Perceived accounting quality and the information content. of prior insider trades

Transcription

1 Perceived accounting quality and the information content of prior insider trades Terrence Blackburne University of Washington Asher Curtis University of Washington July 21, 2016 Anna Elsilä University of Oulu ABSTRACT: We examine how shocks to perceived accounting quality lead to market prices that reflect an increased weighting on prior disclosures of insider trading. Using a sample of firms where industry peers have restated financial information, we find that the short-term market reaction to announcements of restatements by industry peers is significantly more negative for companies with recent insider net selling than for companies with recent insider net buying. We also find that recent insider trading signals are more informative to investors in situations characterized by a greater information asymmetry and when insider trading on private information is more likely. Our findings suggest that insider selling disclosures are an important source of information when perceived accounting quality declines and that a significant portion of the intra-industry information transfers arising from peer restatements is due to a reweighting of other information, rather than an increased discounting of accounting information. Keywords: Insider Trading; Restatements; Disclosure; Re-weighting. JEL Codes: M40; E03. Data Available: All data are available from the sources described in the text. We thank Conference Participants at George Washington University for helpful comments.

2 1. Introduction An important question in the accounting literature is whether the disclosure of insider trading activities is informative to market participants. On the one hand, insider trades are associated with both longer-term future market and financial performance (Ke et al. 2003; Piotroski and Roulstone 2005). On the other hand, the short-term market response to the disclosure of insider trades is limited to insider purchases (Brochet 2010). These findings represent an empirical puzzle; the evidence of an association between insider sales and long-term performance suggests that insider sales should be informative in the short-term. As such, a key challenge to answering whether the disclosure of insider trading activities is informative lies in the ability to identify why, on average, the disclosure of insider sales does not appear to elicit a compelling market response. The standard explanation for why studies do not find evidence of a short-term market response to the disclosure of insider sales is that insider sales are noisy signals. Indeed, a very general principle in disclosure theory is that, in the presence of multiple informative signals, Bayesian investors will weigh the signals according to their beliefs about the signals relative precision. 1 If market participants believe that a large portion of insiders sales are due to liquidity or diversification needs, and thus do not reveal private information about future cash flows, then they will assign them low weight when assessing the value of a firm (Lakonishok and Lee 2001). As such, a more precise signal of future cash flows can be derived from other disclosures, such as earnings reports and periodic filings. Thus, the absence of evidence for a short-term market response to insider sales may be due to low-power of the tests because of researchers inability to separate liquidity trades from trades that are driven by managers private information. The theory also suggests that insider trading disclosures will be more informative to investors when they believe that the signals provided by accounting disclosures are less precise, or in other words, when perceived accounting quality is low. 2 Another factor that limits researchers ability to provide compelling empirical evidence that insider sales are informative is the simultaneity of managers trading and disclosure choices. Subject to certain bounds, managers have flexibility to time their trades around the disclosure of other value relevant information (or vice versa), making insider trading activities and accounting disclosures endogenous. Prior empirical studies provide significant evidence that this simultaneity issue exists. Over 1 For example, see the discussion in Verrecchia (2001) and Veldkamp (2011). 2 Veenman (2012) uses cross-sectional variation in accounting quality to perform a test of the market response to insider trading disclosures consistent with this theory, but does not find evidence that disclosures of insider sales are useful investors. 1

3 long horizons, Ke et al. (2003) provide evidence that insider sales increase three to nine quarters prior to a break in consecutive quarterly earnings increases. Over shorter horizons, Huddart et al. (2007) provide evidence that managers delay trades until after earnings announcements when their exposure to legal penalties is reduced, Billings and Cedergren (2015) find that managers who engage in insider selling are likely to withhold forecasts of bad news, and Aboody et al. (2005) find that managers are able to earn higher abnormal profits from their trading activity when accounting quality is low. Furthermore, theory suggests that the voluntary nature of insider trades will tend to result in pooling equilibria whereby the information content of the trades may only be informative as a conditioning variable (Bagnoli and Khanna 1992). John and Lang (1991) and Badertscher et al. (2011) find evidence consistent with investors using insider sales to condition their response to other signals. We address the simultaneity concern by interpreting the restatement of a peer firm as a negative shock to perceived accounting quality for the non-restating peer. We use this setting to examine the association between the short-run returns of non-restating peer firms and prior insider trading disclosures. This setting is ideal for two reasons. First, several recent studies provide evidence that suggests that a restatement by an industry peer increases the uncertainty about the accounting quality of all firms in that industry (Xu et al. 2006; Kedia and Rajgopal 2011; Kravet and Shevlin 2010; Gleason et al. 2008). 3 As such, a peer restatement has a direct effect on the precision of accounting signals. Second, to the extent that insiders at non-restating peer firms are unable to anticipate the restatement of a peer firm, these events are plausibly exogenous with respect to their prior insider trading activity. We find evidence that investors incorporate information from both insider purchases and sales into prices around peer firm restatements in a manner consistent with disclosure theory. 4 Our results are consistent with our hypothesis that investors will reweight past signals about future cash flows when the perceived precision of a signal declines. Specifically, we hypothesize that investors will increase the weight of insider trading activities there is an exogenous decline in the precision of accounting disclosures about earnings. Our results extend prior literature as our empirical setting allows us to identify evidence of both a positive reaction to prior insider purchases and negative reaction to insider sales, by controlling for the simultaneity bias in the choices to disclose and trade as an insider. 3 More generally, there is a considerable literature following Firth (1996) and Foster (1981) consistent with information transfers within an industry. Specifically, the evidence suggests that when a company within an industry announces material news, investors incorporate it into the stock prices of other firms in the same industry. 4 In our empirical specifications, we include controls for both fixed effects for industry and year and possible confounding variables drawn from the prior literature. 2

4 We next examine whether investors reweighting of prior insider trading varies in the crosssection and time-series as predicted by our theory. We find that the weight investors place on insider trading activities following a peer-restatement is greater for smaller firms, consistent with these firms having a lower number of other information sources; prior to the Sarbanes-Oxley Act (SOX), consistent with a lower level of disclosure pre-sox; and for intentional misstatements in the industry, consistent with these being larger shocks to perceived accounting quality. Finally, we perform additional analysis to rule out alternative explanations for our results. Gleason et al. (2008) find that, conditional on a peer restating, firms in the same industry have an increased likelihood of restating. As such, an alternative explanation for our results is that investors use prior insider trades as a conditioning variable that provides information about the extent of contagion risk. We do not find any compelling evidence consistent with this contagion story, as in our sample prior insider trading activity is not associated with future restatements when conditioned on a peer restatement. Our paper extends Badertscher et al. (2011), who find evidence that investors condition their reaction to the announcement of a firm s own restatement on its past insider trading patterns. Specifically, Badertscher et al. (2011) document compelling evidence that the market reacts significantly more negatively to a firm s restatement when the firm s insiders have recently sold shares. Although Badertscher et al. (2011) undertake efforts to address reverse causality, a limitation of their setting is that the information content of insider trades might be limited to the restatement itself for example, past insider trading may be related to expected future damages that could result from litigation. 5 Our setting allows us to examine information content of insider trades more generally because the information from past trades is presumably unrelated to the peer firm restatement. We contribute to the literature along several dimensions. First, we provide evidence that investors perceive insider sales as having information content. Specifically, by incorporating an exogenous shock in our research design, we are able to identify an association between insider sales and returns. Second, our evidence suggests that a significant simultaneity bias in the disclosure of insider trades at least partially explains the failure to document a significant short-term market reaction to their disclosure. 5 Note that Badertscher et al. (2011) focus on a separate endogeneity issue than our paper, as they control for the simultaneity that arises due to investors receiving two endogenously related signals (insider trades and a restatement) at the same time. Badertscher et al. (2011) solves this problem by examining the insider trades that occurred prior to, but were not disclosed until after, the restatement was made public. 3

5 Thus, while prior evidence that the market regards insider sales as a signal of future cash flows is inconclusive, we find that in the presence of a shock to the perceived precision of accounting disclosures, insider trading appears informative to investors. Our approach could be extended to other disclosure settings that are simultaneously determined and correlated with other signals. Finally, our results also provide a theoretically consistent reason for why so-called stale disclosures appear to elicit market responses. Specifically, we find evidence consistent with Bayesian models where investors reweight multiple signals about future cash flows when there is an exogenous shock to the perceived precision of another signal. 2. Prior literature and hypotheses 2.1. Prior Research A long-standing view of financial economists is that insider trading reduces information asymmetry and improves the efficiency of financial markets and capital resource allocation. Manne (1966) argues that insider trading activity serves as an important mechanism through which corporate insiders communicate their expectations to investors, thereby accelerating price discovery and improving market efficiency. Prior tests of this assertion generally consider whether insiders have private information and test for evidence of this by examining whether insider trades predict future long-term stock returns (e.g., Seyhun 1992, 1986; Rozeff and Zaman 1988). Many of these studies investigate the determinants of returns earned by insiders and focus on whether insiders exploit private information in their trading decisions (e.g., Ravina and Sapienza 2010). While some insider trades may be motivated by a possession of material private information, this literature argues that insiders may trade in the open market to signal their own commitment and beliefs about firm prospects. Several studies explicitly test whether stock market participants react to filings of insider trading transactions. Early studies generally failed to find evidence of a market reaction around days when insiders trades are reported to the Securities and Exchange Commission (SEC) via a form 4 filing. For example, Lakonishok and Lee (2001) conclude that surprisingly, in spite of the extensive coverage the insiders activities receive, the market basically ignores this information when it is reported. Lakonishok and Lee (2001) suggest that insider sales may be perceived as less informative because they may be motivated by reasons unrelated to firm value, such as insiders liquidity or diversification needs. Subsequent studies, however, attribute the lack of a short-term market reaction to a long transaction reporting period, which was allowed before the adoption of SOX. These studies report significantly abnormal returns around filings of insider trades following the enactment of SOX, which shortened the 4

6 reporting period for insider trades (Brochet 2010; Veenman 2012). Nonetheless, empirical evidence of abnormal returns following form 4 filings tends to be observed only for insider purchases Hypotheses A very general principle in disclosure theory is that in the presence of multiple informative signals, Bayesian investors will place weight on signals according to their perceived precision. Disclosure theory suggests that, when information is released sequentially, rational investors will update their beliefs by incorporating the new information and revising the weights that they place on prior signals of firm value (Holthausen and Verrecchia 1988). Consider the following simplified Bayesian framework with two signals (Verrecchia 2001; Veldkamp 2011):,= (1) where, X is the expected cash-flows of the firm, is the aggregation of prior beliefs and accountingbased information, with precision. is the insider trading signal, with precision. We are interested in how the coefficient on insider trades,, responds to changes in beliefs about the precision of accounting-based information, y. Intuitively, if investors perceive accounting-based information with greater precision, then the informativeness of insider trading is low. In terms of Equation (1) the partial derivative of the coefficient on insider trading, <0, confirms this intuition. This suggests that if the perceived precision of accounting declines, the informativeness (i.e., the coefficient) of insider trading is expected to increase. The second partial derivative = is convex, which combined with the first derivative, suggests that the coefficient on insider trading is decreasing at a decreasing rate. Thus, for two firms with different perceived precisions of accounting-based information, the effect of a change in y on the coefficient on insider trading is greater for the firm with the lower level of prior precision. The endogeneity of disclosures and insider actions limits our ability to estimate a consistent coefficient on the short-term market reaction to the insider trading signal. Specifically, if managers delay insider transactions until after public disclosure of their private information then the market will likely have already incorporated it into prices. Alternatively, if managers delay significantly the disclosure of their private information until after their insider transactions then theory suggests that the endogeneity of their trading activity and actions would likely result in pooling equilibria, and thus attenuate the market response to their trades. To overcome this challenge, we require an exogenous source of 5

7 variation in the precision of the accounting signals, y, in order to identify investors updates to the coefficient on insider trading. If the prior insider trades are uninformative (i.e., have a very low precision, z) then the market response to a change in the precision of the accounting signals, y, will be insignificant. Prior literature documents that when one company makes corporate announcements, these announcements often contain information that is useful to investors of other companies in the same industry. Such announcements represent plausibly exogenous news about companies in the same industry as the announcing firm. One particular event, that prior research has examined is the announcement of accounting restatements. Gleason et al. (2008) document a negative price response to peer-firm restatements, but fail to find evidence that analysts update earnings forecasts following these events. They interpret this as evidence that the price response to peer-firm restatements is reflective of investors concerns about accounting quality and not due to news about the mean of future cash flows. These studies suggest that a restatement causes investors to revise their beliefs in the precision of peer firms accounting information. A shock to investors perceptions of accounting quality might be conditioned on previous insider trades through two possible channels. The first is via a re-weighting of various signals of firm value, as described above. The second channel is via updated probabilities of restatement risk, whereby the information content from the trades is limited to their usefulness as a conditioning variable. Therefore, our first hypothesis (H 1 ) is that companies with recent insider purchases will experience higher returns around restatement announcements by industry peers than companies with recent insider selling: H 1 : Returns of non-restating firms at the time of an industry restatement will be positively associated with prior insider trading activities. Specifically, we predict that for non-restating firms, insider selling activity will be associated with negative returns and insider buying activity will be associated with less negative or even positive returns at the time of an industry restatement. Returning to Equation (1) the effects of a peer restatement on the coefficient on prior insider trading varies with the precision of prior beliefs and accounting-based information, y. Using Equation (1) we make predictions relating to cross-sectional and time-series variation in the precision of prior beliefs and accounting-based information, y. First, we expect investors to place greater weight on the insider trading signals when their initial beliefs regarding the precision of accounting information is lower. 6

8 Because smaller companies are generally characterized as having greater information asymmetry and more limited information environments, we expect that insider trading activity will be more informative to investors for these companies. Second, we predict that investors perceive insider trading activity as a more credible signal of earnings quality is when the insiders face lower costs of engaging in earnings management and insider selling (Kedia et al. 2015). In such a situation insider trading combined with earnings management would be more widespread and hence insider trading will better reveal earnings quality and firm value to investors. One such setting is the pre-sox regulatory environment. SOX increased insiders penalties for financial fraud considerably. Among other things, it increased the personal exposure of CEOs and CFOs to liabilities for financial misrepresentation. Empirically, Cohen et al. (2008) document a decrease in accrual-based earnings management following the passage of SOX and Li and Zhang (2006) find evidence of constrained opportunistic trading in restatement announcing companies. In the same vein, the analysis presented in Thevenot (2012) suggests that managers perceive a higher cost of public enforcement in the post-enron period. We therefore expect that insider trading will be a more informative signal of earnings quality before the passage of SOX. Finally, we predict that the coefficient on recent insider trading will be particularly strong when a peer has a restatement due to an intentional misstatement. Prior research suggests that intentional restatements tend to be more severe than unintentional misstatements (e.g. (Palmrose et al. 2004; Hennes et al. 2008). Moreover, recent research suggests that there is earnings management contagion (Kedia et al. 2015). Therefore, the change in market participants beliefs about the precision of accounting information is likely to be larger for these types of restatements. In addition, there is some evidence that suggests managers sell more stock during misstated period (Agrawal and Cooper 2015). Therefore, if market participants believe that prior insider trading is predictive of future restatements then the reaction to intentional restatements will be more pronounced. Collectively, we write these three predictions as Hypothesis 2: H 2 : Returns of non-restating firms at the time of an industry restatement will be significantly more positively associated with prior insider trading activities for (a) smaller firms, (b) pre- SOX, and (c) intentional misstatements. 7

9 3. Empirical Analysis 3.1. Data and sample selection We collect information about accounting restatement announcements from Audit Analytics that occurred during and are not attributable to clerical errors. We collect financial data from Compustat, stock market data from the Center for Research on Security Prices (CRSP) and insider trading data from the Thomson Reuters Insider Filing database. In Table 1 we summarize our sample selection criteria for restatement and peer firms. We restrict our restatement sample based on five requirements. First, we retain only restatement announcements for firms that have non-missing CRSP data and CRSP share codes equal to 10 or 11. Second, we exclude restatements by companies in financial and utilities industries (that is, with GICS sector codes 40 and 55), due to different regulatory regimes for these industries. 6 Third, we exclude 301 observations related to restatement announcements that occurred on the same day or within two trading days after another restatement announcement by a firm in the same industry. Fourth, we follow Gleason et al. (2008) and Kravet and Shevlin (2010) and restrict the restatement announcement sample to materially negative announcements, which we identify as those that result in a three-day market-adjusted return over ( 1; +1) trading days around the announcements of less than 1 percent for restating firms. 7 Fifth, we retain restatement announcements for which there exist peer companies with non-missing control variables. These criteria result in a sample of 1,828 restatement announcements for 1,416 unique firms. For our sample of peer firms we use Compustat s historical eight-digit Global Industry Classification Standard (GICS) codes to classify firms into industries, as Bhojraj et al. (2003) find evidence that GICS industry classifications outperform other industry classifications. We require that peer firms be listed on a major stock exchange (NYSE, AMEX or Nasdaq), and following Gleason et al. (2008) exclude firms with share prices below $5 around the peer restatement announcement to remove the possible confounding effects of smaller illiquid firms. 8 Finally, we exclude firms that are not included in 6 In addition, differences in regulation can make the calculation of control variables (such as discretionary accruals and debt-to-asset ratio) misleading and not comparable with the rest of the sample. The tenor of our results is unchanged when retaining these industries, and using a smaller number of control variables in the regression analysis. 7 Gleason et al. (2008) review press releases related to restatement announcements associated with neutral or favorable market reaction, and find that such cases are often attributable to correction of minor accounting errors, earnings increasing restatements or concurrent announcements of other favorable news. Hence, such announcements contain mixed signals and may not trigger strong information transfers within the industry. 8 Considering all companies or using $2 as an alternative threshold does not quantitatively affect the results reported later. 8

10 the Thomson Reuters Insider Filing database. These restrictions result in a final sample that consists of 73,153 peer firm-event observations and 5,179 unique peer firms. 9 To measure insider trading activity in these peer companies we retain open market sale and purchase transactions (transaction codes P or S in Thomson Reuters) made by corporate insiders. Following Lakonishok and Lee (2001) and Frankel and Li (2004) we exclude insider trades without the Thomson Reuters cleanse indicator A or S, with a transaction price less than $2, with the number of shares traded less than 100, with a transaction price falling outside the range of CRSP transaction day prices (i.e., CRSP variables Bidlo and Askhi ), with the number of shares exceeding the CRSP total daily trading volume or shares outstanding, and those involving transactions of non-common shares (i.e., shares with CRSP codes other than 10 or 11). Finally, we restrict our measure of insider trades to those classified as executive insiders, because they are more likely to be actively and directly participating in corporate decision making Descriptive statistics We present descriptive statistics in Table 2 for the variables used to test our hypotheses. We winsorize continuous variables at the 1 and 99 percentiles to reduce the effect of outliers. Consistent with Gleason et al. (2008), we report that the stock returns of non-restating firms are negative around the announcement of a restatement by an industry peer. For example, the mean (median) 3-day market-adjusted cumulative abnormal return is -0.39% (-0.32%). One interpretation of this result is that we provide out-of-sample evidence of an intra-industry information transfer, or contagion, effect. We also report that intentional misstatements, based on the definition in Audit Analytics, make up slightly more than ten percent of our sample. Roughly eighty percent of our sample restatements occur after the implementation of SOX Preliminary Tests of Hypotheses We present the market reaction to peer-firm restatement announcements partitioned by the direction of pre-announcement insider trading in Table 3. In these tests we use several windows to 9 The sample has good representation across industries and over time. We find 122 industries in the sample, with the Information Technology GICS subindustry category having the most restatements during our sample period (see Table A2 ). We also find that the sample is well distributed over the calendar years in the sample, with a slight increase in 2005 and 2006 relative to the other years (see Table ), we include year fixed effects to remove this variation from our main analyses. 10 Specifically, we retain trades by insiders with Thomson Financial ROLECODE1 equal to CEO, CFO, CI, CO, CT, EVP, O, OB, OP, OS, OT, OX, P, S, SVP, VP. We discuss alternative insider classifications in our robustness analysis. 9

11 aggregate insider trading as we do not have any compelling ex-ante reason to investigate a specific time period. Specifically, in Panel A we report the aggregation of insider s trading activities over the period spanning (-180; -1) days before restatement announcements, in Panel B the period spanning (-90; -1) and (-30; -1) in Panel C. 11 A firm is assigned to the category BUY if the aggregate value of insider purchases during an aggregation window exceeded the aggregate value of insider sales. Similarly firms with insider sales exceeding insider purchases are assigned to the category SELL. Finally, we classify the remaining observations, that is, peer companies with no insider trading during an aggregation period, into the category NO_TRADE. We measure the market reaction using cumulative market-adjusted (size-decile-adjusted) returns over the three days around the peer restatement announcement. The periods over which we measure the main variables of interest are depicted in Figure 1. Consistent with our predictions in Hypothesis 1, we find that the abnormal stock returns at the time of a restatement by an industry peer announcement are significantly less adverse for peer companies with pre-announcement net insider buying relative to peer companies with preannouncement net insider selling. For example, Table 3 panel B shows that when insider trading activity is measured over 90 days before the restatement announcement, the difference in the mean marketadjusted returns between BUY and SELL categories is 0.597% (p-value<0.001). Table 3 further shows that the difference in returns between the BUY and SELL groups becomes more pronounced when insider trading is measured over more recent periods before the restatement announcements. This is consistent with investors viewing the most recent insider trading activity as the most precise signal. Further, the short-term market reaction to the restatement announcement for NO_TRADE category lies in between of abnormal return values for BUY and SELL groups. To evaluate the strength of the immediate market reaction we also calculate abnormal returns of non-restating firms over (+2; +30) days following the restatement announcement. For comparative purposes, these long-run abnormal returns are presented next to the three-day abnormal returns in Table 4. As shown, the difference between abnormal returns for BUY and SELL categories continues to increase over the next month especially for the SELL category. This reinforces our view that investors reaction to the information released at restatement announcements of industry peers includes a significant reweighting of the recent insider trading activity. Figures 2 and 3 display time-series variation 11 In untabulated analyses, we calculate insider trading activities over the periods spanning from (-180; -5), (-90; - 5), and (-30; -5) days prior to the restatement announcement to reduce the possibility that the market reaction was not to the disclosure of the insider trading activities. Our results are very similar when using these alternative windows to identify prior insider trading activities. 10

12 in the returns around peer restatement announcements in for the non-restating peers classified into portfolios based on the direction of recent insider trading. The figures highlight the disproportionately large market reaction for insider sells at the time of a peer restatement. Table 5 presents the differences in mean abnormal returns around restatement announcements of firms with insider purchases versus firms with insider selling during 90 days before restatement announcements for a series of subsamples. 12 These results indicate that the difference in peer firm abnormal returns around restatement announcements are even more dependent on the direction of the insider trading in the pre-announcement period for smaller firms, during the pre-sox period, and for intentional misstatement announcements. For example, difference in the three-day market-adjusted returns between peer companies with pre-announcement insider buying and pre-announcement insider selling before the adoption of SOX is 1.549% (p-value<0.001) while after adoption of SOX the corresponding difference is significantly lower (0.459%, p-value<0.001). Our results are consistent with those in Brochet (2010) and suggest that insider buying is a more informative signal for investors in the pre-sox period relative to the post-sox period. Overall, the results in Tables 4 and 5 provide preliminary support for Hypothesis 1 and Hypotheses 2(a)-2(c). Our evidence is consistent with investors of peer companies using information on the direction pre-announcement insider trading of non-restating firms to gauge the importance of an industry peer restatement Multivariate Tests of Hypotheses To provide a more complete test of Hypothesis 1, we estimate a multivariate linear regression model to control for possible confounding effects. Specifically, we estimate a panel model that includes control variables and year and industry fixed effects: % =! + " #$ + %&'' +( ) *++,-.,/0 +1&23 +4 (2) where, observations are indexed for peer firm i for the time, t, of the announcement of a restatement by an industry peer firm j (for i j). As our focus is on the market response of peer firms, we exclude the 12 In the subsequent analysis we use insider trading aggregated over (-90, -1) days before restatement announcement. Companies often implement policies which prohibit insider trading before earnings announcements (e.g., Bettis et al. 2000). Therefore, aggregating insider trades over 90 days should allow us to capture the most recent insider trading activity which would not otherwise fall into the blackout periods. The results are also statistically significant if we use alternative measurement windows. 11

13 restating firm. RET3d% is the 3-day cumulative market-adjusted or size-decile-adjusted abnormal return measured over (-1; +1) trading days around restatement announcement by an industry peer firm. We measure the insider trading activities of firms as an indicator variable that captures the direction of insider trading in non-restating firms over 90 calendar days before the announcement of restatements. Thus, coefficient estimates on #$ and %&'' measure the average returns around a peer restatement for the non-restating companies with either net buys or net sells in the prior 90 days relative to cases without insider trading (the no trade cases). We include both fixed effects and continuous variables as control variables. We include both industry (+,-.,/0 ) and year (1&23 ) fixed effects. Our choice of control variables (( ) ) is guided both by the broad literature on intra-industry information transfers and by the literature on earnings quality. For parsimony, we do not report the coefficient estimates on control variables (see Appendix). Consistent with prior studies, we measure control variables using the latest available annual data before the restatement announcement. Gleason et al. (2008) argue that the returns to a peer restatement may be related to a firm s growth options, particularly, they argue that the contagion effect induced by restatement announcements should be more severe for companies facing greater capital market pressures. 13 A firm s growth options may also affect insiders propensity to purchase or sell their own stock. Therefore, we control for growth options by including book-to-market ratio (BM), research and development (R&D) expenses deflated by total sales (RD) and firm age (FIRM_AGE). We include the earnings-to-price ratio (EP) as prior research suggests managers act as contrarian traders (Jenter 2005). To control for any potential size effect we include the natural logarithm of total assets (PEER_SIZE) into Equation (2). We also control for firm leverage, calculated as the sum of short- and long-term debt scaled by total assets (LEV). Because the content of restatement announcements is related to earnings quality, Gleason et al. (2008) argue that the restatement-related information transfer is likely to be more adverse in peer firms with indications of earnings management. To control for this effect, we include a measure of absolute discretionary accruals (see Appendix for details). Restatements by large companies are likely to receive greater press coverage and are likely to be more indicative of accounting problems in the same industry relative to restatements by smaller companies. Therefore, we control for the 13 Skinner and Sloan (2002) report that growth companies are penalized by market for reporting negative earnings surprises to a greater extent relative to value stocks. 12

14 restating firm s size by including the natural logarithm of its total assets (RESTATER_SIZE). Finally, we include industry and year fixed effects and cluster standard errors at the peer firm level. All the variables are defined in Appendix 1. The results from estimating Equation (2) are presented in Column (1) of Table 4. Panel A reports the results of specification with market-adjusted returns, while Panel B shows estimates of regression where size-decile-adjusted returns are used as a dependent variable. From Column (1) of Panel A it can be seen that the estimated coefficient on INSIDER_TRADING when it takes the value BUY is (p<0.001) and % for the SELL category. This suggests that peer companies with recent insider trading experience significantly different returns around the announcements of restatement news in the industry relative to peer firms without insider trading. According to this multivariate analysis the spread in announcement returns between firms with net insider buying and net insider selling is 0.738% (calculated as the difference of 0.452% less %). The coefficient estimates on control variables show that several other factors are significantly associated with the peer-firm restatement. Firms with lower growth opportunities, as proxied by higher book-to-market ratios, experience a smaller market reaction to the peer-restatement announcement. We fail to find evidence of an association between the market reaction to a peer restatement and absolute discretionary accruals (untabulated). The results also indicate that restatement announcements by larger companies trigger more negative intra-industry information transfers (coefficient on RESTATER_SIZE in column (1) of Table 5= 0.095, t-value= 2.56). To test H2a-H2c we augment Equation (2) with interaction variables as follows: % =5! +5 " #$ +5 %&'' #$ %&'' +( ) * ++,-.,/0 +1&23 +4 where, 1 is an indicator variable that takes the value of 1 for the sample partitions. The first partitioning variable is a dummy variable equal to 1 if the restatement occurs before the effective day of SOX (August 29, 2002), and zero thereafter (PRE_SOX). If the insider trading represents a more valuable signal to investors during the period of less strict enforcement and lower costs of jointly engaging in earnings manipulation and insider trading, we expect 6 to be significantly positive. Second, announcements of intentional misstatements would likely attract greater attention from industry peer investors making them reassess earnings quality of their companies. Hence the second partitioning variable is a dummy variable equal to 1 if the announcement is related to an intentional (3) 13

15 misstatement, zero otherwise (IRREGULARITY). We define intentional misstatements using the approach developed by Audit Analytics and expect 6 to be significantly positive. Finally, our third partitioning variable is a dummy variable equal to 1 if firm s assets are below the sample median and zero otherwise (SMALL). We expect insider trading signal to be more valuable for investors, when the amount of other information about the company is limited, e.g. in smaller companies. We therefore predict 6 to be significantly positive in this third specification. Table 4 shows the main univariate results while Columns (2)-(4) of Table 4 show the multivariate results of Equation (3) with one interaction variable at a time. The results in Table 5 Column (2) indicate that the coefficient estimate on PRE_SOX*INSIDER_TRADING (= BUY ) is 0.691, t-value=2.44. As expected, the coefficient estimate on PRE_SOX*INSIDER_TRADING (= SELL ) is negative and equal to 0.409, t-value= The sum of these interaction coefficients suggests that the difference in restatement announcement returns for companies with recent insider selling and recent insider buying is 1.100% higher in the pre-sox implementation period than afterwards. The F-value of 6.44 reported in the bottom of this table suggests that the difference is statistically significant. This is consistent with peer restatements having a stronger effect on investors perception of noise in the accounting system before SOX. Column (3) of Table 4, shows that a coefficient estimate on the interaction variable IRREGULARITY*INSIDER_TRADING (= SELL ) is significantly negative of 0.270(t-value= 1.66), consistent with our prediction that insider selling activity is a particularly strong signal for peer firm investors when intentional misstatements are announced. However, the F-test fails to reject the hypothesis that the difference is statistically significant. Finally, results reported in Column (4) of Table 5 indicates that the difference in the short-term announcement returns for peers with recent insider selling versus peers with recent insider buying is stronger for smaller companies. 14 From Panel B the difference for small companies is 0.517% higher (computed as a sum of coefficients SMALL*INSIDER_TRADING (= BUY ) and SMALL*INSIDER_TRADING (= SELL )) than the difference of 0.397% (computed as a sum of coefficients INSIDER_TRADING (= BUY ) and INSIDER_TRADING (= SELL )) for large companies. 14 Because the dummy variable SMALL is based on the partitions of PEER_SIZE, it controls for firm size. We therefore do not include a variable PEER_SIZE in the regression specifications with SMALL. 14

16 4. Further analyses 4.1. Insider trading and restatement risk An alternative to a reweighting of signals that potentially explains our findings is that prior insider trading is predictive of the likelihood that a firm s accounting information will be restated. If this were the case, then our findings provide evidence of insider trading having information content as a conditioning variable rather than as an information event in and of itself. To rule out this explanation, we first construct a subsample of all restatement events that have not been preceded by an industry restatement in the prior six months. This results in a sample of 18,561 observations with 4,756 unique firms. We then test the likelihood of a future restatement using the following specification: &:;2;& < ==! += " #$ += %&'' += 6 % +( ) *+1&23 +4 (4) where, we follow prior literature and include controls for size (PEER_SIZE), growth opportunities (BM), intangible investments (R&D), past accounting performance (ROA, LOSS), firm age (FIRM_AGE), leverage (LEVERAGE), and discretionary accruals (ABS_DACC) (e.g., Armstrong et al. 2013; Burns and Kedia 2006). 15 In Table 6 Panel A, we report estimates of Equation (4) using probit, logit, and linear probability specifications excluding %. We find that in all specifications, the coefficient on #$ is not significant at conventional levels, and the coefficient on %&'' is significantly negative. Further, using an F-test we fail to find evidence that the likelihood of future restatements differ significantly between firms where there has been recent insider trading, for example in Column (1) the F-statistic is 0.04 (p=0.84). This evidence suggests it is unlikely that the difference in the market reaction around restatements for these firms is due to expectations that insider trading signals either higher or lower information risk. In Table 6 Panel B, we report estimates of Equation (4) including %. We find that in all specifications, the coefficient on % is not significant at conventional levels. Further including % has little effect on the coefficients on #$ and %&''. This result is inconsistent with the market reaction to a peer-firm restatement being due to the anticipation of follow-on restatements. Therefore, our results suggest that one potential reason for the negative returns documented in Gleason 15 We note that, because ABS_DACC may be another way to measure accounting quality, it might qualify as a bad control. Our results are robust to its exclusion from our tests. 15

17 et al. (2008) could be due to investors changing the weight of non-accounting-based price information in response to a shock to perceived accounting quality Alternative measures of insider trading activities We consider alternative approaches to the measurement of insider trading activities. First, we partition insider trades used in the main tests into routine and opportunistic trades following the approach suggested by Cohen et al. (2012). Our expectation is that routine insider trading should not be informative and consequently should not affect revisions of investors expectations at the restatement announcements in the industry. To implement the approach of Cohen et al. (2012) each year we identify insiders who traded in the same calendar month during the most recent three consecutive years. We then estimate Equation (2) after excluding these routine trades. We find that in regression on sizeadjusted returns the coefficient estimates are more negative for the sell category (coefficient = , p<0.0001) and more positive for the buy category (coefficient = , p<0.0001) relative to main results when routine trades are excluded (untabulated). These results are consistent with the notion that routine trades are uninformative, at least for the smaller sample for which trades can be classified in this way. 16 Second, in our main analysis we only include the trades of the top executives of the firm following the assumption that inside trades by non-executive insiders are less informative. Ravina and Sapienza (2010), however, find that independent directors appear to earn returns on their trades that are similar to top executives. We include the insider trading activities of all insiders in the Thomson Reuters database and reestimate our regressions. We find that this broader measure of insider trading has little effect on the estimates of Equations (2) and (3), suggesting our analysis is robust to this measurement approach Robustness analysis We run a number of robustness analyses to determine the effects of changes in measurement of our key variables. First, we estimate Equation (2) with several additional control variables. Specifically, we include a 3-day market-adjusted return of a restating firm around restatement announcements to 16 Cohen et al. (2012) also identify opportunistic trades as trades made by those insiders, who traded in each of the three consecutive years, but in a different calendar month. When we use the sample of these trades, we find that the difference between buy and sell categories remains statistically significant. However, these unreported results also suggest that these opportunistic trades are no more informative that the trades which cannot be classified into either category because of the absence of trading history. 16

18 control for the strength of the intra-industry contagion. Unreported results suggest that the coefficient estimate on this variable in the regression on peer s market adjusted return is (p<0.05), suggesting that the reaction of peer firm s investors is more adverse when more severe restatements are announced. Next, we augment Equation (2) with a dummy variable equal to 1 if the peer firm and the restating company uses services of the same auditor (using Compustat item AU ). We expect that the market reacts more negatively when the peers use the same auditor. Unreported results suggest that the dummy variable loads negatively, however, the coefficient estimate is not significant at conventional levels. We also assessed the robustness of our findings to changes in the measurement of industry and the source of restatements. There are many possible industry classifications; especially those based on permutations of SIC codes. Whereas, we rely on the findings in Bhojraj et al. (2003) to choose our industry classification, we also consider a recent approach by Hoberg and Phillips (2010) who base industry classification using similarities of product descriptions in firm s 10-K filings. We find the main inferences of our study are robust to this alternative classification (untabulated). We also consider the use of an alternative source for restatement dates by using the data collected by the GAO, and continue to find similar results (untabulated). 5. Conclusion We examine how shocks to perceived accounting quality lead to changes in the market pricing of prior disclosures of insider trading. Disclosure theory suggests that Bayesian investors weigh different signals according to their precision, which would imply that the market pricing of insider trading activity is based on the precision of insider trading relative to other signals about future cash-flows. Prior literature also highlights the possibility that insiders decisions to disclose information and to trade their own shares are simultaneously determined. This literature informs our hypotheses, which are based on the reasoning that we can identify information content from insider trades when there is a shock to perceived accounting quality. We use a sample of firms where industry peers have restated financial information as a shock to perceived accounting quality. We find that the short-term market reaction to announcements of restatements by industry peers is significantly more negative for companies with recent insider net selling than for companies with recent insider net buying. Firms without recent insider trades, in contrast, have significant negative returns at the time of a peer restatement. Our results are robust to a 17

19 number of changes in the measurement of key proxies and alternative specifications. These results are important and highlight that disclosures of both insider purchases and sales are informative to investors, a result that is expected given theory, but has eluded prior empirical investigations of insider trading. We also conduct tests to examine whether cross-sectional and time-series variation in the observed associations is consistent with our theory. As predicted, we find results that suggest that recent insider trading signals are more informative to investors in smaller firms, prior to the implementation of SOX, and when peer firms disclose intentional material misstatements. The statistical and economic magnitudes of these effects are consistent with predictions based on the re-weighting of signals. In further analysis, we find that it is unlikely that the link between returns and prior insider selling is due to an increase in the likelihood of the firm restating. These results are inconsistent with investors using insider trades as a conditioning variable to predict the extent of accounting contagion. Instead, our evidence is consistent with investors reweighting the importance of insider trading information when the usefulness of accounting as a signal of firm value exogenously declines. Taken together, our findings suggest that insider selling disclosures are an important source of information when perceived accounting quality declines. To the extent that insiders are unable to predict a peer firm s restatement, our analysis avoids the endogeneity problems due to the simultaneity in insiders choices of when to trade relative to the timing of other disclosures. Second, our results suggest a significant portion of the intra-industry information transfers arising from peer restatements is due to a reweighting of other information, rather than an increased discounting of accounting information. As such, the pricing of accounting quality does not appear to be the sole driver of the (on average negative) market reaction to a peer restatement. Instead, we provide an alternative explanation, based on the reweighting of signals that is caused by a reduction in the perceived precision of accounting information. Finally, our results provide an important insight into why stale disclosures might appear to elicit a market reaction, based on investors reweighting prior disclosures when accounting quality exogenously declines. 18