When Are Insider Trades More Informative?

When Are Insider Trades More Informative? ABSTRACT Using a comprehensive insider trading database, we document that US corporate insiders are more likely to sell rather than to buy as the stock price moves closer to the 52-week high. When insiders trade, they buy (sell) disproportionally more as the stock price moves closer to its 52-week low (high). We interpret these findings as supportive evidence that insiders anchor on the 52-week low (high) for purchases (sales). More importantly, we find that the subsequent 60-day stock returns following insider purchases are 4.0 percent higher when such purchases take place when the stock price is far away from its 52-week low than when the stock price is close to its 52-week low. In contrast, the subsequent 60-day stock returns following insider sales are 2.6 percent lower for insider sales taking place when the stock price is far away from its 52-week high than when the stock price is close to its 52-week high. Our interpretation is that when insiders buy and sell at times when it is suggested otherwise by the anchoring bias, insiders have overcome the anchoring bias because of the private information they have access to. Thus, insider trades that are conducted when the stock price is far away from its anchor level are more informative. JEL Classification: G11, G14, G15 Key words: Insider trade, anchoring bias, 52-week high and 52-week low, private information 1

1. Introduction The field of psychology documents that human beings are subject to many psychological biases such as overconfidence and representativeness heuristic etc. Applying such psychological biases to the financial markets has become increasingly popular with the surging behavioral finance research. Many empirical studies have tested the existence of such biases among different kinds of investors and examined the economic consequences of such biases. 1 These studies have generated significant insights into how the financial market is shaped or affected by behavioral biases. This paper focuses on one particular psychological bias: anchoring. Anchoring refers to the psychological pattern that human beings make decisions based on some reference points they have experienced in the past. These reference points are usually some seemingly relevant anchoring variables. The psychology literature provides ample evidence that individuals anchor on salient but often irrelevant information in their decision-making process. Kahneman and Tversky (1974) provides a psychological foundation for the anchoring effect. Since these decisions are not necessarily driven by rational considerations, the anchoring effect is often referred to as anchoring bias. Recently, an emerging strand of literature has applied this notion to the field of investments and examined the implication of anchoring bias on investor behavior and the financial markets. 2 Analysis of the anchoring bias requires the identification of the conditioning variable, i.e., the anchor. The anchor variable should be readily available so that investors can have access to such information without expending too much effort. In this regard, saliency of the anchor variable is often very much desired since the more salient the conditioning information is, the more likely such information stands out from the crowd and gets chosen. 1 See Baker and Nofsinger 2010 and the references therein for a review of the behavioral finance literature. 2 A list of anchoring bias literature includes: Northcraft and Neale 1987, Heath et al. 1999, Grinblatt and Keloharju 2001, Lambson et al. 2004, George and Hwang 2004, Huddart et al. 2009, Driessen et al. 2012, Baker et al. 2012, George et al. 2013 etc. We briefly review the literature on anchoring bias in the financial market in Section 2. 2

While there are many candidates for the anchoring variable, a commonly identified anchor in the stock market is the highest and lowest stock prices in the prior 52 weeks, i.e., the 52-week highs and lows. The 52-week highs and lows are widely reported and can be easily retrieved from financial information outlets such as the Wall Street Journal, Bloomberg, Yahoo Finance, and Financial Times. Stocks that are approaching their 52-week highs and lows also receive extensive media coverage. This suggests that a stock s 52-week highs and lows are salient information and speedily available in the financial world. In addition, anecdotal evidence suggests that investors refer to the highest and lowest points of stock price in the past 52 weeks when they trade. It is for this reason that the 52-week highs and lows conveniently meet the anchor criteria. Existing studies on anchoring bias have greatly enhanced our understanding of the prevalence of anchoring bias in various financial markets and its implications on investor behavior and stock market response to information events. However, very few of them have examined anchoring bias within the context of insider trading. This paper turns to insider trading to study anchoring bias. We argue that insider trading provides a very rich setting for anchoring bias research for a number of reasons. First, on one hand, like other human beings, corporate insiders can suffer from anchoring bias; on the other hand, corporate insiders are actively involved in the operations and management of their firms, which entitles them to private information about the firms. As a matter of fact, corporate insiders are often believed to enjoy certain informational advantage over outside investors. Such informational advantage may mitigate the effect of anchoring bias since insiders are in a better position to value the firms. Second, insider trades can be driven by information reasons or behavioral biases. The interaction between information-driven insider trading and bias-driven insider trading is intriguing. Disentangling these trading motives may shed lights on which insider trades outside investors can learn the most from. Third, insiders are now required to file their trading information in a timely fashion. The massive reported insider trades that have become increasingly pervasive, along with granular information on the transaction prices and shares traded, provides us with fertile ground to examine anchoring bias. 3

Our paper starts with an investigation of whether corporate insiders are subject to anchoring bias. Anchoring bias implies that insiders are less likely to buy when the stock price moves away from its 52-week low than when the stock price gets closer to its 52-week low given that their valuation is anchored at its 52-week low. Similarly, insiders are less likely to sell when the stock price moves away from its 52-week high than when the stock price gets closer to its 52-week high given that their valuation is anchored at its 52-week high. This is precisely what we find in the data. Using a comprehensive insider trading database, we show that corporate insiders exhibit strong anchoring bias when they trade. As far as the trading direction is concerned, insiders are more likely to sell as opposed to buy when the stock price gets closer to its 52-week high. When insiders actually trade, they buy disproportionally more as the stock price moves closer to its 52-week low and sell disproportionally more as the stock price moves closer to its 52-week high. Our interpretation is that insiders use the 52-week low as the anchor when they buy and the 52-week high as the anchor when they sell. Under the influence of anchoring bias, insiders are less likely to buy or sell when the stock price moves away from its anchor level. This is simply because it would be ill timed for insiders to buy when the stock price is away from its 52-week low. It would be equally ill timed for insiders to sell when the stock price is away from its 52-week high. While this logic seems quite intuitive, adding information-driven insider trades on top of it allows us to form very interesting expectations about the subsequent stock returns following insider trades when the stock price moves away from its anchor level. More specifically, if insiders have positive private information about their firms, they might presumably overcome the anchoring bias and buy even when the stock price is far away from its 52-week low. Similarly, if insiders have negative private information about their firms, they might presumably overcome the anchoring bias and sell even when the stock price is far away from its 52-week high. Thus, the confounding effect of anchoring bias and information motives for insider trades allows us to derive testable hypotheses about subsequent stock returns following insider trades. If insiders overcome their anchoring bias due to their access to private information and buy when the stock price is far away 4

from its 52-week low, such purchases are more likely to be driven by positive information. Consequently, we expect more positive stock returns following insider purchases when such purchases take place when the stock price is far away from its 52-week low than when the stock price is close to its 52-week low. Similarly, if insiders overcome their anchoring bias due to their access to private information and sell when the stock price is far away from its 52-week high, such sales are more likely to be driven by negative information. As a result, we expect more negative stock returns following insider sales when such sales take place when the stock price is far away from its 52-week high than when the stock price is close to its 52-week high. The insider trading database we use lends strong support to this conjecture. We find that the subsequent 60- day stock returns following insider purchases are 4.0 percent higher for insider purchases taking place when the stock price is far away from its 52-week low than when the stock price is close to its 52-week low. In contrast, the subsequent 60-day stock returns following insider sales are 2.6 percent lower for insider sales taking place when the stock price is far away from its 52-week high than when the stock price is close to its 52-week high. This result is robust to alternative abnormal return measurements and stock return horizons. Our interpretation is that when insiders buy and sell at times when it is suggested otherwise by the anchoring bias, insiders have overcome the anchoring bias because of the private information they have access to. Thus, insider trades that are conducted when the stock price is far away from its anchor level are more informative. Our paper is closely related to Lee and Piqueira 2016. They also examine whether insiders are subject to anchoring bias. However, their only use the 52-week high as the anchor and does not use the 52-week low. We argue that it is important to examine the role of both 52-week highs and 52-week lows as the anchor variables. Both 52-week highs and lows are readily available from the financial media outlets and are salient information that can serve as the reference points for investors decision making. The human nature of buying low and selling high implies that investors in general will refer to the 52-week high if they want to sell and the 52-week low if they want to buy. 5

While our findings in this paper are consistent with Lee and Piqueira 2016 in that we both document supportive evidence for the existence of anchoring bias when corporate insiders trade, an important and novel contribution of our paper is our focus on the return implications of insiders overcoming their anchor bias. Observing insider purchases and sales taking place when the stock price is far away its anchor level allows us to be more confident that such insider purchases and sales are more likely to be driven by the private information insiders have access to. Our findings that more positive subsequent returns follow insider purchases when the stock price is far from its 52-week low and more negative subsequent returns follow insider sales when the stock price is far from its 52-week high confirm this logic. Note that the stock s 52-week highs and lows, the stock price nearness to the 52-week highs and lows, as well as the reported insider trades are all public information available to outside investors. Thus, our analysis that insider trades are more informative when the stock prices moves away their anchor levels could potentially benefit outside investors in that they can follow closely and mimic insider trades that are conducted when the stock prices move far away from the two extremes to reap trading profits. Needless to say, our findings about when insider trades are more informative are particularly interesting to outside investors and professional money managers. Our paper stands at the intersection of behavioral finance literature and insider trading research. As far as we know, our paper is one of the few attempts to examine anchoring bias under insider trading context. Our paper is also one of the few papers to advocate the use of both 52-week highs and lows as the anchoring variable. Our findings that insiders exhibit anchoring bias confirm the relevance of anchoring bias. This paper is also among the first of its kind to adopt a behavioral-based approach to insider trading research. Given that insider trading has become more pervasive and that the financial profession has paid increased attention to insider trading, extending the insider trading research along the lines of behavioral finance is important and meaningful. Extant insider trading research has primarily focuses on the rational aspects of insider trading, largely ignoring the possibility that insiders could be subject to any psychological biases just like any other human beings. The existence of anchoring bias among insiders supports such a behavioral 6

perspective. In this regard, our paper complements the behavioral finance literature and potentially opens the door for more studies on the psychological biases of corporate insiders. Perhaps the most important of the contribution of this paper is to the investment profession. Our findings that insider trades are more informative when the stock price moves away from its anchor level add significant insights into professional portfolio managers and outside investors who can piggyback on insider trades to reap substantial trading profits. 4 percent higher abnormal returns on the purchase side and 2.6 percent lower abnormal returns on the sale side combine for a total of 6.6 percent for a 60-day window. Such a trading profit is certainly significant by normal economic metrics. The sizeable abnormal returns following such trades can also shed lights on how regulators can expend their limited resources to curb any illegal insider trading. The rest of the paper is organized as follows. We review the behavioral finance literature on anchoring bias in Section 2. In Section 3 we briefly describe the data and methodologies used for our empirical design. Section 4 presents our main empirical analysis about whether corporate insiders exhibit anchoring bias and the economic implications of such bias, if any. We conclude in Section 5. 2. Literature Review on Anchoring Bias Cognitive psychology argues that the limited supply of cognitive resources leads human beings to resort to common heuristics (i.e., shortcuts) in their decision-making process since such heuristics can help economize the substantial cognitive effort that is required to analyze and process the information that is available and relevant. This could be especially important for the financial market since the it is flooded with enormous information about myriads of financial instruments and markets. Two common heuristics are widely discussed in the psychology literature: substitution and anchoring. Substitution heuristics takes place when human beings replace a difficult question with a much simpler one whose answer requires less cognitive effort (Kahneman 2011). Anchoring heuristics takes place when human beings condition or anchor on certain readily available information when making decisions, even though such information could be irrelevant to their decision problem (Kahneman 1992). 7

An emerging strand of the behavioral finance literature has applied the notion of anchoring bias to the financial markets and examined how anchoring bias affects various market participants and their financial decision making. In a nutshell, such applications of anchoring to the financial markets center upon three main themes: the different financial market segments, significant corporate information events, and investor behavior. In the following section we briefly survey the literature around each of the three themes. Many empirical studies have documented evidence consistent with the existence of anchoring bias in the equity market, the options market, the real estate market and the credit market. Huddart, Lang, and Yetman (2009) document volume spikes in the stock market when a stock passes a 52-week high. In an influential paper, George and Hwang (2004) find that a strategy taking a long position in stocks near the 52-week highs and a short position in stocks far from the 52-week highs generates abnormal future returns. Driessen, Lin, and Van Hemert (2012) document that option implied volatility increases after stock prices rise above the 52-week high. In the real estate market, Northcraft and Neale (1987) and Lambson et al. (2004) document evidence that investors in the real estate market are likely to be subject to the anchoring bias. Dougal et al. (2015) examine anchoring bias in the credit markets. They find that if credit spreads have declined from the firms past loan, then firms are charged a higher interest rate than justified by current fundamentals and vice versa. They argue that anchoring to past loan terms can potentially explain their findings. It is worth noting that the majority of existing studies on the anchoring bias in the equity market use only the stock price s 52-week high as the anchoring variable and few of them use the stock price s 52-week low as the reference point. Part of the reason might be that the stock s 52-week high is more eye-catching and salient information. However, to the extent that the only way for investors to generate trading profits is to buy low and sell high, we advocate the use of the 52-week low as the anchor when investors buy and the 52-week high as the anchor when investors sell. From a psychological perspective, using two-sided anchors separately for buying vs. selling makes sense since it seems natural for investors to refer to the 52-week low when they buy. Referring to the 52-week high when investors want to buy would be illogical and far- 8

fetched. Similarly, it appears intuitive for investors to use the 52-week high as the reference point when they want to sell. Moreover, in today s financial market, both 52-week high and 52-week low are equally readily available. For these considerations, we advocate the use of both 52-week high and 52-week low as anchors throughout the paper. Anchoring effect also plays a profound role around significant corporate information events such as mergers and acquisitions and earnings announcements. In the context of mergers and acquisitions, Baker, Pan, and Wurgler (2012) demonstrate that target shareholders are substantially more likely to accept takeover offers if such offers are made at above the target firm s 52-week high. Turning to earnings announcements, George, Hwang, and Li (2013) show that the post-earnings announcement drift occurs when stock prices are anchored near or far from 52-week highs and not otherwise. They argue that it is not the surprise in earnings itself but anchoring on the 52-week high that drives the market s under-reaction to extreme earnings news for individual stocks. A few empirical studies also suggest that anchoring bias affects the behavior of different types of investors. Using the 52-week high stock price as the anchor, Heath, Huddart, and Lang (1999) show that the probability of employee option exercise doubles when the underlying stock price crosses its 52-week high. Grinblatt and Keloharju (2001) document that both retail and institutional investors seem to exhibit anchoring bias in that they are more likely to sell stocks trading at a historical high and buy stocks trading at a historical low. Existing studies on anchoring bias have confirmed the significant role of anchoring bias in investors financial decision making. Our paper attempts to add to the anchoring literature by examining whether corporate insiders exhibit anchoring bias and the economic implications of such anchoring bias for market participants. 3. Data and Methodologies 9

This study utilizes a number of data sources. Stock characteristics such as daily prices and returns are based on the Center for Research in Securities Prices (CRSP) database. In the rest of this section we focus on the details of constructing the final sample of insider trades. Insider trading data are obtained from Thomson Reuters Insider Filing Data Feed (IFDF). The insider trading records are the transactions of persons subject to the disclosure requirements of Section 16(a) of the Securities and Exchange Act of 1934 reported on Form 4 and 5.3 Among the information required on Form 4 are: name and address of the reporting person, issuer name and ticker or trading symbol, relationship of the reporting person to the issuer (officers, directors or other positions held by the reporting persons in issuers), indicator of whether it is a purchase or sale, the date, price and trade size of the transaction. In our empirical tests, we classify insiders into top executives, officers/directors and large block shareholders. To code insiders into any of the three categories, we rely on the identity information of corporate insiders revealed through the role code (role code 1 to role code 4). According to the data manual file of Thomson Reuter IFDF, insiders tend to commit common mistakes and omit pertinent data when filing their documents, and hence, the data has gone through a unique cleansing process. In spite of this, the finance literature has documented a number of data errors in this database. 4 We impose a number of filters to further eliminate such data errors. We require that trading records have a matching CUSIP with data available from CRSP and only open market transactions in equity securities are considered in this sample. Moreover, the transaction price for any reported trades must stay within the daily price range as recorded in CRSP for the corresponding trading day. We further impose a minimum transaction price of five dollars and a minimum share volume of one hundred shares. Following Jeng et al. 3 According the Securities and Exchange Act of 1934, the term corporate insider refers to corporate officers, directors and large shareholders who own more than 10% of the firm s stock. If insiders buy or sell their firm s stock, they are mandated to file with the Securities and Exchanges Commission (SEC) within the first 10 days of the next month after their transactions. Starting from 29 August 2002, insiders are required to report their trades within two business days. 4 See Appendix A in Jeng et al. (2003) for more details. 10

(2003), we also purge duplicate transactions (i.e., those with identical entries in all categories) from the final sample. Table 1 presents the summary statistics for insider trades over our sample period 1986 2015. We slice the 30 years of insider trading into three subsamples and calculate the aggregate number of trades, number of shares reported, and total dollar amount of shares traded. We also calculate the number of insiders who have traded and number of firms that have reported insider trades for each of the three subsample periods. There are many cases where a single insider reports multiple trades on the same day. We aggregate the number of shares traded and calculate the weighted average of the transaction prices in these cases. Panel A: Insider Purchases No. of Trades No. of Shares Dollar Amount of Shares No. of Insiders No. of Firms 1986-1995 2,426 13,624,527 242,737,515 1,177 695 1996-2005 115,723 1,603,129,086 26,208,987,013 30,285 7,011 2006-2015 95,577 2,711,477,907 55,269,748,544 22,582 5,506 Panel B: Insider Sales No. of Trades No. of Shares Dollar Amount of Shares No. of Insiders No. of Firms 1986-1995 4,331 71,241,459 1,940,115,440 1,952 825 1996-2005 362,628 9,858,503,005 348,370,178,140 54,477 7,259 2006-2015 392,799 12,125,782,276 416,473,279,195 46,546 5,680 As we can see from Table 1, insiders have traded a fairly large amount of shares in the most recent two decades. Over the 30-year period, the total dollar amount of shares US corporate insiders have traded adds up to $848 billion dollars. This clearly speaks to the importance of insider trading research. Also notice that the insider trading activity is more pronounced on the sell side than on the buy side. This is consistent with the notion that a lot of insider sales come from shares insiders have acquired through their stock option awards and grants. 4. Empirical Analysis 4.1. Do insiders exhibit anchoring bias when they trade? Our analysis starts with an investigation of whether insiders are subject to anchoring bias when they trade. At first glance, there exist arguments that can both support and challenge the existence of anchoring bias 11

among corporate insiders. On one hand, corporate insiders, like any human beings and economic agents, can be prone to many psychological biases including anchoring bias. On the other hand, insiders are often believed to enjoy much information advantage due to their active involvement in the operations and management of their companies, and hence, they have access to private information about their firms. Consequently, insiders are in a better position to value their firms, which implies that insiders are less likely to exhibit anchoring bias when they trade. Anchoring bias implies that for stocks that are approaching their 52-week high (or alternatively, moving away from their 52-week low), investors are hesitant to believe that the stock prices are likely to go up even further given that investor s perception of the stock valuation is anchored at its 52-week high. On the other hand, for stocks that are far away from its 52-week high, investors are reluctant to believe that the stock prices are likely to go down even further due to their valuation anchoring at its 52-week high. Thus, the anchoring effect is strongest when the stock price is either very close to its 52-week high or far away from its 52-week high. For this reason, the stock price nearness to its 52-week high is often used as a key variable in empirical tests for anchoring bias (George, Hwang, and Li 2013). We follow this empirical methodology and examine the likelihood of insider purchases as compared to insider sales as a function of the stock price nearness to 52-week high by pooling all insider trades (purchases and sales) together. If insiders are subject to anchoring bias, we expect that as the stock price gets closer to its 52-week high, insiders are more likely to sell than to buy given that their valuation of the stock is anchored at 52-week high and that being close enough to the anchor implies a lower probability of breaking the anchoring price level, and hence, it is a good time to sell rather than to buy. This intuition motivates us to examine the insider trade direction as a function of the nearness to its 52-week high. In the following sections we perform both univariate and multivariate tests for anchoring bias. 4.1.1. Univariate tests for anchoring bias 12

To capture insiders propensity to buy or sell, we turn to the fraction of insider purchases or sales relative to the total insider trades. This seems to be a natural choice given that we observe all the reported insider trades and that such trades are conducted at different points in time when the stock price is at different proximity to its 52-week high. Our first set of tests explores whether the fraction of insider purchases relative to the total insider trades changes as the stock price scaled by the 52-week high falls into different ranges. To achieve this purpose, we sort all insider trades into five quintiles based on the stock price nearness to its 52-week high. The stock price nearness is defined as the closing price averaged over the 30-day window before each insider transaction date divided by the 52-week high. For each quintile, we calculate the fraction of insider purchases as a percentage of total insider trades. More specifically, Insider Purchases Insider Purchases %= Insider Purchases+ Insider Sales Insider trading activities are measured along three alternative dimensions: the number of purchases and sales, the total number of shares bought and sold, and the total dollar volume bought and sold. Examining these three dimensions not only captures different aspects of insider trades but also serves as a robustness check. The absence of anchoring bias would suggest that the fraction of insider purchases as a percentage of total insider trading activities should remain relatively constant when the stock price falls into different quintiles. On the other hand, the existence of anchoring bias at the trading direction level implies that the percentage of insider purchases should decrease (increase) as the stock price moves closer to its 52-week high (low). Table 2 presents the frequency distribution of insider purchases as a fraction of all insider trades for each of the five quintiles. 5 Panel A, Panel B, and Panel C examine number of trades, total shares traded and total 5 In our empirical analysis, we also perform robustness check by sorting all insider trades into ten deciles. The patterns documented in Table 2 is robust to this alternative sorting procedure. 13

dollar volume traded respectively. Quintile 1 consists of insider trades when the stock price is far away from its 52-week high (or alternatively, closest to its 52-week low) whereas Quintile 5 consists of insider trades when the stock price is closest to the 52-week high (or alternatively, far away from its 52-week low). Panel A: % of Insider Purchases by No. of Trades Insider Purchases Insider Sales % of Purchases Quintile 1 64,867 131,836 32.98% Quintile 2 51,098 143,598 26.25% Quintile 3 37,684 157,611 19.30% Quintile 4 30,553 163,837 15.72% Quintile 5 29,524 162,876 15.35% Panel B: % of Insider Purchases by Shares Traded Quintile 1 1,613,883,554 4,693,867,611 25.59% Quintile 2 997,190,971 4,110,491,528 19.52% Quintile 3 761,107,280 4,186,315,200 15.38% Quintile 4 533,631,977 4,520,912,667 10.56% Quintile 5 422,417,739 4,543,939,735 8.51% Panel C: % of Insider Purchases by Dollar Volume Traded Quintile 1 22,895,743,268 146,319,142,515 13.53% Quintile 2 18,513,493,551 117,754,931,911 13.59% Quintile 3 16,724,026,005 138,307,141,185 10.79% Quintile 4 12,734,815,505 164,691,001,740 7.18% Quintile 5 10,853,394,743 199,711,355,425 5.15% As we can see from Table 2, insider purchases as a percentage of all insider trades almost decrease monotonically as the stock price moves from far away from the 52-week high to being very close to 52- week high. This pattern is robust to the insider trading activity measured by the number of trades, the number of shares traded and the dollar volume of shares traded. 6 Results in Table 2 suggest that as the stock price is getting closer to its 52-week high, insider purchases account for a lower percentage of their overall trading activity. This is consistent with the notion that insiders are less likely to buy (sell) when the stock price approaches (moves away from) its 52-week high. Overall, results in Table 2 lends preliminary support to anchoring bias. 4.1.2. Multivariate regression tests for anchoring bias 6 Note that since insiders conduct more purchases at low prices than at high stock prices (for instance, no. of purchases amounts to 64,867 for quintile 1 whereas it is only 29,524 for quintile 5), the insider purchase % is much lower when measured by dollar volume traded. 14

While the result from the univariate tests is supportive of the anchoring bias, it is only descriptive and does not control for other factors that can affect insiders decision to buy vs. sell as the stock price draws closer to its 52-week high. In what follows, we aim to examine how insider s trading decision responds in regard to the nearness to 52-week high while controlling for such factors. As we have discussed in the above subsection, anchoring bias implies that as the stock price approaches its 52-week high, insiders are more likely to sell rather than buy if their stock valuation is anchored at 52-week high. On the other hand, as the stock price moves further away from the 52-week high, insiders are more likely to buy rather than sell. Thus, a key prediction from anchoring bias is that the likelihood of insider purchases (sales) should decrease (increase) with respect to the nearness to 52-week high. To test this intuition, we employ a logistics regression framework to examine insiders decision on the trading direction. Our empirical specification is as follows: Prob ( isbuy 1) logit( ) Near52 Near52 IsTopExec Near52 IsOfficDir relshr logprc 0 1 2 3 4 5 1 Prob ( isbuy 1) IsBuy is an indicator variable that takes the value of 1 for insider purchase and 0 for insider sales. Near52 is the stock price nearness to its 52-week high, defined as the average closing price over [t-30, t-1] scaled by the 52-week high, where t is the insider transaction date. We include a number of covariates to control for insider and trade characteristics: IsTopExec is an indicator variable that takes the value of 1 if the insider is a top executive and 0 otherwise; IsOfficDir is an indicator variable that takes the value of 1 if the insider is an officer/director and 0 otherwise; Relshr is the shares traded relative to the daily trading volume on the insider transaction date; Logprc is the natural logarithm of the average closing price over [t-30, t-1]. IsTopExec and IsOfficDir are interacted with Near52 to examine whether different types of insiders exhibit anchoring bias. Relshr and Logprc are included and expected to carry negative signs since we expect that insiders are more likely to sell if their trades account for a larger percentage of the trading volume on that day, or if the average closing price in the 30-day window leading up to the transaction date is higher. Our focal variable is Near52. A negative sign before Near52 is consistent with anchoring effect. 15

Table 3 presents the estimation results. Many firms in our sample have multiple insider trades, hence we calculate firm-clustered standard errors. Variable Estimate Std. Err. Wald Chi-Square Pr > ChiSq Near52-0.454 0.0224 412.265 <.0001 Near52*IsTopExec -1.4449 0.0100 20726.850 <.0001 Near52*IsOfficDir -1.5865 0.0141 12576.910 <.0001 Relshr -0.1874 0.0231 65.8710 <.0001 Logprc -0.2364 0.0054 1950.727 <.0001 Results in Table 3 indicate that all control variables carry the expected signs and are statistically significant at 1 percent level. More importantly, we notice that Near52 carries a negative and highly significant parameter estimate. Thus, insiders are less likely to buy (i.e., more likely to sell) as the stock price gets closer to 52-week high, which is consistent with the predictions of anchoring bias. On a side note, we observe that the parameter estimates before the interaction terms between Near52 and IsTopExec and IsOfficDir are negative and strongly significant, suggesting that both top executives and officers/directors exhibit larger degrees of anchoring bias as compared to large shareholders. 4.2. Anchoring bias for insider purchases and sales The analysis in the above supports the existence of anchoring bias when insiders trade. We now expand our analysis to examine insider trades along each trading direction. A detailed analysis of the anchoring bias when insiders buy or sell not only advances our analysis of the anchoring bias from the trade direction level to the trade level but also helps confirm or refute the role of both 52-week high vs. 52-week low as the anchoring variables. 4.2.1. Anchoring bias in insider sales We start with insider sales. As we have argued in the previous section, it makes sense to use the 52-week high as the anchoring variable when considering insider sales. Anchoring bias suggests that as the stock price gets closer to its 52-week high, insiders are increasingly reluctant to believe that the stock price will 16

break the anchor level. In other words, insiders should sell more when the stock price is near its 52-week high than when the stock is far away from it. To test this intuition, we examine the frequency distribution of insider sales when the stock price falls into different intervals. More specifically, for each insider sale trade, we first retrieve the highest and lowest stock price in the prior 52 weeks. We then calculate the stock price range in the prior 52 weeks by taking the difference between the 52-week high and 52-week low. The stock price range is then sliced into 10 equal intervals with the length of each interval being (52-week high 52-week low)/10. Each insider sale is then classified into any of the 10 intervals based on the reported transaction price. We then calculate the percentage of insider sales as a fraction of the total number of sales for each interval. Table 4 presents the frequency distributions of insider sales for each of the ten intervals. If insiders are not subject to anchoring bias, we would expect that each of the 10 intervals should contain about 10 percent of the insider sales. On the other hand, insiders will sell disproportionally more when the stock price gets closer to its 52-week high if they are prone to anchoring bias. Thus, higher percentages of insider sales concentrating on the higher end of the intervals lend support to the anchoring bias. This is precisely what Table 4 documents. As the stock price gets closer to the 52-week high, insiders seem to be increasingly selling more. Before the stock price moves past the 7 th interval, the percentage of insider sales has been consistently less than 10 percent. As the stock price moves past the 8 th interval and gets closer to its 52-week high, the percentage of insider sales exceeds 10 percent for the first time and reaches the highest when the stock price is closest to its 52-week high. As a matter of fact, the top 4 intervals (intervals 7 to 10) subsume approximately 63 percent of all insider sale trades. This pattern is robust to insider sales measured by the number of trades, or by the number of shares sold, or by the total dollar volume sold. Overall, the results in Table 4 demonstrate that insiders sell disproportionally more when the stock price gets closer to its 52-week high, consistent with their valuation anchoring at 52-week high. 17

Interval No. of Trades No. of Trades (%) No. of shares No. of Shares (%) Dollar Volume Dollar Volume (%) 1 33,298 4.38% 1,589,152,903 7.21% 40,687,826,002 5.31% 2 42,026 5.53% 1,403,601,371 6.36% 41,227,835,089 5.38% 3 46,974 6.18% 1,342,197,214 6.09% 41,572,150,754 5.42% 4 49,006 6.45% 1,369,334,090 6.21% 42,731,529,689 5.57% 5 53,305 7.02% 1,438,803,176 6.52% 44,397,241,496 5.79% 6 59,968 7.89% 1,833,902,071 8.32% 61,936,906,225 8.08% 7 71,356 9.39% 1,881,929,295 8.53% 61,121,733,393 7.97% 8 90,279 11.88% 2,447,888,049 11.10% 84,050,497,516 10.96% 9 126,224 16.61% 3,475,094,970 15.76% 127,872,471,893 16.68% 10 187,329 24.66% 5,273,666,301 23.91% 221,185,822,455 28.85% 4.2.2. Anchoring bias in insider purchases We now turn to examine the presence of anchoring bias when insiders buy. As we have argued before, we use the 52-week low as the anchor. We follow the same methodology and slice the stock price range in the prior 52 weeks into 10 equal intervals. We then classify all insider purchases into any of these 10 intervals based on the reported transaction price. The frequency distribution of insider purchases across the 10 intervals is then calculated. If insiders are free of anchoring bias, we would expect insiders to buy approximately 10 percent of the time for each of the 10 intervals. On the other hand, if insiders are subject to anchoring bias, we would expect insiders to buy disproportionally at the lower ends of the 10 intervals. Table 5 presents the frequency distribution of insider purchases, again measured by the number of purchases, the number of shares purchased, and the total dollar volume of purchases for each of the 10 intervals. Interval No. of Trades No. of Trades (%) No. of shares No. of Shares (%) Dollar Volume Dollar Volume (%) 1 45,427 21.25% 1,032,536,702 23.85% 15,608,383,632 19.10% 2 29,733 13.91% 608,416,072 14.05% 10,070,955,482 12.32% 3 22,621 10.58% 430,004,220 9.93% 7,355,112,801 9.00% 4 18,246 8.54% 337,681,292 7.80% 6,278,560,374 7.68% 5 16,667 7.80% 330,835,910 7.64% 6,328,844,814 7.74% 6 15,980 7.48% 332,567,969 7.68% 7,282,196,724 8.91% 7 15,961 7.47% 322,596,493 7.45% 6,826,855,491 8.35% 8 15,949 7.46% 272,415,435 6.29% 6,910,911,969 8.46% 9 16,458 7.70% 356,719,313 8.24% 7,212,596,548 8.82% 10 16,695 7.81% 305,349,630 7.05% 7,861,926,048 9.62% 18

As we can see from Table 5, insiders buy the most towards the lower ends of the 10 intervals. The bottom three intervals (intervals 1 to 3) collectively account for approximately 46 percent of all insider purchase trades. Again this pattern is robust to alternative measures of insider purchase activities. Overall, the results in Table 5 supports the existence of anchoring bias when insiders buy. 4.3. Return implications of the application of anchoring bias 4.3.1. Deriving the hypotheses The evidence documented in the above is supportive of the anchoring bias when insiders trade. We now turn to the analysis of the economic implications of anchoring bias. After all, unveiling the existence of anchoring bias does not add much value to the investment profession unless it helps shed lights on whether and how outside investors can learn from insider trades. We argue that the existence of anchoring bias actually lays the foundation for our follow-up analysis of when insider trades are more informative. Availing ourselves of interesting stock return dynamics following trades that are driven by anchoring bias and those that are driven by private information, we are able to explore insiders trading motives in more depth and gain additional insights into how outside investors can piggyback on the information-driven trades of corporate insiders. Broadly speaking, the trading motives for corporate insiders can be classified into two main categories: information reasons and non-information reasons such as liquidity considerations and psychological bias. While ex ante we cannot differentiate these trading motives, such trading motives can have their own implications on the trading directions as well as the subsequent stock returns. A careful analysis of the trading directions helps us glean more information about the underlying trading motives. More specifically, if insider trades are entirely driven by information reasons, we expect them to buy if they have positive news and to sell if they have negative news. Consequently, we expect positive stock returns following insider purchases and negative returns following insider sales. On the other hand, anchoring bias is often associated with under-reaction when stock price approaches or moves away from its 52-week high. 19

When the stock price moves close to its 52-week high, usually driven by good news, investors are reluctant to bid the stock price higher even if the information warrants it. Thus, there is under-reaction to good news. However, the positive information will eventually prevail and the stock price goes up. Thus, there is price continuation. Similarly, when the stock price is pushed far away from its 52-week high and draws closer to its 52-week low because of bad news, investors are unwilling to sell the stock at prices that are as low as the information implies. The negative information will eventually prevail and the price goes down. It has been argued that the anchoring effect is the strongest when the stock price is either far away from or close to its anchor. Consequently, our analysis focuses on these two extremes. This being the case, let s consider the following 2 by 2 matrix formed by the direction of insider trades as well as the stock price proximity to the anchor level. Insider trades are classified into purchases and sales whereas the stock price is either far away from or very close to the anchor price. Denote each of the four scenarios by the four matrix elements A, B, C and D respectively as indicated below. Direction of Insider Trades Stock Price Relative to Anchor Level Far Away from Very Close To Buy A B Sell C D Let s start with insider sales where insiders anchor their valuation at the stock s 52-week high. Consider Scenario D first. In this extreme case, the stock price is close to its 52-week high. Anchoring bias suggests that insiders are more likely to sell given that the stock price has moved quite close to its anchor and insiders perceive it to be less likely for the stock price to break the anchor level, and hence, it is a good time to sell. Alternatively, insiders may choose to sell because they have access to negative private information. Thus, while we observe insider sales taking place when the stock price is fairly close to its 52-week high, ex ante it is unclear whether the insider sales are driven by negative private information that insiders have access to or driven by anchoring bias. Insider sales driven by negative information should be followed with negative stock returns. However, anchoring bias in this case suggests that the stock price will continue to 20

drift up due to the under-reaction to good news. Thus, return predictions from anchoring bias driver and private information driver contradicts each other, thus leading to conflicting signals for outside investors. Now consider Scenario C, where insiders sell when the stock price is really far away from its 52-week high. Again consider anchoring bias first. If insiders are subject to anchoring bias, then insiders are less likely to sell given their valuation anchoring at the 52-week high. Thus, if we observe insiders selling when the price is far away from its 52-week high, it is highly likely that insiders have overcome their anchoring bias because of their access to negative private information. Insider sales driven by negative private information should predict negative subsequent returns. In addition, the stock price under-reaction when the stock price moves far away from its 52-week high also suggests subsequent negative returns. Thus, the return predictions from information-driven and bias-driven insider sales reinforce each other and outside investors can be more confident that for insider sales conducted when the stock price moves far away from its 52- week high, such sales are probably driven by insiders who have negative information about the stock and who have overcome the anchoring bias. An intuitive way to summarize the above discussion is to realize the fact that insiders sell when the stock price is far away from the 52-week high essentially implies their overcoming the anchoring bias and revealing their negative private information. Thus, the confounding effect of anchoring bias on top of insider sales suggests that subsequent stock returns should be more negative for Scenario C as compared to Scenario D. Let s now turn to insider purchases. In this case, insiders anchor their valuation at the 52-week low. For the same reason, we examine the two extreme cases when the stock price is either far away from or close to its 52-week low. Now apply the same logic to examine Scenario B. The stock price has drawn very close to its 52-week low. Anchoring bias suggests that insiders should buy since their valuation is anchored at its 52-week low and being close to the anchor level already implies lower probability of breaking the anchor level. Alternatively, insiders may buy in this case because they have access to private positive information about the stock. Thus, ex ante, when outside investors observe insider purchases taking place when the 21

stock price is close to its 52-week low, it is unclear whether such purchases are driven by anchoring bias or private information. The return implications from these two drivers are opposite: bias-driven purchases imply under-reaction and hence price will continue to drop whereas information-driven purchases imply continuous price increase and positive subsequent stock returns. Let s examine Scenario A next. The stock price moves far away from the anchor level (the 52-week low). Thus, anchoring bias implies that insiders under the anchoring bias influence will be less likely to buy since being far away from the anchor level implies that the stock price will possibly fall back to the anchor level and hence, it is not a good time to buy. However, if insiders have access to positive private information and thus overcome the anchoring bias, insiders will end up buying in this case. Ex ante, when insiders buy at a time when it is suggested otherwise by the anchoring bias, outside investors can be more confident that such insider purchases are more likely to be driven by positive private information. To summarize, we want to emphasize the fact that insiders buy when the stock price is far away from the 52-week low essentially implies their overcoming the anchoring bias and revealing their positive private information. Thus, the confounding effect of anchoring bias on top of insider purchases suggests that subsequent stock returns should be more positive for Scenario A as compared to Scenario B. The existence of anchoring bias when corporate insiders trade provides us with a confounding factor on top of the insider trading signaled by their trading direction. Such a factor can facilitate the understanding of the subsequent stock price movements. Note that the stock price proximity to its anchor level and the trading direction of corporate insiders are public information to outside investors, thus, our reasoning in the above could potentially enable outside investors to better piggyback on observed insider trades and reap trading profits by learning from such trades. 4.3.2. Testing the return hypotheses To test the above hypotheses, we first sort all insider purchases and sales into 5 or 10 groups based on the proximity of average stock price in the 30-day window immediately before the insider transaction date 22