Short Selling Around Restatement Announcements

Abstract

This article draws on two distinct literatures—one that investigates the impact of accounting restatements and another that investigates the trading behavior of short sellers—to further our understanding of how sophisticated investors process and respond to news about accounting corrections. We examine short-seller behavior in the days surrounding restatement announcements to determine when short sellers identify restatements and how they trade on restatement news. Our findings suggest that short sellers do not appear to anticipate restatement announcement dates, but we find that abnormal short selling is significantly higher than is typical when restatements are announced, especially for restatements announced transparently (i.e., in press releases or 8-Ks). Short sellers target small companies, whose information environments may be weaker, and companies making restatements that reduce previously reported income. In addition, we find that restating companies targeted most heavily by short sellers experience the most negative subsequent abnormal returns over horizons of up to 40 trading days following the restatement disclosure. Overall, our results suggest that short sellers respond to, but do not anticipate, restatement announcements and trade as if they understand the post-announcement stock price implications of restatement disclosures.

Keywords

restatement short sellers disclosure short interest transparency

Introduction

In this article, we study short selling activity around a sample of accounting restatements made to correct material errors in previously issued financial statements. Restatement announcement dates are likely to be particularly attractive targets for short sellers because short sellers have relatively short investment horizons (Boehmer, Jones, & Zhang, 2008) and because prior literature documents significantly negative short-window returns following restatement announcements (Desai, Krishnamurthy, & Venkataraman, 2006; Gleason, Jenkins, & Johnson, 2008; Myers, Scholz, & Sharp, 2013; Palmrose, Richardson, & Scholz, 2004).

Short sellers are sophisticated market participants (Diamond & Verrecchia, 1987), so examining their trading activities in the days around restatement announcements provides important insights into how markets process these corrective disclosures. One innovation in our approach is that we use daily short selling data to examine the trading behavior of short sellers in very short windows. This allows us to draw inferences with respect to how short sellers trade on the information provided in restatement announcements. Specifically, we investigate at what point short sellers respond to restatement announcement information and whether short selling activity is associated with restatement severity, disclosure transparency, or the restating companies’ information environments. Finally, we analyze subsequent returns to see whether short sellers target restating companies that experience more severe price declines in the weeks following their restatement announcements.

Prior evidence finds elevated levels of short selling in restating companies long before the restatements are announced, perhaps for reasons not directly related to the restatements. Restating companies tend to perform poorly prior to the restatement announcements (Ettredge, Scholz, Smith, & Sun, 2010; Scholz, 2008) and the accounting errors corrected by the restatements may have persisted for several years, allowing sophisticated short sellers to detect and trade on other evidence of questionable reporting, such as poor accrual quality (Desai, Krishnamurthy, & Venkataraman, 2006; Hirshleifer, Teoh, & Yu, 2011).

U.S. Securities and Exchange Commission (SEC) rules require the disclosure of a material restatement within 4 days of discovery. Between the discovery and disclosure, Regulation Fair Disclosure restricts managers from privately disclosing this material information, but several additional parties (e.g., members of the board of directors, auditors, attorneys, and possibly the SEC) may be aware of the impending announcement.¹ Although information leakage from any of these sources violates confidentiality and insider trading rules, short sellers may discover that an announcement is imminent. Short sellers who obtain information about upcoming restatements may decide to trade on that information because the perceived probability of being investigated may be low or, if challenged by the SEC, short sellers could argue that their trades are based on low financial reporting quality rather than on inside information.²

In our analyses, we examine daily abnormal short selling activity in pre-announcement, announcement, and post-announcement windows surrounding the disclosure of 740 restatements announced from 2005 through 2007. We identify which restatement characteristics are associated with short-seller behavior and determine whether disclosure transparency has an effect on the short-seller reaction. Our analyses also consider pre-existing short interest, and so shed light on whether the short-seller response to restatement announcements is driven by their ability to predict the upcoming disclosure and/or by their ability to quickly respond to the information once it is announced.

We find that short volume is abnormally high for restating companies for at least 90 days prior to the restatement announcement, but we find no evidence of an increase in daily abnormal short selling in the 5 days immediately preceding the announcement, nor of an association between pre-announcement short selling activity and announcement returns. These patterns suggest that although short sellers may detect leading indicators of poor accounting quality, they are not informed of imminent restatement announcements (i.e., information leakage is not reflected in short selling behavior prior to restatement announcements) or are unwilling to trade on this information because such trades could be construed as violating insider trading laws. Although short sellers do not appear to front-run restatement announcements per se, our evidence does suggest that in the days just before the restatement is announced short sellers are more active in companies with a history of restating earnings, which is consistent with the notion that short sellers are attuned to overall earnings quality signals.

At the time of the restatement announcement (i.e., in Days [0, +1]), we find a marked increase in abnormal short selling, which reverts to the pre-announcement level in the post-announcement period (i.e., in Days [+2, +5]). The increase in short selling at the time of the restatement announcement is attributable mainly to restatements disclosed transparently (i.e., in press releases or on Form 8-Ks) rather than to more opaque disclosures (i.e., those in other SEC filings such as Form 10-K). Regardless of how transparently restatements are announced, we find that short sellers are more active in companies with the greatest decreases in previously reported income and are less active in companies that increase previously reported income. We also find that among restatements disclosed transparently, short sellers are more active in smaller companies, perhaps because smaller companies have weaker information environments on average and so are more likely to be mispriced. Among restatements announced obscurely, short sellers are more active when the market reaction to the announcement is less negative, suggesting that short sellers look for opportunities where the initial market reaction to the restatement may be incomplete. This quick and adept response is consistent with results from prior work investigating short-seller reactions to other information events (e.g., Ben-David, Drake, & Roulstone, 2013; Edwards & Hanley, 2010; Engelberg, Reed, & Ringgenberg, 2012) and also with the notion that short sellers have prior familiarity with restating companies separate from the restatement announcement.

Our final set of analyses provides evidence that short sellers likely earn significant abnormal returns from their trades around restatement announcements.³ Here, we construct three portfolios of restating companies based on the magnitude of short trading at the restatement announcement (low, medium, and high short trading portfolios). Over holding periods of 10 through 50 trading days, only the high short trading portfolio shows significant negative abnormal returns. These results are consistent with the idea that short sellers are able to profitably trade on information from restatement announcements.

Overall, our evidence suggests that short sellers respond to, rather than anticipate, restatement announcements, and that disclosure transparency and restatement characteristics have important effects on their response. When restatements are announced transparently, short sellers are more active in smaller companies, and when announced non-transparently, short sellers are more active when the announcement return is less negative. Thus, our evidence suggests that short sellers identify instances where the market underreacts to restatement announcements and that their strategies are likely profitable.

This article draws on two distinct literatures—one that investigates the impact of restatements and another that investigates the trading behavior of short sellers—to further our understanding of how sophisticated investors process and respond to news about accounting corrections. Our analyses of daily short volume data in short windows around restatement announcements provide important improvements over prior studies that use monthly short interest levels over long windows (Desai, Krishnamurthy, & Venkataraman, 2006; Efendi & Swanson, 2009), because they increase the likelihood that short-seller activities are directly related to the restatement characteristics rather than to other information about earnings quality (e.g., high accruals or poor overall financial condition). Thus, our analyses are less susceptible to alternative explanations and allow us draw stronger inferences with respect to causality. The results provide information related to recent concerns from regulators and investors by demonstrating how trading by a group of sophisticated investors varies with restatement disclosure transparency.

Prior Literature and Empirical Predictions

Short Selling

The costs and risks associated with short positions suggest that only informed traders will choose to sell short. Consistent with this, Diamond and Verrecchia (1987) argue that the high cost of short selling prevents uninformed traders from taking short positions. Their analytical model shows that unexpected increases in short interest predict future price declines. Building on Diamond and Verrecchia (1987), subsequent empirical studies find a negative association between high short interest and future abnormal returns (e.g., Dechow, Hutton, Meulbroek, & Sloan, 2001). Also consistent with the insight that short sellers will be informed traders, Boehmer et al. (2008) report that institutional investors and hedge funds account for about 74% of all short sales in their sample, that individual investors account for less than 2% of all short sales, and that the balance consists primarily of market-makers shorting for liquidity reasons. They conclude that short sellers are “extremely well informed” about future stock price movements (Boehmer et al., 2008, p. 524). Finally, Cassell, Drake, and Rasmussen (2011) find that short-seller trading decisions are significantly correlated with the decisions of another informed party, the external auditor.

Another line of short selling research investigates whether short sellers appear to anticipate bad news. Using a sample of NASDAQ stocks from September 13 through December 12, 2000, Christophe, Ferri, and Angel (2004) examine daily short interest over the days preceding earnings announcements and find that short sellers increase their positions in advance of negative earnings news. However, using a sample of NYSE stocks from a more recent time period (April 2004 through March 2005), Daske, Richardson, and Tuna (2006) find no evidence that short sellers anticipate negative earnings news in the days leading up to earnings announcements.

Most relevant to our study, prior research investigates whether short sellers seem to anticipate that companies may restate over long horizons. Efendi and Swanson (2009) find that monthly short interest is higher in restating companies than in control companies beginning 15 months before the restatement announcements, and they find that the difference in short interest between restating and control companies is attributable to restatements with accounting irregularities.⁴ Similarly, Desai, Krishnamurthy, and Venkataraman (2006) find that monthly short interest levels are significantly higher for restating companies beginning 18 months before a restatement announcement than for control companies matched on size and book-to-market.⁵ In general, Desai, Krishnamurthy, and Venkataraman (2006) find little evidence of an increase in the level of short interest as the announcement date nears: short interest 1 month before the disclosure does not differ from short interest 6 or 12 months prior, and differs from 18 months prior for only the subsample of companies with high income-increasing accruals. The authors do not investigate short selling activity at the time of the restatement announcement, but they conclude that “short sellers can identify suspect financial reporting in advance of public disclosure” (Desai, Krishnamurthy, & Venkataraman, 2006, p. 89), perhaps by “[paying] attention to information being conveyed by accounting accruals” (Desai, Krishnamurthy, & Venkataraman, 2006, p. 84). Thus, this evidence is also consistent with short sellers trading on indicators of poor earnings quality which are correlated with, but often do not result in, a subsequent restatement.

The long-window designs in these prior studies were necessary because daily short interest data were previously unavailable but more refined measures of short interest are now feasible because of increased data availability. These measures provide more powerful tests of the relations between short selling and events (Henry & Koski, 2010) and allow us to overcome the challenges of a long-window design. Specifically, they allow us to directly examine the extent to which short sellers anticipate imminent restatements and restatement characteristics (rather than the extent to which they react to other information about earnings quality—such as accrual levels or poor overall financial condition). Furthermore, Daske et al. (2006) and Boehmer et al. (2008) suggest that short interest is dominated by relatively short-term trading strategies (i.e., average short positions are estimated at 33 and 37 trading days, respectively). Thus, to the extent that short sellers maintain their positions for only a short period of time, our daily measure better captures their behavior surrounding restatement announcements.

Financial Statement Restatements

The restatements we study are corrections of material errors in previously issued financial statements. Management typically announces a restatement in a press release or in an SEC filing following the discovery of a misstatement or irregularity by management, the SEC, the external auditor, or some combination thereof. Prior literature documents substantial negative returns to restatement announcements (see, for example, Myers et al., 2013; Palmrose et al., 2004). This negative market reaction may be related to changes in expectations of future earnings or growth rates, uncertainty about the credibility of financial statements, or concerns about the competence or integrity of management (Gleason et al., 2008; Hribar & Jenkins, 2004; Lev, Ryan, & Wu, 2007; Palmrose et al., 2004).

Prior studies find that negative market reactions increase with the severity of the restatement. For example, restatements that decrease previously reported net income, involve fraud, or affect core earnings (i.e., revenue, cost of sales, or ongoing operating expenses) are associated with a more negative market response (Palmrose et al., 2004; Scholz, 2008). Myers et al. (2013) also find that the market reacts more negatively to restatements made by companies with material weaknesses in internal controls, when restatements are associated with allegations of fraud, when restatements affect previously reported revenues or other components of core earnings, and by companies under investigation by the SEC. Restatement announcements are also associated with long-term negative market reactions for up to 6 months (Desai, Krishnamurthy, & Venkataraman, 2006; Efendi & Swanson, 2009), and severe restatements often precipitate shareholder litigation (Palmrose & Scholz, 2004), tightening debt covenants (Graham, Li, & Qiu, 2008), SEC Enforcement Actions (Files, 2011), management and director turnover (Desai, Hogan, & Wilkins, 2006; Srinivasan, 2005), and increases in the cost of capital and information risk (Hribar & Jenkins, 2004; Kravet & Shevlin, 2010). Furthermore, Fich and Shivdasani (2007) find that the negative valuation effects of financial misreporting extend to companies that share interlocking directors with companies sued for misreporting.

Recent research also finds that returns are affected by the characteristics of restatement disclosures (Files, Swanson, & Tse, 2009; Gordon, Henry, Peytcheva, & Sun, 2010; Myers et al., 2013). Controlling for restatement severity, these studies find that the market reaction to restatements disclosed prominently is more negative than the reaction to restatements disclosed obscurely,⁶ and Files et al. (2009) document that securities class-action lawsuits are more frequent when restatements are disclosed prominently in company-issued press releases. In summary, evidence from the restatement literature suggests that, on average, restatement announcements represent an important source of material bad news. As noted above, the negative consequences of this news include, but are not limited to, stock price declines, SEC Enforcement Actions, management and director turnover, and shareholder litigation.

Empirical Predictions

The literatures on short sales and on restatements suggest that restatement announcements should be an attractive target for short sellers. Therefore, we use three sets of empirical tests to examine short-seller behavior around restatement announcements. The first set of tests uses daily abnormal short selling and investigates whether, on average, short-seller investment behavior falls into one of three categories. First, short sellers may anticipate specific restatement announcement dates and thus may increase their short positions in the days leading up to these announcements to profit from an expected announcement period price decline. Significant positive abnormal short selling immediately before the announcement date would provide evidence consistent with this possibility. Second, short sellers may increase their short positions once the announcements are made, regardless of whether or not the announcement was anticipated, to profit from expected price declines over the subsequent trading days. Significant positive abnormal short selling during the announcement period would provide evidence consistent with this possibility.⁷ Third, short selling may not increase at the time of the announcement or immediately thereafter. While we posit that short sellers should respond to announcements when they expect subsequent negative returns, it is possible that, on average, short sellers do not expect future returns to be sufficiently negative to justify additional short sales.⁸ It is also possible that although short sellers anticipate specific restatement announcement dates, their existing short positions are sufficient given the uncertainty related to post-announcement returns. Alternatively, short sellers may refrain from making additional short sales immediately prior to the restatement announcements because this could be construed as a violation of insider trading laws.

The second set of tests investigates whether short sales around restatement announcements are contextual. That is, we test whether short sellers appear to anticipate or respond more actively to restatements announced more transparently and/or more severe restatements. The final set of tests investigates whether the future stock price performance of restating companies varies with announcement period abnormal short selling. If subsequent returns are more negative for restating companies with greater abnormal short selling at the time of the announcement, this suggests that short sellers respond to (or target more heavily) restatements announced by companies that subsequently underperform the market. This latter result would suggest that short sellers have a sophisticated understanding of the market implications of accounting restatements.

Data and Descriptive Statistics

Data

We obtain a sample of restatements made to correct financial statements that violated Generally Accepted Accounting Principles (GAAP) from the Audit Analytics (AA) database.⁹ We begin with the subsample of AA restatements for which dollar materiality effects are available because restatement materiality is an important consideration in our study. The AA database includes materiality data for restatements announced by companies trading on the major exchanges (i.e., the NYSE, AMEX, and NASDAQ).¹⁰

To proxy for restatement severity, we form the following variables using data in the AA database: NEGINCOME = an indicator variable that equals 1 for income-decreasing restatements (i.e., for those restatements where reported income was originally overstated), and 0 otherwise; ABSCHGNI = the absolute value of the restatement’s cumulative impact on income, scaled by sales revenue in the year prior to the restatement announcement; FRAUD = an indicator variable that equals 1 if the misstatement involves allegations of fraudulent activities, and 0 otherwise; PAST_RESTATE = an indicator variable that equals 1 if the restating company has already restated at least once in the AA database, and 0 otherwise.¹¹

In addition, the AA database provides the disclosure form used to announce the restatement (e.g., press release, Form 8-K, Form 10-K, etc.). This allows us to partition our sample into two transparency groups labeled the Press release or SEC Form 8-K (PR/8-K) subsample and the “Other Filings” subsample.¹²

We obtain daily short selling data directly from the NYSE, AMEX, and NASDAQ exchanges. The data span January 2, 2005, through August 6, 2007, and are part of an SEC mandate that self-regulated exchanges make tick data on short interest available under Regulation SHO.¹³ The primary focus of our analysis is on daily short selling activity surrounding the restatement announcement. Therefore, for each trading day, we compute daily short selling by aggregating across all transactions for a given company. Consistent with Christophe et al. (2004), we estimate daily abnormal short selling (ABSS) by adjusting each company’s daily short volume by its average level over a benchmark period as follows:

A B S S_{i, t} = [\frac{S S_{i, t}}{A V E S S_{i}}] - 1,

where SS_i,t = the total short volume for company i on Day t and AVESS_i = the average short volume for company i for a benchmark period consisting of the 3-month period that ends 10 days before the restatement announcement.

We emphasize that our short interest variable removes the normal level of short selling in a particular security. This adjustment is important because some short trades are initiated for reasons unrelated to news such as for liquidity or hedging purposes. Non-information-based short selling should occur evenly through a period and would not be expected to cluster around particular events. Thus, by removing the normal level of short selling, we are able to focus our analysis on variability in short selling that is likely to be driven by news.

Our analyses also include short interest levels leading up to the restatement announcement window. Data on the level of short interest (comprised of the number of shares sold short and settled as of the 15th day of each month) are available once per month from the COMPUSTAT Monthly Securities Database.¹⁴ We supplement these data with a proprietary data set obtained directly from the NYSE, AMEX, and NASDAQ exchanges.

We calculate the monthly short interest ratio by dividing the number of shares shorted on the 15th of each month by the number of shares outstanding (from the Center for Research in Security Prices [CRSP] database) on that date. To estimate whether the companies in our sample have relatively high short interest, we calculate abnormal short interest (AB_PRIOR_SI) by subtracting the average short interest ratio for all companies traded on the NYSE, AMEX, and NASDAQ from the short interest ratio calculated for each of our sample companies. Subtracting the average level of short interest removes the positive time trend in short interest (Dechow et al., 2001; Drake, Rees, & Swanson, 2011). In our regression analyses, we include AB_PRIOR_SI measured just prior to our 11-day announcement window.¹⁵ We require that monthly short interest data be available in the month prior to the restatement announcement and that daily short volume be available for at least 3 days: the announcement date, 1 day in the pre-announcement period, and 1 day in the post-announcement period.¹⁶

Prior research finds that short volume and trading volume are positively correlated (Christophe et al., 2004), so our models include abnormal volume, calculated using data from the CRSP daily stock files.

ABVO L_{i, t} = [\frac{VO L_{i, t}}{AVEVO L_{i}}] - 1,

where VOL = the daily trading volume as reported in CRSP for company i on Day t and AVEVOL = the average daily trading volume for company i the benchmark period, as defined previously.

We also estimate daily abnormal returns (ABRET) by subtracting the value-weighted market return from the raw stock return as reported in CRSP. When we accumulate returns over a particular trading horizon, we calculate the buy-and-hold raw return for the period and subtract the buy-and-hold value-weighted return for the same period. We obtain data from COMPUSTAT to calculate two control variables—the market value of equity (MVE = CSHO×PRCC_F) and the book-to-market ratio (BTM = CEQ/MVE), where CSHO is the common shares outstanding, PRCC_F is the closing stock price at the end of the fiscal year, and CEQ is the total common equity. Finally, we obtain data from Thomson Financial to calculate the institutional ownership ratio (INSTOWN), defined as total shares owned by institutions divided by total shares outstanding.

The intersection of the databases described above yields a potential sample of 1,327 restatements. We impose additional data requirements to remove confounds from the analyses.¹⁷ We remove 396 restatements where management announces earnings in the 11-day restatement announcement window. We impose this restriction because Christophe et al. (2004) find that short interest accumulates in advance of earnings announcements and we do not want to confound our tests with short positions taken, either in the pre-announcement period or after the announcement, based on information about earnings news. We also remove 181 restatements made pursuant to the SEC’s Staff Accounting Bulletin (SAB) 108 because these restatements correct errors that were previously deemed immaterial.¹⁸ Finally, we remove 10 restatements where the closing stock price falls below US$1 on any day in the 11-day announcement window to ensure that our results reflect economically meaningful amounts. Our final sample includes 740 restatement announcements made by 648 unique companies.¹⁹

Descriptive Statistics

We present descriptive statistics for our full sample on the restatement announcement date (Day 0) in Table 1, Panel A. We find that mean abnormal short selling (ABSS[0, +1]) is 118% greater than normal on the restatement announcement date and that mean abnormal trading volume (ABVOL[0, +1]) is 106% greater than normal. We find that the average short-window (Days [0, +1]) abnormal stock price reaction to the restatement announcement is −1.3%, which indicates that investors view restatements as bad news.

Table 1.

Descriptive Statistics.

Panel A: Full Sample Distributions.
Variable	M	SD	Lower quartile	Median	Upper quartile
ABSS[0, +1]	1.176	2.290	−0.176	0.453	1.467
ABVOL[0, +1]	1.060	1.988	−0.057	0.439	1.278
ABRET[0, +1]	−0.013	0.056	−0.031	−0.006	0.011
PRIOR_SI	0.049	0.057	0.010	0.030	0.068
AB_PRIOR_SI	0.015	0.057	−0.023	−0.004	0.032
BTM	0.500	0.333	0.269	0.450	0.678
MVE	2,963	7,062	193	588	1,818
INSTOWN	0.585	0.320	0.327	0.628	0.852
PAST_RESTATE	0.345	0.476	0	0	1
FRAUD	0.018	0.131	0	0	0
NEGINCOME	0.542	0.499	0	1	1
ABSCHGNI	0.046	0.132	0.000	0.005	0.023
n	740

Panel B: Comparisons of Means Across High Versus Low Transparency.
Variable	PR/8-K	Other filings	t stat
n	472	268
ABSS[0, +1]	1.341	0.885	2.61***
ABVOL[0, +1]	1.285	0.664	4.13***
ABRET[0, +1]	−0.020	−0.002	−4.29***
PRIOR_SI	0.051	0.046	1.03
AB_PRIOR_SI	0.017	0.012	1.17
BTM	0.490	0.517	−1.07
MVE	2,715	3,400	−1.27
INSTOWN	0.613	0.537	3.13***
PAST_RESTATE	0.360	0.317	1.18
FRAUD	0.021	0.011	0.99
NEGINCOME	0.627	0.392	6.33***
ABSCHGNI	0.051	0.036	1.45

Note. ABSS = abnormal short selling; ABVOL = abnormal trading volume; ABRET[0, +1] = the buy-and-hold raw return over trading Days 0 and +1 less the buy-and-hold value-weighted market return over the same period; AB_PRIOR_SI = the last short interest ratio measured prior to an 11-day announcement window minus the average short interest ratio for all companies traded on the NYSE, AMEX, and NASDAQ stock exchanges; BTM = book-to-market ratio; MVE = market value of equity; INSTOWN = institutional ownership ratio; PAST_RESTATE = an indicator variable that equals 1 if the restating company has already restated at least once in the AA database, and 0 otherwise; AA = Audit Analytics; FRAUD = an indicator variable that equals 1 if the misstatement involves allegations of fraudulent activities, and 0 otherwise; NEGINCOME = an indicator variable that equals 1 for income-decreasing restatements (i.e., for those restatements where reported income was originally overstated), and 0 otherwise; ABSCHGNI = the absolute value of the restatement’s cumulative impact on income, scaled by sales revenue in the year prior to the restatement announcement; t stat = t statistic; PR/8-K = Press release or SEC Form 8-K.

***

, **, and * indicate significance at 1%, 5%, and 10%, respectively.

Consistent with Desai, Krishnamurthy, and Venkataraman (2006), we find that short interest levels for restating companies are abnormally high relative to short interest levels for companies traded on the major exchanges in the month prior to the announcements. The mean abnormal short interest level (AB_PRIOR_SI) of 1.5% is significantly different from zero (t = 6.75). On average, sample companies have a book-to-market ratio (BTM) of 0.50, a market value of equity (MVE) of US$3.0 billion, and the majority of their shares are owned by institutions. With respect to restatement characteristics, we find that 1.8% of restatements involve fraud, 54.2% reduce net income, and the average effect (in absolute value) on net income is 4.6% of revenue. Finally, 34.5% of the restatements are preceded by at least one other restatement since 2000.

In Table 1, Panel B, we present descriptive statistics for subsamples of restatements disclosed with high transparency (PR/8-K) and low transparency (Other Filings), which represent 64% and 36% of the full sample, respectively. We find that abnormal short selling (ABSS[0, +1]) and abnormal trading volume (ABVOL[0, +1]) are significantly higher for high transparency restatement announcements. Consistent with Myers et al. (2013), we find that high transparency restatements are more severe than are low transparency restatements. Specifically, restatements announced in press releases or Form 8-K filings are more likely to reduce net income and have a more negative stock price reaction—they earn abnormal returns of −2.0%, compared with −0.2% for the low transparency group. This suggests that restatements announced in the more transparent manner should be an especially attractive target for short sellers.

Empirical Tests

Test of Daily Abnormal Short Selling Around Restatement Announcements

In this section, we examine short selling activity in the days around the restatement announcement. In Figure 1, we present the average daily abnormal short selling (ABSS) over an extended 181 trading day window (trading days [−90, +90]) for the full sample of restatements. We find that average ABSS is significantly positive across the full time period (the average t statistic is 8.69 and ranges from 6.46 to 12.72, untabulated), revealing that short volume is higher than is typical during this period. We also find that the magnitude of ABSS is stable in the days leading to the announcement, at about 60% higher than in the estimation period. The SEC requires that companies publicly disclose the need for a restatement within 4 days of identifying a misstatement, so these results suggest that short sellers are not privy to, or do not act on, inside information about imminent restatement announcements and that there is no event that systematically “triggers” short sales of the restating companies’ stock. Rather, it appears that short sellers maintain an already heightened interest in these companies because of their poor performance during the pre-announcement period,²⁰ or because they detect poor quality financial reporting and anticipate a stock price decline upon its revelation at an unknown date.

Figure 1.

Mean daily abnormal short trading around restatement announcements.

However, at the restatement announcement, we find a marked increase in abnormal short selling, which peaks on Day 0 at almost 120% higher than is usual, and remains significantly elevated on Day 1. The difference in abnormal short selling between Days [0, +1] and the 5 days immediately preceding the announcement is highly significant (t statistic = 4.98). This evidence, together with the stable level of abnormal short selling during the pre-announcement period, suggests that short sellers are poised and ready to respond quickly to restatement announcements made by these companies because these companies have been followed by short sellers for some time. In the post-announcement period, we find that the magnitude of ABSS drops back to approximately 65% higher than is typical, with volume comparable with the pre-announcement period magnitudes. This continued short-seller interest is consistent with restating companies performing poorly on average (beyond having financial reporting problems). Again, the difference between Days [0, +1] and the 4 days immediately after the restatement announcement is highly significant (t statistic = 3.77).

Next, we examine whether short selling activity varies with the transparency of restatement disclosures. In Figure 2, we plot ABSS for a 41 trading day window (trading days [-20, +20]).²¹Figure 2 indicates that most of the increased short-seller activity at the announcement date is related to restatements announced transparently, in press releases or on Form 8-Ks, rather than in less transparent filings. For the PR/8-K restatements, we observe relatively stable ABSS magnitudes in the pre-announcement period, followed by a dramatic spike on Days 0 and +1. On both days, ABSS is significantly greater for transparently disclosed restatements (t statistic = 2.11 and 2.26, respectively) than for restatements announced in other SEC filings but short selling is abnormally high for restatements announced in other SEC filings throughout the window. In both the pre- and post-announcement windows, ABSS is generally statistically similar for those companies making transparent versus non-transparent restatement disclosures. Prior research indicates that more severe restatements tend to be announced in press releases or in Form 8-K (Myers et al., 2013), so the relation between short selling and disclosure transparency may be due to either the relative severity of transparent restatements, the relative transparency of the disclosure venue, or both. Either way, we find that short sellers focus on the restatements most likely to generate more negative stock price reactions.

Figure 2.

Mean daily abnormal short trading around restatement announcements for the PR/8-K and other filings subsamples.

Regression Tests of Factors Associated With Short-Seller Responses to Restatement Announcements

In this subsection, we use regression analyses to investigate factors associated with short selling activity surrounding restatement announcements over three measurement periods. We define the pre-announcement period as Days [−5, −1]. This window represents the period investigated by Christophe et al. (2004) in their study of short selling activity preceding earnings announcements and includes the SEC’s required 4-day announcement deadline. We follow Palmrose et al. (2004) and define the announcement period as Days [0, +1]. Finally, we define the post-announcement period as Days [+2, +5].

Figures 1 and 2 show overall short volume increases dramatically during the announcement window and also reveal elevated short selling activity in advance of announcement. We begin by investigating whether short selling activity in the 5 days preceding restatement announcements is associated with specific company or restatement characteristics. In particular, if short sellers have information about specific restatements, we expect to see more short selling activity preceding severe restatements. Our multivariate regressions also allow us to control for key variables associated with abnormal short selling. Consistent with Christophe et al. (2004), we estimate the following model in the pre-announcement period:

A B S S [- 5, - 1] = β_{1} A B R E T [0, + 1] + β_{2} A B R E T - 5, 1 + β_{3} A B V O L - 5, 1 + β_{4} A B_P R I O R_S I + β_{5} B T M + β_{6} M V E + β_{7} I N S T O W N + β_{8} P A S T_R E S T A T E + β_{9} F R A U D + β_{10} N E G I N C O M E + β_{11} A B S C H G N I + β_{12} (N E G I N C O M E \times A B S C H G N I) + β_{YEAR} + e,

where all variables are defined previously.

In Model 3, ABRET[0, +1] proxies for the severity of the news in the restatement announcement. Thus, a negative and significant β₁ provides evidence that short sellers take greater short positions during the pre-announcement period in restatements which result in greater stock price declines, suggesting that short sellers are privy to inside information about impending restatement announcements and take positions accordingly. We also include additional variables that provide information about the severity of the restatement (i.e., FRAUD, NEGINCOME, ABSCHGNI, and NEGINCOME×ABSCHGNI).

With respect to control variables, we include ABRET[−5, −1] to represent contemporaneous stock price movements, which may influence the level of short selling, and we include ABVOL[−5, −1] to control for the contemporaneous correlation between overall trading volume and short volume. It is important to note that including these variables may introduce endogeneity because short volume is a component of total trading volume, and thus, short volume is likely to be jointly determined with returns and total volume.²² However, as documented by Christophe et al. (2004), these control variables are necessary to ensure that the model does not incorrectly attribute all pre-announcement short selling to the upcoming restatement announcement. Because not including these variables creates a known correlated omitted variables problem, we follow prior literature and include these variables with the caveat that our results may be influenced by this decision.²³ We also include AB_PRIOR_SI because abnormal short selling may be associated with the level of abnormal short interest prior to the announcement window and we include PAST_RESTATE to indicate prior reporting problems. We include BTM and MVE because prior studies find that short interest is associated with these information characteristics (see, for example, Dechow et al., 2001; Drake et al., 2011). We include INSTOWN to proxy for short selling constraints because low levels of institutional ownership can indicate a low supply of lendable shares (Saffi & Sigurdsson, 2011). Finally, we include year fixed effects (β_YEAR) to account for cross-sectional correlation and we cluster the standard errors by company to account for serial correlation (Petersen, 2009).²⁴

In Table 2, we present the estimation results from Model 3 for the full sample and for the two disclosure transparency partitions. We find that the coefficient on the announcement return, ABRET[0,+1], is not significantly different from zero in any regression, indicating short selling in the pre-announcement period is unrelated to announcement returns. However, results for ABSCHGNI suggest that there is less short selling activity prior to restatements that are arguably benign. In this interaction model, ABSCHGNI captures the association between short selling and the magnitude of non-negative corrections. Thus, the negative coefficient suggests less short selling activity as corrected income increases. There is also evidence of less short selling activity for companies where prior short positions were larger. However, there is more short selling activity for smaller companies, companies that have previously restated, and when returns preceding the announcement are more positive. Overall, these results are consistent with short sellers focusing on companies with weaker information environments, investing against market trends, and correctly interpreting earnings signals. However, there is no evidence that short sellers anticipate and trade immediately in advance of restatement announcements.

Table 2.

Regression of Daily Abnormal Short Trading in the Pre-Announcement Period.

\begin{matrix} Model 3 : ABSS [- 5, - 1] = β_{YEAR} + β_{1} ABRET [0, + 1] + β_{2} ABRET [- 5, - 1] + β_{3} ABVOL [- 5, - 1] \\ + β_{4} AB_PRIOR_SI + β_{5} BTM + β_{6} MVE + β_{7} INSTOWN + β_{8} PAST_RESTATE + β_{9} FRAUD \\ + β_{10} NEGINCOME + β_{11} ABSCHGNI + β_{12} (NEGINCOME \times ABSCHGNI) + e . \end{matrix}

Variable	Panel A: Full sample		Panel B: PR/8-K		Panel C: Other filings
	Coef	t stat	Coef	t stat	Coef	t stat
ABRET[0,+1]	0.086	0.11	−0.253	−0.29	1.238	0.65
ABRET[−5,−1]	1.992	2.51**	1.973	1.99**	1.993	1.62
ABVOL[−5,−1]	1.018	14.87***	0.998	12.06***	1.057	8.98***
AB_PRIOR_SI	−1.455	−2.95***	−0.499	−0.80	−2.918	−3.41***
BTM	−0.012	−0.09	−0.180	−1.12	0.219	0.88
MVE	−0.000	−3.24***	−0.000	−2.95***	−0.000	−1.44
INSTOWN	−0.150	−1.14	−0.187	−1.06	−0.093	−0.44
PAST_RESTATE	0.169	2.06**	0.142	1.36	0.214	1.52
FRAUD	−0.165	−1.16	−0.088	−0.49	−0.358	−1.76*
NEGINCOME	−0.050	−0.61	−0.129	−1.33	0.078	0.54
ABSCHGNI	−0.812	−2.75***	−1.359	−2.95***	−0.571	−1.01
NEGINCOME×ABSCHGNI	0.164	0.48	0.681	1.55	−0.109	−0.15
Adjusted R²	.551		.528		.592
n	740		472		268
Tests of sums of coefficients	Coef	F stat	Coef	F stat	Coef	F stat
β₁₀+β₁₂	0.114	0.13	0.552	1.86	−0.031	0.00

Note. We estimate daily abnormal short selling, ABSS, as [SS / AVESS] − 1, where SS = the total short volume for the day and AVESS = the average short volume for a benchmark period consisting of the 3-month period ending 10 days before the restatement announcement. ABRET[x, y] = the buy-and-hold raw return over trading days x through y less the buy-and-hold value-weighted market return over the same period. ABVOL = abnormal trading volume; AB_PRIOR_SI = the last short interest ratio measured prior to an 11-day announcement window minus the average short interest ratio for all companies traded on the NYSE, AMEX, and NASDAQ stock exchanges; BTM = book-to-market ratio; MVE = market value of equity; INSTOWN = institutional ownership ratio; PAST_RESTATE = an indicator variable that equals 1 if the restating company has already restated at least once in the AA database, and 0 otherwise; AA = Audit Analytics; FRAUD = an indicator variable that equals 1 if the misstatement involves allegations of fraudulent activities, and 0 otherwise; NEGINCOME = an indicator variable that equals 1 for income-decreasing restatements (i.e., for those restatements where reported income was originally overstated), and 0 otherwise; ABSCHGNI = the absolute value of the restatement’s cumulative impact on income, scaled by sales revenue in the year prior to the restatement announcement; PR/8-K = Press release or SEC Form 8-K; Coef = coefficient; t stat = t statistic; F stat = F statistic.

***

, **, and * indicate significance at 1%, 5%, and 10%, respectively.

The significance of these results differs between restatements announced transparently and non-transparently; most of these results are due to transparently disclosed restatements. Because the disclosure venue is not publicly known during this window, these differences likely reflect the greater materiality, on average, of restatements disclosed transparently.²⁵ In all partitions, short selling is positively associated with contemporaneous volume.

Next, we investigate factors associated with abnormal short selling in the 2-day announcement window using the following model:

\begin{matrix} ABSS [0, + 1] = θ_{1} ABRET [0, + 1] + θ_{2} ABVOL [0, + 1] + θ_{3} AB_PRIOR_SI \\ + θ_{4} BTM + θ_{5} MVE + θ_{6} INSTOWN + θ_{7} PAST_RESTATE \\ + θ_{8} FRAUD + θ_{9} NEGINCOME + θ_{10} ABSCHGNI \\ + θ_{11} (NEGINCOME \times ABSCHGNI) + θ_{YEAR} + v, \end{matrix}

where all variables are defined previously.

We present the estimation results for Model 4 in Table 3. In most respects, the results in the announcement window are similar to those in the pre-announcement period. In Panel A, for the full sample, there is more short selling activity for companies with prior restatements and smaller companies, and less short selling for restatements that increase or do not change previously reported income. However, unlike in the pre-announcement period, the coefficient on the interaction term (NEGINCOME×ABSCHGNI) and the sum of the coefficients on NEGINCOME and (NEGINCOME×ABSCHGNI) are both positive and significant.²⁶ These findings suggest that short sellers become more active in companies as adjustments to net income become more negative. Results also suggest more short selling activity when announcement returns are less negative. Together, these results suggest that short sellers look for opportunities where restatements are serious, but companies are likely to have weaker information environments, as proxied for by size and as reflected in less negative initial market reactions.

Table 3.

Regressions of Daily Abnormal Short Trading in the Announcement Period.

\begin{matrix} Model 4 : ABSS [0, + 1] = θ_{YEAR} + θ_{1} ABRET [0, + 1] + θ_{2} ABVOL [0, + 1] \\ + θ_{3} AB_PRIOR_SI + θ_{4} BTM + θ_{5} MVE + θ_{6} INSTOWN + θ_{7} PAST_RESTATE \\ + θ_{8} FRAUD + θ_{9} NEGINCOME + θ_{10} ABSCHGNI + θ_{11} (NEGINCOME \times ABSCHGNI) + v . \end{matrix}

Variable	Panel A: Full sample		Panel B: PR/8-K		Panel C: Other filings
	Coef	t stat	Coef	t stat	Coef	t stat
ABRET[0,+1]	3.191	1.69*	1.813	0.76	5.693	2.19**
ABVOL[0,+1]	0.899	14.61***	0.870	12.64***	0.971	6.76***
AB_PRIOR_SI	−1.010	−1.23	−0.868	−0.75	−1.686	−1.49
BTM	−0.007	−0.04	−0.202	−0.95	0.242	0.86
MVE	−0.000	−2.70***	−0.000	−2.51**	−0.000	−1.06
INSTOWN	−0.227	−1.12	−0.155	−0.51	−0.187	−0.80
PAST_RESTATE	0.259	2.25**	0.298	2.13**	0.195	1.03
FRAUD	−0.040	−0.11	−0.149	−0.33	0.192	0.52
NEGINCOME	0.088	0.77	−0.013	−0.09	0.281	1.32
ABSCHGNI	−2.044	−3.82***	−3.856	−4.49***	−1.293	−2.24**
NEGINCOME×ABSCHGNI	1.593	2.64***	3.010	3.44***	1.655	1.85*
Adjusted R²	.596		.612		.576
n	740		472		268
Tests of sums of coefficients	Coef	F stat	Coef	F stat	Coef	F stat
θ₉+θ₁₁	0.746	8.46***	0.755	12.53***	0.737	5.13**

Note. We estimate daily abnormal short selling, ABSS, as [SS/AVESS] − 1, where SS = the total short volume for the day and AVESS = the average short volume for a benchmark period consisting of the 3-month period ending 10 days before the restatement announcement; ABRET[x, y] = the buy-and-hold raw return over trading days x through y less the buy-and-hold value-weighted market return over the same period; ABVOL[−5,+1] = the abnormal volume over trading Days −5 through +1, where abnormal volume is calculated as [VOL/AVEVOL] −1, where VOL = the daily trading volume as reported in CRSP and AVEVOL = the average daily trading volume for a benchmark period consisting of the 3-month period ending 10 days before the restatement announcement; AB_PRIOR_SI = the last short interest ratio measured prior to an 11-day announcement window minus the average short interest ratio for all companies traded on the NYSE, AMEX, and NASDAQ stock exchanges; PAST_RESTATE = an indicator variable that equals 1 if the restating company has already restated at least once in the AA database, and 0 otherwise; AA = Audit Analytics; FRAUD = an indicator variable that equals 1 if the misstatement involves allegations of fraudulent activities, and 0 otherwise; NEGINCOME = an indicator variable that equals 1 for income-decreasing restatements (i.e., for those restatements where reported income was originally overstated), and 0 otherwise; ABSCHGNI = the absolute value of the restatement’s cumulative impact on income, scaled by sales revenue in the year prior to the restatement announcement; Coef = coefficient; t stat = t statistic; BTM = book-to-market ratio; MVE = market value of equity; INSTOWN = institutional ownership ratio; PR/8-K = Press release or SEC Form 8-K; F stat = F statistic.

***

, **, and * indicate significance at 1%, 5%, and 10%, respectively.

Results for the PR/8-K and Other Filings subsamples in Panels B and C reinforce this interpretation. The association with less negative announcement returns is attributable to restatements announced in Other Filings. Prior studies find that the initial price response is smaller when restatements are less transparent (see, for example, Files et al., 2009; Myers et al., 2013), so these results suggest that short sellers find opportunities in obscurely disclosed restatement announcements. However, the association with smaller companies is found mainly for restatements announced in Form 8-Ks, suggesting that when announcements are transparent, short sellers focus on companies likely to receive less attention. Overall, our results reinforce the notion that short sellers are sophisticated market participants who can identify obscure opportunities and can anticipate future returns.

Our third model examines factors associated with abnormal short selling in the post-announcement period:

\begin{matrix} ABSS [+ 2, + 5] = γ_{0} + γ_{1} ABRET [0, + 1] + γ_{2} ABRET [+ 2, + 5] \\ + γ_{3} ABVOL [+ 2, + 5] + γ_{4} AB_PRIOR_SI + γ_{5} BTM + γ_{6} MVE \\ + γ_{7} INSTOWN + γ_{8} PAST_RESTATE + γ_{9} FRAUD + γ_{10} NEGINCOME \\ + γ_{11} ABSCHGNI + γ_{12} (NEGINCOME \times ABSCHGNI) + γ_{YEAR} + ε, \end{matrix}

where all variables are defined previously.

This model differs from Model 4 by the inclusion of the post-announcement period abnormal return (ABRET[+2,+5]) and by the substitution of abnormal post-announcement volume, ABVOL[+2,+5], for abnormal announcement volume. Significant coefficients on our severity proxies in this model would suggest that even post-announcement abnormal short selling is linked to the information revealed about restatements announced at least two trading days prior.

In Table 4, we present the estimation results from Model 5. Consistent with prior windows, for the full sample, we find that the coefficient on ABSCHGNI is negative and significant, suggesting that the lower level of short selling activity for income-increasing restatements continues through the post-announcement period. We find further evidence of contrarian trading by short sellers as there is more short selling activity when returns in the post-announcement window are less negative. There also continues to be more short selling activity in the post-announcement period for companies that experienced less negative announcement returns. Short sellers continue to focus on companies likely to have weaker information environments (i.e., companies with less institutional ownership and, for restatements disclosed transparently, smaller companies).

Table 4.

Regressions of Daily Abnormal Short Trading in the Post-Announcement Period.

\begin{matrix} Model 5 : ABSS [+ 2, + 5] = γ_{0} + γ_{1} ABRET [0, + 1] + γ_{2} ABRET [+ 2, + 5] + γ_{3} ABVOL [+ 2, + 5] \\ + γ_{4} AB_PRIOR_SI + γ_{5} BTM + γ_{6} MVE + γ_{7} INSTOWN \\ + γ_{8} PAST_RESTATE + γ_{9} FRAUD + γ_{10} NEGINCOME \\ + γ_{11} ABSCHGNI + γ_{12} (NEGINCOME \times ABSCHGNI) + γ_{YEAR} + ε . \end{matrix}

Variable	Panel A: Full sample		Panel B: PR/8-K		Panel C: Other filings
Variable	Coef	t stat	Coef	t stat	Coef	t stat
ABRET[0,+1]	2.002	1.98**	1.397	1.09	1.732	0.96
ABRET[+2,+5]	2.294	2.58***	3.297	2.86***	−0.695	−0.58
ABVOL[+2,+5]	0.900	14.09***	0.808	10.36***	1.177	13.52***
AB_PRIOR_SI	−0.202	−0.22	−0.744	−0.95	0.211	0.12
BTM	0.120	0.96	0.174	1.07	0.065	0.29
MVE	−0.000	−1.61	−0.000	−1.94*	−0.000	−0.91
INSTOWN	−0.429	−2.69***	−0.350	−2.10**	−0.395	−1.48
PAST_RESTATE	0.026	0.32	0.073	0.75	−0.017	−0.12
FRAUD	0.204	1.12	0.129	0.96	0.528	0.77
NEGINCOME	−0.031	−0.35	0.049	0.44	−0.095	−0.69
ABSCHGNI	−0.866	−3.24***	−0.599	−1.17	−0.807	−1.83*
NEGINCOME×ABSCHGNI	0.304	0.90	−0.105	−0.18	0.612	1.03
Adjusted R²	.532		.517		.602
n	740		472		268
Tests of sums of coefficients	Coef	F stat	Coef	F stat	Coef	F stat
γ₁₀+γ₁₂	0.088	0.79	0.203	0.01	−0.032	0.87

Note. We estimate daily abnormal short selling, ABSS, as [SS/AVESS] − 1, where SS = the total short volume for the day and AVESS = the average short volume for a benchmark period consisting of the 3-month period ending 10 days before the restatement announcement; ABRET[x, y] = the buy-and-hold raw return over trading days x through y less the buy-and-hold value-weighted market return over the same period; ABVOL[−5,+1] = the abnormal volume over trading Days −5 through +1, where abnormal volume is calculated as [VOL/AVEVOL] − 1, where VOL = the daily trading volume as reported in CRSP and AVEVOL = the average daily trading volume for a benchmark period consisting of the 3-month period ending 10 days before the restatement announcement; AB_PRIOR_SI = the last short interest ratio measured prior to an 11-day announcement window minus the average short interest ratio for all companies traded on the NYSE, AMEX, and NASDAQ stock exchanges; PAST_RESTATE = an indicator variable that equals 1 if the restating company has already restated at least once in the AA database, and 0 otherwise; AA = Audit Analytics; FRAUD = an indicator variable that equals 1 if the misstatement involves allegations of fraudulent activities, and 0 otherwise; NEGINCOME = an indicator variable that equals 1 for income-decreasing restatements (i.e., for those restatements where reported income was originally overstated), and 0 otherwise; ABSCHGNI = the absolute value of the restatement’s cumulative impact on income, scaled by sales revenue in the year prior to the restatement announcement; Coef = coefficient; t stat = t statistic; BTM = book-to-market ratio; MVE = market value of equity; INSTOWN = institutional ownership ratio; PR/8-K = Press release or SEC Form 8-K; F stat = F statistic.

***

, **, and * indicate significance at 1%, 5%, and 10%, respectively.

In summary, we find no evidence that short sellers are privy to private information about restatement announcement dates: short selling activity remains consistent during the pre-announcement period and spikes significantly at the announcement day. Furthermore, pre-announcement short selling activity is not associated with announcement returns, suggesting that short sellers do not have specific information about restatement attributes or that they choose not to act on information which they may possess. However, consistent with prior studies, our evidence suggests that prior to the announcement, short sellers are attuned to overall earnings quality signals as they tend to avoid companies with understated income but are active in companies with past restatements. Furthermore, it appears that short sellers react to restated earnings information, selling when restatements have a more negative impact on earnings and avoiding companies that restate earnings upward.

Short selling is abnormally high throughout the three windows we investigate, and our evidence suggests short sellers provide important, contrarian information, particularly to segments of the market likely to have weaker information environments. In the pre-announcement window, short sellers are active in companies with more positive contemporaneous returns and in smaller companies. At the announcement, short sellers remain more active in companies with more positive announcement returns and in smaller companies. In the post-announcement period, short sellers continue to be more active when announcement returns or contemporaneous returns are more positive, as well as when companies have less institutional ownership.

Our results also suggest differences in short selling activity for restatements announced in Form 8-Ks or in press releases versus in Other Filings. Most of the significant associations are due to restatements announced transparently, which also tend to be more material restatements. However, short sellers’ focus on companies with less negative announcement returns appears to be due mainly to restatements announced more obscurely, suggesting that short sellers believe that the announcement reaction is incomplete when restatements are disclosed obscurely.

Alternative Measures of Short Selling Activity

In this section, we examine the sensitivity of our results to three alternative proxies for short selling activity around restatement announcements. The first two are related to short volume and the third is based on changes in the level of short interest. First, Christophe et al. (2004) estimate an additional short volume measure, labeled relative short selling (RELSS), calculated as,

RELSS = [\frac{SS}{VOL}],

where all variables are defined previously.

Thus, RELSS is the ratio of short volume to total trading volume. We use this measure and re-estimate Models 3, 4, and 5 using RELSS as the dependent variable. We exclude ABVOL from the models because total trading volume is already captured in the ratio, and we control for the normal level of RELSS by including the average RELSS over the benchmark period, as defined above. We find that the results using this alternative measure (untabulated) are consistent with those tabulated.

Second, our primary measure of short volume, ABSS, does not account for the possibility that some portion of short volume is driven by temporal changes in market-wide short selling due to factors like general market sentiment.²⁷ Thus, we investigate whether market-wide short volume is confounding our results by calculating the average daily ABSS ratio for all companies, which we label MktABSS. We then subtract MktABSS from ABSS to get a measure of abnormal short volume that adjusts for temporal changes in market-wide short volume. We find that the results using this alternative measure (untabulated) are also consistent with those tabulated.

Third, we construct a measure based on the change in the level of short interest in the stock. As discussed previously, an important innovation in this study is that we investigate short volume data, which is available daily; this allows us to observe the short seller trades in very short windows around restatement announcement dates. However, one drawback to using the short volume data is that it does not allow us to observe the level of short interest. Thus, in an additional test, we examine the change in the level of short interest around restatement announcements using the level of short interest as published by the stock exchanges near the 15th of each month. To increase the likelihood that the outstanding short positions are related to the restatement announcements, we focus on a subsample of restatement announcements that occur within 10 days of a subsequent short interest reporting date.²⁸ Using this subsample, we calculate the change in the short interest ratio (CHGSI) as the difference between the short interest ratio measured prior to the restatement announcement and the short interest ratio measured after the restatement announcement. As an alternative measure of the change in short interest, we calculate the change in abnormal short interest (CHGABSI), relative to the average short interest ratio for all companies traded on the NYSE, AMEX, and NASDAQ stock exchanges during the corresponding month. We use these measures as alternative dependent variables in the following model:

\begin{array}{l} C H G S I or C H G A B S I = ω_{0} + ω_{1} B T M + ω_{2} M V E + ω_{3} I N S T O W N \\ + ω_{4} P A S T_R E S T A T E + ω_{5} F R A U D + ω_{6} N E G I N C O M E + \\ ω_{7} A B S C H G N I + ω_{8} (N E G I N C O M E \times A B S C H G N I) + ω_{YEAR} + ε, \end{array}

where all variables are defined previously.

We first estimate Model 8 using CHGSI and find that the results (untabulated) are consistent with those presented above. More specifically, we find a negative and significant coefficient on ABSCHGNI (p < .02) and a positive and significant coefficient on NEGINCOME×ABSCHGNI (p < .10). We also find that the sum of the coefficients ABSCHGNI + NEGINCOME×ABSCHGNI is positive and significant (p < .10). We find similar results using the CHGABSI as the dependent variable. Taken together, these results provide further evidence that short activity is higher in restatement companies with more negative adjustments to net income.

SEC Comment Letters as a Potential Confound

The evidence from our empirical analyses suggests that short sellers respond to, rather than anticipate, restatement announcements, and that disclosure transparency and restatement characteristics have important effects on their responses. We also observe that short interest is abnormally high in the months leading up to the restatement announcements, which suggests that short sellers are trading on other evidence about questionable financial reporting quality. In this subsection, we consider the possibility that short sellers are using information about SEC comment letters to identify impending restatements. The SEC’s Divisions of Corporation Finance and Investment Management review the financial filings of public registrants and issue comment letters asking companies to provide additional information so that the SEC can “better understand the disclosure, revise disclosure in a document on file with the SEC, provide additional disclosure in a document on file with the SEC, or provide additional or different disclosure in a future filing with the SEC” (SEC, 2011). The SEC does not publicly disclose either the existence or content of comment letters until at least 20 days after the issues raised by a comment letter are resolved, so the SEC cannot be the source of public information about a comment letter induced restatement.²⁹

Companies are free, however, to disclose information about comment letters at any time in Form 8-K, and are required to disclose the existence of any outstanding comment letters in their annual 10-K filings. As a practical matter, comment letter disclosures prior to restatement announcements are unusual because most accounting issues are addressed during the year-end audit process. Thus, unresolved comment letter issues typically do not persist to the Form 10-K filing date. Moreover, if a disclosure indicates the comments will be resolved by a restatement, AA considers this as the restatement disclosure date, in which case, the comment letter and restatement disclosure dates are the same. Otherwise, the disclosure of a comment letter is another indication of questionable financial reporting, similar to poor accrual quality, and it is possible that short sellers use this information to take positions in companies that will ultimately restate.

We conduct a number of additional tests to investigate the extent to which comment letters might be influencing our inferences. To do this, we obtain comment letter data from AA, and we identify restatements in our sample that are associated with comment letters. From our sample of 740 companies, we find that 63 (or approximately 8.5%) make restatements associated with comment letters. We then conduct three sets of untabulated analyses. First, we test whether the mean pre-announcement ABSS is significantly greater for sample companies receiving comment letters than for sample companies not receiving comment letters and find that the difference is not statistically significant (t statistic = 1.16, p = .24). Second, we include a comment letter indicator variable in regression Models 3 through 5 to test for increased levels of short selling in the presence of comment letters. Here, we find that the coefficient on the comment letter indicator variable is insignificant in all models and that its inclusion does not change our inferences. Third, we re-estimate all of our tests after excluding restatements associated with comment letters and find that all of our reported results hold. Thus, the results of these tests increase our confidence that our findings are not confounded by comment letters.

Tests of Subsequent Abnormal Returns

In this section, we investigate whether short sellers are most active around restatement announcements associated with subsequent price declines. That is, we investigate whether there is evidence that short sellers earn returns on these positions. We group restating companies into three portfolios based on the magnitude of short trading on the day after the restatement announcement (i.e., we form terciles of ABSS on Day +1).³⁰ Each company remains in its assigned portfolio for trading horizons of 10, 20, 30, 40, and 50 trading days. For each trading day in our sample, we calculate the value-weighted return to each of the three portfolios, at each of the five trading horizons. We then estimate a four-factor model using the time-series of daily returns for each portfolio as follows (Day t subscripts omitted):

Re t_{p} - Re t_{rf} = α + δ_{1} (Re t_{mkt} - R_{rf}) + δ_{2} SMB + δ_{3} HML + δ_{4} UMD + e,

where Ret_p = the value-weighted portfolio return; Ret_rf = the risk-free rate; Ret_mkt = the value-weighted market return; SMB = return on a portfolio of small stocks less the return on a portfolio of big stocks; HML = return on a portfolio of high book-to-market stocks less the return on a portfolio of low book-to-market stocks; and UMD = return on a portfolio of stocks that were past winners less the return on the portfolio of stocks that were past losers.

The intercept, α, is the estimate of the average daily abnormal return for each portfolio over the time-series. If short sellers target those restating companies that subsequently underperform, we expect a negative α for the High Short Trading portfolio only.

In Table 5, we present the estimation results for Model 6 over five different holding periods. In Panel A, using a 10 trading day holding period, we find that all of the intercepts are insignificantly different from zero, so there is no evidence of abnormal returns. In Panel B, using a 20 trading day holding period, we find a significantly negative α for the High Short Trading portfolio only. The α of −0.0016 means that the portfolio yields returns of −16 basis points per day or about 3.2% over a 20-day post-restatement announcement holding period, on average. In Panel C, using a 30 trading day holding period, we find similar evidence of a significantly negative α. The estimate for α remains negative and significant in Panel D, using a 40 trading day holding period, but the magnitude decreases to −9 basis points per day, and becomes insignificant in Panel E, using a 50 trading day holding period.

Table 5.

Four-Factor Model Estimation Results for Portfolios of Restating Companies Partitioned by Short Trading.

Model 6 : Re t_{p} - R_{rf} = α + δ_{1} (Re t_{mkt} - R_{rf} + δ_{2} SMB + δ_{3} HML + δ_{4} UMD + e .

Panel A: 10 Trading Day Holding Period.
	α	Ret_mkt−R_rf	SMB	HML	UMD	No. of days in time-series	r ²
Low short trading	−0.0005	0.8531***	0.1357	0.2697	0.5367***	611	.208
	−0.85	8.23	0.81	1.01	3.44
Medium short trading	−0.0002	1.2515***	0.3153*	0.1259	−0.0953	608	.295
	−0.41	12.27	1.93	0.49	−0.63
High short trading	−0.0010	0.8755***	0.9979***	0.7754*	−0.1007	578	.124
	−1.04	5.67	3.94	1.94	−0.43
Panel B: 20 Trading Day Holding Period.
	α	Ret_mkt−R_rf	SMB	HML	UMD	No. of days in time-series	r ²
Low short trading	−0.0002	1.0534***	0.2363*	0.3223	−0.1396	649	.295
	−0.52	13.58	1.87	1.60	−1.17
Medium short trading	0.0000	1.0342***	0.2382*	−0.0902	0.0240	657	.297
	−0.01	12.92	1.83	−0.44	0.19
High short trading	−0.0016***	1.0527***	0.7913***	0.4679*	−0.1636	626	.294
	−2.87	11.36	5.20	1.93	−1.15
Panel C: 30 Trading Day Holding Period.
	α	Ret_mkt−R_rf	SMB	HML	UMD	No. days in time-series	r ²
Low short trading	−0.0002	1.0681***	0.1784	0.3348*	0.0144	666	.342
	−0.37	15.01	1.52	1.82	0.13
Medium short trading	−0.0005	1.1218***	0.0801	0.1869	−0.0835	667	.318
	−1.00	14.94	0.65	0.97	−0.73
High short trading	−0.0016***	1.0222***	0.6693***	0.4206*	−0.1688	652	.305
	−3.05	12.40	4.91	1.96	−1.34
Panel D: 40 Trading Day Holding Period.
	α	Ret_mkt−R_rf	SMB	HML	UMD	No. days in time-series	r ²
Low short trading	−0.0003	1.0585***	−0.0117	−0.2050	0.3148**	676	.298
	−0.51	12.97	−0.09	−0.99	2.58
Medium short trading	−0.0003	1.0805***	−0.0745	0.0336	−0.0276	677	.351
	−0.76	16.47	−0.69	0.20	−0.28
High short trading	−0.0009*	1.1451***	0.5458***	0.3325*	−0.1696	669	.387
	−1.91	15.75	4.57	1.77	−1.54
Panel B: 20 Trading Day Holding Period.
	α	Ret_mkt−R_rf	SMB	HML	UMD	No. of days in time-series	r ²
Low short trading	−0.0002	1.0534***	0.2363*	0.3223	−0.1396	649	.295
	−0.52	13.58	1.87	1.60	−1.17
Medium short trading	0.0000	1.0342***	0.2382*	−0.0902	0.0240	657	.297
	−0.01	12.92	1.83	−0.44	0.19
High short trading	−0.0016***	1.0527***	0.7913***	0.4679*	−0.1636	626	.294
	−2.87	11.36	5.20	1.93	−1.15
Panel C: 30 Trading Day Holding Period.
	α	Ret_mkt−R_rf	SMB	HML	UMD	No. days in time-series	r ²
Low short trading	−0.0002	1.0681***	0.1784	0.3348*	0.0144	666	.342
	−0.37	15.01	1.52	1.82	0.13
Medium short trading	−0.0005	1.1218***	0.0801	0.1869	−0.0835	667	.318
	−1.00	14.94	0.65	0.97	−0.73
High short trading	−0.0016***	1.0222***	0.6693***	0.4206*	−0.1688	652	.305
	−3.05	12.40	4.91	1.96	−1.34
Panel D: 40 Trading Day Holding Period.
	α	Ret_mkt−R_rf	SMB	HML	UMD	No. days in time-series	r ²
Low short trading	−0.0003	1.0585***	−0.0117	−0.2050	0.3148**	676	.298
	−0.51	12.97	−0.09	−0.99	2.58
Medium short trading	−0.0003	1.0805***	−0.0745	0.0336	−0.0276	677	.351
	−0.76	16.47	−0.69	0.20	−0.28
High short trading	−0.0009*	1.1451***	0.5458***	0.3325*	−0.1696	669	.387
	−1.91	15.75	4.57	1.77	−1.54
Panel C: 30 Trading Day Holding Period.
	α	Ret_mkt−R_rf	SMB	HML	UMD	No. days in time-series	r ²
Low short trading	−0.0002	1.0681***	0.1784	0.3348*	0.0144	666	.342
	−0.37	15.01	1.52	1.82	0.13
Medium short trading	−0.0005	1.1218***	0.0801	0.1869	−0.0835	667	.318
	−1.00	14.94	0.65	0.97	−0.73
High short trading	−0.0016***	1.0222***	0.6693***	0.4206*	−0.1688	652	.305
	−3.05	12.40	4.91	1.96	−1.34
Panel D: 40 Trading Day Holding Period.
	α	Ret_mkt−R_rf	SMB	HML	UMD	No. days in time-series	r ²
Low short trading	−0.0003	1.0585***	−0.0117	−0.2050	0.3148**	676	.298
	−0.51	12.97	−0.09	−0.99	2.58
Medium short trading	−0.0003	1.0805***	−0.0745	0.0336	−0.0276	677	.351
	−0.76	16.47	−0.69	0.20	−0.28
High short trading	−0.0009*	1.1451***	0.5458***	0.3325*	−0.1696	669	.387
	−1.91	15.75	4.57	1.77	−1.54
Panel D: 40 Trading Day Holding Period.
	α	Ret_mkt−R_rf	SMB	HML	UMD	No. days in time-series	r ²
Low short trading	−0.0003	1.0585***	−0.0117	−0.2050	0.3148**	676	.298
	−0.51	12.97	−0.09	−0.99	2.58
Medium short trading	−0.0003	1.0805***	−0.0745	0.0336	−0.0276	677	.351
	−0.76	16.47	−0.69	0.20	−0.28
High short trading	−0.0009*	1.1451***	0.5458***	0.3325*	−0.1696	669	.387
	−1.91	15.75	4.57	1.77	−1.54
Panel E: 50 Trading Day Holding Period.
	α	Ret_mkt−R_rf	SMB	HML	UMD	No. days in time-series	r ²
Low short trading	−0.0001	0.9743***	−0.0883	−0.2384	0.3925***	686	.300
	−0.16	12.93	−0.71	−1.25	3.54
Medium short trading	−0.0001	1.1505***	−0.0523	0.1471	−0.2005**	687	.387
	−0.35	18.60	−0.51	0.94	−2.20
High short trading	−0.0003	1.1408***	0.5137***	0.2364	−0.1979**	683	.444
	−0.88	18.12	4.94	1.48	−2.14

Note. We group restating companies into three portfolios, based on the magnitude of short trading on the day after the restatement announcement (i.e., we form terciles of ABSS on Day +1). Each company remains in its assigned portfolio for trading horizons of 10, 20, 30, 40, and 50 trading days, presented in panels A through E, respectively. For each trading day in our sample, we calculate the value-weighted return to each of the three portfolios, at each of the five trading horizons. We estimated Model 6 using the time-series of daily portfolio returns. Ret_p = the value-weighted portfolio return; Ret_mkt = the value-weighted market return; Ret_rf = the risk-free rate; SMB = return on a portfolio of small stocks less the return on a portfolio of big stocks; HML = return on a portfolio of high book-to-market stocks less the return on a portfolio of low book-to-market stocks; UMD = return on a portfolio of stocks that were past winners less the return on the portfolio of stocks that were past losers.

***

, **, and * indicate significance at 1%, 5%, and 10%, respectively.

It is important to note that the portfolio αs in Table 5 are gross returns that ignore transaction costs such as the cost of borrowing shares and trading costs to rebalance the portfolios. However, Diether and Werner (2011) estimate that lending contract fees are only about 1.64% (3.74%) annually for NYSE (NASDAQ) stocks, which is approximately 13.5 basis points (30.7 basis points) for a holding period of 30 days for NYSE (NASDAQ) stocks, and because the number of stocks in the portfolio at any given time is relatively low (the median is 16), the cost of rebalancing would be relatively low. In Table 5, Panel C, we estimate a portfolio α of −480 basis points (−16 basis points per day × 30 days) for a 30-day trading horizon, which seems more than sufficient to cover borrowing and trading stocks on either the NYSE or the NASDAQ. Thus, the evidence in Table 5 indicates that short positions taken in response to restatement announcements yield abnormal returns in excess of trading costs, and suggests that short sellers understand how the information in restatement announcements affects future stock prices and respond accordingly.

Conclusion

This article examines short selling activity in the days around accounting restatement announcements and investigates whether short sellers are more active in companies with the most severe misstatements before the misstatements are publicly disclosed. We find no evidence of increased short selling activity in the days immediately prior to restatement announcements, suggesting that short sellers are not privy to inside information about upcoming restatement announcements.

We do find significantly greater short volume in the 2-day window beginning with the restatement announcement, suggesting that short sellers respond to restatement announcements by increasing their short positions. Further analyses reveal that the increase in short volume is observed only for companies that announce their restatements transparently, in a press release or on a Form 8-K.

We investigate factors associated with short-seller responses to restatement announcements and find that short sellers are less active before, at, and after restatement announcements when restatements increase previously reported income. We also find significantly higher short selling activity for income-decreasing restatements and in the 2-day announcement window. Our evidence suggests that short sellers are more active when information environments are likely to be weaker as we find more short selling around restatements announced by smaller companies and when obscure restatement announcements elicit less negative returns.

Finally, we estimate the relation between announcement day abnormal short selling and subsequent abnormal returns, and we find that those restatements with the highest announcement day abnormal short selling (i.e., those where short sellers are most active) significantly underperform the market over the next 20 to 40 trading days.

The results of this study provide insights into how short sellers make their investment decisions. We find that short sellers target companies making relatively severe accounting restatements and that these stocks subsequently underperform the market. This evidence adds support to the notion that short sellers play an important role in the capital markets by taking positions, which help to keep stock prices in line with fundamental values. It also adds to emerging evidence that short sellers are sophisticated market participants who trade as if they understand the implications of accounting events for future returns.

Footnotes

Acknowledgements

We thank Kris Allee, Chris Anderson, Andy Call, Mark Hirschey, Bill Mayew, James Myers, K. Ramesh, Jake Thornock, David Wood, and workshop participants at Michigan State University for helpful comments and discussions.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Linda Myers gratefully acknowledges financial support from the Garrison/Wilson Chair at the University of Arkansas.

Notes

References

Ben-David

Drake

M. S.

Roulstone

D. T.

(2013). Acquirer valuation and acquisition decisions. Journal of Financial and Quantitative Analysis.

Boehmer

Jones

C. M.

Zhang

(2008). Which shorts are informed?The Journal of Finance, 63, 491-527.

Cassell

C. A.

Drake

M. S.

Rasmussen

S. J.

(2011). Short interest as a signal of audit risk. Contemporary Accounting Research, 28, 1278-1297.

Cassell

C. A.

Dreher

L. M.

Myers

L. A.

(2013). Reviewing the SEC’s review process: 10-K comment letters and the cost of remediation. The Accounting Review, 88, 1875-1908.

Christophe

S. E.

Ferri

M. G.

Angel

J. J.

(2004). Short-selling prior to earnings announcements. The Journal of Finance, 59, 1845-1875.

Daske

Richardson

Tuna

A. I.

(2006, May). Do short sale transactions precede bad news events? New evidence from NYSE daily data (Working paper). University of Pennsylvania, Philadelphia.

Dechow

Hutton

A. P.

Meulbroek

Sloan

R. G.

(2001). Short-sellers, fundamental analysis, and stock returns. Journal of Financial Economics, 61, 77-106.

Desai

Hogan

C. E.

Wilkins

M. S.

(2006). The reputational penalty for aggressive accounting: Earnings restatements and management turnover. The Accounting Review, 81, 83-112.

Desai

Krishnamurthy

Venkataraman

(2006). Do short sellers target firms with poor earnings quality? Evidence from earnings restatements. Review of Accounting Studies, 11, 71-90.

10.

Diamond

D. W.

Verrecchia

R. E.

(1987). Constraints on short-selling and asset price adjustment to private information. Journal of Financial Economics, 18, 277-311.

11.

Diether

Lee

K. H.

Werner

(2009). It’s SHO time! Short-sale price tests and market quality. The Journal of Finance, 64, 37-73.

12.

Diether

Werner

(2011, July). When constraints bind (Working paper). The Ohio State University, Columbus.

13.

Drake

Rees

Swanson

(2011). Should investors follow the prophets or the bears? Evidence on the use of public information by analysts and short sellers. The Accounting Review, 86, 101-130.

14.

Edwards

A. K.

Hanley

(2010). Short selling in initial public offerings. Journal of Financial Economics, 98, 21-39.

15.

Efendi

Swanson

E. P.

(2009, August). Short seller trading in companies with a severe accounting irregularity (Working paper). Texas A&M University, College Station.

16.

Engelberg

Reed

Ringgenberg

(2012). How are shorts informed? Short sellers, news, and information processing. Journal of Financial Economics, 105, 260-278.

17.

Ettredge

Scholz

Smith

K. R.

Sun

(2010). How do restatements begin? Evidence of earnings management preceding restated financial reports. Journal of Business Finance & Accounting, 37, 332-355.

18.

Fich

E. M.

Shivdasani

(2007). Financial fraud, director reputation, and shareholder wealth. Journal of Financial Economics, 86, 306-336.

19.

Files

(2011). SEC enforcement: Do forthright disclosure and cooperation really matter?Journal of Accounting & Economics, 53, 353-374.

20.

Files

Swanson

E. P.

Tse

(2009). Stealth disclosure of accounting restatements. The Accounting Review, 84, 1495-1520.

21.

Gleason

C. A.

Jenkins

N. T.

Johnson

W. B.

(2008). The contagion effects of accounting restatements. The Accounting Review, 83, 83-110.

22.

Gordon

E. A.

Henry

Peytcheva

Sun

(2013). Discretionary disclosure and market reaction to restatements. Review of Quantitative Finance and Accounting, 41, 75-110.

23.

Graham

J. R.

Qiu

(2008). Corporate misreporting and bank loan contracting. Journal of Financial Economics, 89, 44-61.

24.

Henry

T. R.

Koski

J. L.

(2010). Short selling around seasoned equity offerings. Review of Financial Studies, 23, 4389-4418.

25.

Hirshleifer

Teoh

(2011). Short arbitrage, return asymmetry, and the accrual anomaly. Review of Financial Studies, 24, 2429-2461.

26.

Hribar

Jenkins

(2004). The effect of accounting restatements on earnings revisions and the estimated cost of capital. Review of Accounting Studies, 9, 337-356.

27.

Hunton

J. E.

Rose

J. M.

(2008). Can directors’ self-interests influence accounting choices?Accounting, Organizations and Society, 33, 783-800.

28.

Karpoff

Lou

(2010). Short sellers and financial misconduct. The Journal of Finance, 65, 1879-1913.

29.

Kravet

Shevlin

(2010). Accounting restatements and information risk. Review of Accounting Studies, 15, 264-294.

30.

Lev

Ryan

S. G.

(2007). Rewriting earnings history. Review of Accounting Studies, 13, 419-452.

31.

Myers

L. A.

Scholz

Sharp

N. Y.

(2013, April). Restating under the radar? Determinants of restatement disclosure choices and the related market reactions (Working paper). University of Arkansas, Fayetteville.

32.

Palmrose

Z.-V.

Richardson

V. J.

Scholz

(2004). Determinants of market reactions to restatement announcements. Journal of Accounting & Economics, 37, 59-89.

33.

Palmrose

Z.-V.

Scholz

(2004). The circumstances and legal consequences of non-GAAP reporting: Evidence from restatements. Contemporary Accounting Research, 21, 139-178.

34.

Petersen

M. A.

(2009). Estimating standard errors in finance panel data sets: Comparing approaches. Review of Financial Studies, 22, 435-480.

35.

Reilly

(2006, September21). No more “stealth restating.”The Wall Street Journal. Retrieved from http://online.wsj.com/articles/SB115880258353369631

36.

Saffi

P. C.

Sigurdsson

(2011). Price efficiency and short selling. Review of Financial Studies, 24, 821-852.

37.

Scherbina

(2008). Suppressed negative information and future underperformance. Review of Finance, 12, 533-565.

38.

Scholz

(2008). The changing nature and consequences of public company financial restatements. Washington, DC: The Department of the Treasury.

39.

Securities and Exchange Commission. (2011). Comment letters. Retrieved from http://www.sec.gov/answers/commentletters.htm

40.

Srinivasan

(2005). Consequences of financial reporting failure for outside directors: Evidence from accounting restatements. Journal of Accounting Research, 43, 291-334.