Short-Sellingand the Accrual Anomaly
By
Stephen E. Christophe
GeorgeMasonUniversity
School of Management, MS5F5
Fairfax, VA 22030
Tel. (703) 993-1767, FAX (703) 993-1767
Michael G. Ferri
GeorgeMasonUniversity
School of Management, MS5F5
Fairfax, VA 22030
Tel. (703) 993-1893, FAX (703) 993-1767
and
James J. Angel
GeorgetownUniversity
The McDonoughSchool of Business
Room G-4 Old North
Washington, DC 20057
Tel. (202) 687-3765, FAX (202) 687-4031
February, 2005
______
Direct correspondence to: Stephen E. Christophe, GeorgeMasonUniversity, School of Management, MS5F5, Fairfax, VA 22030, (703) 993-1767, . We thank the Nasdaq Stock Market for providing the data. Angel and Ferri gratefully acknowledge financial support from the Nasdaq Educational Foundation. The views expressed in this paper are those of the authors and do not necessarily reflect the views of the Nasdaq Stock Market, Inc., the Nasdaq Educational Foundation, or anyone else.
Short-Selling and the Accrual Anomaly
ABSTRACT
This study employs newly available information on the daily short-selling of Nasdaq stocks from September 2000 to July 2001 in an effort to analyze how short-sellers respond to the Form 10K disclosure of the accrual component of corporate earnings. Research by Sloan (1996) showed that accrual earnings are less persistent than cash earnings and that the market had failed to recognize that fact. Richardson (2003) found no evidence that short-sellers aggressively pursued high accrual firms. Unfortunately, he was constrained to work with the data of the stock exchanges’ monthly short interest report. The Nasdaq dataset we analyze is far more precise than the short interest and permits tests of daily patterns in short-selling. The test results support the view that investors are ready to use short-selling to exploit the market’s failure to adjust downward the share prices of firms that rely heavily on accrual-based earnings.
Short-Selling and the Accrual Anomaly
I. Introduction
This study utilizes information on the daily short-selling of Nasdaq stocks to analyze how short-sellers respond to disclosure of the accrual component of corporate earnings. That disclosure takes place in the Form 10K, which the Securities and Exchange Act of 1934 requires companies with publiclytraded securities to file annually with the Securities and Exchange Commission (SEC). While the 10K releases data about many aspects of a firm’s financial situation and performance, it has the particularly important function of informing investors about the portion of the company’s reported earnings that takes the form of accruals.
The composition of earnings acquired special significance after Sloan (1996) found that (1) earnings from accruals exhibit lower future persistence than earnings from cash flows, and (2) this difference in persistence is not fully reflected in stock prices. This anomaly presents a potential arbitrage opportunity whereby investors could short the stocks of companieswith high accruals as a portion of total earnings and establish long positions in the stocks of companies with high cash flow as a percentage of total earnings. In fact, Sloan demonstrated that such a strategy can generate positive abnormal returns, though he acknowledged that the strategy could entail substantial costs from both the acquisition of information and the price impact from the action of shorting targeted shares.
Richardson (2003), however,finds no empirical evidence that short-sellers have attempted to exploit the accrual anomaly. His research reveals that short interest (reported for companies once per month) is relatively invariant across quintiles that he based on his sampled companies’ levels of accruals relative to earnings. He considered several explanations for this phenomenon: high accrual firms are more costly and/or difficult to short; high accrual firms are more risky to investors and therefore carry potentially high downside payoff potential; and investors may not be aware of the accrual anomaly.
Anotherpossible explanation forRichardson’s empirical resultsis that the short interest data he utilizes in his study is limited in two important ways. First, itrepresents only a snap-shot of total shorted shares on one day during the month. Further, theexchange’s monthly short interest report aggregates the short positions of market participants as different in motivation and planning horizon as dealers, option-market arbitrageurs, and investors expecting price declines. As a result of either or both limitations, the data-set Richardson worked with is probably much too coarse to capture the day-by-day activities of short-sellers who are engaging in investment strategies aimed at exploiting the accrual anomaly.
Recently, additional and far more detailed data on short-selling in Nasdaq stocks havebecome available, and these data allow a precise examination of daily short-selling specifically by investors who expect price declines.[1] The purpose of our study is toanalyzethese data to examine shorting activity around SEC10Kfiling dates to further explore how short-sellers react to the disclosure of the accrual versus cash component of earnings.
We uncover evidence that short-sellers establish positions consistent with an attempt to exploit the accrual anomaly. For example, we find significantly higher post 10K filing date short-selling for high versus low accrual firms. And, this result is robust to both the inclusion and exclusion of firm’s that did not file their 10K forms in a timely manner at the SEC. Further, we reconcile our results with Richardson’s (2003) by showing that although there is a statistically significant relationship between reported accruals and daily short selling, there is no significant relationship between reported accruals and monthly short interest. We provide first evidence that some of the component parts that are aggregated to obtain short interest are negatively correlated with one another and argue that this negative correlation makes a company’s reported short interest a rather crude statistic for examining whether short sellers establish investment positions in an attempt to exploit the accrual anomaly.
The paper proceeds as follows: Section II more fully describes prior research that examines different aspects of the accrual anomaly and short selling; Section III describes our short-selling data, and the sample of Nasdaq firms examined in our analysis; Section IV contains estimations and results including a reconciliation of ourfindings with prior research that looks at monthly short interest; and, the final Section presents our conclusions.
II. Background
In seminal work, Sloan (1996) finds that investors do not fully recognize the implicationsfor future earningscontained in a company’s disclosure of the accrual and cash flow components that comprise its current earnings. Instead, investors act in a manner consistent with overestimating the persistence of earnings that arise from accruals, and underestimating the persistence of earnings that arise from cash flows. Sloan demonstrates that these mis-estimations lead to apparent stock pricing errors such that a trading strategy based upon establishing a long position in the stocks of companies which report high cash flows relative to earning and a short position in companies reporting high accruals relative to earnings can potentially result in arbitrage profits.
Lev and Nissim (2004) document that this anomaly continues to persist in recent years. Theyalso provide some evidence that institutional investors adjust their portfolio holdings in reaction to accrual disclosures. For example, they examine 13(f) filings and find a negative relationship between a firm’s reported accruals and the reported institutional ownership of that firm’s shares in the subsequent quarter.[2] This suggests that institutions trade on accrual information during the quarter during which that information is released. In addition, they find that in recent years institutions have more intensely altered their share holdings in response to accrual disclosures. This increase in the intensity of the response is consistent with an increasing awareness of the accrual anomaly. However, they also report that extreme accrual firms tend to have characteristics that tend to make them unattractive to most institutional investors. For example, they tend to be small firms with a low stock price and book to market ratio. Consequently, even though institutions appear to adjust holdings in recognition of the accrual anomaly, lack of institutional ownership of many extreme accrual firms may help explain why the anomaly persists.
Consistent with Richardson’s (2003) finding, Mashruwala, Rajgopal, and Shevlin (2004) argue that arbitrageurs may not eliminate the accrual anomaly because the trades required to exploit the anomaly can be risky and/or costly. They find that there is a 49% chance that pursuit of the strategy could be unprofitable in any given year, and that the strategy shows negative profits during 5 months of the median year. In addition, they report that extreme accrual firms lack close substitutes against which to lay-off risk. If arbitrageurs only want to pursue a limited number of perceived mispriced investment opportunities (Shleifer and Vishny (1997)), they may be wary of employing an accrual anomaly-based strategy due to the common occurrence of losses during certain years and/or months during the year (either of which could induce fund investors to withdraw their investments in the fund).
A number of other studies have begun to examine how institutional ownership can lead generally to constraints on short-selling (though these studies do not necessarily focus specifically on short-selling constraints and the accrual anomaly). For example, Chen, Hong and Stein (2002), and Nagel (2004) both suggest that low institutional is a likely indicator of a stock that may be short sale constrained (due to a general unavailability of lendable shares). Nagel suggest that a limited supply of shares may help explain why some types of return predictability are not arbitraged away. For example, consistent with this argument, he reports that return predictability increases sharply for stocks with low institutional ownership. Asquith, Pathak, and Ritter (2005) report that although short sales constraints are not binding for most companies, stocks that exhibit both high short interest and low institutional ownership (which together imply the existence of short sale constraints) exhibit significantly negative abnormal returns when measured from a four-factor model regression with the stocks placed in an equally-weighted portfolio.
Short sales constraints of high accrual firms could help explain Richardson’s (2003) finding that four months after company fiscal year-ends, there is no statistically significant difference between the short interest of high accrual versus other firms. Instead, he finds that short interest across accrual quintiles is relatively constant at about 1.2% to 1.25% of shares outstanding. He concludes that short sellers apparently do not utilize the information contained in accrual disclosures, and are therefore foregoing a potential profit opportunity. However, Richardson’s (2003) utilization of the monthly report of a stock’s short interest as his dependent variable in empirical estimation potentially represents the choice of a very noisy statistic. As described in substantial detail below, short interest is an aggregation of share sold short by many different types of investors/dealers. Therefore, it is worthwhile to further examine the relationship between disaggregated daily short-selling and accruals to provide additional insights into the behavior of investors.
III. Data and Sample
The source of our short-selling data is Nasdaq’s Automated Confirmation Transactions Service (ACT), which processed the vast preponderance of trades in stocks listed on Nasdaq’s National Market System (NMS). Market participants in Nasdaq-listed stocks are required to report their trades to ACT within 90 seconds of execution.[3] In addition to price and volume statistics, each ACT record contains a code which indicates whether the seller is engaged in short selling. From the Nasdaq, we received all ACT records formore than 3,000 NMS stocks on the 207 trading days between September 13, 2000, and July 10, 2001. We aggregated the records of each stock’s trades within each day and conduct our analysis on daily short-selling data.
The ACT records allow us to separate short sales into several different categories. For example, we are able to identify short sales that are exempt from the bid test because the trade represents such activities as options arbitrage.[4] We can also identify the shorting activities of dealers, who short sell mostly to manage their inventories and carry out their normal function of market-making. Finally, we can identify short sales that are designated “customer short” which represent investors establishing short positions (we believe) in anticipation of stock price declines (or relative underperformance)
This final group of short sales (“customer short”) and the associated shares that are shorted represent the focal point of our analysis. Specifically, we examine the data on customer behavior for evidence that short-selling around 10K filings is different from other times and that the difference does indicate a response to the accrual information disclosed in the filings. Thus, a rise in short-selling from a typical to a high level around a filing that reveals high accruals is evidence that investors expect the price decline that Sloan (1996) uncovered. By contrast, if short-selling around the filing is basically the same as at other times and the same for firms of varying amounts of reported accruals, this would tend to undermine the view that investors pursue accrual-based strategies.
We restricted our initial sample from ACT to the stocks that traded every day over the 207 day sample period and averaged at least 50 trades each day. In our view, this criterion mitigates the possibility of making improper inferences that reflect peculiar features of infrequently traded stocks. This initial sample consists of a total of 1,312 stocks for which we obtained daily return data from CRSP.[5]
We next merged this initial sample with financial statement data obtained from Compust where we imposed the additional restriction that Compustat contain sufficient information to estimate each company’soperating accruals for year-end 2000. Following Richardson (2003), we define operating accruals (OA)in this way:
OA = (Earnings before extraordinary items – Cash flow from operations)/AverageAssets[6]
The sample obtained from this merging of data in ACT, CRSP, and Compustat consists of 1,217 stocks.
Although the Compustat database containsa record of each company’s earnings announcement date, it is appropriate for our analysis to identify the firm’s 10K filing date because this date is typically the first point in time when investors can definitively discern the composition of the firm’s earnings (accrual versus cash components). To obtain 10Kfiling dates, we conducted an online search of the records in That search provided filing dates for 1,101 stocks in our sample.
The next set of restrictions to the sample included eliminating24 financial services companies (SICs 6000-6999)and 475companies whose stock prices were less than $10 on the10K filing date. The latter criterion derives from evidence presented by D’Avolio (2002) that companies with low stock prices may be difficult to short. This restricted sample, then, consistsof 602 stocks which we posit represents the opportunity set of Nasdaq stocks from which investors could have chosen, following fiscal year-end 2000 accounting disclosures, if they had attempted to implement an investment strategy based upon accruals.
Our empirical methodology (described in more detail below), however, limit us to examining the short-selling characteristics for a sub-set consisting of 437 of these companies. To examine how a company’s short-selling differs during the 10K filing period compared to its typical amount of short-selling we construct a measure of each company’s normal daily short-selling and normal daily volume. To estimate these variables,we rely on the trading and short-selling of a company’s stock that occurs overthesixcalendar weeks (approximately 30 trading days) ending one week prior to the company’s earnings announcement. In addition, to minimize the possibility that empirical findings are due to the short-selling effects surrounding the earnings announcement rather than the 10K filing,we also restrict the sample to only those firms with a 10Kfiling date at least 7 calendar days after their earnings announcement date (as reported on Compustat).[7]
Because the starting point of our short-selling data is September 13, 2000, these restrictions entail that the 10Kfiling date must be at least (6+1+1) eight weeks after September 13, 2000. Finally, to enable an examination of a stock’s short-selling subsequent to the 10K filing, we require that the filing date is at least 10 days prior to July 10, 2001 (the last day of our short-selling sample data). The imposition of these methodological restrictions results in a final sample of 437 companies.
Similar to Richardson (2003), we classify stocks into quintiles based on their level of OAat fiscal year-end 2000. We have chosen to form those quintiles on the basis of the distribution of OA for the full set of 602 stocks because, as mentioned above, we assume these stocks represent the opportunity set from which investors could have chosen if they were pursuing an investment strategy focused on accruals. The lowest quintile (Q1) consists of firms with OA<= -0.1407; the second lowest quintile (Q2) contains firms with -0.1407 < OA<= -0.0766; the middle quintile (Q3) is comprised of firms with -0.0766 < OA<= -0.0316; the second highest quintile (Q4) consists of firms with -0.0316 < OA<= .0241; and, the highest quintile (Q5) contains all firms with .0241OA.