Short Selling and Earnings Management: A Controlled Experiment *

Transcription

1 Short Selling and Earnings Management: A Controlled Experiment * Vivian W. Fang a University of Minnesota Allen Huang b Hong Kong University of Science and Technology Jonathan Karpoff c University of Washington This draft: June 27, 2013 Abstract We exploit a natural experiment (Regulation SHO s Pilot Program) to examine the effect of short selling on firms reporting behavior. In this experiment an arbitrary subsample of the Russell 3000 index was designated as pilot stocks and exempted from short-sale price tests. Before the program, both pilot and non-pilot firms exhibit similar levels of discretionary accruals. Pilot firms discretionary accruals reduce significantly during the program, and then revert to preexperiment levels when the program ends. This pattern occurs primarily among firms with high institutional ownership and high CEO wealth-performance sensitivity. Within the subsample of firms that engage in financial misconduct and lead to SEC enforcement actions, the conditional probability of detection is significantly higher for pilot firms during the pilot program. Further, pilot firms earnings response coefficients decline and their coefficients of current returns on future earnings increase during the program, and revert to pre-program levels after the experiment. Together, these results indicate that decreases in short selling costs curb firms opportunistic reporting behavior and enhance the informativeness of their share prices. JEL classifications: G14; G18; G19; M41; M48 Keywords: Regulation SHO; Pilot Program; Short Selling; Earnings Management; Price Informativeness * We thank Russell Investments for providing us the list of Russell 3000 index used in this paper. a fangw@umn.edu, Tel: , Carlson School of Management, University of Minnesota, Minneapolis, MN 55455, USA. b allen.huang@ust.hk, The Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong c karpoff@uw.edu, Foster School of Business, University of Washington, Seattle, WA 98195, USA.

2 1. Introduction Short Selling and Earnings Management: A Controlled Experiment Previous research shows that short sellers can identify earnings manipulation and fraud before they are publicly revealed. 1 But this is for earnings manipulations that have already taken place. Might short selling also constrain firms incentives to manipulate or misrepresent earnings in the first place? That is, does short selling, or its prospect, help to improve the quality of firms financial reporting in general? In this paper we exploit a natural experiment that allows us to address this question. In July 2004, the Securities and Exchange Commission (SEC) adopted a new regulation governing short selling activities in the U.S. equity markets Regulation SHO. Regulation SHO contained a Rule 202T pilot program in which every third stock ranked by trading volume within each exchange was drawn from the Russell 3000 index and designated as a pilot stock. From May 2, 2005 to August 6, 2007, pilot stocks were exempted from short-sale price tests, thus decreasing the cost of short selling these stocks. The price tests, which include the tick test for exchangelisted stocks and the bid test for Nasdaq National Market Stocks, were maintained for non-pilot stocks. 2 The pilot program creates an ideal setting to examine the effect of short-sale constraints on corporate financial reporting decisions, for three reasons. First, the pilot program and its impact on short selling costs represent an exogenous shock to the cost of selling short in the 1 See Dechow, Sloan and Sweeney (1996), Desai, Ramesh, Thiagarajan, and Balachandran (2002), Griffin (2003), Christophe, Ferri, and Angel (2004), Efendi, Kinney and Swanson (2005), Desai, Krishnamurphy, and Venkataraman (2006), and Karpoff and Lou (2010). 2 The pilot program was originally scheduled to commence on January 3, 2005 and end on December 31, 2005 (Securities Exchange Act Release No , July 28, 2004). However, the SEC postponed the commencement date to May 2, 2005 (Securities and Exchange Act Release No , November 29, 2004) and extended the end date to August 6, 2007 (Securities and Exchange Act Release No , April 20, 2006). 1

3 affected firms. We can identify no evidence that the firms themselves lobbied for the pilot program, or that any individual firm could know it would be in the pilot group until the program was announced. Second, the elimination of price tests initiated a meaningful decrease in the cost of selling short among the pilot stocks (e.g., see the SEC s Office of Economic Analysis, 2007). The pilot program eliminates the need to estimate short selling costs a notoriously difficult task (see Lamont, 2004) since it defines a treatment group (the pilot stocks) in which short selling costs were decreased relative to the control group (the non-pilot stocks). This setting is superior to a blanket ban on price tests, such as the one imposed in U.S. markets in Third, the pilot program had specific beginning and ending dates, facilitating a difference-in-differences (hereafter, DiD) analysis of the impact of short selling costs on firms financial reporting. The known ending date allows us to investigate whether the effects of the pilot program reversed when it ended an important check on the internal validity of the difference-in-differences tests (e.g., see Roberts and Whited, 2011). We begin by verifying that pilot firms represent a random draw from the Russell 3000 population. In the fiscal year before the pilot program, the pilot and non-pilot firms are similar in size, growth, corporate spending, profitability, leverage, and dividend payout, a finding also reported by Grullon, Michenaud, and Weston (2011). More importantly, although the two groups of firms exhibit similar levels of discretionary accruals pre-program, pilot firms significantly reduce their discretionary accruals once the program starts. After the program ends, pilot firms discretionary accruals revert to pre-program levels. The non-pilot firms, meanwhile, show no significant change in their discretionary accruals during or after the pilot program. We further test the robustness of this finding by controlling for firms R&D expenditures, capital expenditures, and equity issuance, as Grullon, Michenaud, and Weston (2011) document 2

4 a significant reduction in pilot firms investment and equity issuance compared to non-pilot firms during the pilot program. Our results are barely affected, implying that the effect of the pilot program on discretionary accruals is unlikely to be explained by the changes in the difference between two groups of firms investment or equity issuance surrounding the program. We also find that the impact of the pilot program on earnings management is most pronounced among firms with high institutional ownership and high sensitivity of the CEO s wealth to the company s share price. Institutional ownership is associated with the availability of shares that makes short selling feasible (e.g., see Nagel, 2005). So these results indicate that a reduction in short selling costs is particularly effective at deterring earnings management when short selling is most feasible and when managers incentives to manipulate earnings are relatively high. We then examine the effect of the pilot program on the discovery of financial misconduct that eventually leads to enforcement action by the SEC. A priori, a reduction in short sale constraints has competing effects on the discovery of financial misconduct. On one hand, the increased threat of short selling may deter aggressive earnings manipulations and decrease the incidence of financial misconduct, an effect that is consistent with our previous findings. On the other hand, an increase in short selling activity may increase the likelihood of discovery for any misrepresentation that does occur. Consistent with the view that both effects deterrence and an increased rate of discovery are at work, we find that the pilot program has an insignificant effect on the unconditional likelihood that a firm will be revealed to have engaged in financial misrepresentation. Among firms that do misrepresent their earnings, however, the conditional probability of detection is significantly higher for pilot firms during the pilot program. 3

5 Finally, we examine the implications of the pilot program for price efficiency and market quality through its effect on firms reporting practices. We show that the pilot firms earnings response coefficients decrease during the pilot program, while their coefficients of current returns on future earnings increase. These two results indicate that the reduction in pilot firms earnings management during the pilot program corresponds to an increase in the informativeness of their share prices. These findings make four contributions to the literature. First, they show that short selling, or its prospect, has real effects on firms financial reporting. This demonstrates one avenue through which trading in secondary financial markets affects firms decisions. 3 Second, our results highlight one important avenue through which short selling makes prices more informative. Previous research emphasizes how short selling facilitates the flow of private information into prices (e.g., Miller, 1977; Harrison and Kreps, 1978). Our findings suggest that short selling also improves price informativeness by decreasing managers use of discretionary accruals. Third, our findings identify a new determinant of earnings management short-sale constraints in addition to the factors identified in previous research (see Dechow, Ge, and Schrand, 2010). And fourth, these results contribute to the policy debate over the benefits and costs of short selling. Previous research demonstrates that short sellers frequently are good at identifying overpriced shares, and that their trading conveys external benefits to other investors by improving market quality and by accelerating the discovery of financial misconduct (Karpoff 3 See Bond, Edmans, and Goldstein (2012) for a survey of research on the real effects of financial markets. For example, Fang, Noe, and Tice (2009) and Edmans, Fang, Zur (2013) examine the effect of stock liquidity on firm performance and governance; Kang and Kim (2013) examine the effect of liquidity on CEO turnover; Fang, Tian, and Tice (2013) examine the effect of liquidity on innovation, Grullon, Michenaud, and Weston (2011) examine the effect of short selling on equity issuance and investment; and Massa, Zhang, and Zhang (2013b) examine the effect of short selling on governance. A concurrent paper by Massa, Zhang, and Zhang (2013a) also examines the effect of short selling on earnings management using data on the costs of short selling in international markets, and reports conclusions that are similar to ours. 4

6 and Lou, 2010). 4 Our results indicate that the prospect of short selling decreases earnings management and increases price informativeness in general, even among firms that are not charged with financial reporting violations. This indicates that short selling, or its prospect, conveys external benefits to investors by improving the quality of financial reporting and the informativeness of share prices. This paper is organized as follows. Section 2 describes short sale price tests in the U.S. equity markets and how they can affect firms use of discretionary accruals. Section 3 describes the data. Sections 4 and 5 report tests of the effect of the Regulation SHO s pilot program on firms use of discretionary accruals and on the incidence and discovery of firms financial misconduct, and Section 6 reports on tests that examine whether the pilot program coincided with an increase in the informativeness of share prices. Section 7 concludes. 2. The effects of short sale price tests on earnings management 2.1. Short-sale price tests in U.S. equity markets Short sale price tests were initially introduced to the U.S. equity markets in 1930s, ostensibly to avoid bear raids by short sellers in declining markets. The NYSE adopted an uptick rule in 1935, which was replaced in 1938 by a stricter SEC rule, Rule 10a-1, also known as the tick test. The rule mandates that a short-sale can only occur at a price above the most recently traded price (plus tick) or at the most recently traded price if that price exceeds the last different price (zero-plus tick). 5 In 1994 the NASD also adopted its own price test ( bid test ) under Rule 4 See the references in footnote 1, and also the SEC s OEA (2007), Alexander and Peterson (2008), and Diether, Lee, and Werner (2009). 5 Narrow exceptions apply, as specified in SEC s Rule 10a-1, section (e). 5

7 Rule 3350 requires a short-sale to occur at a price one penny above the bid price if the bid is a downtick from the previous bid. 7 Most evidence indicates that the price tests work to constrain short selling. For example, Angel (1997) and Alexander and Peterson (1999) show that Rule 10a-1 impedes short selling by delaying order execution and lowering fill rates, even in rising markets. McCormick and Reilly (1996) find that Rule 3350 curtails short selling for Nasdaq NMS stocks during declining markets. Diether, Lee, and Werner (2009) show that pilot stocks listed on both the NYSE and Nasdaq experienced a significant increase in short selling activity during the pilot program. These stocks had more frequent short-sale trades and a greater short sales-to-share volume ratio during the term of the pilot program. Moreover, NYSE-listed pilot stocks experienced a higher level of order-splitting (i.e., short-sale trades with smaller trade size), suggesting that short sellers applied more active trading strategies. These findings are consistent with those reported by the SEC s OEA (2007). 8 While most research indicates that price tests deter some short selling, the broader impact of these price tests remains hotly contested. On July 6, 2007, the SEC repealed Rule 10a-1 and any rule of exchanges (including NASD s Rule 3350) that applied price tests on short-sale orders. This decision prompted a huge backlash from practitioners and politicians. The former 6 Rule 3350 applies to Nasdaq National Market (Nasdaq NM or NNM) securities; securities traded in the OTC markets, including Nasdaq Small Cap, OTCBB, and OTC Pink Sheets, are exempted. When Nasdaq became a national listed exchange in August 2006, NASD Rule 3350 was replaced by Nasdaq Rule 3350 for Nasdaq Global Market securities (formerly Nasdaq NM securities) traded on Nasdaq, and NASD Rule 5100 for Nasdaq NM securities traded over-the-counter. 7 The Nasdaq switched from fractional pricing to decimal pricing over the interval of March 12, 2001 April 9, Prior to decimalization, Rule 3350 required a short sale to occur at a price 1/8 th dollar (if before June 2, 1997) or 1/16 th dollar (if after June 2, 1997) above the bid. 8 The evidence on the effect of price tests on short selling activity is not unanimous, however. Christophe, Ferri, and Angel (2004) find that short-sale activity is not significantly different between stocks that are covered by Rule 3350 and a matched sample of Nasdaq SmallCap Market stocks that are not. Alexander and Peterson (2008) report a higher number of short-sale trades only for NYSE-listed pilot stocks during the pilot program. They do not detect any significant impact on overall short-sale volume for either the NYSE- or the Nasdaq- listed pilot stocks. 6

8 state banking superintendent of New York argued that the SEC s repeal of the price tests added to market volatility, especially in down markets. 9 The Wall Street Journal argued that the SEC s Office of Economic Analysis was too biased to evaluate the price tests fairly. 10 Wachtell, Lipton, Rosen & Katz, a well-known law firm, argued that the uptick rule should be reinstated immediately, and three members of Congress introduced a bill (H.R. 6517) to require the SEC to reinstate the uptick rule. Presidential candidate Sen. John McCain blamed the SEC for the recent financial turmoil by turning our markets into a casino and called for the dismissal of the chairman of the SEC. On February 24, 2010, the SEC restored a modified uptick rule. Under the new rule, price tests are triggered when a security s price declines by 10% or more from the previous day s closing price The impact of the pilot program on earnings management Executives have incentives to distort their firms reported financial performance for personal gain, e.g., to bolster their compensation, gains through stock sales, job security, operational flexibility, or control. 11 Managers cannot, however, manipulate earnings with impunity (see Hazarika et al., 2012). One constraint on earnings management is provided by short selling. Empirical evidence shows that, short selling tracks a firm s discretionary accruals (e.g., Cao et al., 2006; Karpoff and Lou, 2010; Hirshleifer, Teoh, and Yu, 2011). An increase in discretionary accruals therefore invites greater scrutiny by short sellers and the potential for short 9 Gretchen Morgenson, Why the roller coaster seems wilder, The New York Times, August 26, 2007, Page There s a better way to prevent bear raids, The Wall Street Journal, November 18, 2008, Page A Safdar (2003), Burns and Kedia (2006), Bergstresser and Philippon (2006), Efendi, Srivastava, and Swanson (2007), and Cornett, Marcus, and Tehranian (2008) conclude that managers manage earnings to boost their compensation. Beneish and Vargus (2002) and Bergstresser and Philippon (2006) report that insider stock sales increase during periods in which firms have abnormally high accruals. DeFond and Park (1997), Ahmed, Lobo and Zhou (2006), and Mergenthaler, Rajgopal, and Srinivasan (2009) show that managers career concerns affect earnings management. DeFond and Jiambalvo (1994) and Sweeney (1994) show that CEOs have incentive to manage earnings to gain operational flexibility and control. 7

9 selling pressure on the firm s share price. Short selling also facilitates the flow of unfavorable information into share prices (Miller, 1977), dampens artificial price inflation that is created by earnings management, and helps to uncover aggressive earnings manipulations that eventually draw attention from regulators and lead to enforcement actions (Karpoff and Lou, 2010). These results suggest that short selling decreases the benefits and increases the costs of earnings management for managers. Benefits decrease because short selling increases price efficiency and dampens the artificial inflation in share prices that motivates much earnings management in the first place. Costs increase because short selling helps to monitor managers reporting behavior. As stated above, high levels of discretionary accruals attract scrutiny by short sellers and increase the likelihood that the earnings management will be revealed and disciplined. Indeed, Hazarika et al. (2012) find that the likelihood of forced CEO departure increases with the firms use of discretionary accruals. We hypothesize that both effects of short selling on managers incentives to manage earnings a decrease in benefits and an increase in costs work to reduce the amount of earnings management in which managers engage. Regulation SHO s pilot program (Rule 202T), which eliminated price tests on short selling for the pilot stocks, represents an exogenously imposed reduction in the cost of short selling in the pilot stocks. This implies that the pilot program should be associated with a decrease in the amount of earnings management at the pilot firms compared to the non-pilot firms. The tests below examine this prediction. Our tests also examine four subsidiary implications of the effect of the pilot program on earnings management. First, we investigate whether the impact of the pilot program on discretionary accruals is more pronounced among firms for which the short selling price tests are likely to be a binding constraint on short selling activity, i.e., firms with large institutional 8

10 shareholdings. Second, we investigate whether the impact of the pilot program is more pronounced among firms in which managers have strong incentives to manage earnings because their wealth is closely tied to the company s share price. Third, we examine the impact of the pilot program on the discovery of financial misrepresentation that leads to enforcement action by the SEC. Finally, we examine whether greater short selling, or its prospect, corresponds to an increase in the informativeness of the pilot firms share prices during the pilot program. 3. Data 3.1. Sample On July 28, 2004, the SEC issued its first pilot order (Securities Exchange Act Release No ) and published a list of 986 stocks that would trade without being subject to any price tests during the term of the pilot program. 12 To create this list, the SEC first excluded stocks that were not previously subject to price tests (i.e., not listed on NYSE, AMEX, or Nasdaq NM) and stocks that went public or had spin-offs after April 30, The remaining 2004 Russell 3000 index members were then sorted by each stock s average daily dollar volume computed from June 2003 through May 2004 within each of the three listing markets. Finally, every third stock (beginning with the second one) within each listing market was designated as a pilot stock. To construct our sample, we start with the 2004 Russell 3000 index and also exclude stocks that were not listed on the NYSE, AMEX, or Nasdaq NM, and stocks that went public or had spin-offs after April 30, Based on the SEC s pilot order, we identify an initial sample of 986 pilot stocks and 1,966 non-pilot stocks. An examination of the exchange distribution of these stocks shows that both the pilot and non-pilot groups are representative of the Russell The list can be obtained from the SEC s website: 9

11 index, confirming the statistics reported by the SEC. Specifically, of the 986 pilot stocks, 49.9% (492) are listed on the NYSE, 47.9% (472) on the Nasdaq NM, and 2.2% (22) on the AMEX. The exchange distribution of non-pilot stocks is very similar, with 50% (982) listed on the NYSE, 48% (944) on the Nasdaq NM, and 2% (40) on the AMEX. In our tests we delete firms in the financial services (SIC ) and utilities industries (SIC ) because disclosure requirements and accounting rules are significantly different for these regulated industries. We require data from the Compustat Industrial Annual Files to construct earnings management proxies and calculate control variables. In most tests, we require all firms to have data to calculate firm characteristics across the entire sample period, and The resulting balanced panel sample consists of 388 pilot firms and 709 control firms. If we relax this requirement, our unbalanced panel sample contains pilot firms and 1,504-1,610 control firms in the year immediately before the announcement of the pilot program (i.e., 2003), depending on the data availability to calculate a given firm characteristic. We emphasize the results from the balanced panel sample, but also report results for the unbalanced sample. Throughout, the results are similar using either sample Key test variables We create an indicator variable PILOT to denote firms with pilot stocks (pilot firms). Specifically, PILOT equals one if a firm s stock is designated as a pilot stock under Regulation SHO s pilot program and zero otherwise. Pilot firms constitute the treatment sample and nonpilot firms serve as the control sample. We also construct three variables to indicate time periods: PRE equals one if a firm-year s fiscal end falls between January 1, 2001 and December 31, 2003 and zero otherwise; DURING equals one if a firm-year s fiscal end falls between 10

12 January 1, 2005 and December 31, 2007 and zero otherwise; and POST equals one if a firmyear s fiscal end falls between January 1, 2008 and December 31, 2010 and zero otherwise. In our DiD tests, we skip the year 2004 because the identity of the pilot and non-pilot stocks was made public in July 2004, so it is not clear whether year 2004 should be classified as pre- or during-pilot period. In the Internet Appendix, we report tests that indicate the results reported here are not substantially affected if we include 2004 in the PRE or DURING periods Measures of earnings management We proxy for earnings management using abnormal accruals, defined as the difference between total accruals and the fitted normal accruals derived from a modified version of the Jones model (Jones, 1991). Specifically, for each fiscal year and Fama-French 48 industry, we estimate the following cross-sectional model using all firms with the necessary financial data available in Compustat: (1) where i indexes firms and t indexes fiscal years. The intercept term is added to the original Jones model because Kothari, Leone, and Wasley (2005) argue that it provides additional control for heteroskedasticity and results in a more symmetric measure of discretionary accruals. 13 Total accruals TA i,t+1 are estimated using the statement of cash flow approach and defined as earnings before extraordinary items and discontinued operations (Compustat item IBC) minus operating 13 In the Internet Appendix we show that our results are robust to using a discretionary accrual measure removing the intercept item in the estimation procedure, i.e., the measure of Dechow, Sloan, and Sweeney (1996). 11

13 cash flows (OANCF-XIDOC), ASSET i,t is total assets at the end of fiscal year t (AT), ΔREV i,t+1 is the change in sales revenue (SALE) from fiscal year t to t+1, and PPE i,t+1 is the gross value of property, plant and equipment (PPEGT) at the end of fiscal year t+1. We require at least 10 observations to perform each cross sectional estimation. Next, we use the following model and the estimated coefficients from Eq. (1) to compute the fitted normal accruals NA i,t+1, (2) Following prior studies, the change in accounts receivable (RECT) is subtracted from the change in sales revenue as credit sales might also provide potential opportunity for accounting distortion. After obtaining the fitted normal accruals NA i,t+1 from Eq. (2), we then calculate firm-year specific discretionary accruals as DA i,t+1 = (TA i,t+1 / ASSET i,t ) - NA i,t+1. Finally, as recommended by Kothari, Leone, and Wasley (2005), we adjust the estimated abnormal accruals for performance. We match each sample firm with another firm from the same fiscal year and the same Fama-French 48 industry that has the closest return-on-assets ratio (IB divided by AT) as the given sample firm. The performance-matched abnormal accrual, denoted as DA1, is calculated as the firm-specific abnormal accruals minus the abnormal accruals of the matched firm. To test the robustness of this measure, we again follow Kothari, Leone, and Wasley (2005) and create a second accrual measure, DA2, by replacing with when estimating Eq. (1), taking into account the potential accounting 12

14 distortion in credit sales in both steps of the estimation procedure. Both DA1 and DA2 are signed and constructed to be positively related to income-increasing earnings management Firm characteristics Following Grullon, Michenaud, and Weston (2011), we compare the pilot and non-pilot firms characteristics in the fiscal year immediately before the announcement of the pilot program, Table 1, Panel A reports on the balanced panel sample, in which we require a firm to have financial data available to calculate firm characteristics and accrual measures in all years of the sample period. The mean book value of assets in both groups is $3.7 billion. The two groups also exhibit similar mean and median values of the market-to-book ratio, one-year growth in assets, capital expenditures-to-assets ratio, R&D expenditures-to-assets ratio, annual return-on-assets ratio, cash flow-to-assets ratio, leverage, and the level of cash and dividends (both as a percentage of total assets). In none of these comparisons is the difference statistically significant, which supports our contention that Regulation SHO s pilot program is a wellcontrolled experiment that is suitable for examining the effects of short sales constraints. Panel B of Table 1 reports similar comparisons for the larger unbalanced panel sample. Firms in the unbalanced panel sample are slightly smaller than the firms included in the balanced panel sample, with assets averaging $2.9 billion versus $3.7 billion. As in Panel A, the pilot and non-pilot firms in the unbalanced panel sample are similar to each other in the other financial characteristics we examine. The sole exception is that the median capital expenditure by pilot firms is slightly higher than that for the control firms. 13

15 4. The effect of Regulation SHO s pilot program on discretionary accruals 4.1. Univariate difference-in-differences tests Table 2 reports the results of the DiD tests that examine the impact of the pilot program on firms discretionary accruals. Panel A reports on the balanced panel sample, in which firms are included only if they have data for all years examined in the tests. The mean value of DA1 during the three-year period before the pilot program, , is for both the pilot firms and control firms. The t-statistic for the difference in means is -0.03, and the Wilcoxon z- statistic is During the three-year period of the pilot program, the mean value of DA1 decreases to for the pilot firms while it remains at for the control firms. The t- statistic for the difference in means is and the Wilcoxon z-statistic is -2.23, both significant at 5% level. For the three-year period after the pilot program, DA1 increases for the pilot firms to a mean of zero, while it changes slightly for the control firms to The t-statistic for the difference in means in the post-pilot period is 0.69, and the Wilcoxon z-statistic is Similar results obtain using the alternate measure of discretionary accruals, DA2. These results are illustrated in Figure 1. As the figure shows, the control firms discretionary accruals do not change much over the entire sample period. The pilot firms discretionary accruals are similar to those of the control firms before the pilot program, decrease significantly during the program, and then revert to levels that are similar to those of the control firms after the program. The bottom-right cell of Panel A of Table 2 reports on the DiD estimators. The mean DiD estimator for DA1 from before to during the pilot program is with a t-statistic of This difference is statistically significant only at the 10% level. However, the results from other tests reported below, including multivariate DiD tests and the results from the unbalanced 14

16 panel, are significant at lower levels. Also, the DiD estimator that tracks DA1 from during to after the pilot program is with a t-statistic of Further, the DiD estimator that compares DA1 pre-program to post-program is statistically insignificant with a t-statistic of The last two DiD estimators demonstrate that the effect of the pilot program on discretionary accruals reverses when the program ends an important check on the internal validity of the DiD test. The results are similar using DA2. Table B of Table 2 reports on the changes in DA1 and DA2 using data from the unbalanced panel sample in which we do not require firms to have financial data available for all years of the sample period. The results are similar to those from the balanced panel sample Multivariate difference-in-differences tests In this section we extend the DiD test using multivariate regressions. We retain firm-year observations for both treatment and control firms for a nine-year window ( and ) surrounding Regulation SHO s pilot program and estimate the following model: DA[1,2] i,t = β 0 + β 1 PILOT i DURING t + β 2 PILOT i POST t + β 3 PILOT i + β 4 DURING t + β 5 POST t + ε i,t (3) The dependent variable is one of the two discretionary accrual measures, DA1 i,t or DA2 i,t. PILOT i is an indicator variable that denotes treatment firms. DURING t and POST t are time indicators that denote the three-year period during Regulation SHO s pilot program ( ) and the three-year period after the program ended ( ). The benchmark period consists of the three years before the pilot program ( ). Again, the year 2004 is 15

17 omitted from these tests because the pilot program began midway through The regression results estimating Eq. (3) are reported in the first two columns of Table The key coefficient estimate of interest, β 1 on PILOT i DURING t, is negative and significant at the 5% level in both columns. The magnitude of is consistent with the univariate DiD results reported in Table 2 and indicates that DA1 (i.e., discretionary accruals as a percentage of total assets) is 1 percentage point lower for the treatment group than for the control group during the three-year period of the pilot program compared to the three-year period before the program. The coefficient on PILOT i, β 3, is insignificant, consistent with pilot firms and non-pilot firms exhibiting similar levels of discretionary accruals before the pilot program. The coefficient on PILOT i POST t, β 2, is also insignificant, which once again demonstrates the reverting pattern as the difference between the pilot and non-pilot firms discretionary accruals after the pilot program is not statistically different from that before the program. In Columns (3) and (4), we augment Eq. (3) by including four controls that have been shown to affect a firm s level of discretionary accruals (e.g., Becker et al., 1998; Kothari, Leone, and Wasley, 2005; Zang, 2012): the natural logarithm of total assets (SIZE), market-to-book ratio (MB), return-on-assets (ROA), and leverage (LEV). The results are similar when we include these additional controls. Prior research shows that a firm s discretionary accruals are highly correlated with its growth (e.g., Fairfield, Whisenant, and Yohn, 2003) and its incentives to issue equity (e.g., Teoh, Welch, and Wong, 1998a, 1998b). Since Grullon, Michenaud, and Weston (2011) document that pilot firms significantly reduce their equity issuance and investment during the pilot program in response to a dropped stock price, it is possible that our finding regarding discretionary accruals 14 Consistent with prior research, the R 2 s from the regressions generally are low, indicating that most of the crosssectional differences in discretionary accruals are due to unmodeled or firm-specific factors. 16

18 is driven by changes in the difference between the pilot and non-pilot firms investment or equity issuance surrounding the pilot program. Note that our discretionary accrual measures are performance-matched, following Kothari, Leone, and Wasley (2005), and are intended to remove any variation in accruals due to the changes in performance such as those caused by a reduction in investment. In addition, we include the market-to-book ratio in Eq. (3), a standard control for investment opportunities. Nevertheless, to ensure that the changes in the difference between pilot and non-pilot firms investment is not driving the effect of pilot program on discretionary accruals, we re-estimate Eq. (3) controlling for R&D expenditures (R&D) and capital expenditures (CAPEX), both scaled by beginning-of-year total assets. We further include a control for firms equity issuance (EQUITYISSUE) also scaled by beginning-of-year total assets, as in Grullon, Michenaud, and Weston (2011). The results including these additional controls are reported in Table 4. In Columns (1) and (2), we include R&D and CAPEX in the regressions separately with DA1 and DA2 as the dependent variable. 15 In Columns (3) and (4), we sum the two variables to derive a firm s total investment INVESTMENT. As shown, the coefficients on our variable of interest, PILOT DURING are barely affected by the inclusion of these control variables. The results remain similar when we include EQUITYISSUE in Columns (5)-(6) and both INVESTMENT and EQUITYISSUE in the last two columns. In summary, the results in Tables 2-4 show that the reduction in short-sale constraints from Regulation SHO s pilot program is associated with a significant decrease in discretionary accruals. This implies that an increased threat of short selling curbs opportunistic reporting 15 The results are very similar if we measure capital expenditures as the annual increase in gross fixed assets from the balance sheet. 17

19 behavior. Because the pilot program was designed specifically to compare a random sample of pilot stocks with the non-pilot stocks in a controlled experiment, it is difficult to attribute these findings to reverse causality (which would require the SEC to anticipate the changes in discretionary accruals when assigning the pilot stocks) or an omitted variable (which would suggest the assignment by the SEC was not random). Additional evidence indicates that the effect on discretionary accruals reversed after the pilot program ended and that it is not explained by changes in pilot firms investment or equity financing relative to non-pilot firms. This lends further support to the view that the decreases in pilot firms discretionary accruals are due to the treatment of the pilot program, i.e., that the reduction in short selling constraints caused the drop in discretionary accruals of the affected firms Multivariate difference-in-differences partitioned on institutional ownership Previous researchers argue that the cost of short selling is negatively related to a firm s level of institutional ownership because it is relatively easy to borrow shares from institutional owners to establish short positions (e.g., Nagel, 2005; Hirshleifer, Teoh, and Yu, 2011). Short selling in firms with low levels of institutional ownership, in contrast, can be difficult simply because of a scarcity of shares to borrow. Consistent with this argument, institutional ownership is positively related to short selling in empirical tests (e.g., Karpoff and Lou, 2010). In this section we investigate whether the negative effect of Regulation SHO s pilot program on pilot firms discretionary accruals is related to the level of institutional ownership in the firm. If the removal of price tests works to discourage discretionary accruals because of an increase in the prospect of short selling, the effect should be largest among firms in which short selling is most likely. Stated differently, the threat of increased short selling should have little 18

20 effect in firms with low institutional ownership because the cost of borrowing these firms shares may be prohibitively high. 16 To test this conjecture we partition the balanced panel sample of treatment and control firms into two subsamples based on whether a firm s institutional ownership in the fourth quarter of 2003 is above the sample median. Institutional ownership is retrieved from Thomson's CDA/Spectrum database (form 13F) and aggregated on the firm level. We define institutional ownership as the shares held by all institutions divided by the total shares outstanding at the end of the quarter from CRSP monthly files (adjusted for stock splits and other distributions), following Asquith, Pathak, and Ritter (2005). The percentage of pilot firms is 33.9% in the subsample with high institutional ownership and 37.0% in the subsample with low institutional ownership. Table 5 reports the results from re-estimating Eq. (3) using these two subsamples. Consistent with our conjecture, the impact of Regulation SHO s pilot program on discretionary accruals is much more pronounced among firms with high levels of institutional ownership. In fact, the evidence in Columns (5)-(8) indicates that there is no significant effect among firms with below-median levels of institutional ownership. Among firms with high levels of institutional ownership, the magnitude of the effect is nearly twice that reported in Table 3 for the overall sample. The differences in the magnitudes of the coefficients are statistically significant at the 5% level. 16 A counter-argument is that the incremental impact of Regulation SHO s pilot program will be greatest among firms with low institutional ownership. This could occur if the elimination of price tests is a substitute for institutional ownership in lowering the costs of short selling. This conjecture would imply that the elimination of price tests can offset the high cost of establishing short positions that comes from locating shares to borrow. This argument, while plausible, is inconsistent both with our priors and with the evidence we report in this section. 19

21 We also note that the coefficient on PILOT is positive and significant in the subsample of firms with high institutional ownership, and negative and significant among firms with low institutional ownership. This suggests that the pilot firms and non-pilot firms exhibit different levels of discretionary accruals within the two subsamples before the pilot program. Given that pilot firms have larger discretionary accruals to begin with in the subsample of firms with high institutional ownership, one concern is that the larger drop we observe for the pilot firms during the pilot period merely captures a reversal of previous periods earnings management, as reflected in the higher discretionary accruals. However, the magnitude of the coefficient on PILOT DURING ( in Column (1)) is consistently much larger than that of the coefficient on PILOT (0.008 in Column (1)). This indicates that there is a separate effect on the pilot firms that is over and above any possible reversal of discretionary accruals. Overall, the results in Table 5 indicate that the disciplinary effect of short sellers on earnings management is most pronounced among firms that have high levels of institutional ownership. Previous evidence indicates that the cost of short selling is negatively related to institutional ownership because institutional owners increase the supply of shares that can be borrowed to implement a short sale. Our finding indicates that the impact of Regulation SHO s pilot program on short selling costs is largest among firms for which price tests are the binding constraints on short selling An alternative interpretation is that the pilot program encouraged institutional investors to pressure managers to abstain from engaging in earnings manipulation. It is not clear exactly how the pilot program would have such an effect unless it was through the threat of greater short selling. Even under this interpretation, the changes were prompted by the exogenous reduction in short-sale constraints for pilot firms through Regulation SHO s pilot program. 20

22 4.4. Multivariate difference-in-differences partitioned on wealth-performance sensitivity A prominent motivation for managers to manipulate earnings through discretionary accruals is to increase the value of their equity holdings. 18 If short selling works to discipline equity compensation-motivated earnings management, we should expect the pilot program to have relatively large effects on earnings management where managers wealth is closely tied to share prices. To examine this conjecture, we partition the balanced panel sample of pilot and non-pilot firms into two subsamples based on whether the firm CEO s scaled wealth-performance sensitivity (WPS) in 2003 is above the sample median. The WPS measure, which aims to capture the sensitivity of the CEO s wealth to the stock price, is computed as the dollar change in the CEO s wealth for a 100 percentage point change in the stock price, scaled by annual pay (see Edmans, Gabaix, and Landier, 2009). The WPS measure is calculated using the ExecuComp database. 19 Since ExecuComp covers S&P 1500 firms, this reduces our sample size by 29%. In this reduced sample, pilot firms constitute 39.8% of firms for which WPS is above the median, and 35.5% of the firms for which WPS is at or below the median. The results re-estimating Eq. (3) using the two subsamples partitioned by WPS are reported in Table 6. The coefficient estimate on PILOT DURING is significant and negative only in the subsample of firms with above median WPS, with or without control variables included. That is, DA1 or DA2 is predictively lower for pilot firms during the pilot program, but only in the subsample of firms with above median WPS. This result is consistent with the view 18 For examples, see Safdar (2003), Burns and Kedia (2006), Bergstresser and Philippon (2006), Efendi, Srivastava, and Swanson (2007), and Cornett, Marcus, and Tehranian (2008). 19 The data were kindly made available by Alex Edmans. 21

23 that short selling works to discipline earnings management that is motivated by managers efforts to boost the value of their equity holdings. 5. The effect of Regulation SHO s pilot program on financial misrepresentation The results in Tables 2 6 indicate that the pilot firms tended to decrease their earnings management during the pilot program. In this section we examine whether the pilot program also affected the incidence and discovery of earnings management that is aggressive enough to attract enforcement action by the SEC for financial misrepresentation. A decrease in the cost of short selling has offsetting effects on the likelihood that a firm will be caught misrepresenting its financial statements. On one hand, our results indicate that firms are less likely to manage earnings in general, implying a decrease in the likelihood of extreme earnings manipulations that could potentially attract regulatory scrutiny. On the other hand, Karpoff and Lou (2010) show that short selling accelerates the discovery of financial misconduct, so a decrease in short selling costs should correspond to an increase in the likelihood of discovery for any earnings manipulations that do occur. We therefore expect the pilot firms to experience a decrease in the likelihood of earnings manipulations, but an increase in the conditional probability of detection for those firms that do manipulate earnings. The overall effect on the rate at which pilot firms are caught manipulating earnings is ambiguous. To test these propositions, we conduct two types of tests using data on the incidence of SEC enforcement actions for financial misrepresentations, which are collected by and described in Karpoff et al. (2013). These data contain all enforcement actions initiated by the SEC from 1978 through 2011, and include information on the SEC s determination of when the financial misrepresentation began, ended, and was first revealed to the public. 22

24 We first examine whether pilot firms are more or less likely to be caught misrepresenting their financial statements, by estimating the following model: VIOLATION[1,2] i,t = β 0 + β 1 PILOT i + β 2 SIZE i,t + β 3 MOMENTUM i,t + β 4 RETVOL i,t + β 5 TURNOVER i,t + β 6 FRAUDPROB i,t + ε i,t (4) The dependent variable VIOLATION i,t is defined in two alternative ways. VIOLATION1 i,t equals one if a firm has a reporting violation that was initiated during the pilot period of 2005 through 2007, and zero otherwise. For VIOLATION2 it we expand the definition to code a reporting violation as one if it was either initiated during the pilot period or initiated before the pilot period and continued into this period, and zero otherwise. The control variables include those used by Wang, Winton, and Yu (2012) to examine the determinants of fraud and detection (firm size measured by the log of assets, annual stock buy-and-hold return, daily return volatility, and stock turnover), plus the probability of misstatement measure from Dechow et al. (2012). Panel A of Table 7 reports the results from estimating Eq. (4) using the balanced panel sample of pilot and non-pilot firms, although the results are similar for the larger unbalanced panel sample. The additional data required to estimate Eq. (4) shrink the sample size from 1,097 to 966 firms. As shown in Column (1), the probability of initiating a reporting violation during the pilot period and getting caught (VIOLATION1) is slightly higher among the pilot firms, but the coefficient is statistically insignificant with a p-value of In Column (2), the probability of having a reporting violation that extends into the pilot period is also not significantly different across two groups of firms. These results indicate that the unconditional probability of initiating or continuing financial misrepresentation and getting caught is not 23

25 significantly higher for pilot firms than for non-pilot firms during the pilot program. This is consistent with our expectation that the effect of the pilot program has offsetting effects on the probability of initiating and getting caught for financial misrepresentation. To examine whether the conditional probability of getting caught is affected by the pilot program, we replicate the time-to-discovery tests used by Karpoff and Lou (2010) to examine the effect of short selling on the speed with which financial misrepresentation is uncovered. Specifically, we estimate a Cox proportional hazard model: DISCOVERED i,t = β 0 + β 1 PILOT i DURING t + β 2 PILOT i + β 3 DURING t + β 4 MONTHS i,t + β 5 MONTHS i,t 2 + β 6 SIZE i,t + β 7 BM i,t + β 8 MOMENTUM i,t + β 9 SEVERITY[1,2] i,t + ε i,t (5) The data consist of firm-month observations for all firms in the Karpoff et al. (2013) database that also are in the balanced panel sample of pilot and non-pilot firms and that were publicly revealed to have engaged in financial misrepresentation during or after the announcement of the pilot program (i.e., July of 2004, when the list of the pilot stocks was announced by the SEC). 20 The dependent variable, DISCOVERED i,t, equals one in the month in which firm i is revealed to have engaged in misrepresentation, and zero otherwise. For example, if a firm s violation period is from January 1999 through December 2005 and the misconduct is discovered in December 2005, this firm would have 84 observations (84 months from January 1999 through December 2005). DISCOVERED i,t would equal zero for the first 83 months and 1 for the last month. 20 Eq. (5) is estimated using firm-month observations, allowing us to define the beginning date of the pilot program more finely. We use the month July 2004, when the identity of the pilot firms became publicly known, as the cutoff date to ensure that we do not omit any potential effect of the pilot program. Our results are robust to using January 2005 or May 2005 as the alternative cutoff date. Also, note that, since we restrict our focus to only the reporting violations that took place during or after July 2004 in this test, we only need to include one time dummy, DURING t, to differentiate during- and post-pilot periods. 24