JEOPARDY, NON-PUBLIC INFORMATION, AND INSIDER TRADING AROUND SEC 10-K AND 10-Q FILINGS

Transcription

1 JEOPARDY, NON-PUBLIC INFORMATION, AND INSIDER TRADING AROUND SEC 10-K AND 10-Q FILINGS Steven Huddart, Pennsylvania State University Bin Ke, Pennsylvania State University and Charles Shi, University of California, Irvine Forthcoming, Journal of Accounting & Economics Evidence contrasting U.S. insider trades in high- and low-jeopardy periods and across firms at high and low risk for 10b-5 litigation indicates that insiders condition their trades on foreknowledge of price-relevant public disclosures, but avoid profitable trades when the jeopardy associated with such trades is high, such as immediately before earnings announcements. Insiders avoid profitable trades before quarterly earnings are announced and sell (buy) after good (bad) news earnings announcements. Insiders trade most heavily after earnings announcements and profit from foreknowledge of price-relevant information in the forthcoming Form 10-K or 10-Q filing. JEL Classification: J33 K22 M12 Keywords: accounting standards, government regulation, insider trading, litigation risk, stock-based compensation this draft: June 18, 2006 Seminar participants at the 2005 Canadian Academic Accounting Association meetings, the 2005 Hong Kong University of Science and Technology summer symposium, Monash University, the Pennsylvania State University, Rice University, the University of California Berkeley, the University of Hong Kong, the University of Queensland, and Yale University provided many useful comments. We thank François Brochet, Sandra Chamberlain, Thomas Lys (the editor), Gans Narayanamoorthy, Karen Nelson, Adam Pritchard, Alan Ramsay, Louis-Philippe Sirois, Jake Thomas, Mark Vargus, Serena Shuo Wu, and an anonymous referee for many useful comments. Karen Hennes and Santhosh Ramalingegowda provided able research assistance. We especially thank Karen Nelson for sharing data on litigation risk with us. Part of this research was completed while Steven Huddart was a visiting fellow at the University of Queensland. Send correspondence to: Steven Huddart Smeal College of Business Pennsylvania State University Box 1912 University Park, PA telephone: facsimile: web: huddart@psu.edu

2 1. Introduction We examine and contrast how the frequency and value of insider trades in the United States are associated with two significant information releases that occur in every fiscal quarter: the earnings announcement, which is a summary measure of firm performance that is highly price-relevant, and the subsequent Form 10-K or 10-Q filing, which contains more detailed financial results and also represents price-relevant information. We replicate earlier findings of a weak association between insider trades shortly before the earnings announcement and the subsequent earnings announcement. 1 We provide new evidence that insider trades before earnings announcements are relatively infrequent and occur when the magnitude of the earnings announcement abnormal return is small. In sharp contrast, insiders trade relatively heavily after the earnings announcement and these trades are significantly associated with the stock s returns over narrow windows around both the forthcoming 10-K or 10-Q filing and the preceding earnings announcement. 2 Returns over these windows are proxies for the price-relevant information released to the market by the disclosure that occurs within the window. The pattern of associations is consistent with the interpretation that variation in jeopardy over a fiscal quarter (i.e., the combined risks of unfavorable publicity, civil liability, and criminal prosecution) restrains insiders seeking to profit from foreknowledge of corporate disclosures more before earnings announcements than before 10-K and 10-Q filings. Buttressing this interpretation, we present evidence that variation in insider trading across firms is associated with firm-specific variation in the ex ante risk of 10b-5 litigation. Collectively, these findings are consistent with insiders conditioning their trades on foreknowledge of price-relevant public disclosures but limiting their trades to periods when the jeopardy they face due to trade is low. 1 The results of studies that have examined the relationship between insider trades and corporate news released within the next three months are mixed. Givoly and Palmon (1985) find little association between insider trades and subsequent Wall Street Journal reports, a sample composed mainly of earnings announcements. Sivakumar and Waymire (1994) find trading by insiders in one quarter is not correlated with errors in analysts forecasts of next quarter s earnings. Noe (1999) finds that increases in insider trades in the twenty days prior to disclosure of management forecasts are not correlated with management earnings forecast errors. Other studies, however, find evidence of an association between insider trades and the next earnings announcement. Lustgarten and Mande (1995) present evidence that insiders purchase undervalued stocks (but no evidence that insiders sell overvalued stocks) in the 30 days before earnings announcements. Roulstone (2004) documents a statistically significant but economically small association between insider trades in the two months before an earnings announcement and the abnormal return at the earnings announcement. 2 Insiders trades after the 10-K or 10-Q filing are significantly associated with the stock s returns over narrow windows around both the preceding announcement and filing. 1

3 Since it does not seem possible to elicit from insiders directly and unbiasedly the motives behind their trades, evidence on the litigation avoidance hypothesis necessarily is circumstantial. This paper offers evidence drawn from two related settings that plausibly differ in the seriousness of the jeopardy. Trade that shortly precedes and is conditioned upon a forthcoming earnings announcement has resulted in prosecutions and so is risky. 3 Legal advice that insiders should avoid trading in the period immediately before an earnings announcement is widespread. Many corporations stipulate blackout periods that cover the preannouncement period and during which insiders may not trade. In contrast, we know of no complaint filed against an insider for trading improperly on foreknowledge of the contents of a 10-K or 10-Q. 4 Relatedly, Bettis, Coles, and Lemmon (2000) document that blackout periods typically end on the second trading day after the earnings announcement, so corporate policy typically permits trades in this period. This suggests that the risk stemming from trade after the earnings announcement are lower than that risks from trade before the announcement. According to Kelson and Allen (2004, p. 13): A well-designed insider trading policy that is properly followed creates an effective prophylactic against inadvertent insider trading, and provides a defense for a company s insiders against any allegation that such trading has occurred. Adoption and enforcement of a written insider trading policy also provide a method for the corporation to demonstrate that appropriate steps have been taken to prevent insider trading violations, and to assert a defense against controlling person liability for trades made by its insiders under Sections 20A, 20(a), and 21A of the Securities Exchange Act of 1934 (Exchange Act). They further point out that only a minority of companies keep trading windows closed (or blackout periods in place) until after the filing of their 10-Qs and 10-Ks. Neither legislation nor SEC rules require 3 A relevant case in this area is SEC v. Lipson, No. 97-CV-2661, 129 F. Supp. 2d No centralized database of securities law complaints exists these documents are filed in courthouses around the country so it is not feasible to conduct an exhaustive search. However, a variety of keyword searches uncovered no document alleging improper trade related to any of over 1,000 securities for which documents are available at the Securities Class Action Clearinghouse ( Furthermore, a keyword search of the Lexis/Nexis file SEC Litigation releases, Administrative Releases, and AAERs using the search string insider trading w/50 10-K or 10-Q yields 21 documents, none of which relate to an allegation of improper trade shortly before the filing of a 10-K or 10-K. This suggests the SEC has never pursued an insider for trading on foreknowledge of information contained in a 10-K or 10-Q. 2

4 firms to restrict insider trades to particular periods or circumstances. Thus, firm policies that prohibit or discourage trade at specific times are an endogenous and voluntary response by firms (and their managers and shareholders) to the more fundamental risks that insiders and the corporation face. We term these risks jeopardy. Jeopardy arises from the formal surveillance and policing activities of the exchanges and the SEC; the resulting enforcement actions and precedents; and the less formal disciplinary roles of the business press, who publicize certain insider trades, and the plaintiffs bar, who launch 10b-5 class action lawsuits. Given that the jeopardy an insider faces from trade varies over the fiscal quarter, and, in particular, jeopardy is higher before the earnings announcement than in the period between the announcement and the 10-K or 10-Q filing, it is interesting to ask whether insider trade clusters in low-jeopardy periods and whether such trades are profitable to insiders. Prior research has examined the connection between insider trading and a variety of information releases. 5 We believe our study is the first to examine insider trading around the filing of 10-Ks and 10-Qs. As such, it provides evidence on the specific nature of the private information related to proximate financial disclosures that insiders possess and use in making their trading decisions. The filing of a 10-K or 10-Q is an important and interesting informational event because (i) it occurs frequently and regularly, (ii) the magnitude of the stock price response is sometimes large for 5% of the observations, the abnormal return at the filing is less than 9.78%, and for 5% of the observations, it is more than 9.77% (as we explain in more detail below) and (iii) compared to informed trade before an earnings announcement, there are reasons to believe the jeopardy due to trade before the filing is lower. Johnson, Nelson, and Pritchard (2004) examine whether insider trading is a determinant of subsequent securities litigation. They find that plaintiffs lawyers filings and 5 Among these studies are examinations of insider trades around earnings forecasts (Penman, 1982 and Noe, 1999), news announcements in the Wall Street Journal (Givoly and Palmon, 1985), declarations of bankruptcy (Seyhun and Bradley, 1997), dividend initiations (John and Lang, 1991), seasoned equity offerings (Karpoff and Lee, 1991), stock repurchases (Lee, Mikkelson and Partch, 1992), and takeover bids (Seyhun, 1990). Hillier and Marshall (2002) document the abnormal returns associated with insider trading before and after the two-month long trading ban preceding earnings announcements that applies to companies listed in the United Kingdom. 3

5 allegations point to the level of insider stock sales as evidence of management s fraudulent intent. Our research question is different. Rather than asking if insider trading contributes to litigation, we ask how the combined threats of SEC scrutiny, litigation, and adverse publicity (i.e., jeopardy) may discipline or limit insider trading. We do this by contrasting insider trading decisions in firm- and time-specific situations where jeopardy is either high or low. Jagolinzer and Roulstone (2004) examine the evolution of the distribution of insider trades around earnings announcements in the years Over this 17-year period, they find that insider trades in the month after the earnings announcement, as a fraction of all insider trades, has increased. Their evidence also suggests a shift of insider trades to the period after the earnings announcement. The shift is more pronounced at firms that have certain characteristics: small market capitalizations, low analyst following, low institutional ownership, more volatile earnings surprises, and more volatile returns. Given a trend over time towards greater jeopardy for improper trade and assuming greater jeopardy at firms with those certain characteristics, their time-series analysis complements this study in identifying how jeopardy influences insiders trades. Corporate insiders profit from foreknowledge of price-relevant information disclosures by selling before the disclosure of bad news or after the disclosure of good news. Some argue that instead of altering the time of trade to profit from information releases, insiders may alter the time of the disclosure or the content of the disclosure. 6 Along these lines, Beneish, Press, and Vargus (2001) examine whether earnings management precedes or follows insiders trades. In our setting, the time of disclosure of the 10-K or 10-Q is fixed within narrow limits by regulation. Further, many firms voluntarily disclose the date on which they plan to announce earnings. The evidence in Bagnoli, Kross, and Watts (2002) is that, by and large, firms do announce earnings on the planned-anddisclosed date. 7 This suggests that earnings announcement dates are known in advance 6 Aboody and Kasznik (2000) present evidence that CEOs adjust the timing of voluntary disclosures around fixed stock option award dates, so as to expedite disclosure of bad news and delay disclosure good news, thereby increasing the value of options granted at the money. Bartov and Mohanram (2004) present evidence that is consistent with executives reporting inflated earnings prior to option exercises, followed by a pattern of lower-than-expected earnings after exercise, which represents a reversal of the earlier inflation. 7 Delaying the announcement is associated with a significant price drop, which is personally costly to insiders holding long positions in the stock. 4

6 and sticky. Also, the scope to modify the content of the disclosure likely is limited by the facts that earnings have been reported after either a review by public accountants in the case of 10-Qs, or audited in the case of 10-Ks. As a consequence, the components of earnings have been fixed. Further, consistency over time of the principles used in the preparation of financial reports is required by reporting standards, filing is mandatory, and deadlines are prescribed, so the setting we examine is one in which insiders have discretion to choose the time and quantity of trade, but are limited in their ability to alter the content or timing of their disclosures. For investors, our findings indicate that the informativeness of insider trades depends, in part, on when during the fiscal quarter the trade takes place and characteristics of the firm including the risk of litigation that the firm faces. The link between jeopardy and the timing of insider trades may interest jurists studying how individuals actions change in response to statute, case law, and regulation. These findings also have implications for regulatory choices. For regulators seeking to limit the information rents gathered by informed insider traders, an important question is how to balance the trading needs of top executives who receive substantial stock-based compensation against the profitable trading opportunities created when insiders have private information and substantial trading discretion. We provide a measure of insider trading profits attributable to foreknowledge of the contents of 10-K and 10-Q filings. The paper is organized as follows. Section 2 describes key features of the setting and the data. Section 3 describes our empirical methods and presents our analysis of the distribution of insider trades over the fiscal quarter. Section 4 documents how insider trades vary across firms in response to the risk of litigation the firm faces. Section 5 concludes. 2. Data and institutional background In this section, we describe the data, the strategies insiders may use to profit from private information about corporate disclosures, and the timeline relating the disclosures to the trading periods we examine in each firm-quarter. Next, we define active and passive trading, describe the construction of the variables used in the analysis, and present descriptive statistics on these variables. 5

7 2.1 Data The data used in this study come from three sources. The 10-K and 10-Q filing dates are from the SEC s Edgar online database. Stock returns and financial information come from the CRSP and Compustat files. Insider trading data are from First Call/Thomson Financial Insider Research Services Historical Files. These data are the transactions of persons subject to the disclosure requirements of Section 16(a) of the Securities and Exchange Act of 1934 reported on Forms 4 and 5. Because our objective is to examine trading by insiders motivated by foreknowledge of company disclosures, the transactions included in the study are limited to open market purchases and sales by officers and directors. Non-open market transactions, including, e.g., option grants and exercises, transactions related to dividend reinvestment plans, stock transfers between spouses, and certain pension and other benefit program transactions are excluded. Also, transactions that are reportable solely because the transacting entity is a large shareholder are excluded. Since the requirement to file 10-Ks or 10-Qs through Edgar online became effective on January 1, 1996, our sample includes only calendar years The SEC requires that 10-Qs be filed within 45 days after the fiscal quarter s end and 10-Ks be filed within 90 days after the fiscal year end. 8 In our sample, 95% of the 10-Q (10-K) filings are made within 47 days (92 days) of the fiscal quarter s end (fiscal year s end). Our research questions are best addressed by studying situations where the timing of disclosures is fixed in advance and the content of disclosures is privately known to insiders. Accordingly, we exclude instances where announcements and filings are greatly delayed. Specifically, we require (i) the quarterly earnings reporting date to be no later than 45 days after the fiscal quarter s end, (ii) the 10-Q filing date to be no later than 47 days after the fiscal quarter s end, (iii) the annual earnings reporting date to be no later than 90 days after the fiscal quarter s end, (iv) the 10-K filing date to be no later than 92 days after the fiscal quarter s end. We impose these data restrictions because a delayed filing or announcement likely indicates that the disclosure is not finalized 8 For most companies, the deadlines for filing 10-Qs and 10-Ks are being shortened gradually to 35 and 60 days, respectively, by SEC Release No (issued September 5, 2002 and effective November 15, 2002). The SEC subsequently proposed to delay by one year the implementation of this rule change. See SEC Release No (issued November 17, 2004 and effective December 23, 2004). 6

8 timely. Also, delays generally signal bad news, which Bagnoli et al. (2002) show leaks out in advance of the announcement. Finally, delays may indicate that the disclosure is contentious or lacks representational faithfulness. Situations where either insiders do not know the content of the disclosure, insiders late-adjust the content of the disclosure, the content of the disclosure is leaked, or the disclosure is misleading are inconsistent with the experimental setting we seek to explore. 9 The final sample contains 110,305 firm-quarter observations, representing 7,791 unique firms. We count as an observation each of the three periods of the firm-quarters meeting the data requirements, including quarters in which there are no insider trades. The sample sizes for some regression specifications are smaller due to missing values for certain independent variables. 2.2 Opportunity to profit from foreknowledge of disclosures Consider how an insider may profit from short-lived private information about the contents of a forthcoming public disclosure. When the disclosure contains good (bad) news the insider ought to buy (sell) before the disclosure. Further, when the insider must sell stock for such reasons as personal liquidity needs, but he nevertheless has some discretion over when to trade, the insider benefits from postponing the sale until after good news is released. Likewise, when the insider must purchase stock for such reasons as achieving a stock ownership target established by his compensation contract, the insider benefits by postponing the purchase until after bad news is released. The insider s opportunity to profit from the price impact of a public disclosure ends when the disclosure is made, since at that point the insider has either traded or postponed his intended trade. However, if the insider strategically delays purchases (sales) until after bad (good) news is released, then trades will be correlated with past stock returns. Our tests for an association between insider trades before and after the earnings announcement and 10-Q and 10-Q filings with the short-window stock returns around those events are tests of whether and how insiders exploit specific pieces of short-lived private information, namely foreknowledge of the contents of the announcements and filings. 9 Note, however, that qualitatively similar results obtain if we relax the data requirements so that the 10-Q is filed no later than 60 days after the quarter s end and the 10-K is filed no more than 180 days after year s end. 7

9 Assuming that insiders know in advance the contents of the disclosure and can predict the price reaction that will occur when the disclosure is made, larger absolute returns imply a larger potential profit to insiders from conditioning their trades on the disclosure. 10 To assess whether the profit opportunity before the announcement is comparable to the profit opportunity before the filing, Table 1 presents summary statistics on the distributions of the raw, market-adjusted, and the absolute value of market-adjusted stock returns around the earnings announcement and the subsequent SEC filing. [Table 1] We choose the 10-Q and 10-K event windows to be brief periods of concentrated price reaction associated with these disclosures. In a study of 10-K and 10-Q reports filed electronically with the SEC using the EDGAR system, Griffin (2003, p. 435) observes that the EDGAR document will normally reside in the public domain at zero or low cost within one or two business days following the filing date. Using the absolute excess stock return as a measure of investor response to the filing, he finds that the response is concentrated on the day of and the two days after the filing date. Accordingly, we define RET FD to be the return over days 0 to +2 relative to the filing and call these days the Filing Window. We choose an earnings announcement window that is the same length as the filing window, namely 3 trading days. Since Morse (1981) finds that the price response to earnings announcements (measured as the median absolute return residual) is largest on days 1 to +1 relative to the announcement, accordingly we define RET EA to be the return over days 1 to +1 relative to the announcement and call these days the Announcement Window. ARET EA and ARET FD are the corresponding abnormal returns computed as the difference between the buy-and-hold raw returns and buy-and-hold value-weighted market index. Not surprisingly, the mean and median values of these returns are near zero. 10 This seems a reasonable assumption since there is a long line of research documenting stock price reactions to the unexpected component of these announcements beginning with Ball and Brown (1968), who study price movements around earnings announcements; and including Balsam, Bartov, and Marquardt (2002), who study price movements around 10-Q filings. 8

10 The active component of an insider s trading gain from a stock purchase is the product of the value of the shares he buys and the subsequent abnormal return. Likewise, the trading loss an insider avoids from a stock sale equals the value of shares he sells times the subsequent abnormal return. To assess the frequency with which insiders are faced with the opportunity to receive a gain by buying or avoid a loss by selling, and to make potential gains comparable to potential losses, in Table 1 we also report the absolute value of returns. Comparing the magnitude of the trading opportunity insiders face before the announcement with the opportunity they face before the filing, observe that the absolute value of the market-adjusted return over the earnings announcement, abs(aret EA), has a mean of 6.40% and a median of 3.96%, while the absolute value of the return over days 0 to +2 relative to the 10-K or 10-Q filing, abs(aret FD), has a mean of 4.38% and a median of 2.80%. Comparing the ratio of either the means or medians, it is apparent that the average or typical profit opportunity at the filing is about 70% of the profit opportunity at the announcement. The smaller magnitude of the return at the filing implies that, all else equal, insiders incentive to trade on foreknowledge of the filing is smaller than the incentive to trade on foreknowledge of the announcement. This makes it less likely that our test will detect an association between insider trades and the return at the filing than at the announcement. Nevertheless, large returns at the filing are frequent: For 5% of the observations, the abnormal return at the filing is less than 9.78%, and for 5% of the observations, it is more than 9.77%. The magnitude of these returns implies that insiders potential gains from well-timed trade are significant. It is useful to compare the profit opportunity at these events with the opportunities at other events. Jensen and Ruback (1983) point out that the abnormal returns of target firms subject to a merger or acquisition is more than 20% over the month leading up to the announcement. Case law establishes a clear obligation for insiders to avoid trade in this period. Seyhun and Bradley (1997) point out that the abnormal return experienced by a firm in the year leading up to a bankruptcy filing is about 50% and that insider selling during this period is abnormally high. Ke, Huddart, and Petroni (2003) document increased insider selling before breaks in a string of consecutive earnings increases; the average abnormal return over the 32 days before and including this event is 9

11 4.29%, which is comparable to the means of abs(aret EA) and abs(aret FD) measured over only three trading days. Thus, the announcements around which we study insider trades are associated with price movements that in many cases are as large as those associated with major corporate events like mergers, bankruptcy filings, and extreme earnings surprises. Therefore, these events should be large enough to prompt insiders to trade. Moreover, the events we study occur four times each year for every public company. Given a difference in jeopardy and economically significant profit opportunities from private information, a test of the litigation avoidance hypothesis is therefore possible by comparing the strength of the associations between (i) insider trades preceding the announcement with the announcement return and (ii) insider trades between the announcement and the filing with the filing return. 2.3 Timeline In our experimental design, we focus on insider trading in three non-overlapping periods within every fiscal quarter. The first period is the twenty calendar days ending two days before a quarterly earnings announcement. The second period, begins on the second trading day after the announcement date and ends on the earlier of (i) the 20th calendar day after this day and (ii) the calendar day before corresponding 10-K or 10-Q is filed. The third period is the twenty calendar days beginning on the third day after the 10-K or 10-Q filing date. Because the stock market reaction to earnings announcements typically occurs on the three days centered on the earnings announcement date while the market reaction to SEC filings occurs on the three days beginning on the filing date, we construct Periods 1, 2, and 3 so that they surround the Announcement and Filing Windows, but not overlap them. To facilitate the comparison of coefficient estimates across regressions based on observations from different periods, we construct the periods so that they are as close in length as possible. In the variable definitions introduced below, the index p {1,2,3} denotes the period. Figure 1 plots each of the periods on a timeline representing a typical firm-quarter. [Figure 1] 10

12 Because of variation in the time between the announcement date and filing date, Period 2 is less than 20 calendar days for some firm-quarters. 11 In other firm-quarters, the Filing Window begins more than 20 calendar days after the Announcement Window ends, so the last days before the filing are excluded from Period 2. As a robustness check, we repeated the analysis reported below after redefining Period 2 to include all the days between the end of Announcement Window and the beginning of the Filing Window. Results are qualitatively unchanged. 2.4 Active and passive trading If a trade is driven by the insider s desire to profit from a particular disclosure, the direction and magnitude of insider trades both before and after the event may be correlated with the price reaction to the disclosure because the insider may engage in both active and passive trading strategies. The definitions of active and passive trading below parallel those in Seyhun (1998, p. 50). Take first the case of stock sales. To profit from foreknowledge of a public announcement of bad news, an insider may trade actively: anticipating the stock price fall after the bad news is disclosed, the insider may sell before the announcement, receiving an active return from his action. Empirically, active returns are indicated by more-thanexpected sales transactions and more-than-expected sales value before bad news. To profit from foreknowledge of a public announcement of good news, an insider may also capture a passive profit: anticipating that the stock price will rise after the good news is disclosed, the insider may delay selling until after the announcement, receiving a passive return from postponing action. Empirically, passive returns are indicated by more-thanexpected sales transactions and more-than-expected sales value after good news. Correspondingly in the case of stock purchases, active returns are indicated by more-than-expected purchase transactions and higher-than-expected purchase value before good news, while passive returns are indicated by more-than-expected purchase transactions and more-than-expected purchase value after bad news. 11 Firm-quarters in which the filing date is 0, 1, or 2 trading days after the announcement date are excluded from the analysis. Period 2 does not exist in such cases because there is no interval between the Announcement Window and the Filing Window, which represent 12% of firm-quarters. This situation arises, e.g., if a firm does not formally announce its earnings in press release that precedes the filing of a 10-K or 10-Q. 11

13 Combining these observations, a positive association between net purchase transactions or net purchase value and future abnormal returns indicates active trading, while a negative association with past abnormal returns is consistent with passive trading. This interpretation is offered, e.g., by Seyhun (1986). It is possible, however, that the association between net purchase transactions and past returns is not driven by private information that prompts insiders to delay trading. Instead, it may be that insiders make contrarian trades in response to recent stock returns. Lakonishok and Lee (2001) and Cheng and Lo (2005) present evidence that insiders buy (sell) when the abnormal stock return is negative (positive) over the previous 3 to 6 months. In examining the relation between insider trade and abnormal returns in narrow windows surrounding the announcement and filing, we include as a regressor the stock s return over a six-month window preceding these disclosures. Despite this, our tests do not allow us to rule out either contrarian tendencies or private information as drivers of the association between insider trades and abnormal returns at prior disclosures. 2.5 Definition of test variables The principal dependent variables in our analyses are the signed frequency and the signed value of net insider trading in the three specified periods of each firm-quarter. So that the measures of trade reflect the net direction, frequency, and value of trade in a given period when some insiders may be selling while others are buying, trade frequency and trade value in a firm-quarter-period are defined as follows: FREQp fq is the number of purchase transactions minus the number of sale transactions in period p at firm f in quarter q. VALUEp fq is the value, in millions of dollars, of purchase transactions minus the value of sale transactions in Period p at firm f in quarter q. 12 In over half the periods we examine, there are no insider trades. In these cases FREQp and VALUEp, are set to zero. In a smaller number of cases FREQp (respectively, VALUEp) is zero when the number (respectively, value) of purchase transactions in a firm-quarterperiod equals the number (respectively, value) of sales transactions. Overall, FREQp and VALUEp are zero for 91% of Period 1 observations, 71% of Period 2 observations, and 77% of Period 3 observations. 12 Results are similar if FREQp fq and VALUEp fq are scaled by the number of insiders at that firm who trade in that period. 12

14 [Table 2] Panel A of Table 2 presents descriptive statistics on the dependent variables. Because sales transactions exceed purchase transactions in number and value and because sales transactions are coded as negative values, the mean values of FREQp and VALUEp are negative. The large standard deviations of these variables are driven by the small number of quarters where insiders trade heavily. For instance, the smallest and largest values of FREQ2 are 986 and 218, meaning that in one firm-quarter there are 986 stock sale transactions (net of stock purchase transactions) in that firm-quarter during Period 2 while in another there are 218 stock purchase transactions (net of stock sale transactions) in that firm-quarter during Period 2. The 1st and 99th percentiles of these FREQ2 are only 16 and 7. Similarly, the smallest and largest values of VALUE2 are far from zero, although the 1st and 99th percentiles of these variables, and 0.400, respectively, are much closer to zero. The distributions of both FREQp and VALUEp are highly leptokurtic. Because of the potential for extreme values of the dependent variable to be highly influential in the regression analysis, in the reported regression results we eliminate outliers using Cook s (1977) distance statistic. As well, in untabulated supplemental analyses, we perform rank regressions. The results of the rank regressions are qualitatively similar to the reported results. It is also useful to consider how much trade takes place in the three 20-calendar-day periods we consider and what fraction of all trades takes place in these periods rather than at other times in the quarter. Our analysis is based on 406,357 trades that take place over 110,305 firm-quarters, which implies that the average number of trades, both purchases and sales, is 3.68 trades per firm-quarter. The percentages of these trades that occur in Periods 1, 2, and 3, respectively, are 7.8%, 37.1%, and 24.6%, for a total of 69.5%. Thus, about two-thirds of trades occur in the three periods we examine, which represent 60 of the 91 calendar days in a typical quarter. It is notable that insider trades are unevenly distributed across periods. The number of trades in Period 2 is more than four times the number of trades in Period 1. 13

15 2.6 Definition of other independent variables Other independent variables used in the basic regression are: PRIOR RETp fq, the buy-and-hold return for the six calendar months before the beginning of the period; MV fq, the market value of equity, in billions of dollars; and BM fq, the ratio of the book value to the market value of equity. MV and BM are computed as of the end of the fiscal quarter to which the announcement and filing relate. Prior returns are included because insiders tend to be contrarian (i.e., insider buying is greater after low stock returns and lower after high stock returns), as documented by Rozeff and Zaman (1998) and Lakonishok and Lee (2001). Based on this research, the coefficient estimate on PRIOR RETp is predicted to be negative in all periods. The firm s market value is included as a control because Seyhun (1986) reports that insider trading varies cross-sectionally with firm size. 13 Additionally, portfolio rebalancing for reasons unrelated to private information is often cited by executives as a reason for trading stock. For this reason, larger stock trades, both purchases and sales, are likely to be associated with larger insider stock and stock option positions. Hence, a positive relation between holdings and trade size would be expected. Insider stock and option holdings are available in machine-readable form on Execucomp for only a small number of firms. Since executive compensation and the value of executive stock and option holdings are strongly positively correlated with firm size, firm size could be used in place of insider stockholdings as a measure of non-information based motives for trade. To preserve sample size, we adopt this approach. Findings in prior research also suggest that the coefficient estimate on MV should be negative in all periods because insiders purchase less stock at large firms. For this reason, we expect that MV is negatively related to FREQp and VALUEp in each period. The book-to-market ratio controls for the effect documented by Rozeff and Zaman (1998) that insider buying climbs as stocks change from growth to value categories. Accordingly, the coefficient estimate on BM is predicted to be positive. Panel B of Table 2 presents descriptive statistics on the explanatory variables. The distributions of PRIOR RET1, PRIOR RET2, and PRIOR RET3 are very similar 13 Specifically, Seyhun (1986, Table 3) documents that the absolute value of stock trades by insiders (scaled by the value of all stock trades) is negatively related to firm size. 14

16 because they are computed over similar periods. Our data are drawn from the years From the beginning of this period until the third quarter of 2000, prices generally rose. In the following period, prices generally fell. Despite the retreat that began in 2000, from June 1995 to December 2002, the period over which prior returns are computed, the S&P 500 Index and the NAS/NMS Composite Index both rose by more than 40%, implying that the median 6-month returns of between 2% and 3% reported in Table 2 are consistent with overall market movements. The median observation has a market value of common equity at the beginning of the fiscal year of $254.2 million and the mean MV is nearly 10 times greater. Because MV is highly skewed, we report regression results for ln(mv), which is the natural logarithm of MV. The book-to-market ratio, BM, over sample firm-quarters has mean and median values of and , respectively. 3. Analysis of trade over time Finding that the net number of insider purchases, FREQp, is positively correlated with the abnormal return associated with a forthcoming announcement means that insiders are more likely to buy stock before good news is released and more likely to sell stock before bad news is released. For a given price reaction (i.e., abnormal return) to an announcement by a firm, insiders profits are proportional to the value of stock traded, VALUEp. Finding that the value of stock traded increases in the magnitude of the abnormal return at the announcement means the value of insider trades and, hence, insider trading profits increase when their information advantage (as proxied by the abnormal return) is greater. 14 To test for these associations, we regress FREQp and VALUEp on the information content of the announcements and the filings, as proxied by the abnormal returns at these events, and the other independent variables: FREQp fq = β 0 + β 1 ARET FD fq + β 2 ARET EA fq + β 3 PRIOR RETp fq + β 4 ln(mv fq ) + β 5 BM fq + ǫ fq and (1) VALUEp fq = β 0 + β 1 ARET FD fq + β 2 ARET EA fq + β 3 PRIOR RETp fq + β 4 ln(mv fq ) + β 5 BM fq + ǫ fq. (2) 14 Relative to the value measure, the signed frequency measure equally weights the insider s trading decisions. If high-value insider trades are more likely to be motivated by liquidity needs or a desire to diversify or are subject to greater scrutiny by regulators (as is suggested by Seyhun, 1998, p. 75), then an equal-weighting of trades may better capture the informed component of insider trade. 15

17 Separate regressions are run for each of the periods, p {1,2,3}. These regressions pool data across firms and quarters. The chief interests in these regressions are the coefficient estimates on ARET FD and ARET EA, i.e., β 1 and β 2. Regressions control for firm, calendar year quarter, and fiscal quarter fixed effects. Periods 1 and 3 are 20 days long in every case. Because some firms file 10-Ks or 10-Qs shortly after announcing earnings, Period 2 is less than 20 days in some cases. 15 To control for the possibility that either FREQ2 or VALUE2 is correlated with the number of days in period 2 (e.g., when period 2 is less than 20 days there might be fewer insider trades simply because there are fewer days on which to trade), the set of explanatory variables for period 2 is augmented with indicator variables (coefficients not reported). 16 We analyze interim and fourth quarter observations separately because it is possible that the relation between insider trade and the abnormal returns at the announcement and the filing may differ across quarters for at least four reasons. First, the time between an interim earnings announcements and the filing of the 10-Q is shorter than the time between the fourth quarter earnings announcement and the filing of the 10-K: on average the elapsed days between the Announcement Window and the Filing Window for these events are 16 and 41 calendar days, respectively. 17 Second, Salomon and Stober (1994) find that the stock price response to earnings surprises is greater at interim earnings announcements that at fourth quarter announcements. Third, Griffin (2003) finds that the price response to 10-K filings is greater, on average, than the price response to 10-Q filings. Finally, for both announcements and filings, two-sample Wilcoxon ranksum tests indicate significant differences (at better than the level) in the distributions of absolute abnormal returns for the fourth quarter compared to interim quarters. These differences in price reactions and time between information events could lead 15 Over the first three fiscal quarters of the firm-year, the median, mean and minimum number of calendar days between the Announcement Window and the Filing Window are 17, 16 and 1 days, respectively. For 25% of the observations, there are 11 or fewer days between the Announcement Window and the Filing Window. For the last fiscal quarter, the gap between the earnings announcement date and the 10-K filing date is longer: the median, mean and minimum number of days between the Announcement Window and the Filing Window are 43, 41 and 1 days, respectively. 16 Specifically, 19 indicator variables, Dn for n = 1,..., 19, are created. Let n be the length of period 2 in calendar days. For an observation for firm f in quarter q where n < 20, the indicator variable Dn fq is equal to 1; otherwise, it is zero. 17 Recall that for the years we study, the SEC requires 10-Qs to be filed no later than 45 days after the end of the quarter, while 10-Ks must be filed no later than 90 days after the end of the quarter. In the sample, 10-Qs, on average, are filed 43 days after the quarter s end, while 10-Ks are filed, on average, 84 days after the fiscal year-end. 16

18 insiders to adopt different trading strategies across interim and fourth quarter disclosures; however, the signs, magnitudes, and significance levels of coefficient estimates (reported below in Table 3) are quite consistent between the annual 10-K and three quarterly 10-Qs. 3.1 Predicted associations of trade with announcement and filing returns If trades are prompted by an insider s desire to actively exploit foreknowledge of the content of either the earnings announcement or the regulatory filing, then there should be a positive association between either the number or value of purchase transactions (net of the number or value of sales transactions) and the forthcoming information releases. If trades are prompted by an insider s desire to passively exploit foreknowledge of the content of either the earnings announcement or the regulatory filing, then there should be a negative association between either the number or value of purchase transactions (net of the number or value of sales transactions) and a prior information release. Thus, if insiders trade actively to profit from private information in a period, then the coefficient estimate on the abnormal returns at a disclosure (i.e., either the announcement of earnings, or the filing of a 10-K or 10-Q) that follows the period should be positive. If insiders trade passively to profit from private information in a period, then the coefficient estimate on the abnormal returns at a disclosure that precedes the period should be negative. Prior research reports mixed evidence on the association between insider trades before the earnings announcement and the announcement return. Accordingly we predict no association between the trade measures (i.e., FREQ and VALUE) and ARET EA in Period 1. Because jeopardy is high before the announcement, we further predict no association between the trade measures and ARET FD in Period 1. Given the lower jeopardy insiders face for active trade after the earnings announcement, we predict (i) positive associations between the trade measures and ARET FD in Period 2, (ii) negative associations between the trade measures and ARET EA in Period 2, and (iii) negative associations between the trade measures and both ARET EA and ARET FD in Period 3. 17

19 3.2 Active and passive trading within Periods 1, 2, and Main tests The first three columns of Table 3 lay out the predicted signs on the event return coefficient estimates that follow from the discussion above for each of the three periods. In Table 3, specifications (1a) (1c) correspond to Period 1, (2a) (2c) to Period 2, and (3a) (3c) to Period 3. Thus, the dependent variable is FREQ1 in Panel A and VALUE1 in Panel B in specifications (1a) (1c); it is FREQ2 in Panel A and VALUE2 in Panel B in specifications (2a) (2c); and the dependent variable is FREQ3 in Panel A and VALUE3 in Panel B in specifications (3a) (3c). In the (a) specifications, results are presented for all quarters pooled. In the (b) and (c) specifications, results are reported for the interim quarters and the fourth quarters separately. [Table 3] Regarding Period 1 (i.e., the period before the earnings announcement), Table 3 indicates that associations between insider trades and the forthcoming announcement and filing are insignificant, with the exception of a positive coefficient estimate on ARET EA in specifications (1a) and (1b) of Panel A, where the coefficient estimate is significantly positive at the 5% level using a two-tailed test. Thus, there is some evidence that insiders buy (sell) more often before good (bad) news earnings announcements in quarters 1 3, but no evidence of this in quarter 4 and no evidence that the value of shares traded varies with the abnormal return at the announcement. The lack of any significant association with ARET FD in Period 1 suggests that the value of insiders trades in Period 1 is not increased by the desire to profit from foreknowledge information in the filing not conveyed by the earnings announcement. 18 The results for Period 2 are sharply different. The pattern of insider trades in this period suggests that insiders use their private information to derive both active and passive profits. Consistent with the realization of active profits, insider trades, measured in Panel A by the signed frequency of trade, is positively associated at better than the 18 Coefficient estimates and standard errors are very similar when market-adjusted returns are replaced with raw returns in this and subsequent regressions. 18

20 5% level with the abnormal return of the forthcoming filing. Note from Panel A that the coefficient estimates for Period 2 on ARET FD in specifications (2a), (2b), and (2c) are more than 10 times the coefficient on ARET EA in specifications (1a), (1b), and (1c), which implies that for a given abnormal return at the disclosure, the effect on insider trades is more than 10 times greater in Period 2 compared to Period 1. The significantly positive coefficient estimate on ARET FD in Period 2 implies that insiders buy before filings interpreted by the market as good news and sell before filings interpreted as bad news. 19 We turn next to Panel B, where the dependent variable is the signed value of shares traded. When all quarters are pooled, the coefficient estimate on ARET FD is significantly positive at the 5% level using a two-tailed test. Thus, there is evidence that the value of shares purchased by insiders is higher before a good news filing (and lower before a bad news filing). When the observations for quarters 1 3 and quarter 4 are analyzed separately, the sign and magnitude of the coefficient estimates on ARET FD is similar, but the relationship is insignificant. Consistent with the realization of passive profits, insider trades, measured either by the signed frequency or signed value of trade, is negatively associated at the 1% level with the abnormal return at the preceding announcement. The significantly negative coefficient estimate on ARET EA in Period 2 is consistent with the notion that insiders sell after announcements interpreted by the market as good news and buy after announcements interpreted as bad news. 20 For Period 3, the coefficient estimates on the abnormal returns at the preceding filing and announcement both are significantly negative (with the exception of ARET EA for quarter 4 analyzed separately), which is consistent with trade in Period 3 being driven in part by insiders passive use of private information. Such an association with past news also could arise from a contrarian trading strategy under which insiders condition their trades on past stock price movements so that they buy after bad news events 19 The positive coefficient on ARET FD in Period 2 is not a result of insiders trading on the post earnings announcement drift because the coefficient on ARET FD is unaffected by the inclusion of the earnings surprise in the immediately preceding earnings announcement, defined as the seasonal difference in earnings per share scaled by stock price at the end of fiscal quarter 4 relative to the observation quarter (results not reported). 20 To alleviate the concern of cross-sectional dependence of insider trades, we also run a Fama-MacBeth regression of specification (2a) of Table 3 by calendar year and draw similar inferences. 19