SSRN Id4592419

Hedge Fund Option Usage and Skewness Risk
Premium*
Shuaiyu Chen Shuaiqi Li
Abstract
We study how hedge fund option usage can aect the skewness risk
premium in the cross-section of individual stock options. We nd that stocks
with more of their hedge fund holders employing the long naked put strategy
(long put option without the underlying stock) have more positive returns of
their skewness assets comprised of options, whose payo (price) resembles
the realized (risk-neutral) skewness of the underlying stock return. We
document evidence consistent with a price-pressure channel: Those stocks
face larger demand on their out-of-the-money put options, which makes
their risk-neutral skewness more negative and lowers the price of skewness
asset; In the meanwhile, the long naked put positions of hedge funds cannot
predict the realized skewness. The channel works through the idiosyncratic
rather than systematic component of skewness. Other hedge fund option
strategies cannot robustly predict skewness asset returns.
Keywords: skewness risk premium, option, hedge fund.
* We are grateful for the comments from George Aragon, Steven Heston, Christopher Jones, Gang
Li (discussant), and conference participants at the 9th Hong Kong Joint Finance Research Conference.
All errors are our own.
Mitchell E. Daniels, Jr. School of Business, Purdue University, chen4144@purdue.edu
Department of Economics and Finance, City University of Hong Kong, shuaiqli@cityu.edu.hk.
Electronic copy available at: https://ssrn.com/abstract=4592419

1 Introduction
Classic portfolio theory assumes that investors only care about the mean and variance
of stock returns, but the question of whether higher moments are priced has been of
long-standing interest to nancial economists. Early studies such as Rubinstein (1973),
Kraus and Litzenberger (1976), and Harvey and Siddique (2000) develop theoretical
frameworks in which only systematic skewness carries a negative risk premium as it
cannot be hedged away, while recent theories (see Brunnermeier, Gollier, and Parker
(2007), Barberis and Huang (2008), and Mitton and Vorkink (2007)) show that total
skewness, including the idiosyncratic skewness, plays a critical role because investors
with skewness preference may choose to hold under-diversied portfolios.
The empirical pursuit of measuring skewness premium is divided into two ap-
proaches. One strand of the literature uses stocks as the testing asset: They sort
stocks into portfolios based on future skewness estimated from some statistical models,
1
and then compare the stock portfolio returns. . The other approach is to construct a
skewness asset that typically involves a long position on out-of-the-money (OTM) call
options and a short position on OTM put options. The asset has a payo resembling
2
or replicating the realized skewness of the underlying stock return. Therefore, the
average return of this asset is a direct measure of the skewness risk premium.
In this paper, we study how hedge fund (HF hereafter) usage of individual stock
options can aect the skewness risk premium, measured with options, in the cross-
section. There are ve reasons why this design can be important: First, the leverage
1 See Harvey and Siddique (2000), Boyer, Mitton, and Vorkink (2010), Conrad, Dittmar, and Ghysels
(2013), Chang, Christoersen, and Jacobs (2013), Amaya, Christoersen, Jacobs, and Vasquez (2015),
and Langlois (2020)
2 See Bali and Murray (2013), Kozhan, Neuberger, and Schneider (2013), Schneider and Trojani
(2019), Pederzoli (2017), and Orªowski, Schneider, and Trojani (2023).

embedded in option contracts makes them particularly attractive to those investors
with lottery preference and leverage constraint, and oers an ideal arena to measure
skewness risk premium (Boyer and Vorkink (2014), and Frazzini and Pedersen (2022));
Second, HF is commonly viewed as an important user of options. Since option demand
pressure can signicantly aect option prices (Garleanu, Pedersen, and Poteshman
(2008), and Muravyev (2016)), the option usage of HF can potentially aect the price
level of skewness asset and hence the magnitude of skewness risk premium; Third, HF
is supposed to be important marginal investor in pricing the skewness risk, given the
ample evidence in the extant literature documenting that HF uses options to exploit
tail risk, which is closely related to skewness risk. For example, Agarwal, Ruenzi, and
Weigert (2017) nd that HFs change their option positions to time the systematic tail
risk prior to the 2008 nancial crisis. Gao, Gao, and Song (2018) document evidence
consistent with some HFs exploiting market-level ex-ante disaster concerns by selling
insurance; Forth, the individual stock option positions of HFs are supposed to represent
their preference for idiosyncratic, rather than systematic, skewness. Otherwise, they
should use the relatively more liquid index options. This can oer new evidence on
whether idiosyncratic skewness is priced; Fifth, the nature of dierent option strategies
could reveal the speculative demand of HF for dierent sides of the distribution. For
example, long put (call) strategy should signal the desire for exposure to the left (right)
side of the distribution. This helps us understand that HF option demand aects
skewness premium through which side of the distribution.
HF option demand may aect the return of skewness asset via two opposing chan-
nels: price-pressure channel versus informed-trading channel. Take the long naked put
position as an example: In a price-pressure channel, HF demand for put option will
make put more expensive relative to call and thus lower the price of skewness asset

due to its short position on put option. This positively predicts the return of skewness
asset. In an informed-trading channel, HFs take the long naked put position because
they anticipate large downward movements. If they have skills in timing crashes at the
individual-stock-level, their naked put position will negatively predict future skewness,
which is analogous to the payo of skewness asset, and thus negatively predict the
return of skewness asset.
We utilize the individual stock option positions of HF disclosed in the quarterly
SEC 13F llings. Since HFs are only required to disclose their long positions, we focus
on strategies involving long option positions and ignore those involving short positions
due to the data constraint. We classify option positions into four dierent types of
strategies: long naked put, protective put, straddle/strangle, and long call. Due to the
fact that HFs are not required to report the details of their option contracts such as
market value, strikes, and maturities, we follow Aragon and Martin (2012) in using the
proportion of a stock's HF holders disclosing a certain option strategy on the underlying
stock to proxy for hedge fund option demand. This measure is at the stock-quarter level.
In each month, we match the monthly skewness asset return with the option demand
proxy in the previous quarter, and run Fama-MacBeth regressions to predict returns.
We construct the skewness asset following Bali and Murray (2013).
Our main nding is that stocks with a larger proportion of HF holders employing
the long naked put tend to have higher returns on their skewness assets. This nding
supports the price-pressure but not informed-trading channel. The positive relation
remains robust after controlling for a wide range of rm characteristics and persists for
only one quarter. This lines up with our nding that the majority of HF option positions
last up to one quarter. The relatively short holding period suggests that HFs use the
naked put to bet on short-term information and that the price pressure should be fairly

short-lived. A long-short strategy formed on the naked put usage generates a monthly
return of 1.53% and an annual Sharpe ratio of 0.68. The strategy remains protable
after considering reasonable option bid-ask spreads. The return spread is insignicant
(largest) among the subset of stocks with the lowest (highest) option-hedging costs,
consistent with the price-pressure channel in which option demand pressure aects
option price through the frictions from market makers.
To further disentangle the price-pressure channel from the informed-trading chan-
nel, we separately investigate how the long naked put position predicts the risk-neutral
and realized skewness, which closely resemble the price and payo of the skewness asset,
respectively. The skewness asset in Bali and Murray (2013) is delta- and vega-neutral
just at the inception of portfolio formation. Even though this static hedge minimizes
transaction cost and enhances tradeability, large stock price and volatility movements
after the formation date may reduce the hedging performance and the resemblance
between asset payo and realized skewness. As a robustness check, we construct risk-
neutral skewness following Orªowski, Schneider, and Trojani (2023) (OST hereafter).
Unlike the conventional method in Bakshi, Kapadia, and Madan (2003) (BKM here-
after), OST dene skewness in a way such that we can form a linear portfolio of options,
whose price equals the risk-neutral skewness, to generate a payo equal to the realized
skewness. We use naked put positions to predict the synthetic return of the skewness
asset in OST and nd the same positive predictive power with a similar t-statistic as
using the asset in Bali and Murray (2013).
To investigate how HF naked put usage aects the risk-neutral skewness, we rst
verify that the proxy constructed from long naked put positions is indeed positively
related to option demand pressure. Specically, we nd that market makers face larger
buying pressure on the put options written on the stocks with more HFs using the

naked put. Therefore, the proxy captures not only the extensive margin of HF option
usage, but also the intensive margin. After we split option contracts into OTM and
other moneynesses, we nd that the proxy can only positively predict the demand on
OTM put options that take up a short position in the risk-neutral skewness. This
specic pattern of demand pressure is supposed to make these stocks' put options more
expensive and lower their risk-neutral skewness. Consistent with this conjecture, we
nd that the naked put positions negatively predict cross-sectional risk-neutral skewness
3
constructed in OST. The evidence supports the price-pressure channel in which HF
naked put usage lowers the price of skewness asset and positively predicts skewness
asset return.
Next, we investigate whether HF naked put usage can predict realized skewness.
HFs supposedly use naked put to exploit the large downward movements in future stock
prices, i.e., crash risks. If they have the skill to time the stock-level crashes, their naked
put positions should negatively predict future skewness. If their timing skill were strong
enough, it could outweigh the decrease of risk-neutral skewness in the price-pressure
channel and negatively predict the return of skewness asset. We nd limited evidence
for this hypothesis: Used alone, the coecient of naked put is negative but signicant
only at the 10% level; After controlling for other predictors in the literature, it becomes
insignicant. The evidence contradicts the informed-trading channel and suggests that
HFs using naked put do not possess skills at timing crashes at the individual-stock-level.
In summary, the ndings that HF naked put positions lower risk-neutral skewness, have
little predictive power on realized skewness, and positively predict the return of skewness
asset combined lend supports to the price-pressure channel.
Intuitively, HFs use individual stock options to exploit the idiosyncratic component
3 The negative relation is robust to alternative construction in BKM.

of stock price movements. Consistent with this intuition, after decomposing the total
risk-neutral skewness into idiosyncratic and systematic components, we nd that naked
put usage only lowers the idiosyncratic but not the systematic component. It suggests
that the desire for exposures to idiosyncratic downward stock movements motivates HFs
to long naked put, which makes put options overpriced and then aects the skewness
premium embedded in options. This is consistent with the empirical ndings in Boyer,
Mitton, and Vorkink (2010) and Boyer and Vorkink (2014) that idiosyncratic skewness
is priced and carries a negative expected return. Our nding also complements the HF
literature that typically focuses on the exposure to the market-level tail risk.
Other option strategies, including protective put, straddle/strangle, and long call,
cannot robustly predict skewness asset return. We nd that those option positions fail
to predict the option demand pressure faced by market makers. Therefore, there are no
evidence that their usage imposes enough price pressure to aect option return. Even
though the market value of option position is not reported, HFs report the market value
of the stocks underlying the option contract. In terms of stock value, naked put has
an average more than twice of protective put and straddle/strangle. Thus, it is not
surprising that those two strategies do not create demand pressure as large as naked
put. However, for long call position, it has an average more than twice of naked put. To
explore why it do not translate into large buying pressure for call options, we use rm
characteristics to explain option usage. What we nd is that turnover, which proxies
for dierence of opinions according to Hong and Stein (1999), is the most important
determinant of call usage other than rm size. Its coecient magnitude is twice of short
interest, which is the most important determinant of naked put usage. In comparison,
when we explain naked put usage, the coecient of short interest is more than twice of
turnover. Therefore, it is likely that other investors trade against HFs and absorb the

buying orders on call options from HFs. Another possibility is that covered call (writing
call option and long the underlying stock) is a popular strategy employed by mutual
fund for the purpose of income generation (Cici and Palacios (2015) and Kaniel and
Wang (2022)). This would neutralize the buying pressure from HF to some extent. To
test this possibility, we match the long option positions of HF with the corresponding
4
short option positions of mutual fund at the stock-quarter level. We nd that 24.56% of
the long call positions from HF have at least one mutual fund taking the corresponding
short call positions on the same underlying rm in the same quarter. In contrast, only
9.6% of long naked put positions from HF correspond to the short put positions from
mutual fund. Therefore, the liquidity provided by mutual funds that use covered call
may dampen the price pressure from the HFs that long call. Overall, we nd that HF
speculative demand for the right-side of distribution, proxied by long call positions, is
not strong enough to aect the skewness premium. This is in contrast with long naked
put positions, which proxy for the demand on the left-side of distribution.
We also provide empirical evidence against alternative interpretations. First, the
SEC only requires HFs to report long positions but not short positions. This asymmetric
regulatory requirement raises the concern that the long naked put positions we observe
may be used to hedge the unobservable short positions on put options of HFs. In
this case, long naked put positions of HF are used for risk management, instead of
speculation, purpose, and it would be unsurprising that they cannot predict future
skewness. However, if risk management were the main purpose, the size of those long
naked put positions should be smaller or at most equal to the short positions. In
this case, the net demand of HFs on put options should be negative, and our proxy is
4 In addition to long positions, mutual funds report their short positions to data vendors such as
Morningstar and Lipper.

expected to negatively predict the open interest of put options. This predicted sign is
in sharp contrast with our earlier ndings. If we multiply put open interest by delta
(vega), we nd that the long naked put usage negatively (positively) predicts the delta-
(vega-) adjusted put open interest, suggesting that market makers have to sell put to
clear the market. This further corroborates our earlier ndings and contradicts the risk
management interpretation.
Second, we do several things to tease out the eect of volatility from skewness. The
skewness asset in Bali and Murray (2013) is vega-neutral at the formation day, and we
control for volatility in regressions. In addition, the skewness asset in OST we use as a
robustness check can separate the jump component of skewness from the leverage eect
component. The naked put usage can negatively predict the return of jump component
with highly signicant t-statistic. We also control for variance risk premium and nd
that it barely aects the predictive power of naked put usage on skewness asset return.
Our paper contributes to the literature on the skewness risk premium by connecting
it with the literature studying how HF exploits tail risk. The HF literature focuses on
5
how higher-moment risks can explain HF performance. However, no one examines the
other direction, i.e., how HF activities such as option usage can aect the pricing of
higher-moment risks. To the best of our knowledge, we are the rst to study the role
of HF as a marginal investor in pricing skewness risk.
Aragon and Martin (2012) nd that HF positions of individual stock options predict
the direction and volatility of the underlying stock return, indicating that HFs are
informed about the rst- and second-order moments. We extend the framework to study
the pricing of third-order moment, after noticing the fact that HFs are sophisticated
5 See Agarwal, Bakshi, and Huij (2009), Bali, Brown, and Caglayan (2012), Kelly and Jiang (2012),
and Agarwal, Ruenzi, and Weigert (2017) for example.

enough to utilize options to exploit higher-order moment risks as suggested by the HF
literature and that the activity itself can create option demand pressure and thus aect
the pricing of those risks.
The rest of this paper is organized as follows. Section 2 describes the data used in
our empirical work. Section 3 investigates how HF option positions predict skewness
asset returns. Section 4 presents evidence supporting that the return predictability is
caused by price pressure. Section 5 performs robustness checks. Section 6 concludes.
2 Data
This section describes how we measure HF option demand and skewness asset return.
2.1 Hedge Fund Option Holdings
Our main database of portfolio holdings contains historical Form 13F lings. Since
1978, all institutional investors (including hedge fund investment advisors) who exercise
investment discretion over accounts holding at least $100 million are required by Section
13(f ) of the Exchange Act of 1934 to make quarterly disclosures of portfolio holdings
to the SEC in Form 13F within 45 days of the quarter end. The types of securities that
are required to be disclosed include exchange-traded stocks, ETFs, equity options and
warrants, convertible bonds, and shares of closed-end investment companies. All long
positions in these securities with more than ten thousand shares or with market values
greater than $200,000 must be reported on Form 13F. Short positions, open-end funds,
and private securities do not need to be disclosed. The items in Form 13F include
the issuers of the securities, the type of security, CUSIP, the number of shares, and
the market value of each security held by institutional investors. In the case of option

positions, institutional investors must provide the CUSIP and the dollar amount owned
in terms of the securities underlying the options, not the options themselves. They
are required to report whether options are calls or puts, but are not required to report
the strike price or expiration date of an option. In addition, institutional investors
report aggregated holdings in all individual funds managed by the same management
company.
The Form 13F lings are available for download from the SEC's EDGAR system,
but need considerable further processing due to manual formatting. While the widely
used institutional holdings databases in academic studies (e.g., Thomson Reuters and
Factset) only provide stock holdings, the commercial database WhaleWisdom oers
a complete set of reported 13F positions, including stock, option, and other types
of securities. Therefore, our empirical analysis is based on the 13F holdings data of
WhaleWisdom.
The Form 13F lings of HF investment advisors are identied as Agarwal, Jiang,
Tang, and Yang (2013) and Agarwal, Ruenzi, and Weigert (2017), based on the names
of HF companies that appear in four commercial HF databases: Eurekahedge, Hedge
6
Fund Research (HFR), Morningstar, and Lipper TASS. We match these HF company
names with the names of institutions reported in WhaleWisdom. Our sample consists
of portfolio holdings disclosed by 770 HF advisers from December 2004 to September
7
2020.
6 We thank Vikas Agarwal for providing the list of hedge fund rms. Joenväärä, Kaupila,
Kosowski, and Tolonen (2021) show that the use of multiple hedge fund databases rather than one
helps mitigate data biases. Some hedge fund rms are doing market-making in the option mar-
kets. According to the list of market makers in the derivative markets reported by the Financial
Conduct Authority (see, https://www.fca.org.uk/publication/documents/market-makers-authorised-
primary-dealers.pdf.), we remove six names from the hedge fund list: CREDIT SUISSE, EVNINE &
ASSOCIATES INC, GEMSSTOCK LTD., MITSUBISHI UFJ ASSET MANAGEMENT LTD., MOR-
GAN STANLEY, and TD ASSET MANAGEMENT INC.
7 Hedge fund holdings data in WhaleWisdom start from March 2001. However, due to the consid-
eration detailed in the next section, we pick this specic sample period.
10

Since HFs are only required to disclose their long positions, we focus on strategies
involving long option positions. We classify option positions into six strategies: (1).`Op-
tion' means either long put or call. All option positions fall into this category, and it
is the most general category. (2).`Call' means that the fund only long call without put
on the stock. It is a directional bet on the upside information. (3).`Put' means long
put. In this case, the fund can either long only the put to make a directional bet on the
downside information, or long both put and call on the stock to make a non-directional
bet. The union of (2) and (3) equals (1). (4).`Straddle' means long both put and call on
8
the same underlying stock. It is a non-directional bet on the volatility. (5).`Protective
Put' means long both the put and the underlying stock without the call. HF uses it for
hedging purpose. (6).`Naked Put' means long put without the underlying stock or the
call. It is used for speculative purpose. The union of (4), (5), and (6) equals (3).
Table 1 presents the summary statistics of HF option holdings. In Panel A, there
are 22, 556 quarterly fund portfolios, 5, 110 of which contain at least one option holding.
In those 5, 110 portfolios, there are on average 119 rms on which the fund holds stocks
or options. On average, a fund holds options on 16.61% of the rms in its portfolio.
After we classify option positions into sub-categories, a fund long call (naked put) on
8.72% (5.25%) of the rms in its portfolio. A caveat is that the distributions are right
skewed, with medians much smaller than means.
In Panel B, we rst aggregate the market value (in millions of dollars) of the stocks
underlying the option contracts at the fund-quarter level, and then report the market
value conditioning on that a fund reported at least one option holding in that quarter.
On average, the market value of the stocks underlying all types of option contracts
8 Since funds do not report the strikes, we cannot dierentiate straddle from strangle. So we use
`Straddle' hereafter just for simplicity.
11

equals 98.31 million dollars in a HF portfolio, with long naked put positions contributing
23.42 million dollars.
Panel C reports the number of quarters each option position lasts. It is at the
granular fund-quarter-option level. For example, assume that a fund used options on
Tesla in March and June of 2019. It didn't report naked put in September, but started
to use it on Tesla in December 2019 again. Then we have two observations: One is the
naked put in March and June, and it equals 2 because it lasts for two quarters; The
other observation is the naked put in December, and it equals 1 because it only lasts
for one quarter. In the panel, the medians equal 1 for all categories, suggesting that
HFs use options to bet on relatively short-lived information.
To measure the option demand of HF industry for a given stock-quarter, we use the
proportion of HFs disclosing a certain type of option strategy on the stock, computed
as the number of funds using the strategy divided by the total number of funds holding
the stock or its options. For example, assume that there are 100 HFs holding the stock
or option of Tesla in a given quarter, and that 50 of them use naked put on Tesla.
Then:
50
N aked P ut = = 0.5.
100
Since naked put requires a long position on put options, we expect the measure to
be positively related to the demand pressure created by HF industry on Tesla's put
9
options. We construct the measure for each option strategy and for each stock-quarter.
Then we match this quarterly measure with the monthly skewness asset returns in the
next quarter. Therefore, the option holdings from December 2004 to September 2020
correspond to the returns from January 2005 to December 2020. By construction, the
9 If a stock or its option is not held by any HF in some quarter, the measure is not well-dened. We
follow Aragon and Martin (2012) and exclude those stock-quarters.
12

measure lines in [0, 1]. We pick it because funds do not report the details of their
option contracts. In addition, it is a well-established measure in the literature: Aragon
and Martin (2012) and Aragon, Chen, and Shi (2022) use it to show that HF option
positions predict volatilities and returns on the underlying stock.
2.2 Skewness Asset Return
To construct skewness asset, we use data from the OptionMetrics Ivy DB database,
which provides daily closing bid and ask quotes for the U.S. equity options. We use the
interest rate of appropriate maturity (interpolated when necessary) from OptionMetrics
as the risk-free rate. We obtain information about stock prices, returns, dividends,
trading volume, and market capitalization from CRSP. Accounting data and short
interest are from Compustat.
Following Bali and Murray (2013), the skewness asset comprises of a long position
on OTM call and a short position on OTM put. The OTM put (call) is the contract
with a delta closest to −0.1 (0.1). To isolate the third order movement from the rst
and second order moments, the skewness asset is constructed to be delta- and vega-
10
neutral. Let ∆P (∆C ) be the delta of put (call), and νP (νC ) be the vega of put (call).
The asset consists of 1 contract in the OTM call, −νC /νP contracts in the OTM put,
and −(∆C − νC /νP · ∆P ) shares in the underlying stock. We form the asset on the third
Friday of each month and hold it to the third Friday of the subsequent month, which
is the typical expiration day of monthly options. Therefore, skewness asset return is
10 To further tease out the relation between skewness and volatility, we control for the implied
volatility when predicting skewness asset returns later.
13

measured at monthly frequency as following:
P ayof f − P rice
rSkew = − rf .
|P rice|
Bali and Murray (2013) use the absolute value of the skewness asset price because the
price is not guaranteed to be positive. We construct the price using bid-ask midpoints
of options. rf is the risk-free interest rate.
The skewness asset is designed to increase in value if the skewness increases, and
it represents long skewness positions. If we hold the asset to expiration, it realizes a
high (low) payo when high (low) stock return is realized, but it is largely insensitive
to small stock moves.
We apply a series of lters: (1).Our sample includes only common shares (CRSP
share codes of 10 and 11). (2).We exclude stock-month observations if the underlying
stock price is less than $5 on the formation date or if the stock has a split or pays a
dividend during the remaining life of the option. Thus, the early exercise premium is
small. (3).We further remove option contracts with zero open interest or bid prices.
(4).We also discard options with missing delta or vega. (5).We then delete options
whose ask price is lower than the bid price, and eliminate those whose prices violate
arbitrage bounds. (6).Even though we pick the OTM options with deltas closest to
−0.1 (0.1) with an intention to capture the tail movements, the nal options picked
may not be as OTM as we desire after applying all the lters above. Thus, we include
only the observations with ∆P ≥ −0.25 and ∆C ≤ 0.25.11
Since HFs do not report the details of their option positions, there is no way for us
to compare the strikes of their option holdings with those options in the skewness asset,
11 Removing this lter does not qualitatively aect the return predictability of naked put.
14

and it is very likely that those options do not have the same strike and maturity. This
heterogeneity does not pose a concern on the price-pressure channel in the setting of
option markets, because Garleanu, Pedersen, and Poteshman (2008) show that the price
pressure created by demand imbalance on one option contract could spillover to other
options written on the same stock via correlated unhedgeable risks such as stochastic
volatility. Therefore, HFs can still aect the return of skewness asset even though they
are not holding the exact option contracts in the skewness asset.
We match each stock's skewness asset return with HF usage of the same stock's
options in the previous quarter. Figure 1 plots the number of stocks with positive HF
option demand for the 4 types of option strategies, respectively. To ensure adequate
statistical power and a fair comparison among those strategies, we require a month to
have at least 30 stocks with positive option demand and thus start the sample from
January 2005. The sample ends in December 2020.
Table 2 presents the summary statistics. Our nal sample contains 145, 067 stock-
month observations. rSkew has a mean of −0.77% per month with a standard deviation
of 16.33%. On average, a stock has 28 HF holders. HF option usage is heavily skewed:
Most of the stocks have no HF holders demanding their options, but the 90th percentiles
of most option strategies are positive. This leaves us enough stocks to form a reasonably
diversied decile portfolio.
15

3 Hedge Fund Option Demand and Skewness Asset
Return
This section investigates how dierent HF option strategies predict cross-sectional skew-
ness asset returns. Section 3.1 examines the return predictability using the Fama and
MacBeth (1973) regression. Section 3.2 explores the protability by forming long-short
portfolios. Section 3.3 checks the persistence of return predictability.
3.1 Cross-sectional Regressions
We run the following monthly Fama-MacBeth regression,
Skew
ri,t+1 = αt + γt · Hedge F und Option U sei,t + θt · Controlsi,t + ϵi,t+1 ,
Skew
where: ri,t+1 is stock i's skewness asset return in month t+1. Hedge F und Option U sei,t
is the latest available proportion of HFs disclosing a certain type of option position on
stock i. To ensure the robustness of our regression results, we include a variety of con-
trol variables that are common in the literature, including option return predictors and
stock characteristics. The construction details for all controls used in this paper can
be found in Appendix A. First, we control for the average implied volatility of the op-
tions in the skewness asset, as an eort to tease out the relation between the third and
second- order moments. Second, we control for option liquidity by using the average
percentile bid-ask spread of those options. Third, we include the short interest to make
sure that HF option demand is not a mere proxy for it. The rest variables are those
that have been shown by the literature to be related to the future or risk-neutral skew-
ness, including lagged skewness, turnover, maximum stock return in the past month,
16

beta, log market capitalization, log book-to-market ratio, idiosyncratic volatility, past
stock returns, Amihud illiquidity measure, and rm leverage ratio. We do not include
some important straddle return predictors such as volatility deviation in Goyal and
Saretto (2009), because we nd that they do not predict skewness asset return. This
suggests that skewness asset is constructed to capture higher-moment risk and diers
from volatility asset.
Table 3 reports the time-series averages of the coecients in Fama-MacBeth regres-
sions, together with their t-statistics in parentheses corrected for heteroskedasticity and
autocorrelation, following Newey and West (1986) with three lags. In column (1), the
variable Option positively predict rSkew with a t-statistic of 3.87. When it increases
from 0 to 1, rSkew increases by 5.2% per month. It remains signicant after controls. If
we split Option into Call and Put in column (3), they both signicantly predict rSkew .
However, they are no longer signicant at the conventional level after controls in column
(4). If we further split Put into Naked Put, Protective Put, and Straddle in column
(5), only Naked Put is signicant, indicating that the entire predictive power of put
option positions comes from naked put motivated by the speculative purpose. When it
increases from 0 to 1, rSkew increases by 15% per month. In column (6), we examine
the predictive power of all subcategories together with controls. The only subcategory
signicant at the conventional level is naked put, with a t-statistic of 2.86.
3.2 Portfolio Sort
The previous section documents that Naked Put displays a robust and positive predic-
tive power on skewness asset return. Next, we employ portfolio sort to examine the
performance of the option strategies formed by Naked Put and evaluate the impact of
17

transaction costs. The sample period is from January 2005 to December 2020.
We form long-short portfolio of skewness assets. We long the stocks in the top
decile with highest Naked Put. If the number of stocks with positive Naked Put is less
than 10% of the total number of stocks in a month, we only long those with positive
Naked Put. When we pick the sample period, we require a month to have at least 30
stocks with positive Naked Put. This lter enables us to have a reasonably diversied
portfolio. If the number is larger than 10% in that month, we long those stocks in the
top 10%. As to the short-side, we short all those stocks with zero Naked Put.
We value-weight stocks in the long and short portfolios on the third Friday of each
month based on their option dollar open interest. Since there is no well-established
factor model for skewness assets, we calculate risk-adjusted returns using the Fama
and French (2015) ve factors, stock momentum, the S&P 500 Index skewness return,
and the cross-sectional average of skewness returns, weighted by their option open
12
interests.
Panel A in Table 4 presents the average monthly returns of portfolios formed by
Naked Put. The (%) symbol after a variable means that it is reported as a percentage.
The long-short strategy generates a monthly return of 1.53% and an annualized Sharpe
ratio of 0.68. Its alpha equals 1.42% with a t-statistic of 2.81. A nice feature is that
the strategy has limited crash risk: It has a positive skewness of 5.60 and a minimum
monthly return of −8.83%.
We then analyze the protability over dierent subsamples. Figure 2 plots the ve-
year moving averages of monthly returns (Panel A) and annualized Sharpe ratio (Panel
B). The strategy performance is positive in each ve-year subsample, and performs
especially well during the nancial crisis and COVID period.
12 The risk factors are taken from Kenneth French's website.
18

So far, our results assume that options can be bought and sold at the midpoint of
bid and ask quotes. To evaluate how bid-ask spreads can aect the prots, we consider
eective spreads equal to 25%, 50%, 75%, and 100% of the quoted spreads in Panel
B. The eective spread is twice the dierence between the traded price and midpoint.
An eective-to-quoted spread ratio of 50% is equivalent to paying half of the quoted
spread. The column MidP corresponds to zero eective spread, i.e., options are traded
at midpoints. The column 100% refers to the case in which traders buy options at
ask and sell at bid. Muravyev and Pearson (2020) show that algorithmic traders who
time executions pay 20% of the spread. As an attempt to reduce the impact of bid-ask
spreads, we also look at a sub-sample with only stocks in the short-leg whose percentage
bid-ask spreads are lower than the median of that month..
Panel B in Table 4 presents the average monthly returns of long-short strategies
under dierent spread ratios. When we use all stocks, the strategy performance is
no longer signicant at the conventional level as early as the 25% spread ratio kicks
in. But when we exclude those stocks with higher-than-median option spreads, the
strategy remains protable with a monthly return of 1.25 and a t-statistic of 2.21 under
the spread ratio of 25%. All strategies incur losses if the entire quoted spread must be
paid. Therefore, reducing transaction cost is essential to maintain the protability of
the strategy.
3.3 Persistence of the Return Predictability
This section checks how long the return predictability persists. We use Naked Put with
lags up to 4 quarters to predict rSkew in Fama-MacBeth regressions including controls.
Table 5 shows that only the one-quarter-lagged Naked Put signicantly predicts
19

rSkew with a t-statistic of 2.88. Thus, the predictability only persists for one quarter.
This is consistent with the position-level evidence in Panel C of Table 1 that most of HF
option holdings last for one quarter. The coecients of Naked Put become insignicant
beyond the past quarter but stay positive. This pattern is dierent from the price
pressure eect in the stock market, which is typically followed by a reversal of signs.
The reason is that after the monthly options in skewness asset expire, there cannot be
any price reversal on those contracts. Thus, the price pressure eect in option markets
does not lead to a switch of signs in predicting option returns over dierent horizons.
4 A Price-Pressure Channel
This section proposes a price-pressure channel to explain the predictive power of Naked
Put: When more HFs use naked put on a stock, they create larger demand on the
stock's put options, make them more expensive, and lower the price of the skewness
asset that takes a short position on put option. If naked put positions do not contain
information on future skewness, which closely resembles the payo of skewness asset,
Naked Put will positively predict rSkew .
Section 4.1 conrms that Naked Put is positively related to the buying pressure on
put options and especially the OTM ones. Section 4.2 explores an alternative explana-
tion that HFs use naked put to do risk management rather than speculation. Section
4.3 investigates the determinants of HF option usage at the stock-quarter level. Section
4.4 proposes a potential reason why long call positions do not translate into buying
pressure on call options by looking at how mutual funds trade against HFs in option
markets. Section 4.5 shows that option hedging costs are necessary for HF option usage
to aect option return.
20

4.1 Option Demand Pressure
The proxies constructed for dierent option strategies capture the extensive margin of
HF option usage, but not necessarily the intensive margin. In addition, option demand
from HFs may be absorbed by other investors and thus not translate into the net
demand pressure faced by market makers. This section aims to establish the positive
relation between Naked Put and the net buying pressure on put option.
We use the net signed open interest to measure net buying pressure on option.
We use the data from The International Securities Exchange (ISE) Open/Close Trade
Prole, which records the information on signed option volumes of the open/close and
13
buy/sell orders from public customers, broker/dealer, and proprietary traders. Fol-
lowing the literature, we group together the three types of investors, and their aggregate
option demand is the demand pressure faced by market makers. We adopt the method
in Ni, Pearson, Poteshman, and White (2021) to construct the net signed open interest
for option contract j on trading day t, as follows:
OpenInterestBuy Buy OpenBuy

j,t = OpenInterestj,t−1 + V olumej,t − V olumeCloseSell
j,t ,
OpenSell
OpenInterestSell Sell
j,t = OpenInterestj,t−1 + V olumej,t − V olumeCloseBuy
j,t ,
N etOpenInterestj,t = OpenInterestBuy Sell

j,t − OpenInterestj,t .
Then we aggregate put and call options, respectively, into stock-month-level by sum-
ming up all available puts/calls written on the same stock on the third Friday of each
month, in order to align with the time point when we construct skewness asset.
To account for heterogeneous rm sizes, we scale the net signed open interest by
the shares outstanding of the underlying stock. We then use all option positions to
13 Ge, Lin, and Pearson (2016) provide a detailed description of the dataset.
21

predict the scaled net open interests of put and call, respectively, in Fama-MacBeth
regressions. The control variables largely follow those in previous exercises, except that
we use the average implied volatility of at-the-money (ATM) options instead of those in
the skewness asset as before, because: (1). The dependent variables aggregate the open
interests of all option contracts; (2). ATM options are most sensitive to volatility. Also
due to reason (1), we use the average percentage bid-ask spread of all options written on
the same stock. To facilitate interpretation, we standardize all independent variables to
have a mean of 0 and a standard deviation of 1. Constrained by the availability of ISE
data, the sample period in this section is from January 2006 to May 2016. To enlarge
sample and be general, we do not restrict to the stocks with enough valid options to
construct rSkew in this section.
Table 6 reports the regression results. To ease visualization, we multiply all de-
pendent variables by 103 and omit the coecients of controls. In Panel A, Naked Put
positively predicts the net buying pressure on put options faced by market makers: In
column (1) without controls, it has a t-statistic of 3.44, and a one-standard-deviation
increase will increase the scaled open interest by 0.00058. After including controls in
column (2), the magnitude drops to 0.000328 while remaining highly signicant with
a t-statistic of 3.37. Therefore, when more HFs use naked put, market makers face a
higher buying pressure on put, which is expected to make put more expensive. An-
other interesting observation is that Naked Put negatively predicts the demand for call
options. This is not surprising given that a higher value of Naked Put reects a more
bearish view of HFs. Some HFs would probably sell calls instead of buying puts. Since
HFs do not report their short positions, we cannot provide direct empirical evidence.
Nevertheless, the negative relation between Naked Put and call option demand will also
lower the price of skewness asset that takes a long position on call.
22

Protective Put cannot signicantly predict open interests. The signicance of Strad-
dle varies inconsistently depending on whether we include controls. In terms of under-
lying stock value in Panel B of Table 1, Protective Put and Straddle have an average
less than half of Naked Put. Hence, it is not surprising that these two do not create
demand pressure as large as naked put and cannot predict skewness asset return. For
long call positions, it has an average more than twice of naked put. However, when we
predict the open interest of call option in columns (3) and (4), the coecients of Call
are positive but insignicant at the conventional level, suggesting that HF demand for
call options also do not translate into large buying pressure. We will revisit the poten-
tial reason for this nding in depth in later sections. Call negatively predicts the open
interest of put options in columns (1) and (2), suggesting that some HFs may express
their bullish views by selling put. However, the magnitude is much smaller than the
coecient of Naked Put. This may explain why Call cannot robustly predict skewness
asset return.
We further explore the moneyness of options potentially used by HF. The leverage
of OTM options makes them attractive to speculators, and HFs are likely to use OTM
put options to bet on the large downside movement in stock prices. To check this,
we classify option contracts into OTM and other moneyness (at- and in-the-money)
following Bollen and Whaley (2004): The category OTM Put includes put options with
deltas larger than −0.375; The category ATM+ITM Put includes those with deltas
smaller than −0.375; OTM Call includes call options with deltas smaller than 0.375;
ATM+ITM Call includes those with deltas larger than 0.375. Then we compute open
interests by summing up options in each category.
In Panel B of Table 6, we use HF optoin usage to predict the open interests of
dierent moneynesses. In columns (1) and (2), Naked Put positively predict the open
23

interest of OTM put with highly signicant t-statistics. However, it loses signicance
when we predict the open interest of ATM and ITM put with controls in column
(4). This evidence suggests that HFs using naked put are particularly interested in
purchasing OTM put options. This demand pattern can aect option return especially
given that OTM put is part of the skewness asset.
In summary, we nd that the measure Naked Put is indeed positively related to the
net buying pressure on put options, especially the OTM ones, faced by market makers,
which is a necessary condition for the price-pressure channel. Other option strategies do
not translate into order imbalances, either because HFs do not use them heavily enough
(such as protective put and straddle) or because option demand of HFs is absorbed by
other investors as in the case for long call positions we will show in later sections.
4.2 Risk Management vs. Speculation
Since the SEC only requires HFs to report long positions but not short positions, this
makes it possible that the long put positions we observe may be used to hedge the
unobservable short positions on put options taken by HFs. In this case, long naked put
positions are used for risk management, instead of speculation, purpose.
If risk management were the main purpose, the size of those long naked put positions
is supposed to be smaller or at most equal to the corresponding short positions. In this
case, the net demand from HFs on put options should be negative, and Naked Put is
expected to negatively predict the open interest of put options. This predicted sign is
in sharp contrast with our ndings in the previous section, which contradicts the risk
management story.
Another possibility is that HFs hedge their option positions by delta or vega and
24

the way we compute open interest does not take this into account. To check this, we
compute the delta- (vega-) adjusted open interest: We multiply the open interest of
each option contract by its delta (vega) and then aggregate the adjusted open interest
14
at the stock-month level.
In Panel C of Table 6, we use HF optoin usage to predict the delta- or vega-adjusted
open interests. In columns (1) and (2), Naked Put negatively predict the delta-adjusted
open interest of put option with signicant t-statistics. In columns (5) and (6), Naked
Put signicantly predict the vega-adjusted open interest of put option with a positive
coecient. Since put options have negative delta and positive vega, both evidence
suggest that market makers face a larger buying pressure on put when Naked Put
increases. This contradicts the risk management story that the observed long naked
put positions are used by HFs to hedge some unobservable short positions, and further
corroborates our earlier ndings .
4.3 The Determinants of Hedge Fund Option Usage
This section explores the determinants of HF option usage at the stock-quarter level.
We explain the HF option positions in the next quarter by running quarterly Tobit
regressions, because the dependent variables are censored at 0 and 1. To account for
auto-correlation, we control for the one-quarter-lagged dependent variables. We include
the time-xed eects and cluster standard errors at the quarter level.
Table 7 presents the results. We standardize all independent variables to help com-
parison. The strongest determinant for all option strategies is the rm's log market
capitalization, followed by lag dependent. Other than the two, short interest is the
strongest determinant for Naked Put, suggesting that HFs use naked put as a substi-
14 Delta and vega are taken from OptionMetrics.
25

tute for short-selling. In column (1), a one-standard-deviation increase would increase
naked put usage by 0.021. The strongest determinant for Call is turnover, which proxies
for dierence of opinions according to Hong and Stein (1999). In column (2), a one-
standard-deviation increase in turnover would increase long call usage by 0.012, twice of
the coecient of short interest. In comparison, the magnitude of short interest is more
than twice of turnover in column (1). Hence, compared with long naked put position,
it is much more likely that other investors trade against HFs and absorb the buying
pressure from HFs on call options. Dierence of opinions could be a reason why HF
demand for call options do not translate into the net buying pressure faced by market
makers.
4.4 Mutual Fund Option Holdings
Covered call is a popular strategy employed by mutual funds for the purpose of income
generation (Cici and Palacios (2015)). Their short position on call option could neu-
tralize part of the buying pressure from HFs for the call options written on the same
stock. This could also oer an explanation for why the HF demand for call does not
translate into net buying pressure.
To test this possibility, we match the long option positions of HF with the corre-
sponding short option positions of mutual fund at the stock-quarter level. We extract
15
option holdings of mutual funds from a Morningstar dataset. Since the dataset ends
in June 2015, the sample period in this section is from December 2004 to June 2015. An
advantage of the dataset is that mutual funds report the market value of their option
positions: A positive (negative) value means a long (short) position. However, a chal-
lenge is that mutual funds report option holdings in a nonstandard way: First, unlike
15 We thank David Hunter for sharing this dataset. It is used in Hunter (2015).
26

stock holdings, most of the option holdings reported do not have common identiers
like CUSIP, making it impossible to directly link option positions with their underlying
stocks. Second, the names of underlying stocks are included in the security name item
for option holdings. However, mutual funds abbreviate the underlying stocks' names in
an arbitrary way, and sometimes they use tickers instead of names, making the match-
ing with underlying stocks dicult. Finally, mutual funds do not report the important
characteristics of option contracts other than whether it is a call or put. It is impossible
to tell the strikes and maturities for most of the option positions.
To extract option holdings, we follow the procedures in Cici and Palacios (2015).
We begin by using the security names of holdings as the main input and identify the
16
observations that contain the Call or Put text strings in the names. Next, we use
visual inspection to remove misclassications because some observations contain the
above text strings but are not option holdings, such as Output and Computer. We
link individual stock option positions with the names of underlying stocks. We use a
name-matching algorithm based on spelling distance to match security names with rm
names. Then we use visual inspections to pick the observations in which mutual funds
report tickers instead of rm names and match those security names with rm tickers.
In the last step, we use visual inspections again to lter out mismatches.
After extracting the option holdings of mutual funds, we match the long option
positions of HFs with the corresponding short option positions of mutual funds at the
stock-quarter level. In Table 8, there are 20, 969 long call positions from HFs at the
fund-quarter-stock level. 5, 151 of them can be matched with at least one mutual fund
taking a short call position on the same underlying stock in the same reporting quarter.
16 Mutual funds can be arbitrary in capitalizations and spelling: In addition to Call, they might
write CALL, call, Calloption, etc. The same is true for put options. We also consider these
variants in the search.
27

Therefore, 24.56% of long call positions from HFs have mutual funds trading against
them. In comparison, 12.04% of general long put positions from HF correspond to the
short put positions from mutual funds, a proportion only half of that for long call. If
we further zoom into the long naked put positions from HF, only 9.6% of them have
corresponding short put positions from mutual funds.
The above evidence suggests that the liquidity provided by mutual funds that use
covered call may dampen the price pressure from the HFs that long call. In addition
to dierence of opinions, this could be another explanation why HF demand for call
options do not translate into net buying pressure and predict option return.
4.5 Option Hedging Costs
Garleanu, Pedersen, and Poteshman (2008) show that for option demand imbalance to
aect option return, a necessary condition is the frictions for market makers to hedge
their option positions. If the price-pressure channel were true, the long-short portfolio
sorted by Naked Put should be more protable among the subset of stocks whose
options are more dicult to hedge. We use double-sort to test this: Each month we
rst sort stocks into terciles by a proxy for option hedging cost. Then we further form
long-short portfolios based on Naked Put within each tercile. Specically, we long/short
the skewness assets of all stocks with positive/zero Naked Put and value-weight them
by option dollar open interests.
We consider three forms of hedging costs in Garleanu et al. (2008): stochastic
volatility, jump risk, and delta-hedging cost. Following Cao, Vasquez, Xiao, and Zhan
(2023), we measure stochastic volatility by estimating an EGARCH(1,1) model on the
28

third Friday each month using rolling-one-year daily stock returns as follows:
2 1
rt = σt zt ; lnσt2 = ω + αrt−1
2 2
+ βlnσt−1 + γ[|zt−1 | − ( ) 2 ],
π
where: rt is the stock return; σt is the conditional volatility; zt is the innovation term.
We set the maximum number of iterations at 500, and 97.3% of cases successfully
converge. After we generate a series of time-varying volatilities for each day in the
estimation window, we calculate the standard deviation of the model-implied volatilities
for each rm every month with successful convergence and use it as the proxy for
stochastic volatility. To measure jump risk, we follow Gao, Gao, and Song (2018) and
use a disaster concern index constructed with OTM puts as follows:
Z
ln(St /K)
Jump Risk = 2 P (K, t, T )dK.
K<St K2
The portfolio assigns larger weights to more deeply OTM puts and thus protects in-
vestors against negative price jumps. The larger the index, the greater the downward
jump risk, which means more risks for market makers who write put options. To ap-
proximate the continuous integral, we follow the procedure in Carr and Wu (2009) by
extrapolating the strike range to −5 standard deviations from the current stock price
and generating a ne grid of 500 implied volatility points. Then we transform implied
volatilities into option prices via Black-Scholes formula. To nd a proxy for delta-
hedging cost, we follow Boulatov, Eisdorfer, Goyal, and Zhdanov (2022) and use the
average daily stock bid-ask spread in the previous month.
Table 9 presents the returns (in percentage) of the double-sorted portfolios. For all
three hedging costs, the returns are low and insignicant among the subset of stocks
29

with the lowest costs. In contrast, when the costs become the highest, the prots
are the largest in magnitude and highly signicant. The pattern is consistent with the
price-pressure channel in which Naked Put aects option return because market makers
charge a higher price as a compensation for bearing inventory risk when they write put
options to HFs.
5 Other Empirical Tests
The previous section presents evidence that HF usage of long naked put could lead to
net buying pressure on put options. We hypothesize that this demand pressure will
make put more expensive and lower the risk-neutral skewness, which resembles the
price of skewness asset. Section 4.1 tests this hypothesis using an alternative skewness
asset in OST and also investigates the informed-trading channel. Section 4.2 constructs
the risk-neutral skewness following BKM as a robustness check and decomposes it into
systematic and idiosyncratic components. Section 4.3 examines whether our result is
related to variance risk premium.
5.1 Price Pressure vs. Informed Trading
In a price-pressure channel, HF naked put usage negatively predicts risk-neutral skew-
ness and positively predicts skewness asset return. While in an informed-trading chan-
nel, it negatively predicts both the future realized skewness and the skewness asset re-
turn, if the crash-timing skill of HFs were strong enough to outweigh the price-pressure.
To test the two channels, we construct an alternative skewness asset following OST
because it has the following advantages: First, OST dene realized skewness in such a
way that it is tradeable as the payo of a model-free portfolio of options and stocks.
30

Therefore, the risk-neutral and realized skewness are internally consistent, which oers
a clean economic interpretation; Second, we pick as the payo the jump component of
skewness in OST dened as:
T
Realized, OST
X rt2
Skew = (ert − rt − 1 − ), (1)
t=1
2
3
17
PT
which equals t=1 ( r3! + o(rt4 )) based on Taylor expansion. This conceptual frame-
work enables us to separate jump eect from leverage eect and better tease out volatil-
ity from higher moment; Third, this asset involves a daily rebalancing position of stock
and option to replicate the dened jump skewness as payo. This strengthens the close
resemblance between the price (payo ) of skewness asset and the risk-neutral (realized)
skewness.
The price of the portfolio that can generate the payo in Equation 1 equals:
Z ∞
Implied, OST ln(K/F0 )
Skew = O(K, t, T )dK, (2)
0 K2
where F0 is the forward price at formation day; O(K, t, T ) is the price of an OTM
option. We apply to individual stocks the option lters (1)-(5) in Section 2.2, and then
follow Carr and Wu (2009) in computing the price in Equation (2): We extrapolate
the strike range to ±5 standard deviations from the current stock price, generate a ne
grid of 1000 implied volatility points, and then transform implied volatilities into option
18
prices via Black-Scholes formula.
17 r is the daily log stock return and T is the option expiration date.
t
18 The standard deviation is approximated by the average implied volatility of OTM options. Before
we do the extrapolation, we require a stock-month to have at least three valid options after applying
the option lters, with at least one OTM put and one OTM call.
31

The return of this asset is dened as:
SkewRealized, OST − SkewImplied, OST

rSkew, OST = . (3)
|SkewImplied, OST |
19
There exists extreme outlier due to very small denominator. The estimation of higher
moments using individual stock options could be sensitive to microstructure noise. To
make the empirical test more reliable, we trim observations with |SkewImplied, OST | at
the bottom 5% of each month. Since the option-implied skewness is known at the
formation day, we do not use forward-looking information here.
We use Naked Put to predict cross-sectional rSkew, OST , the price and payo of
skewness asset, respectively. Control variables largely follow those in Table 3. The
asset is dierent from that in Bali and Murray (2013) in that it includes all OTM
options. Therefore, we use as controls the average implied volatility of ATM put and
call as it is the most sensitive to volatility, and the average percentage bid-ask spread
of all OTM options as this better reect the liquidity of this asset.
Table 10 presents the result. In columns (1) and (2), Naked Put positively pre-
dicts the return with highly signicant t-statistics, consistent with the sign in Table 3.
Thus, the positive predictive power of Naked Put is robust to dierent constructions of
skewness asset and supports the price-pressure channel. Next, to dissect the return pre-
dictability, we decompose the return into price and payo, and use them as dependent
variables. In columns (3) and (4), Naked Put negatively predicts risk-neutral skewness,
which is the price of asset. Again, this is consistent with the prediction of price-pressure
channel. When we predict future skewness, the payo of asset, in columns (5) and (6),
Naked Put has negative signs. However, the coecients are insignicant at the conven-
19 The smallest absolute risk-neutral skewness is at the order of 10−8 . When we scale realized
skewness by this number, option return blows up.
32

tional level, and there is very limited evidence supporting the informed-trading channel.
In summary, we nd that HF naked put usage lowers risk-neutral skewness, fails
to predict future skewness, and positively predicts the return of skewness asset. These
evidence combined lends support to the price-pressure channel and contradicts the
informed-trading channel.
5.2 Systematic vs. Idiosyncratic Skewness
As a robustness check for the negative predictability of risk-neutral skewness, we con-
20
struct the risk-neutral skewness following BKM. In addition, we follow BKM and
decompose the skewness into systematic and idiosyncratic components. Intuitively,
HFs use individual stock options to exploit the idiosyncratic component of stock price
movements. If so, naked put usage is expected to lower the idiosyncratic but not the
systematic component.
On the third Friday of each month, we construct the risk-neutral skewness, SkewImplied, T otal ,
using Equation (5) in BKM. BKM assume that risk-neutral stock return process follows
a one-factor market-model:
ri,t = ai + βRN,i rm,t + ϵi,t ,
where ri,t and rm,t are the excess returns of the stock and the market portfolio, respec-
3
20 BKM dene risk-neutral skewness based on the scaled skewness measure: E[(rt −Ert ) ]
. The
E[(rt −Ert )2 ]3/2
option portfolio is nonlinear and nontradeable, and hence does not allow a return decomposition into
risk-neutral and realized skewness as in the previous section. In addition, this measure of realized
skewness can be heavily inuenced by the leverage eect as shown in Neuberger (2012).
33

tively; βRN,i is the risk-neutral beta. The systematic skewness of stock i is as follows:
3 3
βRN,i σm
SkewiImplied, Sys. = Implied
Skewm ,
σi3
Implied
where Skewm is the risk-neutral skewness of the S&P 500 Index. The square of σi
(σm ) is the risk-neutral variance of stock i (the S&P 500 Index) constructed following
Equation (7) in BKM. To estimate βRN,i , we use the methodology in French, Groth, and
Kolari (1983). It incorporates the information in options and is more appropriate than
a rolling-based historical measure for this exercise. We dene idiosyncratic skewness as
the dierence between the total and systematic skewness:
SkewiImplied, Idio. = SkewiImplied, T otal − SkewiImplied, Sys. .
We use Naked Put to predict the three types of skewness in monthly Fama-MacBeth
regressions and standardize all independent variables. Table 11 reports the result.
In column (1) and (2), Naked Put negatively predicts SkewImplied, T otal with highly
signicant t-statistics. The negative sign is consistent with that in columns (3) and (4)
of Table 10, and is robust to alternative constructions of risk-neutral skewness. After
decomposition, we nd that Naked Put negatively predicts SkewImplied, Idio. in columns
(3) and (4) but positively predicts SkewImplied, Sys. in columns (5) and (6), suggesting
that the negative predictability of risk-neutral skewness comes from the desire of HF
for exposure to the idiosyncratic rather than systematic component of stock return.
This is consistent with the empirical ndings in Boyer, Mitton, and Vorkink (2010) and
Boyer and Vorkink (2014) that idiosyncratic skewness is priced.
34

5.3 Variance Risk Premium
Kozhan, Neuberger, and Schneider (2013) nd that skewness risk is closely related to
variance risk at the aggregate market level. This section investigates the possibility
that HF naked put usage aects skewness premium via its inuence on variance risk
premium (VRP). We follow Bollerslev, Tauchen, and Zhou (2009) and dene VRP
as the dierence between implied and realized variance. On the third Friday of each
month, we estimate realized variance using the high-frequency 15-minutes-interval stock
returns in the previous month, and construct a model-free implied variance using the
Volatility Surface database.
We use Naked Put and VRP to predict the cross-sectional return of skewness asset
in Bali and Murray (2013). Table 12 presents the result. Without controls in column
(2), VRP signicantly predicts return. When we include both of them in column (3)
without controls, Naked Put still positively predicts return and its coecient is very
similar to that in column (1). When we include controls in column (4), VRP becomes
insignicant while Naked Put remains highly signicant. The above evidence suggests
that VRP cannot explain the predictive power of naked put usage on skewness premium.
6 Conclusion
While nance literature attempts to estimate skewness risk premium in various ways,
no one explores the role of HFs in pricing skewness risk. According to the rich literature
studying how HFs exploit tail risk, they are supposed to be pivotal. This paper lls the
gap by examining how HF option usage can aect skewness premium. Specically, we
nd that naked put usage positively predicts cross-sectional return of skewness asset.
This predictability is robust to the inclusion of a battery of control variables and to
35

alternative constructions of skewness asset. A long-short strategy generates a monthly
return of 1.53% and an annual Sharpe ratio of 0.68. The protability of this strategy
survives reasonable option bid-ask spreads.
To understand the underlying economic mechanism, we document evidence consis-
tent with a price-pressure channel: (1).Stocks with more HFs employing naked put
on them face larger buying pressure on their put options, especially the OTM ones;
(2).Those stocks tend to have more negative risk-neutral skewness, which corresponds
to a lower price of skewness asset. (3).Naked put usage cannot predict future skewness,
which resembles the payo of skewness asset; (4).Option hedging costs are necessary
for HF naked put usage to aect option return.
After we decompose risk-neutral skewness into the systematic and idiosyncratic
components, we nd that the price pressure from naked put positions mainly exerts
on the idiosyncratic but not systematic component. The nding suggests that the
desire for exposures to idiosyncratic downward stock movements motivates HFs to long
naked put, which makes put options overpriced and then aects the skewness premium
embedded in options. This is consistent with the empirical ndings in the literature that
idiosyncratic skewness is priced and carries a negative expected return. Our nding also
complements the HF literature that typically focuses on the exposure to the market-
level tail risk.
We nd that the usage of other option strategies fails to predict skewness asset
return and do not translate into net buying pressure faced by option market makers.
To explore the reason, we show that HFs do not use protective put and straddle as
heavily as they do for naked put and long call in terms of underlying stock value. For
long call strategy, HFs tend to use it on stocks with large dierence of opinions, which
raises the possibility that other investors trade against them and absorb their buying
36

pressure. In addition, we nd that the popularity of covered call strategy among mutual
funds partially counteracts the demand of HFs on call options.
37

References
Agarwal, V., G. Bakshi, and J. Huij, 2009, Do higher-moment equity risks explain hedge fund re-
turns?, Robert H. Smith school research paper no. RHS, pp. 06153.
Agarwal, V., W. Jiang, Y. Tang, and B. Yang, 2013, Uncovering hedge fund skill from the portfolio
holdings they hide, The Journal of Finance, 68(2), 739783.
Agarwal, V., S. Ruenzi, and F. Weigert, 2017, Tail risk in hedge funds: A unique view from portfolio
holdings, Journal of Financial Economics, 125(3), 610636.
Amaya, D., P. Christoersen, K. Jacobs, and A. Vasquez, 2015, Does realized skewness predict the
cross-section of equity returns?, Journal of Financial Economics, 118(1), 135167.
Amihud, Y., 2002, Illiquidity and stock returns: cross-section and time-series eects, Journal of
nancial markets, 5(1), 3156.
Ang, A., R. J. Hodrick, Y. Xing, and X. Zhang, 2006, The cross-section of volatility and expected
returns, The journal of nance, 61(1), 259299.
Aragon, G. O., S. Chen, and Z. Shi, 2022, Volatility Timing Using ETF Options: Evidence from
Hedge Funds, Available at SSRN 4246146.
Aragon, G. O., and J. S. Martin, 2012, A unique view of hedge fund derivatives usage: Safeguard or
speculation?, Journal of Financial Economics, 105(2), 436456.
Bakshi, G., N. Kapadia, and D. Madan, 2003, Stock return characteristics, skew laws, and the dier-
ential pricing of individual equity options, The Review of Financial Studies, 16(1), 101143.
Bali, T. G., S. J. Brown, and M. O. Caglayan, 2012, Systematic risk and the cross section of hedge
fund returns, Journal of Financial Economics, 106(1), 114131.
Bali, T. G., N. Cakici, and R. F. Whitelaw, 2011, Maxing out: Stocks as lotteries and the cross-section
of expected returns, Journal of nancial economics, 99(2), 427446.
Bali, T. G., and S. Murray, 2013, Does risk-neutral skewness predict the cross-section of equity option
portfolio returns?, Journal of Financial and Quantitative Analysis, 48(4), 11451171.
38

Barberis, N., and M. Huang, 2008, Stocks as lotteries: The implications of probability weighting for
security prices, American Economic Review, 98(5), 20662100.
Bollen, N. P., and R. E. Whaley, 2004, Does net buying pressure aect the shape of implied volatility
functions?, The Journal of Finance, 59(2), 711753.
Bollerslev, T., G. Tauchen, and H. Zhou, 2009, Expected stock returns and variance risk premia,
The Review of Financial Studies, 22(11), 44634492.
Boulatov, A., A. Eisdorfer, A. Goyal, and A. Zhdanov, 2022, Limited attention and option prices,
Swiss Finance Institute Research Paper, (20-64).
Boyer, B., T. Mitton, and K. Vorkink, 2010, Expected idiosyncratic skewness, The Review of Finan-
cial Studies, 23(1), 169202.
Boyer, B. H., and K. Vorkink, 2014, Stock options as lotteries, The Journal of Finance, 69(4),
14851527.
Brunnermeier, M. K., C. Gollier, and J. A. Parker, 2007, Optimal beliefs, asset prices, and the
preference for skewed returns, American Economic Review, 97(2), 159165.
Cao, J. J., A. Vasquez, X. Xiao, and X. E. Zhan, 2023, Why Does Volatility Uncertainty Predict
Equity Option Returns?, The Quarterly Journal of Finance, 13(01), 2350005.
Carr, P., and L. Wu, 2009, Variance risk premiums, The Review of Financial Studies, 22(3), 1311
1341.
Chang, B. Y., P. Christoersen, and K. Jacobs, 2013, Market skewness risk and the cross section of
stock returns, Journal of Financial Economics, 107(1), 4668.
Cici, G., and L.-F. Palacios, 2015, On the use of options by mutual funds: Do they know what they
are doing?, Journal of Banking & Finance, 50, 157168.
Conrad, J., R. F. Dittmar, and E. Ghysels, 2013, Ex ante skewness and expected stock returns, The
Journal of Finance, 68(1), 85124.
39

Fama, E. F., and K. R. French, 1992, The cross-section of expected stock returns, the Journal of
Finance, 47(2), 427465.
, 2015, A ve-factor asset pricing model, Journal of nancial economics, 116(1), 122.
Fama, E. F., and J. D. MacBeth, 1973, Risk, return, and equilibrium: Empirical tests, Journal of
political economy, 81(3), 607636.
Frazzini, A., and L. H. Pedersen, 2022, Embedded leverage, The Review of Asset Pricing Studies,
12(1), 152.
French, D. W., J. C. Groth, and J. W. Kolari, 1983, Current investor expectations and better betas,
The Journal of Portfolio Management, 10(1), 1217.
Gao, G. P., P. Gao, and Z. Song, 2018, Do hedge funds exploit rare disaster concerns?, The Review
of Financial Studies, 31(7), 26502692.
Garleanu, N., L. H. Pedersen, and A. M. Poteshman, 2008, Demand-based option pricing, The Review
Ge, L., T.-C. Lin, and N. D. Pearson, 2016, Why does the option to stock volume ratio predict stock
returns?, Journal of Financial Economics, 120(3), 601622.
Goyal, A., and A. Saretto, 2009, Cross-section of option returns and volatility, Journal of Financial
Economics, 94(2), 310326.
Harvey, C. R., and A. Siddique, 2000, Conditional skewness in asset pricing tests, The Journal of
nance, 55(3), 12631295.
Hong, H., and J. C. Stein, 1999, Dierences of opinion, rational arbitrage and market crashes, .
Hunter, D. L., 2015, Mind the Gap: The Portfolio Eects of 'Other' Holdings, Asian Finance Asso-
ciation (AsianFA) 2015 Conference Paper, Available at SSRN: https://ssrn.com/abstract=2565274
or http://dx.doi.org/10.2139/ssrn.2565274.
Jegadeesh, N., and S. Titman, 1993, Returns to buying winners and selling losers: Implications for
stock market eciency, The Journal of nance, 48(1), 6591.
40

Joenväärä, J., M. Kaupila, R. Kosowski, and P. Tolonen, 2021, Hedge fund performance: Are stylized
facts sensitive to which database one uses?, Critical Finance Review, 10, 170.
Kaniel, R., and P. Wang, 2022, Unmasking mutual fund derivative use, Available at SSRN 3692838.
Kelly, B. T., and H. Jiang, 2012, Tail risk and hedge fund returns, Chicago Booth Research Paper,
(12-44).
Kozhan, R., A. Neuberger, and P. Schneider, 2013, The skew risk premium in the equity index market,
Kraus, A., and R. H. Litzenberger, 1976, Skewness preference and the valuation of risk assets, The
Journal of nance, 31(4), 10851100.
Langlois, H., 2020, Measuring skewness premia, Journal of Financial Economics, 135(2), 399424.
Mitton, T., and K. Vorkink, 2007, Equilibrium underdiversication and the preference for skewness,
Muravyev, D., 2016, Order ow and expected option returns, The Journal of Finance, 71(2), 673708.
Muravyev, D., and N. D. Pearson, 2020, Options trading costs are lower than you think, The Review
Neuberger, A., 2012, Realized skewness, The Review of Financial Studies, 25(11), 34233455.
Newey, W. K., and K. D. West, 1986, A simple, positive semi-denite, heteroskedasticity and auto-
correlationconsistent covariance matrix, .
Ni, S. X., N. D. Pearson, A. M. Poteshman, and J. White, 2021, Does option trading have a pervasive
impact on underlying stock prices?, The Review of Financial Studies, 34(4), 19521986.
Orªowski, P., P. Schneider, and F. Trojani, 2023, On the nature of (jump) skewness risk premia,
Management Science.
Pederzoli, P., 2017, Crash risk in individual stocks, Available at SSRN 2862372.
41

Rubinstein, M. E., 1973, The fundamental theorem of parameter-preference security valuation, Jour-
nal of Financial and Quantitative Analysis, 8(1), 6169.
Schneider, P., and F. Trojani, 2019, Divergence and the Price of Uncertainty, Journal of Financial
Econometrics, 17(3), 341396.
42

Figure 1: Number of Stocks with Positive Hedge Fund Option Demand
This gure plots the number of stocks in each month with positive HF option usage for four types
of option strategies (Naked Put, Protective Put, Straddle, and Call), respectively, after we match
monthly skewness asset return of each stock with its HF option usage in the previous quarter. The
sample is from April 2001 to December 2020.
Number of US Common Stocks with Hedge Fund Option Demand

Naked Put
Protective Put
Straddle
400
Call
300
200
100
0
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
43

Figure 2: Performance of the Long-Short Portfolio Sorted by Naked Put
The gure plots the rolling-ve-year performance of the long-short (L/S) portfolio sorted by Naked
Put. The upper panel plots the rolling-average monthly return (in percentage). The lower panel plots
the rolling annualized Sharpe ratio.
Plot 1. 60-Month Moving Average Returns of the L/S Portfolio of Skewness Assets Sorted by Hedge Fund Naked Put Demand
2.5
2.0
1.5
%
1.0
0.5
Plot 2. 60-Month Moving Sharpe Ratios of the L/S Portfolio of Skewness Assets Sorted by Hedge Fund Naked Put Demand
0.9
0.8
0.7
Annual
0.6
0.5
0.4
0.3
2010 2012 2014 2016 2018 2020
44

Table 1: Summary Statistics: Option Positions
This table reports the summary statistics of hedge fund holdings from December 2004 to September
2020: Panel A presents the fraction of option holdings per hedge fund portfolio and reporting quarter,
conditioning on that a fund reported at least one option holding. There are 22, 556 portfolios at the
fund-quarter level, and 5, 110 of them contain at least one option holding. Number of Positions is the
total number of rms on which a fund holds stocks or options. Option Fraction is the fraction of rms
on which a fund holds options. For example, if a fund discloses 10 dierent rms in its portfolio and
holds options on 2 of them, Option Fraction equals 0.2. (%) means that a variable is in percentage.
Panel B reports the market value (in millions of dollars) of the stocks underlying the option contracts
per fund portfolio and quarter, conditioning on that a fund reported at least one option holding.
Panel C reports the number of quarters an option position lasts. For example, a fund used options on
Tesla in March and June of 2020. It didn't use naked put on Tesla in September, but started to use it
in December 2020 again. Then we have 2 observations: One is the naked put in March and June, and
it equals 2 because it lasts for 2 quarters; The other observation is the naked put in December, and it
equals 1 because it only lasts for 1 quarter.
Panel A: Fraction of options per fund portfolio and report date (N = 5110).
Mean StdDev P1 P10 Median P90 P99
Number of Positions 119.43 239.09 3 12 46 271 1418
Option Fraction (%) 16.61 18.65 0.16 1.41 9.52 42.86 83.33
Call Fraction (%) 8.72 12.22 0 0 4.50 22.22 58.82
Put Fraction (%) 7.89 13.79 0 0 2.50 24.07 66.67
Naked Put Fraction (%) 5.25 11.43 0 0 0.64 14.89 57.14
Protective Put Fraction (%) 1.39 4.19 0 0 0 4.12 20.00
Straddle Fraction (%) 1.25 5.48 0 0 0 2.33 26.20
Panel B: `Market value' of options per fund portfolio and report date.

Option 98.31 249.18 0.003 0.21 19.42 251.47 1153.34
Call 57.12 185.96 0 0 5.81 140.27 682.66
Put 41.18 151.14 0 0 1.62 96.39 591.12
Naked Put 23.42 75.02 0 0 0.04 57.81 364.29
Protective Put 9.50 95.85 0 0 0 12.80 127.54
Straddle 8.27 81.62 0 0 0 2.16 160.55
Panel C: The number of quarters option positions last.

Option 2.07 2.24 1 1 1 4 12
Call 1.52 1.23 1 1 1 3 6
Put 2.14 2.32 1 1 1 4 12
Naked Put 1.46 1.15 1 1 1 3 6
Protective Put 1.41 1.07 1 1 1 2 6
Straddle 2.37 2.60 1 1 1 5 13
45

Table 2: Summary Statistics
This table reports the summary statistics: rSkew is the monthly return of the skewness asset in Bali
and Murray (2013); Number of Holders is the total number of hedge funds that hold the stock or its
options. Other variables are the proportions of hedge funds disclosing a certain type of option strategy
on the rm, dened as the number of funds using the strategy divided by the total number of funds
holding the stock or its option. We match rSkew with its option variables in the past quarter. We
classify 6 option strategies: (1).`Option' means either long put or call; (2).`Call' means only long call
but without put; (3).`Put' means long put. Fund may also long call in this case; (4).`Naked Put'
means long put without the underlying stock and call; (5).`Protective Put' means long put and the
underlying stock, without call; (6).`Straddle' means long both put and call. Sample is from January
2005 to December 2020.
Observation Mean StdDev P1 P10 Median P90 P99

rSkew 145067 -0.0077 0.1633 -0.4618 -0.1163 -0.0053 0.1099 0.3127
Number of Holders 145067 28 20 4 10 23 52 101
Option 145067 0.0296 0.0461 0 0 0 0.0833 0.2000
Call 145067 0.0113 0.0272 0 0 0 0.0417 0.1250
Put 145067 0.0184 0.0351 0 0 0 0.0571 0.1538
Naked Put 145067 0.0059 0.0219 0 0 0 0.0204 0.1000
Protective Put 145067 0.0036 0.0143 0 0 0 0 0.0667
Straddle 145067 0.0089 0.0212 0 0 0 0.0333 0.0909
46

Table 3: Predict cross-sectional skewness asset returns
We run the monthly Fama-MacBeth regression:
Skew
ri,t+1 = αt + γt · Hedge F und Option U sei,t + θt · Controlsi,t + ϵi,t+1 ,
Skew
where: ri,t+1 is rm i's skewness asset return in month t + 1. Hedge F und Option U sei,t is the latest
available proportion of hedge funds disclosing a certain type of option position on rm i. Control
variables include: average implied volatilities and percentage bid-ask spreads of options in the skewness
asset; short interest of the stock; realized skewness of daily log stock returns in the past month;
daily average turnover in the past month; the maximum stock return in the past month; market beta
estimated using rolling-1-year daily stock return; logarithm of market cap; logarithm of book-to-market
ratio; idiosyncratic volatility of Fama-French 3 factors in past month (IVOL); past month stock return
(RETt−1,t ); stock momentum (RETt−12,t−1 ); Amihud illiquidity measure; rm leverage. T statistics
are Newey-West adjusted with 3 lags. Sample is from January 2005 to December 2020.
(1) (2) (3) (4) (5) (6)

Option 0.052 0.029
(3.87) (2.42)
Call 0.050 0.046 0.044 0.041
(2.34) (1.94) (2.04) (1.72)
Put 0.059 0.025
(3.04) (1.64)
Naked Put 0.150 0.097
(4.38) (2.86)
Protective Put 0.008 -0.012
(0.26) (-0.37)
Straddle 0.078 -0.015
(0.72) (-0.48)
Implied Volatility 0.029 0.029 0.030
(4.12) (4.11) (4.20)
Option BA Spread -0.000 -0.000 -0.001
(-0.16) (-0.17) (-0.34)
Short Interest 0.047 0.047 0.047
(4.51) (4.37) (4.37)
Lag Skew -0.001 -0.001 -0.001
(-2.09) (-2.10) (-2.04)
Turnover -0.169 -0.170 -0.163
(-2.32) (-2.31) (-2.19)
Max. Stock Return 0.007 0.009 0.006
(0.23) (0.32) (0.22)
β -0.006 -0.006 -0.006
(-3.04) (-3.06) (-3.05)
log(ME) 0.001 0.001 0.001
(2.47) (2.36) (2.46)
log(B/M) 0.000 0.000 0.000
(0.44) (0.24) (0.35)
IVOL -0.047 -0.049 -0.042
(-0.38) (-0.41) (-0.35)
RETt−1,t 0.029 0.029 0.030
(3.91) (3.92) (4.03)
RETt−12,t−1 0.002 0.002 0.001
(0.64) (0.65) (0.59)
Amihud 0.721 0.708 0.684
(3.11) (2.99) (2.89)
Leverage -0.005 -0.005 -0.006
(-0.94) (-0.92) (-1.02)
Intercept -0.009 -0.043 -0.009 -0.042 -0.008 -0.042
(-1.92) (-2.64) (-1.92) (-2.56) (-1.84) (-2.63)
Observation 145067 130191 145067 130191 145067 130191
Adjusted R2 0.002 0.039 47 0.004 0.041 0.008 0.043

Table 4: Strategy performance and transaction costs
This table reports the monthly average returns of long-short strategies. In Panel A, the Long portfolio
includes rms in the top decile sorted by the variable `Naked Put'. We also require those rms to have
positive Naked Put. The Short portfolio includes all rms with Naked Put equal to 0. We weight rms
by the dollar open interests of their options. We adjust alpha for the Fama and French ve factors,
stock momentum, the S&P 500 Index skewness return in excess of the risk-free rate, and the average
skewness return across all rms weighted by their open interests.
In Panel B, we compare the long-short strategy returns computed from the midpoint price (MidP)
with those computed from the eective bid-ask spread, estimated to be 25%, 50%, 75%, and 100%
of the quoted spread. To lower the impact of option bid-ask spread, we also compute returns using
only rms in the short-leg whose percentage bid-ask spreads are lower than the median of that month.
T -statistics are in parentheses. The sample period is from January 2005 to December 2020.
Panel A: Long-short portfolios.
Long Short Long - Short

Return (%) 0.79 -0.74 1.53
(1.26) (-1.55) (2.73)
Alpha (%) 1.08 -0.34 1.42
(2.72) (-2.76) (2.81)
Standard Deviation (%) 8.71 6.61 7.75
Annual Sharpe Ratio 0.31 -0.39 0.68
Skewness 2.24 -9.58 5.60
Kurtosis 43.68 114.46 46.80
Minimum (%) -59.18 -80.80 -8.83
Panel B: Option bid-ask (BA) spread.
MidP 25% 50% 75% 100%

All Firms:
Return (%) 1.53 0.94 0.35 -0.24 -0.83

(2.73) (1.68) (0.62) (-0.43) (-1.47)
Alpha (%) 1.42 0.85 0.27 -0.30 -0.88
(2.81) (1.68) (0.54) (-0.60) (-1.74)
Firms with low BA spread:
Return (%) 1.68 1.25 0.82 0.38 -0.06

(2.98) (2.21) (1.44) (0.67) (-0.10)
Alpha (%) 1.59 1.17 0.75 0.32 -0.11
(3.18) (2.34) (1.49) (0.64) (-0.23)
48

Table 5: Persistence of skewness asset return predictability
We use the naked put positions of hedge funds, lagged from 1 to 4 quarters, to predict skewness asset
returns in Fama-MacBeth regressions. T statistics are Newey-West adjusted with 3 lags. Sample is
from January 2005 to December 2020.
(1) (2) (3) (4)

N aked P ut−1 0.095
(2.88)

(1.41)

(1.63)

(1.33)
Implied Volatility 0.030 0.030 0.029 0.030

(4.23) (4.27) (4.21) (4.27)
Option BA Spread -0.000 -0.001 -0.001 -0.001
(-0.23) (-0.41) (-0.56) (-0.41)
Short Interest 0.045 0.050 0.051 0.052
(4.48) (4.85) (4.94) (4.83)
Lag Skew -0.001 -0.001 -0.001 -0.001
(-2.05) (-1.98) (-1.97) (-1.88)
Turnover -0.160 -0.164 -0.173 -0.175
(-2.12) (-2.16) (-2.28) (-2.36)
Max. Stock Return 0.006 0.008 0.006 0.003
(0.20) (0.28) (0.20) (0.10)
β -0.006 -0.006 -0.006 -0.006
(-3.06) (-3.03) (-2.89) (-2.90)
log(ME) 0.001 0.002 0.002 0.002
(2.62) (2.87) (2.79) (2.76)
log(B/M) 0.000 0.000 0.000 0.000
(0.42) (0.48) (0.53) (0.68)
IVOL -0.047 -0.052 -0.039 -0.029
(-0.37) (-0.42) (-0.30) (-0.23)
RETt−1,t 0.029 0.028 0.029 0.028
(3.95) (3.65) (3.67) (3.54)
RETt−12,t−1 0.001 0.001 0.001 0.001
(0.58) (0.56) (0.50) (0.48)
Amihud 0.750 0.746 0.728 0.730
(3.17) (3.18) (3.10) (3.01)
Leverage -0.005 -0.005 -0.005 -0.005
(-0.95) (-0.84) (-0.91) (-0.92)
Intercept -0.044 -0.047 -0.047 -0.047
(-2.73) (-2.93) (-2.87) (-2.81)
Observation 130191 130072 129803 129304

Adjusted R2 0.039 0.038 0.039 0.039
49

Table 6: Net signed option open interest
This table reports the results of monthly Fama-MacBeth regressions. The dependent variable is a
rm's net signed open interest of put (call), scaled by its shares outstanding and then multiplied by
103 , on the third Friday of each month. In Panel B, we classify option contracts into OTM (out-of-
the-money) and ATM+ITM (at- and in-the-money). In Panel C, we multiply open interest by the
delta and vega, respectively. All independent variables are standardized. T-statistics, in parentheses,
are computed using the Newey-West standard errors with three lags. Constrained by the availability
of the ISE data, the sample begins in January 2006 and ends in May 2016.
Panel A: Open interest.
Put Call
(1) (2) (3) (4)

Naked Put 0.580 0.328 -0.609 -0.439
(3.44) (3.37) (-5.21) (-3.60)
Call -0.106 -0.155 0.137 0.136
(-2.00) (-2.55) (1.72) (1.72)
Protective Put 0.201 0.161 -0.165 -0.125
(1.56) (1.60) (-1.64) (-1.43)
Straddle -0.091 -0.188 -0.264 -0.107
(-1.23) (-2.13) (-3.20) (-1.21)
Controls No Yes No Yes
Observation 135516 119727 135516 119727
Adjusted R2 0.020 0.058 0.017 0.056
Panel B: Open interest by moneynesses.
Put Call
OTM ATM+ITM OTM ATM+ITM
(1) (2) (3) (4) (5) (6) (7) (8)

Naked Put 0.301 0.227 0.279 0.101 -0.249 -0.186 -0.360 -0.253
(2.79) (2.91) (3.39) (1.47) (-2.93) (-1.95) (-4.59) (-3.80)
Call -0.027 -0.049 -0.078 -0.106 -0.000 0.058 0.137 0.078
(-0.65) (-1.02) (-2.62) (-3.07) (-0.00) (0.77) (1.87) (1.07)
Protective Put 0.125 0.126 0.076 0.035 -0.115 -0.077 -0.050 -0.048
(1.43) (1.81) (1.21) (0.64) (-1.74) (-1.33) (-0.90) (-0.92)
Straddle -0.048 -0.042 -0.044 -0.147 -0.193 -0.051 -0.071 -0.056
(-0.72) (-0.54) (-0.86) (-3.00) (-3.34) (-0.75) (-1.25) (-1.04)
Controls No Yes No Yes No Yes No Yes
Observation 135516 119727 135516 119727 135516 119727 135516 119727
Adjusted R2 0.017 0.048 0.018 0.058 0.017 0.049 0.014 0.048
Panel C: Weighted by delta and vega.
Delta-adjusted Vega-adjuested
Put Call Put Call
(1) (2) (3) (4) (5) (6) (7) (8)

Naked Put -0.203 -0.095 -0.264 -0.186 2.835 2.248 -3.951 -2.688
(-3.05) (-2.17) (-5.18) (-4.11) (2.44) (2.61) (-4.46) (-3.54)
Call 0.046 0.069 0.058 0.036 -0.432 -0.600 -0.790 -0.219
(1.99) (3.09) (1.27) (0.79) (-1.21) (-1.67) (-1.33) (-0.33)
Protective Put -0.094 -0.059 -0.066 -0.059 0.397 0.441 -1.689 -0.695
(-1.85) (-1.42) (-1.63) (-1.57) (0.84) (1.11) (-2.63) (-1.52)
Straddle 0.054 0.096 -0.087 -0.056 -0.404 -0.000 -3.517 -0.780
(1.96) (3.45) (-2.18)
50(-1.46) (-0.68) (-0.00) (-3.26) (-0.81)
Controls No Yes No Yes No Yes No Yes
Observation 135516 119727 135516 119727 135516 119727 135516 119727
Adjusted R2 0.021 0.059 0.015 0.054 0.008 0.024 0.008 0.025
Table 7: Explain hedge fund option usage
We run the quarterly Tobit regression:
Hedge F und Option U sei,t+1 = α + θ · V ariablesi,t + ϵi,t+1 .
Dependent variables are left censored at 0 and right censored at 1. We include quarter-xed eects and
cluster standard errors at the quarter level. We control for the lagged-one-quarter dependent variable.
Independent variables are standardized and winsorized at 1% level. Sample is from December 2004 to
September 2020.
(1) (2) (3) (4)

Naked Put Call Protective Put Straddle
Short Interest 0.021 0.006 0.010 0.008
(26.79) (11.30) (9.97) (12.54)
Turnover 0.008 0.012 0.007 0.011
(9.59) (16.20) (8.16) (15.31)
log(ME) 0.054 0.048 0.053 0.044
(41.93) (33.63) (39.45) (36.55)
Implied Volatility 0.012 0.011 0.009 0.004
(8.51) (8.87) (5.11) (3.03)
RETt−1,t 0.002 0.003 -0.003 0.003
(1.95) (4.25) (-2.08) (5.84)
RETt−12,t−1 0.006 -0.004 -0.002 -0.001
(6.02) (-5.23) (-1.70) (-2.00)
IVOL -0.000 0.005 0.004 0.004
(-0.09) (4.40) (3.13) (5.20)
Max. Stock Return 0.000 -0.002 -0.001 -0.002
(0.09) (-1.72) (-0.87) (-1.92)
β -0.000 -0.000 -0.000 0.001
(-0.68) (-0.73) (-0.35) (0.67)
log(B/M) -0.000 -0.000 -0.000 0.000
(-0.41) (-0.87) (-0.62) (0.78)
Leverage 0.002 0.000 -0.002 -0.000
(1.70) (0.60) (-2.38) (-0.74)
Amihud 0.002 0.004 0.007 0.004
(0.66) (2.22) (2.82) (3.62)
Lag Skew -0.000 0.001 0.001 -0.000
(-0.48) (1.24) (1.19) (-0.96)
Lag Dependent 0.024 0.026 0.017 0.018
(40.35) (30.50) (28.06) (29.00)
Intercept -0.119 -0.137 -0.178 -0.114
(-55.62) (-73.83) (-30.92) (-32.74)
Observation 133607 133607 133607 133607

Pseudo R2 0.576 0.734 0.530 1.616
51

Table 8: Hedge fund versus mutual fund
This table matches the long option positions of hedge funds with the short option positions of mutual
fund based on the underlying stock and reporting quarter. The column `Hedge Fund' is the number
of option positions for a certain type of strategy in the row variable, and is at the fund-quarter-stock
level. The column `Mutual Fund' is the number of hedge fund option positions with at least one mutual
fund taking the corresponding short option positions on the same underlying rm in the same quarter.
The column `Percentage' is dened as the ratio of Mutual Fund to Hedge Fund. Constrained by the
availability of mutual fund option holdings data, the sample is from December 2004 to June 2015.
Hedge Fund Mutual Fund Percentage

Call 20969 5151 24.56%
Put 25591 3081 12.04%
Naked Put 8166 784 9.60%
Table 9: Option hedging costs
This table reports return means and t-statistics from sequential double sorts on skewness assets. Every
third Friday, we sort skewness assets into 3 portfolios (Low, Middle, High) based on a proxy for the
option hedging cost as shown in the row variable and then, within each tercile, form long-short portfolio
based on the variable `Naked Put'. We take long (short) positions on all rms with positive (zero)
Naked Put and value-weight them by option dollar open interests. Numbers reported are the long-short
return spreads within each tercile. T-statistics are in parentheses. Following Cao et al. (2019), we
measure stochastic volatility by estimating an EGARCH(1,1) model on the third Friday using rolling-
1-year daily stock returns and then computing the standard deviation of the model-implied volatilities.
To measure jump risk, we use the disaster concern index in Gao, Gao, and Song (2018). The larger
the index, the greater the downward jump risk faced by market makers who write put options. We
measure stock bid-ask spread as the average daily bid-ask spread in the previous month. Sample is
Low Middle High

Stochastic volatility 0.39 1.06 2.18
(1.01) (1.50) (2.74)
Jump risk 0.17 0.16 2.16
(0.73) (0.42) (2.40)
Stock bid-ask spread -0.06 1.19 1.79
(-0.17) (2.47) (2.19)
52

Table 10: Skewness asset in Orªowski, Schneider, and Trojani (2023)
We run monthly Fama-MacBeth regression to predict the return, price, and payo of the skewness
asset in Orªowski, Schneider, and Trojani (2023). T statistics are Newey-West adjusted with 3 lags.
Sample is from January 2005 to December 2020.
rSkew, OST SkewImplied, OST SkewRealized, OST

(1) (2) (3) (4) (5) (6)
Naked Put 0.349 0.464 -0.433 -0.167 -0.009 -0.006
(4.74) (4.41) (-5.53) (-3.40) (-1.83) (-1.25)
Implied Volatility -0.281 -0.283 -0.007

(-7.75) (-13.86) (-4.14)
Option BA Spread -0.209 -0.017 -0.000
(-9.81) (-5.08) (-0.08)
Short Interest 0.219 0.020 0.001
(6.00) (2.98) (0.55)
Lag Skew -0.008 0.002 0.000
(-2.80) (3.22) (1.17)
Turnover -1.519 -0.045 -0.002
(-2.81) (-0.69) (-0.14)
Max. Stock Return 0.160 -0.109 -0.005
(1.07) (-3.54) (-1.49)
β 0.042 0.023 0.000
(4.35) (7.58) (2.37)
log(ME) 0.007 -0.010 -0.000
(2.74) (-11.44) (-3.76)
log(B/M) 0.001 -0.002 -0.000
(0.49) (-3.76) (-0.28)
IVOL -0.708 0.631 0.046
(-1.01) (5.23) (3.20)
RETt−1,t 0.466 -0.030 0.001
(9.25) (-5.42) (0.40)
RETt−12,t−1 0.072 0.009 0.000
(6.09) (4.45) (0.35)
Amihud -0.732 -0.754 -0.030
(-0.64) (-3.32) (-0.64)
Leverage 0.026 -0.026 -0.000
(1.41) (-7.17) (-0.29)
Intercept 0.891 0.893 -0.018 0.314 -0.000 0.007
(88.67) (12.78) (-6.82) (10.62) (-5.71) (3.99)
Observation 154999 138839 154999 138839 154999 138839

2
Adjusted R 0.000 0.045 0.018 0.380 0.003 0.090
53

Table 11: Predict option-implied skewness
We run monthly Fama-MacBeth regression to predict the option-implied skewness in Bakshi, Kapadia,
Implied, T otal
and Madan (2003). We consider three types of skewnesses: the total skewness (Skew ),
Implied, Idio. Implied, Sys.
the idiosyncratic skewness (Skew ), and the systematic skewness (Skew ). All
independent variables are standardized. T statistics are Newey-West adjusted with 3 lags. Sample is
Implied, T otal Implied, Idio. Implied, Sys.

Skewi,t+1 Skewi,t+1 Skewi,t+1
(1) (2) (3) (4) (5) (6)
Naked Put -0.039 -0.035 -0.062 -0.042 0.023 0.007
(-13.35) (-12.32) (-16.46) (-12.44) (10.70) (5.45)
Implied Volatility 0.189 -0.106 0.294

(14.45) (-7.43) (19.34)
Option BA Spread -0.083 -0.086 0.002
(-12.20) (-12.51) (2.29)
Short Interest -0.097 -0.106 0.009
(-17.80) (-18.28) (4.80)
Lag Skew 0.003 -0.024 0.027
(0.93) (-6.38) (10.65)
Turnover 0.000 -0.007 0.007
(0.11) (-1.31) (3.59)
Max. Stock Return -0.019 0.068 -0.088
(-2.04) (5.35) (-11.85)
β -0.151 0.127 -0.278
(-11.69) (10.32) (-19.01)
log(ME) -0.279 -0.181 -0.098
(-13.35) (-8.92) (-12.74)
log(B/M) 0.000 0.014 -0.013
(0.10) (3.96) (-9.07)
IVOL 0.002 -0.092 0.094
(0.23) (-7.19) (12.86)
RETt−1,t -0.076 -0.092 0.017
(-17.19) (-18.60) (9.40)
RETt−12,t−1 -0.021 -0.034 0.013
(-4.47) (-6.71) (7.08)
Amihud -0.078 -0.021 -0.057
(-10.54) (-2.93) (-12.99)
Leverage -0.029 -0.027 -0.002
(-8.50) (-7.57) (-0.82)
Intercept -0.642 -0.656 -0.195 -0.222 -0.447 -0.434
(-29.31) (-21.56) (-10.16) (-10.96) (-16.24) (-12.77)
Observation 163066 146039 163066 146039 163066 146039

Adjusted R2 0.005 0.252 0.010 0.158 0.004 0.640
54

Table 12: Variance risk premium
This table checks whether the predictive power of long naked put positions is robust to the inclusion of
variance risk premium, dened as the dierence between implied variance and rolling-one-month his-
torical variance following Bollerslev, Tauchen, and Zhou (2009). T statistics are Newey-West adjusted
with 3 lags. Sample is from January 2005 to December 2020.
(1) (2) (3) (4)

Naked Put 0.153 0.147 0.094
(4.69) (4.46) (2.82)
Variance Risk Premium 0.013 0.013 0.006
(3.95) (3.74) (1.26)
Implied Volatility 0.021

(2.71)
Option BA Spread -0.000
(-0.12)
Short Interest 0.045
(4.41)
Lag Skew -0.001
(-2.22)
Turnover -0.145
(-1.89)
Max. Stock Return 0.013
(0.41)
β -0.006
(-2.88)
log(ME) 0.001
(2.60)
log(B/M) 0.000
(0.42)
IVOL 0.020
(0.14)
RETt−1,t 0.028
(3.69)
RETt−12,t−1 0.001
(0.59)
Amihud 0.766
(3.17)
Leverage -0.006
(-0.99)
Intercept -0.008 -0.008 -0.009 -0.044
(-1.70) (-1.67) (-1.84) (-2.63)
Observation 145067 144898 144707 129933

Adjusted R2 0.002 0.004 0.006 0.042
55

A Variable Construction
This section presents the construction details for the control variables used in this paper:
Implied Volatility: In Table 3, 5, and 12, we use the average implied volatility of
the options in the skewness asset of Bali and Murray (2013). In Table 6, 7, 10,
and 11, we use the average implied volatility of ATM put and call with 30 days
maturity in Volatility Surface database of OptionMetrics.
Option Bid-Ask Spread: The average percentage option bid-ask spread.
Short Interest: The number of shares shorted in the Compustat database, scaled
by the shares outstanding.
Lag Skew: The lagged realized skewness calculated from daily log stock returns
in the previous months.
Turnover: The average daily turnover (volume/shares outstanding) in the previ-
ous month.
Max. Stock Return: the maximum daily stock return in the previous month (Bali,
Cakici, and Whitelaw (2011)).
β: The CAPM beta calculated from rolling-one-year daily stock returns.
Log(ME): The natural logarithm of a rm's market capitalization on the third
Friday of each month.
Log(B/M): The natural logarithm of the book equity for the scal year-end of a
calendar year divided by the market equity at the end of December of that year,
as in Fama and French (1992).
56

IVOL: The idiosyncratic volatility of stock returns estimated from the Fama-
French three factors using rolling one-month daily returns, following Ang, Ho-
drick, Xing, and Zhang (2006).
RETt−1,t : Short-term stock return reversal, calculated as the cumulative stock
return from the start of a month to the third Friday of that month.
RETt−12,t−1 : Stock return momentum, calculated as the cumulative stock return
over the past 11 months ending at the end of the previous month (Jegadeesh and
Titman (1993)).
Amihud: Amihud illiquidity measure (Amihud (2002)), calculated using the past
year daily data and multiplied by 106 to adjust for the scale.
Leverage: The leverage ratio of a rm computed from the quarterly Compustat
database, dened as Book Debt / (Book Debt + Market Cap.). Book Debt equals
the sum of DLTTQ and DLCQ in Compustat.
57

SSRN Id4592419

Uploaded by

Copyright:

Available Formats

SSRN Id4592419

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

SSRN Id4592419

Uploaded by

Copyright:

Available Formats

Hedge Fund Option Usage and Skewness Risk

Keywords: skewness risk premium, option, hedge fund.

Electronic copy available at: https://ssrn.com/abstract=4592419

long-standing interest to nancial economists. Early studies such as Rubinstein (1973),

frameworks in which only systematic skewness carries a negative risk premium as it

with skewness preference may choose to hold under-diversied portfolios.

Electronic copy available at: https://ssrn.com/abstract=4592419

Second, HF is commonly viewed as an important user of options. Since option demand

skewness premium through which side of the distribution.

position as an example: In a price-pressure channel, HF demand for put option will

Electronic copy available at: https://ssrn.com/abstract=4592419

return of skewness asset.

We utilize the individual stock option positions of HF disclosed in the quarterly

proportion of a stock's HF holders disclosing a certain option strategy on the underlying

We construct the skewness asset following Bali and Murray (2013).

Electronic copy available at: https://ssrn.com/abstract=4592419

option price through the frictions from market makers.

To further disentangle the price-pressure channel from the informed-trading chan-

using the asset in Bali and Murray (2013).

Electronic copy available at: https://ssrn.com/abstract=4592419

enough, it could outweigh the decrease of risk-neutral skewness in the price-pressure

asset combined lend supports to the price-pressure channel.

3 The negative relation is robust to alternative construction in BKM.

Electronic copy available at: https://ssrn.com/abstract=4592419

Electronic copy available at: https://ssrn.com/abstract=4592419

We also provide empirical evidence against alternative interpretations. First, the

Electronic copy available at: https://ssrn.com/abstract=4592419

of HF as a marginal investor in pricing skewness risk.

Electronic copy available at: https://ssrn.com/abstract=4592419

the pricing of those risks.

caused by price pressure. Section 5 performs robustness checks. Section 6 concludes.

2.1 Hedge Fund Option Holdings

Electronic copy available at: https://ssrn.com/abstract=4592419

the strike price or expiration date of an option. In addition, institutional investors

of HF companies that appear in four commercial HF databases: Eurekahedge, Hedge

of portfolio holdings disclosed by 770 HF advisers from December 2004 to September

Electronic copy available at: https://ssrn.com/abstract=4592419

Table 1 presents the summary statistics of HF option holdings. In Panel A, there

skewed, with medians much smaller than means.

Electronic copy available at: https://ssrn.com/abstract=4592419

23.42 million dollars.

HFs use options to bet on relatively short-lived information.

be positively related to the demand pressure created by HF industry on Tesla's put

Electronic copy available at: https://ssrn.com/abstract=4592419

option contracts. In addition, it is a well-established measure in the literature: Aragon

positions predict volatilities and returns on the underlying stock.

2.2 Skewness Asset Return

interest rate of appropriate maturity (interpolated when necessary) from OptionMetrics

interest are from Compustat.

Electronic copy available at: https://ssrn.com/abstract=4592419

of options. rf is the risk-free interest rate.

it represents long skewness positions. If we hold the asset to expiration, it realizes a

to small stock moves.

only the observations with ∆P ≥ −0.25 and ∆C ≤ 0.25.11

Electronic copy available at: https://ssrn.com/abstract=4592419

January 2005. The sample ends in December 2020.

of 16.33%. On average, a stock has 28 HF holders. HF option usage is heavily skewed:

diversied decile portfolio.

Electronic copy available at: https://ssrn.com/abstract=4592419

long-standing interest to nancial economists. Early studies such as Rubinstein (1973),

with skewness preference may choose to hold under-diversied portfolios.

diversied decile portfolio.

Table 3 reports the time-series averages of the coecients in Fama-MacBeth regres-

signicant at the conventional level is naked put, with a t-statistic of 2.86.

especially well during the nancial crisis and COVID period.

An eective-to-quoted spread ratio of 50% is equivalent to paying half of the quoted

paid. Therefore, reducing transaction cost is essential to maintain the protability of

to aect option return.

corroborates our earlier ndings .

visual inspection to remove misclassications because some observations contain the

skewness in OST dened as: