Peer stock short interest and future returns
`
`Ferhat Akbas, Ekkehart Boehmer, Egemen Genc*
`
`
`
`ABSTRACT
`
`Firm-level monthly short interest is positively and significantly related to the returns of firms that
`compete in the same product markets. This finding is robust to standard controls and cannot be
`explained by industry momentum, industry lead-lag relationships, or industry information
`spillover effects. Short interest also contains information about the fundamentals of competing
`firms. Trading cost reductions are an important driver for trading a firm’s competitors rather than
`the firm’s own stocks. Our findings suggest that short sellers’ trades play an important role in the
`price discovery of competing firms, beyond their direct effects documented previously.
`
`February, 2015
`
`
`
`
`
`*Akbas is from the University of Kansas (akbas@ku.edu), Boehmer is from Singapore Management University
`(eboehmer@smu.edu.sg). Genc is from Rotterdam School of Management, Erasmus University (egenc@rsm.nl). We
`thank seminar participants at the University of Kansas for their insightful comments.
`
`
`
`
`
`CFAD VI 1070 - 0001
`CFAD VI v. CELGENE
`IPR2015-01102
`
`

`
`1. Introduction
`
`Informed traders influence contemporaneous and future security prices. Depending on
`
`both the aggressiveness of the informed traders and the amount traded, prices will converge more
`
`or less quickly to the new equilibrium price. These dynamics are quite well understood (Kyle,
`
`1985). However, firms are not independent entities and information signals other than the ones
`
`from its own trading environment potentially affect their stock prices and improve the price
`
`discovery. While recent studies point out that order flows containing industry or market wide
`
`information can contemporaneously affect the return of more than one security (Tookes, 2008;
`
`Pasquariello and Vega, 2013), relatively little is known about whether and how informed trading
`
`in stocks of competing firms affects a firm’s own stock price. To narrow this gap, we study
`
`information transmission between economically linked stocks through informed trading on firm
`
`specific information.
`
`Our experiment involves studying the effect of trades of short sellers, who are generally
`
`considered to be well-informed traders, on the future share prices of product-market competitors.
`
`Controlling for firm and industry characteristics, we show that short interest in a particular stock,
`
`despite its negative effect on its own future price, is significantly and positively related to future
`
`returns and earnings surprises of the closest competitor. We define the closest competitor as a
`
`firm’s closest neighbor in terms of product market share. Accounting for about two thirds of the
`
`average absolute effect of short interest, cross-price impact of a firm’s short position is
`
`economically large. Using only stocks in the top short-interest quintile, a trading strategy that
`
`goes short in the competitors of the least-shorted quintile of stocks and long in the competitors of
`
`the most-shorted quintile earns a significant 71 basis points per month, or 8.52% per year.
`
`
`
`1
`
`CFAD VI 1070 - 0002
`
`

`
`We conjecture that cross-price impacts arise because stock market frictions, especially
`
`short selling constraints, prevent short sellers from fully trading on their information. In the
`
`presence of short selling constraints, short sellers find it costly to trade on all their information
`
`using stocks of only one firm, so that part of their information is revealed through their trades in
`
`competing firms. To test this conjecture, we examine the role of short selling constraints and find
`
`that more binding constraints appear to let traders prefer a long position in the competing firm.
`
`Specifically, standard measures of shorting costs—such as the absence of listed put options,
`
`small firm size, and high idiosyncratic volatility—are all associated with greater cross-price
`
`impacts. Thus the cross-price impact increases when a firm has higher shorting costs.
`
`The positive cross-firm impact suggests that the information content is competitive and
`
`firm specific, rather than industry-wide. These results are robust to controlling for industry-wide
`
`short interest, further supporting this view. Moreover, industry short interest has no effect on
`
`future returns. Indeed, the economic link as a close competitor is crucial in finding the positive
`
`cross-firm effect. We demonstrate the importance of this economic link by running simulations
`
`in which we randomly select firms from the same or from other industries, rather than identifying
`
`close competitors. In contrast to Pasquariello and Vega (2013), who look at trading in general
`
`rather than at short selling, we find no cross-price impact for economically unrelated firms.
`
`Instead, the strength of the economic link between two firms drives our results. Moreover, the
`
`results are not short-term price impacts that are later reversed. We show that the cross-price
`
`impact survives up to one year, and the competing firm’s short interest predicts future earnings
`
`surprises. Both results lend support to the premise that the cross-firm effect is driven by informed
`
`trading stemming from firm specific information.
`
`
`
`2
`
`CFAD VI 1070 - 0003
`
`

`
`We focus on the price impact across product-market competitors for two reasons. First, as
`
`formalized by Tookes’ (2008), product market competition provides incentives for trading in
`
`competing firms.1 In a competitive environment, firm-specific news that improves the value of
`
`one firm can negatively affect the value of other firms, and vice versa. For example, the success
`
`of a car manufacturer with a particular model is likely to reduce sales for competing models, or a
`
`pharmaceutical company’s breakthrough drug will reduce the sales of competing drugs. In this
`
`setting, when short sellers obtain firm-level information for a company, they can infer the impact
`
`of that same information for the firm’s competitor, and can strategically chose to split their trades
`
`between the firm and the competitor to minimize the overall costs of short selling.
`
`To examine this idea empirically, we focus on short positions around analyst
`
`recommendation downgrades, known to be important events with negative price reactions
`
`(Womack, 1996; Barber, Lehavy, McHichols, and Trueman, 2001). We isolate cases in which an
`
`analyst downgrade occurs in a firm, while its competitor does not experience such an event. We
`
`find that prior to analyst downgrades, short interest increases in the event firm.
`
`Contemporaneously, short sellers decrease their short positions in the firm’s competitor before
`
`the downgrade even though the competitor does not have any such event over a twelve-month
`
`window around the downgrade announcement. This finding suggests that short sellers can extract
`
`useful information that was originally about one firm to take a position not only in the affected
`
`firm but also in its competitors.
`
`
`1 In her model, informed traders decide to trade privately obtained information about a firm’s future production costs
`either in that firm’s stock or in the stock of competitors in an oligopolistic market. Empirically, using a small sample
`of earnings announcements, she shows that non-announcing firms’ order flows, measured over short term intervals
`such as five minutes, contain information about the announcing firm.
`
`
`
`
`3
`
`CFAD VI 1070 - 0004
`
`

`
`Second, short sellers can hedge their positions by taking an opposite position in a close
`
`competitor when trading on their information in a stock. If short sellers cannot fully exploit their
`
`information in a stock due to short selling constraints, then their hedge positions in a close
`
`competitor would signal part of their information. It should be noted that, in this second
`
`mechanism, short sellers’ trade in a particular stock is still motivated by information and they
`
`trade in the close competitor for hedging purposes. Nonetheless, in either strategic trading or
`
`hedging mechanisms, short sellers are more likely to trade in close product-market competitors.
`
`This provides us a natural experimental that allows us to test our view that short sellers’
`
`information is partially revealed through their trades in close competitors.
`
`There are several reasons for focusing on short interest to examine the information flow
`
`across competing firms. First, a substantial literature demonstrates that short sellers are typically
`
`informed traders and that their trades predict future stock prices and fundamental firm specific
`
`information.2 Second, short sellers have superior information processing skills, being able to
`
`distinguish between bad news and seemingly neutral or positive news (Engelberg, Reed, and
`
`Ringgenberg, 2012). Third, because low levels of short interest predict significantly higher
`
`returns and upcoming good news (Boehmer, Huszar, Jordan, 2010; Akbas, Boehmer, Erturk, and
`
`Sorescu, 2013), short sellers are also good at avoiding upcoming good news. Therefore, short
`
`interest is also informative about a simultaneous long position that short sellers might take in a
`
`competing firm. This argument suggests that short sellers also trade on positive information.
`
`We assess the robustness of our findings by showing that our results are not sensitive to
`
`using different industry classifications, alternative ways of defining competing firms, and using a
`
`
`2 Among others see Diether, Lee, and Werner, 2009; Boehmer, Jones, and Zhang, 2008; Asquith, Pathak, and Ritter,
`2005, Christophe, Ferri, and Angel, 2004, Christophe, Ferri, and Hsieh 2009; Karpoff, 2010, Akbas, Boehmer,
`Erturk, and Sorescu, 2013.
`
`
`
`4
`
`CFAD VI 1070 - 0005
`
`

`
`portfolio of competing firms rather than just one competing firm. The results survive after we
`
`control for various risk factors and controlling for firm, competitor, and industry characteristics
`
`does not change our inference. Our results are not driven by the lead-lag relationship between
`
`small and big firms (Lo and MacKinlay, 1990; Hou and Moskowitz, 2005; Hou, 2007), by
`
`industry momentum (Moskowitz and Grinblatt, 1999), or by complicated firm effects (Cohen
`
`and Lou, 2012). Finally, controlling for each firm’s short-interest does not affect the predictive
`
`power of competing-firm short interest.
`
`Our analysis contributes to five strands of the financial economics literature. First, the
`
`paper adds to the studies that explore financial and product market interactions.3 Our results
`
`contribute to this literature by showing that the product market links informed trading and future
`
`price changes across related firms due to trading frictions in the stock market. A related stream
`
`of research studies the effect of information releases around a firm’s announcements on the value
`
`of related firms in the same industry. For example, Foster (1981) and Freeman and Tse (1992)
`
`show that a firm’s earnings announcements evoke price reactions and changes in the income of
`
`other firms in the same industry. Additional events associated with information transfers include
`
`bankruptcy announcements (Lang and Stulz, 1992), stock repurchases (Hertzel, 1991),
`
`accounting restatements (Gleason, Jenkins, and Johnson, 2008), managers’ voluntary earnings
`
`forecasts (Han, Wild, and Ramesh, 1989), dividend announcements (Laux, Starks, and Yoon,
`
`1998), initial acquisitions (Song and Walkling, 2000), antitrust actions (Bittlingmayer and
`
`Hazlett, 2000), and going private transactions (Slovin, Sushka, and Bendeck, 1991). In contrast,
`
`we study the trading activity of a group of investors and complement this literature by showing
`
`
`3 See Titman (1984), Brander and Lewis (1986), Maksimovic and Titman (1991), Chevalier and Scharfstein (1996),
`Allen and Phillips (2000), Gaspar and Massa (2006), Hou and Robinson (2006), Tookes (2008), Irvine and Pontiff
`(2009), Aguerrevere (2009), Hoberg and Phillips (2010), Peress (2010), and Lyandres and Watanabe (2011).
`
`
`
`5
`
`CFAD VI 1070 - 0006
`
`

`
`that the transactions of these informed traders reveal information about the prices of competitors,
`
`even without a major announcement about that competitor. Moreover, most of these studies find
`
`contagion effects. Specifically, the price reaction of the announcing and related (rival) firm go in
`
`the same direction suggesting that some common factors (i.e., industry-wide) are at work. We
`
`document, however, a cross-price impact in the opposite direction, suggesting that in our setting
`
`competitive effects are more important than industry-wide effects.
`
`Second, we contribute to the literature on price discovery in stock markets with related
`
`securities.4 This literature examines the commonality of liquidity (e.g., Chordia, Roll, and
`
`Subrahmanyam, 2000), the commonality of order flows (Harford and Kaul, 2005), or common
`
`factors affecting prices, order flow, and liquidity. We complement this literature by showing that
`
`informed trading in competing firms, a special case of commonality, also improves price
`
`discovery. Our results suggest that to understand a firm’s price discovery process fully, one not
`
`only needs to consider information about the firm itself but also must evaluate information about
`
`its competitors and about market frictions such as short selling constraints.
`
`Third, our paper contributes to the literature on short selling, which suggests that short
`
`selling is informative about future returns on the shorted stock and improves the price discovery
`
`process (Boehmer, Jones, and Zhang, 2008; Asquith, Pathak, and Ritter, 2005; Boehmer and Wu,
`
`2013; Saffi and Sigurdsson, 2011). Diamond and Verrecchia (1987) hypothesize that short sale
`
`constraints reduce the speed of price adjustments to private information, a hypothesis that finds
`
`empirical support in Reed (2007). Importantly, in this paper we show that information about
`
`
`4 A vast literature addresses price discovery in a single security or in multiple securities that represent different
`claims on the same underlying asset, such as stocks and options or stocks and futures. See, among others, Hasbrouck
`(1991), Madhavan, Richardson, and Roomans (1997), Dufour and Engle (2000), Biais and Hillion (1994), Easley,
`O’Hara, and Srinivas (1998), Chan, Chung, and Fong (2002), Chakravarty, Gulen, and Mayhew (2004), Cao, Chen,
`and Griffin (2005), Harris (1989), and Chan (1992).
`
`
`
`6
`
`CFAD VI 1070 - 0007
`
`

`
`short selling is important even beyond the firm in which it takes place. Short selling in one firm
`
`predicts the future returns and earnings surprises of competing firms, revealing a cross-firm price
`
`discovery mechanism particularly in the presence of short sale constraints.
`
`Fourth, the cross-price impact that we document is related to the concept of pairs trading.
`
`In its simplest form, this strategy involves the simultaneous purchase and (short) sale of two
`
`securities whose prices historically move together to exploit the temporary price spread between
`
`two co-integrated stocks (Gatev, Goetzmann, and Rouwenhorst, 2006). The cross-firm trading
`
`that we analyze in this paper is a type of pairs trading, involving the trading of two stocks that
`
`compete in the product market in opposite directions. The link between those stocks, however, is
`
`not purely statistical, and possibly resulting in a high correlation of returns. Instead, it is driven
`
`by an economic link driven by product market competition. We find that the cross-price impact
`
`is more pronounced among competing firms with lower return correlation (or higher negative
`
`correlation). Therefore, our results do not merely indicate a pure statistical arbitrage strategy, as
`
`described by Gatev, Goetzmann, and Rouwenhorst (2006) that relies on the mean reversion of
`
`the price spread. Instead, cross-firm trading originates from informed traders and based on
`
`changes in the fundamentals of the firms.
`
`Fifth, this paper contributes to the literature on investors’ delayed reaction to publicly
`
`available information. In this literature, delayed reaction might be due to investors’ limited
`
`attention, physiological biases, or limits to arbitrage.5 We show that information that is readily
`
`publicly available, i.e. competing firms’ monthly short interest, is only slowly incorporated into
`
`
`5 Among others, please see the following papers for an extensive theoretical argument and empirical support for the
`delayed reaction view: Kahneman and Tversky (2000), Foster, Olsen, and Shevlin (1984), Bernard and Thomas
`(1989), Daniel, Hirshleifer, and Subrahmanyam (1998), Hong and Stein (1999), Hong, Lim and Stein (2000),
`Huberman and Regev (2001), Ikenberry and Ramnath (2002), Hirshleifer and Teoh (2003), Hirschey and
`Richardson (2003), Barber and Odean (2008), DellaVigna and Pollet (2006), Hou (2007), Menzly and Ozbas (2006),
`Hong, Torous, and Valkanov (2007), Zhang (2006), Boehmer, Jones and Zhang (2008), and Duffie (2010).
`
`
`
`7
`
`CFAD VI 1070 - 0008
`
`

`
`the stock prices of a firm. While prior studies find delays in the processing of cross-industry
`
`information flows (Menzly and Ozbas, 2006), information flows from firms that have strong
`
`economic links (Cohen and Frazzini, 2008), and industry wide information flows when firms are
`
`difficult to evaluate (Cohen and Lou, 2012), we show that information flows about competitors
`
`are also processed with a delay.
`
`The paper proceeds as follows. Section 2 describes the selection of competing firms and
`
`the construction of the sample. Section 3 presents the main results on the effect of short interest
`
`in competing firms. Section 4 explores the nature of this cross-price impact, and Section 5
`
`discusses short sellers’ incentives to trade in the competing firms. Section 6 performs a battery of
`
`robustness tests using alternative competitor firm and industry definitions, and Section 7
`
`concludes.
`
`
`
`2. Data and variables
`
`2.1. Computing competing firms’ short interest
`
`We measure short sellers’ activity using monthly short interest (SHORT) for each stock,
`
`defined as total shares shorted divided by the total shares outstanding measured mid-month. We
`
`use the Kth Neighborhood method to determine a firm’s closest competitor within an industry.
`
`First, we define industries using the SIC3-industry classification. In robustness tests, we obtain
`
`qualitatively identical results using 4-digit SIC, 3- or 4-digit NAICS industry definitions, and
`
`Hoberg and Phillips’s (2010) industry classifications based on text-based analysis of firms’
`
`product descriptions.
`
`Second, for all firms within an industry, we calculate the absolute distance of its market
`
`share to all other firms in that industry. The closest competitor of a firm is determined, with
`
`
`
`8
`
`CFAD VI 1070 - 0009
`
`

`
`replacement, as the one that has the smallest absolute difference in market share. In rare cases,
`
`more than one competitor shares the same absolute difference with a firm. In these cases, we take
`
`all these peers as competitors. Finally, C_SHORT is defined as the short interest in the
`
`competing firm (or the average, if there is more than one closest competitor). Although all firms
`
`in the same industry compete with one another to some extent, we argue that firms with similar
`
`market share in an industry should have the greatest informational connection and, in turn,
`
`influence each other’s stock prices more.
`
`In robustness tests, we use different ways of classifying competing firms. First, we use a
`
`firms’ price-cost margin (the Lerner Index), instead of market share, to find a firm’s closest
`
`competitor. Second, we define C_SHORT as the average of the two, three, and four closest
`
`competitors’ short interest or as the average of all firms’ short interest within the same market
`
`share cluster6 or within groups formed simply by ranking based on market share.7 The results for
`
`each of these approaches are qualitatively the same and even larger in magnitude.
`
`2.2. Data
`
` The sample consists of all NYSE or NASDAQ-listed common stocks for which monthly
`
`short interest reports are available over the period from June 1988 to December 2012. To ensure
`
`that the results are not driven by extremely illiquid stocks, we exclude stocks with a previous
`
`month-end price below five dollars. We also require the availability of at least 12 months of past
`
`return data from CRSP and accounting data with industry classification codes from
`
`COMPUSTAT for both a firm and its closest competitor.
`
`
`6The clustering procedure forms groups of firms so that the firms within a group, or cluster, have market shares
`more similar to each other than to firms outside the group. In other words, the difference between members of
`different clusters is greater than the differences between members of the same cluster. We use a classification
`algorithm similar to Gitman (1973) and Huizinga (1978) and provide more details in the robustness section.
`7 Ranking creates groups of equal frequency independent of the distribution of market shares within the industry. In
`contrast, clustering relies on non-parametric density estimation and can form groups of different sizes, depending on
`how market shares are distributed within the industry.
`
`
`
`9
`
`CFAD VI 1070 - 0010
`
`

`
`Our control variables are defined as follows. SIZE is the market value of equity,
`
`calculated as the previous month-end number of shares outstanding times share price. BM is the
`
`ratio of the previous quarter-end book to market value of equity. TURN is the monthly share
`
`turnover ratio, measured as the number of shares traded divided by the number of outstanding
`
`shares. C_TURN is the monthly turnover ratio of the close competitor firm and controls for the
`
`trading activity in the competitor firm. In regression analyses, we use the log of SIZE and BM
`
`because both variables display considerable skewness. IVOL is idiosyncratic daily return
`
`volatility, calculated as the standard deviation of daily residuals from the Fama-French (1993)
`
`model over a month as in Ang, Hodrick, Xing, and Zhang (2006). To calculate this variable, we
`
`require at least 15 days of daily return data. IO measures institutional ownership, defined as the
`
`sum of the holdings of all institutions for each stock in each quarter divided by the number of
`
`shares outstanding from CRSP. Stocks with available return data but no reported institutional
`
`holdings are assumed to have zero IO. Data on IO come from 13-F filings through Thomson
`
`Financial.
`
`We control for lagged returns of each stock, MOM, its competitors, C_MOM, and its
`
`industry, IND_MOM. MOM is the cumulative return over the past twelve months and captures
`
`the momentum effect documented by Jegadeesh and Titman (1993). IND_MOM controls for
`
`industry momentum (Moskowitz and Grinblatt, 1999) and is computed as the equally-weighted
`
`industry return over the past twelve months. Finally, to control for within-industry lead-lag
`
`effects (Hou, 2007), we calculate the corresponding returns to small firms, IND_MOM
`
`_SMALL, and large firms, IND_MOM _BIG, within each industry. To calculate returns to small
`
`and big firms in each industry, every month we group firms into three groups by their previous
`
`month-end market capitalization. Then the return to big firms is the equally weighted average of
`
`
`
`10
`
`CFAD VI 1070 - 0011
`
`

`
`firm returns in the top size tercile and the return to small firms is the equally weighted average of
`
`firm returns in the bottom size tercile.
`
`Table 1 shows the summary statistics and correlations for these variables. In Panel A, we
`
`show the time-series averages of cross-sectional statistics. We have on average 1893 stocks per
`
`month in our sample. The average (median) short interest (SHORT) is 3.19% (1.91%) for the
`
`sample. For the set of closest competitors, the average (median) short interest (C_SHORT) is
`
`2.96 % (1.73%). The average firm size is 2.98 billion dollars, and firm size exhibits considerable
`
`variation, suggesting that our sample contains both small and big firms.
`
`Panel B shows the correlation matrix. The correlation between short interest, SHORT,
`
`and competitor firms’ average short interest, C_SHORT, is 0.282. The moderate positive
`
`correlation suggests that the two measures largely capture different information. Our main
`
`variable of interest, C_SHORT, is positively correlated with size and turnover and negatively
`
`correlated with book-to-market. These correlations are smaller than the correlation between
`
`firm’s own short interest, SHORT, and firm size and turnover. Correlations of competitor firms’
`
`average short interest, C_SHORT, with other stock and industry variables are less than 0.3.
`
`In Table 2, we present portfolio sorts as a precursor to our main regression analysis. Each
`
`month, we first group our sample into short interest (SHORT) quintiles and then rank them based
`
`on their competing firms’ average short interest (C_SHORT) within each short interest (SHORT)
`
`quintile. All sorts are independent. We report time series averages of equally weighted monthly
`
`portfolio returns and alphas obtained from Carhart’s (1997) four-factor model. We skip a month
`
`between the portfolio formation period and the holding period. Our results show a positive and
`
`significant relationship between future returns and the short interest in its closest competitor.
`
`This relationship is most pronounced for high short interest stocks, and it survives after we adjust
`
`
`
`11
`
`CFAD VI 1070 - 0012
`
`

`
`for risk using the Carhart four factor models. The raw (risk adjusted) return difference between
`
`the highest and lowest C_SHORT quintiles ranges from 18 (9) to 71 (70) basis points per month,
`
`suggesting a high economic impact particularly among stocks in the top short-interest quintile.
`
`Similar return difference for short interest of the firm ranges from 37 (43) to 102 (118) basis
`
`points per month. This suggests that the predictive power of competing firm short interest is
`
`about two thirds of the predictive power of a firm’s own short interest. Thus competing-firm
`
`short interest is economically important, both in an absolute sense and relative to the well-
`
`established short-interest effect on future returns.
`
`
`
`3. Competing firm short interest and future returns
`
`In this section, we use Fama-MacBeth (1973) regressions to test the main hypothesis of
`
`our paper: that short interest contains information about competing firms. Specifically, we
`
`regress a firm’s future returns on the short interest in its closest competitor, controlling for
`
`characteristics that might differentiate the two firms. To avoid potential bid-ask bounce effects
`
`on our estimates, we skip one month between formation and holding period in all our regression
`
`tests. The main variable of interest is competing-firm short interest (C_SHORT), and we test
`
`how it is related to a firm’s future return at different horizons. The regressions include various
`
`firm and industry controls that are known to affect returns. To account for autocorrelation in the
`
`monthly time-series of cross-sectional Fama-MacBeth coefficients, we use Newey-West standard
`
`errors with six lags.
`
`Table 3 displays determinants of future one-month returns. The first specification is
`
`analogous to our portfolio double sorts, except that we now use continuous variables to measure
`
`short interest (SHORT) and competing-firm short interest (C_SHORT). The second specification
`
`
`
`12
`
`CFAD VI 1070 - 0013
`
`

`
`is our baseline model with controls for the log of market capitalization (SIZE), the log of the
`
`book-to-market ratio (BM), share turnover (TURN), idiosyncratic return volatility (IVOL),
`
`institutional ownership (IO), and past twelve-month return (MOM).
`
`Cohen and Lou (2011) argue that price reaction of an easy-to-analyze firm may lead the
`
`price reaction of a complicated firm, when both are subject to a common shock. If a competing
`
`firm is an easy-to-analyze firm, we could find a relationship between a firm’s future return and
`
`competing-firm short interest just because short interest and past returns are correlated for the
`
`competing-firm (Diether, Lee, and Werner, 2009). In the third specification, we address this
`
`possibility by including firms’ past 12-month cumulative return (C_MOM) in our regressions.
`
`Moreover, short interest in competing firm might simply proxy for overall trading interest in a
`
`set of firms including the competitor and the firm. In the literature several papers (e.g. Chordia
`
`and Swaminathan, 2000, Hou and Moskowitz, 2005) document that interest in a firm as
`
`measured by trading activity affects the lead-lag affects in returns and predicts future returns. To
`
`address this possibility, we control for trading activity measured as share turnover, C_TURN, in
`
`the competing firm.
`
`In the fourth specification, to ensure that our results are not driven by the industry
`
`momentum effect documented by Moskowitz and Grinblatt (1999), we use as a control the
`
`lagged 12-month returns of the firm’s industry, IND_MOM. In the fifth specification, we control
`
`for within-industry lead-lag effects (Hou, 2007) by adding average past returns for small firms
`
`(IND_MOM_SMALL) and for large firms (IND_MOM_BIG) in each industry. Finally, although
`
`the correlation between SHORT and C_SHORT is not large, in the sixth specification we
`
`
`
`13
`
`CFAD VI 1070 - 0014
`
`

`
`exclude a firm’s own short interest (SHORT) to make sure that our results are not affected by a
`
`potential multicollinearity problem.8
`
`The results in Table 3 show that, regardless of specification, a firm’s future one-month
`
`return is positive and significantly related to competing-firm short interest. The coefficients on
`
`C_SHORT range from 0.015 to 0.019, with corresponding t-values of 2.67 to 3.21. As expected,
`
`and as consistent with prior studies, SHORT is always negative and significantly related to future
`
`returns. In our main specification (column 2), the coefficients on C_SHORT and SHORT imply
`
`that, in magnitude, after controlling for other firm characteristics the predictive power of
`
`competing firm short interest is about one third of the predictive power of a firm’s own short
`
`interest. Thus competing-firm short interest is economically important, both in an absolute sense
`
`and relative to the well-established short-interest effect on future returns.
`
`When we examine the control variables, SIZE is negatively and IO is positively related to
`
`future returns, and both are significant. IVOL has a marginally significant negative coefficient,
`
`while BM has a positive but insignificant coefficient. We further estimate positive and
`
`significant coefficients for IND, MOM, and BIG. Most importantly, C_SHORT preserves its
`
`significance after we control for other firm characteristics, including size, IO, BM, and the
`
`competing firm momentum, trading activity and, industry returns. Thus, our results are not
`
`driven by complicated-firm effects, lead-lag effects, or competing-firm momentum.
`
`Overall, the results in Table 3 suggest that future stock returns are positively related to
`
`competing firms’ short interest level. This effect is robust to the inclusion of various controls
`
`such as the firm’s own short interest, stock price momentum, size, book-to-market, institutional
`
`
`8 Because the correlation between IND_MOM and IND_MOM_BIG is high, we do not include these variables
`together in the same model. However, the results on the coefficient of C_SHORT and SHORT are similar if we
`include all these variables simultaneously.
`
`
`
`14
`
`CFAD VI 1070 - 0015
`
`

`
`ownership, and idiosyncratic volatility. Moreover, industry lead-lag effects, competitors past
`
`returns and trading activity, or industry momentum do not explain our findings.
`
`
`
`4. Exploring the nature of information in competing firm short interest
`
`Our previous tests establish a positive relationship between the competing firm short
`
`interest and that firm’s returns at monthly frequency. To better understand the natur