Are corporate managers savvy about their stock price? Evidence from insider trading after earnings announcements

J. Account. Public Policy 29 (2010) 27–44

Contents lists available at ScienceDirect

J. Account. Public Policy journal homepage: www.elsevier.com/locate/jaccpubpol

Are corporate managers savvy about their stock price? Evidence from insider trading after earnings announcements q Adam Kolasinski a, Xu Li b,*,1 a b

Michael G. Foster School of Business, University of Washington, 335 Mackenzie Hall, Seattle, WA 98195-3200, USA School of Management, University of Texas at Dallas, 2601 N. Floyd Road, Office 4.427 (SM 41), Richardson, TX 75083-0688, USA

a r t i c l e

i n f o

JEL classification: G14 M41 Keywords: Insider trading Market timing Post-earnings announcement drift Underreaction anomolies

a b s t r a c t We find that insiders trade as if they exploit market underreaction to earnings news, buying (selling) after good (bad) earnings announcements when the price reaction to the announcement is low (high). We also find that insider trades attributable to public information about earnings and the price reaction generate abnormal returns. By demonstrating that managers spot market underreaction to earnings news, our results imply that managers are savvy about their company’s stock price. Ó 2009 Published by Elsevier Inc.

1. Introduction Most studies on insider trading focus on whether insiders trade based on their private information. In contrast, we study how public information about prices and earnings influences insiders’ trading strategies. We find that insiders tend to buy more and sell less when a stock underreacts to good earnings news than when it underreacts to bad earnings news. Specifically, insiders engage in more net buying after release of good earnings news accompanied by a low price reaction than they do after release of bad earnings news accompanied by a high price reaction. We also find that when our model of insider trading based on public earnings and price information predicts high insider net buying in a stock, the stock subsequently experiences positive abnormal returns. We thus find evidence that

q We are indebted to Richard Frankel, Jarrad Harford, Dirk Jenter, SP Kothari, Suresh Radhakrishnan, Jeremy Stein and Joseph Weber for their helpful comments. We also thank MIT and UT-Dallas workshop participants and participants of First Lone Star Accounting Research Conference for their helpful suggestions. We have benefited from the editor, Martin Loeb and two anonymous reviewers for their comments. We thank Bin Ke for assistance in computing abnormal returns after earnings announcements. * Corresponding author. Tel.: +1 206 543 8737. E-mail addresses: [email protected] (A. Kolasinski), [email protected] (X. Li). 1 Tel.: +1 972 883 6385.

0278-4254/$ - see front matter Ó 2009 Published by Elsevier Inc. doi:10.1016/j.jaccpubpol.2009.10.004

28

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44

insiders trade in response to market underreaction to earnings, confirming that they have some skill at judging whether their firm’s stock price correctly reflects public information. These results support the notion that managers are savvy about their company’s stock price, a largely untested assumption critical to some models in the finance literature (e.g., Baker and Wurgler, 2002; Schleifer and Vishny, 2003). Our paper is one of the first to directly study insiders’ ability to use public information to earn abnormal returns. Seyhun (1988) finds that aggregate insider trading patterns forecast aggregate market returns. These results indicate that insiders have a superior ability to identify macro-economic trends that tend to move the market, but it is possible that private information about trends in new orders, sales, or other fundamentals correlated with macro-economic factors may give them this ability. Seyhun (1990) finds evidence that insider purchases increased substantially following the 1987 stock market crash, and that such purchases were highly profitable, indicating that insiders exploited market overreaction to the crash. The study, however, is limited to only one brief episode, whereas ours covers a broad time period. On their face, the findings of Sivakumar and Waymire (1994) that insiders sell (buy) after good (bad) earnings news may seem to indicate that insiders attempt to exploit overreaction to such news. However, an equally plausible explanation is that insiders are using private information to identify earnings peaks and troughs. Other studies on insider trading document that insiders profitably trade in a contrarian fashion in anticipation of events that signal a reversal in firm trends. Ke et al. (2003) document that insiders make profitable trades in anticipation of breaks in earnings trends, selling just before an earnings peak and buying just before a trough. Karpoff and Lee (1991) and Kahle (2000) find that insiders sell in anticipation of new issue announcements, which typically result in price declines and occur after price run-ups. Noe (1999), Cheng et al. (2007) and Givoly and Palmon (1985) find that managers trade on their superior foreknowledge of future earnings patterns as well as other private information. These results, and others like it, are important because they demonstrate that insiders successfully anticipate the direction in which prices will move in response to future disclosures about which they have private information. The results, however, have little bearing on insiders’ ability to determine whether prices correctly reflect public information at a given point in time. In contrast, our results demonstrate that they have this ability. While our results indicate that insiders use public signals to enhance their returns, it is likely that they also use private information when deciding their trading strategies. For example, Lustgarten and Mande (1995) find that insiders trade on private information about forthcoming earnings. Park and Park (2004) document a relation between insider sales and earnings management. Since there are many corporate events that we do not observe, the insider trading patterns we observe are likely based on some combination of both private and public information.2 Nevertheless, we believe the fact that insiders do use public information, and in particular, information about prices, to enhance their returns, is an important result, because it provides evidence that insiders are price-savvy, at least when it comes to underreaction to the disclosure earnings news. Our results complement Jenter (2005), who finds that insiders profit from buying when book-tomarket ratios are high and selling when they are low, thus apparently exploiting the value effect. If the mispricing of value and glamour stocks causes the value effect, Jenter’s results suggest that insiders are indeed skilled at detecting mispricing in their stock. By finding similar results in a different setting, namely, post-earnings announcement drift, we bolster Jenter’s (2005) inference that insiders are indeed savvy about mispricing in their own firm’s sock. Our evidence has important public policy implications. The degree to which insider trading should be permitted is a matter of heated debate (see Hu and Noe (1997) for a summary). Some opponents of insider trading cite high insider trading profits as evidence that insider trading hurts liquidity traders. Others, such as Ronen et al. (2006), argue that insider trading may encourage earnings management. Our finding that insiders trade, at least in part, based on their interpretation of public information, suggests that not all insider trades have such negative effects, even during a

2 Our results are robust by controlling for future earnings surprises in our tests, so insider trading on future earnings information cannot be driving our results.

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44

29

sample period during which there were few blackout date restrictions. Hence our findings support Hagerman and Healy (1992), who find that insider trading does not adversely impact bid-ask spreads. That insiders trade based on misevaluation of public information suggests another channel through which their trading may cause information to get more accurately reflected in stock prices. Hence we also complement the results of Lustgarten and Mande (1998), who find that insider trading reveals information about future earnings to market participants. Our results suggest a means by which insider trading may improve the extent to which market participants correctly interpret public information. The rest of the paper is organized as follows. In Section 2 we discuss our hypotheses pertaining to the relationship between insider trading, earnings news, and the initial price reaction to such news. Section 3 describes our sample, data and research design. In Section 4 we present our results. In section 5, we test how sensitive our results are to alternative specifications and variable definitions. Section 6 concludes.

2. Discussion of hypotheses In this section, we develop hypotheses regarding insider trades made in response to the release and price reaction to earnings news, as well the trades’ abnormal returns. If insiders are market savvy, they should recognize market underreactions to earnings news and adjust their trading strategies accordingly. Therefore, if insiders realize that the market has underreacted to good news, ceteris paribus, we expect them to buy more and sell less than they otherwise would. Likewise, we expect them to sell more and buy less than they otherwise would if the market has underreacted to bad news. Many studies document that stock prices underreact to unexpectedly high or low earnings news (Ball and Brown, 1968; Bernard and Thomas, 1989, 1990; Foster et al., 1984; Watts, 1978; see Kothari (2001) for a summary). In particular, they find that positive earning shocks forecast positive abnormal returns, and negative earnings shocks forecast negative abnormal returns. Sivakumar and Waymire (1994), however, document that insiders tend to sell after good earnings news and buy after bad earnings news. Thus, at first glance, it would seem that insiders do not attempt to exploit underreaction to earnings news. Sivakumar and Waymire (1994), however, fail to take into consideration the effect of announcement returns on insider trading decisions after earnings announcements. To see how insider trades might be influenced by the announcement returns, we consider two plausible scenarios. Insiders may actively try to exploit underreaction to earnings news. Insiders may also delay or accelerate previously planned trades, in a manner similar to what Huddart et al. (2007) call ‘‘passive trading,” in order to accommodate underreaction. We now discuss each in turn. An insider actively exploiting underreaction buys and sells when the market underreacts to good and bad news, respectively. It is plausible that insiders believe the market is more likely to have underreacted to good and bad earnings if the announcement returns are low and high, respectively. Thus active insiders will buy following a good earnings announcement accompanied by a low announcement return. They will sell following a bad earnings announcement accompanied by a high announcement return, though they may be reluctant to sell in this case to avoid sending a negative signal to the stock market which may affect the firm and insiders unfavorably. They are less likely to trade following good and bad earnings announcements accompanied by high and low announcement returns. Panel A of Fig. 1 summarizes these predictions. Some insiders likely trade for reasons other than active exploitation of underreaction, but we expect them to adjust the timing of their trades to accommodate underreaction, in a ‘‘passive trading” of sorts. Stock-based compensation forces insiders to hold undiversified positions in their own firm, giving them an incentive to sell as shares vest. Similarly, insiders often have private information that positive news will come in the future, giving them an incentive to buy. They might also want to buy to signal confidence to the markets. However, market underreaction gives insiders an incentive to alter the timing of such trades. Insiders have an incentive to use their private information to time their planned sales to coincide with earnings peaks. However, if the market underreacts to a good earnings

30

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44

Predicted Insider Trading Intensity Panel A: Active insiders Initial market return High Low Net insider purchases Earnings Good No insider transactions Bad Net insider sales/no insider transactions No insider transactions news

Panel B: Passive insiders Earnings Good Bad news

Initial market return High Low Net insider sales No insider transactions No insider transactions Net insider purchases

Panel C: Both active and passive insiders Initial market return High Low Net insider purchases Earnings Good Net insider sales Bad Net insider sales/no insider transactions Net insider purchases news Fig. 1. Predicted insider trading intensity.

announcement, insiders have an incentive to delay such sales. Now consider an insider who wants to buy because he has some private information about positive future news. Ke et al. (2003) find that insiders trade on private information as early as two years prior to its eventual disclosure. The insider will want to use his private information to time this purchase to coincide with an earnings trough. However, if the market underreacts to the announcement, the insider will delay this purchase. Thus with passive insider trading, we expect more insider selling after good earnings when the announcement return is high, since in this case earnings are more likely to be at a high point and the market is unlikely to have underreacted. Similarly, we expect buying after bad earnings when the announcement return is low. With passive trading, we expect less trading in cases where the market underreacts, i.e., when earnings are good but the return is low and earnings are bad and the return is high. Panel B of Fig. 1 summarizes these predictions. It is likely that insiders trade both actively and passively in response to underreaction to earnings announcements. We thus expect some combination of panels A and B of Fig. 1, which we label panel C. Following a good earnings announcement, we expect net selling if the announcement return is high and net buying if the announcement return is low. Following a bad earnings announcement, we expect little trading (perhaps some selling) if the announcement return is high and buying if it is high. In short, we expect the magnitude of the earnings surprise and announcement return to have an interactive effect on insider trading. Formally: H1: After a good earnings announcement, we expect net insider purchases to be low (possibly negative) if the announcement return is high and net insider purchases to be high if the announcement return is low. After a bad earnings announcement, we expect insider net purchases to be high if the announcement return is low and net purchases to be low (possibly negative) if the announcement return is high. We will henceforth refer to H1 as the ‘‘insider awareness hypothesis”3. Confirmation of H1 would not necessarily imply that insiders are correctly adjusting their trading strategies in response to market underreaction. It would, at best, suggest that they trade as if they be3 Since this hypothesis predicts an interaction effect, it can also be interpreted as the effect of stock price reaction on net purchases is different for high versus low unexpected earnings. We thank an anonymous referee for pointing it out.

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44

31

lieve the market has underreacted, but it would provide no indication whether their beliefs are valid. To test the validity of these beliefs, we must test whether the adjustments insiders make to their trading strategies based on the magnitude of unexpected earnings and the announcement return will increase their trading profits. We thus formulate the following hypothesis: H2: Insider net purchases (net sales) predicted by the interaction between the magnitude of unexpected earnings and the stock price reaction to the release of earnings news should in turn forecast positive (negative) abnormal returns. We label H2 as the ‘‘insider returns hypothesis”. Modern portfolio theory, at first glance, appears to undercut the motivation for our hypotheses. Inside information is not per se necessary to identify firms whose stock prices have likely underreacted to earnings news. Insiders could exploit such underreaction by trading in a diversified portfolio of stocks other than their own, a seemingly preferable strategy given that their human capital and financial wealth tends to be concentrated in their own firm. Modern portfolio theory notwithstanding, however, there are still plausible economic reasons for insiders to exploit underreaction by trading in their own stock. Insiders may use private information to supplement public signals, allowing them to more accurately assess the magnitude of underreaction. Alternatively, as Jenter (2005) points out, insiders may not be well-versed in modern portfolio theory, and hence may trade in their own shares despite lack of diversification. After all, insiders are selected based on their ability to manage a firm, not a portfolio. Finally, as discussed above, insiders trading for exogenous reasons, such as selling to diversify or buying based on private information of future positive news, have an incentive to adjust the timing of their trades to accommodate underreaction. Whether insiders trade in response to underreaction to earnings news, therefore, is an open empirical question worthy of examination. 3. Sample, data, and research design In this section, we first discuss our sample and data (Section 3.1). We describe our measures of insider trading intensity, unexpected earnings, and returns (Section 3.2). We also detail our method of testing both the insider awareness hypothesis, or H1 (Section 3.3) and the insider returns hypothesis, or H2 (Section 3.4). 3.1. Sample and data We obtain data from five sources. Financial statement data are from Compustat. Stock return data are from the Center for Research in Security Prices (CRSP). Insider trading data are from the Security and Exchange Commission’s (SEC) Ownership Reporting System (ORS) data files. Institutional ownership data are from Thomson Financial. Analyst forecast data are from I/B/E/S. From the ORS data file, we download all transactions by insiders from 1980 to 1997, all the years available to us in the database. Insiders in ORS are defined by Section 16(a) of the Securities and Exchange Act of 1934. In our paper, we only include transactions made by managerial insiders, such as any director or officer of the issuer, because Lakonishok and Lee (2001) document that these trades are more informative. Following previous literature, such as Frankel and Li (2004), we apply several filters to the sample to ensure that these insider transactions are both meaningful and consistent with the stock return data from CRSP.4 After applying these filters, we are left with 346,120 insider transactions for 6925 firms. We sort the transactions by firm and calendar quarter. For each firm-quarter in our sample, we obtain quarterly earnings announcements from the Compustat quarterly data file. We delete firm-quarters for which no Compustat data are available. For each 4 These filters include eliminating non-common shares, duplicate transactions, transactions with missing transaction date, private transactions, transactions with price less than $2 a share or price falling out of Bidlo and Askhi of CRSP daily file, transactions with number of shares less than 100 or more than CRSP daily volume and transactions made by beneficial owners.

32

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44

firm-quarter, we calculate the number of insider purchases and sales that occur within the insider trading period, as defined in the following section. If no trade occurs, we code the number as zero. We also obtain for each firm-quarter earnings per share data (item 19) for that fiscal quarter as well as for the same fiscal quarter for the previous year. We then calculate unexpected earnings for a firmquarter by subtracting the previous fiscal year’s quarterly earnings per share number from the current year’s, and normalize by the stock price as of 2 days before the earnings announcement date. From CRSP we obtain return data, and generate a size-adjusted return by subtracting the return on the NYSE–AMEX–NASDAQ size decile to which a particular firm belongs. In the sensitivity analysis, we obtain institutional ownership data from Thomson Financial and analyst earnings forecast data from I/ B/E/S. After the data merging process is complete, we have 163,018 firm-quarter observations. Observations are further reduced in later tests because some variables have missing values. 3.2. Measures of insider trading, earnings and stock returns For each earnings announcement, we define three time periods. Designating the earnings announcement date as day 0 and referring to any other date by the number of days from day 0, we define our periods as follows: 1. The earnings announcement period: the period beginning on day 1 and ending on day 1. 2. The insider response period: the period beginning on day 0 and ending on day 30. 3. The subsequent return period: the period beginning on day 30 and ending on day 180. During the earnings announcement period, we measure the market’s reaction to the earnings news. We begin this period on day 1 to allow for information leakage. We end it on day 1, instead of day 0, in order to ensure that the market has had enough time to react to the earnings news. Following Bernard and Thomas (1990), we compute the cumulated size-adjusted abnormal return during this period as R1 and use it as our measure of the market’s reaction to earnings news. During the insider response period, we measure the intensity of insider trading as the number of net purchases or Net P, defined as P minus S, where P is number of purchases and S is number of sales. We choose the starting and ending dates of this period so that it broadly coincides with the period to which most firms restrict insider trading. The insider response period also broadly coincides with the period during which insiders are observed to trade most frequently (see Bettis et al., 2000; Jeng et al., 2003; Roulstone, 2004) and are most likely attempting to rebalance their portfolios to accommodate underreaction to earnings news. Our measure of insider trading intensity is similar to Lakonishok and Lee (2001), with the difference that they scale net purchases by total transactions (purchases + sales) and we do not. We believe that in our context, unscaled net purchases are a superior measure because our trading window is short, only one month, so many firms in our sample have only sales or only purchases. If we used scaled net purchases, firms with multiple purchases but no sales, or multiple sales but no purchases, would be coded as having the same level of insider trading intensity as firms with only one purchase or sale. This seems unreasonable to us. Lakonishok and Lee (2001) do not have this issue because their insider trading window is much longer, six months, during which time few firms have just purchases or just sales. Nevertheless, as discussed in Section 5, we confirm that our results are robust to the Lakonishok and Lee (2001) measure as well as other measures of insider trading intensity. The subsequent return period, day 30 to day 180, roughly coincides with the period during which the bulk of abnormal returns following an earnings announcement are realized (i.e., ‘‘drift”). We define the cumulative size-adjusted return over this period as R2. We exclude from the subsequent return period the first 30 days that follow the earnings announcement because insider trades tend to be distributed throughout this time, and we want to capture the return during the period after the trades have been made. We present Fig. 2 to better illustrate the timeline of our variables’ definition. We assume earnings follow a seasonal random walk process and define unexpected earnings as earnings per share in quarter t less earnings per share in quarter t  4, scaled by the stock price. We use the seasonal random walk model, rather than analyst forecasts, because many firms in our

33

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44

Timeline Insider trading activity period

Earnings announcement Day 0

Day -1 Day 1

Day 30

Day 180

Earnings announcement Period R1

Subsequent return R 2

Fig. 2. Timeline.

sample are not covered by analysts. Hence we would lose an unacceptably large number of observations if we were to use analyst forecasts to estimate unexpected earnings. We do not use Bernard and Thomas’ (1989, 1990) standardized unexpected earnings measure because it requires a minimum of ten quarters to compute and they only use NYSE and AMEX firms, so it would also cause us to lose many observations when our sample includes NASDAQ firms. Such a measure has been used in recent studies, for instance, see Ng et al. (2008). Our results, however, are robust to these alternative definitions of earnings surprise, as we discuss in Section 5. 3.3. Testing the insider awareness hypothesis (H1) For each calendar quarter, we rank firms according to their unexpected earnings and compute decile breakpoints. For each calendar quarter, we assign firms into earnings deciles using the breakpoints from the previous quarter’s unexpected earnings distribution. Not all firms announce earnings on the same day, so the distribution of the current quarter’s unexpected earnings is not observable for much of the period during which the insiders in our sample are trading. Hence using the current quarter’s breakpoints might introduce look-ahead bias. We set UEp equal to the firm’s assigned earnings decile in a given calendar quarter. Using a two-way sort method, we next assign deciles for R1. Within each unexpected earnings decile for the previous quarter, we compute breakpoints for R1 deciles in that quarter. We then assign current period values of R1 to deciles using these breakpoints. We set R1p equal to the rank of the assigned decile. Since we sort by R1 in descending order, R1p equals 1 for the firm-quarter observations that have the highest price reactions, and R1p equals 10 for the firm-quarter observations that have the lowest (or most negative) price reactions. We estimate the following equations by running cross-sectional Fama–MacBeth type regressions in which the data are partitioned by calendar quarter:

Net P ¼ a þ b1  UEp þ b2  R1 p þ b3  UEp  R1 p þ b4  Size þ b5  B=M þ b6  Holdings X bi  Industryi þ e: þ b7  Momentum þ b8  Net P Pre þ

ð1Þ

Variables are defined as follows: Net P: Net number of purchases is calculated as (number of purchases – number of sales), where number of purchases is number of insider purchases that occur for a firm-quarter within thirty days after an earnings announcement and number of sales is number of insider sales that occur for a firm-quarter within thirty days after an earnings announcement. UEp: Decile rank by unexpected earnings, which is measured as actual earnings of quarter t less actual earnings of quarter t  4 scaled by stock price two days before the earnings announcement of quarter t. We use the previous quarter’s breakpoints in unexpected earnings decile breakpoints to assign the rank. R1p: Inverse decile rank of the current quarter’s R1, based on the previous quarter’s R1 breakpoints, where R1 is defined as cumulated size-adjusted return in 3-trading-days centered on earnings

34

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44

announcement date for each firm-quarter, so firms with the highest announcement returns belong to decile 1 and firms with the lowest belong to decile 10. This variable is a product of sequential sorting, in which we first sort firm-quarters into earnings deciles, then within each earnings decile, we sort the return into deciles. UEp*R1p: Interaction of UEp and R1p. Size: log of market value of equity measured two days prior to the earnings announcement. B/M: book value measured at the end of the quarter divided by market capitalization. Holdings: Average number of shares held by the insiders who trade in this quarter divided by the total number of share outstanding. Momentum: 12-month buy-and-hold return ending 2 days before the earnings announcement date. Pre-Net P: Net P measured in the window of 6 month before the earnings announcement date. Industryi: Industry indicator, 49 industries are defined as Fama and French (1997). Since insiders have been documented to be contrarian on average, (e.g., Jenter, 2005; Sivakumar and Waymire, 1994), we expect b1 to be negative and b2 to be positive in Eq. (1). H1 predicts a positive sign of b3: if insiders exploit underreaction, they will tend to buy (sell) after good (bad) earnings news when the price reaction is small in absolute magnitude. We include the two main effects, UEp and R1p in the equation and expect the coefficients on them to be significant, since insiders may trade for many reasons other than taking advantage of underreaction to earnings news. If, after controlling for the two direct effects, we still find a positively significant b3, it would suggest that market underreaction does drive some insider trades, but not all of them. We use the following variables to control for other factors that may affect the frequency of trading. Size: Log of market value of equity measured two days prior to the earnings announcement. The larger the firm, the more insiders it may have, and hence the more likely for it to have insider trades, both purchases and sales. Sales are often made for reasons unrelated to the insider’s perceptions of firm value (personal liquidity needs, portfolio rebalancing, etc.) and there are more sales transactions than purchase transactions for insiders on average. The more insiders lead to the more sales transactions than purchase transactions, so we expect a negative b4 for Eq. (1). B/M: Book value measured at the end of the quarter divided by market capitalization. As documented in Jenter (2005), insiders tend to buy when B/M is high and sell when B/M is low, so we expect b5 to be positive for both Eq. (1). Holdings: Average number of shares held by the insiders who trade in this quarter divided by the total number of share outstanding. If insiders hold more of the firm’s stock, then for reasons related to portfolio rebalancing they are more likely to sell but less likely to buy. Therefore, we expect b6 to be negative for both Eq. (1). Momentum: To control for past returns’ effect on insiders trading decisions since prior literature document that insiders trades are contrarian, for example, see Lakonishok and Lee (2001). We expect a negative sign on momentum in Eq. (1). Pre-Net P: Net P measured in the window of 6 month before the earnings announcement date. It is well recognized that insiders are restricted by the short-swing rules. According to the restriction, insiders who sold stocks prior to the earnings announcement cannot buy stocks within the subsequent 6 months and vice versa. We predict a positive coefficient on Net P Pre.5 Because we control for proxies for risk (B/M and size) as well as momentum, we have a clean test of whether market underreaction to the public release of earnings news affects insider trading. Underreaction to earnings news tends to be pronounced in the more extreme earnings deciles (e.g., Bernard and Thomas, 1990). In addition, brokerage commissions, lack of diversification mentioned above, and the inability to unwind positions for 180 days, all present significant trading costs for insiders. Insider trading to exploit underrecation may thus be stronger in extreme earnings deciles. We

5

We thank an anonymous referee for suggesting this variable.

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44

35

therefore estimate the following equation for observations in the top earnings news deciles, that is, observations for which UEp equals 9 or 10, and again for observations in the bottom earnings news deciles, that is, observations for which UEp equals 1 or 2. As before, we run cross-sectional Fama–MacBeth type regressions in which the data are partitioned by calendar quarter:

NetP ¼ a þ b1  R1 p þ b2  Size þ b3  B=M þ b4  Holdings þ b5  Momentum þ b6 X  Net P Pre þ bi  Industryi þ e:

ð2Þ

Since we limit our sample to firm-quarters in extreme deciles, we drop the interaction term of UEp*R1p from the specification. For the subsample of observations in the top earnings news deciles, a higher R1p means a greater market underreaction, so H1 predicts more net insider purchases for higher R1p and hence a positive sign on b1. For observations in the bottom earnings news deciles, a lower (less negative) R1p means less underreaction, so H1 also predicts a positive b1 for this subsample. H1 also suggests that the coefficient on R1p is higher in the top earnings news subsample than that in the bottom earnings news subsample. This specification, because it lacks an added interaction term, also has the advantage of making it easier to assess the economic significance of coefficient estimates. 3.4. Tests of the insider returns hypothesis (H2) We test H2 by showing that the portion of insider trades attributable to insider attempts to exploit underreaction generate trading profits. Recall that it is likely that insiders’ primary motivation for trading something other than exploitation of such underreaction. Therefore, we cannot test H2 by simply testing whether insider trades as a whole are associated with abnormal returns. Furthermore, such a test would be superfluous since a plethora of studies have documented that insider trades forecast stock returns.6 Instead, we examine the abnormal return forecasting power of the portion of insider trades attributable to the magnitude of earnings surprise and the market’s reaction to such surprise. To accomplish this task, we first run a pooled version of regression model (1) and obtain the coefficient estimates. We then use these coefficient estimates to estimate the portion of net insider purchases attributable to insider awareness of underreaction to earnings news. We label this quantity d and compute it as follows: NetP

^2  R1 p þ b ^3  UEp  R1 p: ^1  UEp þ b d ¼a ^þb NetP

ð3Þ

d we estimate the following equation using the quarterly Fama–MacBeth Having calculated NetP method:

d þ b  Size þ b  B=M þ b  Momentum þ e: R2 ¼ a þ b1  NetP 2 3 4

ð4Þ

The dependent variable is the size-adjusted cumulative abnormal return from day 30 to day 180. The d as well as the ubiquitous size, book-to-market, and momentum independent variables include NetP controls. If H2 is supported, we predict a positive and significant b1. As a robustness check, we rerun the above specification using industry dummies both in Eqs. (3) and (4). 4. Results 4.1. Descriptive statistics Our sample consists of insider trades made within the one-month period after the earnings announcement. In the aggregate, such insider trades consist of over 45% of the total. Table 1 presents equal-weighted unexpected earnings decile portfolio returns for several different periods after earnings announcements. UE decile 1 is the extreme bad news portfolio, which contains firms whose quarterly unexpected earnings were the lowest. UE decile 10 is the extreme good news 6 A partial list of studies that document that insider trades are profitable includes Jaffe (1974), Seyhun (1986), Lin and Howe (1990) and Jeng et al. (2003). Lakonishok and Lee (2001) conclude that only insider purchase in small firms are informative.

36

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44

Table 1 Abnormal buy-and-hold returns of decile portfolios.

UE UE UE UE UE UE UE UE UE UE

decile decile decile decile decile decile decile decile decile decile

1 2 3 4 5 6 7 8 9 10

60 Day return (%)

90 Day return (%)

120 Day return (%)

180 return (%)

1.59 0.85 0.51 0.51 0.19 0.36 1.05 1.55 1.81 1.61

2.45 1.53 1.01 0.84 0.36 0.56 1.52 2.68 3.02 2.98

2.86 1.66 1.00 0.81 0.27 0.81 2.14 3.35 3.58 3.44

3.96 2.16 1.21 0.86 0.33 1.30 2.78 4.40 5.03 4.84

This table reports average abnormal returns of firms in various unexpected earnings deciles for various time periods following the earnings announcement. Each time period begins 2 days after the earnings announcement and lasts 60, 90, 120 or 180 days later. Deciles are computed using the quarterly ranking of unexpected earnings, which is defined as (earnings of quarter t minus earnings of quarter t  4) scaled by price two days before quarter t’s earnings announcements. Abnormal return is defined as buy-and-hold return of the stock less the buy-and-hold return over the same period of the CRSP NYSE-AMEX-NASDAQ valueweighted size decile index to which the stock belongs. UE decile 1 consists of firms with the lowest unexpected earnings and UE decile 10 consists of firms with the highest unexpected earnings.

portfolio, which contains firms whose quarterly unexpected earnings were the highest. If the market on average underreacts to earnings releases, the abnormal returns should move from being most negative in decile 1 to most positive in decile 10. Previous literature on the earnings underreaction anomaly also shows that the absolute magnitude of the return of decile 1 is smaller than that of decile 10. In Table 1, returns over the four horizons are consistent with the predicted pattern in general. Returns monotonically increase from decile 1 to decile 10 in most cases, as documented in prior literature. More importantly, the magnitudes of abnormal returns are smaller for the bad news portfolio, while they are considerably larger for the good news portfolio. The pattern indicates that there is less underreaction to bad earnings news than good earnings news. Table 2 shows the proportion of firm-quarters with insider trades across earnings news deciles and the significant tests of the difference between each decile and decile 1. The proportion of quarters with insider purchases nearly monotonically decrease from decile 1 to decile 10 and the t-statistics show that other than decile 2, the difference are all significant at least at 5% level, implying that insiders

Table 2 Portion of quarters with insider trades across deciles portfolios.

UE UE UE UE UE UE UE UE UE UE

decile decile decile decile decile decile decile decile decile decile

1 2 3 4 5 6 7 8 9 10

Portion of firm-quarters with purchases (%)

T-test of difference from decile 1

Portion of firm-quarters with sales (%)

T-test of difference from decile 1

9.36 9.23 8.74 7.99 8.31 8.21 7.91 7.52 7.40 7.40

– 0.42 1.96 4.41 3.32 3.32 4.66 5.96 6.39 6.37

9.85 12.96 15.01 16.35 17.99 20.81 21.00 18.25 15.61 11.35

– 8.84 14.47 17.47 21.39 27.74 28.21 21.97 17.03 5.79

This table reports the portion of quarters with insider trades for each unexpected earnings (UE) decile portfolio. Unexpected earnings (UE) are defined as earnings of quarter t less earnings of quarter t  4 scaled by price two days before quarter t’s earnings announcement. UE decile 1 is the extreme bad news decile and UE decile 10 is the extreme good news decile. Number of insider purchases (sales) is calculated as number of insider purchases (sales) made within thirty days after the earnings announcement. Each quarter is characterized as with or without insider purchases (sales). The portions of quarters with insider purchases (sales) are obtained for each decile. The t-statistics are calculated for difference between the portion of each decile and decile 1.

37

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44 Table 3 Descriptive statistics for variables.

Net P UEp R1p UEp*R1p Size B/M Holdings Momentum Pre-net P R2

N

Mean

Standard deviation

Maximum

Q3

Median

Q1

Minimum

163,018 163,018 162,768 162,768 163,018 163,018 163,018 163,011 163,018 162,798

0.29 5.50 5.50 30.27 5.10 0.78 0.04 0.20 0.99 0.03

1.84 2.87 2.86 23.78 1.82 0.62 0.17 0.57 4.81 0.25

47.00 10.00 10.00 100.00 12.37 51.43 0.99 27.80 110.00 5.99

0.00 8.00 8.00 45.00 6.33 0.98 0.00 0.39 0.03 0.15

0.00 5.00 6.00 24.00 4.96 0.65 0.00 0.12 0.00 0.02

0.00 3.00 3.00 10.00 3.75 0.40 0.00 0.12 1.00 0.10

53.00 1.00 1.00 1.00 0.13 0.00 0.00 0.98 137.00 3.33

This table provides descriptive statistics for variables used in subsequent tests. To be included in this table, a firm-quarter observation must be accompanied by sufficient data to compute the variables displayed below. Firm-quarter observations are drawn from the period between 1980 and 1997. Variable definitions: Net P: net number of purchases is calculated as (number of purchases  number of sales), where number of purchases is number of insider purchases that occur for a firm-quarter within thirty days after an earnings announcement and number of sales is number of insider sales that occur for a firm-quarter within thirty days after an earnings announcement; UEp: decile rank by unexpected earnings, which is measured as actual earnings of quarter t less actual earnings of quarter t  4 scaled by stock price two days before the earnings announcement of quarter t. We use the previous quarter’s breakpoints in unexpected earnings decile breakpoints to assign the rank; R1p: inverse decile rank of the current quarter’s R1, based on the previous quarter’s R1 breakpoints, where R1 is defined as cumulated size-adjusted return in 3-trading-days centered on earnings announcement date for each firm-quarter, so firms with the highest announcement returns belong to decile 1 and firms with the lowest belong to decile 10; UEp*R1p: interaction of UEp and R1p; Size: log of market value of equity measured two days prior to the earnings announcement of quarter t; B/M: book value measured at the end of quarter t divided by market value of equity measured two days prior to the earnings announcement of quarter t; Holdings: average number of shares held by the insiders who trade in this quarter divided by the total number of share outstanding; Momentum: 12-month buy-and-hold return ending 2 days before the earnings announcement date; Pre-net P: net P measured in the window of 6 month before the earnings announcement date; R2: cumulated size-adjusted return beginning 30-day after earnings announcement date and ending 180-day after earnings announcement for each firm-quarter.

are more likely to buy stocks when unexpected earnings are negative, a finding consistent with the findings in Sivakumar and Waymire (1994). It is also interesting that the proportion of quarters with sales is larger for the middle deciles (decile 6 and decile 7) than for the extreme deciles (decile 1 and decile 10). Compared with decile 1, all other deciles have significantly more quarters with insider sales as indicated by the significant t-statistics. This pattern suggests that insiders are more likely to sell stocks when the current quarter’s earnings outperform that of quarter’s t  4, and the outperformance is relatively small. Descriptive statistics for all variables are shown in Table 3. Given the results in Table 2, it is not surprising to observe a large number of quarters with Net P equal to zero. The mean of Net P is negative, which is what one would expect given the personal liquidity and portfolio rebalancing needs of insiders. In addition, in our sample the unreported mean of number of purchases and sales are 0.16 and 0.43, respectively. These numbers contrast with those reported in Table 1 of Lakonishok and Lee (2001), where the mean number of management purchases and sales are 2.77 and 4.74, respectively. There are two reasons for this seemingly large difference. First, we measure the number of transactions within one-month window of an earnings announcement, but they measure the number of transactions within a one-year window. Second, they include private transactions, while we include only open market transactions. The descriptive statistics on size and B/M indicate that the two variables are approximately normally distributed. The Pearson and Spearman correlation coefficients presented in Table 4 support the insider awareness hypothesis (H1). Consistent with the predictions, correlations between Net P and UEp*R1p are significantly positive. These correlations suggest that overall, insiders tend to buy (sell) when earnings news is good (bad) and the market’s reaction to the earnings announcements is small in absolute magnitude. It is possible, however, that the impact of R1p drives the results of the univariate test, so it is necessary to perform a multivariate analysis. As predicted, the correlations between UEp*R1p and its components, UEp and R1p, are very high (0.6–0.7). That again calls for the control of the two direct

38

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44

Table 4 Correlation among variables.

Net P UEp R1p UEp*R1p Size B/M Holdings Momentum Pre-net P R2

Net P

UEp

R1p

UEp*R1p

Size

B/M

Holdings

Momentum

Net P Pre

R2

1.000 0.041* 0.055* 0.010* 0.138* 0.110* 0.130* 0.120* 0.351* 0.041*

0.044* 1.000 0.000 0.665* 0.000 0.060* 0.007* 0.238* 0.020* 0.044*

0.063* 0.000 1.000 0.662* 0.009* 0.015* 0.022* 0.018* 0.009* 0.054*

0.009* 0.680* 0.677* 1.000 0.016* 0.053* 0.015* 0.174* 0.021* 0.061*

0.149* 0.002+ 0.002 0.068* 1.000 0.301* 0.059* 0.114* 0.179* 0.093*

0.156* 0.057* 0.016* 0.080* 0.33* 1.000 0.038* 0.251* 0.149* 0.032*

0.303* 0.011* 0.029* 0.008* 0.227* 0.128* 1.000 0.037* 0.059* 0.001

0.128* 0.282+ 0.017* 0.234* 0.210* 0.316* 0.076+ 1.000 -0.151* 0.012*

0.285* 0.016* 0.015* 0.030+ 0.184* 0.212* 0.166* 0.170* 1.000 0.021*

0.048* 0.044* 0.055* 0.061* 0.081* 0.037* 0.016* 0.032* 0.022* 1.000

This table contains correlations between each of the variables used in subsequent tests. Spearman correlations are above the diagonal, and Pearson correlations are below. To be included in this table, a firm-quarter observation must be accompanied by sufficient data to compute the variables displayed below. Therefore, the statistics for all variables are based on 163,018 firmquarter observations except R2, R1p, UEp*R1p and momentum. Firm-quarter observations are drawn from the period between 1980 and 1997. Variable definitions: Net P: Net number of purchases is calculated as (number of purchases  number of sales), where number of purchases is number of insider purchases that occur for a firm-quarter within thirty days after an earnings announcement and Number of sales is number of insider sales that occur for a firm-quarter within thirty days after an earnings announcement; UEp: Decile rank by unexpected earnings, which is measured as actual earnings of quarter t less actual earnings of quarter t  4 scaled by stock price two days before the earnings announcement of quarter t. We use the previous quarter’s breakpoints in unexpected earnings decile breakpoints to assign the rank; R1p: Inverse decile rank of the current quarter’s R1, based on the previous quarter’s R1 breakpoints, where R1 is defined as cumulated size-adjusted return in 3-trading-days centered on earnings announcement date for each firm-quarter, so firms with the highest announcement returns belong to decile 1 and firms with the lowest belong to decile 10; UEp*R1p: interaction of UEp and R1p; Size: log of market value of equity measured two days prior to the earnings announcement of quarter t; B/M: book value measured at the end of quarter t divided by market value of equity measured two days prior to the earnings announcement of quarter t; Holdings: average number of shares held by the insiders who trade in this quarter divided by the total number of share outstanding; Momentum: 12-month buy-and-hold return ending 2 days before the earnings announcement date; Pre-net P: net P measured in the window of 6 month before the earnings announcement date; R2: cumulated size-adjusted return beginning 30-day after earnings announcement date and ending 180-day after earnings announcement for each firm-quarter. * Denotes two-tailed significance at the .01 levels, respectively. + Denotes two-tailed significance at the .05 levels, respectively.

effects in a multivariate analysis. Other correlations are generally smaller than 0.3 in absolute magnitude. 4.2. Hypothesis test results7 4.2.1. The insider awareness hypothesis (H1) We estimate Eq. (1) and present the results in Table 5. All our coefficient estimates have the same sign as predicted. The direct effect of UEp is negative and significant, just as was found in Sivakumar and Waymire (1994). Thus if we ignore the price reaction to the earnings announcement, insiders will on average sell after good earnings news and buy after bad news. This result indicates that market underreaction is not the sole driver of insider trading in our sample. However, the coefficient on UEp*R1p is positive and statistically at the 1% level. Thus when earnings are unexpectedly high and the price reaction to earnings news is low, insiders engage in more net buying than when earnings are unexpectedly low and the price reaction high. We conclude that insiders take into account underreaction to earnings when determining their trading strategies. To understand the economic significance of our findings, we consider two extreme cases: the case where earnings news is bad and the price reaction is high (UEp is 1 and R1p is 1) compared to the case 7 The hypothesis test results are based on Fama–MacBeth regressions and we reported Newey–West corrected t-statistics in Tables 5–8. Our results are robust to firm fixed effects regression by pooling all firm-quarters together.

39

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44 Table 5 Regression analysis for hypothesis 1. Expected Sign

Net P

Intercept

?

UEp



R1p

+

UEp*R1p

+

Size



B/M

+

Holdings



Momentum



Pre-net P

+

0.272*** (5.95) 0.024*** (5.25) 0.022*** (7.31) 0.002*** (3.58) 0.102*** (9.49) 0.037*** (5.08) 1.010*** (10.95) 0.143*** (5.38) 0.118*** (21.42) 17.5%

Adjusted R2

This table presents the time-series means of coefficients and adjusted R2s produced by quarterly cross-sectional regressions using the model (1), i.e., these results are based on quarterly Fama–MacBeth regressions using a total of 162,768 observations. This table contains parameter estimates and Newey–West corrected t-statistics computed by taking the sample average of parameters estimated in 72 cross-sectional regressions, one for each calendar quarter on our sample. The time period is from 1980 to 1997. Coefficients on 49 industry indicators are not reported.

Net P ¼ a þ b1  UEp þ b2  R1 p þ b3  UEp  R1 p þ b4  Size þ b5  B=M þ b6  Holdings þ b7  Momentum X bi  Industryi þ e; þ b8  Net P Pre þ Variable definitions: Net P: net number of purchases is calculated as (number of purchases  number of sales), where number of purchases is number of insider purchases that occur for a firm-quarter within thirty days after an earnings announcement and Number of sales is number of insider sales that occur for a firm-quarter within thirty days after an earnings announcement; UEp: decile rank by unexpected earnings, which is measured as actual earnings of quarter t less actual earnings of quarter t  4 scaled by stock price two days before the earnings announcement of quarter t. We use the previous quarter’s breakpoints in unexpected earnings decile breakpoints to assign the rank; R1p: inverse decile rank of the current quarter’s R1, based on the previous quarter’s R1 breakpoints, where R1 is defined as cumulated size-adjusted return in 3-trading-days centered on earnings announcement date for each firm-quarter, so firms with the highest announcement returns belong to decile 1 and firms with the lowest belong to decile 10; UEp*R1p: interaction of UEp and R1p; Size: Log of market value of equity measured two days prior to the earnings announcement of quarter t; B/M: book value measured at the end of quarter t divided by market value of equity measured two days prior to the earnings announcement of quarter t; Holdings: average number of shares held by the insiders who trade in this quarter divided by the total number of share outstanding; Momentum: 12-month buy-and-hold return ending 2 days before the earnings announcement date; Pre-net P: net P measured in the window of 6 month before the earnings announcement date. *** Denotes two-tailed significance at the .01 level.

where earnings news is good and the price reaction is small (UEp is 10 and R1p is 10). Holding other variables constant, the difference in net purchases between the two cases is going to be:

EðNetP=UEp ¼ 10; R1 p ¼ 10Þ  EðNetP=UEp ¼ 1; R1 p ¼ 1Þ ¼ b1 DUEp þ b2 DR1 p þ b3 DðR1 p  UEpÞ ¼ 0:024ð10  1Þ þ 0:022ð10  1Þ þ 0:002ð10  10  1  1Þ ¼ 0:18 Thus net purchases are on average higher by 0.18 when a stock has underreacted to an extreme positive earnings shock (UEp = 10, R1p = 10) than when it has underreacted to a negative earnings shock (UEp = 1, R1p = 1). An increase in Net P of 0.18 is economically large compared with the mean of Net P, which is 0.29. The control variables also turn out to have the same sign as predicted and are consistent with prior literature. For example, the positive coefficient on B/M is consistent with Jenter (2005), and the

40

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44

negative coefficient on Momentum supports the notion that insiders are contrarian on average, selling after periods of positive returns and buying after periods of negative returns. Finally positive sign on Net P Pre indicates that the short-swing rule impacts insider trading. Table 6 presents the results from estimating Eq. (2) in subsamples. The results in this table are more intuitive, since we do not have the interaction variable based on decile ranks. For the extreme good earnings news subsample, the coefficient on R1p is significantly positive. This result indicates that when the price reaction is small (the R1p is inversely ranked) and unexpected earnings are high, insiders make more purchase transactions and fewer sale transactions. For the extreme bad earnings news subsample, the coefficient on R1p is also significantly positive. This result indicates that insiders facing extreme bad news and a price reaction that is close to zero or positive tend to sell more and buy less. In addition, further unreported test indicates that the difference of coefficients on R1p in the two subsamples is statistically significant. The results in Table 6 are also economically significant. In the good earnings news subsample, the coefficient on R1p is 0.028. Thus net purchases, after a good earnings announcement, will tend to be higher by 0.252 when the price reaction is low (R1p = 10) relative to when it is high (R1p = 1). Together, the results presented in Tables 5 and 6 provide strong evidence that insiders adjust their trades to accommodate perceived underreaction to earnings news, though our results do not rule out that other factors might affect insider trading.

Table 6 Regression analysis for hypothesis 1 – subsamples. Expected Sign

Top two deciles

Expected Sign

Bottom two deciles

Intercept

?

?

R1p

+

Size



B/M

+

Holdings



Momentum



0.063 (0.66) 0.028*** (6.81) 0.102*** (6.83) 0.048*** (2.95) 1.165*** (7.19) 0.139*** (4.96) 0.118*** (19.43) 19.3%

0.055 (0.48) 0.015*** (5.82) 0.064*** (6.03) 0.017 (1.49) 0.008 (0.14) 0.134** (2.65) 0.098*** (11.65) 12.2%

Pre-net P

+ 2

Average adjusted R

+  +   +

This table presents the time-series means of coefficients and adjusted R2s produced by quarterly cross-sectional regressions using the model (2) shown above, i.e., these results are based on quarterly Fama–MacBeth regressions using a total of 32,218 and 32,232 observations for top two deciles and bottom two deciles, respectively. This table contains parameter estimates and Newey–West corrected t-statistics computed by taking the sample average of parameters estimated in 72 cross-sectional regressions, one for each calendar quarter on our sample. The time period is from 1980 to 1997. Coefficients on 49 industry indicators are not reported.

Net P ¼ a þ b1  R1p þ b2  Size þ b3  B=M þ b4  Holdings þ b5  Momentum þ b6  Net P Pre þ

X

bi  Industryi þ e;

Net P: net number of purchases is calculated as (number of purchases  number of sales), where number of purchases is number of insider purchases that occur for a firm-quarter within thirty days after an earnings announcement and number of sales is number of insider sales that occur for a firm-quarter within thirty days after an earnings announcement; R1p: inverse decile rank of the current quarter’s R1, based on the previous quarter’s R1 breakpoints, where R1 is defined as cumulated sizeadjusted return in 3-trading-days centered on earnings announcement date for each firm-quarter, so firms with the highest announcement returns belong to decile 1 and firms with the lowest belong to decile 10; Size: log of market value of equity measured two days prior to the earnings announcement of quarter t; B/M: book value measured at the end of quarter t divided by market value of equity measured two days prior to the earnings announcement of quarter t; Holdings: average number of shares held by the insiders who trade in this quarter divided by the total number of share outstanding; Momentum: 12-month buy-and-hold return ending 2 days before the earnings announcement date; Pre-net P: net P measured in the window of 6 month before the earnings announcement date. ** Denotes two-tailed significance at the .05 level. *** Denotes two-tailed significance at the .01 level.

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44

41

4.2.2. The insider returns hypothesis (H2) Table 7 presents estimates of the parameters of regression model 4. Notice that the estimated coefd is positive and highly statistically significant in both specifications, with and withficient on the NetP out industry fixed effects. We thus infer that the incremental trades made in response to perceived underreaction to earnings news enhance the abnormal returns earned by insiders. The magnitude of the coefficient in model 4 is also economically significant. Taking a value of 0.074, it implies that a net insider purchase made in response to perceived market underreaction to earnings news enhances the abnormal returns to an insider’s trading strategy by 7.4% during the 30 to 180 day window following the earnings announcement. d is calculated using parameters estimated from the full sample of data, its coefficient Since NetP estimate in regression model 3 is potentially susceptible to look-ahead bias. We do not, however, use any return data over any substantial time horizon to compute the estimates. Furthermore, our purpose is not to test a trading strategy that an outsider might have replicated in the past, but to see whether a component of insiders’ overall trading strategy has increased their abnormal returns. Hence we do not believe the potential of look-ahead bias is of concern. Table 7 Regressions for hypothesis 2. Expected sign

R2 (1) Based on Eqs. (3) and (4)

R2 (2) Added industry dummies into Eqs. (3) and (4)

Intercept

?

d NetP

+

0.019*** (3.21) 0.074***

0.012 (0.97) 0.080***

Size



B/M

+

Momentum

+

(5.48) 0.000*** (6.72) 0.022*** (3.90) 0.021*** (3.49)

(6.25) 0.000** (2.02) 0.024*** (4.79) 0.016*** (3.03)

Average adjusted R2

1.7%

7.4% 2

This table presents the time-series means of coefficients and adjusted R s produced by quarterly cross-sectional regressions using the model shown above, i.e., these results are based on quarterly Fama–MacBeth regressions using a total of 162,768 observations. This table contains parameter estimates and Newey–West corrected t-statistics computed by taking the sample average of parameters estimated in 72 cross-sectional regressions, one for each calendar quarter on our sample. In column (1), d corresponds to fitted values of NetP generated from coefficients estimated using regression model (1), the results of which NetP are presented in Table 5. In column (2), industry effects are added into both Eqs. (3) and (4), although the coefficients on d indicates that the portion of insider trading strategy industries are not reported to conserve space. A positive coefficient on NetP attributable to public information about earnings and the price reaction to the earnings announcement generates positive abnormal returns.

d þ b  Size þ b  B=M þ b Momentum þ e R2 ¼ a þ b1  NetP 2 3 4 where

^  UEp þ b ^ R pþb ^  UEp  R p d ¼a ^þb NetP 1 2 1 3 1 Variable definitions: UEp: Decile rank by unexpected earnings, which is measured as actual earnings of quarter t less actual earnings of quarter t  4 scaled by stock price two days before the earnings announcement of quarter t. We use the previous quarter’s breakpoints in unexpected earnings decile breakpoints to assign the rank.; R1p: Inverse decile rank of the current quarter’s R1, based on the previous quarter’s R1 breakpoints, where R1 is defined as cumulated size-adjusted return in 3-tradingdays centered on earnings announcement date for each firm-quarter, so firms with the highest announcement returns belong to decile 1 and firms with the lowest belong to decile 10; UEp* R1p: Interaction of UEp and R1p; Size: Log of market value of equity measured two days prior to the earnings announcement of quarter t; B/M: Book value measured at the end of quarter t divided by market value of equity measured two days prior to the earnings announcement of quarter t; Momentum: 12-month buyand-hold return ending 2 days before the earnings announcement date; R2: Cumulated size-adjusted return beginning 30-day after earnings announcement date and ending 180-day after earnings announcement for each firm-quarter. *Denotes two-tailed significance at the .1 level. ** Denotes two-tailed significance at the .05 level. *** Denotes two-tailed significance at the .01 level.

42

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44

5. Sensitivity analyses In this section, we test whether our results are sensitive to specification (Section 5.1), as well as alternative definitions of key variables and other miscellaneous robustness checks (Section 5.2). 5.1. Specification Piotroski and Roulstone (2004) find that the relation between insider trading decisions and future earnings information is non-linear. In addition, Net P is a truncated integer variable, whereas our linear specification assumes a continuous dependent variable, potentially resulting in biased and inconsistent parameter estimates. We thus test H1 using a Tobit model by setting the lower bound and

Table 8 Regression analysis for hypothesis 1 – subsamples by institutional ownership. Expected sign

High ownership net P

Low ownership net P

Intercept

?

UEp



R1p

+

UEp*R1p

+

Size



B/M

+

Holdings



Momentum



Pre-net P

+

0.241*** (3.71) 0.027*** (5.04) 0.023*** (4.91) 0.002*** (3.03) 0.108*** (10.30) 0.032*** (3.60) 1.107*** (9.36) 0.158*** (5.97) 0.117*** (20.53) 18.6%

0.130 (0.91) 0.021** (2.48) 0.013* (1.83) 0.002** (2.05) 0.077*** (6.16) 0.070*** (7.10) 0.698*** (3.39) 0.092*** (2.77) 0.108*** (11.76) 17.1%

2

Adjusted R

This table presents the time-series means of coefficients and adjusted R2s produced by quarterly cross-sectional regressions using the model (1), i.e., these results are based on quarterly Fama–MacBeth regressions using the two subsamples by institutional ownership (high vs. low). This table contains parameter estimates and Newey–West corrected t-statistics computed by taking the sample average of parameters estimated in 72 cross-sectional regressions, one for each calendar quarter on our sample. The time period is from 1980 to 1997. Coefficients on 49 industry indicators are not reported.

NetP ¼ a þ b1  UEp þ b2  R1 p þ b3  UEp  R1 p þ b4  Size þ b5  B=M þ b6  Holdings þ b7  Momentum X bi  Industryi þ e; þ b8  Net P Pre þ Variable definitions: Net P: net number of purchases is calculated as (number of purchases  number of sales), where number of purchases is number of insider purchases that occur for a firm-quarter within 30 days after an earnings announcement and number of sales is number of insider sales that occur for a firm-quarter within thirty days after an earnings announcement; UEp: decile rank by unexpected earnings, which is measured as actual earnings of quarter t less actual earnings of quarter t  4 scaled by stock price two days before the earnings announcement of quarter t. We use the previous quarter’s breakpoints in unexpected earnings decile breakpoints to assign the rank; R1p: inverse decile rank of the current quarter’s R1, based on the previous quarter’s R1 breakpoints, where R1 is defined as cumulated size-adjusted return in 3-trading-days centered on earnings announcement date for each firm-quarter, so firms with the highest announcement returns belong to decile 1 and firms with the lowest belong to decile 10; UEp*R1p: interaction of UEp and R1p; Size: log of market value of equity measured two days prior to the earnings announcement of quarter t; B/M: book value measured at the end of quarter t divided by market value of equity measured two days prior to the earnings announcement of quarter t; Holdings: Average number of shares held by the insiders who trade in this quarter divided by the total number of share outstanding; Momentum: 12-month buy-and-hold return ending 2 days before the earnings announcement date; Pre-net P: net P measured in the window of 6 month before the earnings announcement date. * Denotes two-tailed significance at the .1 level. ** Denotes two-tailed significance at the .05 level. *** Denotes two-tailed significance at the .01 level.

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44

43

upper bound of Net P at 1 and 1, respectively. The results in Table 5 are robust to this specification and the estimates of the coefficients in this specification have the same signs as in the linear specification, and they are also statistically significant at the 1% level. 5.2. Alternate variable definitions and other robustness checks For the remaining robustness checks, to conserve space most of the results are untabulated and available upon request. Unexpected earnings: We use two alternative methods of defining unexpected earnings: (1) standardized unexpected earnings as defined by Bernard and Thomas (1989); and (2) the difference between actual EPS and the median consensus EPS of analysts forecasts, normalized by price. Our results are identical when we use the Bernard and Thomas definition. When we use analyst forecasts, the coefficient on UEp*R1p in Table 5 is still positive but is now less statistically significant, taking a pvalue of 5.77%. We suspect the reduced significance is due to reduced sample size owing to the fact that many firms in our original sample are not followed by analysts. Insider trading intensity: We use three alternate definitions of insider trading intensity: (1) total net purchases within 15 days after earnings announcement date rather than 30 days; (2) the difference in net purchases as a percent of total insider transactions; and (3) the dollar amount of net purchases standardized by market capitalization. In all three cases, our results are qualitatively similar. Announcement return decile: We redefine R1p using an independent sort, not conditional on earnings surprise deciles. Our results remain the same. Other checks: Ke and Ramalingegowda (2005) show that transient institutional investors trade to take advantage of the post-earnings announcement drift. It is possible that institutional holdings may affect insider trading in a non-linear manner. We thus partition the sample into high institutional ownership vs. low institutional ownership subsamples and re-run our test. The results are similar for both subsamples. We present the results in Table 8. 6. Conclusion This study investigates insider trades following earnings announcements using a sample of quarterly earnings announcements over the period from 1980 to 1997. We find that insiders adjust their trading strategies when it is likely they believe the market underreacts to earnings news. In particular, insiders buy more and sell less when the earnings news is good and the market’s initial return following the earnings announcement is small. We also find that insiders sell more and buy less when earnings news is bad and the market’s initial return is less negative (or more positive). It is clear, therefore, that insiders are trading as though they believe the market underreacts to earnings announcements and are not just trading based on their private information about future earnings. We also find that insider trades made in response to public information about earnings and the price reaction to the earnings announcement generate abnormal returns. Together, these results have important implications for corporate finance. They imply that insiders, on average, can spot instances wherein the market fails to correctly incorporate an important piece of public information into prices. This implication, in turn, bolsters capital structure theories that assume corporate managers are savvy enough about prices in order to successfully use securities offerings and repurchases to exploit mispricing. References Baker, M., Wurgler, J., 2002. Market timing and capital structure. Journal of Finance 57, 1–32. Ball, R., Brown, P., 1968. An empirical evaluation of accounting income numbers. Journal of Accounting Research 6, 159–177. Bernard, V., Thomas, J., 1989. Post-earnings-announcement drift: delayed price response or risk premium? Journal of Accounting Research 27, 1–48. Bernard, V., Thomas, J., 1990. Evidence of stock prices do not fully reflect the implications of current earnings for future earnings. Journal of Accounting & Economics 13, 305–340. Bettis, J., Coles, C., Lemmon, M.L., 2000. Corporate policies restricting trading by insiders. Journal of Financial Economics 57, 191–220.

44

A. Kolasinski, X. Li / J. Account. Public Policy 29 (2010) 27–44

Cheng, S., Nagar, V., Rajan, M., 2007. Insider trades and private information: the special case of delayed disclosure trades. Review of Financial Studies 20, 1833–1864. Fama, E., French, K., 1997. Industry cost of equity. Journal of Financial Economics 43, 153–193. Foster, G., Olsen, C., Shevlin, T., 1984. Earnings releases, anomalies and the behavior of security returns. The Accounting Review 59, 574–603. Frankel, R., Li, X., 2004. Characteristics of a firm’s information environment and the information asymmetry between insiders and outsiders. Journal of Accounting & Economics 37, 229–259. Givoly, D., Palmon, D., 1985. Insider trading and the exploitation of inside information: some empirical evidence. Journal of Business 58, 69–87. Hagerman, R., Healy, J., 1992. The impact of SEC-required disclosure and insider-trading regulations on the bid/ask spreads in the over-the-counter market. Journal of Accounting and Public Policy 11, 233–243. Hu, J., Noe, T., 1997. The insider trading debate. Federal Reserve Bank of Atlanta Economic Review 82 (4), 34–45. Huddart, S., Ke, B., Shi, C., 2007. Jeopardy, non-public information, and insider trading around DEC 10-K and 10-Q filings. Journal of Accounting and Economics 43, 3–36. Jaffe, J., 1974. Special information and insider trading. Journal of Business 47, 410–428. Jeng, L., Metrick, A., Zeckhauser, R., 2003. Estimating the returns to insider trading: a performance-evaluation perspective. Review of Economics and Statistics 85, 453–471. Jenter, D., 2005. Market timing and managerial portfolio decisions. Journal of Finance 60, 1903–1949. Kahle, K., 2000. Insider trading and the long-run performance of new security issues. Journal of Corporate Finance 6, 25–53. Karpoff, L., Lee, D., 1991. Insider trading before new issue announcements. Financial Management 20, 18–26. Ke, B., Ramalingegowda, S., 2005. Do institutional investors exploit the post-earnings announcement drift? Journal of Accounting & Economics 39, 25–55. Ke, B., Huddart, S., Petroni, K., 2003. What insiders know about future earnings and how they use it: evidence from insider trades. Journal of Accounting & Economics 35, 315–346. Kothari, S., 2001. Capital markets research in accounting. Journal of Accounting & Economics 31, 105–231. Lakonishok, J., Lee, I., 2001. Are insider trades informative? Review of Financial Studies Spring 79, 111. Lin, J., Howe, J., 1990. Insider trading in the OTC market. Journal of Finance 45, 1273–1284. Lustgarten, S., Mande, V., 1995. Financial analysts’ earnings forecasts and insider trading. Journal of Accounting and Public Policy 14, 233–261. Lustgarten, S., Mande, V., 1998. The effect of insider trading on financial analysts’ forecast accuracy and dispersion. Journal of Accounting and Public Policy 17, 311–327. Ng, J., Rusticus, T., Verdi, R., 2008. Implications of transaction costs for the post-earnings announcement drift. Journal of Accounting Research 46, 661–696. Noe, C., 1999. Voluntary disclosures and insider transactions. Journal of Accounting & Economics 27, 305–326. Park, M., Park, T., 2004. Insider sales and earnings management. Journal of Accounting and Public Policy 23, 381–411. Piotroski, J., Roulstone, D., 2004. The influence of analysts, institutional investors, and insiders on the incorporation of market, industry, and firm-specific information in stock prices. The Accounting Review 79, 1119–1151. Ronen, J., Tzur, J., Yaari, V., 2006. The effect of directors’ equity incentives on earnings management. Journal of Accounting and Public Policy 25, 359–389. Roulstone, D., 2004. The relation between insider-trading restrictions and executive compensation. Journal of Accounting Research 20, 551–578. Schleifer, A., Vishny, R., 2003. Stock market driven acquisitions. Journal of Financial Economics 70, 295–311. Seyhun, N., 1986. Insiders’ profits, costs of trading, and market efficiency. Journal of Financial Economics 16, 189–212. Seyhun, N., 1988. The information content of aggregate insider trading. Journal of Business 61, 1–24. Seyhun, N., 1990. Overreaction or fundamentals: some lessons from insiders’ response to the market crash of 1987. Journal of Finance 45, 1363–1368. Sivakumar, K., Waymire, G., 1994. Insider trading following material news events: evidence from earnings. Financial Management 23, 23–32. Watts, R., 1978. Systematic abnormal returns after quarterly earnings announcements. Journal of Financial Economics 6, 127– 150.