Why does the Option to Stock Volume Ratio Predict Stock Returns?

Li Ge†, Tse-Chun Lin‡, and Neil D. Pearson§

August 9, 2015

Abstract

We use data on signed option volume to study which components of option volume predict stock
returns and resolve the seemingly inconsistent results in the literature. We find no evidence that
trades related to synthetic short positions in the underlying stocks contain more information than
trades related to synthetic long positions. Purchases of calls that open new positions are the
strongest predictor of returns, followed by call sales that close out existing purchased call
positions. Overall our results indicate that the role of options in providing embedded leverage is
the most important channel why option trading predicts stock returns.

JEL Classification: G12, G13, G14

Keywords: Option trading volume, stock return predictability, information, leverage

                                                            
*
We are thankful for valuable comments from an anonymous referee, Utpal Bhattacharya, Jonathan Brogaard,
Kewei Hou, Michelle Lowry, Dimitris Papanikolaou, Joshua Pollet, and seminar participants at Cubist Systematic
Strategies, National Taiwan University, and the University of Hong Kong. Tse-Chun Lin gratefully acknowledges
research support from the Faculty of Business and Economics at the University of Hong Kong and the Research
Grants Council of the Hong Kong SAR government. All errors remain our responsibility.
†
Monash Business School, Monash University. Tel.: +61 (3) 9903-2123. E-mail: li.ge@monash.edu.
‡
Faculty of Business and Economics, The University of Hong Kong. Tel.: +852 2857-8503. E-mail:
tsechunlin@hku.hk.
§
Department of Finance, University of Illinois at Urbana-Champaign. Tel.: +1 (217) 244-0490. E-mail:
pearson2@illinois.edu.
 Why does the Option to Stock Volume Ratio Predict Stock Returns?

Abstract

We use data on signed option volume to study which components of option volume predict stock
returns and resolve the seemingly inconsistent results in the literature. We find no evidence that
trades related to synthetic short positions in the underlying stocks contain more information than
trades related to synthetic long positions. Purchases of calls that open new positions are the
strongest predictor of returns, followed by call sales that close out existing purchased call
positions. Overall our results indicate that the role of options in providing embedded leverage is
the most important channel why option trading predicts stock returns.

JEL Classification: G12, G13, G14

Keywords: Option trading volume, stock return predictability, information, leverage

1. Introduction

There has been long-standing interest in whether and how investors who possess valuable

private information might use derivatives markets to execute trades based on that information.

Recently, Roll, Schwartz, and Subrahmanyam (2010) introduced to the literature a measure of

the ratio of option trading volume to stock trading volume (O/S) and empirically explored some

of its properties. Their findings suggest that cross-sectional and time-series variation in O/S

might be driven by informed trades. Johnson and So (2012) pursue this point and find that O/S

predicts the returns of the options’ underlying stocks over a one-week horizon, with high O/S

predicting negative returns. They argue that “the negative relation between O/S and future

returns is driven by short-sale costs in equity markets, which make option markets an attractive

venue for traders with negative news.” In their model explaining their empirical findings,

Johnson and So (2012) emphasize the same role of options and argue that “the costs associated

with short-selling make informed traders more likely to use options for bad signals than for good

ones and, as a result, high O/S indicates negative private information and low O/S indicates

positive private information.”

The O/S ratio is constructed from unsigned option trading volume and can be high

because option traders are establishing or closing out positions that involve either synthetic short

or long exposure to underlying stock prices. Given that purchases of calls and sales of puts

comprise about one-half of option trading volume, Johnson and So’s (2012) finding and

interpretation are surprising in light of previous research showing that option trades that create

synthetic long exposure predict positive stock returns. For example, Pan and Poteshman (2006)

find that high and low values of the put-call ratio computed from directional information about

order flow predict low and high returns, respectively, suggesting that investors with positive

1 
 
information about stock prices trade on that information using options.1 One possible resolution

of the apparent inconsistency might lie in the fact that the signed option order flow data used by

Pan and Poteshman (2006) were not available to market participants, in contrast to the O/S ratio.

However, a recent empirical literature shows that price-based measures such as the

implied volatility spread and implied volatility skew are able to predict future stock returns

(Cremers and Weinbaum, 2010; Xing, Zhang, and Zhao, 2010), while other researchers explore

the return predictability of these price-based measures around earnings announcements,

corporate events, and events such as the initiation of analyst coverage and analyst

recommendation changes (Jin, Livnat, and Zhang, 2012; Lin and Lu, 2015; Chan, Ge, and Lin,

2013; and Hayunga and Lung, 2014). This literature provides evidence that measures constructed

from option prices contain positive as well as negative information about future stock prices. A

similar conclusion can be drawn from Cao, Chen, and Griffin (2005) who sign option order flow

using a version of the Lee-Ready algorithm. In light of this evidence, it is puzzling why the O/S

ratio predicts negative stock returns. As indicated above, purchases of calls and sales of puts

comprise about half of option trading volume, and they predict positive returns.

In this paper, we reconcile the apparent inconsistency by exploiting data from the

International Securities Exchange (ISE) on the signed option volumes of public customers to

study which components of option order flow predict underlying stock returns. Similar to the

OptionMetrics data used to compute the O/S ratio and in contrast to the data used by Pan and

Poteshman (2006), the data we use are available to any market participants willing to pay the

license fee. The data allow us to study whether option trades that provide synthetic short

exposure to underlying stock prices are more informative than those that provide synthetic long

                                                            
1
The idea that the embedded leverage provided by options might be appealing to investors interested in long
exposure to underlying stock prices dates at least to Black (1975).

2 
 
exposure, as claimed by Johnson and So (2012). Our data are also disaggregated according to

whether the signed volume comes from the opening of new option positions or the closing of

existing positions.

Using these data, we find no evidence that option trades related to synthetic short

positions in the underlying stocks are more informative than option trades that provide synthetic

long exposure. The most informative option volume is that due to trades that open new bought

call positions, which we refer to as open buy call volume. This result holds in all of the

subsamples we examine and is robust to the specifications of the regression models. In our main

results, the point estimates indicate that the second most informative component of option

volume consists of option volume stemming from sell trades that close previously purchased call

positions. Even though these are sales of calls, they do not involve synthetic short positions in

the underlying stocks because they involve the closing of previously establish long positions, not

the establishment of new short positions. Meanwhile, open sell put volume also predicts positive

returns.

Unsurprisingly, we also find evidence that volume associated with new synthetic short

positions in the underlying stock negatively predicts returns. But the finding that the O/S ratio

negatively predicts returns is not driven so much by these components of volume as by the result

that the unwinding of bought call positions also negatively predicts returns. That is, volume that

opens synthetic short positions negatively predicts returns, and volume that opens synthetic long

positions positively predicts returns, with larger coefficients. In addition, volume resulting from

the unwinding of bought call positions also negatively predicts returns. Therefore, there are more

components of option volume negatively predicting returns than positively predicting returns.

Taken as a whole, our evidence indicates that option trades related to synthetic long positions in

3 
 
the underlying stock either contain about the same quantity of or more information than option

trades related to synthetic short positions. Given the embedded leverage and information content

are similar in both synthetic long and short positions, we can conclude that the short sale costs do

not play the most important role in explaining the stock return predictability of options trading

volume. The embedded leverage is the most important channel why O/S predicts stock returns.

Meanwhile, compared with the opening trades, we find that closing trades are generally

less informative, which is consistent with the findings in Pan and Poteshman (2006). The

exception to the statement that closing trades are generally less informative is the significant

predictability from volumes due to unwinding of bought call positions. This finding is

unsurprising given that open buy call volume has the strongest predictive power among the

various opening volumes. If open buy call volumes are the strongest predictor among the

opening volumes, it is to be expected that the trading volumes that unwind these positions would

be the strongest predictor among the closing volumes.

We obtain these results using weekly Fama-Macbeth (1973) regressions that include

signed O/S measures (in decile ranks) constructed from the signed volume data plus a wide range

of control variables. Our main results are based on a sample that excludes option expiration

weeks because a large fraction of expiration week option volume is due to “rolling” of option

positions into later expiring contracts, which might not reflect new information. When we

examine expiration weeks separately, we do not find a significant relation between the overall

O/S measure based on unsigned volumes of both calls and puts used by Johnson and So (2012)

and returns. However, our signed O/S measures constructed from signed volumes during the

expiration weeks continue to predict returns during the subsequent week.

4 
 
 These results suggest that the higher embedded leverage of options, compared with that

of stocks, is the most important channel why options market attracts informed trading which

predicts stock returns. However, the closing relative bid-ask spread is monotonically increasing

in the option’s leverage, which is linked to option moneyness. The higher transaction costs for

out-of-the money (OTM) options might lead some traders to capitalize on their private

information by trading at-the-money (ATM) or in-the-money (ITM) options, depending on the

content of the private information. Therefore, which moneyness category of options has the

strongest predictability is an empirical question. For example, both Cao, Chen, and Griffin (2005)

and Pan and Poteshman (2006) find that the predictability of signed options volume is stronger

for OTM options. However, Johnson and So (2012) show that predictability of unsigned O/S is

stronger for ITM options.

Consistent with Cao, Chen, and Griffin (2005) and Pan and Poteshman (2006), we find

evidence that signed O/S measures constructed from higher leverage options have stronger stock

return predictability. In particular, OBC/S, CSC/S, OSP/S, and CBP/S constructed from volumes

of OTM options have stronger predictability than the same measures constructed from volumes

of ATM or ITM options. OBP/S does not predict returns for ITM options, but does predict

returns for ATM and OTM options, with the estimated coefficient being slightly smaller (but

more significant) for OTM options. In contrast, ITM OSC/S predicts returns, but ATM and OTM

OSC/S do not.

In the model of Miller (1977), investor disagreement, combined with short-sale

constraints, leads to overvaluation which in turn implies negative abnormal returns. Options do

not fully eliminate short-sale constraints and costs because many professional investors are

subject to investment mandates that do not permit the use of options. These considerations

5 
 
suggest the possibility that O/S measures based on unsigned call and put volumes might predict

negative returns if overall O/S is highly correlated with measures of disagreement. However,

when we include disagreement measures based on options trading, stock trading, and analyst

forecast dispersion in our regression specifications, the measures of disagreement based on ISE

options volumes do not predict returns, and the ability of the measures of option order flow to

predict returns is unaffected.

In addition, we analyze the predictability of signed options trading volume from different

investor types (firm proprietary traders and public customers) and trade size buckets (small,

medium, and large trades from public customers) reported by the ISE. The proportions of trading

volume from firm proprietary traders, customer small, medium and large trades are 33%, 51%,

4%, and 12%, respectively. We find that option volumes from customer small trades have the

strongest return predictability in both magnitude and statistical significance, followed by those

from customer large trades. The volumes from customer medium trades show weaker

predictability which could be due to the small volumes in this category. Consistent with the

findings in Pan and Poteshman (2006), the signed volumes from firm proprietary traders are not

informative about future stock returns.

We also find some evidence that the signed O/S measures have higher stock return

predictability when contemporaneous stock returns are in the middle 20% of the distribution of

returns over the past previous months. The results are consistent with those in Conrad, Hameed,

and Niden (1994) and Gervais, Kaniel, and Mingelgrin (2001), who show that the predictive

power of (unsigned) extreme equity trading volume on subsequent returns is strongest when

contemporaneous returns are non-extreme.

6 
 
 Finally, we show that our results are robust to variations in the construction of the signed

O/S measures and the use of raw values of O/S instead of decile ranks.

Our paper contributes to the literature of informed options trading. We find that

embedded leverage plays an important role, which is at least as important as short-sale costs

argued in Johnson and So (2012), in explaining why options trading volume predict future equity

returns. This is consistent with the implications of the theoretical model in Easley, O’Hara, and

Srinivas (1998) that informed investors prefer to trade in options market due to higher leverage.

Our results thus reconcile the findings in Pan and Poteshman (2006) and Johnson and So (2012).

The remainder of the paper is organized as follows. The next section describes the data

we use, focusing on the signed option volume data. Section 3 first verifies that the Johnson and

So (2012) result that the O/S measure based on unsigned option volumes predicts returns is

found in our data. It then presents the main results regarding which components of option

volume predict returns. Subsection 3.3 presents evidence about how the return predictability is

related to option moneyness. Section 4 first addresses the possibility that O/S predicts returns

because it proxies for disagreement. It then discusses whether the predictive power differs among

different investor types and trade size buckets and how the predictability could be affected by

non-extreme stock returns. Section 5 presents the results of various robustness tests and Section

6 briefly concludes.

2. Data and descriptive statistics

The main data we use are from the International Securities Exchange (ISE) Open/Close

Trade Profile, which provides daily buy and sell trading volume for each option series traded at

the ISE, disaggregated by whether the trades open new option positions or close existing

7 
 
positions.2 Because we focus on the ability of options volume to predict the cross section of

stock returns, we restrict the sample to include only individual equity options, dropping the

options on exchange-traded funds (ETFs) and indexes.3 Trades reported in the ISE Open-Close

profile represent about 30% of the total trading volume in individual equity options during our

sample period spanning May 2005 to August 2012.4

The ISE data include volumes due to trades of both firm proprietary traders and public

customers. Market maker trades are only reported indirectly, in that in most cases a market

maker is on the other side of the trade by a public customer or firm proprietary trader. Trades in

which market makers are on both sides are not included in the data. The public customer trading

volumes are further disaggregated into volumes due to small, medium, and large trades, defined

by cutoffs determined by the ISE. For each option series and trader type or trade size bucket, the

option trading volume data are broken down into four categories: volume from buy orders that

open new purchased positions (open buy volume), volume from sell orders that open new written

positions (open sell volume), volume from buy orders that close existing written positions (close

buy volume), and volume from sell orders that close existing purchased positions (close sell

volume). Our main analyses combine the volumes from the different trader types and trade size

buckets, while in Section 4.2 we look at each type and bucket separately. For each underlying

stock and volume type we also aggregate the volume for the different call and put series, yielding

four call volumes and four put volumes for each underlying stock. Considering the different

characteristics, i.e., open/close, buy/sell, and call/put, we have eight categories of options volume,
                                                            
2
A detailed description of the ISE data is available at http://www.ise.com/market-data/products/put-call-data/ise-
open-close-trade-profile/.
3
An exception to this statement is that Section 5 of the Internet Appendix reports the results of regressions that
explore whether O/S measures for ETFs predict ETF returns. We find that there is no significant predictability in
this sample of ETFs. This result is consistent with the argument in Subrahmanyam (1991) that there is likely to be
less information-based trade in the markets for index derivatives than in the markets for individual securities.
4
The ISE Open-Close data do not include trades between two market makers and include both sides of the trade in
cases in which public customers or firm proprietary traders trade directly with each other.

8 
 
which are open buy call (OBC), open sell call (OSC), close buy call (CBC), close sell call (CSC),

open buy put (OBP), open sell put (OSP), close buy put (CBP), and close sell put (CSP).

We obtain the total option trading volumes across all option exchanges from

OptionMetrics and obtain the daily returns and trading volume data for the underlying stocks

from the Center for Research in Security Prices (CRSP). Option implied volatilities used as

control variables in some regressions are from OptionMetrics. Balance sheet data used to

construct book-to-market ratios are from COMPUSTAT.

The overall O/S variable is defined as the ratio of total option trading volume (in number

of shares) to total stock trading volume during a certain period, e.g., a week.5 Analogous call and

put measures, C/S and P/S, are defined using call and put option trading volumes, respectively.

We use OMO/S, OMC/S, and OMP/S to denote O/S, C/S, and P/S calculated using total volume

across all option exchanges from OptionMetrics. Similarly, ISEO/S, ISEC/S, and ISEP/S

represent O/S, C/S, and P/S calculated using trading volumes from the ISE Open/Close Trade

Profile. ISEO/S is further broken down into the signed components, i.e., OBC/S, OSC/S, CBC/S,

CSC/S, OBP/S, OSP/S, CBP/S, and CSP/S, where, for example, OBC/S is the ratio of open buy

call volume to stock volume. We refer to all of these option volume measures as O/S ratios. Both

these variables and the control variables we use are listed and defined in Appendix A.6

Most of our analyses use weekly decile O/S measures that are constructed as follows. For

each stock and week, we first compute the various weekly O/S ratios by combining the

appropriate option volumes with weekly stock volumes. For each week we then partition the O/S

                                                            
5
The option trading volume provided by the ISE is in terms of the number of contracts. To calculate the O/S ratio
we convert the option trading volume data from numbers of contracts to numbers of underlying shares. Ordinarily,
each contract is on 100 shares of stock, though sometimes (e.g., following stocks splits) the terms of the contracts
are adjusted and the number of shares is different from 100. Our calculations of option trading volume reflect these
adjustments to the terms of the option contracts.
6
Section 1 of the Internet Appendix presents evidence about the determinants of signed O/S measures. The results
are generally consistent with the findings in Roll, Schwartz, and Subrahmanyam (2010).

9 
 
ratios of the various stocks into deciles and set the decile variables equal to 1, 2, …, 10,

accordingly, where 1 corresponds to the smallest decile and 10 to the largest. These decile

variables are indicated by appending “Dec” to the variable, for example DecOMO/S and

DecOBC/S denote the decile variables constructed from OMO/S and OBC/S, respectively. The

use of decile variables is consistent with Johnson and So (2012) and prevents the results from

being driven by a small number of extreme values. The robustness checks also present results

based on alternative definitions of the weekly O/S measures.

[Table 1 about here]

Table 1 presents summary statistics for the weekly O/S ratios used in the main results, the

one-week cumulative abnormal return CAR(1) used as the dependent variable in many of our

regressions, and the weekly values of other variables used as controls. The dependent variable

CAR(1) is the cumulative abnormal return in the week following the week in which the O/S

measures are calculated. We use the CRSP value-weighted market return as the benchmark in

calculating the abnormal return.

The mean of ISEO/S is about one-third of the mean of OMO/S, consistent with the ISE’s

share of total option trading volume. Comparison of the means and standard deviations of

ISEC/S and ISEP/S reveals that call volume is on average both larger and more variable than put

volume. The O/S ratios computed from the open and close volumes reveal that open trading

volumes are two to three times as large as close trading volumes, which is unsurprising because

option positions can be held to expiration. For example, OBC/S has a mean of 0.01 which is

about three times of the mean of CBC/S, 0.003. The abnormal return CAR(1) has a mean of

0.071% and a standard deviation of 6.619%. The abnormal return from the same week as the O/S

measures, CAR(0), is used as a control variable and unsurprisingly is of similar magnitude to

10 
 
CAR(1). The Amihud (2002) illiquidity measure (Amihud), Deceqvol, Decopvol, Impl_vol,

B/M, Size, Momen, and Skewness are used as control variables in our regression analyses. These

variables are defined in the variable list in Appendix A. Table 1 also includes summary statistics

for the measures of disagreement that are discussed and used in Subsection 4.1 below.

[Table 2 about here]

Table 2 shows the correlations among the various measures of option trading volume and

the two measures of disagreement based on option trading volumes. Several things are worth

noting. First, even though the volumes reported in the ISE data account for only about 30% of

total option trading volumes the correlation between OMO/S and ISEO/S is high, 0.61,

suggesting that ISE trading volumes are representative of total option trading volumes. Second,

both OMC/S and OMP/S are highly correlated with OMO/S. Meanwhile, OMC/S and OMP/S

have a positive correlation of 0.45, indicating that call and put volumes often move together.

Third, opening buy volumes are highly correlated with opening sell volumes in the same option

class—the correlations between OBC/S and OSC/S and OBP/S and OSP/S are both 0.76.

Opening buy (sell) call volumes are also positively correlated with opening buy (sell) put

volumes. To the extent that the option volume measures capture the private information held by

investors, it is clear that investors often disagree. Fourth, open volumes are highly correlated

with close volumes in the same option class. Some of this may be due to “rolling” of positions

into later expirations. Finally, close volumes are positively correlated, but not as positively

correlated as open volumes.

[Table 3 about here]

We further explore the positive dependence among the variables through a non-

parametric analysis. In Panels A to C of Table 3, we sort our sample into deciles by OMO/S,

11 
 
OMC/S, and OMP/S, respectively, where decile 1 includes the smallest values and decile 10 the

largest. Then for each decile, we present the average decile of each of the eight components of

ISEO/S, that is the average values of the O/S decile variables. Panels D to F are similar but show

the average raw value of each of the eight components of ISEO/S. There is a clear pattern that

the average deciles of each of the components of ISEO/S are monotonically increasing with

OMO/S, OMC/S, and OMP/S. For example, when OMO/S is in decile 10, each of the

components is on average in either decile 8 or 9. The monotonic relations hold regardless of

whether the component of ISEO/S represents buying or selling calls or puts, imply that different

investors are trading in different directions at the same time. Thus, when the variables indicating

positive information, such as OBC/S, are high (low), the variables indicating negative

information, such as OBP/S, typically are also high (low).

3. Which components of option order flow predict returns?

We exploit the buy/sell and open/close information in the ISE data to study which

components of option volume predict returns and therefore understand the ability of the overall

O/S ratio to negatively predict returns. Lakonishok, Lee, Pearson, and Poteshman (2007) show

that only a small fraction of trades in individual equity options are parts of complicated strategies

such as straddles, strangles, and spreads. Thus, it is reasonable to interpret most option volume,

especially opening volume, as containing directional information about future stock prices.7

3.1. Regression analyses showing that O/S measures predict future stock returns

                                                            
7
We hypothesize that informed investors trade on their private information in the options market; thus their trading,
in particular their signed trading volume, predicts future equity returns. Investors can also trade for hedging
purposes, in which case we would find no return predictability. If investors use complicated strategies, part of the
trading volume may have a relation with future equity return which is opposite to that predicted by the informed
trading story. This only makes it less likely to find evidence of predictability. In fact, there could be a large
proportion of noise trading; however, we still manage to find supporting evidence from the part of options trading
which contains information.

12 
 
 We begin by verifying that the overall O/S ratio predicts future stock returns during our

sample period. Following Johnson and So (2012) who examine returns and Roll, Schwartz, and

Subrahmanyam (2010) who examine the absolute values of returns, we estimate weekly cross-

sectional regressions of the form

CAR(1) = α  β1DecOMO/S + γ1Amihud + γ 2 CAR(0) + γ 3 Deceqvol + γ 4 Decopvol

(1)
+ γ 5 Impl_vol + γ 6 B/M + γ 7 Size + γ 8 Momen + γ 9Skewness   .

The dependent variable CAR(1) is the cumulative abnormal return (CAR) in the week

subsequent to the week from which trading volumes are used to construct OMO/S are taken. The

main independent variable of interest is DecOMO/S, the decile variable constructed from

OMO/S.

Following Johnson and So (2012) and Xing, Zhang, and Zhao (2010), we control for the

Amihud (2002) illiquidity ratio (Amihud), the cumulative abnormal return in the week in which

O/S measures are constructed (CAR(0)), a decile variable constructed from the change in equity

trading volume (DecΔeqvol), a decile variable constructed from the change in option trading

volume (DecΔopvol), option implied volatility (Impl_vol), the book-to-market ratio (B/M), firm

size (Size), six-month momentum (Momen), and historical return skewness (Skewness). As

argued in Battalio and Schultz (2011) and Grundy, Lim, and Verwijmeren (2012), the short-sale

ban during the financial crisis affected option market trading volumes. As a result, we exclude

observations from the last quarter of 2008.

[Table 4 about here]

In Table 4, we obtain the coefficients by estimating Eq. (1) for each week and averaging

the resulting coefficient estimates. In Panel A, we restrict our sample to those observations with

the O/S ratios in non-expiration weeks, where the expiration week is defined as the week that

contains the third Friday in each month. Panel B shows the regressions for the expiration weeks.

13 
 
The t-statistics in parentheses below the average coefficient estimates are based on Newey-West

(1987) standard errors.8 We consider non-expiration and expiration weeks separately because a

large fraction of option trading volume in expiration weeks appears to stem from “rolling”

positions into options with later expirations (Muravyev, 2015), and the option trading volume

during non-expiration and expiration weeks might have different information content.

The results in the first column of Panel A confirm Johnson and So’s (2012) finding that

the overall O/S ratio strongly and negatively predicts the subsequent week’s stock returns.

Models 2 and 3 present results for specifications in which DecOMO/S is replaced by similar

variables DecOMC/S and DecOMP/S based on OptionMetrics call and put volumes, respectively.

These variables also negatively predict future returns, with coefficients about the same size as

the coefficient on DecOMO/S. This is consistent with the descriptive statistics showing that the

raw values of OMO/S, OMC/S, and OMP/S are highly correlated with each other.

Models 4 through 6 of Panel A show the results for specifications in which DecOMO/S is

replaced by DecISEO/S, DecISEC/S, and DecISEP/S, the decile O/S measures computed from

the ISE data. They also predict returns, although the coefficients and t-statistics are slightly

smaller than those in Models 1 through 3. These results suggest that ISE trading volume is

representative of the whole market in terms of its ability to predict future stock returns. The

slightly smaller coefficients are also consistent with the fact that the ISE data are only part of

total trading volume so that O/S measures based on ISE volume can be interpreted as O/S

measures based on aggregate option trading volume plus errors.

Comparing Panels A and B, one can see that the unsigned O/S measures DecOMO/S,

DecISEO/S, DecOMP/S, and DecISEP/S lose their predictive power during expiration weeks,

                                                            
8
In the Fama-MacBeth (1973) regressions, we use four lags when the O/S measures are from non-expiration weeks
(Panel A), and three lags when the O/S measures are from expiration weeks (Panel B).

14 
 
while the measures DecOMC/S and DecISEC/S constructed from call option volume become

positive predictors during expiration weeks. As shown in Table 5 below, the predictive power of

the various signed components of trading volume changes during expiration weeks. Recognizing

that the unsigned O/S measures aggregate signed trading volumes, these changes in the

predictive power of signed trading volumes shown in Table 5 lead to changes in the predictive

power of the unsigned O/S measures.9

3.2. Regression analyses showing that measures of signed option volume predict future stock

returns

This section presents the results of cross-sectional regressions showing that signed O/S

ratios predict future stock returns.10 The most general specification we estimate is

CAR(1) = α  β1DecOBC/S + β 2 DecOSC/S + β 3 DecCBC/S + β 4 DecCSC/S

+ β 5 DecOBP/S + β 6 DecOSP/S + β 7 DecCBP/S + β 8 DecCSP/S + β 9 DecOMO/S
(2)
+ γ1Amihud + γ 2 CAR(0) + γ 3 Deceqvol + γ 4 Decopvol + γ 5 Impl_vol
+ γ 6 B/M + γ 7 Size + γ 8 Momen + γ 9Skewness  ε

where the independent variables of interest are the decile O/S measures computed from signed

option volume, i.e., DecOBC/S, DecOSC/S, DecCBC/S, DecCSC/S, DecOBP/S, DecOSP/S,

DecCBP/S, and DecCSP/S. Specifications that include all eight of these variables allow us to

                                                            
9
For example, the most notable change in the predictive power of the signed O/S measures across the two panels of
Table 5 is that the variable DecCSC/S, which is a negative predictor during non-expiration weeks, loses its
predictive power during expiration weeks. Thus three of decile variables constructed from the components of
ISEC/S, these being, DecOBC/S, DecOSC/S and DecCSC/S, are related to returns during non-expiration weeks, but
only two, DecOBC/S and DecOSC/S, are related to returns during expiration weeks. Because DecOBC/S,
DecOSC/S and DecCSC/S are positive, negative, and negative predictors, respectively, the net effect is that
DecISEC/S is a negative predictor during non-expiration weeks. Then because DecOBC/S is a stronger predictor
than DecOSC/S (and DecCSC/S loses its predictive power) during expiration weeks, the net effect is that
DecISEC/S is a positive predictor during expiration weeks. The sign of the coefficient on DecOMC/S changes with
the change in the sign of the coefficient on DecISEC/S because DecOMC/S and DecISEC/S are highly correlated. 
10
Section 2 of the Internet Appendix provides additional evidence that the signed O/S measures predict returns
using sequential double-sorted portfolios. Section 4 of the Internet Appendix shows that the signed O/S measures
before earnings announcements and other corporate events (SEOs, M&A activities, repurchases, stock splits, and
dividends) predict the two-day cumulative abnormal returns on the announcement or event date and the following
date.

15 
 
examine the effect of each of the components of option volume, controlling for the effect of the

others. As before, the dependent variable CAR(1) is the cumulative abnormal return in the week

subsequent to the week of the option volumes used to construct the independent variables. All of

the specifications include the same control variables as used in Eq. (1), and some also include

DecOMO/S as an additional control variable.

[Table 5 about here]

Following the approach in Table 4, we estimate a cross-sectional regression for each

week and report the means of the weekly regression coefficients along with the associated t-

statistics based on Newey-West (1987) standard errors. As before, we drop observations from the

last quarter of 2008. Panel A of Table 5 presents the regression results excluding option

expiration weeks. In each of Models 1 to 4, we include two out of the eight measures of signed

option trading volume that are expected to contain information with opposite implications for

stock returns, and exclude the other measures. For example, Model 1 includes DecOBC/S and

DecOSC/S and excludes the other measures of option trading volume. Consistent with our

expectations, DecOBC/S is positively related to stock returns in the next week, while a higher

value of DecOSC/S predicts a lower return. In Model 2, DecCSC/S is significantly negatively

related to stock returns. In Model 3, DecOBP/S is negatively associated with future returns,

consistent with a role of options in allowing traders to avoid costs of shorting the underlying

stocks. Model 4 finds no evidence that closing put trades predict stock returns.

Models 5 and 6 include the four components of signed option trading volume for calls

and puts, respectively. Model 7 shows the results for a regression with all eight components, and

we focus on Model 8 which further controls for the overall OMO/S using the decile variable

DecOMO/S.

16 
 
 As shown in the last column in Panel A (Model 8), the opening trades that establish

synthetic long positions in the underlying stock positively predict subsequent stock returns.

DecOBC/S has an estimated coefficient of 0.039 with a very large t-statistic of 5.24 and is the

strongest predictor of subsequent stock returns. DecOSP/S also positively predicts stock returns,

with a positive coefficient of 0.018 and a t-statistic of 2.68.

As expected, the opening trades that establish synthetic short positions in the underlying

stock predict negative stock returns. DecOSC/S has an estimated coefficient of −0.029 with a

very large t-statistic of −5.18 and DecOBP/S has a negative coefficient of −0.019 with a t-

statistic of −2.80. The negative coefficient on DecOBP/S is consistent with the hypothesis that

options play a role in allowing investors to avoid short-sale constraints or costs. The negative

coefficient on DecOSC/S is also consistent with this, though one should recognize that many

written calls are part of covered call strategies that involve net long exposures to the underlying

stock (Lakonishok, Lee, Pearson, and Poteshman, 2007).

Turning to the closing trades, we find that closing sells of calls negatively predict

subsequent stock returns as shown by the negative coefficient on DecCSC/S of −0.034 with a t-

statistic of −4.38. This is the second strongest predictor in magnitude after DecOBC/S. Even

though the variable DecCSC/S captures volume due to sales of calls, the coefficient estimate is

not evidence of the importance of synthetic short positions because the closing sells of calls close

out previously established long positionsthere was never a synthetic short position. None of

the other closing volume variables is significantly related to future stock returns.

The finding that closing volumes generally are weaker predictors than opening volumes

is consistent with the argument in Pan and Poteshman (2006, pp.904) that closing trades are

expected to be less informative. They point out that “traders can only use information to close

17 
 
positions if they happen to have appropriate positions open at the time they become informed,”

making it less likely that closing volumes reflect information. Consistent with this, in Table 5

Panel A the coefficients on DecCBC/S and DecCBP/S are of the expected signs but small in

magnitude and insignificant, and the coefficient on DecCSP/S is close to zero in the main

specifications Models 7 and 8. The only exception to the general pattern that closing trades show

little predictability is DecCSC/S, which significantly predicts returns. The ability of DecCSC/S

to predict returns seems likely to be driven by the combination of two factors. First, in our ISE

sample open interest of calls is about 1.5 times that of puts, suggesting that more investors hold

calls than puts. Since investors can close their positons only if they have existing open positions,

it is expected that call closing volumes are more informative than put closing volumes. Second,

as shown in Panel A of Table 5, DecOBC/S has the strongest predictive power among the

opening volumes. Given these two factors, it is not surprising that the closing sell call volumes

that unwind opening call trades contain the most information compared with other closing

volumes.

These results of Model 8 are robust to the regression specification. The results from

Model 7 that does not include DecOMO/S are almost identical to those from Model 8, and those

from Models 5 and 6 are very similar. Looking across the results for all models, one can see that

the four variables DecOBC/S, DecOSC/S, DecCSC/S, and DecOBP/S are significant at the 1%

level in every specification in which they appear. DecOSP/S, the fifth signed volume measure

that is significant in Models 7 and 8, is insignificant in only Model 3, where it carries a t-statistic

of 1.07. Some of the measures of closing volume that are insignificant in Models 7 and 8 are

significant at the 10% level in Models 4 and 6 that do not include all of the decile signed O/S

measures.

18 
 
 These results provide no evidence that option trades related to synthetic short positions in

the underlying stocks contain more information than option trades related to synthetic long

positions. The two strongest predictors, DecOBC/S and DecCSC/S, measure volume due to

establishing or unwinding long call positions. Considering the point estimates in Model 8, the

coefficients on the open volume measures associated with synthetic long positions in the

underlying stocks, DecOBC/S and DecOSP/S, are 0.039 and 0.018, which are larger than or

almost the same size as the coefficients on the open volume measures associated with synthetic

short positions, DecOSC/S and DecOBP/S, which are 0.029 and 0.019. The absolute

magnitude of the coefficient on DecOBC/S is statistically larger than that on DecOBP/S, with the

absolute mean difference of 0.020 and t-statistic of 1.935. 11 The other highly significant

coefficient is the coefficient of 0.034 on DecCSC/S. The overall O/S ratio negatively predicts

returns not because it only captures synthetic short positions, but rather because more

components of option volume negatively predict returns than positively predict returns due to the

fact that trading volume stemming from the unwinding of bought call positions (CSC/S)

negatively predicts returns. This evidence indicates that options’ role in allowing investors with

positive information about future stock prices to obtain synthetic leverage is important. Based on

the point estimates the embedded leverage plays a greater role compared to the role of options in

helping investors to circumvent short-sale constraints.

Panel B of Table 5 presents results for the same set of specifications estimated using the

observations from expiration weeks. Even though much of the expiration-week trading volume is

likely due to “rolling” positions into later expiring options, the results in Panel B are similar to

                                                            
11
The difference between coefficients of DecOBC/S and DecOBP/S is tested by (|Mean(DecOBC/S)| 
|Mean(DecOBP/S)|)/ sqrt(Stderr(DecOBC/S)2 + Stderr(DecOBP/S)2). Other pairs are mostly insignificantly different.
For example, the coefficients of DecOSC/S and DecOSP/S have a mean difference of 0.011 with a t-statistic of
1.237.

19 
 
those in Panel A. The four variables measuring opening volumes, DecOBC/S, DecOSC/S,

DecOBP/S, and DecOSP/S, have significant coefficients with the expected signs. In fact, the

point coefficients on these opening volume variables are somewhat larger than the corresponding

point estimates in Panel A. Differences are that in Models 7 and 8 of Panel B the coefficient on

DecCSC/S is now small and insignificant, and the coefficient on DecCBP/S is 0.021 and highly

significant, while it is negative and insignificant in Models 7 and 8 in Panel A. Note that closing

buys of puts do not open new synthetic short positions in the underlying stock but rather close

out previously established synthetic long positions, i.e., they close out written put positions.

These results are also evidence that options’ role in allowing investors with positive information

about future stock prices to obtain synthetic leverage is just as important as their role in helping

investors with negative information avoid short-sale costs, and perhaps more important.

3.3. Option leverage and O/S return predictability

The results above and our interpretation of them as evidence that the synthetic leverage

provided by options is important suggest that options providing higher leverage will attract more

informed trades, causing volume in these options to more strongly predict stock returns.

However, the high transactions costs of high leverage options might lead some traders to

capitalize on their private information by trading low leverage options. Therefore, which type of

options has the highest predictability is an empirical question. For example, Cao, Chen, and

Griffin (2005) and Pan and Poteshman (2006) find that the predictability of signed options

volume is stronger for high leverage options, while Johnson and So (2012) present evidence that

the predictability of unsigned O/S is stronger for low leverage options. In this subsection, we

explore whether the return predictability of the unsigned and signed O/S measures are sensitive

to option leverage.

20 
 
 [Table 6 about here]

To capture the effect of option leverage, we classify each option into one of the three

moneyness categories according to option delta, as in Bollen and Whaley (2004). A call option is

defined to be out-of-the-money (OTM) if 0.02 < delta ≤ 0.375; at-the-money (ATM) if 0.375 <

delta ≤ 0.625; and in-the-money (ITM) if 0.625 < delta ≤ 0.98. A put option is defined to be out-

of-the-money (OTM) if –0.375 < delta ≤ –0.02; at-the-money (ATM) if –0.625 < delta ≤ –0.375;

and in-the-money (ITM) if –0.98 < delta ≤ –0.625. We then calculate firm-week unsigned and

signed O/S measures using the options in each moneyness category and estimate the Fama-

MacBeth (1973) regression as specified in Eq. (2) separately for each moneyness category. We

require the weekly number of firms in each moneyness category to be at least 30.

In Panel A of Table 6, we summarize the mean bid-ask spread for the options in each

moneyness category. The bid-ask spread increases with the level of leverage, with OTM options

having an average relative spread of 35.98% and ITM options having a much lower average

relative spread of 7.93%.

For the signed O/S measures, we find evidence that trading volume in higher leverage

options is more strongly predictive, consistent with results in Cao, Chen, and Griffin (2005) and

Pan and Poteshman (2006). Table 6 Panel B shows that DecOBC/S, DecCSC/S, DecOSP/S, and

DecCBP/S in the OTM category are stronger predictors of future stock returns than the same

variables in the ATM or ITM categories, both in the magnitudes of the point estimates and in

their statistical significance. Because the deltas of the OTM options are by definition smaller

than those of the ATM or ITM options, the differences in predictability are even greater on a

delta-equivalent basis.

21 
 
 Turning to the other variables, DecOBP/S does not predict returns for ITM options, but

does predict returns for ATM and OTM options, with the estimated coefficient being slightly

smaller (but more significant) for OTM options. Recognizing that the deltas of OTM options are

smaller than those of ATM options, the similarity of the coefficients implies that on a delta-

equivalent basis OTM volume is more strongly predictive than ATM volume. For DecOSC/S,

the coefficient estimates for the ITM category is larger than the estimates in the ATM and OTM

categories, and significant, while the coefficients for the ATM and OTM categories are not. This

result might be due to the differences in deltas in the different categories or higher relative bid-

ask spreads for OTM options that cause traders with information to avoid using these options.

In Panel C of Table 6, we find that DecOMO/S constructed from volumes of ITM options

has slightly stronger return predictability as compared to OTM and ATM DecOMO/S. This is

consistent with Johnson and So (2012)’s finding that unsigned O/S has stronger predictive power

if the options leverage is lower, although we do not find the same monotonicity in option

leverage that they do. Meanwhile, the ordering of the magnitudes of the coefficients is reversed

in our ISE options sample. This contrast between the results for the signed and unsigned O/S

measures highlights the limitations of using the unsigned O/S measures.

4. Additional evidence

This section provides additional evidence about the information contained in option

trades. Subsection 4.1 addresses an alternative explanation that O/S may capture disagreement

leading to overvaluation (see, e.g., Miller, 1977), which may lead to subsequent negative returns.

Subsection 4.2 examines the trades executed by different investor types or falling into different

trade size buckets. Subsection 4.3 analyzes how the predictability of signed O/S measures differs

depending on whether past stock returns were extreme or non-extreme.

22 
 
4.1. Disagreement

Previous sections present results consistent with the hypothesis that the overall O/S ratio

predicts future returns because various components of it reflect the private information held by

option traders. However, a possible alternative hypothesis is that the overall O/S ratio is a

measure of disagreement rather than a measure of informed trading. For example, Table 2 shows

that the correlations between OBC/S and OSC/S and between OBP/S and OSP/S are both 0.76,

indicating that investors are trading in different directions. In Table 3, we find that whenever one

of the decomposed variables is in a high decile, all of the others are also in high deciles. Thus, it

is possible that the overall O/S ratio measures disagreement among investors.

In Miller (1977), if there exist both short-sale constraints and differences of information

or opinions, the stock price reflects only the views of investors with positive information or

opinions. As a result, the stock will be overvalued and have negative subsequent returns. Options

ameliorate short-sale constraints but cannot eliminate them because the investment mandates of

many professional equity money managers prevent them from using options. If the overvaluation

due to disagreement is corrected over a short horizon such as one week, we may see a negative

relation between disagreement and future returns.

To explore this hypothesis, we construct two measures of disagreement, as follows. We

first define a measure of disagreement based on ISE opening volumes termed ODisagmt:

ODisagmt  (OBC/S  OSP/S)  (OSC/S  OBP/S)

(3)
 | (OBC/S  OSP/S)  (OSC/S  OBP/S) | .

Here OBC/S + OSP/S measures open volume that provides positive exposure to the stock price,

and OSC/S + OBP/S measures open volume that provides negative exposure to the stock price.

ODisagmt is maximized if OBC/S + OSP/S = OSC/S + OBP/S. It is zero if all of the volume

23 
 
provides either positive or negative exposure, i.e. if either OBC/S + OSP/S = 0 or OSC/S +

OBP/S = 0. In this case there is no disagreement.

Similarly, a measure of disagreement based on both opening and closing volumes is

OCDisagmt  (OBC/S  CBC/S  OSP/S  CSP/S)  (OSC/S  CSC/S  OBP/S  CBP/S)

(4)
 | (OBC/S  CBC/S  OSP/S  CSP/S)  (OSC/S  CSC/S  OBP/S  CBP/S) |

where the sum OBC/S + CBC/S + OSP/S + CSP/S measures the opening and closing volume

that provides positive exposure to the stock price, and the sum OSC/S + CSC/S + OBP/S +

CBP/S measures the opening and closing volume that provides negative exposure to the stock

price.

The summary statistics in Table 1 show that the mean of OCDisagmt, 0.039, is smaller

than but close to the mean of ISEO/S, 0.043. Table 2 shows that the disagreement measures are

highly correlated with the measures of option trading volume. In particular, they are very highly

correlated with the O/S measure based on ISE volume, ISEO/S—the correlations between

ODisagmt and ISEO/S and between OCDisagmt and ISEO/S are 0.91 and 0.99, respectively. To

the extent that investor disagreement can cause overvaluation and thus negative future returns,

these high correlations indicate that this mechanism is a plausible explanation or partial

explanation for why the overall O/S ratio predicts negative stock returns.

In addition to the above two measures constructed using ISE options volume, we also

consider disagreement measures based on stock trading. Following Danielsen and Sorescu

(2001), Boehme, Danielsen, and Sorescu (2006), and Chatterjee, John, and Yan (2012), we use

the idiosyncratic volatility of stock returns as one proxy for disagreement. We also follow Chang,

Cheng, and Yu (2007) and use turnover rate as another stock-trading-based measure of

disagreement. Finally, we use analyst dispersion, which is the proxy for disagreement proposed

by Diether, Malloy, and Scherbina (2002).

24 
 
 [Table 7 about here]

Table 7 reports the results of regressions in which we incorporate decile versions of the

disagreement measures into Eq. (2) and conduct similar Fama-MacBeth (1973) regressions of

CAR(1) on the disagreement measures and the components of ISEO/S. We only report results

using the non-expiration weeks and drop the observations from the fourth quarter of 2008.

Models 1, 3, 5, and 7 show the results using DecODisagmt and Models 2, 4, 6, and 8 show the

results using DecOCDisagmt. We include idiosyncratic volatility in Models 1 and 2, turnover in

Models 3 and 4, and analyst dispersion in Models 5 and 6. Models 7 and 8 include either

DecODisagmt or DecOCDisagmt as well as all of the other three proxies. In these regressions,

only turnover is significantly related to the subsequent stock returns. Controlling for the

disagreement measures, the predictive power of the ISEO/S decomposed variables remains the

same as in Panel A of Table 5.12 These results make it clear that the negative return predictability

of overall O/S that we show in Table 4 is not due to disagreement per se.

These results do not rule out the possibility that overvaluation due to disagreement is

corrected slowly, perhaps over several weeks or even months rather than within one week.

Regardless, they imply that the ability of O/S to predict returns shown in Johnson and So (2012)

and Table 4 above is not caused by the disagreement channel.

4.2. Analysis for different types of investors and trade size buckets

Our main results focus on the aggregate options trading volume from all investor types

and trade size buckets. In this subsection, we examine whether the return predictability of the

decile O/S measures differs across investor types and trade size buckets.

                                                            
12
In Section 3 of Internet Appendix, we find similar results using disagreement measures lagged either one, two,
three, or four weeks.

25 
 
 The ISE trading volumes are classified into four types, i.e., volumes from firm

proprietary investors, volumes from customer small trades (trades of less than 100 contracts

each), customer medium trades (trades of between 101 and 199 contracts each), and customer

large trades (trades of over 200 contracts each). We calculate firm-week decile signed O/S

measures using the trading volume from each type and trade size bucket. Table 8 Panel A repeats

the analysis in Table 5 Panel A for each category individually, while Panel B reports the results

of a specification that includes the signed volumes from all four categories in one regression.

These results are mixed. Option volumes from customer small trades generally, but not

uniformly, provide the strongest return predictability in both magnitude and statistical

significance.13 The volumes from customer medium and large trades show some predictability

but not as much as that from customer small trades. These results are not surprising because

sophisticated traders who desire to establish large options positions can do this using multiple

small trades, and large trades may be negotiated in which liquidity-taking traders are forced to

reveal their identities and trading interest. Most ISE trading volume, about 51%, stems from

customer small trades, while customer medium and large trades account for only 4% and 12%

respectively.

[Table 8 about here]

Similar to Pan and Poteshman (2006), we also find that the signed volumes from firm

proprietary traders are not informative about future stock returns. One possible explanation is

that firm proprietary traders who possess private information have access to actual leverage

while trading in the stock market and do not need the synthetic leverage available in the option

                                                            
13
For example, based on the regression in Panel B of Table 8, customer small trades are statistically different from
customer medium trades for DecOBC/S, DecOSC/S and DecOSP/S, with the respective coefficient differences of
0.024 (t-statistic = 2.12), –0.029 (t-statistic = –2.81), and 0.032 (t-statistic = 3.17).

26 
 
market. Alternatively, they might make trading profits via complicated spread and volatility

strategies, the components of which do not contain directional information about stock prices.

4.3. The interaction of signed O/S measures and non-extreme stock returns

Prior literature shows that the predictive power of unsigned extreme stock trading volume

is strongest when contemporaneous returns are non-extreme (Conrad, Hameed, and Niden, 1994;

Gervais, Kaniel, and Mingelgrin, 2001). In this subsection, we examine how the

contemporaneous returns affect the predictability of options trading volume. We define non-

extreme returns similarly to Gervais, Kaniel, and Mingelgrin (2001). For each firm-week, we

define a dummy variable “Midret” which equals 1 if the return during the week the O/S variables

are calculated is in the middle 20% of the distribution of returns over the previous six months,

and 0 otherwise. The regression setting follows Panel A of Table 5 except that the specifications

include the “Midret” dummy interacted with each of the decile signed O/S measures. The results

are shown in Section 6 of the Internet Appendix.

With the 20% window, we find some evidence that the predictability is stronger on days

with non-extreme returns, consistent with Conrad, Hameed, and Niden (1994) and Gervais,

Kaniel, and Mingelgrin (2001). For example, the interactions of DecOBC/S and DecCBC/S with

“Midret” are significantly positive, indicating that they have stronger predictability when

contemporaneous returns are non-extreme. This result is also consistent with Roll, Schwartz, and

Subrahmanyam’s (2010) argument that the option volume’s return predictability is stronger

when the private information has not been reflected in the contemporaneous stock returns. We

find qualitatively similar but statistically weaker result when we define the non-extreme returns

using a wider band from the middle of the returns distribution. For example, the interaction term

27 
 
of DecOBC/S and Midret becomes insignificant when we use middle 35% of the distribution of

returns over the previous six months as the non-extreme returns.

5. Robustness checks

In Sections 3 and 4, we used the weekly cross-sectional decile ranks of the O/S measures

as the main independent variables. To take into account the time-series variation of options

trading volume, here we employ several alternative O/S measures.

[Table 9 about here]

Using open buy call volume as an example, following Johnson and So (2012) we first

construct ∆OBC/S as the difference between OBC/S and its average over the prior six months,

scaled by the average. Second, to account for within-stock volatility of the O/S measures, we

construct a standardized OBC/S measure std∆OBC/S, defined as the difference between OBC/S

and its average over the prior six months, scaled by the standard deviation over the prior six

months. Each week, firms are then sorted into deciles based on the cross-section of weekly

∆OBC/S or std∆OBC/S, giving us decile variables Dec∆OBC/S (Model 1) and Decstd∆OBC/S

(Model 2). Third, we take into account the time-series variation in OBC/S within each firm using

the decile variable DecΩOBC/S (Model 3). This is constructed by ranking OBC/S relative to the

firm’s OBC/S time series over the prior six months and then sorting each firm-week into an

ΩOBC/S decile, where a firm-week with its OBC/S above the 90th percentile of the firm’s own

OBC/S distribution is ranked into decile 10, for example. By doing so, the decile rank of

ΩOBC/S is determined by the firm’s own time-series of O/S. The signed measures for the other

components of option trading volume are defined in the same way. Table 9 reports the results of

Fama-Macbeth (1973) regressions using these alternative O/S measures.

28 
 
 The results in Table 9 indicate that a higher OBC/S, relative to its recent time series, is

related to a higher future equity return. Similarly, the new decile variables constructed from

OSC/S, CSC/S, and OBP/S also show statistically significant predictability, as expected. Thus,

our results hold when O/S measures are constructed according to either cross-sectional or time-

series ranks.

[Table 10 about here]

In addition, instead of using decile ranks of signed O/S measures as in our main results,

we obtain qualitatively similar results using their raw values (Table 10). A one standard

deviation increase in raw OBC/S (OBP/S) leads to 1.209% increase (0.539% decrease) in the

next week’s stock return, while using decile variables a one notch increase in DecOBC/S

(DecOBP/S) leads to 0.039% increase (0.019% decrease) in the next week’s stock returns.

As shown in Sections 7 and 8 of the Internet Appendix, our main results are robust to

using an alternative two-stage regression setting and to using a shorter two-day return interval

rather than the one-week horizon used in the main results. Examining longer horizons, the return

predictability of decile signed O/S measures remains significant in the first few weeks and

becomes insignificant beyond one month.

6. Conclusion

Roll, Schwartz, and Subrahmanyam (2010) construct the O/S ratio from option and stock

trading volume and present evidence that cross-sectional and time-series variation in the O/S

ratio is due to variation in informed trading. Pursuing this idea, Johnson and So (2012) find that

the O/S ratio negatively predicts the returns of options’ underlying stocks over a one-week

horizon and claim that this is due to the role of options in ameliorating short-sale costs. In this

29 
 
paper, we take advantage of the ISE Options Open/Close data on signed option volume to study

which components of option volume predict stock returns.

We construct eight categories of option volume from the ISE data, which are open buy

call (OBC), open sell call (OSC), close buy call (CBC), close sell call (CSC), open buy put

(OBP), open sell put (OSP), close buy put (CBP), and close sell put (CSP), and from them eight

corresponding signed O/S measures. OBC volume is the strongest predictor of returns, followed

by CSC volume. Even though the CSC volume involves sales of calls, the results for the decile

variable DecCSC/S are evidence that synthetic long positions contain information because the

CSC trades close out existing purchase call positions but do not involve synthetic short positions

in the underlying stocks. Overall, our results indicate that the role of options in providing

embedded leverage is at least as important as their role in ameliorating short-sale constraints and

costs. The O/S ratio negatively predicts returns because more components of option volume

negatively predict returns than positively predict returns, due to the fact that trading volume

stemming from the unwinding of bought call positions negatively predicts returns.

Besides, we show that for most categories of option volume the signed O/S measures

most strongly predict future returns for OTM options. This is consistent with the findings in Cao,

Chen, and Griffin (2005) and Pan and Poteshman (2006). We also find that option trading

volume contains information about future stock returns by showing that including measures of

disagreement in the regressions does not mitigate the return predictive power of the components

of O/S, despite the fact that disagreement measures constructed using ISE data are highly

correlated with O/S.

Our results mainly come from the impact of option volumes from public customers,

where customer small trades show the strongest effect. Trading volumes from firm proprietary

30 
 
traders are not informative. We also find some evidence that the predictive power of signed O/S

measures on next-week stock returns is greater when the contemporaneous stock return is non-

extreme. The main results are robust to using alternative definitions of O/S measures.

31 
 
Appendix A

This appendix defines the variables used in the cross-sectional regressions.

Variable Name Definition

OMO/S Ratio of total options volume in OptionMetrics to stock volume

OMC/S Ratio of call options volume in OptionMetrics to stock volume

OMP/S Ratio of put options volume in OptionMetrics to stock volume

ISEO/S Ratio of total options volume in ISE to stock volume

ISEC/S Ratio of ISE call options volume to stock volume

ISEP/S Ratio of ISE put options volume to stock volume

OBC/S Ratio of ISE open buy call volume to stock volume

OSC/S Ratio of ISE open sell call volume to stock volume

CBC/S Ratio of ISE close buy call volume to stock volume

CSC/S Ratio of ISE close sell call volume to stock volume

OBP/S Ratio of ISE open buy put volume to stock volume

OSP/S Ratio of ISE open sell put volume to stock volume

CBP/S Ratio of ISE close buy put volume to stock volume

CSP/S Ratio of ISE close sell put volume to stock volume

Most of the regressions use decile versions of the above variables, denoted
DecOMO/S, DecOMC/S, DecOBC/S, etc. With one exception, the decile
variables are constructed by ranking each of the “raw” variables relative to
the cross-section of the variable on the same date. The exception is the
decile variables DecΩOBC/S, DecΩOSC/S, etc. used in Model 3 of Table
9. These are constructed by ranking the corresponding “raw” O/S variable
relative to the firm’s own time series of the raw variable over the prior six
months and then sorting each firm-week into a decile, where a firm with
the raw variable above the 90th percentile of its own distribution is ranked
into decile 10, for example.

CAR(1) The underlying stock's cumulative abnormal return in the week

subsequent to the week of the volumes used to calculate the O/S measures
used as explanatory variables. The CRSP value-weighted market return is

32 
 
 used as the benchmark in calculating the abnormal returns. CAR(1) is in
percent.

Amihud Amihud illiquidity ratio in the same week as the O/S measures. We
multiply the value by one billion for ease of presentation.

CAR(0) The underlying stock's cumulative abnormal return in the week of the
volumes used to calculate the O/S measures used as explanatory variables.
The CRSP value-weighted market return is used as the benchmark in
calculating the abnormal returns. CAR(0) is in percent.

Δeqvol Change in equity trading volume, defined as the difference between total
equity volume traded in the week of the O/S measures and the weekly
average equity trading volume over the prior six months, scaled by this
average. Deciles of Δeqvol, denoted DecΔeqvol, are used in the
regressions.

∆opvol Change of option trading volume, defined as the difference between total
OptionMetrics trading volume in the week of the O/S measures and the
weekly average option trading volume over the prior six months, scaled
by this average. Deciles of Δopvol, denoted DecΔopvol, are used in the
regressions.

Impl_vol The underlying stock’s average daily implied volatility during the one
month prior to the week of the O/S measures. The daily implied volatility
is calculated as open interest weighted implied volatility for all options
traded on that day.

B/M Natural logarithm of book-to-market ratio measured at the firm's last

quarterly financial statement report date.

Size Natural logarithm of market capitalization measured at the firm's last

quarterly financial statement report date.

Momen Momentum is calculated as the underlying stock's cumulative abnormal

return over the past six months, where we use CRSP value-weighted
market return as the benchmark in calculating the abnormal return.
Momen is in percent.

Skewness The historical skewness is calculated using daily return during the one
month prior to the week of the O/S measures.

ODisagmt ODisagmt is a measure of disagreement computed from ISE opening

volumes:

ODisagmt = (OBC/S + OSP/S) + (OSC/S + OBP/S)

 | (OBC/S + OSP/S)  (OSC/S + OBP/S) |

33 
 
 The regressions use a decile version of this variable, DecODisagmt.
OCDisagmt OCDisagmt is a measure of disagreement computed from both opening
and closing ISE volumes:

OCDisagmt = (OBC/S + CBC/S + OSP/S + CSP/S)

+ (OSC/S + CSC/S + OBP/S + CBP/S)
 | (OBC/S + CBC/S + OSP/S + CSP/S)
 (OSC/S + CSC/S + OBP/S + CBP/S) |

The regressions use a decile version of this variable, DecOCDisagmt.

IdioVol Idiosyncratic volatility is calculated as the standard deviation of the daily

abnormal return, based on CAPM model, over the six months up to and
including the week of the O/S measures.

Turnover Turnover over rate is calculated as the daily trading volume (scaled by the
number of shares outstanding), averaged over the six months up to and
including the week of the O/S measures.

Analyst Dispersion We calculate analyst forecast dispersion from the IBES detail file with the
forecast period of one year. For each firm-week, we measure the analyst
dispersion as the standard deviation of all analyst earnings forecasts
during the previous six months, scaled by the stock price at the end of
week 0.

Midret For each firm-week, the dummy variable “Midret” equals 1 if the stock
return during the week that O/S measures are calculated is in the middle
20% of the distribution of returns over the previous six months, and 0
otherwise.

34 
 
References

Amihud, Y., 2002. Illiquidity and stock returns: cross-section and time-series effects. Journal of
Financial Markets 5, 31–56.
Battalio, R., Schultz, P., 2011. Regulatory uncertainty and market liquidity: the 2008 short sale
ban's impact on equity option markets. Journal of Finance 66, 2013–2053.
Black, F., 1975. Fact and fantasy in the use of options. Financial Analysts Journal 31, 36–72.
Boehme, R., Danielsen, B., Sorescu, S., 2006. Short sale constraints, differences of opinion, and
overvaluation. Journal of Financial and Quantitative Analysis 41, 455–487.
Cao, C., Chen, Z., Griffin, J., 2005. Informational content of option volume prior to takeovers.
Journal of Business 78, 1073–1109.
Chan, K., Ge, L., Lin, T., 2013. Informational content of options trading on acquirer
announcement return. Journal of Financial and Quantitative Analysis, forthcoming.
Chang, E., Cheng, J., Yu, Y., 2007. Short-sales constraints and price discovery: evidence from
the Hong Kong market. Journal of Finance 62, 2097–2121.
Chatterjee, S., John, K., Yan, A., 2012. Takeovers and divergence of investor opinion. Review of
Financial Studies 25, 227–277.
Conrad, J., Hameed, A., Niden, C., 1994. Volume and autocovariances in short-horizon
individual security returns. Journal of Finance 49, 1305–1329.
Cremers, M., Weinbaum, D., 2010. Deviations from put–call parity and stock return
predictability. Journal of Financial and Quantitative Analysis 45, 335–367.
Danielsen, B., Sorescu, S., 2001. Why do option introductions depress stock prices? A study of
diminishing short sale constraints. Journal of Financial and Quantitative Analysis 36,
451–484.
Diether, K., Malloy, C., Scherbina, A., 2002. Differences of opinion and the cross section of
stock returns. Journal of Finance 57, 2213–2241.
Easley, D., O'Hara, M., Srinivas, P., 1998. Option volume and stock prices: evidence on where
informed traders trade. Journal of Finance 53, 431–465.
Fama, E., MacBeth, J., 1973. Risk, return, and equilibrium: empirical tests. Journal of Political
Economy 81, 607–636.
Gervais, S., Kaniel, R., Mingelgrin, D., 2001. The high-volume return premium. Journal of
Finance 56, 877–919.
Grundy, B., 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, 331–348.
Hayunga, D., Lung, P., 2014. Trading in the options market around financial analysts' consensus
revisions. Journal of Financial and Quantitative Analysis 49, 725–747.
Jin, W., Livnat, J., Zhang, Y., 2012. Option prices leading equity prices: do option traders have
an information advantage? Journal of Accounting Research 50, 401–432.

35 
 
Johnson, T., So, E., 2012. The option to stock volume ratio and future returns. Journal of
Financial Economics 106, 262–286.
Lakonishok, J., Lee, I., Pearson, N., Poteshman, A., 2007. Option market activity. Review of
Financial Studies 20, 813–857.
Lin, T., Lu, X., 2015. Why do options prices predict stock returns? Evidence from analyst
tipping. Journal of Banking and Finance 52, 17–28.
Livnat, J., Mendenhall, R., 2006. Comparing the post-earnings announcement drift for surprises
calculated from analyst and time series forecasts. Journal of Accounting Research 44,
177-205.
Miller, E., 1977. Risk, uncertainty, and divergence of opinion. Journal of Finance 32, 1151–1168.
Muravyev, D., 2015. Order flow and expected option returns. Journal of Finance, forthcoming.
Newey, W., West, K., 1987. A simple, positive semi-definite, heteroskedasticity and
autocorrelation consistent covariance matrix. Econometrica 55, 703–708.
Pan, J., Poteshman, A., 2006. The information in option volume for future stock prices. Review
of Financial Studies 19, 871–908.
Roll, R., Schwartz, E., Subrahmanyam, A., 2010. O/S: The relative trading activity in options
and stock. Journal of Financial Economics 96, 1–17.
Subrahmanyam, A., 1991. A theory of trading in stock index futures. Review of Financial
Studies 4, 17–51.
Xing, Y., Zhang, X., Zhao, R., 2010. What does the individual option volatility smirk tell us
about future equity returns? Journal of Financial and Quantitative Analysis 45, 641–662.

36 
 
 Table 1. Summary statistics

Summary statistics for weekly O/S measures and other firm characteristics. Variables are defined in Appendix A. The statistics
reported are for the “raw” (not decile) values of the various O/S and disagreement variables.

Variable No. Obs. Mean Std Q1 Median Q3

OMO/S 557,846 0.136 0.274 0.029 0.072 0.161
OMC/S 557,845 0.083 0.225 0.016 0.041 0.094
OMP/S 557,846 0.053 0.102 0.008 0.024 0.062
ISEO/S 558,629 0.043 0.415 0.006 0.018 0.049
ISEC/S 546,155 0.026 0.261 0.003 0.010 0.028
ISEP/S 517,207 0.019 0.212 0.002 0.006 0.020
OBC/S 546,155 0.010 0.156 0.001 0.003 0.010
OSC/S 546,155 0.009 0.095 0.001 0.003 0.009
CBC/S 546,155 0.003 0.018 0.000 0.001 0.003
CSC/S 546,155 0.004 0.035 0.000 0.001 0.004
OBP/S 517,207 0.007 0.128 0.000 0.002 0.007
OSP/S 517,207 0.007 0.070 0.000 0.002 0.006
CBP/S 517,207 0.002 0.018 0.000 0.000 0.002
CSP/S 517,207 0.003 0.036 0.000 0.000 0.002
CAR(1) (%) 558,770 0.071 6.619 -2.549 -0.034 2.577
Amihud 558,765 5.723 77.457 0.165 0.550 2.078
CAR(0) (%) 558,863 0.096 6.687 -2.557 -0.024 2.605
Δeqvol 554,467 -0.072 0.838 -0.449 -0.180 0.136
∆opvol 555,351 0.067 2.950 -0.656 -0.326 0.206
Impl_vol 544,202 0.491 0.221 0.342 0.442 0.584
B/M 558,863 -1.461 3.284 -1.431 -0.863 -0.360
Size 558,863 7.979 1.638 6.813 7.872 9.090
Momen (%) 554,688 3.056 31.631 -12.127 2.388 17.469
Skewness 558,747 0.137 0.907 -0.325 0.117 0.583
ODisagmt 504,733 0.026 0.31 0.002 0.009 0.028
OCDisagmt 504,733 0.039 0.368 0.004 0.016 0.044
IdioVol 553,842 0.024 0.016 0.014 0.02 0.029

Turnover 556,385 0.015 0.019 0.008 0.012 0.018

Analyst Dispersion 515,046 0.025 0.647 0.003 0.006 0.015

37 
 
 Table 2. Correlation matrix

Correlations between the various O/S and disagreement measures. We calculate correlations in each week, and then average them across all weeks. All correlations are computed
from the “raw” values of the variables. ODisagmt is the disagreement measure constructed from the open trading volumes in the ISE Open/Close data, while OCDisagmt is the
disagreement measure constructed from both opening and closing trading volumes. Detailed definitions of the variables are in Appendix A.

Correlation OMO/ OMC/ OMP/ ISEO/ ISEC/ ISEP/ OBC/ OSC/ CBC/ CSC/ OBP/ OSP/ CBP/ CSP/ ODisagmt OCDisagmt
S S S S S S S S S S S S S S

OMO/S 1.00 0.92 0.74 0.61 0.54 0.50 0.47 0.46 0.37 0.41 0.44 0.43 0.36 0.37 0.56 0.61

OMC/S 0.92 1.00 0.45 0.52 0.54 0.31 0.49 0.46 0.36 0.41 0.27 0.27 0.24 0.24 0.48 0.52

OMP/S 0.74 0.45 1.00 0.60 0.39 0.67 0.34 0.34 0.30 0.32 0.60 0.57 0.45 0.48 0.57 0.61

ISEO/S 0.61 0.52 0.60 1.00 0.87 0.81 0.78 0.77 0.54 0.58 0.71 0.72 0.51 0.52 0.91 0.99

ISEC/S 0.54 0.54 0.39 0.87 1.00 0.42 0.90 0.87 0.60 0.65 0.37 0.38 0.30 0.29 0.78 0.85

ISEP/S 0.50 0.31 0.67 0.81 0.42 1.00 0.37 0.38 0.29 0.31 0.89 0.89 0.60 0.63 0.74 0.79

OBC/S 0.47 0.49 0.34 0.78 0.90 0.37 1.00 0.76 0.33 0.52 0.33 0.34 0.24 0.24 0.76 0.77

OSC/S 0.46 0.46 0.34 0.77 0.87 0.38 0.76 1.00 0.46 0.33 0.35 0.36 0.25 0.23 0.79 0.76

CBC/S 0.37 0.36 0.30 0.54 0.60 0.29 0.33 0.46 1.00 0.50 0.23 0.24 0.27 0.26 0.39 0.55

CSC/S 0.41 0.41 0.32 0.58 0.65 0.31 0.52 0.33 0.50 1.00 0.25 0.25 0.27 0.28 0.41 0.60

OBP/S 0.44 0.27 0.60 0.71 0.37 0.89 0.33 0.35 0.23 0.25 1.00 0.76 0.32 0.49 0.73 0.70

OSP/S 0.43 0.27 0.57 0.72 0.38 0.89 0.34 0.36 0.24 0.25 0.76 1.00 0.44 0.33 0.73 0.70

CBP/S 0.36 0.24 0.45 0.51 0.30 0.60 0.24 0.25 0.27 0.27 0.32 0.44 1.00 0.48 0.36 0.50

CSP/S 0.37 0.24 0.48 0.52 0.29 0.63 0.24 0.23 0.26 0.28 0.49 0.33 0.48 1.00 0.37 0.52

ODisagmt 0.56 0.48 0.57 0.91 0.78 0.74 0.76 0.79 0.39 0.41 0.73 0.73 0.36 0.37 1.00 0.92

OCDisagmt 0.61 0.52 0.61 0.99 0.85 0.79 0.77 0.76 0.55 0.60 0.70 0.70 0.50 0.52 0.92 1.00

38 
 
 Table 3. Non-parametric analysis of the dependence between decile O/S measures
Non-parametric analysis of the dependence between some of the O/S measures. Panels A to C summarize the average deciles
of all eight signed components of ISEO/S for each decile of OMO/S, OMC/S, and OMP/S. Panels D to F summarize the
average raw values of all eight signed components of ISEO/S for each decile of OMO/S, OMC/S, and OMP/S. Deciles take the
values 1, 2, 3, ..., 10, where 1 is the smallest and 10 is the largest.

Decile: 1 2 3 4 5 6 7 8 9 10
Panel A. By DecOMO/S
DecOBC/S 2 3 4 4 5 6 6 7 8 9
DecOSC/S 2 3 4 4 5 6 6 7 8 9
DecCBC/S 3 4 4 5 5 6 6 7 8 8
DecCSC/S 3 3 4 5 5 6 6 7 8 8
DecOBP/S 3 3 4 4 5 6 6 7 8 8
DecOSP/S 3 3 4 4 5 6 6 7 8 8
DecCBP/S 3 4 4 5 5 6 6 7 7 8
DecCSP/S 3 4 4 5 5 6 6 7 7 8
Panel B. By DecOMC/S
DecOBC/S 2 3 4 4 5 6 6 7 8 9
DecOSC/S 2 3 4 4 5 6 6 7 8 9
DecCBC/S 3 3 4 5 5 6 6 7 8 8
DecCSC/S 3 3 4 4 5 6 6 7 8 8
DecOBP/S 3 4 4 5 5 6 6 7 7 8
DecOSP/S 3 4 4 5 5 6 6 7 7 8
DecCBP/S 3 4 4 5 5 6 6 7 7 8
DecCSP/S 3 4 4 5 5 6 6 7 7 8
Panel C. By DecOMP/S
DecOBC/S 3 4 4 5 5 6 6 7 7 8
DecOSC/S 3 4 4 5 5 6 6 7 7 8
DecCBC/S 3 4 4 5 5 6 6 7 7 8
DecCSC/S 3 4 4 5 5 6 6 7 7 8
DecOBP/S 2 3 3 4 5 6 6 7 8 9
DecOSP/S 2 3 4 4 5 6 6 7 8 9
DecCBP/S 3 3 4 4 5 5 6 7 8 8
DecCSP/S 3 3 4 4 5 6 6 7 8 8
Panel D. By DecOMO/S
OBC/S 0.001 0.002 0.003 0.004 0.005 0.007 0.009 0.012 0.017 0.035
OSC/S 0.001 0.002 0.002 0.003 0.005 0.006 0.008 0.011 0.015 0.031
CBC/S 0.000 0.001 0.001 0.001 0.002 0.002 0.003 0.004 0.006 0.012
CSC/S 0.000 0.001 0.001 0.002 0.002 0.003 0.004 0.005 0.008 0.016
OBP/S 0.000 0.001 0.002 0.002 0.003 0.004 0.006 0.008 0.012 0.025
OSP/S 0.001 0.001 0.002 0.002 0.003 0.004 0.006 0.008 0.011 0.024
CBP/S 0.000 0.000 0.001 0.001 0.001 0.002 0.002 0.003 0.004 0.009
CSP/S 0.000 0.000 0.001 0.001 0.001 0.002 0.002 0.003 0.005 0.010

39 
 
 Panel E. By DecOMC/S
OBC/S 0.001 0.001 0.002 0.003 0.005 0.006 0.009 0.012 0.018 0.038
OSC/S 0.001 0.001 0.002 0.003 0.004 0.006 0.008 0.010 0.015 0.032
CBC/S 0.000 0.000 0.001 0.001 0.002 0.002 0.003 0.004 0.006 0.012
CSC/S 0.000 0.001 0.001 0.001 0.002 0.003 0.004 0.005 0.008 0.017
OBP/S 0.003 0.002 0.003 0.004 0.004 0.005 0.007 0.008 0.011 0.019
OSP/S 0.003 0.002 0.003 0.003 0.004 0.005 0.006 0.008 0.010 0.018
CBP/S 0.001 0.001 0.001 0.001 0.001 0.002 0.002 0.003 0.004 0.007
CSP/S 0.001 0.001 0.001 0.001 0.002 0.002 0.003 0.003 0.004 0.008
Panel F. By DecOMP/S
OBC/S 0.005 0.004 0.005 0.006 0.006 0.008 0.010 0.012 0.015 0.025
OSC/S 0.004 0.003 0.004 0.005 0.006 0.007 0.008 0.010 0.013 0.023
CBC/S 0.001 0.001 0.001 0.002 0.002 0.003 0.003 0.004 0.005 0.009
CSC/S 0.002 0.002 0.002 0.002 0.003 0.004 0.004 0.005 0.007 0.012
OBP/S 0.000 0.001 0.001 0.002 0.002 0.004 0.005 0.008 0.012 0.029
OSP/S 0.000 0.001 0.001 0.002 0.002 0.004 0.005 0.007 0.011 0.026
CBP/S 0.000 0.000 0.000 0.001 0.001 0.001 0.002 0.003 0.004 0.010
CSP/S 0.000 0.000 0.000 0.001 0.001 0.001 0.002 0.003 0.005 0.011

40 
 
 Table 4. Fama-MacBeth regressions of weekly returns on unsigned O/S measures

Results of weekly Fama-MacBeth (1973) regressions of CAR(1) on deciles of the unsigned O/S measures OMO/S, OMC/S,
OMP/S, ISEO/S, ISEC/S, and ISEP/S. For each firm, we first sum daily option and stock volumes to obtain the weekly volumes.
The O/S measures are then computed from the weekly volumes, and the O/S deciles are formed at the end of each week. The
dependent variable CAR(1) is the cumulative abnormal return (CAR) in the week subsequent to the week in which the O/S
measures are calculated, and is in percent. All other variables are defined in Appendix A. We estimate a cross-sectional
regression for each week, and then report the time-series means of the regressions coefficients. We exclude observations from the
last quarter of 2008 to avoid the potential effect of the short-sale ban on option trading volume. Panel A presents the results of
regressions for which the O/S measures are from non-expiration weeks, where the expiration week is the week containing the
third Friday of each month. Panel B presents the results of regressions for which the O/S measures are from the expiration weeks.
t-statistics based on Newey-West (1987) adjusted standard errors, with four lags in Panel A and three lags in Panel B, are shown
in parentheses below the mean coefficient estimates. N is the number of firm-weeks. ***, **, and * indicate statistical
significance at the 1%, 5%, and 10% levels, respectively.

Panel A. O/S, C/S and P/S in non-expiration weeks

Model 1 2 3 4 5 6
DecOMO/S -0.020**
(-2.52)
DecOMC/S -0.018**
(-2.13)
DecOMP/S -0.019***
(-2.89)
DecISEO/S -0.014*
(-1.94)
DecISEC/S -0.013*
(-1.65)
DecISEP/S -0.014**
(-2.53)
Amihud -0.021*** -0.021*** -0.022*** -0.021*** -0.021*** -0.022***
(-4.66) (-4.60) (-4.77) (-4.68) (-4.60) (-4.71)
CAR(0) -0.007 -0.007 -0.008* -0.007 -0.007 -0.008
(-1.56) (-1.51) (-1.66) (-1.57) (-1.57) (-1.63)
DecΔeqvol -0.007 -0.006 -0.006 -0.005 -0.004 -0.004
(-1.26) (-1.09) (-1.00) (-0.87) (-0.74) (-0.70)
Dec∆opvol 0.002 0.000 -0.001 -0.003 -0.004 -0.004
(0.36) (0.05) (-0.18) (-0.43) (-0.67) (-0.79)
Impl_vol -0.374 -0.397 -0.392 -0.415 -0.427 -0.424
(-1.01) (-1.06) (-1.06) (-1.13) (-1.15) (-1.15)
B/M -0.004 -0.004 -0.004 -0.004 -0.004 -0.004
(-0.89) (-0.88) (-0.79) (-0.78) (-0.79) (-0.71)
Size -0.036* -0.038* -0.036* -0.041* -0.042** -0.039*
(-1.77) (-1.88) (-1.69) (-1.97) (-2.02) (-1.83)
Momen -0.001 -0.001 -0.001 -0.001 -0.001 -0.001
(-0.65) (-0.65) (-0.67) (-0.66) (-0.65) (-0.68)

41 
 
Skewness -0.006 -0.006 -0.005 -0.005 -0.006 -0.005
(-0.34) (-0.35) (-0.31) (-0.31) (-0.34) (-0.28)
Intercept 0.587** 0.607** 0.599** 0.623** 0.630** 0.615**
(2.21) (2.29) (2.23) (2.33) (2.36) (2.29)
2
Adjusted R 0.067 0.067 0.066 0.066 0.066 0.066
N 340,758 340,758 340,758 340,758 340,758 340,758

42 
 
 Panel B. O/S, C/S and P/S in expiration weeks
Model 1 2 3 4 5 6
DecOMO/S 0.010
(0.94)
DecOMC/S 0.024**
(2.22)
DecOMP/S -0.011
(-1.17)
DecISEO/S 0.009
(1.18)
DecISEC/S 0.018**
(2.20)
DecISEP/S -0.010
(-1.23)
Amihud -0.015 -0.014 -0.015 -0.015 -0.015 -0.015
(-1.51) (-1.51) (-1.58) (-1.54) (-1.51) (-1.56)
CAR(0) -0.022** -0.023*** -0.022** -0.022** -0.023** -0.022**
(-2.51) (-2.64) (-2.49) (-2.51) (-2.59) (-2.50)
DecΔeqvol 0.034** 0.038** 0.030** 0.033** 0.034** 0.030**
(2.34) (2.62) (2.07) (2.29) (2.37) (2.11)
Dec∆opvol -0.032*** -0.040*** -0.021** -0.030** -0.033*** -0.021**
(-2.73) (-3.46) (-2.03) (-2.63) (-2.96) (-2.07)
Impl_vol 0.516 0.429 0.629 0.564 0.513 0.633
(0.99) (0.83) (1.24) (1.11) (1.00) (1.27)
B/M 0.004 0.004 0.004 0.004 0.004 0.004
(0.48) (0.53) (0.44) (0.49) (0.52) (0.46)
Size -0.043 -0.051 -0.033 -0.042 -0.047 -0.033
(-1.14) (-1.34) (-0.87) (-1.14) (-1.25) (-0.88)
Momen 0.003 0.003 0.003 0.003 0.003 0.003
(0.90) (0.87) (0.92) (0.90) (0.89) (0.91)
Skewness 0.013 0.012 0.014 0.014 0.013 0.014
(0.41) (0.36) (0.45) (0.42) (0.41) (0.44)
Intercept 0.181 0.224 0.124 0.155 0.178 0.122
(0.40) (0.49) (0.27) (0.34) (0.39) (0.27)
Adjusted R2 0.059 0.059 0.058 0.058 0.059 0.058
N 108,188 108,188 108,188 108,188 108,188 108,188

43 
 
 Table 5. Fama-MacBeth regressions of weekly returns on signed O/S measures

Results of weekly Fama-MacBeth (1973) regressions of CAR(1) on deciles of the signed O/S measures OBC/S, OSC/S, CBC/S,
CSC/S, OBP/S, OSP/S, CBP/S, and CSP/S. For each firm, we first sum daily option volumes from the ISE Open/Close data to
obtain the weekly volumes for each category of option trades, and sum the daily stock volumes from CRSP to obtain the weekly
stock volumes. The signed O/S measures are then computed from the weekly volumes, and the O/S deciles are formed at the end
of each week. The dependent variable CAR(1) is the cumulative abnormal return (CAR) in the week subsequent to the week in
which the O/S measures are calculated, and is in percent. All other variables are defined in Appendix A. We estimate a cross-
sectional regression for each week, and then report the time-series means of the regressions coefficients. We exclude
observations from the last quarter of 2008 to avoid the potential effect of the short-sale ban on option trading volume. Panel A
presents the results of regressions for which the O/S measures are from non-expiration weeks, where the expiration week is
defined as the week containing the third Friday of each month. Panel B presents the results of regressions for which the O/S
measures are from the expiration weeks. t-statistics based on Newey-West (1987) adjusted standard errors, with four lags in Panel
A and three lags in Panel B, are shown in parentheses below the mean coefficient estimates. N is the number of firm-weeks. ***,
**, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

Panel A. O/S decompositions in non-expiration weeks

Model 1 2 3 4 5 6 7 8
DecOBC/S 0.021*** 0.036*** 0.037*** 0.039***
(2.97) (5.51) (5.32) (5.24)
DecOSC/S -0.033*** -0.030*** -0.029*** -0.029***
(-5.96) (-5.90) (-5.71) (-5.18)
DecCBC/S 0.000 0.004 0.005 0.006
(0.01) (0.82) (1.05) (1.08)
DecCSC/S -0.029*** -0.037*** -0.035*** -0.034***
(-3.67) (-4.90) (-4.42) (-4.38)
DecOBP/S -0.022*** -0.018*** -0.020*** -0.019***
(-3.47) (-2.96) (-2.96) (-2.80)
DecOSP/S 0.007 0.014* 0.018** 0.018***
(1.07) (1.89) (2.50) (2.68)
DecCBP/S -0.011* -0.012* -0.008 -0.007
(-1.83) (-1.90) (-1.27) (-1.14)
DecCSP/S -0.010* -0.006 -0.001 -0.001
(-1.75) (-1.10) (-0.11) (-0.12)
DecOMO/S -0.004
(-0.33)
Amihud -0.021*** -0.022*** -0.022*** -0.022*** -0.022*** -0.022*** -0.022*** -0.022***
(-4.58) (-4.71) (-4.72) (-4.74) (-4.72) (-4.76) (-4.68) (-4.56)
CAR(0) -0.007 -0.006 -0.007 -0.008* -0.006 -0.008 -0.006 -0.006
(-1.54) (-1.38) (-1.58) (-1.71) (-1.25) (-1.63) (-1.27) (-1.27)
DecΔeqvol -0.005 -0.003 -0.003 -0.002 -0.003 -0.001 -0.001 -0.002
(-0.84) (-0.43) (-0.45) (-0.28) (-0.52) (-0.25) (-0.25) (-0.34)
Dec∆opvol -0.005 -0.002 -0.004 -0.006 -0.004 -0.004 -0.003 -0.002
(-0.88) (-0.34) (-0.78) (-1.11) (-0.67) (-0.75) (-0.51) (-0.37)
Impl_vol -0.412 -0.326 -0.454 -0.376 -0.323 -0.398 -0.324 -0.316

44 
 
 (-1.11) (-0.89) (-1.22) (-1.03) (-0.88) (-1.08) (-0.88) (-0.86)
B/M -0.004 -0.004 -0.004 -0.004 -0.004 -0.004 -0.004 -0.004
(-0.72) (-0.77) (-0.72) (-0.72) (-0.71) (-0.73) (-0.71) (-0.78)
Size -0.039* -0.030 -0.039* -0.031 -0.029 -0.032 -0.024 -0.024
(-1.91) (-1.44) (-1.83) (-1.50) (-1.39) (-1.52) (-1.18) (-1.16)
Momen -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001
(-0.66) (-0.56) (-0.64) (-0.71) (-0.52) (-0.67) (-0.50) (-0.48)
Skewness -0.007 -0.003 -0.005 -0.004 -0.004 -0.005 -0.004 -0.005
(-0.39) (-0.20) (-0.32) (-0.24) (-0.24) (-0.29) (-0.21) (-0.27)
Intercept 0.616** 0.559** 0.620** 0.566** 0.553** 0.573** 0.527** 0.519*
(2.31) (2.09) (2.30) (2.12) (2.06) (2.15) (1.97) (1.96)
Adjusted R2 0.066 0.067 0.066 0.066 0.067 0.067 0.069 0.070
N 340,758 340,758 340,758 340,758 340,758 340,758 340,758 340,758

45 
 
 Panel B. O/S decompositions in expiration weeks
Model 1 2 3 4 5 6 7 8
DecOBC/S 0.053*** 0.057*** 0.061*** 0.058***
(4.42) (4.37) (4.55) (4.34)
DecOSC/S -0.030*** -0.021** -0.020* -0.023**
(-2.74) (-2.01) (-1.88) (-2.08)
DecCBC/S -0.019** -0.020** -0.014 -0.016*
(-2.14) (-2.33) (-1.55) (-1.83)
DecCSC/S 0.019* -0.003 0.001 -0.002
(1.86) (-0.27) (0.07) (-0.17)
DecOBP/S -0.052*** -0.047*** -0.052*** -0.054***
(-4.00) (-3.50) (-3.88) (-3.87)
DecOSP/S 0.042*** 0.050*** 0.046*** 0.043***
(4.22) (5.53) (5.20) (4.66)
DecCBP/S -0.001 -0.014* -0.017** -0.021***
(-0.18) (-1.76) (-2.36) (-2.87)
DecCSP/S -0.018* -0.008 -0.010 -0.012
(-1.72) (-0.73) (-0.90) (-1.11)
DecOMO/S 0.024
(1.36)
Amihud -0.015 -0.014 -0.016* -0.016 -0.015 -0.017* -0.018* -0.019*
(-1.53) (-1.50) (-1.69) (-1.62) (-1.54) (-1.76) (-1.80) (-1.81)
CAR(0) -0.023** -0.023** -0.021** -0.023** -0.022** -0.021** -0.022** -0.022**
(-2.61) (-2.61) (-2.44) (-2.57) (-2.58) (-2.43) (-2.56) (-2.59)
DecΔeqvol 0.031** 0.031** 0.032** 0.032** 0.031** 0.033** 0.033** 0.038**
(2.21) (2.27) (2.30) (2.32) (2.27) (2.42) (2.42) (2.62)
Dec∆opvol -0.035*** -0.026** -0.020** -0.020** -0.034*** -0.019** -0.029*** -0.037***
(-3.22) (-2.36) (-2.08) (-2.07) (-3.03) (-2.00) (-2.76) (-3.23)
Impl_vol 0.488 0.598 0.555 0.664 0.538 0.596 0.541 0.532
(0.95) (1.16) (1.10) (1.33) (1.03) (1.18) (1.05) (1.02)
B/M 0.004 0.004 0.003 0.004 0.004 0.004 0.005 0.005
(0.52) (0.43) (0.40) (0.52) (0.52) (0.50) (0.58) (0.64)
Size -0.045 -0.033 -0.036 -0.025 -0.037 -0.029 -0.028 -0.026
(-1.19) (-0.89) (-0.96) (-0.68) (-0.97) (-0.77) (-0.73) (-0.67)
Momen 0.003 0.003 0.003 0.003 0.003 0.003 0.003 0.003
(0.92) (0.88) (0.99) (0.93) (0.94) (1.00) (1.01) (1.01)
Skewness 0.014 0.013 0.016 0.014 0.013 0.016 0.014 0.015
(0.42) (0.39) (0.51) (0.42) (0.41) (0.49) (0.44) (0.47)
Intercept 0.172 0.106 0.158 0.081 0.130 0.127 0.121 0.122
(0.38) (0.23) (0.35) (0.18) (0.29) (0.28) (0.26) (0.26)
2
Adjusted R 0.059 0.059 0.059 0.059 0.060 0.060 0.062 0.062
N 108,188 108,188 108,188 108,188 108,188 108,188 108,188 108,188

46 
 
 Table 6. Option moneyness and return predictability

Results of weekly Fama-MacBeth (1973) regressions of CAR(1) on deciles of the signed O/S measures OBC/S, OSC/S, CBC/S,
CSC/S, OBP/S, OSP/S, CBP/S, and CSP/S for options in different moneyness categories. We use option delta to classify options
into different moneyness categories, as in Bollen and Whaley (2004). A call option is defined to be out-of-the-money (OTM) if
0.02 < delta ≤ 0.375; at-the-money (ATM) if 0.375 < delta ≤ 0.625; and in-the-money (ITM) if 0.625 < delta ≤ 0.98. A put option
is defined to be out-of-the-money (OTM) if -0.375<delta ≤ 0.02; at-the-money (ATM) if -0.625<delta ≤ 0.375; and in-the-
money (ITM) if 0.98 < delta ≤ 0.625. Panel A shows the average relative bid-ask spread (in percent) for options in each of the
moneyness categories. In Panels B and C, for each firm-week, we calculate signed and unsigned O/S measures, respectively,
using the options in each moneyness category. As in Table 5, we estimate weekly Fama-MacBeth (1973) regressions and report
the time-series means of the coefficient estimates, where in each week we require that there are at least 30 firms in each
moneyness category. Panel B presents regression results for the signed O/S measures, and Panel C presents results for the
unsigned O/S measures. We exclude observations for which the O/S measures are from an expiration week and also exclude
observations from the last quarter of 2008. N is the number of firm-weeks. t-statistics based on Newey-West (1987) adjusted
standard errors with four lags are shown in parentheses below the mean coefficient estimates. ***, **, and * indicate statistical
significance at the 1%, 5%, and 10% levels, respectively.

Panel A. Relative bid-ask spread (%)

Moneyness Call Put All
OTM 38.81 29.48 35.98
ATM 12.80 9.98 12.44
ITM 7.74 6.81 7.93

47 
 
 Panel B. Predictability of signed O/S measures by moneyness
OTM ATM ITM
Model 1 2 3
DecOBC/S 0.020** 0.012 0.007
(2.56) (1.25) (0.76)
DecOSC/S -0.011 -0.013 -0.025***
(-1.60) (-1.40) (-3.24)
DecCBC/S -0.001 0.002 0.003
(-0.16) (0.25) (0.32)
DecCSC/S -0.040*** -0.015 -0.005
(-5.08) (-1.64) (-0.58)
DecOBP/S -0.017** -0.018* -0.006
(-2.56) (-1.74) (-0.62)
DecOSP/S 0.023*** 0.021*** 0.016**
(3.66) (2.71) (2.23)
DecCBP/S -0.023*** 0.002 -0.007
(-2.68) (0.28) (-0.83)
DecCSP/S 0.005 -0.004 -0.004
(0.56) (-0.51) (-0.36)
DecOMO_S 0.011 -0.022 -0.009
(0.82) (-1.44) (-0.65)
Amihud -0.009 -0.043 -0.079
(-0.15) (-1.37) (-1.03)
CAR(0) -0.001 0.002 -0.002
(-0.19) (0.22) (-0.37)
Dec∆eqvol -0.007 -0.009 -0.017
(-0.68) (-0.76) (-1.30)
Dec∆opvol -0.009 0.005 -0.007
(-0.88) (0.42) (-0.76)
Impl_vol 0.121 -0.126 -0.552
(0.26) (-0.30) (-1.18)
B/M -0.008 -0.005 -0.006
(-1.33) (-0.48) (-0.87)
Size -0.033 -0.058** -0.077***
(-1.11) (-2.00) (-2.84)
Momen -0.001 -0.001 -0.002
(-0.66) (-0.27) (-0.96)
Skewness -0.011 -0.031 0.039
(-0.45) (-1.11) (1.18)
Intercept 0.501 0.840* 1.267***
(1.06) (1.95) (2.88)
Adjusted R2 0.089 0.091 0.100
N 172,039 158,024 125,599

48 
 
 Panel C. Predictability of unsigned O/S measures by moneyness
OTM ATM ITM OTM ATM ITM
Model 1 2 3 4 5 6
DecOMO/S -0.017*** -0.016*** -0.020***
(-2.81) (-2.73) (-3.67)
DecISEO/S -0.013** -0.015* -0.012
(-2.13) (-1.93) (-1.56)
Amihud -0.020*** -0.018*** -0.018*** 0.055 -0.033 -0.054
(-5.15) (-4.38) (-4.69) (0.45) (-1.12) (-0.70)
CAR(0) -0.007 -0.006 -0.006 -0.001 0.001 -0.002
(-1.36) (-1.31) (-1.20) (-0.16) (0.12) (-0.23)
Dec∆eqvol -0.006 -0.005 -0.005 -0.013 -0.008 -0.018
(-0.90) (-0.81) (-0.70) (-1.15) (-0.76) (-1.50)
Dec∆opvol 0.000 0.000 -0.001 -0.005 -0.005 -0.010
(0.05) (0.05) (-0.23) (-0.59) (-0.44) (-1.04)
Impl_vol -0.464 -0.456 -0.434 -0.048 -0.214 -0.650
(-1.34) (-1.32) (-1.26) (-0.10) (-0.45) (-1.40)
B/M -0.005 -0.004 -0.006 -0.008 -0.005 -0.006
(-1.23) (-0.92) (-1.31) (-1.32) (-0.50) (-0.94)
Size -0.035* -0.034* -0.032* -0.054** -0.070*** -0.087***
(-1.92) (-1.82) (-1.71) (-2.04) (-2.63) (-3.45)
Momen -0.001 -0.001 -0.001 -0.001 -0.001 -0.002
(-0.49) (-0.41) (-0.56) (-0.54) (-0.29) (-1.06)
Skewness -0.007 -0.006 -0.002 -0.009 -0.034 0.031
(-0.44) (-0.40) (-0.16) (-0.38) (-1.26) (1.03)
Intercept 0.590** 0.567** 0.571** 0.654* 0.891** 1.295***
(2.17) (2.07) (2.10) (1.82) (2.21) (3.11)
Adjusted R2 0.063 0.063 0.063 0.084 0.085 0.095
N 374,589 371,027 374,582 172,039 158,024 125,599

49 
 
 Table 7. Fama-MacBeth regressions including disagreement measures

Results from Fama-MacBeth (1973) regressions of CAR(1) on both the signed O/S measures and disagreement measures to
assess whether the ability of the signed O/S measures to predict abnormal returns is affected by including disagreement measures
in the regressions. The disagreement measure ODisagmt is constructed from opening ISE volumes, while the measure
OCDisagmt is constructed from both opening and closing ISE volumes. IdioVol is the idiosyncratic volatility calculated as the
standard deviation of the daily abnormal return based on the one-factor market model over the six months prior to week 1, the
week over which the left-hand side variable CAR(1) is measured. Turnover is calculated as the daily trading volume (scaled by
number of shares outstanding) averaged over the six months prior to week 1. Analyst dispersion is calculated using forecasts
from the IBES detail file with a forecast period of one year. For each firm-week, we measure analyst dispersion as the standard
deviation of all analyst earnings forecasts during the previous six months, scaled by the stock price at the end of week 0. All other
variables are defined in Appendix A. As in Table 5, we estimate weekly Fama-MacBeth (1973) regressions and report the time-
series means of the coefficient estimates. Observations for which the O/S measures are from expiration weeks and observations
from the last quarter of 2008 are excluded. N is the number of firm-weeks. t-statistics based on Newey-West (1987) adjusted
standard errors with four lags are shown in parentheses below the mean coefficient estimates. ***, **, and * indicate statistical
significance at the 1%, 5%, and 10% levels, respectively.

Model 1 2 3 4 5 6 7 8
DecOBC/S 0.038*** 0.038*** 0.038*** 0.038*** 0.037*** 0.037*** 0.039*** 0.039***
(4.73) (4.93) (4.52) (4.87) (4.58) (4.74) (4.84) (5.16)
DecOSC/S -0.028*** -0.029*** -0.028*** -0.028*** -0.028*** -0.028*** -0.028*** -0.028***
(-3.74) (-4.32) (-3.80) (-4.26) (-3.77) (-4.28) (-3.86) (-4.35)
DecCBC/S 0.006 0.006 0.008 0.008 0.006 0.006 0.008 0.008
(1.16) (1.08) (1.54) (1.45) (1.20) (1.09) (1.52) (1.41)
DecCSC/S -0.035*** -0.035*** -0.034*** -0.034*** -0.036*** -0.036*** -0.033*** -0.033***
(-4.39) (-4.25) (-4.35) (-4.18) (-4.55) (-4.41) (-4.31) (-4.17)
DecOBP/S -0.018** -0.019** -0.019** -0.018** -0.019** -0.019** -0.017* -0.017**
(-2.11) (-2.43) (-2.13) (-2.36) (-2.09) (-2.43) (-1.96) (-2.21)
DecOSP/S 0.021*** 0.020*** 0.020** 0.021*** 0.022*** 0.021*** 0.021*** 0.021***
(2.68) (2.63) (2.58) (2.70) (2.73) (2.69) (2.67) (2.68)
DecCBP/S -0.007 -0.007 -0.005 -0.005 -0.007 -0.007 -0.005 -0.005
(-1.12) (-0.95) (-0.85) (-0.69) (-1.19) (-1.02) (-0.81) (-0.68)
DecCSP/S 0.001 0.001 0.002 0.002 0.000 -0.000 0.003 0.003
(0.18) (0.16) (0.34) (0.35) (0.01) (-0.01) (0.41) (0.41)
DecOMO_S -0.003 -0.002 -0.005 -0.005 -0.004 -0.003 -0.004 -0.004
(-0.25) (-0.20) (-0.42) (-0.38) (-0.30) (-0.26) (-0.37) (-0.33)
DecODisagm -0.002 0.001 -0.002 -0.000
t
(-0.17) (0.07) (-0.14) (-0.01)
DecOCDisag -0.000 -0.001 -0.000 -0.000
mt
(-0.02) (-0.09) (-0.00) (-0.02)
IdioVol -6.435* -6.524* -4.170 -4.255
(-1.86) (-1.89) (-1.18) (-1.21)
Turnover -6.000*** -5.999*** -5.892*** -5.875***
(-3.27) (-3.27) (-3.18) (-3.18)

50 
 
Analyst Dispersion 1.343 1.348 1.433 1.439
(1.15) (1.16) (1.23) (1.24)
Amihud -0.022*** -0.023*** -0.026*** -0.026*** -0.023*** -0.023*** -0.027*** -0.027***
(-3.45) (-3.49) (-3.66) (-3.70) (-3.57) (-3.63) (-3.90) (-3.96)
CAR(0) -0.004 -0.004 -0.004 -0.004 -0.003 -0.003 -0.003 -0.003
(-0.80) (-0.80) (-0.77) (-0.76) (-0.64) (-0.64) (-0.64) (-0.64)
Dec∆eqvol -0.001 -0.001 -0.002 -0.002 -0.003 -0.003 -0.003 -0.003
(-0.15) (-0.17) (-0.34) (-0.34) (-0.43) (-0.45) (-0.41) (-0.42)
Dec∆opvol -0.004 -0.004 -0.004 -0.005 -0.003 -0.003 -0.006 -0.006
(-0.58) (-0.64) (-0.60) (-0.67) (-0.41) (-0.47) (-0.91) (-0.97)
Impl_vol -0.052 -0.049 -0.108 -0.109 -0.443 -0.443 -0.068 -0.064
(-0.15) (-0.14) (-0.29) (-0.30) (-1.23) (-1.23) (-0.20) (-0.19)
B/M -0.004 -0.004 -0.003 -0.003 -0.004 -0.004 -0.004 -0.004
(-0.80) (-0.78) (-0.60) (-0.58) (-0.80) (-0.78) (-0.76) (-0.74)
Size -0.029 -0.030 -0.027 -0.028 -0.025 -0.025 -0.036 -0.037*
(-1.33) (-1.37) (-1.30) (-1.33) (-1.16) (-1.20) (-1.63) (-1.65)
Momen -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001
(-0.40) (-0.40) (-0.38) (-0.39) (-0.34) (-0.35) (-0.37) (-0.38)
Skewness -0.004 -0.004 -0.005 -0.006 -0.007 -0.008 -0.007 -0.007
(-0.24) (-0.26) (-0.30) (-0.33) (-0.43) (-0.45) (-0.41) (-0.42)
Intercept 0.512* 0.525* 0.480* 0.484* 0.528* 0.539* 0.587** 0.594**
(1.80) (1.83) (1.72) (1.73) (1.93) (1.95) (2.06) (2.07)
Adjusted R2 0.075 0.075 0.076 0.075 0.078 0.078 0.084 0.084
N 322,210 322,210 322,210 322,210 322,210 322,210 322,210 322,210

51 
 
 Table 8. Different investor types and trade size buckets

Results from Fama-MacBeth (1973) regressions of CAR(1) on decile signed O/S measures for different types of investors and
trade size buckets. The different categories are firm proprietary traders, customer small trades (trades of less than 100 contracts
each), customer medium trades (trades of between 101 and 199 contracts each), and customer large trades (trades of 200 or more
contracts each). We calculate firm-week signed O/S measures using the trading volumes from the various categories as the main
independent variables. As in Table 5, we estimate weekly Fama-MacBeth (1973) regressions and report the time-series means of
the coefficient estimates. Observations for which the O/S measures are from expiration weeks and observations from the last
quarter of 2008 are excluded. In Panel A, we estimate a separate regression for each category. In Panel B, we estimate one
regression including the signed O/S measures from all categories. The regression reported in Panel B includes all of the control
variables used in the regressions reported in Panel A, but Panel B does not report the coefficient estimates on the control
variables to save space. N is the number of firm-weeks. t-statistics based on Newey-West (1987) adjusted standard errors with
four lags are shown in parentheses below the mean coefficient estimates. ***, **, and * indicate statistical significance at the 1%,
5%, and 10% levels, respectively.

Panel A. Trades from each type of investor or trade size bucket

Firm Proprietary Customer_Small Customer_Medium Customer_Large
Model 1 2 3 4
DecOBC/S -0.004 0.031*** 0.017** 0.027***
(-0.76) (3.95) (2.01) (3.24)
DecOSC/S 0.007 -0.028*** -0.011 -0.015*
(1.14) (-4.44) (-1.56) (-1.92)
DecCBC/S -0.002 0.006 0.005 0.021**
(-0.39) (1.08) (0.56) (2.20)
DecCSC/S -0.009* -0.036*** -0.016* -0.028***
(-1.69) (-4.73) (-1.89) (-2.79)
DecOBP/S 0.003 -0.022*** -0.015* -0.021**
(0.79) (-3.26) (-1.84) (-2.05)
DecOSP/S -0.004 0.017*** -0.006 0.002
(-0.69) (2.79) (-0.87) (0.31)
DecCBP/S -0.001 -0.006 -0.004 0.011
(-0.16) (-0.91) (-0.39) (0.78)
DecCSP/S -0.006 0.001 0.017* 0.008
(-1.20) (0.16) (1.70) (0.82)
DecOMO/S -0.014 0.001 -0.018* -0.020**
(-1.38) (0.06) (-1.92) (-2.24)
Amihud -0.022*** -0.022*** -0.021*** -0.021***
(-4.64) (-4.53) (-4.60) (-4.64)
CAR(0) -0.007 -0.005 -0.007 -0.006
(-1.44) (-0.99) (-1.56) (-1.38)
Dec∆eqvol -0.004 -0.002 -0.007 -0.008
(-0.72) (-0.32) (-1.13) (-1.36)
Dec∆opvol 0.002 -0.002 0.002 0.002
(0.25) (-0.36) (0.34) (0.29)
Impl_vol -0.357 -0.319 -0.378 -0.379
(-0.98) (-0.87) (-1.03) (-1.03)
B/M -0.004 -0.004 -0.004 -0.005
(-0.85) (-0.71) (-0.90) (-0.94)
Size -0.030 -0.025 -0.032 -0.036
(-1.38) (-1.21) (-1.42) (-1.57)
Momen -0.001 -0.001 -0.001 -0.001
(-0.56) (-0.45) (-0.61) (-0.59)
Skewness -0.004 -0.003 -0.006 -0.008
(-0.25) (-0.20) (-0.38) (-0.45)
Intercept 0.566** 0.547** 0.610** 0.570**

52 
 
 (2.15) (2.09) (2.28) (2.14)
Adjusted R2 0.069 0.071 0.069 0.069
N 340,758 340,758 340,758 340,758

Panel B. Trades from all types of investors and trade size buckets
Firm Proprietary Customer_Small Customer_Medium Customer_Large
DecOBC/S 0.007 0.029*** 0.005 0.023***
(1.30) (3.69) (0.62) (2.60)
DecOSC/S -0.003 -0.028*** -0.001 -0.013*
(-0.45) (-4.20) (-0.08) (-1.65)
DecCBC/S 0.002 0.005 -0.007 0.027***
(0.46) (0.94) (-0.73) (2.78)
DecCSC/S -0.009* -0.034*** -0.006 -0.022**
(-1.86) (-4.48) (-0.66) (-2.19)
DecOBP/S 0.000 -0.021*** -0.005 -0.013
(0.09) (-3.01) (-0.59) (-1.37)
DecOSP/S 0.002 0.018*** -0.014* 0.003
(0.38) (2.90) (-1.76) (0.39)
DecCBP/S -0.000 -0.007 -0.007 0.020
(-0.07) (-1.00) (-0.81) (1.39)
DecCSP/S -0.002 -0.000 0.016 0.003
(-0.46) (-0.07) (1.46) (0.29)
Controls Yes
Intercept 0.510*
(1.95)
Adjusted R2 0.076
N 340,758
 

53 
 
 Table 9. Alternative signed O/S measures

Results of weekly Fama-MacBeth (1973) regressions of CAR(1) on alternative decile signed O/S measures. Using OBC/S as an
example, following Johnson and So (2012) we define ∆OBC/S as the difference between OBC/S and its average over the prior
six months, scaled by the average. The measure std∆OBC/S is standardized OBC/S, defined as the difference between OBC/S
and its average over the prior six months, scaled by the standard deviation over the prior six months. The measure ΩOBC/S is the
rank of OBC/S relative to the firm’s OBC/S time series over the prior six months, and thus takes into account the time-series
variation in OBC/S within each firm. Each week we sort firms based on weekly ∆OBC/S (Model 1) or std∆OBC/S (Model 2) into
deciles, and use the decile variables in the regressions. In Model 3, we sort ΩOBC/S into deciles for each firm-week, where a
firm-week is ranked into decile 10 if its OBC/S is above the 90th percentile of its own time-series distribution, for example. In
this way, we construct three types of signed O/S measures using ISE data, and three types of unsigned OMO/S measures using
OptionMetrics data. The other variables are defined in Appendix A. As in Table 5, we estimate weekly Fama-MacBeth (1973)
regressions and report the time-series means of the coefficient estimates. We exclude observations for which the O/S measures
are from expiration weeks, and also exclude observations from the last quarter of 2008. N is the number of firm-weeks. t-statistics
based on Newey-West (1987) adjusted standard errors, with four lags, are shown in parentheses below the mean coefficient
estimates. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

Model 1 2 3
Dec∆OBC/S 0.026*** Dec std∆OBC/S 0.025*** Dec ΩOBC/S 0.026***
(4.61) (4.35) (4.42)
Dec∆OSC/S -0.017*** Dec std∆OSC/S -0.015*** Dec ΩOSC/S -0.017***
(-3.65) (-3.16) (-3.48)
Dec∆CBC/S -0.001 Dec std∆CBC/S 0.002 Dec ΩCBC/S 0.000

(-0.23) (0.62) (0.08)

Dec∆CSC/S -0.021*** Dec std∆CSC/S -0.016*** Dec ΩCSC/S -0.021***
(-3.23) (-2.61) (-3.54)
Dec∆OBP/S -0.010** Dec std∆OBP/S -0.010** Dec ΩOBP/S -0.016***
(-2.12) (-2.56) (-3.31)
Dec∆OSP/S 0.004 Dec std∆OSP/S 0.002 Dec ΩOSP/S 0.004
(0.73) (0.44) (0.75)
Dec∆CBP/S -0.007 Dec std∆CBP/S 0.002 Dec ΩCBP/S -0.004
(-1.50) (0.42) (-0.78)
Dec∆CSP/S -0.005 Dec std∆CSP/S -0.001 Dec ΩCSP/S -0.003

(-0.97) (-0.36) (-0.78)

Dec∆OMO/S 0.002 Dec std∆OMO/S 0.025** Dec ΩOMO/S 0.019
(0.18) (2.55) (1.63)
Amihud -0.015* Amihud -0.014* Amihud -0.016*
(-1.74) (-1.72) (-1.88)
CAR(0) -0.009* CAR(0) -0.009* CAR(0) -0.010*
(-1.72) (-1.75) (-1.80)
Dec∆eqvol -0.002 Dec∆eqvol 0.009 Dec∆eqvol 0.008
(-0.22) (1.15) (0.81)
Dec∆opvol 0.002 Dec∆opvol -0.026** Dec∆opvol -0.017
(0.15) (-2.55) (-1.48)
Impl_vol -0.465 Impl_vol -0.515 Impl_vol -0.461

54 
 
 (-1.18) (-1.30) (-1.17)
B/M -0.005 B/M -0.006 B/M -0.006
(-0.96) (-1.05) (-1.05)
Size -0.031 Size -0.045** Size -0.032
(-1.42) (-2.01) (-1.43)
Momen -0.002 Momen -0.002 Momen -0.002
(-0.82) (-0.89) (-0.85)
Skewness -0.007 Skewness -0.009 Skewness -0.008
(-0.40) (-0.50) (-0.46)
Intercept 0.571* Intercept 0.571** Intercept 0.504*
(1.97) (1.98) (1.72)
Adjusted R2 0.072 Adjusted R2 0.072 Adjusted R2 0.072
N 296,814 N 296,814 N 296,814

55 
 
 Table 10. Fama-MacBeth regressions using raw values of signed O/S measures

Results of weekly Fama-MacBeth (1973) regressions of CAR(1) on raw values of signed O/S measures. As in Table 5, we
estimate weekly Fama-MacBeth (1973) regressions and report the time-series means of the coefficient estimates. We exclude
observations for which the O/S measures are from expiration weeks, and also exclude observations from the last quarter of 2008.
N is the number of firm-weeks. t-statistics based on Newey-West (1987) adjusted standard errors, with four lags, are shown in
parentheses below the mean coefficient estimates. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels,
respectively.

Model 1 2 3 4 5 6 7 8
OBC/S 4.143*** 7.618*** 7.412*** 7.750***
(3.68) (5.76) (5.61) (5.88)
OSC/S -4.844*** -8.494*** -7.986*** -7.671***
(-3.71) (-5.86) (-5.66) (-5.51)
CBC/S 3.591 9.479*** 9.194*** 10.365***
(1.37) (3.42) (3.42) (3.94)
CSC/S -5.620*** -10.013*** -9.880*** -8.981***
(-2.81) (-4.29) (-4.11) (-3.76)
OBP/S -4.025** -5.144*** -4.793*** -4.213**
(-2.53) (-2.84) (-2.64) (-2.26)
OSP/S 2.106 3.219* 3.305* 3.736**
(1.34) (1.71) (1.80) (2.08)
CBP/S -2.114 -2.871 -1.568 -0.792
(-0.81) (-1.01) (-0.57) (-0.27)
CSP/S -1.452 2.188 2.533 2.752
(-0.55) (0.77) (0.85) (0.89)
OMO/S -0.230**
(-2.16)
Amihud -0.022*** -0.022*** -0.022*** -0.022*** -0.022*** -0.022*** -0.022*** -0.022***
(-4.70) (-4.71) (-4.74) (-4.76) (-4.78) (-4.75) (-4.75) (-4.69)
CAR(0) -0.008* -0.007 -0.007 -0.008 -0.007 -0.007 -0.006 -0.006
(-1.65) (-1.50) (-1.56) (-1.63) (-1.39) (-1.58) (-1.37) (-1.34)
Dec∆eqvol -0.004 -0.003 -0.003 -0.003 -0.004 -0.003 -0.004 -0.005
(-0.59) (-0.59) (-0.53) (-0.52) (-0.61) (-0.57) (-0.60) (-0.78)
Dec∆opvol -0.010** -0.008 -0.006 -0.009* -0.009* -0.007 -0.007 -0.005
(-1.98) (-1.60) (-1.33) (-1.84) (-1.77) (-1.36) (-1.45) (-0.96)
Impl_vol -0.433 -0.456 -0.449 -0.454 -0.416 -0.431 -0.405 -0.369
(-1.15) (-1.22) (-1.21) (-1.22) (-1.10) (-1.15) (-1.07) (-0.97)
B/M -0.004 -0.004 -0.004 -0.004 -0.004 -0.004 -0.004 -0.004
(-0.75) (-0.75) (-0.75) (-0.72) (-0.73) (-0.75) (-0.75) (-0.82)
Size -0.045** -0.046** -0.045** -0.045** -0.044** -0.044** -0.043** -0.039*
(-2.16) (-2.18) (-2.11) (-2.13) (-2.07) (-2.09) (-2.04) (-1.88)
Momen -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001
(-0.68) (-0.61) (-0.68) (-0.69) (-0.59) (-0.70) (-0.62) (-0.62)

56 
 
Skewness -0.006 -0.006 -0.005 -0.005 -0.006 -0.005 -0.006 -0.006
(-0.33) (-0.35) (-0.28) (-0.32) (-0.34) (-0.33) (-0.33) (-0.37)
Intercept 0.631** 0.641** 0.623** 0.630** 0.615** 0.611** 0.598** 0.557**
(2.34) (2.36) (2.31) (2.34) (2.25) (2.27) (2.20) (2.05)
2
Adjusted R 0.067 0.067 0.067 0.066 0.069 0.068 0.072 0.072
N 340,758 340,758 340,758 340,758 340,758 340,758 340,758 340,758

57