Accounting Accruals and Short Selling

Transcription

1 Accounting Accruals and Short Selling Bilal Erturk Assistant Professor of Finance Spears School of Business Oklahoma State University Giorgio Gotti Assistant Professor of Accounting College of Business University of Texas at El Paso Tony Kang Associate Professor of Accounting Spears School of Business Oklahoma State University Ramesh Rao Professor of Finance Spears School of Business Oklahoma State University The paper has benefited from comments received at the 2013 AAA conference and presentations at the following workshops: Virginia Tech, University of North Texas, King Fahd University of Petroleum and Minerals, American University of Sharjah, IBS-Hyderabad, Chulalongkorn University, and University of Hong Kong.

2 Accounting Accruals and Short Selling Abstract We find that the positive association between short interest and total accruals is attributable to discretionary accruals. We do not find that short interest is positively related to non-discretionary accruals. Further, our results are stronger for firms that meet or narrowly beat earnings threshold using discretionary accruals. For highly shorted stocks, we document a significant abnormal return differential between high and low accrual stocks when ranked by total and discretionary accruals but not non-discretionary accruals. We provide large scale cross-sectional evidence that short sellers play a useful role in uncovering opportunistic earnings management. Key words: Accounting accruals, Discretionary Accruals, Short Selling JEL classification: M40, M41, O16

3 Accounting Accruals and Short Selling 1. Introduction The main objective of this study is to evaluate whether short sellers effectively take trading positions based on different components of accruals that are known to have different implications for future earnings and stock returns. Prior studies that examine the association between financial reporting attributes and short sellers behaviors provide mixed evidence (e.g., Richardson 2003; Desai, Krishnamurthy, and Venkataraman 2006; Karpoff and Lou 2010; Hirshleifer et al. 2011). In a recent study, Hirshleifer, Teoh and Yu (2011) find a positive association between short selling and total accruals in a cross-section of firms. They also document that the accrual anomaly (Sloan 1996) is exacerbated when constraints on short arbitrage are more severe. 1 However, these studies do not examine whether short sellers differentiate between different components of accounting accruals, even though they are shown to have different implications for future earnings and stock returns (e.g., DeFond and Park 2001; Xie 2001). Since not all portions of accruals are equally opportunistic, this is an important distinction that enables us to tell whether short sellers really play a role in detecting financial misreporting, i.e., opportunistic earnings management. To the extent that short sellers are sophisticated investors that understand the implications of different components of earnings for future firm performance, they are most likely to take short positions for the component of earnings that will experience the greatest reversal. Our study fills this void in this literature. A key innovation of this study is that we examine the associations between short interest and discretionary and nondiscretionary components of accruals separately. The discretionary component of accruals is likely to reflect opportunistically managed portion of earnings, whereas 1 Similar to Hirshleifer et al. (2011), Karpoff and Lou (2010) suggest that short sellers anticipate the eventual discovery and severity of financial misconduct, helping to uncover misconduct and to keep share prices closer to fundamental values. 1

4 the non-discretionary component is expected to capture manager s assessment of future cash flows from current operations. Nevertheless, prior studies that examine the relation between accounting accruals and short selling solely focus on total accruals. To the extent short sellers are a sophisticated group of investors who can process and benefit from financial statement information, one would expect them to take more short positions in stocks with greater discretionary accruals that are shown to be less persistent and reverse to a greater extent in the future periods than non-discretionary accruals (e.g., DeFond and Park 2001; Xie 2001). Accrual accounting is at the heart of earnings measurements and financial reporting (Barth, Beaver, Hand and Landsman 2005). As stated in SFAC No.1 (FASB, 1978), accounting accruals are intended to convey useful signals about future cash flows that are not captured in current period cash flows. In estimating accruals, managers use their judgment and exercise discretion. To the extent that managers exercise best accounting judgment and use the reporting discretion, accounting accruals will signal future cash flows (e.g., Dechow et al. 1998; Barth, Beaver, Lang and Landsman 1999; Barth, Cram and Nelson 2001). In this case, current accruals will relate positively to future stock returns. However, prior research suggests that managers sometimes use their reporting discretion in an opportunistic manner (e.g., Dechow and Sloan 1991; Sweeney 1994; Becker et al. 1998; Rangan 1998; Teoh et al. 1998a, 1998; Guidry et al. 1999), leading to a negative association between current accruals and future cash flows and stock returns (e.g., Sloan 1996; DeFond and Park 2001; Xie 2001; Cheng and Thomas 2006). Those studies show that managers often use their discretion in reporting accruals to either opportunistically inflate to meet certain earnings 2

5 thresholds or deflate to smooth out earnings over time. 2 In line with this idea, Sloan (1996) finds that accruals reverse in the future and that current period accruals are negatively associated with future stock returns. Xie (2001) shows that this association is primarily driven by the discretionary component of accruals. His results show that the market overprices discretionary accruals because investors overestimate the persistence of these accruals. DeFond and Park (2001) also suggest that the market overprices discretionary accruals because investors underanticipate the future reversal of those accruals. While critics argue that short sellers undermine investors confidence in financial markets 3, advocates claim that short selling facilitates market efficiency and the price discovery process (Boehmer and Wu 2012), as investors who identify overpriced firms can sell short, thereby incorporating their unfavorable information into market prices (Karpoff and Lou 2010). If short sellers effectively trade on information contained in current period accruals, short interest, i.e., the amount of short position taken in a stock, will be positively associated with the discretionary component of accruals, which has shown to be negatively associated with future stock returns. In contrast, short interest is likely to be higher in firms with low non-discretionary accruals, to the extent that non-discretionary accruals signal future cash flows (e.g., DeFond and Park 2001; Xie 2001). In order to measure the incremental effect of accruals on short interest, we calibrate short interest by controlling for other known determinants of short interest such as firm size, book-tomarket, momentum, and institutional ownership. Our evidence shows a positive association 2 For example, while the extant literature suggests that most of discretionary accruals are reversed in the future, Tucker and Zarowin (2006) suggest that not all earnings discretions are necessarily opportunistic. Their evidence suggests that income smoothing can improve earnings informativeness. 3 For instance, some argue that short sellers can spread false rumors about a firm in which (s)he has a short position and profit from the resulting decline in the stock price (Karpoff and Lou 2010). The former SEC Chairman Christopher Cox called such behavior distort and short (The WSJ, July 24, 2008). 3

6 between short interest and discretionary accruals but no such relationship is observed for nondiscretionary accruals. In many cases, we observe a negative relation between short interest and nondiscretionary accruals. We further observe that short interest ratio is greater for firms with the highest discretionary accruals relative to those with the lowest discretionary accruals. We also find that short interest ratio is actually lower for firms with the highest non-discretionary accruals compared to those with the lowest non-discretionary accruals. This is consistent with the idea that ungarbled accruals signal future profitability (e.g., Barth et al. 2001). Moreover, examining changes in short interest reveals that short sellers increase their positions after observing high discretionary accruals are reported. However, we don t find such a relationship for non-discretionary accruals. These results suggest that short sellers discern the implications of discretionary and nondiscretionary components of accruals for future returns and take trading positions accordingly. This result suggests that Hirshleifer et al. s (2011) finding is largely driven by the discretionary component of accruals. To bolster this claim, further analysis shows that the positive association between short interest and total accruals (Hirshleifer et al. 2011) is observed only when the proportion of discretionary accruals in total accruals is high. This also suggests that our finding is not simply driven by the mechanical relation between total and discretionary accruals. We further find that the positive relationship between short interest and accruals is stronger for firms using accruals (both total and discretionary) to meet or narrowly beat the earnings threshold. This provides corroborating evidence to our earlier crosssectional test results, as these firms are more likely to have engaged in opportunistic earnings management to avoid reporting negative earnings surprises (Matsumoto 2002). These results are robust to controlling for institutional monitoring and short selling constraints proxied by institutional ownership. 4

7 Our final set of results examines the effect of accruals and their components on short seller profitability. For purposes of this test, we focus on stocks with the greatest short interest following quarterly earnings announcements. Our portfolio analysis shows a significant differential in the buy-and-hold abnormal returns between extreme portfolios classified by total and discretionary accruals but not by non-discretionary accruals. Short sellers experience abnormal return spreads of 4.5 (3.8) percent between high and low total (discretionary) accrual portfolios, while the spread is insignificant in the case of non-discretionary accruals. This suggests that trading by short sellers on information contained in discretionary accruals yields significant incremental returns but that short seller trading profitability is likely unrelated to nondiscretionary accruals. Our study makes several important contributions to the literature. First, this study contributes to the long-standing debate over whether short sellers improve market efficiency or encourage harmful manipulations of corporate performance (Jones and Lamont 2002). Our finding extends prior studies that document short sellers profit from financial misconducts (e.g., Karpoff and Lou 2010, Desai, Krishnamurthy, and Venkataraman 2006) by providing more direct, large-scale evidence that short sellers effectively take trading positions based on opportunistic accrual-based earnings management actions that may not ultimately be subject to legal enforcement actions. 4 In contrast to the prior studies that examine a set of firms that were clearly overpriced ex-post, i.e., firms that were subject to SEC enforcement for financial misrepresentation (Karpoff and Lou 2010) and restated earnings (Desai et al. 2006), we focus on 4 Desai et al. (2006) use the GAO restatement data, but the reliability of these data is often questioned. For example, Hennes, Leone, and Miller (2008) report that 76% of the restatement data in the GAO database are simple errors rather than misrepresentation or fraud. 5

8 firms that are overpriced ex-ante. 5 Our approach enables us to provide large scale evidence on short sellers role in market price discovery their ability to identify overpriced stocks in a setting that is less obvious and likely more difficult to characterize as financial misconduct. In this regard, we provide more powerful evidence over short-sellers ability to capitalize on overpriced stocks through financial statement analysis. This study also contributes to prior literature on investors capital allocation decisions based on accounting accruals (e.g., Sloan 1996; DeFond and Park 2001; Xie 2001; Desai et al. 2006; Hirshleifer et al. 2011). Our results show that the positive relation between total accruals and short interest (e.g., Hirshleifer et al. 2011) is primarily driven by the discretionary component of accruals, i.e., short sellers target firms with large discretionary accruals, which are known to experience greater reversals in the future periods, but not the ones with high nondiscretionary accruals. There is also weak evidence that non-discretionary accruals are associated negatively with short interest, consistent with the signaling role of accruals for future cash flows. Supporting the idea that short sellers represent a sophisticated group of investors, our evidence suggests that they are able to see through the different implications of the two components of accounting accruals, and correct capital misallocations from opportunistic financial reporting. The remainder of this article is organized as follows. In the next section, we review relevant literature and develop empirically testable hypotheses. We describe the sample and research design in Section 3. We discuss the results in Section 4. We conclude in Section Related Literature and Hypotheses 5 We examine a large sample of panel data than the ones based on samples of firms that were subject to SEC enforcement action or have available GAO restatement data. For example, Karpoff and Lou s (2010) sample contains 454 firms during , only a small fraction of publicly traded firms. 6

9 Prior literature finds that short sellers are able to predict short-horizon abnormal returns (Diether et al., 2009). There are a few possible reasons for how they are able to achieve this result. The first explanation suggests that market frictions (among many, Diamond and Verrecchia, 1987) and investor behaviors (Daniel et al., 1998; Hong and Stein, 1999) allow the market price in the short horizon to deviate from the underlying fundamental value, and that short sellers as more sophisticated investors trade in these situations. Indeed there is evidence (Boehmer et al. 2008) that 75% of short sales are executed by institutional investors, despite the fact that many institutions (for instance many mutual funds) are not allowed to short stocks. An alternative explanation on why short sellers are able to predict short-horizon abnormal returns is that they provide liquidity to the market, when there is a significant short term imbalance between buy and sell orders. When the buying pressure reverts, prices also revert to the fundamental value and the short sellers cover their positions at a profit. In this case short sellers are compensated for providing immediacy (Grossman and Miller, 1988; Campbell et al. 1993). Finally, another possible explanation considers short sellers to step in and absorb part of the market risk in periods of high volatility, and for that they are compensated with a trading profit. Short selling requires an investor to borrow shares from another invesor who is willing to lend. The lender generally requires cash collateral, equal to 102% of the market value of the borrowed shares, from the borrower. As Hirshleifer et al. (2011) note, Federal Reserve Regulation T requires short sellers to post an additional 50% in margin when the lender is a U.S. broker-dealer. The lender pays the short seller interest, i.e., the rebate rate, on the collateral. 6 6 As Hirshleifer et al. (2011) point out, the spread between the rebate rate and the market interest rate on cash funds (the loan fee) constitutes a direct cost to the short seller. In addition, the lenders have the right to call a loan at their disposal. 7

10 While there is a rather extensive literature that shows how short sellers are informed traders (e.g., Diether et al., 2009), some studies suggest otherwise (Henry and Koski 2010). For instance there is evidence that short sellers target their trading on companies characterized by market over-pricing compared with their fundamental ratios (Dechow et al., 2001), earnings restatements and high accruals (Efendi et al., 2005; Desai et al. 2006), analysts forecast revisions and earnings surprises (Francis et al. 2006, Christophe et al. 2004), analyst recommendation changes (Christophe et al., 2010), financial misconduct as represented in SEC enforcement actions (Karpoff and Lou, 2010), and that they are trading to exploit the earnings announcement drift and the accrual anomaly (Cao et al., 2006). In contrast, Henry and Koski (2010) find no evidence of informed short selling around SEO announcements. The Financial Accounting Standards Board (FASB) states in Objectives of Financial Reporting by Business Enterprises (1978) that the primary objective of accounting data is to provide information to help present and potential investors, creditors, and others assess the amounts, timing, and uncertainty of prospective net cash inflows to the related enterprise (para 37). Prior research generally shows that future cash flow prediction is enhanced by considering accounting accruals data in addition to current period cash flow information. Finger (1994) finds that used alone or together with current period cash flows, contemporaneous earnings are a significant predictor of future cash flow. Lorek and Willinger (1996) examine the time-series properties and predictive ability of cash-flow data, and conclude that accounting accruals have predictive ability for future cash flows incremental to current cash flows. Dechow, Kothari and Watts (1998) show that accrual earnings better predict future operating cash flows than current operating cash flows, confirming the idea that earnings convey information about future cash flows not contained in current period cash flows. Barth et al. 8

11 (1999) find that accounting accruals are incrementally informative for future cash flows over and beyond current accruals. Barth et al. (2001) find that accruals and its components predict future cash flows after controlling for current cash flows, consistent with the idea that accounting accruals convey useful signals for future cash flows. Specifically, they find that accrual components such as change in accounts payable/receivable, change in inventory, and depreciation and amortization, are useful in predicting future cash flows, incremental to current cash flow. While a primary role of accounting accruals is enhancing the prediction of future cash flows, research evidence shows that managers at times use their discretion to report earnings in an opportunistic manner (e.g., Dechow and Sloan 1991; Becker et al. 1998; Rangan 1998; Teoh et al. 1998a, 1998b; Healy and Wahlen 1999). The evidence suggests that managers use accruals to opportunistically increase or decrease earnings in various contexts, e.g., prior to initial public offerings and seasoned equity offerings (Rangan 1998; Teoh et al. 1998a, 1998b), when the firm s compensation contract is tied to the earnings (Dechow and Sloan 1991; Guidry et al. 1999; Healy 1985; Holthausen et al. 1995), and when the firm is likely to violate debt covenants (Sweeney 1994). Thus, a more realistic view of accounting accruals is that it reflects both managers candid assessment of future cash flows as well as opportunistically biased accrual judgments driven by various conflicts of interest they face. If accounting accruals predict future cash flows, they will also relate to future stock returns which reflects expectation of future cash flows. Interestingly, however, Sloan (1996) finds that total accruals are negatively associated with future stock returns because investors tend to over-estimate the persistence of accruals and over-price them as a result. Xie (2001) shows that Sloan s (1996) findings are primarily driven by the discretionary component of accruals, 9

12 which is likely to capture the opportunistically managed portion of accruals. He does not find that non-discretionary portion of accruals are negatively associated with future stock returns. These findings suggest that different components of accruals have different implications for future stock returns. To the extent that short sellers are a group of sophisticated investors who are able to accurately process financial statement information, they are likely to distinguish different components of accounting accruals that have different implications for future cash flows. Specifically, they are more likely to sell short shares of a firm with inflated earnings due to reporting opportunism. Accordingly, we formulate our first hypothesis as follows: H1: Short interest are greater for firms with greater discretionary accruals. To rule out alternative explanations, we also examine the sensitivity of discretionary accrual -short selling relationship to the underlying incentives for short artbitrage. It is possible that the profitability of short selling arbitrage is greater in cases of stronger external monitoring and where short selling constraints are lower. Using institutional ownership as a proxy for external monitoring and short selling constraints, we predict that the positive relationship between short selling and discretionary accruals will be increasing in insititutional ownership. Next, we examine whether short sellers are more cautious about accruals when the firm narrowly meets or beats the analyst forecasts. Matsumoto (2002) reports that managers tend to manage earnings to avoid negative earnings surprises. Her results suggest that managers have greater incentives to opportunistically manage earnings when they are likely to miss the earnings threshold. This implies that when discretionary accruals are used to meet or narrowly beat the analyst expectation, the discretionary accruals are more likely to be opportunistic in nature. In 10

13 this regard, we examine whether the positive relationship between accruals and short interest is stronger for suspect firms, i.e., firms that just meet or narrowly beat the analyst expectation using discretionary accruals, than non-suspect firms. To the extent that short sellers are able to see through the manager s intention, short interest is likely to be higher for suspect firms. This leads us to predict the following. H2: The relationship between accruals and short interest is stronger for suspect firms (firms that would not have met the analyst expectation without discretionary accruals). For our final hypothesis we examine the question of what role accruals and their components play in short seller profitability. Specifically, focusing on stocks with the highest short interest following earnings announcements, we evaluate differential abnormal returns available to short sellers from trading the two extreme accrual portfolios. If our primary hypothesis is valid, we should find significant return spreads from trading on the extreme discretionary accrual stocks and less significant or insignificant spread returns from trading on non-discretionary accruals. A higher return from shorting high discretionary accrual stocks relative to low discretionary accrual stocks would imply that the information uncovered from high discretionary accruals is economically meaningful and enhances short sellers profits. On the other hand, a finding that short seller returns are not different between low and high nondiscretionary accrual portfolios would imply short seller trades vis-à-vis these stocks are unrelated to any information that may be contained in non-discretionary accruals. More formally this hypothesis may be stated as: 11

14 H3: Short sellers will experience significantly higher (less significant or insignificant) returns by trading on high discretionary (non-discretionary) accrual stocks relative to low discretionary (non-discretionary) accrual stocks. 3. Data and Research Design The sample consists of stocks listed on NYSE, Amex, and NASDAQ with data between January 1988 and December We gather data from several different sources. Quarterly accounting information is obtained from Compustat XFeed US. Monthly short interest is obtained from the individual exchanges. Short interest is defined as the open short position as of settlement on the 15 th of each month (or the preceding day if it is not a business day) as a percentage of total shares outstanding. This information represents short sale trades that occurred three or five business days prior as the settlement period changed from five to three days on June 7, After establishing the short trade date, we relate each firm-quarter accrual observation to the first available short interest information that comes at least one week after the earnings announcement date. We impose this minimum one week lag between earnings announcements and short sale trade dates so that short sellers have at least one week to execute their trades. Market information including stock prices, shares outstanding, and returns are taken from CRSP. Our measurement interval is quarterly, so all variables are expressed in this frequency. Appendix 1 lists the variables used in the study and their definitions. Each firmquarter observation is required to have sufficient data to calculate the models in the paper. To estimate discretionary accruals we adopt a performance matched Jones model, as presented first by Kothari et al. (2005) and used, among many, by Tucker and Zarowin (2006): 1 & (1) 12

15 Total accruals, change in sales, and gross property plant and equipment are deflated by the previous quarter total assets. We also control for return on assets as Kothari et al. (2005) found that the traditional Jones model is misspecified for firms with extremely good/bad performances. We estimate the regression in model (1) and measure non-discretionary accruals as the fitted values from the regression, and the discretionary accruals as the residuals from the regression. [Insert Table I about here] Table I presents sample descriptive statistics. The average short interest in our sample is 2.4 percent while the 90 th percentile is about 5.8 percent demonstrating a significant difference between highly and lightly shorted stocks. The difference between the 10 th and 90 th percentile of firms based on total accruals is almost 10 percent of total assets while the average total accruals in our sample is only percent. We observe that total accruals account for about 5.1 percent of total assets for firms that are aggressive in using accruals (the 90 th percentile). On the other hand, non-discretionary accruals are not as much dispersed with only percent difference between the 10 th and 90 th percentile of firms. We use two measures of short interest: raw short interest (SI) and abnormal short interest (ABSI). We define ABSI it for firm i in quarter t as the difference between SI and and expected SI for the quarter: ABSI it = SI it E(SI it ) (2) We calculate expected short interest, E(SI) in a manner similar to the one employed in Karpoff and Lou (2010). Using prior evidence as a basis, Karpoff and Lou suggest that E(SI) is a function of firm size, book-to-market (BM), momentum, and institutional ownership. Following their approach, each quarter, we classify all stocks into tertiles independently by size, book-tomarket, momentum, and institutional ownership yielding a total of 81 portfolios. Each of the 81 13

16 portfolios is further classified into industry groups based on two-digit SIC codes. Size is defined as the natural logarithm of the number of shares outstanding multiplied by the price per share. Book-to-market ratio is calculated by dividing book value of equity by market value of equity. Momentum is the mean stock return performance over the previous 4 quarters. Institutional ownership (IO) is the number of shares held by institutions as reported by the 13-F filings divided by total number of shares outstanding. We then estimate the following cross-sectional regression on a quarterly basis: 1 (3) Where SI it is the short interest for firm i in quarter t and the independent variables are dummy variables for Size, BM, Momentum, and IO. The dummy variables equal one if the variable is in the medium or high tertiles, and zero otherwise. Industry dummy variable, Ind it,is equal to 1 if firm i belongs to industry k in quarter t. The fitted value for SI it in the above equation serves as our estimate for E(SI it ) for any given stock and quarter depending upon where the stock falls with respect to the 81 portfolios classified by size, book-to-market, momentum, institutional ownership, and industry group. [Insert Table II about here] Table II presents the time-series averages of the estimates of quarterly cross-sectional regressions of equation (3) using the Fama-MacBeth procedure. The t-statistics are corrected for serial correlation using three lags per Newey-West (1987). We note that all of the dummy variable coefficients are statistically significant. We observe that short interest for small size firms is signficantly less than for the medium and large size firms. In the case of book-to-market 14

17 and momentum, short interest for the lowest tertile stocks is greater than for stocks in the medium and high tertiles though the pattern is not monotonic between the medium and high tertiles. Finally, short interest is monotinically increasing in institutional ownership. Similar patterns are observed by Karpoff and Lou (2010). 4. Results We first present univariate results followed by the multivariate results. The univariate results are shown in Table III. For the univariate results we examine short interest levels and its changes for sample firms classified into tertiles by accruals in each quarter. The table presents time-series averages of short interest and its changes for each tertile and consists of three panels, one for each measure of accruals: total, discretionary, and non-discretionary accruals. We examine short interest (SI), abnormal short interest (ABSI), and their changes ( SI, ABSI). The table also shows the difference in the short interest variables between the low and high accruals portfolios and their significance. For total accruals, we observe that, with the exception of ABSI, all other measures of short interest increase as we go from the low to medium to high accrual portfolios. In the case of ABSI, short interest for the high accrual portfolio exceeds the short interest for both the low and medium accrual portfolios with the medium portfolio showing values below that of the low accrual portfolio. Across all measures of short interest, we find that short interest for the high acrrual firm portfolio is always significantly greater than short interest for the low accrual firm portfolio. [Insert Table III about here] The next two panels disaggregate the results into accrual components. We generally observe a monotonic increase in short selling as we go from the low to medium to high discretionary accrual portfolios. Similar to the case of total accruals, in every instance the 15

18 difference between the high and low accrual portfolios is positive and statistically significant at conventional levels. In the case of non-discretionary accruals it is clear that short interest is not increasing in accruals. If anything, there may be an inverse relationship. Overall, the univariate results indicate that the previously documented positive relationship between short selling and accruals is attributable to discretionary accruals and not to non-discretionary accruals. As noted earlier, accounting accruals can convey useful signals about future cash flows but managers may also use their discretion to opportunistically manage earnings using accruals. Such opportunistic behavior manifests itself in the discretionary portion of accruals. Consistent with this view, our univarirate results indicate that short sellers focus on firms with high discretionary accruals, which are subject to opportunistic earnings management, rather than simply firms with high total accruals. [Insert Table IV about here] Next, we discuss the multivariate results. Table IV presents multivariate results using ABSI as the dependent variable. The independent variables consist of total accruals or its components.... (4) We present results for the full sample as well as subsamples classified by institutional ownership. We examine differential sensitivity to institutional ownership because prior evidence suggests that the intensity of short selling is a function of short selling constraints. Hirshleifer et al. (2011) and others find that short selling is greater where the availability of loanable shares is greater. These studies find that institutional ownership is a good instrument to proxy for availability and liquidity of shares. In the context of our study, we expect that the short selling 16

19 accrual relationship is going to be stronger for the subset of firms with greater institutional ownership. Table IV Panel A presents pooled regression results using actual values of firm level variables as the independent variables and using fractile ranks of short interest, i.e., decile rankings independently constructed each quarter for the independent variables, and then standardized to take values ranging between zero and one. We use the decile rankings to ensure that our results are not affected by a few extreme observations and to be able to compare incremental contribution of each variable to short selling. The results for the two approaches are qualitatively similar therefore our discussion will mainly focus on using actual values for the independent variables. The first three columns present regression estimates separately for each of the three accrual measures (Total, Discretionary, and Non-Discretionary accruals). Column four presents estimates with the discretionary and non-discretionary accruals considered simultaneously. The last two columns present regression estimates for firms classified into low and high subsets by IO. The low and high subsets are identified relative to the median value. Across all columns, consistent with the univariate results, we observe that ABSI is positively related to total and discretionary accruals and negatively related to non-discretionary accruals. The D.Accruals coefficient tends to be larger in the high IO group, consistent with the idea of the availability of loanable shares constraining short sellers arbitrage behaviors. The difference in the D.Accruals coefficient between the high and low IO groups has a t-statistic of 2.36 (twotailed). Table IV Panel B replicates Table IV Panel A analysis but using Fama-MacBeth regressions. The results are qualitatively similar. In particular, we find that stocks in the top decile based on discretionary accruals have about 0.20% more short interest ratio relative to 17

20 stocks in the bottom decile depending on different regression specifications. Considering the unconditional average short interest ratio of 1.2% in our sample, this result is economically significant since the mean short interest in the highest discretionary accrual decile would be about 18% higher than the mean short interest of the lowest discretionary accrual decile. On the other hand, short interest is about 0.60% lower for stocks in the top decile non-discretionary accrual portfolio compared to those in the bottom decile. 7 [Insert Table V and VI about here] Table IV provides convincing evidence that short sellers target the dicretionary portion of total accruals to identify potential opportunistic earnings management effects. To further ensure the relevance of discretionary accruals as the driving element in the accruals-short seller relationship, we conduct additional tests using subsamples of firms with certain discretionary accrual characteristics that are likely to be of interest to short sellers. Tables V and VI present results replicating the analysis included in Table IV limiting the sample to firms with positive (income increasing) accruals (Table V) and to firms in the highest accrual decile (Table VI). Since positive accruals firms are on average more likely to experience negative returns in the future (e.g., Sloan 1996; Xie 2001), short sellers are likely to target these firms. We also separately examine the highest decile accrual firms as short sellers may particularly target these firms because of their greater susceptabiliy to opporuntistic behavior (Hirshleifer, Teoh and Yu (2011)). Overall across all columns in both tables the results are, again, consistent with the 7 In untabulated results we test for the sensitivity of our findings to NASDAQ vs. NYSE and AMEX listed firms. Hirshleifer et al. (2011) find that short-selling of accounting accruals is more pronounced in NASDAQ firms than in NYSE firms. They argue that NASDAQ firms are generally harder to value as they tend to be smaller, informationally more opaque, and more growth oriented. When actual values of accruals are used, we observe similar results in both samples, indicating that the documented association between short interest and discretionary accruals are not limited to the NASDAQ sample. However, consistent with Hirshleifer et al. (2011), NASDAQ firms demonstrate a stronger association between short interest and accruals when decile rankings are used. This result stems from the fact that NASDAQ firms display twice as large cross-sectional variation in accruals relative to NYSE firms. Therefore, our study shows that short sellers do not seem to discriminate between accrual values of NASDAQ and NYSE firms. 18

21 univariate results and regression results shown in table IV: we observe that ABSI is positively related to total and discretionary accruals and negatively related to non-discretionary accruals. The results reported in Table V and VI are consistent with this idea, and the coefficient values on T.Accruals and D.Accruals are greater than the ones reported in the previous tables and the t- statistics are more significant. For example, in Table V we find that stocks in the top decile based on discretionary accruals have about 0.50% more short interest ratio relative to stocks in the bottom decile, which is more than the twice the differential observed in Table IV. Next, we classify our sample by proportion of discretionary accruals to total accruals. If short sellers are uncovering opportunistic earnings management evident from discretionary accruals, we would expect stronger results for firms with greater proportions of discretionary accruals to total accruals. Table VII presents the results. We document that short selling is significantly positively related to discretionary accruals only for the subset of firms with the highest third in proportion of discretionary accruals to total accruals. The relationship is insignificant in the other two or of opposite sign. [Insert Table VII about here] The regression results presented to this point are based on level variables. It is generally acknowledged that the more powerful results are obtained when variables are measured in their change form. Table VIII is structured similarly to Table IV with the exception that all the variables are in their change form. The dependent variable is the change in ABSI over two quarters ( ABSI) and the independent variables are measured as changes over two quarters preceding the measurement period for short interest. [Insert Table VIII about here] 19

22 The results validate the conclusions from the level regressions and show that abnormal short interest increases following an increase in total accruals and discretionary accruals. One departure between the levels and change regressions is that non-discretionary accruals are no longer significant; earlier the coefficient was significantly negative. Table VIII also presents results using changes as standardized decile ranked variables for the independent variables. The results are similar to what we observed using pooled regressions. Our results here suggest that the relationship between short selling and accruals manifests itself more generally across the broad sample of firms. Hirshleifer et al. (2011) also use decile rankings but focus only firms whose total accrual changes from any given decile in the base year to the tenth decile in the following year. For this subset of firms they find a positive relationship between total accrual changes and short interest. Their study does not provide a clear explanation as to why they examine this subset of firms. It is not clear why short sellers would not be interested in a firm that, for example, goes from decile 1 to decile 8 compared to one going from decile 3 to decile 10. Indeed our results indicate that the positive short selling - accruals relationship likely holds across the spectrum of changes in accruals regardless of the particular decile. Our results in Table VIII indicate that the positive short seller accrual relationship is evidenced across the broad sample of inter-decile changes in accruals. In the spirit of Hirshleifer et al. (2011) in Table IX we present results for the association between abnormal short interest and accruals for firms that changed to and from the highest accrual deciles. In this specification of the basic model, the independent variables take the value of -1 if a firm ranks among the highest decile in quarter q-1 but not in quarter t, +1 if a firm does not rank among the highest decile in quarter q-1 but becomes so in quarter q, and 0 otherwise. The qualitative results are similar to those discussed earlier in Table VIII where we consider 20

23 changes across all deciles. This provides additional support that the findings reported initially by Hirshleifer et al. (2011) hold more generally across the cross section firms and not just confined to changes to and from the highest accrual decile. [Insert Table IX about here] In Table X we present results for our second hypothesis on whether firms that just meet or beat analysts EPS forecast exhibit a stronger association between abnormal short interest and accruals. We compute JustMBF following extant literature (Koh et al. 2008) as a dummy variable equal to one if the firm s actual EPS just meets or exceeds the last analysts forecast at least 3 days before the quarterly earnings announcement by a cent per share or less. We exclude firms in regulated industries because they likely have different incentives than those in nonregulated industries: financial institutions (SIC codes ), utilities (SIC codes ), and other quasi-regulated industries (SIC codes ) (Matsumoto 2002). We winsorized at +/-1% the difference between the earnings per share forecast error, computed as the difference between actual EPS minus last analysts EPS forecast at least 3 days before the quarterly earnings announcement. The model we adopt for the analysis follow Fan et al. (2010): (5) We test for the hypothesis that Accruals coefficient is higher for JustMBF firms than for NonJustMBF testing whether 3 4. For each quarter t, short interest is regressed on firm characteristics obtained at the end of quarter t-1. JustMBF (NonJustMBF) is an indicator variable that equals 1 (0) if a firm-quarter observation has actual EPS meets or exceeds the last analysts 21

24 forecast at least 3 days before the quarterly earnings announcement by a cent per share or less, and 0 (1) otherwise. [Insert Table X about here] Results from Table X show that firms that just meet or beat the EPS forecasts have a more strong association between abnormal short interest and accruals. However the results are different across columns: while for both total and discretionary accruals (column 1 and 2) the coefficients for both sub-samples (firms that just meet-beat EPS forecast vs. firms that don t) are positive and significant, with a statistically higher coefficient for the JustMBF sub-sample, for non-discretionary accruals (column 3) we have a non significant coefficient for the JustMBF sub-sample, and a negative and significant coefficient for the sub-sample of firms that do not just meet-beat the EPS forecast. Overall the results from Table X show a strong positive association between abnormal short selling and discretionary accruals, stronger for firms that just meet or beat by 1 cent or less the last analysts forecast at least 3 days before quarterly earnings announcement. Our third hypothesis focuses on the role accruals and their components play in impacting short seller profitability. For this test we first rank all the stocks independently based on their accruals and short interest. For the subset of stocks with the greatest short interest, we examine the one-year abnormal return differential between the highest and lowest accrual portfolios. 8 If short interest is driven by the potential for uncovering information related to discretionary accruals (and not non-discretionary accruals) and that such information yields differential returns, abnormal return spreads for high vs.low accrual portfolios should be signficant for 8 The two subsets were generated by independently ranking the firms on short interest and on accruals. In order to make sure short sellers had sufficient time to access and process information, we use the first available short interest information that comes at least one week after the earnings announcement date. The abnormal returns are annual buy-and-hold returns starting one month after the short date following the quarterly earnings announcement relative to the benchmark portfolio of firms in the same industry/quarter. 22

25 stocks ranked by discretionary accruals and less significant or insignificant for stocks ranked on non-discretionary accruals. We report these results in Table XI. [Insert Table XI about here] We start by examining the abnormal return spread between the two extreme accrual portfolios without regard to short seller holdings. We do this to confirm previous findings that accruals contain information about future returns and that the strength of this relationship varies for the two components of accruals. This is reported in Panel A. We then examine the return spreads for the subset of firms that have the highest short interest. These are presented in Panel B. From Panel A we note that the return differental between low and high total accrual firms in the cross-section of firms is approximately 5.5%. This result is in line with what prior studies report, suggesting that low total accrual firms earn substantially higher returns than high total accrual firms. For discretionary accruals the 5% return spread between the two extreme portfolios is in line with what we observed for total accruals. In contrast, consistent with what prior studies report, the non-discretionary sorting results in a much smaller differential of 1.4%. Overall, the return patterns in Panel A confirm evidence from prior research that current period accruals are negatively associated with future stock returns and that this is mostly due to discretionary accruals (Xie 2001). In Panel B we report the returns for the sub-sample of high short-selling firms (top quintile short selling firms). As can be seen, the spread in abnormal returns between the high and low accrual portfolios is approximately 4.5% for total accruals and 3.8% for discretionary accruals, both significant at conventional levels. Interestingly, the return differential is not significantly different from zero in the case of non-discretionary accruals. Note, however, that the returns for the low and high non-discretionary acccruals are sizeable at approximately 4%. 23

26 The insignificant difference suggests that for these stocks short seller motivation is unrelated to any information that may be contained in the non-discretionary component of accruals. Our results imply that the signficant differential in returns between high and low accrual portfolios with high short interest is related to information contained in high discretionary accruals and not non-discretionary accruals and that short sellers can reap signficant abnormal returns by trading on this information. These results are consistent with the notion that short sellers engage in informed trading in anticipation of future reversal of accruals, especially discretionary accruals. 5. Conclusions The role of short sellers in uncovering meaningful information about firms is a controversial one. We focus on the relation between short selling behavior and the quality of accruals inherent in firms reported earnings. Prior literature is ambiguous in this regard. Richardson (2003) does not find that short sellers trade on the basis of information contained in accounting accruals. Hirshleifer et al. (2011) on the other hand find that short selling and accruals are positively related though this relation primarily exists only in the top decile of accrual firms. The focus of these studies was total accruals. However, the literature on accruals suggests that accruals have both informative and opportunistic elements. If short sellers indeed are sophisticated investors that serve to improve the price discovery process, we should expect them to trade actively in stocks associated with greater opportunistic earnings management. By disaggregating accruals into discretionary and non discretionary components we are able to priovide a more complete picture about the role short sellers play in uncovering opportunistic earnings management. 24