1.72 1.94 1.92

54
 
 Table II
Ex-Ante and Ex-Post Skewness
On each portfolio formation date we sort options into ex-ante skewness quintiles, where ex-ante skewness is
defined by equations (1) through (4). We then calculate the median ex-ante skewness for each quintile. Panel A
reports the time-series average of these medians, which increases monotonically across quintiles by construction.
Panel B reports the time-series skewness of each portfolio. The last two rows of Panel B report the difference in
portfolio skewness across the low and high ex-ante skewness quintiles, and the t-statistic for this difference. This
t-statistic is computed by GMM to account for cross-sectional dependence using the approach of Newey and West
(1987) to account for inter-temporal dependence. Panel C reports the average cross-sectional skewness for these
same portfolios. The last two rows of Panel C report differences across the high and low skewness qunitiles along
with Newey-West (1987) t-statistics.

Calls: Puts:
Skew Days to Expiration Days to Expiration
Quintile 7 18 48 7 18 48
Panel A. Average Ex-Ante Skewness
Low 0.40 0.55 0.86 0.24 0.31 0.42
2 1.03 1.18 1.51 1.00 1.12 1.27
3 1.82 1.93 2.20 1.91 2.03 2.15
4 3.36 3.11 3.23 3.75 3.77 3.65
High 24.94 7.31 6.27 15.78 13.52 10.04
Panel B. Portfolio Skewness
Low 0.07 -0.30 0.38 0.38 0.62 1.28
2 0.57 0.16 0.76 0.77 1.20 1.86
3 1.09 0.71 0.96 1.35 1.76 2.80
4 1.82 1.37 1.17 2.09 2.70 3.94
High 2.68 1.64 1.48 2.71 4.55 7.01
Low-High -2.61 -1.93 -1.10 -2.33 -3.93 -5.73
(t-stat) -(4.92) -(7.14) -(3.24) -(6.34) -(5.15) -(5.42)
Panel C. Average Cross-Sectional Skewness
Low 0.86 0.99 1.56 0.95 1.07 1.61
2 1.52 2.03 2.84 1.89 2.35 2.54
3 2.95 3.47 3.96 3.61 3.88 3.65
4 5.04 5.46 5.50 6.00 6.13 5.20
High 9.11 9.58 8.23 10.08 10.24 8.37
Low-High -8.29 -8.62 -6.73 -9.13 -9.19 -6.75
(t-stat) -(25.66) -(23.72) -(21.44) -(29.75) -(27.24) -(20.75)

55
 
 Table III
Liquidity
This table reports additional summary statistics for the options that survive the data filter outlined in Section B of
the Appendix. For Panels A and B we first measure the average characteristic (either bid-ask spread or volume)
across options within each ex-ante skenwess quntile on each portfolio formation date, and then report the time-
series average of this measure. Panel A reports the average bid-ask spread, defined as (ask-bid)/midpoint for the
portfolio formation date, Panel B reports average trading volume defined as the number of contracts traded on the
portfolio formation date. For Panel C we first measure the total dollar volume across options within each ex-ante
skenwess quntile on each portfolio formation date, and then report the time-series average of this measure. Total
dollar volume is the sum of closing price volume on the portfolio formation date across all options within each
ex-ante skewness quintile. Each panel reports results separately for calls and puts, as well as differences in each
characteristic across the high and low skewness quintiles for each maturity. We also report Newey-West (1987)
standard errors for these differences.

Calls: Puts:
Days to Expiration Days to Expiration
Skew Quintile 7 18 48 7 18 48
Panel A. Average Bid-Ask Spreads
Low 0.07 0.06 0.06 0.07 0.07 0.07
2 0.13 0.11 0.10 0.14 0.12 0.10
3 0.23 0.19 0.14 0.25 0.20 0.13
4 0.46 0.34 0.20 0.47 0.35 0.19
High 1.01 0.84 0.48 0.93 0.77 0.41
Low-High -0.94 -0.78 -0.42 -0.86 -0.70 -0.34
(st. error) (0.01) (0.01) (0.01) (0.01) (0.02) (0.01)
Panel B. Average Daily Volume per Contract
Low 182.70 139.39 89.56 160.29 157.39 96.58
2 342.48 264.05 139.70 290.57 223.87 134.49
3 488.52 371.36 170.56 413.04 312.30 152.45
4 483.47 362.87 175.63 392.38 291.13 145.77
High 318.59 247.20 145.51 253.60 180.14 108.35
Low-High -135.89 -107.81 -55.95 -93.31 -22.75 -11.77
(st. error) (14.72) (13.12) (8.18) (8.72) (12.77) (5.16)
Panel C. Average Daily Total Dollar Volume (Millions)
Low 52.27 48.67 30.20 25.76 50.61 26.06
2 29.87 47.03 25.09 20.07 25.29 13.66
3 22.18 33.87 18.76 17.54 23.96 12.37
4 9.62 15.54 11.69 8.52 12.35 9.06
High 2.78 4.06 4.81 2.58 3.27 3.69
Low-High 49.49 44.61 25.39 23.18 47.34 22.37
(st. error) (2.30) (10.50) (11.68) (7.81) (6.33) (3.65)  

   

56
 
 Table IV
Average Weekly Returns
This table reports the average holding-period returns for individual equity option portfolios taken from the Ivy
database over the period 1996 to 2009. The portfolios are constructed by sorting on ex-ante skewness as in
equations (1) to (4) and the returns are holding-period returns to expiration as in equation (3) for calls, using the
midpoint of the bid and ask prices as the proxy for price. The final two rows report differences in average returns
across the high and low skewness qunitiles along with Newey-West (1987) t-statistics that test whether these
differences are equal to zero. Statistical significance at the 10%, 5%, and 1% levels is indicated by *, **, and
***, respectively.

Call Options: Put Options:

Skew Days to Expiration Days to Expiration
Quintile 7 18 48 7 18 48
Low 1.87 0.22 0.83 -5.38 ** -0.90 0.02
2 1.34 0.83 1.03 -8.51 ** -0.74 -0.33
3 -0.78 0.97 1.01 -15.52 *** -2.24 -1.00
4 -3.92 0.56 0.09 -31.78 *** -4.67 -1.77
High -35.25 *** -8.76 *** -2.58 *** -59.98 *** -13.73 *** -3.16
Low-High 37.11 *** 8.98 *** 3.40 *** 54.60 *** 12.83 *** 3.18 *
(t-stat) (7.56) (4.07) (5.05) (14.13) (3.92) (1.94)  

   

57
 
 Table V
CAPM Pricing Errors
Panel A reports CAPM pricing errors as in equation (5) for portfolios of individual equity options taken from the
Ivy database over the period 1996 to 2009. Portfolios are formed by sorting on ex-ante skewness as in equations
(1) - (4) and returns are holding-period returns to expiration as in equation (3) for calls, using the mid-point of the
bid and ask prices as the proxy for price. Panel B reports CAPM pricing errors for the underlying stocks. We
obtain stock CAPM pricing errors by regressing stock portfolio excess returns on excess market returns over the
same time horizon as we do for the option portfolios in Panel A. In the final rows of each panel we report
differences in CAPM pricing errors across the high and low skewness quintiles along with GMM t-statistics
calculated using the approach of Newey and West (1987). Statistical significance at the 10%, 5% and 1%
significance levels is indicated, respectively by *, **, and ***.

Calls: Puts:
Skew Days to Expiration Days to Expiration
Quintile 7 18 48 7 18 48
Panel A. Option CAPM Pricing Errors
Low -1.20 -0.01 0.37 -2.42 ** -0.79 0.41
2 -3.44 * 0.53 0.43 -3.56 ** -0.52 0.26
3 -6.90 ** 0.62 0.32 -9.30 *** -1.93 -0.27
4 -10.71 ** 0.17 -0.61 -25.59 *** -4.27 ** -0.93
High -40.15 *** -9.10 *** -3.22 *** -55.68 *** -13.37 *** -2.12
Low-High 38.95 *** 9.09 *** 3.58 *** 53.27 *** 12.58 *** 2.53
(t-stat) (9.00) (4.27) (5.36) (12.73) (4.43) (1.42)
Panel B. Underlying Stock CAPM Pricing Errors
Low -0.02 0.00 0.05 -0.13 -0.10 -0.21 ***
2 -0.15 -0.03 0.00 -0.20 * -0.06 -0.12 **
3 -0.22 * -0.06 -0.06 -0.25 * -0.02 -0.05
4 -0.23 * -0.06 -0.13 ** -0.23 * -0.01 0.02
High -0.03 -0.04 -0.21 *** -0.13 0.01 0.08
Low-High 0.01 0.05 0.26 *** -0.01 -0.11 -0.29 **
(t-stat) (0.08) (0.44) (2.71) -(0.04) -(0.89) -(2.53)  

   

58
 
 Table VI
CAPM Pricing Errors Holding Underlying Assets Fixed
This table reports CAPM pricing errors as in Table V with the exception that the underlying assets are held
constant across ex-ante skewness quintiles, and across contract type (call/put) for options with the same maturity.
On a given formation date we classify options with a given maturity into one of four groups: call in high quintile
(CH), call in low quintile (CL), put in high quintile (PH), and put in low quintile (PL). For each underlying asset,
we count the number of option contracts within each of these four groups, and then identify the group with the
least number of contracts. We then force the number of option contracts written on a given underlying asset to be
the same across all four groups, by eliminating contracts from groups that have more than the minimum with
independent uniform probability. By conducting this exercise for every portfolio formation date we develop a
sample in which the set of underlying assets is identical across ex-ante skewness bins and contract type (call/put)
for options of the same maturity. Panel A reports CAPM pricing errors as in equation (5) for the individual equity
option portfolios. Panel B reports CAPM pricing errors for the underlying stocks. We obtain stock CAPM pricing
errors by regressing stock portfolio excess returns on excess market returns over the same time horizon as we do
for the option portfolios in Panel A. In the final row of Panel A we report differences in CAPM pricing errors
across the high and low skewness quintiles along with GMM t-statistics calculated using the approach of Newey
and West (1987). Statistical significance at the 10%, 5%, and 1% level is indicated by *, **, and ***,
respectively. Pricing errors for stocks in Panel B across the high and low ex-ante skewness quintiles and across
contract type are identical by construction.

Calls: Puts:
Skew Days to Expiration Days to Expiration
Quintile 7 18 48 7 18 48
Panel A. Option CAPM Pricing Errors
Low 0.21 -0.01 -0.41 -2.54 * -1.01 0.09
High -39.27 *** -14.03 *** -4.15 *** -54.49 *** -16.79 *** -2.45

Low-High 39.48 *** 14.02 *** 3.74 *** 51.95 *** 15.78 *** 2.54
(t-stat) (4.76) (4.00) (3.02) (6.22) (4.50) (1.00)
Panel B. Underlying Stock CAPM Pricing Errors
Low 0.02 0.03 -0.09 0.02 0.03 -0.09
High 0.02 0.03 -0.09 0.02 0.03 -0.09

Low-High 0.00 0.00 0.00 0.00 0.00 0.00  

59
 
 Table VII
Pricing Errors, Three-Factor Model
This table reports pricing errors as in equation (5) for portfolios of individual equity options taken from the Ivy
database over the period 1996 to 2009. Portfolios are formed by sorting on ex-ante skewness as given in
equations (1) to (4) and returns are holding-period returns to expiration as in equation (3) for calls, using the
midpoint of the bid and ask prices as the proxy for price. Here we report pricing errors for a three-factor model
where the three factors are the excess market return, the return on the zero-investment coskewness portfolio of
Harvey and Siddique (2000), and the excess return on a zero-delta S&P 500 index straddle. In the final rows of
each panel we report differences in pricing errors across the high and low skewness quintiles along with GMM t-
statistics calculated using the approach of Newey and West (1987). Statistical significance at the 10%, 5%, and
1% level is indicated by *, **, and ***, respectively.

Calls: Puts:
Skew Days to Expiration Days to Expiration
Quintile 7 18 48 7 18 48
Low -1.37 0.09 0.52 -2.24 ** -0.96 * 0.41
2 -3.07 1.00 0.70 -2.57 -0.52 0.30
3 -5.75 * 1.42 0.65 -6.92 ** -1.49 -0.12
4 -8.21 * 1.36 -0.34 -21.53 *** -3.25 -0.62
High -36.69 *** -7.89 *** -3.00 *** -52.13 *** -12.07 *** -1.37
Low-High 35.31 *** 7.98 *** 3.52 ** 49.88 *** 11.11 *** 1.78 **
(t-stat) (7.45) (3.78) (5.57) (10.28) (3.95) (2.04)  

   

60
 
 Table VIII
Fama-McBeth Regressions – 18 Days to Maturity
This table reports the time-series average of cross-sectional regression parameters following Fama and MacBeth
(1973) using individual equity options taken from the Ivy database over the period 1996 to 2009. Each month we
construct 100 portfolios by sorting options on ex-ante skewness as given in equations (1) to (4) and regress these
portfolio returns on a set of risk controls and other portfolio characteristics. Risk controls include MKT, which
corresponds to a regression market beta as defined by equation (5) in the paper, MOM, which corresponds to a
regression momentum beta on the momentum factor provided by Ken French, VOL, the volatility risk beta of the
option portfolio obtained by regressing portfolio returns on returns of a zero-delta index straddle, and CSKL, the
coskewness beta of the option portfolio obtained by regressing portfolio returns on squared excess market returns.
We also include ex-ante skewness, skt:T, as an explanatory variable in our cross-sectional regressions, measured as
is the ex-ante skewness rank of the portfolio, which can take a value from zero to 99. To control for other
characteristics, we first independently sort options into 100 bins by moneyness, volume, spread, and smirk as
defined in the paper. We then measure the average characteristic rank across options within each of the 100
skewness-sorted portfolios for each characteristic, and use these average rank measures as additional explanatory
variables in our cross-sectional regressions. Newey-West (1987) t-statistics are in parentheses. In Panel A we
report results for portfolios of call options with 18 days to maturity. In Panel B, we report analogous results for
portfolios of put options. Statistical significance at the 10%, 5%, and 1% level is indicated by *, **, and ***,
respectively.

Panel A. Call Options

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
skt:T -0.097 *** -0.411 ***
-(3.23) -(4.07)
mkt 0.804 *** 2.107 ***
(3.03) (7.06)
mom
 -0.059 0.128
-(0.11) (0.24)
vol
 -17.668 *** 2.426
-(6.69) (1.02)
csk
 -0.208 *** -0.016
-(5.36) -(0.38)
X/S -0.098 *** 0.258 ***
-(3.06) (2.58)
Volume 0.209 *** 0.008
(3.41) (0.21)
Spread -0.134 *** -0.068
-(3.62) -(1.25)
Smirk 0.005 -0.029
(0.14) -(1.17)

61
 
 Table VIII (continued)

Panel B. Put Options

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
skt:T -0.142 *** -0.437 ***
-(2.81) -(3.46)
mkt
 0.194 -1.151 ***
(0.66) -(3.29)
mom 3.538 *** 0.448
(5.46) (0.79)
vol -2.052 4.130 *
-(1.19) (1.91)
csk -0.019 -0.040 *
-(1.44) -(1.79)
X/S 0.144 *** -0.242 *
(2.71) -(1.89)
Volume 0.139 0.067 *
(1.45) (1.79)
Spread -0.189 *** -0.080
-(2.87) -(1.56)
Smirk -0.184 *** -0.017
-(4.92) -(0.78)

62
 
 Table IX
Double Sorts on Moneyness
This table reports CAPM pricing errors for portfolios of individual equity options taken from the Ivy database
over the period 1996 to 2009. In Panel A we adopt a double sort procedure to net out the influence of moneyness.
For a given portfolio formation date, we first sort options by moneyness (X/S) into 10 portfolios and then within
each moneyness decile sort options into two portfolios by ex-ante skewness. We then average the one-period
returns across all moneyness-sorted portfolios to create returns of two portfolios with similar levels of moneyness
but different skewness. In Panel B we reverse this procedure, and first sort options by ex-ante skewness into10
bins, and then within each ex-ante skewness bin sort options by moneyness into two bins. We then average the
one-period returns across all ex-ante skewness-sorted portfolios to create returns of two portfolios with similar
levels of skewness but different moneyness. We obtain CAPM pricing errors by regressing option portfolio
returns on excess market returns as in equation (5). In the final rows of each panel we report differences in CAPM
pricing errors across the two conditionally sorted portfolios along with Newey-West (1987) t-statistics. Statistical
significance at the 10%, 5%, and 1% level is indicated by *, **, and ***, respectively.

Panel A: Controlling for Moneyness

Skew Calls: Puts:
Rank Days to Expiration Days to Expiration
7 18 48 7 18 48
Low -8.05 *** 0.54 0.25 -14.43 *** -1.63 0.35
High -16.86 *** -3.64 *** -1.32 ** -24.13 *** -6.70 *** -1.40 *

Low-High 8.81 ***4.17 *** 1.57 *** 9.70 *** 5.07 *** 1.76 ***
(t-stat) (4.93) (3.73) (2.58) (4.05) (4.66) (2.68)
Panel B: Controlling for Ex-Ante Skewness
Moneyness Calls: Puts:
Rank Days to Expiration Days to Expiration
7 18 48 7 18 48
Low -10.73 *** -1.61 -0.58 -20.32 *** -3.60 ** 0.07
High -14.20 *** -1.50 -0.51 -18.29 *** -4.74 *** -1.13 **

Low-High 3.48 * -0.11 -0.07 -2.04 1.14 1.20 **

(t-stat) (1.91) -(0.09) -(0.12) -(0.91) (1.01) (2.41)

63
 
 Table X
Double Sorts on Ex-ante Coskewness
This table reports CAPM pricing errors for portfolios of individual equity options taken from the Ivy database
over the period 1996 to 2009. In Panel A we adopt a double sort procedure to net out the influence of ex-ante
coskewness. For a given portfolio formation date, we first sort options by ex-ante coskewness (as given in
Section A.2 of the Appendix) into 10 portfolios and then within each ex-ante coskewness decile sort options into
two portfolios by ex-ante skewness. We then average the one-period returns across all ex-ante coskewness-sorted
portfolios to create returns of two portfolios with similar levels of coskewness but different total skewness. In
Panel B we reverse this procedure, and first sort options by ex-ante skewness into10 bins, and then within each
ex-ante skewness bin sort options by ex-ante coskewness into two bins. We then average the one-period returns
across all ex-ante skewness-sorted portfolios to create returns of two portfolios with similar levels of skewness
but different coskewness. We obtain CAPM pricing errors by regressing option portfolio returns on excess
market returns as in equation (5). In the final rows of each panel we report differences in CAPM pricing errors
across the two conditionally sorted portfolios along with Newey-West (1987) t-statistics. Statistical significance at
the 10%, 5%, and 1% level is indicated by *, **, and ***, respectively.

Panel A: Controlling for Ex Ante Coskewness

Skew Calls: Puts:
Rank Days to Expiration Days to Expiration
7 18 48 7 18 48
Low -2.36 0.89 0.38 -4.42 ** -0.86 0.08
High -22.50 *** -3.98 ** -1.46 ** -34.03 *** -7.47 *** -1.13
Low-High 20.14 *** 4.87 *** 1.84 *** 29.61 *** 6.60 *** 1.21
(t-stat) (8.79) (4.51) (4.73) (10.68) (4.07) (1.26)
Panel B: Controlling for Ex-Ante Skewness
Coskew Calls: Puts:
Rank Days to Expiration Days to Expiration
7 18 48 7 18 48
Low -15.90 *** -2.83 ** -0.85 * -21.03 *** -3.96 *** -0.31
High -9.09 *** -0.29 -0.23 -17.59 *** -4.38 *** -0.74
Low-High -6.81 *** -2.54 ** -0.62 -3.44 0.42 0.43
(t-stat) -(2.59) -(2.14) -(1.38) -(1.34) (0.33) (0.66)

64
 
 Table XI
CAPM Pricing Errors from Writing Options at the Bid
This table reports the estimated CAPM pricing errors for portfolios of individual equity call options taken
from the Ivy database over the period 1996 to 2009. The portfolios are constructed by sorting on expected
skewness as in equation (4) and the returns are holding-period returns to expiration as in equation (3) for calls,
multiplied by -1 to indicate returns from writing options, using the bid price in the denominator. We obtain
betas by regressing option portfolio returns on excess market returns as in equation (5). Panel A reports
results for calls while Panel B reports results for puts. In the final rows of each panel we report differences in
CAPM pricing errors across the high and low skewness quintiles along with GMM t-statistics calculated
using the approach of Newey and West (1987) that test whether these differences are equal to zero.

Calls: Puts:
Skew Days to Expiration Days to Expiration
Qunitile 7 18 48 7 18 48
Low -2.44 *** -1.42 *** -0.88 ** -1.29 -0.77 -1.00 **
2 -3.88 ** -2.99 *** -1.29 ** -4.00 ** -2.49 *** -1.15 **
3 -6.42 ** -5.20 *** -1.52 ** -5.85 * -3.32 ** -0.91
4 -15.45 *** -9.61 *** -1.20 -0.69 -4.26 * -0.69
High -4.80 -10.26 *** -0.78 24.46 *** -1.29 -0.73
Low-High 2.35 8.84 *** -0.10 -25.75 *** 0.52 -0.27
(t-stat) (0.32) (2.76) -(0.11) -(3.84) (0.14) -(0.13)

65