Forthcoming in the Financial Analysts Journal

Exploiting Option Information

in the Equity Market

GUIDO BALTUSSEN, BART VAN DER GRIENT, WILMA DE GROOT, CFA,

ERIK HENNINK AND WEILI ZHOU, CFA *

This draft: 31 December 2011

ABSTRACT
Public option market information contains exploitable information for equity investors
for an investable universe of liquid large-cap stocks. Strategies based on several option
measures predict returns and alphas on the underlying stock. Transaction costs are an
important factor given the high turnover of these strategies, but significant net alphas
can be obtained when using a simple transaction cost reducing approach. These findings
suggest that information diffuses from the option market into the underlying stock
market.

Key words: stock selection, behavioral finance, alpha strategies, equity returns, option
markets, stock markets, information spillover, asset pricing.
JEL: G11, G12

*
Guido Baltussen is at the Erasmus School of Economics, Rotterdam, the Netherlands and Structured
Investment Strategies, ING Investment Management, The Hague, the Netherlands. E-mail:
[email protected]. Bart van der Grient, Wilma de Groot, CFA, and Weili Zhou, CFA, are at Robeco
Quantitative Strategies, Rotterdam, The Netherlands. Erik Hennink is at Rabobank International, Utrecht, the
Netherlands.

Electronic
Electroniccopy
copyavailable
availableat:
at:https://ssrn.com/abstract=2001461
http://ssrn.com/abstract=2001461
1 Introduction
In this paper we examine whether public information contained in the option market
predicts cross-sectional stock returns for a well-investable universe of highly liquid U.S.
large-cap stocks and thus provides valuable, exploitable information for equity investors.
Equity options have become an increasingly popular investment alternative over the past
decades. They have asymmetric payoff characteristics and allow investors to take highly
leveraged positions, making them important instruments for speculation or hedging.
Options allow investors to take a view on the price development and risk of the underlying
stocks. In fact, option prices reflect the expectations and worries investors have about
future stock-price developments. Therefore, many practitioners view the equity option
market as a primary source of information about the expected return, risk and sentiment of
individual stocks and the equity market in general. The question is whether this public
information also provides valuable information to investors. That is, can investors exploit
this information?
Standard economic theory suggests not. In complete markets options are redundant
securities and the public information they contain should already be reflected in the prices
of other assets. Moreover, in efficient markets stock prices should adjust immediately to
public information. However, empirical work and intuition suggest otherwise. Empirical
studies generally find that options are non-redundant securities (see for example Buraschi
and Jackwerth (2001)). Intuitively, the option market may lead the equity market if an
investor with positive or negative information on a stock chooses to invest in the option
market rather than in the stock itself. For example, Black (1975) argues that traders prefer
to exploit private information by trading in the option market, because the option market
provides reduced transaction costs, increased financial leverage and a lack of short selling
constraints. If equity-market investors fail to trade on this information a lead-lag
relationship will emerge between the option market and stock market. In fact, Hong and
Stein (1999) argue that information diffuses gradually into and across markets, a finding
empirically confirmed by Hong, Torous and Valkanov (2007) 1 . Similarly, Chakravarty,
Gulen and Mayhew (2004) find that the equity option market contains information that is

1 More specific, Hong, Torous and Valkanov (2007) find that information diffuses from several industry stock
indices into the remainder of the stock market for up to two months.

Electronic
Electroniccopy
copyavailable
availableat:
at:https://ssrn.com/abstract=2001461
http://ssrn.com/abstract=2001461
later on reflected in stock prices. These findings suggest that publicly available information
contained in the option market affects future stock prices, causing stock return
predictability.
Indeed, several recent studies propose option market measures that contain
economically and statistically significant information for subsequent returns on the
associated stocks. Xing, Zhang and Zhao (2010) employ the difference between the implied
volatilities of out-of-the-money put and at-the-money call options, commonly referred to as
the out-of-the-money volatility skew, a measure that reflects the (informed) worries
investors have about negative price movements. They find that stocks with the largest skew
underperform stocks with the smallest skew. Bali and Hovakimian (2009), and Goyal and
Saretto (2009) utilize the difference between realized and implied volatilities, a measure
that captures the volatility risk of a stock. They find that a strategy that buys stocks with
the lowest realized versus implied volatility spread and shorts stocks with the highest
spread produces significant positive returns. Bali and Hovakimian (2009) and Cremers and
Weinbaum (2010) employ the spread between implied volatilities of at-the-money put
versus call options, also known as the at-the-money volatility skew, which they argue
captures the trading activity of informed investors or jump risk. Stocks with a low spread
(i.e. stocks that have higher call than put implied volatilities) outperform stocks with a high
spread. In addition, Cremers and Weinbaum (2010) employ the recent change in the spread
between the implied volatilities of at-the-money put and call options, which might capture
the change in informed trading, and find a negative relation with stock returns.
The aforementioned studies reveal strong predictive power of public option market
information for stock returns between 1996 and 2005. However, all studies focus on a broad
universe of stocks, which might not be exploitable for most practitioners due to their
liquidity constraints and needs. In addition, they do not analyze the impact of transaction
costs when evaluating the profitability. Moreover, some of the studies show a declining
performance towards the end of their sample and omit the highly volatile period around the
subprime crisis, a period in which many equity funds were closed. Given the relatively
short sample period of the papers on this topic, these extra years are highly relevant.
Hence, from a practitioner’s perspective, the added value of these studies is yet unclear.
In this paper we examine whether the above four measures: i) out-of-the-money
volatility skew, ii) realized versus implied volatility spread, iii) at-the-money volatility
skew, and iv) the change in the at-the-money volatility skew, provide valuable information

Electronic copy available at: https://ssrn.com/abstract=2001461

that is exploitable by equity investors. For this end, we (i) study these strategies on a well-
investable universe of U.S. large caps, i.e. the stocks that are most interesting for
practitioners due to their liquidity, (ii) extend the sample period to include the recent,
volatile crisis period, (iii) examine the combined predictive power of the aforementioned
variables in an integrated option information strategy (as from a practitioner’s perspective
it is important to know whether performance improves when combining the individual
variables), and (iv) thoroughly analyze the impact of transaction costs. In addition, we
examine the robustness of the strategy in different market conditions, as well as its
interaction with information uncertainty.
Our findings are as follows. We find that publicly available information in the option
market has been relevant for equity investors. Trading strategies based on the out-of-the-
money volatility skew, realized versus implied volatility spread, at-the-money volatility
skew, and the change in the at-the-money volatility skew all yield significant returns and
alphas, using a well-investable universe of liquid large caps. Although some studies report
that the predictive power decreases over time until 2004-2005, we also find significant
returns in the recent crisis years, which can serve as out-of-sample evidence. Combining all
four variables into a combined option information strategy yields even larger profit
opportunities of 10% per year for the long-short portfolio, thereby strengthening the
relevance of the publicly available information contained in option prices for equity
investors. These results are robust over bull, bear, volatile and calm markets and are
generally of similar magnitude for stocks with low or high information uncertainty.
Furthermore, the documented anomalies are at least as strong when applied to the 100
largest U.S. stocks. Exploiting the option information measures requires an extremely high
turnover. As a consequence, all profitability is estimated to be consumed by transaction
costs on our investable universe. However, still more than half of the documented
profitability persists for a strategy based on only the 100 largest stocks when we take
simple steps to control the transaction costs of managing the portfolio. The net return is
economically strong and statistically significant with a value of above 7% per year for the
long-short portfolio. This leads us to conclude that the documented strategies are
exploitable by practitioners.
In the remainder of this paper we will provide more detail on these results. Section 2
describes the option measures and our well-investable stock universe. Section 3 outlines the
methodology. Section 4 reports the results of each of the four individual variables, the

Electronic copy available at: https://ssrn.com/abstract=2001461

results of the combined option information strategy, its results over market conditions and
information uncertainty portfolios, and the impact of transaction costs. Finally, Section 5
concludes.

2 Data and option market measures

This section describes first our data sample, followed by the option measures we
employ, and the descriptive statistics of our sample and variables.

2.1 Sample
To examine the investability of the option information strategies, we limit our
universe to the 1,250 largest stocks in the S&P/Citigroup U.S. Broad Market Index during
the period between January 1996 and October 2009. This corresponds to a minimum
market capitalization of approximately USD 1 billion in 2009, resulting in highly liquid
stocks that can easily be traded at limited transaction costs. This requirement makes the
stocks investable for many equity investors. Daily stock returns, including dividends and
market capitalizations, are retrieved from FactSet Prices and accounting data from
Compustat.
For the stocks in this universe we extract option data from OptionMetrics, which
contains end-of-day bid and ask quotes, open interests and trading volumes for all equity
options traded in the U.S. For individual equity options, which are American,
OptionMetrics calculates implied volatilities and Greeks using a binominal-tree model
based on the algorithm of Cox, Ross, and Rubinstein (1979). This algorithm copes with
discrete dividend payments and the possibility of early exercise. We apply the following
screens on all options to ensure that we select liquid and heavily traded options that we
believe contain the most reliable information. We filter out options with zero volume or
open interest. As most activity in options is concentrated in the short end, we select options
with a remaining maturity of approximately one month by requiring a maturity between 10
and 40 trading days. We follow Xing et al. (2010) when separating options into at-the-
money (ATM) and out-of-the-money (OTM). A put or call option is defined as ATM when the
ratio of the strike price to the stock price is between 0.95 and 1.05 and a put option as OTM

Electronic copy available at: https://ssrn.com/abstract=2001461

when the ratio is lower than 0.95, but higher than 0.80. When multiple options fall into the
same group, we select options with moneyness closest to 1.00 (ATM) or 0.95 (OTM).2

2.2 Option measures

The first option market measure we employ is the OTM volatility skew, which is
thought to reflect worries about negative price movements, for example due to an
informational advantage (Xing et al. (2010)). For example, Gârleanu, Pedersen and
Poteshman (2009) argue that companies for which investors have relatively pessimistic
perceptions, investors would tend to buy put options either for protection against, or
speculation on future stock price drops. This increase in the demand for put options leads to
a higher price and implied volatility, yielding a steeper volatility smile. Stocks with a high
volatility skew should therefore underperform stocks with a lower skew. Following Xing et
al. (2010), we compute the OTM volatility skew as follows:

SKEWi ,OTM
t = IVi ,OTMP
t − IVi ,ATMC
t (1)

where IVi ,OTMP

t ( IVi ,ATMC
t ) denotes the implied volatility of the OTM put (ATM call) option on

stock i in week t. We use the weekly average of the IV variables to reduce the effect of noise,
while requiring at least two non-missing values during the past five days to compute the
measure.
The second measure is the realized (historical) versus implied volatility spread,
which is thought to capture the volatility risk of a stock; Bali and Hovakimian (2009) show
that stocks with a higher spread between realized and implied volatility have higher
volatility risk. Moreover, Bakshi and Kapadia (2003a, 2003b) show that the realized versus
implied volatility spread bears a negative volatility risk premium. Stocks with a high
realized versus implied volatility spread should therefore underperform. We measure the
realized versus implied volatility spread by the difference between the realized volatility of
the past 20 daily stock returns and the implied volatility:

RVIVi ,t = RVi ,t − IVi ,ATM

t (2)

2 We also replicated our analysis using 0.925 as the OTM boundary, which does not change our conclusions.

Electronic copy available at: https://ssrn.com/abstract=2001461

where RVi,t is the realized volatility of stock i measured in week t and IVi ,ATM
t is the average

of the implied volatility of the ATM call and ATM put option on stock i in week t. As before,
we employ a weekly average of the IV values to reduce the effect of noise. We require at
least two non-missing values during the past five days to compute the measure.
The third measure is the ATM volatility skew, which relates to the trading activity
of informed investors and jump premia. Bali and Hovakimian (2009) and Cremers and
Weinbaum (2010) argue that more informed trading activity of pessimistic (optimistic)
investors and lower (higher) positively priced jump risk lead to higher (lower) implied
volatilities of ATM put as compared to ATM call options. Stocks with a high ATM volatility
skew should therefore underperform. We take the difference between the implied
volatilities of ATM put and call options as our ATM volatility skew measure:

SKEWi ,ATM
t = IVi ,ATMP
t − IVi ,ATMC
t (3)

where IVi ,ATMP

t denotes the implied volatility of the ATM put option on stock i in week t. As

before, we take the weekly average of the IV variables to reduce the effect of noise, and we
require at least two non-missing values during the past five days to compute the measure.
The fourth measure is the change in the ATM volatility skew, which is thought to
reflect the change in informed trading. For example, Cremers and Weinbaum (2010) argue
that an increase in the informed trading activity of pessimistic (optimistic) investors is
likely to result in an increasing (decreasing) spread between the implied volatilities of ATM
put and call options, which therefore should predict lower (higher) stock returns. Since we
focus on a weekly frequency, we compute the change in the ATM volatility skew variable as
the weekly change in the volatility spread between the ATM put and call options:3

∆SKEWi ,ATM
t = SKEWi ,ATM
t − SKEWi ,ATM
t −1 (4)

3 One may argue that our measures are biased towards highly volatile stocks. Therefore, we additionally
investigate a relative (instead of absolute) definition of these variables. This does not affect our conclusions
reported in the following sections, albeit the results become slightly weaker.

Electronic copy available at: https://ssrn.com/abstract=2001461

 A remark seems to be in order here about the above option measures. We prefer to use
a consistent methodology for all option measures. As a consequence, our definitions of the
measures differ slightly from the ones employed by previous studies, due to the use of
different option screens, the use of weekly averages to reduce the effect of noise, or our
focus on weekly investment frequencies. For example, besides the use of different option
screens in the studies, our measures for RVIV and SKEWATM differ from the measures
proposed by Bali and Hovakimian (2009), who employ an average of all near-the-money call
and put options at the end of a month. Similarly, our measures for SKEWATM and
∆SKEWATM differ from the ones put forward by Cremers and Weinbaum (2010), who employ
the open-interest weighted average difference between the implied volatilities of the call
and put options across several option pairs with the same strike and maturity, and its daily
change. These choices are however unlikely to be important, since our definitions capture
the same economic ideas as the original measures proposed by the aforementioned studies.

2.3 Descriptive statistics

Panel A of Table 1 presents the coverage statistics of the option measures averaged
for each week in our sample (Average), at the start of our sample (Start), and at the end of
our sample (End).4 The following points emerge. First, the coverage for the SKEWOTM is the
lowest, with on average 619 out of the 1,250 stocks covered, as the availability of out-of-the-
money option data is lower compared to at-the-money option data. Second, the coverage
increases over our sample for all option measures. Third, the coverage is already
substantial at the start of our sample period, witnessing coverage between 317 stocks (for
SKEWOTM) and 625 stocks (for RVIV). Fourth, when we investigate a strategy that
combines the option variables, we require at least one option variable to be available, which
leads to an average coverage of 901 stocks, starting with 690 stocks at the beginning and
1045 stocks at the end of our sample period.5

[Insert Table 1 about here]

4 The Start and End figures are very much in line with the lowest and highest coverage percentiles.
5 Note that the requirement of all four variables having data available would lead to a lower coverage of on
average 546 stocks. This additional requirement however does not alter our conclusions. Results are available
from the authors upon request.

Electronic copy available at: https://ssrn.com/abstract=2001461

 Panel B of Table 1 reports the descriptive statistics for the stocks and options
included in our universe. Shown are the time-series averages of the cross-sectional means,
standard deviations and quartiles. The first two rows of the table report the market
capitalizations and book-to-market ratios of the companies included in our universe for
which option information is available. As expected, the firms in our universe are large
compared to the average firm listed on NYSE/AMEX/NASDAQ. The average (median)
market capitalization equals USD 9.19 (USD 3.33) billion, compared to average (median)
values of USD 2.19 (USD 0.17) billion for all stocks on NYSE/AMEX/NASDAQ and USD
7.53 (USD 2.13) billion for all stocks on the NYSE/AMEX/NASDAQ with information
available in OptionMetrics over the same time period. The next rows describe our option
measures. The SKEWOTM is 4.72% on average, the RVIV is on average negative (-0.82%),
the SKEWATM is on average a positive 0.79% and the ∆SKEWATM is on average 0.00% per
week. All these measures display substantial variability over the cross-section, witnessing
average standard deviations between 2.08% (SKEWATM) and 10.67% (RVIV).
Panel C of Table 1 reports the average rank correlations between our four option
price variables. All variables have low to moderate correlations with each other. Most
notably, high values of SKEWATM tend to coincide with high past week ∆SKEWATM
witnessing their positive correlation of 53%. Similarly, SKEWATM values tend to correlate
with SKEWOTM values (43%), as also found recently by Doran and Krieger (2010).6 These
numbers indicate there might be some information overlap in the option variables, to which
we will pay special care in the analyses reported in Section 4.2. The overlap, however, does
not affect our conclusions concerning the exploitability of publicly available option market
information by practitioners.

3 Methodology
To evaluate the information contained in option prices we employ the following
procedure. Every Tuesday, we measure each variable given the latest close information

6 Doran and Krieger (2010) note that SKEWOTM consists partly of the SKEWATM and argue the use of
DKSKEWi OTM
,t = IVi ,OTMP
t − IVi ,ATMP
t
to capture crash worries. We have also investigated their measure, which does

not affect subsequent conclusions about the relevance of publicly available option market information for equity
investors. Section 4.2 discusses the results in more detail.

Electronic copy available at: https://ssrn.com/abstract=2001461

from the option market. 7 Subsequently, we sort firms on each variable and form five
quintile portfolios, from quintile 1 (Q1) to quintile 5 (Q5). Q1 (Q5) contains the stocks with
the lowest (highest) variable value, for which we expect the highest (lowest) return. Of
these portfolios we compute the equally-weighted return over the following week, since we
have no micro or small caps in our sample (which are generally hard to invest in and which
are therefore less interesting for practitioners), and since equally-weighted portfolios are
generally better diversified on a sample of large cap stocks.8 We stress however that the
conclusions of our analyses do not change when we compute value-weighted returns, as
reported in the Appendix.9
In our empirical analyses we include a one day lag between the strategy signals and
portfolio construction. Therefore, stocks are bought at Wednesday close prices. This
implementation lag allows for sufficient time to implement the portfolios, even for less
technologically advanced investors, and avoids spurious findings caused by non-
synchronous trading between options and stocks due to slightly different closing times of
the exchanges (see Battalio and Schultz (2006)). We rebalance these portfolios every week,
and calculate their returns in excess of the risk-free rate and their outperformance relative
to the equally-weighted market portfolio. Moreover, we compute the performance of a long-
short portfolio, computed as the return difference between the top (Q1) and bottom quintile
(Q5).
Subsequently, we control for market, size, value and momentum exposures by
correcting the portfolio returns for the market return, the Fama and French (1992, 1993)
size (SMB) and book-to-market (HML), and the Carhart (1997) momentum (UMD) factors
(retrieved from Kenneth French’s website). The resulting alpha, or estimated abnormal
return, of portfolio j is the constant αj in the regression

7 Our results are not driven by the use of Tuesday close information to construct our variables. When we
compute our results for other days of the week we obtain comparable results.
8 Moreover, for investors who manage long-short portfolios or active portfolios against a benchmark, there is no
need to take active positions in line with the market cap weight of stocks, as this would imply that the absolute
risk-adjusted expected return on large-cap stocks is higher than on small-cap stocks.
9 In unreported analyses we also adjust the raw variables for their industry medians to avoid unintended
industry bets (e.g. caused by negative worries about entire industries). The performances (available upon
request) are in general comparable to the ones reported in the current version, but sometimes at lower
volatility.

10

Electronic copy available at: https://ssrn.com/abstract=2001461

 r j = α j + β j ⋅ rm + s j ⋅ SMB + h j ⋅ HML + u j ⋅ UMD + ε j (5)

where rj is the excess return of portfolio j, rm the excess return on the market portfolio and
βj, sj, hj and uj the estimated factor exposures. In addition, we compute CAPM alphas by
only including the excess market return in (5).10
Furthermore, we conduct Fama and MacBeth (1973) regressions to examine the
predictive power of the variables while controlling for other return predicting variables.
This procedure first estimates each week a cross-sectional regression of stock returns on
predicting variables to obtain estimated effects (slope coefficients) of the tested variables.
To ensure comparability across variables and limit the influence of outliers, we standardize
each variable each week using the approach described in detail in the next paragraph. In
the second stage, the slope coefficients are averaged over time and their t-statistics are
calculated. We correct the t-statistics for heteroskedasticity and autocorrelation by applying
Newey-West (1987) standard errors.
After studying the predictive power of each of the four individual variables, we test
their joint profitability (before and after transaction costs), in an aggregate option
information strategy. To this end, we transform the values of each individual variable into
a cross-sectionally standardized score (z-score) that is comparable across variables. More
specifically, the z-score of a variable is constructed by subtracting its cross-sectional median
from the values of the variable and dividing by its median absolute deviation. We use the
median and median absolute deviation instead of the mean and standard deviation to limit
the influence of outliers. The effect of outliers is further reduced by winsorizing the z-scores
at values of ±3. Subsequently, we obtain the combined z-score as the simple, naïve average
of all variables 11 and rank the stocks into quintile portfolios following the same
methodology we used for each individual variable.

10 Moreover, we have also computed three factor alphas that correct for market, size and value exposures, and
five factor alphas that correct for market, size, value, momentum and short-term reversal exposures. These
results are omitted from the tables and main text for brevity, but the results are in line with the CAPM and four
factor alphas. Results are available from the authors upon request.
11 We require at least one option variable to be available for a stock to compute the average. In addition, when a
stock has no coverage on a particular variable we assign a zero, neutral z-score to that variable.

11

Electronic copy available at: https://ssrn.com/abstract=2001461

4 Results
This section presents our results on the relevance of public information contained in
option prices. We first consider the portfolio sorts based on each variable in isolation, then
we show the results of Fama and MacBeth (1973) cross-sectional firm-level regressions that
control for other factors that may impact returns, followed by the results of a simple
combined option information strategy. Next, we analyze the performance over different
market conditions, and we examine the interaction between the combined option
information strategy and information uncertainty. Finally, we examine the profitability of
our option information strategy when incorporating transaction costs, to determine whether
our findings are caused by non-exploitable return patterns.

4.1 Individual portfolios sorts

Table 2 reports the results of sorting each of our individual option variables in five
portfolios and computing the subsequent weekly returns. For each variable we show the
average annualized excess returns and Sharpe ratios for the quintiles. For both the long-
short and quintile portfolios we additionally show the outperformances relative to the
general equity market and information ratios as defined by the ratio of outperformance and
the volatility of outperformance. Additionally, we present the alphas relative to the market
model (CAPM alpha) and the four-factor model (4F alpha). The asterisks *, ** and ***
indicate the significance of the outperformance and alphas at a 10%, 5% and 1%
significance level respectively. Quintile 1 (Q1) contains the stocks which rank lowest on a
measure, while quintile 5 (Q5) contains the stocks which rank highest on a measure. Hence,
Q1 is composed of stocks with the lowest OTM volatility skew, the lowest realized versus
implied volatility spread, the lowest ATM volatility skew and the lowest change in the ATM
volatility skew, and therefore we expect Q1 to generate higher returns than Q5. Figure 1
shows the cumulative performance of the long-short portfolios through time.

[Insert Table 2 about here]

[Insert Figure 1 about here]

12

Electronic copy available at: https://ssrn.com/abstract=2001461

 Panel A of Table 2 reveals that stocks with the lowest (highest) SKEWOTM, i.e. the
lowest (highest) crash worries, typically experience the highest (lowest) returns. Quintile
portfolio 1 generates on average 5.46% a year over the risk-free rate (3.23% over the
market), decreasing to a negative -1.87% (-3.96%) for quintile portfolio 5. These outcomes
result in a 7.48% return a year for the Q1-Q5 long-short portfolio. These results align with
the results found by Xing et al. (2010) on a different universe between 1996 and 2005, and
are highly economically and statistically significant. The resulting 4F alpha (CAPM alpha),
or abnormal return, is 7.96% (7.49%). Hence, market beta, size, value and momentum
effects cannot explain the observed return spread. Figure 1 shows that the performance is
economically strong in the early years of our sample, weaker in the middle years of our
sample, but economically strong again during the post Xing et al. (2010) sample period,
thereby providing out-of-sample evidence of their idea on an investable universe.
As shown in Panel B of Table 2, stocks with the lowest (highest) RVIV, i.e. the lowest
(highest) volatility risk, typically experience the highest (lowest) returns. Quintile portfolio
1 generates on average an excess return of 5.02% a year and an outperformance over the
market of 1.06%. The returns decrease to an excess return of –2.36% and outperformance of
-6.04% for quintile portfolio 5. These outcomes result in an economically and statistically
significant return of 7.56% a year for the long-short portfolio, which is the highest of the
four strategies investigated. The resulting 4F and CAPM alphas are a highly significant
7.06% and 9.13% respectively, indicating that market, size, value and momentum effects
can only explain a small fraction of the observed return spread. Our findings confirm the
ideas of Bali and Hovakimian (2009) on an investable universe that extends the sample
beyond December 2004. The top line in Figure 1 shows that the performance is strongest
during their sample period, but also positive in the subsequent years.12
Panel C of Table 2 reveals that stocks with the lowest (highest) SKEWATM, i.e. the
least pessimistic informed trading or highest jump risk premia, generally experience the
highest (lowest) returns. Quintile portfolio 1 generates on average 6.83% a year over the
risk-free rate (2.81% over the market), decreasing to 0.42% (-3.37%) for quintile portfolio 5.
These outcomes result in a highly economically and statistically significant return of 6.40%
a year for the Q1-Q5 portfolio. As for the SKEWOTM and RVIV the returns on this portfolio

12 Unlike the other variables, the result of this variable substantially improves when applying an industry
adjustment, especially during the recent crisis years.

13

Electronic copy available at: https://ssrn.com/abstract=2001461

cannot be explained by market, size, value and momentum exposures, witnessing a 4F
alpha (CAPM alpha) of 7.96% (6.51%). These results extend the ideas of Bali and
Hovakimian (2009) and Cremers and Weinbaum (2010) to a well-investable universe.
However, Cremers and Weinbaum show that results deteriorate over the later part of their
sample period. Figure 1 shows that the performance is economically strong in the earliest
years of our sample, weak around the burst of the tech bubble, but economically strong
again in the more recent years not considered by the Bali and Hovakimian and Cremers
and Weinbaum studies.
Finally, Panel D of Table 2 contains the results of our fourth variable, the
∆SKEWATM, which proxies the change in informed trading. We may expect an increase
(decrease) in the informed trading activity of optimistic (pessimistic) investors to lead to
higher subsequent returns. Indeed, we find that quintile portfolio 1 generates higher
returns than quintile portfolio 5 (5.55% versus 0.11%). This results in an economically and
statistically significant return of 5.45% a year for the Q1-Q5 long-short portfolio. As for the
previous variables, the returns on this portfolio are not explained by market, size, value
and momentum exposures, witnessing a 4F alpha (CAPM alpha) of 5.52% (5.69%). Figure 1
shows that the performance is flat in the early years of our sample, but steady and
economically positive since 1999, including the post Cremers and Weinbaum (2010) sample
period.
The results reported in this sub-section have shown that the option strategies deliver
significant results when applied to an investable universe of large-cap stocks. Apart from
RVIV, the strategies show a stable performance in the recent crisis period. Furthermore,
Figure 1 shows that the return patterns of the four variables are quite different from each
other, indicating potential diversification benefits when combined in one option strategy. A
correlation analysis confirms this, as the correlations (not presented in tabular form)
between the weekly Q1-Q5 quintile portfolio returns are low to moderate, ranging between -
3% (between RVIV and SKEWOTM) and 44% (between SKEWATM and SKEWOTM). .

4.2 Firm-level regressions

The above results reveal that forming portfolios based on each of the four option
price variables would have generated significant profits. To control for heretofore
unincluded variables that may impact returns, and to test for the option price variables’
joint significance we continue our analysis with Fama and MacBeth (1973) regressions.

14

Electronic copy available at: https://ssrn.com/abstract=2001461

 [Insert Table 3 about here]

Table 3 reports the results in terms of estimated standardized coefficients and their
significance. All coefficients are annualized by multiplying them with 52, so that the values
represent the annualized return changes caused by a one standard deviation shock in the
underlying variables. For the sake of brevity, we largely focus on models 6-10, which
include the control variables logarithm of the market capitalization (ME), book-to-market
(BM), 9 minus 1 month momentum (MoM), market beta (Beta) and short-term reversal
(STR) (although the same picture emerges from models 1-5). We first consider the firm-level
regressions using each variable in isolation. The results confirm our earlier portfolio
results. Models 6-9 show that in isolation, each option variable displays statistically
significant negative predictive power for subsequent weekly stock returns after correcting
for control variables. In economic terms the coefficients also imply strong predictive power,
with a three-standard deviation change in the variables (roughly comparable to the
difference between the top minus bottom portfolio) resulting in a 4.11% to 5.64% change in
annualized returns. Moreover, the regression results also reveal that the option strategies
are substantially different from the other well-known stock selection strategies:
momentum, reversal, beta, size and value. This may not come as a full surprise, given that
the motivation of the option variables we study (i.e., volatility risk, jump risk, and informed
trading) differ fundamentally from the motivation of these traditional stock-selection
strategies (i.e., herding, under- and overreaction, (distress) risk and under- or
overvaluation).
Next, we consider the effect of all option price variables jointly. Model 5 of Table 3
reveals that all option variables jointly contain predictive information for stock returns,
witnessing a significant R2 of 2.6%. Similarly, the R2 of model 10 (which includes all control
variables) is 13.7%, compared to R2’s of 11.8% to 12.3% for models 6-9, where only a single
option measure is considered. The individual variables RVIV, SKEWATM and ∆SKEWATM
remain significant and of roughly similar magnitude as in models 6-9. By contrast, the
coefficient of SKEWOTM becomes insignificant and positive. This result may be caused by
the correlation between the SKEWATM and SKEWOTM measures. In fact, Doran and Krieger
(2010) note that the SKEWOTM of Xing et al. (2010) consists partly of the SKEWATM and

15

Electronic copy available at: https://ssrn.com/abstract=2001461

argue for using DKSKEWi OTM
,t = IVi OTMP
,t − IVi ,ATMP
t = SKEWi OTM
,t − SKEWi ,ATM
t instead to capture

worries about negative price movements. Interestingly, their findings reveal that higher
values of the latter measure result in higher instead of lower returns.
To examine this effect in more detail, we rerun the individual portfolio sort and
Fama-McBeth regressions using DKSKEWOTM, which has a correlation of 75% with
SKEWOTM. The unreported results from these analyses are qualitatively similar as those
reported in Table 3. Additionally, we do not confirm the positive relation between values of
the DKSKEWOTM and subsequent returns, as documented by Doran and Krieger (2010). In
fact, we find a negative (but still insignificant) coefficient in both the univariate as well as
multivariate Fama-McBeth regressions, and an insignificant long-short return (4F alpha) of
-2.65% (-2.04%) using a portfolio sort.
To conclude, the Fama-McBeth regressions confirm our statement that information
contained in publicly available option prices predicts returns on the underlying stocks, even
after controlling for a set of other return predicting variables.

4.3 Combined option information strategy

The above results reveal predictive power of the option variables for stock returns over
the subsequent week. From an investor’s perspective, the joint value of the option market
information variables is especially interesting. Therefore, we aggregate the option variables
into a simple combined option information strategy, using the procedure outlined in Section
3.

[Insert Table 4 about here]

Even though not all option variables are individually significant in the multivariate
Fama-McBeth regressions, we choose to combine all of our four option variables, since all
variables have revealed substantial predictive power in the individual portfolio sorts and
their strategy returns are not highly correlated. 13 Table 4 contains the results. Sorting
stocks based on the combined option information measure leads to large spreads in

13 We endorse the view that a model with higher in-sample risk-return characteristics may be found. However,
to avoid any in-sample optimization we leave this up to the user.

16

Electronic copy available at: https://ssrn.com/abstract=2001461

subsequent weekly returns, larger than observed for each individual option market
variable. Quintile portfolio 1 generates on average 8.09% a year over the risk-free rate
(4.37% over the market), monotonically decreasing to a negative -1.64% (-5.03%) for quintile
portfolio 5. These outcomes result in a highly economically and statistically significant
return of 9.90% a year for the Q1-Q5 long-short portfolio with an information ratio of 1.13.
The 4F and CAPM alphas reveal similar results, with values of 10.06% and 10.09%
respectively. Figure 2 shows the cumulative performance of the long-short portfolio over
time. Apart from a drawdown in 2009, the strategy delivers a strong, increasing stable
outperformance, also in the recent, previously out-of-sample crisis years.14

[Insert Figure 2 about here]

4.4 Robustness Analyses

The above results have shown that the long-short portfolio based on the combined
option information strategy yields economically and statistically significant returns. We
will further analyze the robustness of these findings by examining the portfolio
performance over different market conditions, and by investigating the interaction of the
combined strategy with various measures of information uncertainty.

4.4.1 Performance in different market conditions

To examine the stability of the combined option strategy’s performance in different

market conditions, we split our sample into (i) bull and bear markets based on positive and
negative monthly market returns, and (ii) volatile and calm markets based on the VIX
being above or below its median value. Table 5 reports the results. The performance of the
strategy is strong in both bull and bear markets (although even higher in bear markets)
witnessing top minus bottom returns (4F alphas) of 11.94% (11.82%) in bear versus 8.57%
(9.34%) in bull markets. The results are also stable across volatility regimes, with the
strongest performance in volatile markets. The top minus bottom returns (4F alphas) are
11.34% (10.67%) in volatile markets versus 8.36% (6.19%) in calm markets. These results

14 In addition, we may wonder if the performance of the combined option information strategy improves if we
consider the stocks with stronger signals, for example by employing decile instead of quintile portfolios. We
report these results in the Appendix, which indeed, reveal slightly stronger results in terms of outperformance
and alpha as compared to the quintile results.

17

Electronic copy available at: https://ssrn.com/abstract=2001461

confirm that the strategy is not only working over the whole sample, but also in different
sub-samples and market environments.

[Insert Table 5 about here]

4.4.2 The impact of information uncertainty

Next, we analyze how the profitability of the combined option information strategy
interacts with information uncertainty. In a recent paper, Zhang (2006) argues and finds
that behavioral biases exacerbate if information uncertainty is higher, and that, as a
consequence, momentum profits are stronger among stocks surrounded with high
information uncertainty than among low uncertainty stocks. Information uncertainty may
also be an amplifier of the combined option information strategy. Part of its predictive effect
may be due to the signaling of private information, and the advantage of private
information may be affected by the uncertainty regarding the usefulness and value of
information.

[Insert Table 6 about here]

To investigate whether the results of our combined option information strategy are
indeed affected by information uncertainty, we perform the following analysis. We rank the
stocks with option information available in five quintiles based on three measures for
information uncertainty: market capitalization, past 52-weeks volatility, and the number of
analysts covering the stock. Within each of these groups, we sort the stocks on our
combined option signal and examine the performance of a long-short quintile portfolio. The
results are presented in Table 6. Lower market capitalization, higher volatility and a lower
number of analysts covering a stock indicates higher information uncertainty. The
combined option information strategy yields a significant outperformance and 4F alpha for
all long-short portfolios for all three information uncertainty measures. In addition, the last
column of Table 6 reveals that the profitability of the combined option information strategy
is not significantly stronger for high versus low information uncertainty portfolios. For
example, the long-short combined option information strategy has an outperformance (4F

18

Electronic copy available at: https://ssrn.com/abstract=2001461

alpha) of 13.87% (14.62%) for the quintile of smallest stocks in our sample, versus 12.85%
(13.05%) for the quintile of largest stocks in our sample. These results reveal that
information uncertainty does not amplify the profitability of the combined option variables.

4.5 The impact of transaction costs

The above analyses reveal that public option market information contains valuable
information for equity investors. From an investor’s perspective, it is important to know
whether this conclusion also holds after accounting for transaction costs. In fact, several
recent studies show that many anomalies are not profitable after transaction costs (see for
example Lesmond, Schill and Zhou (2004), and Avramov, Chordia and Goyal (2006)).
Therefore, we next examine the net profitability of our combined option information
strategy.
To this end, we need an estimate of the transaction costs per individual stock.
Commonly, researchers and practitioners employ the model of Keim and Madhavan (1997).
These authors regress total transaction costs (including commissions paid and an estimate
of price impact) for trading NYSE-AMEX stocks during 1991 to 1993 on several
characteristics of the trade and the traded stock. However, as pointed out by De Groot,
Huij, and Zhou (2012), the transaction cost estimates resulting from the Keim and
Madhavan model should be interpreted with caution when applied to the most recent
decades. The reason is that Keim and Madhavan (1997) estimate their model over all stocks
using data between 1991 and 1993, while market microstructures and transaction costs
have substantially altered since. In fact, De Groot et al. find that the Keim and Madhavan
model systematically yields negative cost estimates for a large group of liquid stocks over
our sample period. For example, the median single-tip transaction cost estimates of the
Keim and Madhavan (1997) model for the S&P500 stocks are -9 basis points over our
sample period, substantially lower than the 9bp based on estimates of De Groot et al. This
implies that we would clearly underestimate transaction costs (and hence overestimate the
net return) when applying the Keim and Madhavan transaction cost estimates to our
universe. Furthermore, during our sample period, the transaction cost estimates of De
Groot et al. declined for the S&P500 universe from median single-trip transaction costs of
approximately 12bp in 1996 to a low of 6.5bp in 2008, indicating the general increase in
liquidity over time.

19

Electronic copy available at: https://ssrn.com/abstract=2001461

 Therefore, we follow the procedure proposed by De Groot et al. to estimate each
stock’s transaction costs. They propose to estimate transaction costs by ranking a stock
based on its dollar volume in a given quarter, and applying the transaction cost estimates
for the matching quarterly dollar volume sorted decile portfolio of S&P1500 or S&P500
stocks, as presented in their Table 1, Panel B and Table 2, Panel B, respectively. These
estimates are obtained from Nomura Securities, one of the major brokers in the cash equity
market, and include both estimates of commissions and the price impact of trades. The
assumed trade size for these estimates is one million USD per stock by the end of 2009,
deflated back in time with 10% per annum. Consequently, the estimates are valid for a
sizable strategy.15

[Insert Table 7 about here]

Table 7 presents the results when incorporating transaction costs. The first column
of Table 7 provides the results of the long-short combined option information strategy
applied to our investable universe consisting of the largest 1,250 U.S. stocks. Clearly,
transaction costs can have a dramatic impact on the profitability of the combined option
information strategy, due to a high turnover of 150% single-counted per week (compared to
a maximum possible turnover of 200% per week). As a result, the outperformance drops
from a positive 9.90% per year to a negative -9.88%, suggesting that the strategy is not
exploitable.
Since the turnover of the combined option information strategy is very high, we may
wonder what happens if we only focus on stocks with the lowest transaction costs.
Therefore, we next repeat the same exercise on the extremely liquid universe of the 100
largest stocks in terms of market capitalization, generally the stocks with the highest
liquidity and lowest transaction costs. For this universe, De Groot et al. employ average
single-trip transaction costs of approximately 7bp over our sample period, versus 13bp for
the S&P1500 stocks. The second column of Table 7 shows that the gross outperformance
increases from 9.90% to 13.57% per year when applied to these 100 largest stocks. Hence,

15 For example, a strategy that invests an equal amount in each of the 20% most or least attractively ranked
stocks in our universe (the largest 1,250 U.S. stocks) would be able to employ a capital of 250 million by the end
of 2009 at these transaction costs.

20

Electronic copy available at: https://ssrn.com/abstract=2001461

the gross profitability of the combined option information strategy does not decrease for the
largest and generally most liquid and most followed stocks. At the same time turnover
remains similar, and as a consequence, the impact of transaction costs substantially
reduces from 19.78% to 11.18%, leading to a positive net outperformance of 2.39% a year.
Although still sizable in economic terms, this number is not statistically significant.
Still, the above analysis deals relatively naively with transaction costs. In fact,
portfolio optimization theory prescribes to efficiently trade-off the decay in predictive power
versus reduction in transaction costs, in order to maximize expected net risk-adjusted
performance. We apply this principle by using a simple turnover reducing portfolio
construction approach to the above two universes. More specifically, at the beginning of our
sample period we form a long-short portfolio based on the option strategy signal. Once
included in the long (short) portfolio, a stock is held until the day that it is ranked as the
most unattractive (attractive), that is, it has fallen to the bottom (for long positions) or top
(for short positions) quintile. On the day a stock falls out of the portfolio, it is replaced by
the most attractive (for long positions) or least attractive (for short positions) ranked stock
not yet included in the portfolio. Hence, trades only occur the moment that a stock migrates
from the most (least) attractive 20% stocks to the least (most) attractive 20% stocks. This
not only limits the turnover of the portfolio to stocks that are expected to move strongly into
the adverse direction, but also ensures that the number of stocks in the portfolio is equal to
the long-short portfolio in the previous analysis.
The results of such a dynamic portfolio are presented in the last two columns of
Table 7. We first examine this approach in our investable universe of 1,250 U.S. large-cap
stocks. Gross outperformance deteriorates marginally, as the performance for the 1,250
largest stocks is 9.84% annually (9.90% for the static long-short portfolio). However,
turnover decreases by more than one-third from 150% to 97%, leading to a substantial
reduction in transaction costs from 19.78% to 14.02%. Still, the net outperformance of the
strategy is negative (-4.18%). Next, we repeat the same exercise on the extremely liquid
universe of the 100 largest stocks (see column 4 of Table 7). Again, gross outperformance
deteriorates marginally, from 13.57% previously to 12.95%. Simultaneously, the impact of
transaction costs reduces to a relatively small 5.34%, which leads to an economically and
statistically significant outperformance of 7.61% per year and a 4F alpha of 7.53%. Hence,

21

Electronic copy available at: https://ssrn.com/abstract=2001461

significant outperformance remains when accounting for transaction costs by means of a
turnover reducing approach on a low transaction cost universe.16
In the above analysis we investigate a long-short strategy, and hence may also face
shorting costs. The net returns of the above long-short strategy are large enough to cover
realistic shorting costs of 50bp to 100bp. As a comparison, D’Avolio (2002) estimates the
shorting costs to average 60bp for a much broader universe. In addition, an investor can
reduce, or even avoid shorting costs, by taking active positions based on the combined
option information strategy against a certain benchmark index, as commonly done by many
institutional equity managers.
To summarize, although the gross return of the combined option information
strategy is high, its turnover is also substantial. As a consequence, the impact of
transaction costs can substantially diminish its net profitability. However, large significant
net returns can be obtained when avoiding stocks that have high transaction costs and
employing turnover reducing portfolio construction rules. This leads us to conclude that
publicly available information contained in option prices can be profitably exploited by
investors.

5 Conclusion
We show that publicly available information extracted from traded equity options
contains valuable information for future stock returns. Trading strategies based on worries
about negative price movements (i.e. out-of-the-money volatility skew), volatility risk (i.e.
realized versus implied volatility spread), informed trading and jump risk (i.e. at-the-money
volatility skew), and the change in informed trading (i.e. change in the at-the-money
volatility skew) yield significant returns and alphas. The performances remain significant
after correcting for market, size, value, momentum, reversal and other return predicting
factors. Hence, we find that the option information strategies are substantially different
from other well-known stock selection strategies. These findings extend the results of

16 We may wonder how sensitive these results are to our particular choice of rebalancing rules. Comfortably,
when we sell (buy) a long (short) position when a stock has fallen to the bottom (top) 50%, 60% or 70% percentile
(instead of the 80% percentile), we also find a substantial reduction in turnover to 142%, 118% and 95%
respectively, and significant, positive net outperformance (4F alpha) of 11.12% (11.62%), 9.44% (10.47%), and
9.18% (9.14%) respectively when applied to the largest 100 stocks.

22

Electronic copy available at: https://ssrn.com/abstract=2001461

earlier studies to a well-investable universe of liquid U.S. large caps, a universe highly
relevant for equity investors. Moreover, a combined option information strategy shows even
stronger results with an annualized performance of around 10%, thereby strengthening the
relevance of the publicly available information contained in option prices for equity
investors. Although several studies report that the predictive power of option market
variables decreases over time, we find significant returns also in recent out-of-sample
years. These results are robust over bull, bear, volatile and calm markets and are generally
of similar magnitude for stocks with low or high information uncertainty. In addition, the
documented anomalies are at least as strong when applied to the 100 largest stocks. We
further find that the profitability of the combined option information strategy can be
dramatically reduced by transaction costs, since exploiting the option information measures
requires extremely high turnover. However, the strategy remains profitable when focusing
on a low transaction cost universe and employing simple procedures to reduce transaction
costs, as annual net returns above 7% can be achieved. This suggests that information
diffuses gradually from the equity option market into the underlying stock market.
We have documented that the four option variables have been a strong predictor of
individual stock returns that seems of relevance for practitioners. Interestingly, the option
information strategies differ substantially from other well-known stock-selection strategies,
like momentum and value, strategies commonly applied by practitioners. Finally, we may
raise the question if one can expect these effects to persist going forward? Although time
has to ultimately answer this question, several points may be worth considering here. First,
we provide out-of-sample evidence on the predictive power of the four option variables,
suggesting that their profitability is unlikely to be caused by data mining. Second, the
explanations of the anomalies (proposed by the original studies) are volatility and jump risk
compensation, and information trading. Although these risks may materialize at some
points, their premia tend to be structural components of the economic system. The latter
may also be argued for the presence of private information. If some investors indeed will
possess private information in the future, the question becomes if and when they would be
willing to express their views in the option market (such that price discovery will take place
in that market). The model of Easley, O’Hara and Srinivas (1998), which provides a
theoretical framework for understanding where informed investors will trade, may be
useful here. In their model, informed traders who want to maximize profits, choose to trade
in the option market if the leverage or liquidity in the options is high, if the liquidity in the

23

Electronic copy available at: https://ssrn.com/abstract=2001461

stock is low, or if there are already many informed investors in the stock market. This
suggests that the option prices that will continue to signal private information are the ones
that have high leverage or liquidity and of which the stock has low liquidity or a large
informed investor base. Third, these anomalies may have been less well-known compared to
for example value and momentum strategies, but their documentation may promote a
greater awareness. This may result in the entrance of new investors in these anomalies,
which could potentially decrease their profitability.

---

We thank two anonymous referees, the editor, Sjoerd van Bekkum, David Blitz and Henk
Grootveld for extremely helpful comments. Any remaining errors are our own.

24

Electronic copy available at: https://ssrn.com/abstract=2001461

References
Avramov, D., T. Chordia, and A. Goyal, 2006, Liquidity and autocorrelations in individual
stock returns, Journal of Finance 61, 2365-2394.
Bakshi, G., and N. Kapadia, 2003a, Delta-hedged gains and the negative volatility risk
premium, Review of Financial Studies 16, 527-566.
Bakshi, G., and N. Kapadia, 2003b, Volatility risk premium embedded in individual equity
options, Journal of Derivatives 16, 527-566.
Bali, T., and A. Hovakimian, 2009, Volatility spreads and expected stock returns,
Management Science 55, 1797-1812.
Battalio, R, and P. Schultz, 2006, Options and the bubble. Journal of Finance 61, 2071-
2102.
Black, F., 1975, Fact and fantasy in the use of options, Financial Analysts Journal 31, 36-
41.
Buraschi, A., and J. Jackwerth, 2001, The price of a smile: Hedging and spanning in option
markets, Review of Financial Studies 14, 495-527.
Carhart, M., 1997, On persistence in mutual fund performance, Journal of Finance 52, 57-
82.
Chakravarty, S., H. Gulen, and S. Mayhew, 2004, Informed trading in stock and option
markets, Journal of Finance 59, 1235-1257.
Cox, J. C., S. A. Ross, and M. Rubinstein, 1979, Option pricing: A simplified approach,
Journal of Financial Economics 7, 229-263.
Cremers, K. J. M., and D. Weinbaum, 2010, Deviations from put-call parity and stock
return predictability, Journal of Financial and Quantitative Analysis 42, 335-367.
D’Avolio, G., 2002, The market for borrowing stock, Journal of Financial Economics 66,
271–306.
De Groot, W., J. Huij, and W. Zhou, 2012, Another look at trading costs and short-term
reversal profits, Journal of Banking and Finance, 36, 371-382.
Doran, J. S., and K. Krieger, 2010, Implications for asset returns in the implied volatility
skew, Financial Analyst Journal 66, 65-76.
Easley, D., M. O'Hara, and P.S. Srinivas, 1998, Option volume and stock prices: evidence on
where informed traders trade, Journal of Finance 53, 431-465.

25

Electronic copy available at: https://ssrn.com/abstract=2001461

Fama, E. F., and K. R. French, 1992, The cross-section of expected stock returns, Journal of
Finance 47, 427-465.
Fama, E. F., and K. R. French, 1993, Common risk factors in the returns on stocks and
bonds, Journal of Financial Economics 33, 3-56.
Fama, E. F., and J. D. MacBeth, 1973, Risk, return, and equilibrium: Empirical tests,
Journal of Political Economy 81, 607-636.
Gârleanu, N., L. H. Pedersen, and A. Poteshman, 2009, Demand-based option pricing,
Review of Financial Studies, 22, 4259-4299.
Goyal, A., and A. Saretto, 2009, Cross-section of option returns and volatility, Journal of
Financial Economics, 94, 310-326.
Hong, H., and J. C. Stein, 1999, A unified theory of underreaction, momentum trading and,
overreaction in asset markets, Journal of Finance 54, 2143-2184.
Hong, H., W. Torous, and R. Valkanov, 2007, Do industries lead stock markets?, Journal of
Financial Economics 83, 367-396.
Keim, D. B., and A. Madhavan, 1997, Transaction costs and investment style: An inter-
exchange analysis of institutional equity trades, Journal of Financial Economics 46,
265-292.
Lesmond, D. A., M. J., Schill, and C. Zhou, 2004, The illusory nature of momentum profits,
Journal of Financial Economics 71, 349-380.
Newey, W., and K. West, 1987, A simple positive semi-definite, heteroskedasticity and
autocorrelation consistent covariance matrix, Econometrica 29, 229-256.
Xing, Y., X. Zhang, and R. Zhao, 2010, What does individual option volatility smirk tell us
about future equity returns?, Journal of Financial and Quantitative Analysis 45, 641-
662.
Zhang, F. X., 2006, Information uncertainty and stock returns, Journal of Finance 61, 105-
137.

26

Electronic copy available at: https://ssrn.com/abstract=2001461

Table 1: Descriptive statistics
The table shows coverage, summary statistics and rank correlations for our universe of U.S. large-
cap stocks for the period January 1996 to October 2009. Data are obtained from FactSet and
Compustat (for stocks) and OptionMetrics (for options). Panel A shows the number of stocks in our
universe (of 1,250 stocks) with coverage for each of the variables, averaged over time and at the start
and end of the sample period. The last column shows the number of stocks with at least one option
variable available. Panel B contains summary statistics. Shown are the mean, cross-sectional
standard deviation, 25% percentile, the median, and 75% percentile of the market capitalization
(ME), book-to-market (BM), and the four variables, the OTM skew (SKEWOTM), the volatility spread
(RVIV) (the difference between the realized volatility and implied volatility of ATM call options), the
ATM skew (SKEWATM), and the weekly change in the ATM skew (∆SKEWATM). Panel C contains
Spearman’s rank correlations between the four variables we employ. We first calculate the summary
statistics or rank correlations over the cross-section and then average the resulting numbers over
the time-series.

Panel A: Stock coverage

Time period SKEWOTM RVIV SKEWATM ∆SKEWATM All
Average 619 842 795 708 901
Start 317 625 494 387 690
End 865 1012 999 924 1045
Panel B: Summary statistics
Characteristic Mean Stdev 25% 50% 75%
ME ($ bln) 9.19 16.80 1.62 3.33 8.66
BM 0.37 0.24 0.20 0.33 0.50
SKEWOTM (%) 4.72 3.47 2.50 4.23 6.41
RVIV (%) -0.82 10.67 -7.54 -1.84 4.56
SKEWATM (%) 0.79 2.08 -0.31 0.62 1.74
∆SKEWATM (%) 0.00 2.15 -1.12 0.00 1.12
Panel C: Rank correlations
Variable SKEWOTM RVIV SKEWATM ∆SKEWATM
SKEWOTM 100%
RVIV 2% 100%
SKEWATM 43% 1% 100%
∆SKEWATM 21% 1% 53% 100%

27

Electronic copy available at: https://ssrn.com/abstract=2001461

Table 2: Portfolio returns of individual option market variables
This table shows the portfolio returns of the four option variables for our universe of U.S. large-cap
stocks. We show the results of the quintile portfolios (Q1 to Q5) and the long-short portfolio (Q1-Q5)
that are constructed by sorting in increasing order on SKEWOTM (Panel A), RVIV (Panel B),
SKEWATM (Panel C), and ∆SKEWATM (Panel D). The equally-weighted portfolios are constructed using
Tuesday close information, implemented with a one-day lag and held for one week. We report the
geometric excess return, Sharpe ratio, outperformance, and information ratio of these portfolios on
an annual basis. In addition, we estimate the CAPM and four-factor (4F) alphas of the quintile
portfolios and report the alphas on an annual basis. The sample period is January 1996 to October
2009. Asterisks *, ** and *** indicate significance at 10%, 5% and 1% respectively.

Panel A: SKEWOTM
Statistic Q1 Q2 Q3 Q4 Q5 Q1-
Q1-Q5
Excess Return (%) 5.46 4.18 2.36 0.34 -1.87
Sharpe Ratio 0.21 0.18 0.10 0.01 -0.08
Outperformance (%) 3.23** 1.97* 0.18 -1.79* -3.96** 7.48***
Information Ratio 0.58 0.45 0.05 -0.44 -0.67 0.76
CAPM Alpha (%) 3.38 2.08 0.23 -1.71 -3.86* 7.49***
4F Alpha (%) 2.87* 1.40 -0.01 -2.13 -4.73*** 7.96***
Panel B: RVIV
Statistic Q1 Q2 Q3 Q4 Q5 Q1-
Q1-Q5
Excess Return (%) 5.02 6.56 5.79 3.69 -2.36
Sharpe Ratio 0.22 0.32 0.27 0.16 -0.08
Outperformance (%) 1.06 2.54* 1.81 -0.22 -6.04** 7.56*
Information Ratio 0.16 0.47 0.38 -0.05 -0.54 0.49
CAPM Alpha (%) 2.96 4.47*** 3.68*** 1.57 -4.13 9.13***
4F Alpha (%) 2.02 3.12** 2.52** 0.89 -3.10 7.06**
Panel C: SKEWATM
Statistic Q1 Q2 Q3 Q4 Q5 Q1-
Q1-Q5
Excess Return (%) 6.83 6.97 3.52 1.62 0.42
Sharpe Ratio 0.28 0.31 0.16 0.07 0.02
Outperformance (%) 2.81** 2.95*** -0.38 -2.21** -3.37*** 6.40***
Information Ratio 0.61 0.82 -0.11 -0.67 -0.73 0.91
CAPM Alpha (%) 4.63*** 4.70*** 1.37 -0.48 -1.58 6.51***
4F Alpha (%) 4.93*** 4.25*** 0.59 -1.47 -2.66 7.96***

28

Electronic copy available at: https://ssrn.com/abstract=2001461

Panel D: ∆SKEWATM
Statistic Q1 Q2 Q3 Q4 Q5 Q1-
Q1-Q5
Excess Return (%) 5.55 4.92 4.16 4.05 0.11
Sharpe Ratio 0.22 0.22 0.19 0.18 0.00
Outperformance (%) 1.67 1.06 0.32 0.22 -3.58*** 5.45***
Information Ratio 0.35 0.28 0.08 0.06 -0.74 0.78
CAPM Alpha (%) 3.37* 2.81* 2.05 1.91 -2.08 5.69***
4F Alpha (%) 3.00* 2.21 1.10 1.36 -2.25 5.52***

29

Electronic copy available at: https://ssrn.com/abstract=2001461

Table 3: Firm-
Firm-level regressions
This table shows the results of cross-sectional regressions of weekly excess returns using the Fama-
MacBeth (1973) methodology that controls for firm characteristics based on our universe of U.S.
large-cap stocks from January 1996 to October 2009. We examine the predictability of the SKEWOTM,
RVIV, SKEWATM and ∆SKEWATM while controlling for the logarithm of the market capitalization
(ME), book-to-market (BM), 9 minus 1 month momentum (MoM), market beta (Beta) and short-term
reversal (STR). Raw signals are first transformed into standardized z-scores. The estimated
coefficients are multiplied by 52 to represent the annualized return effects caused by a one standard
deviation increase in the variable. We report the average estimated coefficients, R-squared (R2), and
the average number of cross-sectional stock observations of each model. Asterisks *, ** and ***
indicate significance at 10%, 5% and 1% respectively, based on Newey-West (1987) adjusted
standard errors.

Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Constant 9.15 10.53* 10.43* 10.19* 9.99* 9.50* 11.13** 11.43** 11.27** 9.90*
SKEWOTM -1.93*** -0.58 -1.37*** 0.19
RVIV -1.93* -2.60** -1.68*** -2.06***
SKEWATM -1.74*** -0.87 -1.88*** -1.41**
∆SKEWATM -1.45*** -1.25** -1.40*** -1.07**
log(ME) -0.08 -0.68 -0.98 -0.99 -0.55
BM 0.88 0.92 0.94 0.96 0.74
MoM 1.47 1.39 1.52 1.39 1.38
Beta -2.29 -1.96 -2.38 -2.26 -1.76
STR -3.63*** -3.30*** -3.42*** -3.82*** -3.33***
R2 0.5% 1.1% 0.2% 0.3% 2.6% 12.3% 12.0% 11.8% 12.1% 13.7%
N 619 842 795 708 546 607 825 780 697 536

30

Electronic copy available at: https://ssrn.com/abstract=2001461

Table 4:
4: Portfolio returns of the combined option information strategy
This table shows the portfolio returns of the option information strategy based on
SKEWOTM, RVIV, SKEWATM and ∆SKEWATM for our universe of U.S. large-cap stocks. We
show the results of the quintile portfolios (Q1 to Q5) and the long-short portfolio (Q1-Q5).
Individual variables are standardized using a cross-sectional z-score. The stocks are then
sorted in increasing order into portfolios based on the average of the four individual z-
scores. The portfolios are constructed using Tuesday close information, implemented with a
one-day lag and held for one week. We report the equally-weighted geometric excess return,
Sharpe ratio, outperformance and information ratio of these portfolios on an annual basis.
In addition, we estimate the CAPM and four-factor (4F) alphas of the quintile portfolios and
report the alphas on an annual basis. The sample period is January 1996 to October 2009.
Asterisks *, ** and *** indicate significance at 10%, 5% and 1% respectively.

Statistic Q1 Q2 Q3 Q4 Q5 Q1-
Q1-Q5
Excess Return (%) 8.09 6.17 3.18 1.85 -1.64
Sharpe Ratio 0.34 0.29 0.15 0.08 -0.06
Outperformance (%) 4.37*** 2.52** -0.38 -1.66* -5.03*** 9.90***
Information Ratio 0.92 0.64 -0.11 -0.49 -0.86 1.13
CAPM Alpha (%) 5.79*** 4.01*** 1.10 -0.24 -3.71* 10.09***
4F Alpha (%) 5.38*** 2.78** 0.20 -0.90 -4.03** 10.06***

31

Electronic copy available at: https://ssrn.com/abstract=2001461

Table 5: Performance in different market conditions
This table shows the performance of the long-short portfolio of the combined option information
strategy for our universe of U.S. large-cap stocks over bull (positive monthly return) and bear
(negative monthly return) markets, as well as volatile (VIX above sample median) and calm (VIX
below sample median) markets. The equally-weighted portfolios are constructed using Tuesday close
information, implemented with a one-day lag and held for one week. We report the geometric
outperformance and information ratio of these portfolios on an annual basis. In addition, we
estimate the CAPM and four-factor (4F) alphas of the portfolio and report the alphas on an annual
basis. The sample period is January 1996 to October 2009. Asterisks *, ** and *** indicate
significance at 10%, 5% and 1% respectively.

Statistic Bull Bear Volatile Calm

Outperformance (%) 8.57*** 11.94*** 11.34*** 8.36***
Information Ratio 1.20 1.32 1.15 1.57
CAPM Alpha (%) 8.81*** 11.25*** 10.33*** 6.20***
4F Alpha (%) 9.34*** 11.82*** 10.67*** 6.19***

32

Electronic copy available at: https://ssrn.com/abstract=2001461

Table 6: Double sorts on information uncertainty measures
This table shows the performance of our combined option signal within different segments of the
market, based on a conditional double sort. First, we construct five groups based on market
capitalization, past 52 week volatility or the number of analysts. Second, within each group, we sort
the stocks on our combined option signal and calculate the returns of a long-short quintile portfolio.
These equally-weighted portfolios are constructed using Tuesday close information, implemented
with a one-day lag and held for one week. We report the geometric outperformance, information
ratio, CAPM alpha and 4F alpha of these portfolios on an annual basis. In the last column we report
the same statistics when we compare the long-short portfolio returns between the high and low
groups based on market capitalization, volatility or analyst coverage. The sample period is January
1996 to October 2009. Asterisks *, ** and *** indicate significance at 10%, 5% and 1% respectively.

Panel A: Market capitalization

Statistic Low 2nd 3rd 4th High HML
Outperformance (%) 13.87*** 8.10** 9.96*** 8.27*** 12.85*** -0.87
Information Ratio 0.93 0.56 0.72 0.65 1.00 -0.06
CAPM alpha (%) 14.22*** 8.97*** 10.55*** 9.03*** 13.01*** 0.49
4F alpha (%) 14.62*** 10.48*** 9.78*** 8.83*** 13.05*** 0.21
Panel B: Volatility
Statistic Low 2nd 3rd 4th High HML
Outperformance (%) 4.83*** 5.78*** 11.41*** 13.44*** 10.90** 5.76
Information Ratio 0.71 0.65 1.08 0.99 0.59 0.32
CAPM alpha (%) 5.03*** 5.91*** 11.51*** 13.23*** 12.05** 7.06
4F alpha (%) 4.96*** 6.07*** 11.67*** 14.98*** 11.00** 6.09
Panel C: Number of analysts
Statistic Low 2nd 3rd 4th High HML
Outperformance (%) 12.11*** 10.29*** 12.01*** 7.13* 11.66*** -0.41
Information Ratio 0.82 0.72 0.91 0.51 0.78 -0.02
CAPM alpha (%) 12.79*** 10.81*** 12.57*** 8.61*** 12.08*** 1.49
4F alpha (%) 14.6*** 9.25*** 12.85*** 7.63** 11.64*** -0.81

33

Electronic copy available at: https://ssrn.com/abstract=2001461

Table 7: Gross and net performance of strategies
strategies based on the option information
information
This table presents the gross and net outperformance of the long-short portfolio based on the
combined option strategies within the universe of the 1,250 and 100 largest U.S. stocks. The
portfolios are equally-weighted, constructed using Tuesday close information and implemented with
a one day lag. The static portfolios only contain stocks which are ranked among the top (bottom) 20%
of the universe and are rebalanced every week. The dynamic portfolios are constructed using an
approach that does not directly sell (buy back) stocks that are no longer in the top (bottom) quintile,
but waits until the day that these stocks are ranked among the bottom (top) 20% of stocks. Reported
are the annualized geometric (net) outperformance, (net) information ratio, weekly single-counted
turnover (TO) and the average holding period (HP) in days. In addition, we present the annualized
net CAPM and four-factor (4F) alphas of the long-short portfolios. Transaction cost estimates used
for calculating net outperformance are based on Panel B of Table 1 and 2 presented in De Groot,
Huij and Zhou (2012). The sample period is January 1996 to October 2009. Asterisks *, ** and ***
indicate significance at 10%, 5% and 1% respectively.

Static Portfolios Dynamic Portfolios

Statistic 1,250 Large 100 Large 1,250 Large 100 Large
Outperformance (%) 9.90*** 13.57*** 9.84*** 12.95***
Information Ratio 1.13 0.95 1.07 0.92
TO (%) 150 150 97 70
HP (days) 6.7 6.7 10.3 14.3
Net Outperformance (%) -9.88 2.39 -4.18 7.61**
Net Information Ratio -1.09 0.17 -0.45 0.54
Net CAPM Alpha (%) -9.69 3.48 -3.53 8.53***
Net 4F Alpha (%) -9.87 3.08 -4.00 7.53***

34

Electronic copy available at: https://ssrn.com/abstract=2001461

Figure 1: Cumulative performance of individual option market variables
This figure shows the cumulative (out)performance of the long-short strategy based on
SKEWOTM, RVIV, SKEWATM or ∆SKEWATM for our universe of U.S. large-cap stocks. Our
sample runs from January 1996 to October 2009. The equally-weighted portfolios are
implemented with a one-day lag and held for one week.

35

Electronic copy available at: https://ssrn.com/abstract=2001461

Figure 2: Cumulative performance of the combined option information strategy
This figure shows the cumulative (out)performance of the long-short combined option information
strategy for our universe of U.S. large-cap stocks. Our sample runs from January 1996 to October
2009. The equally-weighted portfolio is implemented with a one-day lag and held for one week.

36

Electronic copy available at: https://ssrn.com/abstract=2001461

Appendix
Table A1:
A1: Value-
Value-weighted portfolio returns of individual and combined option market
variables
This table shows the portfolio returns of the individual and combined option variables for our
universe of 1,250 U.S. large-cap stocks. We show the results of the quintile portfolios (Q1 to Q5) and
the long-short portfolio (Q1-Q5) that are constructed by sorting in increasing order on SKEWOTM
(Panel A), RVIV (Panel B), SKEWATM (Panel C), ∆SKEWATM (Panel D) and four combined option
variables (Panel E). The value-weighted portfolios are constructed using Tuesday close information,
implemented with a one-day lag and held for one week. We report the geometric excess return,
Sharpe ratio, outperformance, and information ratio of these portfolios on an annual basis. In
addition, we estimate the CAPM and four-factor (4F) alphas of the quintile portfolios and report the
alphas on an annual basis. The sample period is January 1996 to October 2009. Asterisks *, ** and
*** indicate significance at 10%, 5% and 1% respectively.

Panel A: SKEWOTM
Statistic Q1 Q2 Q3 Q4 Q5 Q1-
Q1-Q5
Excess Return (%) 7.82 3.72 0.21 1.44 -2.84
Sharpe Ratio 0.33 0.17 0.01 0.06 -0.11
Outperformance (%) 5.75*** 1.72 -1.72 -0.52 -4.71* 10.98***
Information Ratio 0.65 0.28 -0.34 -0.09 -0.48 0.75
CAPM Alpha (%) 5.25*** 1.39 -2.16 -0.85 -5.12*** 11.24***
4F Alpha (%) 5.25*** 1.54 -1.85 -0.09 -5.04*** 11.28***
Panel B: RVIV
Statistic Q1 Q2 Q3 Q4 Q5 Q1-
Q1-Q5
Excess Return (%) 7.89 5.78 2.32 1.38 -3.18
Sharpe Ratio 0.34 0.29 0.12 0.06 -0.11
Outperformance (%) 5.22* 3.15* -0.22 -1.15 -5.59 11.45**
Information Ratio 0.51 0.49 -0.04 -0.21 -0.41 0.60
CAPM Alpha (%) 5.45** 3.33** 0.12 -0.98 -5.38** 12.87***
4F Alpha (%) 4.95* 2.84 0.17 -0.39 -3.20 10.18**
Panel C: SKEWATM
Statistic Q1 Q2 Q3 Q4 Q5 Q1-
Q1-Q5
Excess Return (%) 5.8 4.64 0.34 0.13 0.28
Sharpe Ratio 0.24 0.22 0.02 0.01 0.01
Outperformance (%) 3.17 2.04 -2.16* -2.36 -2.22 5.51*
Information Ratio 0.39 0.41 -0.45 -0.43 -0.24 0.45
CAPM Alpha (%) 3.40 2.19 -2.01 -2.20 -1.93 6.48*
4F Alpha (%) 4.91*** 2.85** -1.71 -2.61 -3.22 9.58***

37

Electronic copy available at: https://ssrn.com/abstract=2001461

Panel D: ∆SKEWATM
Statistic Q1 Q2 Q3 Q4 Q5 Q1-
Q1-Q5
Excess Return (%) 3.44 3.06 2.57 1.80 -2.76
Sharpe Ratio 0.14 0.14 0.12 0.08 -0.11
Outperformance (%) 1.13 0.76 0.28 -0.47 -4.94** 6.39**
Information Ratio 0.14 0.14 0.06 -0.09 -0.58 0.56
CAPM Alpha (%) 1.11 0.72 0.20 -0.57 -5.19*** 7.17***
4F Alpha (%) 1.62 1.53 0.15 -0.51 -5.44*** 8.06***
Panel E: Combined Option Information
Statistic Q1 Q2 Q3 Q4 Q5 Q1-
Q1-Q5
Excess Return (%) 8.45 6.37 1.51 -0.44 -3.79
Sharpe Ratio 0.37 0.31 0.07 -0.02 -0.14
Outperformance (%) 5.83*** 3.80*** -0.94 -2.85* -6.12** 12.73***
Information Ratio 0.78 0.68 -0.19 -0.51 -0.59 0.90
CAPM Alpha (%) 5.91*** 3.85*** -0.81 -2.80* -6.17*** 13.50***
4F Alpha (%) 6.69*** 3.37*** -0.60 -2.04 -5.79*** 14.10***

38

Electronic copy available at: https://ssrn.com/abstract=2001461

 Table A2:
A2: Decile portfolio returns of combined option information strategy
This table shows the portfolio returns of the combined option variables for our universe of 1,250 U.S.
large-cap stocks. We show the results of the decile portfolios (D1 to D10) and the long-short portfolio
(D1-D10) that are constructed by sorting in increasing order on the combined option variables. The
equally-weighted portfolios are constructed using Tuesday close information, implemented with a
one-day lag and held for one week. We report the geometric excess return, Sharpe ratio,
outperformance, and information ratio of these portfolios on an annual basis. In addition, we
estimate the CAPM and four-factor (4F) alphas of the quintile portfolios and report the alphas on an
annual basis. The sample period is January 1996 to October 2009. Asterisks *, ** and *** indicate
significance at 10%, 5% and 1% respectively.

Combined Option Information

Statistic D1 D2 D3 D4 D5 D6 D7 D8 D9 D10 D1-
D1-D10
Excess Return (%) 7.92 8.10 7.34 4.92 1.84 4.45 1.82 1.76 0.82 -4.24
Sharpe Ratio 0.31 0.35 0.33 0.23 0.08 0.20 0.08 0.07 0.03 -0.15
Outperformance (%) 4.20** 4.37*** 3.64*** 1.30 -1.67 0.85 -1.69 -1.75 -2.66* -7.53*** 12.69***
Information Ratio 0.61 0.80 0.69 0.27 -0.36 0.18 -0.36 -0.37 -0.47 -0.88 1.13
CAPM Alpha (%) 5.31** 5.37*** 4.72*** 2.42 -0.63 1.94 -0.66 -0.68 -1.51 -6.57*** 12.98***
4F Alpha (%) 5.21*** 4.64*** 3.57** 1.12 -1.65 1.15 -1.53 -1.15 -1.95 -6.77*** 13.20***

39

Electronic copy available at: https://ssrn.com/abstract=2001461