Financial Restatement Announcements and Insider Trading

Transcription

1 Financial Restatement Announcements and Insider Trading Oliver Zhen Li University of Notre Dame Yuan Zhang Columbia University October, 2006 ABSTRACT We examine insider trading activities around financial restatement announcements and find strong evidence of informed trading by insiders. Focusing on the association between net insider selling and restatement announcement abnormal returns, we provide evidence of net insider selling before the restatement announcements, little net insider selling immediately around the announcements, and net insider buying after the announcements. The passage of the Sarbanes- Oxley Act constrains informed insider selling before the restatement announcements. Trading before the restatement announcements enables insiders to make a significant profit, especially pre-sarbanes-oxley. Overall, our results suggest that insiders trade on privileged knowledge about the forthcoming restatement announcements to their advantage and that they trade in a pattern that minimizes the possibility of insider trading allegations or violating internal corporate insider trading policies. We thank Mei Cheng, Jim Ohlson, and workshop participants at the University of Florida for helpful comments and suggestions. All errors are ours.

2 Financial Restatement Announcements and Insider Trading 1. Introduction Financial restatements are salient negative firm-specific events. Prior studies have shown that financial restatements are associated with a decrease in firm value (Anderson and Yohn, 2002; Palmrose, Richardson and Scholz, 2004), a decrease in future earnings prospect and an increase in the cost of equity capital (Hribar and Jenkins, 2004). The negative valuation impact of financial restatements may induce firm insiders, who presumably possess advance knowledge about the timing, likelihood and severity of financial restatements, to trade shares in a pattern that avoids the negative impact of financial restatements on their personal investment in the restating firms. In this paper, we examine insider trading activities around financial restatement announcements. Specifically, using the Financial Statement Restatement Database complied by the United States Government Accountability Office (formerly the United States General Accounting Office; GAO, 2003 and 2006), 1 we examine two issues: first, the direction and the magnitude of insider trading activities around financial restatement announcements; and second, the timing of their trades. We focus on the relation between net insider selling activities and restatement announcement abnormal returns and find strong evidence of informed insider trading activities around the announcements. We show that net insider selling is negatively related to restatement announcement abnormal returns in Quarters -8 to -2 before the announcements. During the quarter immediately preceding the restatement announcement (Quarter -1) and the 1 Effective July 7, 2004, the GAO's legal name changed from the General Accounting Office to the Government Accountability Office. 1

3 quarter immediately after the announcement (Quarter 0), insiders cease restatement-related net selling of shares. The relation between net insider selling and restatement announcement abnormal returns is insignificant. During Quarters 1, 2 and 4 after the announcements, insiders reverse the direction of their trades and start net buying of shares. The relation between net insider selling and restatement announcement abnormal returns is positive. Further, the passage of the Sarbanes-Oxley Act of 2002 appears to constrain information-motivated insider selling activities before the restatement announcements. We also show that insiders make an economically significant profit by trading before financial restatement announcements, especially prior to the passage of the Sarbanes-Oxley Act. Overall, our results suggest that: 1) insiders trade on private information about the forthcoming restatement announcements in a direction that allows them to avoid the negative wealth impact of the news on their investment in the restating firms; 2) insiders refrain from trading too close to the announcements to minimize insider trading litigation risks and/or to avoid violating internal corporate insider trading restrictions. Our paper is related to studies that examine insider trading activities during periods of fraudulent financial reporting (Beneish, 1999; Dechow, Sloan and Sweeney, 1996; Summers and Sweeney, 1998). We note that financial restatements, often involving aggressive accounting, are not necessarily fraudulent, although their announcements frequently trigger significantly negative market reactions. More importantly, these prior studies focus on the manipulation period (as opposed to the public disclosure of the manipulation) only and find mixed evidence. 2 It is ambiguous whether insiders will engage in abnormal insider selling during periods of earnings manipulations. On the one hand, insiders may sell stock during the misstated periods 2 For example, Beneish (1999) and Summers and Sweeney (1998) find evidence that insiders increase their selling during periods of fraudulent financial reporting, while Dechow, Sloan and Sweeney (1996) find no or very weak evidence of informed insider selling. 2

4 while the stock price is supported by inflated earnings. On the other hand, insiders may postpone selling in the hope that the accounting problems will not be revealed and that the stock price will continue to rise (Agrawal and Cooper, 2006). A concurrently developed study by Agrawal and Cooper (2006) also examines restatementrelated insider trading activities. They find evidence that top managers of restating firms sell more stocks for some subsamples where insiders have greater incentives to sell before the revelation of accounting problems, but not for the full sample of restating firms. Their research methodology is different from ours. Specifically, they focus on the period from the first day of the restated quarter to the restatement announcement date, which includes both the manipulation period and the period prior to the public disclosure of the restatements. As discussed above, insider trading activities during the manipulation period is difficult to predict. Further, prior research has shown that insiders usually trade well in advance of, as opposed to immediately prior to, major corporate disclosures (Seyhun and Bradley, 1997; Ke, Huddart and Petroni, 2003). Thus, if there are muted informed insider trading activities during the test period or if there are some informed insider trading activities in the control period, Agrawal and Cooper s (2006) research design may not be able to detect abnormal insider trading activities associated with the announcements of financial restatements. In contrast, we focus on the quarters around the public disclosure of financial restatements. Since the stock prices of the firms announcing restatements usually drop significantly at the time of the announcements (Anderson and Yohn, 2002), we expect insiders to have incentives to engage in informed selling to avoid material personal losses in their investments in the restating firms. Specifically, we link net insider selling activities to the market response to restatement announcements in order to detect informed insider trading, if any. The reason for doing this is 3

5 simple: It is the negative market response to financial restatements, not necessarily the restatements per se, that motivates insider trading. Our paper also differs from Agrawal and Cooper (2006) in that we present the timing pattern of insider trading activities. Further, our sample includes restatements announced after June 2002, which enables us to investigate whether there is any change in insider trading around restatement announcements after the passage of the Sarbanes-Oxley Act of Our paper makes a contribution to the insider trading literature. While it has been robustly documented that insiders earn abnormal returns on trades of their firms stocks, there is relatively limited evidence linking these trades to particular types of private information (Ke, Huddart, and Petroni, 2003). Our paper improves our understanding of insider trading in this respect by providing evidence that insiders trade on private information: specifically, the timing, likelihood and severity of financial restatements. It has implications for insider trading regulations, both at federal level and corporate level, regarding the timing of informed trading. Our results suggest that insiders not only trade on private information, but also carefully time their trades around the financial restatement announcements to avoid possible insider trading allegations and/or violating internal corporate insider trading restrictions, consistent with Seyhun and Bradley (1997) and Ke, Huddart and Petroni (2003). Our paper also provides some initial evidence on the impact of the Sarbanes-Oxley Act on insider trading activities. The Act aims to restore investor confidence in the capital market. While its main provisions address issues related to financial reporting and auditing, the Act has directly or indirectly affected almost every aspect of the capital market, including insider trading. Our results show little evidence of information-motivated insider trading activities prior to the public disclosure of financial restatements for periods after the passage of the Sarbanes-Oxley 4

6 Act. This is in sharp contrast to periods before the Act, suggesting that the post-sarbanes-oxley Act environment has at least constrained insiders from trading on privileged information to some extent. Our paper is subject to the caveat that we use insider trades reported under the filing requirements of Section 16 of the U.S. Securities and Exchange Act of Thus, they represent open market and legal insider trades (Seyhun and Bradley, 1997) as opposed to illegal insider trading (Meulbroek, 1992). To the extent that there are unreported insider trades, we measure insider trading activities with errors and perhaps biases. However, reporting errors and biases are expected to be relatively small because insiders have reputation and employment concerns and are subject to penalties imposed by insider trading legislations. Further, such errors or biases, if any, should bias against finding information-motivated insider trading activities. We also acknowledge that insider trading is not necessarily information-motivated. For example, insiders may trade due to their liquidity needs. However, we argue that insiders are more likely than outsiders to possess private information about their firms and to trade on that information. Even if insiders trade due to their liquidity needs, they can still time their trades to benefit from the private information they possess (Garfinkel, 1997; Huddart, Ke and Shi, 2006). The directional magnitude of insider trading activities and the timing of their trades around restatement announcements, and in particular, the statistical association between these trades and the event period abnormal returns found in our study suggest strong evidence of insider trading on privileged knowledge. This paper is organized as follows. In Section 2, we discuss insider trading activities around financial restatement announcements and develop hypotheses. In Section 3, we describe our data 5

7 and sample selection process. In Section 4, we discuss the empirical results. We conclude in Section Insider Trading Around Financial Restatement Announcements Insider trading is often linked to trading based on privileged knowledge and earns firm insiders abnormal profits. The literature has examined insider trading around major corporate events (Elliot, Morse and Richardson, 1984; John and Lang, 1991; Sivakumar and Waymire, 1994; Seyhun and Bradley, 1997) and insider trading in asymmetric information environment (Aboody and Lev, 2000; Beneish and Vargus, 2002; Frankel and Li, 2004; Aboody, Hughes and Liu, 2005; Piotroski and Roulstone, 2005; Huddart and Ke, 2006). These studies suggest that insiders take advantage of outsiders by trading on advance knowledge of specific firm events or privileged understanding of certain firm characteristics, such as R&D. In this study, we examine insider trading activities and profit around a major negative corporate event, the announcement of financial restatement. 2.1 DIRECTION AND TIMING OF INSIDER TRADING There are two essential aspects of informed insider trading: the direction of their trades and the timing of their trades. Notwithstanding federal rules or corporate policies against informed insider trading, insiders, with advance knowledge of future firm prospects, have a natural tendency to trade to benefit from material positive private information and to avoid expected losses from material negative private information. Since financial restatement announcements are associated with significant negative abnormal equity returns (Anderson and Yohn, 2002; 6

8 Palmrose, Richardson and Scholz, 2004; Hribar and Jenkins, 2004) and insiders have incentives to avoid losses in their investment in the restating firms, we expect them to sell restating firms shares before the news of the restatements is announced to the general public. More importantly, given their possession of private information, the intensity of their selling activities are expected to be positively associated with the magnitude of the restatement announcement abnormal returns, an indicator of the severity of a specific financial restatement. That is, the more severe a financial restatement, the more shares insiders will sell before its announcement. Thus, our first hypothesis deals with the directional magnitude of insider trading before the announcements of financial restatements: Hypothesis 1: The magnitude of the net insider selling activities before the financial restatement announcements is negatively related to the restatement announcement abnormal returns. While insiders generally have the incentives to trade to benefit from the positive valuation impact of good private information or to avoid the negative valuation impact of bad private information, their trades are subject to insider trading regulations that trading on the basis of material nonpublic information is prohibited. Such regulations can mitigate the kind of trading activities that we predict in Hypothesis 1 above. 3 For example, Garfinkel (1997) shows that after the passage of the Insider Trading and Securities Fraud Enforcement Act of 1988, insiders significantly alter their trading pattern by postponing liquidity sales until after negative earnings announcements. Further, insiders are also subject to the filing requirements of Section 16 of the Securities Exchange Act of Thus, firm insiders have to balance the gains from trading on privileged information and possible sanctions from insider trading allegations. Apart from 3 The "manipulative and deceptive devices" prohibited by Section 10(b) of the Securities Exchange Act of 1934 and Rule 10b-5 thereunder include, among other things, the purchase or sale of a security of any issuer, on the basis of material nonpublic information about that security or issuer, in breach of a duty of trust or confidence that is owed directly, indirectly, or derivatively, to the issuer of that security or the shareholders of that issuer, or to any other person who is the source of the material nonpublic information. 7

9 securities laws that restrict insider trading, most firms also have corporate policies that regulate insider trading activities. Bettis, Coles and Lemmon (2000) show that many firms have blackout periods relative to earnings announcements or other important corporate events during which the company prohibits trading by insiders. They provide evidence that these blackout periods are effective in suppressing insider trading activities. Based on the above discussion, the timing of trades is very important to insiders who trade around a certain event, to regulators who legislate and enforce insider trading regulations, and to researchers who examine insider trading activities. In fact, the timing of trades is an essential aspect of insider trading. To avoid insider trading allegations and/or violating internal corporate policies restricting insider trading activities, insiders have a tendency to avoid trading very close, and especially just prior, to major corporate events. Prior studies have found that insiders trade as early as two years before the break of a string of consecutive earnings increases (Ke, Huddart and Petroni, 2003), and five years before bankruptcy (Seyhun and Bradley, 1997). We argue that if insiders sell shares in restating firms before the financial restatement announcements, they will trade well in advance of the announcements and will refrain from trading very close to the announcements (Noe, 1999; Huddart, Ke and Shi, 2006). By following this trading pattern, they minimize the risk of insider trading allegations and/or violating internal corporate insider trading policies (Ke, Huddart and Petroni, 2003). Insiders have opportunities to trade well in advance of the actual restatement announcements because of their knowledge of the ongoing investigation/audit that leads to the restatements. Thus, our second hypothesis deals with the timing of insider trading: Hypothesis 2: Insiders commence trading well in advance of the financial restatement announcements and they avoid trading very close to the financial restatement announcements. 8

10 When insiders actually commence trading and when they cease trading around financial restatement announcements is an empirical question. Another issue related to insider trading activities around financial restatement announcements is that insiders may reverse the direction of their trades after the news of the financial restatements has been made public. 4 There are three possible reasons for this reversal: (i) market overreaction to the restatement announcements that insiders can further exploit (Seyhun, 1990); (ii) insiders need to rebalance their investment portfolios after net selling of shares in the restating firms; or (iii) a form of passive insider trading (Huddart, Ke and Shi, 2006). As Huddart, Ke and Shi (2006) discuss, to avoid the negative impact of restatement announcements, insiders can sell shares before the announcements or delay purchases until after the announcements. Selling before material negative news events is a form of active insider trading while delaying purchases until after the release of negative news is a form of passive insider trading. Therefore, we also examine insider trading activities subsequent to the announcements. This kind of insider trading may or may not be based on privileged information. 2.2 SARBANES-OXLEY ACT OF 2002 The Sarbanes-Oxley Act was signed into law in July This Act was initiated in response to numerous accounting scandals that surfaced in 2001 and The Act aims to restore investor confidence in and assure the integrity of the capital market. The reforms initiated by the Act address nearly every aspect and participant in the capital market. 5 For example, the Act establishes the Public Company Accounting Oversight Board, which oversees, regulates, 4 For example, in August 2003, with the stock of Tyco International Ltd. dipping after the company announced an accounting restatement, ten company executives spent $1.1 million to buy Tyco stock (Cooke, 2005). 5 See testimony by SEC Chairman William H. Donaldson in front of the House Committee on Financial Services on April 21, 2005 ( 9

11 inspects, and disciplines accounting firms in their roles as auditors of public companies. The Act also covers auditor independence, corporate governance and enhanced financial disclosure. While the Sarbanes-Oxley Act does not specifically address insider trading, it has implications for firm insiders who trade around corporate events. For example, the Act increases criminal penalties for defrauding shareholders of publicly traded companies; it requires management and principal stockholders to report their trades before the end of the second business day following the day on which the transaction has been executed; 6 it requires the disclosure of whether the issuer has adopted a code of ethics for senior financial officers. Thus, after the passage of the Sarbanes-Oxley Act, insiders are possibly subject to higher penalty and scrutiny for informed trading, which can potentially restrain informed insider trading activities around financial restatement announcements. We empirically investigate whether the passage of the Sarbanes- Oxley Act affects the magnitude and the timing pattern of insider trading activities. We expect that the Sarbanes-Oxley Act likely constrains information-motivated insider trading activities before financial restatement announcements. 3. Sample Selection and Variable Definitions 3.1 SAMPLE SELECTION Information on financial restatements is obtained from the Financial Statement Restatement Database complied by the United States Government Accountability Office (formerly the United States General Accounting Office; GAO, 2003 and 2006). This database focuses on financial restatements resulting from accounting irregularities and excludes those from business 6 Prior to the Act, insiders were subject to Section 16 of the Securities Exchange Act which stipulated that, directors, officers and 10% or greater stockholders report changes in their beneficial ownership every month within 10 days after the close of the calendar month in which the transaction occurred. 10

12 transactions such as mergers and acquisitions and those from general accounting changes or bookkeeping errors. Thus, the restatements in this database are more likely due to aggressive financial reporting, and hence are more likely to trigger negative market reactions upon their announcements. GAO (2003) includes 919 restatement announcements by 845 firms from January 1, 1997 to June 30, 2002 and GAO (2006) includes 1,390 restatements announcements by 1,121 firms from July 1, 2002 to September 30, Since we require insider trading data for five quarters after the restatement announcements and our insider trading database terminates by the end of 2005, we keep restatements announced between January 1, 1997 and December 31, 2004 for our analysis. The majority of the database is compiled using the Lexis-Nexis Power Search command and the U.S. News, Combined database, with the keyword of restate, restated, restating or restatement within 50 words of financial statement or earnings. The database also includes some restatements identified through other sources, such as the SEC (GAO, 2003). For each restatement announced, the database provides information on the date of the announcement, the reasons for the restatement, and the party who initiated the restatement. Insider trading information is obtained from Thomson Financial. The Insider Filing Database compiled by Thomson Financial is designed to capture all insider activities as reported on SEC forms 3, 4, and 5. Following prior studies, we focus on open market purchases and sells as reported in Table 1 of Form 4 (Ke, Huddart and Petroni, 2003). To ensure data quality, we delete all insider trading records that are assigned cleanse code A or S by Thomson Financial. 7 7 Thomson Financial verifies the accuracy and reasonableness of insider reported figures by reference to external sources. The cleanse code S indicates no cleansing attempted and security not meeting Thomson Financial s collection requirement; the cleanse code A indicates numerous data elements were missing or invalid and that reasonable assumptions could not be made. 11

13 Consistent with Ke, Huddart and Petroni (2003), we restrict the definition of insiders to officers and directors and exclude non-officer insiders (such as large shareholders, retired officers) because officers and directors are more likely to possess private information regarding the aggressive accounting, ongoing investigation, and the subsequent decision to restate prior financial statements. As discussed earlier, we focus on quarters around the restatement announcements as opposed to the periods of the aggressive accounting. Insiders are expected to have information regarding the forthcoming announcements because the decision to restate prior financial statements usually comes after a period of investigation or audit of the company s financial statements by the SEC, auditor, or the firm itself. 8 However, information about the duration of investigation or audit usually is not publicly available. We choose to focus on eight quarters prior to the restatement announcements to capture any potential informed trading. 9 We also analyze five quarters after the announcement date to understand insiders trading behavior subsequent to the significant market reaction at the announcement. Specifically, as illustrated in Figure 1, we define the restatement announcement date as the event day, and define thirteen 91- calendar-day periods around the event day: eight quarters before the event day and five quarters after the event day. For each quarter, we keep restating firms that have at least one insider transaction filed in Table 1 of Form 4 in the corresponding calendar year in the Thomson Financial Insider Filing database. A total of 1,274 restating firms with non-missing restatement announcement abnormal return RACAR are represented in at least one quarter. Table 1 provides descriptive statistics regarding these 1,274 restatement announcements. Panel A reports the magnitude of RACAR, the 8 For example, the restatement announcement by Xerox in April 2002 came in the wake of an audit ordered by the company of its books after SEC investigations in June 2000 and May 2001 of the company s revenue and lease accounting. 9 A majority of our sample firms announced that they would restate financial statements filed one or two years ago, which suggests that the investigation period is not likely longer than two years for these firms. 12

14 abnormal return from calendar Day -7 prior to the announcement to calendar Day 7 after the announcement. The abnormal return is calculated as the difference between the firm return and the equal-weighted market return. Consistent with prior research (Anderson and Yohn, 2002; Palmrose, Richardson and Scholz, 2004), market reactions to the announcements are significantly negative. The mean is -7.3% and the median is -3.8% for the full sample. Prior to the passage of the Sarbanes-Oxley Act, the mean is -10.3% and the median is -5.5%. After the passage of this Act, the mean is -4.0% and the median is -2.7%. 10 Thus, market reaction to restatement announcements is significantly lower after the passage of the Sarbanes-Oxley Act. Overall, the significant decrease in market values for these restating firms provides incentives for insiders to trade prior to the announcements to avoid significant losses in their personal investment in the restating firms. Panel B of Table 1 reports the frequency of different initiators of the restatements. For the full sample, 52% of the restatements are initiated by the company itself, 11% are initiated by the SEC, and 13% by the auditor. About 2% of the restatements are initiated by some other parties such as IRS or external parties. A significant percentage (29%) of the 1,274 restatement announcements did not identify an initiator. 11 After the passage of the Sarbanes-Oxley Act, auditor and company initiated restatements increase, from 8% prior to the passage of the Sarbanes-Oxley Act to 18% after the Act and from 42% prior to the Act to 64% after the Act, respectively, while SEC initiated restatements decrease from 14% prior to the Act to 7% after the Act In Table 1, restatement announcements are classified as pre-sarbanes-oxley and post-sarbanes-oxely based on the announcement date. In all remaining tables, firm-quarters are classified as pre-sarbanes-oxley and post- Sarbanes-Oxley based on the starting and ending dates of the corresponding insider trading quarter. 11 Since certain restatement announcements identify multiple initiators, the sum of the percentages is greater than 100%. 12 Note that the percentage of restatement announcements not identifying an initiator dropped from 36% to 21% after the Act. 13

15 Panel C of Table 1 reports the indicated reasons for the restatements. Consistent with Anderson and Yohn (2002) and Palmrose, Richardson and Scholz (2004), the most frequent reason for restating financial statements is revenue recognition. 36% of the 1,274 restatements involve revenue recognition. This percentage decreases from 42% pre-sarbanes-oxley to 29% post-sarbanes-oxley. About 28% of the restatements affect cost or expenses. This percentage increases from 20% pre-sarbanes-oxley to 37% post-sarbanes-oxley. Some other more frequent reasons include accounting for mergers and acquisitions (6%), restructuring (16%), securities (11%), and in-process R&D (2%). For the majority of our tests, we exclude observations with no insider trading or zero net selling in a particular quarter (Seyhun, 1986; Piotroski and Roulstone, 2005). This is because without insider trading activities during a specific quarter, we cannot define our insider trading metric as well as two important control variables, prior return and subsequent return. Within each quarter, we aggregate all insider sells and purchases respectively for each firm and obtain total numbers of shares sold or purchased. We delete observations within the top one percentile of either shares bought or shares sold to mitigate the influence of extreme observations or potential data errors. These procedures yield 639 firms in Quarter -8 and 514 firms in Quarter 4 relative to the restatement announcement. 3.2 REGRESSION MODEL AND VARIABLE DEFINITIONS To determine whether insiders engage in information-motivated trading around the restatement announcements, we estimate the following regression for each of the thirteen quarters around financial restatement announcements: SNIS i,t+q = β 0 + β 1 RACAR i,t + β 2 RM i,t + β 3 SEC i,t + β 4 COM i,t + β 5 PRERET i,t+q + β 6 POSTRET i,t+q + β 7 LCAP i,t + β 8 BM i,t + ε i,t. (1) 14

16 Note that in the above regression i denotes a restating firm, t denotes the time of the restatement announcement and Q [-8, 4] denotes a quarter relative to t. The dependent variable SNIS is the scaled net insider selling activities. We measure SNIS based on the number of shares traded. SNIS is the number of shares sold by insiders minus the number of shares bought by insiders scaled by the sum of total shares bought and sold by insiders for a specific quarter Q, similar to John and Lang (1991) and Beneish (1999). It is computed as (subscripts i and t omitted) SNIS Q S s= 1 # SOLD Q, s s= 1 b= 1 = S B # SOLD Q, s + B b= 1 # BOUGHT # BOUGHT Q, b Q, b, (2) where # SOLD Q,s is the number of shares sold by insiders in a sale transaction indexed by s [1, S] and # BOUGHT Q,b is the number of shares bought by insiders in a buy transaction indexed by b [1, B]. The above measure of net insider selling captures the directional intensity of insider trading activities for a specific quarter relative to the restatement announcement. It is likely superior to measures using unscaled net shares sold/bought or dollar value of shares sold/bought (Agrawal and Cooper, 2006), since scaling captures the intensity of net insider selling (John and Lang, 1991; Beneish, 1999). Also note that we do not define a control period for measuring normal net insider selling. Based on the literature, information-motivated insider trading activities are often spread out over a long period of time and are absent immediately prior to the public disclosure of material information (Seyhun and Bradley, 1997; Ke, Huddart and Petroni, 2003). Thus, the control period and event period are often difficult to identify. The difference-in-difference approach used in Agrawal and Cooper (2006) may not adequately capture abnormal insider trading activities. We endeavor to detect information-motivated insider trading activities by 15

17 linking net insider selling around the restatement announcements to abnormal announcement period returns. If net insider selling before restatement announcements is related to the restatement announcement abnormal returns in a pattern that reduces the negative wealth impact of restatements on insiders investments in the restating firms, we provide evidence of insiders trading on privileged information. The independent variables that we use to detect informed insider trading activities include the abnormal returns around the financial restatement announcements and certain characteristics of the restatements. Since the nature and characteristics of the restatements are only made public at the time of the announcements, any statistical link between net insider selling before the financial restatement announcements and abnormal announcement returns as well as these restatement characteristics is an indication of informed insider trading around the announcements of financial restatements. This statistical link, if exists, suggests that insiders have private information on the timing, probability, and severity of financial restatements and trade upon it. As discussed above, the restatement announcement abnormal return RACAR is defined as restating firm s return from calendar Day -7 to Day 7 relative to the restatement announcement minus the equal-weighted market return during the same period. Based on our argument earlier, if insiders trade in a pattern to avoid the negative valuation impact of the financial restatements, we expect net insider selling to be negatively associated with restatement announcement abnormal return RACAR several quarters before the restatement announcements. In the quarter immediately before the restatement announcement, we expect a reduced level of or no association between net insider selling and restatement announcement abnormal return, if insiders refrain from trading too close to the restatement announcements to avoid insider trading allegations or violating corporate insider trading policies (Huddart, Ke and Shi, 2006; Bettis, 16

18 Coles and Lemmon, 2000). In quarters after the restatements, we may see a positive association between net insider selling and restatement announcement abnormal returns, if insiders rebalance their portfolios after net selling of shares in the restating firms, detect possible market overreactions to the restatements, or engage in passive insider trading. We include three variables for the characteristics of financial restatements that are expected to affect how insiders trade around the restatement announcements. RM equals one if the restatement involves revenue manipulation and zero otherwise; SEC equals one if the restatement is initiated by the SEC and zero otherwise; and COM equals one if the restatement is initiated by the company itself and zero otherwise. Since revenue related restatements and SEC initiated restatements are potentially more severe financial restatements (Anderson and Yohn, 2002; Palmrose, Richardson and Scholz, 2004), we expect them to be positively related to net insider selling before the restatement announcements. We include an indicator variable COM for company initiated restatements to test whether such restatements are more likely to trigger insider selling, especially in earlier periods, as insiders are more likely to possess information about the company s own investigation. Seyhun (1986) shows that insider selling is positively associated with past returns and negatively associated with future returns (also Rozeff and Zaman, 1998; Lakonishok and Lee, 2001; and Piotroski and Roulstone, 2005). We define past return for Quarter Q s insider trading activities as the cumulative return from the start of Quarter Q to the last insider trade in the quarter minus the equal-weighted market return for the same period. It is denoted as PRERET. 13 We define future return for Quarter Q s insider trading activities as the cumulative return for a 91-calendar day period beginning on the last insider trade in Quarter Q minus the equal-weighted 13 We also measure PRERET over a 91-calendar day period that ends on the last insider trading day during a specific quarter. The results based on this measure are qualitatively similar. 17

19 market return for the same period. 14 It is denoted as POSTRET. See Figure 1 for a better understanding of the definitions for PRERET and POSTRET. These two variables control for the magnitude of a normal level of insider trading that is not necessarily related to any particular event. Rozeff and Zaman (1988) and Ke, Huddart and Petroni (2003) show that insider selling is positively related to firm size. We thus include the logarithm transformation of market capitalization LCAP at the end of the year during which the restatement announcement is made to control for the size effect. We expect size LCAP to have a positive effect on net insider selling. Following Rozeff and Zaman (1998) and Ke, Huddart and Petroni (2003), we also include bookto-market ratio BM measured at the same time as LCAP and expect it to have a negative effect on net insider selling. 4. Empirical Results 4.1 DESCRIPTIVE STATISTICS Table 2 presents descriptive statistics of net insider selling for the thirteen quarters around restatement announcements. In Panel A, we report means and medians of the two measures of net insider selling: unscaled and scaled measures based on shares traded, for the full sample. In general, all means and medians indicate net insider selling as opposed to net insider buying (i.e., NIS > and SNIS > 0) for all quarters. More restating firms have non-zero net trading in quarters prior to the announcements than quarters after. From Quarter -8 to Quarter -2, the number of firms with non-zero net trading ranges from 630 to 666 and from 53% to 57% as a percentage of 14 Seyhun (1986) shows that most of the abnormal returns related to insider trading occur within 100 days after the trading. 18

20 all restating firms included in the Thomson Financial database for a particular quarter. In contrast, starting from Quarter -1, insider trading activities slow down, with the number of firms with nonzero net trading ranging from 514 to 553 and from 46% to 48% as a percentage of all restating firms included in the Thomson Financial database. 15 These numbers indicate that trading is heavier before than after the restatement announcements, consistent with the notion that trading on private information is possibly lucrative before the information is released to the public. While the frequency of insider trading appears to be higher before the restatement announcements than after the announcements, it is not obvious whether net selling is necessarily higher before the announcements than after the announcements. In fact, Quarters 2, 3 and 4 have the highest level of scaled net insider selling activities. Panels B and C of Table 2 report net insider selling before and after the passage of the Sarbanes-Oxley Act, respectively. We use the end of June 2002 as the cutoff for classifying pre- Sarbanes-Oxley and post-sarbanes-oxley insider trading quarters. 16 Specifically, if an insider trading quarter as defined ends before the end of June 2002, we include it in the pre-sarbanes- Oxley subsample and if an insider trading quarter starts after the end of June 2002, we include it in the post-sarbanes-oxley subsample. In Panel B for the pre-sarbanes-oxley subsample, unscaled net insider selling is generally higher in quarters prior to the announcements than that in quarters after the announcements. As for scaled net insider selling, it is relatively high during Quarters -8 to -4 (mean ranging from to 0.220). It slows down in Quarters -3 to -1 (mean ranging from to 0.162) and drops further in Quarters 0 to 4 (mean ranging from to 15 Note that the number of firms with non-missing RACAR included in the insider trading database varies across different quarters. 16 The Sarbanes-Oxley Act of 2002 was signed into law by President George W. Bush on July 30, 2002 and became effective on August 29. We use the end of June 2002 as the cutoff for classifying pre- and post-sarbanes Oxley insider trading quarters because by the end of June, the passage of the Act became relatively certain and insiders may have started altering their trading behavior accordingly. For sensitivity analysis, we also use the end of July and the end of August 2002 respectively as the cutoff for classifying pre- and post-sarbanes-oxley insider trading quarters and obtain qualitatively similar descriptive and regression results. 19

21 0.099). In Panel C for the post-sarbanes-oxley subsample, the scaled net insider selling is relatively low in Quarters -8 through -6 (mean ranging from to 0.106). While net insider selling is high in Quarters -5 to -1 before the announcements (mean ranging from to 0.447), it is similarly high in Quarters 1 to 4 after the announcements (mean ranging from to 0.467). In a comparison of insider trading activities between the two subsamples, we find that in general, a smaller percentage of restating firms engage in insider trading post-sarbanes-oxley than pre-sarbanes-oxley, especially in quarters prior to the announcements. In terms of the scaled and unscaled net insider selling, both are lower in Quarters -8 through -6 post-sarbanes- Oxley than pre-sarbanes-oxley. However, starting from Quarter -5, net insider selling is generally higher post-sarbanes-oxley than pre-sarbanes-oxley, with both the economic and statistical differences largest for Quarters 1 through 4. Overall, it appears that insider trading activities pre-sarbanes-oxley and post-sarbanes-oxley are noticeably different. While there is a marked decline in net insider selling activities after the restatement announcements from before the restatement announcements pre-sarbanes-oxley, this decline is not obvious post-sarbanes- Oxley. In Table 3, we provide univariate tests of whether net insider selling is associated with market reactions to the restatement announcements. Specifically, for each quarter, we classify all firms with non-missing RACAR that are included in the Thomason Financial database into three equal-sized portfolios based on the value of RACAR. We then report the number of firms with non-zero net insider selling, and the mean and median of SNIS of these firms for each quarter and each portfolio. The Low portfolio includes firms with the most negative market reactions to the restatement announcements and the High portfolio includes firms with the highest market 20

22 reaction. We also report t-test (Wilcoxon test) for means (medians) for the difference in net insider selling between the High and Low portfolios. In general, there is some evidence that insiders net selling activities prior to the restatement announcements (Quarters -8 to -2) are negatively correlated with abnormal returns around the announcements. In Panel A based on the full sample, net insider selling for the Low RACAR portfolio is higher than that for the High RACAR portfolio. However, it is only significant in Quarter -3. There is no difference in net insider selling between Low RACAR and High RACAR portfolios in Quarter -1 right before the announcements. After the restatement announcements, net insider selling for the Low RACAR portfolio is significantly lower than that for the High RACAR portfolio in Quarters 0, 1, 2 and 4. In Panel B based on the pre-sarbanes-oxley subsample, net insider selling for the Low RACAR portfolio is significantly higher than that for the High RACAR portfolio in Quarters -5 and -3 based on the t-test (in Quarters -7, -5 and -3 based on the Wilcoxon test) prior to the restatement announcements. There is no difference in net insider selling between Low RACAR and High RACAR portfolios in Quarter -1 right before the announcements. After the announcements, net insider selling for the Low RACAR portfolio is significantly lower than that for the High RACAR portfolio in Quarters 0, 1, 2 and 4 based on the t-test (in Quarters 0, 1 and 4 based on the Wilcoxon test). In Panel C based on the post-sarbanes-oxley subsample, net insider selling for the Low RACAR portfolio is significantly higher than that for the High RACAR portfolio in Quarter -8 prior to the restatement announcements. There is no difference in net insider selling between Low RACAR and High RACAR portfolios in Quarter -1 right before the announcements. After the announcements, net insider selling for the Low RACAR portfolio is significantly lower than that for the High RACAR portfolio in Quarters 0 and 1. 21

23 To summarize, results in Table 3 suggest some evidence of net insider selling based on restatement announcement abnormal returns prior to the announcements (supporting Hypothesis 1); no association between net insider selling and restatement announcement abnormal returns immediately before the announcements (supporting Hypothesis 2); and net insider buying based on restatement announcement abnormal returns after the announcements. There is limited difference qualitatively between the pre-sarbanes-oxley subsample and the post-sarbanes-oxley subsample. 4.2 MAIN REGRESSION ANALYSIS We focus on the relation between scaled net insider selling SNIS and restatement announcement abnormal return RACAR in the regression analysis. All regression t-statistics are based on White s (1980) heteroskedasticity consistent variance-covariance matrix. Panel A, Table 4 presents results of regression analysis of Equation (1) based on the full sample. The coefficients on RACAR from Quarters -8 to 4 are also depicted in Figure 2. From Quarter -8 to Quarter -2 before the restatement announcement, the coefficient on RACAR is always significantly negative, ranging from (t = -3.69) in Quarter -3 to (t = -2.21) in Quarters -2. This suggests that for these seven quarters before the restatement announcements, insiders sell shares when they expect that the restatement announcement will trigger a drop in firm value. This result supports Hypothesis 1 that net insider selling is negatively related to restatement announcement abnormal returns. During Quarters -1 and 0, the coefficients on RACAR are insignificant, suggesting that insiders cease trading activities that are related to restatement announcement abnormal returns. This result is consistent with Ke, Huddart and Petroni (2003), who find little evidence of a higher 22

24 frequency of insider sales in the two quarters immediately preceding the announcement of a break in an earnings increase string. It is also consistent with Noe (1999) and Huddart, Ke and Shi (2006). Noe (1999) shows that insiders are careful not to make insider transactions immediately before management earnings forecasts to avoid being seen as profiting from a piece of news before it is made public. Huddart, Ke and Shi (2006) find scanty evidence of an association between insider trades and announcement returns in a short window before the earnings announcements. The above timing pattern of insider trading supports Hypothesis 2 that insiders trade well in advance of restatement announcements and that they avoid trading very close to the announcements. This trading pattern potentially minimizes the risk of insider trading allegations and/or violating corporate insider trading policies. In summary, the above directional magnitude and timing pattern of insider trading activities provide strong evidence of insiders trading on privileged knowledge of firm-specific events, such as financial restatements. The coefficients on RACAR are positive for Quarters 1 to 4 relative to the restatement announcement and are significant for Quarters 1, 2 and 4. This result suggests that insiders reverse the direction of their trades after the restatement announcements. As discussed earlier, there are three possible explanations for this result. First, if the market over-reacts to the restatement announcements, insiders may reap further gains by exploiting this over-reaction. This conjecture is consistent with Seyhun (1990), who shows increases in insider purchases after the stock market crash of 1987 as evidence of overreaction to the crash. It should be noted that securities laws prohibit round trip insider trading within a six-month period (Section 16b of the Securities Exchange Act of 1934). However, insiders can, for example in Quarter 1, reverse trades made in Quarter -2 and earlier, which is consistent with our evidence that insiders 23

25 informed selling occurs in Quarter -2 and prior relative to the announcements while their buying starts from Quarter 1. Second, insiders sell restating firms shares just temporarily to avoid possible losses due to the restatement announcements and then they rebalance their portfolios by buying back restating firms shares after the restatement announcements. Third, the reversal may be a manifestation of passive insider trading, i.e., the delaying of purchases until after the negative news is released to the public (Huddart, Ke and Shi, 2006). As for the three major characteristics of restatements, RM is positively associated with net insider selling in Quarter -8. SEC is negatively associated with net insider trading in Quarters -1, 0 and 4. The coefficients on COM are never significant. Following Seyhun (1986), we use PRERET and POSTRET to control for the portion of normal insider trading that is not necessarily due to any particular event. Consistent with Seyhun (1986), the coefficient on PRERET is always positive and significant and the coefficient on POSTRET is always negative and significant. These results suggest that insiders tend to sell shares when the stock is doing well in the past and their selling predicts a drop in stock value in the future (Piotroski and Roulstone, 2005). The coefficient on LCAP is positive and significant in all quarters, suggesting that insider selling tends to occur in large firms, consistent with Ke, Huddart and Petroni (2003). The coefficient on BM is negative and significant in Quarters -5, -3 and 3, consistent with Ke, Huddart and Petroni (2003). It is insignificant in all other quarters. 4.3 EFFECT OF THE SARBANES-OXLEY ACT OF 2002 To determine whether the Sarbanes-Oxley Act of 2002 affects insider trading patterns around the financial restatement announcements, we run regression Equation (1) for the pre-sarbanes- 24