You are on page 1of 23

Available online at www.sciencedirect.

com

Journal of Financial Markets 16 (2013) 308330


www.elsevier.com/locate/nmar

Short sales and put options: Where is the bad news


rst traded?$
Xiaoting Haoa, Eunju Leeb, Natalia Piqueirac,n
a

Sheldon B. Lubar School of Business, University of Wisconsin-Milwaukee, Milwaukee, WI 53211, USA


b
C.T. Bauer College of Business, University of Houston, Houston, TX 77204, USA
c
The Brattle Group, Washington, DC 20036, USA.
Received 9 March 2012; received in revised form 21 August 2012; accepted 12 September 2012
Available online 2 November 2012

Abstract
Although the literature provides strong evidence supporting the presence of informed trading in
both the option and the short equity markets, it is not clear which market attracts more informed
trading. Using a unique dataset that covers intraday transaction data in the option and short equity
markets, we investigate informed trading in a cross-market environment by explicitly studying the
leadlag relationship between the put net trade volume and short sales of the underlying stock. Our
high frequency analysis shows that in general short sales contain more information. However, put
options become more informative before the release of negative earnings announcements.
& 2012 Elsevier B.V. All rights reserved.
JEL classification: G14
Keywords: Put option; Short sales; Informed trading; Earnings announcements

We thank the Editor Amit Goyal, an anonymous referee, Alex Boulatov, Tom George, Kris Jacobs, Archana
Jain, Praveen Kumar, Bruce N. Lehmann, Stuart Turnbull, and participants at the 2010 Financial Management
Association Annual Meeting and the 2011 Midwest Finance Association Annual Meeting for helpful comments
and suggestions. We thank the C.T. Bauer College of Business for providing access to the BAuer Research
DataSet (BARDS), and Yadira Taylor for her assistance with this database.
n
Corresponding author.
E-mail addresses: haox@uwm.edu (X. Hao), elee@bauer.uh.edu (E. Lee),
Natalia.Piqueira@brattle.com (N. Piqueira).
1386-4181/$ - see front matter & 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.nmar.2012.09.005

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

309

1. Introduction
Accumulating evidence in the literature supports the presence of informed trading in
option markets. Black (1975) suggests that informed investors may choose to trade in the
option market since it is typically characterized by lower capital requirements and trading
restrictions, and by higher leverage. A formal model developed by Easley, OHara, and
Srinivas (1998) suggests the existence of a pooling equilibrium in which informed investors
choose to trade in both equity and option markets. Subsequent empirical research further
supports the presence of informed trading in the option market, by showing that option
trading volume predicts subsequent returns of the underlying stock (Pan and Poteshman,
2006; Chakravarty, Gulen, and Mayhew, 2004; Cao, Chen, and Grifn, 2005, etc.).
At the same time, extensive recent research focuses on the role of short sellers in
conveying valuable information about the stocks they short. Most of the empirical
evidence suggests that short sellers are indeed informed traders and therefore they play an
important role in the price discovery process (e.g. Boehmer, Jones, and Zhang, 2008;
Diether, Lee, and Werner, 2009; Boehmer and Wu, in press, among others).
Although the literature provides evidence supporting the presence of informed trading in
both the option and the short equity markets, it is not clear which market attracts more
informed trading. The purpose of this paper is to compare the information content of short
sales and put options trading, thus helping to understand which market is more conducive
to price discovery and information incorporation. The practical implications of this
research are also relevant given recent temporary changes in short selling regulation and
the possibility of further changes in the near future.1 If indeed short sellers play the most
important role in contributing to price efciency, what would be the impact for market
efciency of a more strict regulatory system?
In this paper, we investigate the role of short sales and put option trading in revealing
negative information of the underlying stock, by explicitly studying the effects of short
sales and put option trading imbalance on subsequent quote revisions and trading volume
in the short equity and put option markets. We use a unique dataset that covers intraday
transaction data in the option and equity markets for a sample of NYSE stocks from
March 2005 to June 2007 to estimate, using 5-minute intervals, a structural model based on
Hasbrouck (1991). We extend his bivariate VAR model of stock market trades and quote
revisions to also include the option and short equity markets. By doing so, we can directly
observe the price impact and the leadlag relationships of put option trading and short
sales, which allows us to compare the information content between put option trading and
short sales.
Our high frequency analysis shows that during the sample period, short sales appear to
contain more information since they predict subsequent stock and option returns and that
the predictive power of short sales on subsequent put net trade volume is larger than that
of put net trade volume on subsequent short sales. This result suggests more information
can be learnt from the short equity market than from the put option market. We also
perform subgroup analysis using the information shares measurement (Hasbrouck, 1995;
Chakravarty, Gulen, and Mayhew, 2004) and the relative liquidity in the put option and
short equity markets. Consistent with our expectation, put option trading is more
1

See, for example, Boehmer, Jones, and Zhang (2011) for an analysis of the effects of the September 2008
shorting ban for nancial stocks.

310

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

informative for stocks with higher information shares in the option market. Meanwhile,
short sales have predictive power over subsequent put net trade volume in spite of the
relative liquidity levels across the put option and short equity markets. In all, our results
provide strong support for the idea that informed investors will choose to trade in the short
equity market rst when they are pessimistic about the stocks performance. The only
different results are documented in our sub-period analysis, in which we study the trading
in multiple markets within 3 days before the release of an unexpected negative earnings
announcement on the underlying stocks. With a pending event that will drive down the
stocks fundamental value, more informed trading is observed in the put option market.
This result suggests that the nding of Cao, Chen, and Grifn (2005) that the call option
market plays a more important price discovery role before a takeover announcement is
similarly observed in the put option market.
To our knowledge, this paper serves as the rst attempt to use intraday data to
investigate and compare, in a multi-market setting, the informed trading in the put option
and the short equity markets, contributing to our knowledge of how negative information
is incorporated into prices. Our results show that, while the short equity market seems to
be in general more informative, the option market is more critical in the price discovery
process before negative events, suggesting that both markets have their different roles in
conveying information.
The remainder of this paper is organized as follows. Section 2 provides a brief review of
related literature and our contribution. Section 3 describes the methodology. Section 4
describes the data and presents summary statistics. In Section 5 we discuss the main results.
In Section 6 we present sub-sample results and in Section 7 we discuss our analysis related
to negative earnings announcements. In Section 8 we conduct robustness tests. Section 9
concludes.
2. Related literature
Our paper is related to two main elds of literature. First, we contribute to the literature
investigating informed trading across different markets, in particular the studies
comparing informed trading in the option and stock markets. Second, we contribute to
the literature on informed short selling, in particular the studies linking return
predictability and short selling activity. In this section, we briey summarize the most
relevant ndings in each of these two elds and present our contribution to the literature.
Many empirical studies investigate the presence of informed trading in the option
market and the role of option trading in revealing valuable information about the
underlying stock. Blacks (1975) argument that informed investors would prefer to trade
options due to the higher leverage has been tested by several subsequent studies. For
example, Easley, OHara, and Srinivas (1998) show that informed trading is also observed
in option markets. They propose and test a model that allows investors to choose whether
they want to trade in stock or option markets, showing that a pooling equilibrium i.e.
an equilibrium in which informed traders trade in both option and stock markets is
associated with high leverage in the option market, low liquidity in the stock market, or a
high proportion of overall informed investors. Their empirical results show that the option
net trade volume predicts future stock price movement, rejecting the separating
equilibrium in which informed investors only trade in the stock market. Pan and
Poteshman (2006) also provide evidence that option trading volume predicts stock returns.

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

311

More specically, they nd that stocks with the lowest put-call ratios (positive signal)
outperform those with the highest put-call ratios (negative signal) in the near future.
Chakravarty, Gulen, and Mayhew (2004) measure the price discovery in the stock and
option markets by applying Hasbroucks (1995) methodology to a sample of 60 rms,
using intraday data. Their results provide evidence in favor of informed trading occurring
in both equity and option markets since they nd signicant price discovery in the option
market. Using intraday data, Cao, Chen, and Grifn (2005) study the information content
of call option trading imbalance and underlying stock trading imbalance before takeovers.
They nd that during the pre-announcement period, call option trading imbalance has
higher predictability over the next-day stock returns (takeover premiums) than stock
trading imbalance, providing further evidence on the presence of informed trading in the
option market.
The results, however, are mixed when the focus is on which market leads the other, i.e.
where informed investors would trade rst. Anthony (1988), using total (not signed) daily
volume, studies the relationship between equity and call option markets. His results show
that call option trading volume predicts trading in the underlying stock on the next day,
i.e. options seem to lead stocks. On the other hand, Chan, Chung, and Fong (2002), using a
sample of 14 stocks, intraday data and a similar methodology used in our paper, nd
evidence that the stock market leads the option market, i.e. informed investors prefer to
trade in the stock market.2
Overall, the literature nds strong evidence supporting the presence of informed
trading in options markets but fails to form a consensus on which market contains
more information. Although most of the studies use signed volume in both markets to
compare their relative informativeness, none of them attempt to explicitly incorporate the
trading volume from another market with strong informed investor presencethe
short equity market. Our paper aims to investigate and compare informed trading between
the equity and option markets, by explicitly distinguishing short sellers in the equity
market.
Our paper is also related to the literature on informed short selling, in particular the
studies linking return predictability and short selling activity. Many empirical studies test
the predictions of the theoretical model proposed by Diamond and Verrecchia (1987), in
particular the implication that short sellers are informed. Most of the empirical evidence
related to return predictability with monthly short interest data or intraday short trading
(ow) data suggests that short sellers are informed traders and, by having valuable bad
news, are able to predict negative returns.3 For example, Boehmer, Jones, and Zhang
(2008) show that short sellers are well informed, by using proprietary daily short trading
data from NYSE from 2000 to 2004. In particular they show that stocks that are heavily
shorted underperform lightly shorted stocks by 1.16% (in risk-adjusted terms) in the
subsequent 20 trading days. Diether, Lee, and Werner (2009) nd that portfolios formed
by buying lightly shorted stocks and selling heavily shorted stocks achieve positive
abnormal returns, by using short sales intraday data in 2005. This is again, evidence that
short sellers have valuable information regarding the stock. Boehmer and Wu (in press)
also use daily short trading data to show that short sellers increase the informational
efciency of stock prices, according to several efciency measures.
2
3

This is consistent with separating equilibrium in Easley, OHara, and Srinivas (1998).
In general, intraday shorting ow data are used to construct a daily series of short trading.

312

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

Using monthly short interest data, Asquith, Pathak, and Ritter (2005) show that stocks
with high short interest underperform stocks with low short interest for equally-weighted
portfolios. Desai, Ramesh, Thiagarajan, and Balachandran (2002) also show that
abnormal negative returns are observed for stocks with high monthly short interest.
Although the empirical evidence summarized above suggests that short sellers are
informed traders, these studies do not attempt to investigate short selling in a multi-market
setting, in particular in an environment where informed traders might also choose to trade
options.4 We contribute to this strand of literature by investigating the effects of short
selling in a multi-market environment, aiming to shed light on how information ows
across markets. Our unique dataset, which covers high frequency trading data in options
and short equity markets, allows us to achieve this goal. To our knowledge, this is the rst
paper that uses intraday data to investigate the effects of both short selling and option
trading on stock returns and subsequent trading volume.
3. Methodology and empirical predictions
3.1. Systems of regression equations for multiple markets
The model we use to describe the dynamic relationship between trades and quote
revisions in the stock, call, put and short markets is based on Hasbrouck (1991) and Chan,
Chung, and Fong (2002). In Hasbrouck (1991), a bivariate VAR model of trades and
quote revisions for the stock market is used to study the information content of stock
trades, while in Chan, Chung, and Fong (2002), in order to compare the information roles
of stock and option trades and quote revisions, this model is extended to include the option
market. We further extend the structural model proposed by Chan, Chung, and Fong
(2002) to also include trades in the short equity market. This is done by assuming that the
information in short transactions cannot be fully conveyed from the imbalance of stock
trading volume.
The basic bivariate VAR model for a single market is specied as follows:5
rt ALrt B0zt BLzt e1,t

zt CLrt DLzt e2,t

where rt is the quote revision at transaction time t, which is calculated as the change of bid
ask midpoint from these quotes following transaction t1 to the quotes following
transaction t, and zt is the total trading imbalance (positive if buy-initiated, and negative if
sell-initiated) between transaction time t1 and t. By assumption, the error terms in the
two equations have zero means and are independent from each other.
Chan, Chung, and Fong (2002) extend Eqs. (1) and (2) to include trades and quotes in
multiple (stock, call and put) markets. We follow the same line of reasoning to further
4
In a recent paper, Grundy, Lim, and Verwijmeren (2012) study the effects of the 2008 short-sale ban on the
option market trading, and nd that price discovery is more likely to occur in the short equity market instead of
the options market. Our study differs from this study in several ways. First, our study uses intraday data, while
Grundy, Lim, and Verwijmeren (2012) use daily data. Second, Grundy, Lim, and Verwijmeren (2012) focus on a
special period of time, i.e., the period around the 2008 short-sale ban, while our study uses 2.5 years of data and
reveals the relationship between short sales and put options before the 2008 nancial crisis.
5
See Hasbrouck (1991) for details on this specication (Eqs. (2) and (4), pp. 183184).

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

313

include the trades in the short equity market into our structural model, and by doing so, we
assume that it conveys additional information compared to the net stock trading volume.
More specically, we dene rt rs,t rc,t rp,t 0 and zt zs,t zc,t zp,t zss,t 0 in Eqs. (1)
and (2), where rs,t, rc,t and rp,t represent quote revisions in the stock, call, and put market
during time interval t, and zs,t ,zc,t ,zp,t and zss,t represent net trade volume in the stock,
call option, put option markets and the total short trading volume during time interval t.
Three lags of each explanatory variable are included in the structural model.
AL3  3, B0 3  4, BL3  4, CL4  3, and DL4  4 (L 1, 2, 3) are
coefcients to be estimated. Consistent with the methodology used in prior literature,
lagged values of dependent variables on the right hand side are used to capture serial
correlation effects, so that the disturbances can be assumed to be serially independent from
each other. Compared to the system of two equations in Hasbrouck (1991) and the system
of six equations in Chan, Chung, and Fong (2002), we have a system of seven equations in
total. We believe that by explicitly distinguishing the short trading effects, we will be able
to further investigate the role of short equity trading and put option trading in revealing
information about the stock.
3.2. Empirical predictions
Since the purpose of this paper is to compare the informational role of put option and
short selling volume, we focus on studying the effects of put net trade volume and short
size on the subsequent return and trading volume in the put and short equity markets.
First, with respect to the effects of net trade volume on subsequent quote revisions, we
expect that put net trade volume (short size) should be more signicant in predicting the
subsequent stock and put market returns if there is more informed trading in the put
(short) market. This methodology has been utilized in Chan, Chung, and Fong (2002), in
which they compare the informational role of stock and option volume by studying their
effects on subsequent returns in the market. In similar research, Cao, Chen, and Grifn
(2005) study the information content of call options by relating the option trading
imbalance to the next day stock return. In the analysis of whether short sellers have
valuable information about the stock, Diether, Lee, and Werner (2009) study the
correlations between short sales and the stocks future returns.
Second, we study the effects of net trade volume on subsequent net trade volume, across
markets. We expect that if there is more informed trading in the short market (put option
market), we should be able to observe an information ow from the short (put) market to the
put option (short) market, thus short sales (put net trade volume) should lead the subsequent
put net trade volume (short sales). The leadlag relationship of trading volume in different
markets has been studied in related research. For example, Anthony (1988) nds that trading
in the call option market leads the trading of underlying shares with one-day lag. He reasons
that this represents an information ow from the option market to the stock market. It is also
widely documented in the literature that option underwriters hedge their positions in the stock
market (Grundy, Lim, and Verwijmeren, 2012; Battalio and Schultz, 2011). In this case, option
trading is more informative and leads the stock trading.
We also perform subsample analyses. Specically, we divide our sample according to
their stock market and option market characteristics, and test whether these characteristics
affect the informational role of short and put option trading. Since it is documented in the
prior literature that call option trading tends to be more informative before corporate

314

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

events (Cao, Chen, and Grifn, 2005), we also conduct a sub-period analysis during 3 days
before the release of negative unexpected earnings to see whether the same effect holds for
put options.
4. Data and summary statistics
The empirical analysis in this paper employs several different databases. The timestamped option quotes and trades data are retrieved from the BAuer Research DataSet
(BARDS) database. This dataset tracks through Reuters the nearest-maturity options
written on over 60 stocks starting from March 2005 to June 2009.6 Not all near-maturity
options for a given stock are tracked; only those whose strike prices fall within a band of
plus and minus $10 of the underlying stock price on the Friday when options are identied
for tracking. Non-missing observations in BARDS are exactly as reported by Reuters to
ensure that the data are representative of what traders see in real time. Each intraday
observation contains the stock symbol, option symbol, option type, strike price, expiration
date, transaction date, transaction time, ask price, bid price, trade size, transaction price,
quoted ask depth, and quote bid depth. Intraday data are recorded only when a
transaction occurs or the ask price, bid price, quote ask depth or quote bid depth changes.
To describe main characteristics of the BARDS database, we compare the summary
statistics of the stocks in BARDS with the statistics of all the NYSE and NASDAQ stocks
with options in the OptionMetrics database.7 Appendix B presents the summary statistics
of daily turnover and daily trading volume for NYSE/NASDAQ stocks (Panel A) and
only NYSE stocks (Panel B) in the two databases at the end of 2002, when the BARDS
database was constructed. As shown in Panels A and B, the cross-sectional distributions of
daily turnover for the stocks in BARDS and OptionMetrics suggest that, based on the
turnover ratio, the samples are similar: the average of daily turnover for the NYSE stocks
in BARDS is 0.73%, which is close to the average of daily turnover for the NYSE stocks in
OptionMetrics (0.76%). Meanwhile, the summary statistics of trading volume between the
two groups show that the stocks in the BARDS database are more actively traded
compared to the stocks in OptionMetrics.
Intraday stock trading data are obtained from the Trade and Quote (TAQ) database of
NYSE, which provides a complete history of time-stamped quotes and trades data for all
the underlying stocks whose options are tracked by BARDS. Intraday short sales data for
these stocks are also obtained from the TAQ database and are made available by NYSE as
part of the requirements under Regulation SHO (January 2005).8 For each short-sale
transaction, this dataset contains its transaction time, trade size, and an indicator that
identies short sales that are exempt from the Uptick Rule.9 We maximize the sample
period, and study the intraday transaction data in the option and stock markets between
March 21, 2005 and June 8, 2007.10
6

The options were tracked for 1 month preceding their expiration. See Appendix A for the list of stocks tracked
in BARDS between 2005 and 2009.
7
OptionMetrics contains data on all US exchange-listed equities and market indices, as well as all US listed
index and equity options.
8
For more information on Regulation SHO, see http://www.sec.gov/spotlight/shopilot.htm.
9
The Uptick Rule is a former rule established by the SEC that requires that every short sale transaction be
entered at a price that is higher than the price of the previous trade.
10
The Regulation SHO dataset is available until July 6, 2007.

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

315

Since the Regulation SHO dataset only covers stocks listed on NYSE, we are left with 45
stocks whose options are also continuously tracked during our sample period in the
BARDS dataset. Following Chan, Chung, and Fong (2002), each day the most actively
traded put and call options are selected for each stock. When the most active option has 5
days or less to maturity, it is deleted from our sample since abnormal trading in the option
market on the days near expiration is documented by prior literature.
As for stock transaction data, only the trades and quotes originated from NYSE are
included in our analysis since it is shown in Hasbrouck (1995) that the price discovery for
NYSE stocks is more likely to take place on the NYSE rather than on other exchanges. At
the same time, since we need trading volume measurements over short (5-minute) intervals,
option days with thin trading are deleted from our sample. More specically, following
Chan, Chung, and Fong (2002), option trading days with fewer than 20 trades for the
stock, the most active call, the most active put, or the short-sale are deleted.11 After that,
we are left with 8,520 option days in total.
In order to classify the trading direction for each transaction on each stock and option in
our sample, we use Lee and Ready (1991) algorithm. Specically, if a trade is executed at a
price above (below) the quote midpoint, it is classied as buy-initiated (sell-initiated).
Trades at the quote midpoint are classied using the tick test, which determines the
direction by comparing the trade price to the price of preceding trades.
To calculate the summary statistics of our sample, we obtain daily prices, returns, trading
volumes of the underlying stocks from the CRSP database and book values of the rms from
COMPUSTAT. Earnings announcement data are collected from I/B/E/S for the pre-event
period analysis. Unexpected earnings are dened as the difference between the actual earnings
and the last estimated earnings for the same quarter. All the options days that are within three
days before negative unexpected earnings announcements are chosen for our pre-event studies.
Table 1 provides summary statistics of the most active call and put options and
underlying stocks in our sample.12 Consistent with prior literature, the average daily
volume of the underlying stocks is larger than the daily volumes of the most active put and
call options. The number of trades and daily volume for the most active call options are
larger than those for the most active put options, which suggests that the call option
market is more active than the put option market. For the stocks included in the sample,
short volume on average accounts for 14.11% of total trading volume over the sample.13

5. Main results
Five-minute intervals are used to estimate the structural model (1)-(2). Each option day is
partitioned into 78 successive 5-minute intervals during the time period when both the option
and the stock markets are open (from 9:30 AM to 4:00 PM EST). For each of these intervals, we
generate 5-minute quote revisions using the midpoint of the last bid and ask quotes for each of
the stocks, the most active calls, and the most active puts. If no quote is available for an interval,
there is no quote change for that interval, and we keep the quote from the previous interval. The
11
We also change the lter with fewer than 15 trades as a robustness test, and the results are overall similar. See
Section 8 for more details.
12
The statistics are based on the entire sample that includes all trading days.
13
This is lower than the reported number (19.75%) in Boehmer and Wu (in press) possibly due to the sample
difference.

316

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

Table 1
Summary statistics.

Call options
Option price
Number of trades
Option volume
Put options
Option price
Number of trades
Option volume
Underlying stocks
Price
Volume (in thousands)
Short volume (in thousands)
Market cap (in millions)
Book value (in millions)
Turnover (volume/total shares)
Market-adjusted returns(value-weighted) (%)

Mean

Median

Std. Dev.

Min.

Max.

1.089
52.331
1,985

0.750
29
754

1.299
73.653
3,854

0
1
0

25.950
1,859
100,779

0.927
31.103
1,362

0.625
17
462

1.213
44.008
2,737

0
1
0

34.500
869
56,518

41.549
7,905
1,116
87,892
29,716
0.011
0.014

37.500
5,454
775
58,283
17,074
0.006
0.017

19.052
8,567
1,103
89,939
32,798
0.011
0.052

13.365
21
1
954
252
0.002
0.166

86.161
226,983
18,448
396,325
120,076
0.048
0.100

Table 1 reports summary statistics across all option trading days for the most actively traded call/put options and
for the underlying stock trading volume and short volume, and cross sectional summary statistics of time series
means for the remaining variables related to the 45 NYSE stocks during March 2005June 2007. Option price is
calculated as the midpoint of the best closing bid and ask prices, Number of trades is the sum of the number of
daily trades, and option volume is the daily trading volume for the most active calls and puts. Price is the daily
close price in dollars, Volume is the daily number of shares traded, and short volume is the daily number of shares
shorted. Market cap is the market value of equity in millions of dollars, and book value in millions of dollars is
calculated as book value of equity plus balance sheet deferred taxes minus the book value of preferred stock.
Turnover is daily trading volume divided by total shares outstanding, and market-adjusted returns, denoted in
percent, are daily returns minus value-weighted market returns.

return of each interval is calculated as the log of the ratio of quote midpoints in successive
intervals. We also calculate the net trade volume of the stocks and the most active calls and puts,
and the total volume of short-sales of these stocks for every 5-minute interval. Following Easley,
OHara, and Srinivas (1998) and Chan, Chung, and Fong (2002), we use the standardized return
and net trade volume variables to control for cross-sectional variations across different stocks
and options. For each option day we rst calculate the mean and the standard deviation of the
returns and net trade volume for each variable. The variable is then standardized by subtracting
the mean and dividing by the standard deviation. By doing this, we control for stock xed
effects, and this allows us to pool the 8,520 option days for later analyses.
Pooled regression is used to estimate the structural model (1)-(2). Since we use
standardized returns and net trade volume for each of variable in this model, we can
assume that the error terms are homoscedastic. Furthermore, since we include lagged
values of the dependent variables to control for serial correlation, we can also assume that
the error terms are serially independent. The seven regression equations are estimated
together using structural models, and thus we control for the correlations between the
equations. We include three lags for each explanatory variable, but for simplicity, we only
report the rst two lags in our results. The results of the main model are presented in
Table 2 and we will discuss them in detail in the next two sub-sections.

Table 2
Regression analysis of the relationship between standardized 5-minutes returns and standardized 5-minutes net trade volume of stocks, calls, puts and short sales.
Explanatory variables
Lagged stock
return
Lag1

Stock returns

0.003 0.021n 0.001


0.59
4.68
0.61
0.231n 0.086n 0.16n
48.94 18.43 82.6
0.22n 0.08n 0.09n
45.32 16.58 44.06
0.01n 0.005 0.005n

Put returns
Stock net trade
volume
Call net trade
volume
Put net trade
volume
Short size

Lag1

Lagged put
return

Lagged stock net trade


volume
Lag1

Lag2

Lag1

Lag2

Lag0

Lag1

Lag2

Lag0

Lagged short size

Lag1

Lag2

0.002
0.82
0.06n
32.63
0.04n
21.24
0.004

0.002
0.92
0.08n
42.85
0.15n
76.7
0.002

0.002 0.042n 0.000 0.001 0.003n 0.001 0.001 0.003n 0.001


0.001 0.01n 0.01n
0.82 34.79
0.02
0.77
2.49 0.72 0.73
2.83
1.09
0.55 6.05 6.74
0.03n 0.054n 0.060n 0.022n 0.066n 0.020n 0.009n 0.04n 0.000 0.003n 0.029n 0.011n
16.94 42.72 45.75
16.5 54.56 16.49
7.43 28.57
0.18
2.83 19.21
6.73
0.07n 0.06n 0.06n 0.019n 0.04n 0.000 0.005n 0.061n 0.018n 0.006n 0.006n 0.001
34.92 43.27 42.26 14.53 31.13
0.38
4.31
50.2 14.82
4.94
3.67 0.86
0.000
0.036n 0.018n
0.001 0.000
0.001 0.002
0.060n

Lag1

Lag2

0.00n
2.36
0.01n
6.17
0.011n
6.64
0.025n

13.3
0.007n

1.1
0.027n

0.14
0.005n

1.16
0.003n

1.27
0.001

36.43
0.006n

14.36
0.000

12.47
2.24 4.34 3.37 13.82 4.64
0.04n 0.02n 0.03n 0.01n 0.02n 0.01n

12.12
4.87
0.02n 0.008n

21.13
0.003n

4.05
0.000

2.48
0.032n

1.15
0.008n

3.36
0.008n

0.22
0.002

7.99
4.9 12.96 5.99
0.02n 0.005 0.003 0.001
5.19 1.19 1.96
0.5

12.98
5.45
0.01n 0.003n
11.43
2.46

2.33
0.001
1.17

0.09
0.001
1.29

2.1
0.062n

7.82
0.005n
2.62

6.05
0.002
1.27

26.8
0.017n

Lag0

Lagged put net trade volume

Lag2

1.03
2.64
1.92 0.87 0.15
0.011n 0.01n 0.01n 0.03n 0.01n

Lag0

Lagged call net trade


volume

24.94
6.39
0.003n 0.000
2.39 0.36

4.86
1.32
0.354n 0.144n
255.71 99.43

This table presents the regression results of the following multivariate VAR model:
rt ALrt B0zt BLzt e1,t
zt CLrt DLzt e2,t


where rt rs,t rc,t rp,t and zt zs,t zc,t zp,t zss,t . rs,t ,rc,t and rp,t represent quote revisions in the stock, call, and put market during time interval t, and
zs,t ,zc,t ,zp,t and zss,t represent net trade volume in the stock, call, put and short market during time interval t. AL3  3, B0 3  4, BL3  4, CL4  3, and
DL4  4 (L 1, 2, 3) are coefcients to be estimated. Each variable is then standardized by subtracting the mean and dividing by the standard deviation of the day.
Sample days are deleted where there are less than 20 trades in the put option, call option, or stock market on that day. We use contemporaneous and three lags for the
explanatory variables, and report the regression coefcients for the contemporaneous and rst two lags with nindicating signicance at the 5% level.

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

Dependent
variable

Call returns

Lag2

Lagged call
return

317

318

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

5.1. Comparison with Chan, Chung, and Fong (2002)


In this sub-section we compare our results in Table 2 relating options and the stock
market with Chan, Chung, and Fong (2002). First, we conrm their result that the stock
market net trading volume has predictive power over contemporaneous and subsequent
option returns, while the option net trade volume is only signicant in explaining
contemporaneous stock returns.14 This is consistent with Chan, Chung, and Fong (2002)
interpretation that stock market trading conveys more information than option market
trading.
Second, in terms of the relationship among returns we also nd that stock market
returns have predictive power over subsequent quote returns in the option markets, which
is consistent with Chan, Chung, and Fong (2002): for example, the coefcients on lagged
stock returns are 0.231 and 0.22 for call and put returns, respectively. Within the option
markets, a negative relationship between returns and their own lags is observed as in Chan,
Chung, and Fong (2002).
The main tests of our paper are (1) comparing the effect of short sales and put net trade
volume on the subsequent quote revisions in the two markets, and (2) studying the leadlag
relationship between short sales and put net trade volume. The purpose of these analyses is
to compare the relative price discovery role of the short and put markets. We will discuss
them in detail in the next sub-section.
5.2. The informational role of short sales and put option trading
First, with respect to the effects of short sales and put net trade volume on the
subsequent quote revisions in these two markets, Table 2 suggests that while put net trade
volume only predicts option returns, short sales have predictive power over both future
stock and option returns. We show in Table 2 that short sales have predictive power over
stock market returns in subsequent 5-minute intervals. For example, the coefcient for the
rst lagged short size is 0.01 with a t-statistic of 6.74. This evidence indicates again that
short selling possibly conveys important negative information regarding the stock, which is
not necessarily revealed by looking only at the stock net trading volume. Notice that it has
been well documented in the literature that short sellers are able to spot overvalued stocks
and their trading predicts negative future stock returns (e.g. Diether, Lee, and Werner,
2009; Asquith, Pathak, and Ritter, 2005; Boehmer and Wu, in press). Our nding,
however, is different from previous literature since it is the rst study to use intraday data
(with 5-minute intervals) to examine the information content of short sales in a multimarket setting. The same results cannot be found for the put net trade volume. As
discussed earlier, put net trade volume is only signicant in explaining its own subsequent
quote revisions. This result suggests that there is more information contained in short sales
in the equity market, compared to put option trading.
Second, with respect to the relationship between the net trading volume of short and put
markets, we predict that if the put market (short equity market) has leading information,
put net trade volume (short sales) should have predictive power over subsequent short sales
14
For contemporaneous (subsequent) option returns, the coefcient estimates on stock net trading volume are
0.05 (0.06) and 0.06 (0.06) for calls and puts. Meanwhile, for contemporaneous stock returns, the coefcient
estimates on option net trade volume are 0.003 and 0.003 for calls and puts, respectively.

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

319

(put net trade volume). If the trades in both markets contain different information and are
unrelated with each other, we should not observe any leadlag relationship between the
trading in the two markets. Table 2 provides the main results regarding the leadlag
relationship between the trading in different markets. We notice that stock net trade
volume predicts subsequent stock and option net trade volume. Meanwhile, option net
trade volume predicts subsequent option volume, but does not predict stock net trade
volume. When we include short sales in the regression, we observe signicant explanatory
power of short sales on subsequent stock net trading volume and subsequent put net trade
volume. The coefcient on lagged short size for put net trade volume is 0.008 for the rst
lag and it is statistically signicant. Given that all the dependent and independent variables
are standardized, this result suggests that a one standard deviation increase in lagged short
size results in a 0.008 standard deviation increase in put net trade volume for the next
5-minute interval. Meanwhile, the coefcient on lagged put net trade volume for short sales
(0.003 for the rst lag) suggests that a one standard deviation increase in lagged put net
trade volume leads to a 0.003 standard deviation increase in short selling. Although the
effect of put net trade volume on subsequent short selling is statistically signicant, it is
smaller in magnitude than the effect of short selling on subsequent put trade volume. This
indicates that there is more informed trading in the short equity market, and therefore
information ows from the short equity market to the put option market.
Both the analyses of the price impacts of put and short trading and the leadlag
relationship between the two markets indicate that compared to put options, short selling
volume may be more informative. Notice that prior literature suggests that informed
investors will prefer to trade options because of the higher leverage offered by this
instrument (Black, 1975). However, our results suggest that without controlling for rmspecic events, short sales volume contains more information compared to the put option
trade volume.
6. Subsample analysis
To further compare the information role of put option trading and short sales, we divide
the 45 stocks into two groups according to the information shares measurement
(Hasbrouck, 1995) and the relative liquidity in the put option and equity market, since
these characteristics are likely to be correlated with informed investors decision as to
which market to trade in. We apply the structural model to each group and compare the
results from the subsample analysis.
6.1. Information shares
For stocks traded in multiple markets, Hasbrouck (1995) develops an econometric
approach to study each markets proportional contribution to the price discovery process.
Using a modied information share approach, Chakravarty, Gulen, and Mayhew (2004)
study the contribution of option markets to price discovery. They rst calculate the implied
stock price using option trading data, and then apply the methodology in Hasbrouck
(1995) to the implied and the actual stock price processes. In this paper, we follow the
methodology in Chakravarty, Gulen, and Mayhew (2004) and measure the information
shares in the option market for the 45 sample stocks between February 20, 2007 and April

320

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

20, 2007.15 In particular, using a 1-second interval, we estimate the implied stock intraday
price using the put-call parity and the intraday put and call option prices, and then apply
Hasbrouck (1995) to estimate the option market information shares.
We then apply our main methodology to the stocks with high and low information
shares. The results are reported in Table 3 (Panel A). We expect put options to be more
informative for stocks with higher option market information shares, since by denition,
higher price discovery should be observed in the option markets for these stocks. The
results conrm our expectations. During the subsample period, short sales are not
signicant in predicting subsequent put net trade volume, but for stocks with higher option
market information shares, put net trade volume leads short sales volume. In all, these
results suggest that our main methodology is supported by and complements the
information shares methodology, since by considering explicitly the impact of short sales
volume, it allows us to study the informational role of both put option trading and short
sales in the equity market.
6.2. Liquidity
The relative liquidity between the equity and option markets may be an important factor
when investors decide which market to trade in. As discussed above, one of the drawbacks
of choosing to trade in the option market is its illiquidity. In order to address this issue, we
group the options and underlying stocks in our sample according to the relative liquidity in
the put option and short equity markets.
First we calculate the turnover put-short ratio for each stock by dividing the put option
turnover by total short sales volume during our sample period (as in Roll, Schwartz, and
Subrahmanyam, 2010). Higher put-short ratio indicates higher liquidity in the put option
market relative to the short equity market. It is not clear how the relative liquidity in the short
and put option markets affects the informed trading in these two markets, since on one hand,
in a more liquid option market, investors face less transaction costs related with illiquidity; on
the other hand, a liquid option market may also indicate more hedging-related or uninformed
trading.16 Table 3 (Panel B) suggests that rst, put option trading is not signicant in
predicting stock returns, while short sales predict subsequent stock returns in stocks with high
put-short ratio. In terms of trading volume, put net trade volume does not predict subsequent
short sales regardless of the magnitude of the put-short ratio, while short sales are signicant in
predicting put net trade volume for stocks with both high and low put-short ratios. Moreover,
the predictability of short sales is more pronounced than the predictability of put volume even
in stocks with high put-short ratio. Thus, we conclude that higher turnover in the put option
market cannot indicate more informed trading in this market.
The second liquidity measure we use is the bid-ask spread. We construct a put-short
spread ratio by dividing the average bid-ask spread in the option market by the average
bid-ask spread in the short equity market, hence a higher put-short spread ratio represents
lower liquidity in the put option market compared to the short equity market. Since higher
information asymmetry also leads to a higher bid-ask spread, it is not clear whether put
option trading contains more information for stocks with high or low put-short spread
15

We thank the referee for suggesting the analysis using information shares.
For example, Roll, Schwartz, and Subrahmanyam (2010) nd that lower delta, which implies high hedge
ratios, is correlated with higher trading volume in the option market.
16

Table 3
Subgroup analysis based on stock and option characteristics.
Explanatory variables
Lagged put net trade volume
Dependent
variable

Lag1

Lag2

Lag0

Lag1

Lagged put net trade volume

Lag2

Stock returns

Lag0

Lag1

Lag2

Lag0

Lag1

Lag2

Lagged short size


Lag0

Lag1

Panel B

Panel C

Low information shares from option markets

Low put/short volume ratio

Low put/short bidask spread ratio

0.001
0.001
0.008 0.02n 0.001
0.13
0.21
0.98 2.14 0.07
0.03n
0.009 0.004
0.02n
0.03n
4.19
1.28 0.46
2.21
2.94
0.02n
0.02n
0.006 0.008
2.66
2.98
0.56 0.78
0.003
0.006
0.38n
0.13n
0.47
1.09
46.72
14.99
High information shares from option markets

0.010
1.88
Put returns
0.06n
10.51
Put net trade volume

Short size

Lag2 Lag0

Lagged put net trade volume

Panel A

0.005
0.76
Put returns
0.06n
9.26
Put net trade volume
Short size

Lag0 Lag1

Lagged short size

0.004
0.82
0.01n
2.06
0.03n
4.66
0.01n
2.05

0.003
0.56
0.010
1.76
0.000
0.05
0.005
1.01

0.05n
8.13
0.04n
4.97

0.001 0.001
0.002
0.02n
0.000
0.74
0.8
1.17
12.79 0.12
0.06n 0.02n
0.01n
0.01n
0.000
35.82 10.99
4.81
2.35
0.08
n
n
0.03
0.01
0.01n
16.45
5.72
2.25
0.002
0.001
0.35n
1.68
0.68
194.15
High put/short volume ratio

0.02n
0.003 0.0n
2.6
0.48 3.46
0.04n
0.03n 0.065
5.52
3.81 35.25
0.004 0.008
0.45 0.97
0.39n
0.13n
59.87
18.21

Lag2

0.01n
4.06
0.01n
5.02
0.01n
2.46
0.15n
76.94

0.01n
3.42
0.06n
39.07

0.002
0.000 0.11n 0.03n 0.01n
1.23
0.22 56.34 14.9 3.95
0.02n
0.00n
0.01n
0.000
0.01n
10.51
2.78
5.21
0
5.72
0.03n
0.01n
0.01n 0.001
18.04
3.26
3.86 0.24
0.001
0.001
0.37n
0.14n
0.59
0.82
204.46
73.09
High put/short bid-ask ratio

0.002 0.00 0.06n 0.03n 0.02n


0.92 0.47 24.98 10.61 8.81
0.018 0.003
0.01n 0.004
0.01n
9.75
1.9
2.93 1.39
4.1
0.036 0.01n
0.01n 0.003
18.75
3.04
4.89 0.97
0.003 0.002
0.36n
0.14n
1.69
1.39
166.51
62.79

0.001
0.37
0.06n
31.41

0.001
0.36
0.02n
10.31
0.03n
16.91
0.01n
4.32

0.000
0.13n
0.02n
0.09
57.33
10.01
0.01n 0.001 0.003
4.39 0.49 1.33
0.01n
0.01n
5.98
3.08
0.003
0.33n
1.65
154.69

0.01n
2.39
0.01n
3.46
0.01n
2.06
0.15n
66.67

This table presents the subgroup panel regression results of the last two equations (with dependent variables put net trade volume and short size) of the following
multivariate VAR model:
rt ALrt B0zt BLzt e1,t

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

Stock returns

Lag0

Lagged short size

zt CLrt DLzt e2,t

321

where rt rs,t rc,t rp,t 0 and zt zs,t zc,t zp,t zss,t 0 . rs,t ,rc,t and rp,t represent quote revisions in the stock, call, and put market during time interval t, and zs,t ,zc,t ,zp,t
and zss,t represent net trade volume in the stock, call, put and short market during time interval t. AL3  3, B0 3  4, BL3  4, CL4  3, and DL4  4
(L 1, 2, 3) are coefcients to be estimated. Each variable is then standardized by subtracting the mean and dividing by the standard deviation of the day. Sample days are
deleted when there are less than 20 trades in the call option, put option, or stock market on that day. We use contemporaneous and three lags for the explanatory variables, and
report the regression coefcients for the contemporaneous and rst two lags with n indicating signicance at the 5% level.
We divide all the 45 NYSE sample stocks according to their stock-market related characteristics and apply the VAR model for each of the subgroups. The sample period of
Panel B and Panel C is from March 2005 to June 2007, while the sample period of Panel A is from February 2007 to April 2007.

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

322

ratio. According to Table 3 (Panel C), put option trading is signicant at predicting subsequent
short sales for stocks with higher put-short spread ratio. However, for stocks with both high
and low put-short spread ratio, short sales are signicant in predicting subsequent put option
trading, suggesting that short sales are more informative compared to put option trading
regardless of the relative bid-ask spreads in the short equity and the put option market.
In all, to further understand the different informational roles of short sales and put
option trading, we perform the subgroup analysis using the information shares
measurement (Hasbrouck, 1995) and the relative liquidity in the short equity and put
option markets. Our results suggest that short sales play a more important price discovery
role than put option trading.

7. Sub-period analysis
Prior literature (Cao, Chen, and Grifn, 2005) nds that call option trading contains
more information before company takeovers in the sense that during the preannouncement period, call option trading imbalance has greater predictability over the
next-day stock returns (takeover premiums) than stock trading imbalance, while the same
effect cannot be found during non-takeover period. In particular, they nd that prior to
announcements, buying activity is highest in the short-term out-of-the-money call options
(with the highest leverage). This suggests that the leverage advantage in option markets is
more signicant and may attract informed traders when there is a pending event that will
change the companys fundamental value. Thus, in this section, we also conduct pre-event
analysis to test whether the same effect can be found between put option trading imbalance
and short sales before a companys negative earnings announcement is released.
We check all the quarterly earnings announcements of our 45 stocks during the sample
period and calculate their unexpected earnings (the difference between actual and predicted
earnings). There are 89 negative earnings announcements in total. For each of them, we
keep three days before the announcement day and apply our structural model to the 267
option days. We expect to nd that during this pre-announcement period, put option
trading should be more informative and lead short sales volume.
The results in Table 4 are consistent with the argument above. During the three days
prior to negative earnings announcements, put net trade volume is signicant in predicting
subsequent stock market returns.17 Moreover, we nd that put net trade volume is
signicant in predicting subsequent short sales, while short sales are no longer signicant in
predicting subsequent returns and put net trade volume, suggesting that in a pre-event
period, put option trading contains more information since it is a better predictor for
subsequent stock market returns.
Our results indicate that the effect documented in Cao, Chen, and Grifn (2005) may
also be similarly found in the put option market. Cao, Chen, and Grifn (2005) show that
with a pending extreme informational event (takeover), the option market plays an
important role in the price discovery process, while stock market trading imbalances are
predictors of next day stock returns and option volume is uninformative in normal times.
Our results suggest that while over the entire sample period and without controlling for
rm-specic events there seems to be more informed trading in the short equity market,
17

The coefcient is 0.01, with a t-statistic of 1.96 for the second lag put net trade volume.

Table 4
Regression analysis of the relationship between standardized 5-minutes returns and standardized 5-minutes net trade volume of stocks, calls, puts and short sales
(3 days before earnings announcements).
Explanatory variables
Lagged stock
return

Stock returns

Lag1

Lag2

0.008 0.022
0.4
1.19
Call returns
0.15n 0.08n
7.58
3.96
Put returns
0.14n 0.12n
6.9
6.2
Stock net
0.013 0.030
trade volume
0.61 1.43
Call net trade
0.009 0.05n
volume
0.4
2.45
Put net trade 0.006 0.06n
0.28 2.67
volume
Short size
0.019 0.013
1.06
0.75

Lag1

Lag2

Lagged put
return
Lag1

0.014 0.011 0.006


1.34 1.01 0.52
0.16n 0.05n 0.08n
14.46 4.76 6.97
0.06n 0.04n 0.13n
5.73 3.48 11.52
0.013 0.004 0.018
1.09
0.35 1.55
0.008 0.020 0.002
0.64
1.73
0.15
0.023 0.001 0.020
1.89
0.05 1.72
0.003 0.006 0.013
0.28
0.55 1.26

Lag2

Lagged stock net trade Lagged call net trade


volume
volume
Lag0

0.002 0.06n
0.16
8.53
0.03n 0.05n
2.27
6.84
0.04n 0.07n
4.04 9.16
0.003
0.27
0.003
0.3
0.03n
2.9
0.001
0.15

Lag1

Lag2

Lag0

Lag1

Lagged put net trade


volume

Lag2 Lag0

Lag1

Lag2

Lagged short size

Lag0 Lag1

Lag2

0.02n 0.005 0.007 0.001 0.007 0.02n 0.011 0.01n 0.05n 0.015 0.013
2.43
0.7 1.05
0.12 1.06 2.16 1.63 1.96 5.96
1.68 1.48
0.05n 0.02n 0.05n 0.02n 0.001 0.03n 0.002 0.011 0.008 0.02n 0.004
6.83
2.76
7.27
2.85 0.08
3.5 0.28
1.55
1 2.44 0.41
0.03n 0.007 0.005 0.009 0.008 0.06n 0.02n 0.010 0.003 0.009 0.012
3.3
0.95
0.7
1.29 1.07
8.59
2.59
1.47 0.38
0.99 1.33
0.009 0.004
0.02n
0.011 0.007
0.06n 0.010
1.13
0.47
2.61
1.43
0.89
6.15
1.02
0.013 0.008
0.03n
0.006 0.009
0.009 0.03n
1.58
1.04
4.13
0.83 1.21
0.94
2.66
0.02n 0.006
0.007
0.03n 0.011
0.001 0.009
2.1 0.73
0.94
3.75
1.48
0.15
0.94
0.02n 0.001
0.005
0.01n 0.001
0.35n 0.13n
3.21 0.12
0.76
2.24
0.12
43.84 14.86

This table presents the panel regression results of the following multivariate VAR model during 3 days before the sample stocks negative earnings announcements:
rt ALrt B0zt BLzt e1,t
zt CLrt DLzt e2,t

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

Dependent
variable

Lagged call
return

where rt rs,t rc,t rp,t 0 and zt zs,t zc,t zp,t zss,t 0 . rs,t ,rc,t and rp,t represent quote revisions in the stock, call, and put market during time interval t, and
zs,t ,zc,t ,zp,t and zss,t represent net trade volume in the stock, call, put and short market during time interval t. AL3  3, B0 3  4, BL3  4, CL4  3, and
DL4  4 (L 1, 2, 3) are coefcients to be estimated. Each variable is then standardized by subtracting the mean and dividing by the standard deviation of the day.
We use contemporaneous and three lags for the explanatory variables, and report the regression coefcients for the contemporaneous and rst two lags with n
indicating signicance at the 5% level.
323

324

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

during the pre-event period, informed investors will turn to put option markets and the put
option trading becomes more informative.

8. Robustness tests
In the main analysis, we apply the structural model dened by Eqs. (1) and (2) to the
entire sample in a pooled regression. As a robustness test, we rst apply time-series
regressions to estimate the coefcients of the structural model for each individual stock,
and then cross-sectionally average the estimated coefcients.18
Table 5 reports the results of this robustness test. The results are similar to those in the
main analysis (Table 2). The results show that short sales are signicant in predicting
subsequent put option trading, but not vice versa, suggesting that short sales are indeed
more informative and lead the put option trading.
In the main analysis, we apply the 20-trade lter following Chan, Chung, and Fong
(2002), and delete option trading days with fewer than 20 trades for the stock, the most
active call, the most active put, or the short sales. After applying the lter, there are 8,520
out of 20,427 sample days left for the main analysis. In Table 6, we present the comparison
statistics of some main variables on sample and discarded option days. As expected,
trading volume and number of trades are both higher in all the markets on sample days,
since the discarded option days are deleted due to the thin trading problem. In particular,
there are on average 56 put option trades on sample days, while there are only 13 put
option trades on discarded days. Since we use 5-minute interval for our main analysis,
deleting thin trading days helps to validate the high-frequency methodology.19 However,
we acknowledge that by applying the 20-trade lter, we keep only 41.7% of the total
available option days, which may bring questions to the robustness of our test results. To
address this issue, we conduct several robustness tests.20
First, since we focus on the leadlag relationship between the put option and short
equity market, we drop all the variables related with call options from the VAR model as a
robustness test. By doing this, we avoid applying the 20-trade lter to the call option
market. After applying the 20-trade lter to the put option market only, we retain 9,441
option days for the robustness test, i.e., 46.2% of the total available option days are kept.
Table 7 presents the results of panel regressions using the 9,441 option days. The results are
similar to those in Table 2. More specically, we nd that a one standard deviation
increase in lagged short size results in a 0.008 standard deviation increase in put net trade
volume for the next 5-minute interval, while a one standard deviation increase in lagged
put net trade volume only results in a 0.003 standard deviation increase in the subsequent
short size, suggesting that there is more informed trading in the short equity market.
Second, we replace the 20-trade lter with the 15-trade lter and widen the 5-minute
interval to the 10-minute interval, and then repeat the main analysis. By doing this, 10,639
option days (52.1% of the total available trading days) are kept for the robustness test. In
unreported results, we conrm that short sales contain more information compared to the
18
As mentioned in Chordia, Roll, and Subrahmanyam (2005), this method does the opposite of the Fama and
MacBeth (1973) methodology in which coefcients from cross-sectional regressions are averaged over time.
19
As stated by Easley, OHara, and Srinivas (1998), the lack of trading dictates that the price-volume relation
for these series may not be reliable. (p. 450).
20
We thank the referee for this suggestion.

Table 5
Regression analysis of the relationship between standardized 5-minutes returns and standardized 5-minutes net trade volume of stocks, calls, puts and short sales
(cross-sectional averages of coefcients from individual time-series regressions).
Explanatory variables
Lagged stock
return

Stock returns

Lag1

0.186
0.82
Call returns
4.40n
5.25
Put returns
4.45n
5.46
Stock net trade
0.337
volume
1.88
Call net trade volume
0.84n
5.34
Put net trade volume 1.13n
4.82
Short size
0.375
1.59

Lagged put
return

Lagged stock net trade


volume
Lag0

Lag1

Lag2

Lagged call net trade


volume
Lag0

Lag1

Lag2

Lagged put net trade


volume

Lag2

Lag1

Lag2

Lag1

Lag2

Lag0

0.056
0.65
2.74n
5.21
2.75n
5.27
0.081
0.64
0.57n
3.83
0.59n
4.49
0.27n
2.47

0.004
1.24
0.18n
18.92
0.05n
10.41
0.001
0.23
0.02n
3.84
0.02n
3.41
0.001
0.31

0.001
0.19
0.08n
12.12
0.03n
7.23
0.001
0.45
0.004
0.95
0.003
0.66
0.001
0.43

0.003
1.16
0.05n
10.04
0.17n
26.12
0.005
1.81
0.02n
7.41
0.03n
5.65
0.004
1.5

0.001 0.040 0.000 0.001 0.005 0.001 0.002 0.003


0.26
3.24 0.04
0.3
1.96 0.68 1.04 1.55
0.02n
0.05n 0.03n
0.01n 0.06n
0.02n
0.01n 0.03n
5.15 12.66
5.65
2.83 20.99
7.69
5.78 12.06
0.08n 0.05n 0.03n 0.01n 0.03n
0.01n
0.01n
0.05n
23.42 14.14 6.23 2.31
8.4
3.25
3.7 15.17
0.000
0.03n
0.01n
0.003 0.001
0.12
13.11
5.58
1
0.6
0.004
0.01n
0.01n
0.03n
0.01n
1.01
4.33
2.37
10.09
2.93
0.01n
0.01n 0.01n
0.01n 0.002
3.04
5.11 2.98
2.26
0.9
0.003
0.01n 0.003
0.002 0.002
0.73
3.31 1.71
1.01
1.31

Lag1 Lag2

Lagged short size

Lag0

Lag1

Lag2

0.001 0.001 0.010 0.005 0.004


0.47 0.46 0.29 0.54 0.84
0.001
0.01n
0.03n 0.004 0.01n
0.71
3.34
4.79
1.15
4.2
0.02n
0.01n 0.005 0.006
0.01n
7.95
3.45 0.78
1.28
4.7
0.003 0.002
0.06n
0.02n
1.74 0.93
9.13
5.48
0.00n 0.002
0.002 0.001
2.15
1.42
0.72 0.19
0.04n
0.01n
0.01n 0.003
12.36
3.74
4.34
1.3
0.002 0.002
0.35n
0.14n
1.36
1.13
64.07 43.29

This table reports the cross-sectional average of coefcients from individual time-series regressions. We rst estimate the following multivariate VAR model for each
stock and then average the estimated coefcients over 45 NYSE stocks:
rt ALrt B0zt BLzt e1,t
zt CLrt DLzt e2,t
where rt rs,t rc,t rp,t 0 and zt zs,t zc,t zp,t zss,t 0 . rs,t ,rc,t and rp,t represent quote revisions in the stock, call, and put market during time interval t, and
zs,t ,zc,t ,zp,t and zss,t represent net trade volume in the stock, call, put and short market during time interval t. AL3  3, B0 3  4, BL3  4, CL4  3, and
DL4  4 (L 1, 2, 3) are coefcients to be estimated. Each variable is then standardized by subtracting the mean and dividing by the standard deviation of the day.
Sample days are deleted where there are less than 20 trades in the call option, put option, or stock market on that day. Stocks are deleted when there are less than 10
sample days. We use contemporaneous and three lags for the explanatory variables, and report the mean and the t-statistics of the regression coefcients for the
contemporaneous and rst two lags with n indicating signicance at the 5% level.

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

Dependent variable

Lagged call
return

325

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

326

Table 6
Comparison of sample days and discarded days.
Sample days (8,520)
Mean
Call options
Daily trading volume
Daily number of trades
Put options
Daily trading volume
Daily number of trades
Underlying stock
Daily trading volume
Daily number of trades
Short sales of the underlying stock
Daily trading volume
Daily number of trades

Std. Dev.

Discarded days (11,907)


Mean

Std. Dev.

3,415
88

5,333
84

903
24

2,227
26

2,557
56

3,584
51

495
13

1,653
13

10,176,307
8,544

9,696,026
3,940

6,279,569
5,299

7,232,700
2,699

1,451,010
1,814

1,250,084
1,291

875,987
1,089

912,246
764

This table compares the daily trading volume and number of trades of the call options, put options, underlying
stocks, and short sales on sample days and discarded days. The overall sample includes 45 NYSE stocks from
March 2005 to June 2007. Sample days are deleted when there are less than 20 trades in the call option, put option,
or stock market on that day.

put option trading, since the predictive power of short sales on subsequent put option
trade volume (0.008 with a t-statistic of 3.69) is robustly larger in magnitude than the
predictive power of put option on subsequent short sales (0.004 with a t-statistic of 2.98)
with these alternative lter and high-frequency intervals.
Finally, we repeat the analysis after subtracting the shorting series from the signed stock
volume series. In our main analysis (Table 2), we argue that short sales contain more
information than put options, since we observe that short sales can predict subsequent
stock returns and stock trading volume, while put option trading cannot. However, one
may argue that the predictive power of short sales over future stock returns and trading
volume might be due to the persistence of short sales and the fact that about 20% of stock
trades include short sales.21 To rule this out, we subtract the shorting series from the signed
stock volume series. We rst match trades data from TAQ with short sales data by symbol,
price, size, date, and time. After we identify the trading direction of each trade using the
Lee and Ready algorithm, we subtract the shorting volume from the total stock volume,
generating the adjusted signed stock volume series. The unreported results suggest that the
effect of short sales on subsequent stock returns (0.01 with a t-statistic of 6.97) is very
similar to our main results (Table 2). Additionally, short sales can still predict subsequent
stock trading volume (0.01 with a t-statistic of 4.24), although the effect is smaller in
magnitude when compared to our main analysis. This result suggests that the persistence of
short sales helps to but cannot explain the predictive power of short sales on subsequent
stock trading volume.
Overall, our main result that short sales contain more negative information still holds,
after performing several robustness checks.
21

We thank the reviewer for this suggestion.

Table 7
Regression analysis of the relationship between standardized 5-minutes returns and standardized 5-minutes net trade volume of stocks, puts and shorts (the equations
and variables related with call options are dropped).
Explanatory variables

Stock returns
Put returns
Stock net trade volume
Put net trade volume
Short size

Lagged put return

Lagged stock net trade volume

Lagged put net trade volume

Lag1

Lag1

Lag0

Lag0

0.002
0.48
0.223n
46.97
0.010n
2.05
0.041n
8.35
0.022n
5.26

Lag2
n

0.021
4.71
0.091n
19.4
0.004
0.88
0.028n
5.7
0.005
1.32

0.001
0.77
0.109n
61.99
0.004n
2.43
0.004n
2
0.006n
3.82

Lag2
0.001
0.5
0.044n
25.35
0.003
1.48
0.005n
2.85
0.002
1.53

Lag1
n

0.042
34.81
0.056n
43.85

0.000
0.06
0.059n
44.94
0.037n
26.95
0.019n
13.65
0.013n
11.6

Lag2
0.001
0.77
0.023n
17.69
0.019n
13.55
0.009nn
6.33
0.003n
2.54

Lag1
n

0.003
2.82
0.062n
50.57

0.001
1.1
0.019n
15.21
0.001
1.09
0.032n
25.15
0.003n
2.46

Lag2

Lagged short size


Lag0

Lag1
n

Lag2
n

0.001 0.008 0.01 0.004n


0.59
5.96 6.75
2.35
0.005n
0.006n 0.002
0.011n
4.39
3.66 1.45
6.37
0.002
0.060n
0.025n
1.28
36.46
14.4
0.008n
0.008n
0.002
6.23
4.71
1.29
0.000
0.354n
0.144n
0.39
255.8
99.52

This table presents the panel regression results of the following multivariate VAR model using 45 NYSE stocks from March 2005 to June 2007:
rt ALrt B0zt BLzt e1,t

zt CLrt DLzt e2,t

where rt rs,t rc,t rp,t 0 and zt zs,t zc,t zp,t zss,t 0 . rs,t ,rc,t and rp,t represent quote revisions in the stock, call, and put market during time interval t, and
zs,t ,zc,t ,zp,t and zss,t represent net trade volume in the stock, call, put and short market during time interval t. AL3  3, B0 3  4, BL3  4, CL4  3, and
DL4  4 (L 1, 2, 3) are coefcients to be estimated. Each variable is then standardized by subtracting the mean and dividing by the standard deviation of the day.
Sample days are deleted where there are less than 20 trades in the call option, put option, or stock market on that day. Stocks are deleted when there are less than 10
sample days. We use contemporaneous and three lags for the explanatory variables, and report the mean and the t-statistics of the regression coefcients for the
contemporaneous and rst two lags with n indicating signicance at the 5% level.

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

Dependent variable

Lagged stock return

327

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

328

9. Conclusion
This study provides an empirical comparison of the price discovery roles between put net
trade volume (buyer-initiated trading volume minus seller-initiated trading volume) and short
sales of the underlying stocks for a group of actively traded NYSE stocks and their
corresponding put options. It has been documented in the literature that both put option trading
and short sales convey valuable negative information about the underlying stocks. However, it is
Table A1
The companies stocks and near-maturity options tracked in BARDS from March 2005 to June 2009.
Stock
symbol

Stock name

Stock
symbol

Stock name

AIG

AMERICAN INTERNATIONAL
GROUP INC
ALLSTATE CORP
APPLIED MATERIALS INC
ADVANCED MICRO DEVICES
A M R CORP DEL
AMAZON COM INC

JNJ

JOHNSON & JOHNSON

JPM
KLAC
KO
LU
LXK
MER
MMM

JPMORGAN CHASE & CO


KLA-TENCOR CORP
COCA COLA CO
LUCENT TECHNOLOGIE
LEXMARK INTERNATIONAL INC
NEW
MERRILL LYNCH & CO INC
3M CO

MO
MRK
MSFT
MWD
MXO
NEM
NOK
NXTL
ORCL
OVTI
PFE
PG
QCOM
S
SEBL

ALTRIA GROUP INC


MERCK & CO INC
MICROSOFT CORP
MORGAN STANLEY
MAXTOR CORP
NEWMONT MINING CORP
NOKIA CORP
NEXTEL COMMUNICATIONS INC
ORACLE CORP
OMNIVISION TECHNOLOGIES
PFIZER INC
PROCTER & GAMBLE CO
QUALCOMM INC
SPRINT NEXTEL CORP
SIEBEL SYSTEMS INC

SNDK

SANDISK CORP

SUNW
TWX
TXN
TYC
UPS
USG
VZ
WFC

SUN MICROSYSTEMS INC


TIME WARNER INC
TEXAS INSTRUMENTS INC
TYCO INTERNATIONAL LTD NEW
UNITED PARCEL SERVICE INC
U S G CORP
VERIZON COMMUNICATIONS INC
WELLS FARGO & CO NEW

XOM
YHOO

EXXON MOBIL CORP


YAHOO INC

ALL
AMAT
AMD
AMR
AMZN
BAC
BCGI
BEAS
BRCM
BSX
CA
CHTR
CMCSK
COF
CSCO
DELL
DNA
DUK
EBAY
EMC
ETFC
FNM
FRE
FRX
GE
GM
GP
HAL
HDI
HPQ
IBM
IMCL
INTC

BANK OF AMERICA CORP


BOSTON COMMUNICATION GROUP
INC
BEA SYSTEMS INC
BROADCOM CORP
BOSTON SCIENTIFIC CO
COMPUTER ASSOCIATES INTL INC
CHARTER COMMUNICATIONS INC
COMCAST CORP NEW
CAPITAL ONE FINANCIAL CORP
CISCO SYSTEMS INC
DELL INC
GENENTECH INC
DUKE ENERGY CORP NEW
EBAY INC
E M C CORP MA
E TRADE FINANCIAL CORP
FEDERAL NATIONAL MORTGAGE
ASSN
FEDERAL HOME LOAN MORTGAGE
CORP
FOREST LABS INC
GENERAL ELECTRIC CO
GENERAL MOTORS CORP
GEORGIA-PACIFIC CORP
HALLIBURTON COMPANY
HARLEY DAVIDSON INC
HEWLETT PACKARD CO
INTERNATIONAL BUSINESS MACHS
COR
IMCLONE SYSTEMS INC
INTEL CORP

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

329

not clear which market attracts more informed investors. Our results show that short sales
predict subsequent stock and put option returns, while put option imbalance only predicts its
own future returns, suggesting that short sales convey more negative information about the
stocks than put option trading. The analysis of the leadlag relationship between put option
imbalance and short sales also conrms the above argument since the predictive power of short
sales on subsequent put option is larger in magnitude than the predictive power of put option on
subsequent short sales. Subsample tests are also conducted to provide further support for the
leading informational role of short sales. The only different results are found in the analysis
before a companys negative unexpected earnings are released. With a pending event that can
drive down stocks fundamental value, more informed trading is observed in the put option
market, since in this case put option imbalance is signicant in predicting subsequent short sales,
and short sales lose the predictive power. Our results suggest that the short market seems to be
more important in conveying negative information, while the put option market may be more
critical and contain more information before certain corporate events. Overall, this paper
contributes to our understanding of the price discovery roles of the equity and options markets,
and their different functions in improving nancial market efciency.

Appendix A
See Table A1.
Table B1
Summary statistics of stocks in BARDS and OptionMetrics.
Panel A. NYSE and NASDAQ stocks

Turnover (%)
BARDS
OptionMetrics
Volume (in thousands)
BARDS
OptionMetrics

Min.

25th

No.

Mean

Std. Dev.

67
1,399

1.124
0.891

2.369
4.910

0
0

0.361
0.275

67
1,399

11,561
1,398

17,133
4,517

0
0

2,692
141

No.

Mean

Std. Dev.

Min.

25th

45
822

0.727
0.757

0.816
4.869

0.052
0

0.321
0.273

45
822

8,830
1,400

12,933
3,986

23
0

2,626
168

Median

75th

Max.

0.625
0.491

1.169
0.907

307.204
901.933

5,939
361

12,979
1,049

267,041
267,041

75th

Max.

0.510
0.459

0.828
0.786

19.424
901.933

5,299
425

10,042
1,187

223,848
244,746

Panel B. NYSE stocks

Turnover (%)
BARDS
OptionMetrics
Volume (in thousands)
BARDS
OptionMetrics

Median

Panels A (Panel B) reports the cross-sectional averages of trading volume and turnover for all the NYSE/
NASDAQ (NYSE) stocks with options in BARDS and OptionMetrics as of the end of 2002, when the BARDS
database was constructed. Turnover is daily trading volume divided by total shares outstanding, and Volume is the
daily number of shares traded in thousands of dollars. The summary statistics include the number of stocks (No.),
mean values (Mean), standard deviation (Std. Dev.), minimum and maximum (Min. and Max.) and quartiles
(25th, Median, and 75th) of the two variables for each group.

330

X. Hao et al. / Journal of Financial Markets 16 (2013) 308330

Appendix B
SeeTable B1.

References
Anthony, J.H., 1988. The interrelation of stock and options market trading volume data. Journal of Finance 43,
949964.
Asquith, P., Pathak, P.A., Ritter, J.R., 2005. Short interest, institutional ownership, and stock returns. Journal of
Financial Economics 78, 243276.
Battalio, R., Schultz, P., 2011. Regulatory uncertainty and market liquidity: the 2008 short sale bans impact on
equity option markets. Journal of Finance 66, 20132053.
Black, F., 1975. Fact and fantasy in the use of options. Financial Analysts Journal 31, 3641 & 6172.
Boehmer, E., Jones, C.M., Zhang, X., 2008. Which shorts are informed? Journal of Finance 63, 491527.
Boehmer, E., Jones, C.M., Zhang, X., 2011, Shackling short sellers: the 2008 shorting ban, Working Paper,
Columbia University.
Boehmer, E., Wu, J., Short selling and the price discovery process. Review of Financial Studies, 10.1093/rfs/
hhs097, in press.
Cao, C., Chen, Z., Grifn, J., 2005. Informational content of option volume prior to takeovers. Journal of
Business 78, 10731109.
Chakravarty, S., Gulen, H., Mayhew, S., 2004. Informed trading in stock and option markets. Journal of Finance
59, 12351257.
Chan, K., Chung, Y.P., Fong, W., 2002. The informational role of stock and option volume. Review of Financial
Studies 15, 10491075.
Chordia, T., Roll, R., Subrahmanyam, A., 2005. Evidence on the speed of convergence to market efciency.
Journal of Financial Economics 76, 271292.
Desai, H., Ramesh, K., Thiagarajan, S.R., Balachandran, B.V., 2002. An investigation of the informational role
of short interest in the Nasdaq market. Journal of Finance 57, 22632287.
Diamond, D.W., Verrecchia, R.E., 1987. Constraints on short-selling and asset price adjustment to private
information. Journal of Financial Economics 18, 277311.
Diether, K., Lee, K., Werner, I., 2009. Short-sale strategies and return predictability. Review of Financial Studies
22, 25312570.
Easley, D., OHara, M., Srinivas, P.S., 1998. Option volume and stock prices: evidence on where informed traders
trade. Journal of Finance 53, 431465.
Fama, E., MacBeth, J., 1973. Risk, return, and equilibrium: empirical tests. Journal of Political Economy 81 (3),
607636.
Grundy, B.D., Lim, B., Verwijmeren, P., 2012. Do option markets undo restrictions on short sales? Evidence from
the 2008 short-sale ban. Journal of Financial Economics 106, 331348.
Hasbrouck, J., 1991. Measuring the information content of stock trades. Journal of Finance 46, 179207.
Hasbrouck, J., 1995. One security, many markets: determining the location of price discovery. Journal of Finance
50, 11751199.
Lee, C., Ready, M., 1991. Inferring trade direction from intraday data. Journal of Finance 46, 733746.
Pan, J., Poteshman, A.M., 2006. The information in option volume for future stock prices. Review of Financial
Studies 19 (3), 871908.
Roll, R., Schwartz, E., Subrahmanyam, A., 2010. O/S: the relative trading activity in options and stock. Journal
of Financial Economics 96, 117.