Indirect Insider Trading

Brad Goldie◊, Chao Jiang†, Paul Koch‡, M. Babajide Wintoki§

November, 2019

Abstract

Insiders must disclose indirect trades made through accounts they control, including family, trust, retirement, and foundation accounts. We find that trades made in these indirect accounts are more profitable than direct trades in their own accounts. Hedge portfolios replicating large purchases and sales made through family accounts outperform analogous portfolios of trades made through direct accounts. Indirect trades better predict earnings surprises and large price changes, and are associated with opportunistic insiders who make profitable trades before earnings announcements or have a short investment horizon. Insiders are less likely to trade through indirect accounts following periods of intense regulatory scrutiny.

JEL Classification: G12, G14, G18. Key Words: Insider trading, informed trading, information asymmetry, opportunism, trusts, retirement accounts, foundations, family networks.

*◊Miami University, School of Business; [email protected]. †University of South Carolina, Darla Moore School of Business; [email protected]. ‡Iowa State University, Ivy College of Business; [email protected]. §University of Kansas, School of Business; [email protected]. We acknowledge the helpful comments of Ferhat Akbas, Lezgin Ay, Jamie Brown, Gjergji Cici, Matthew Denes, Bob DeYoung, Truong Duong, Tyler Jensen, Hugh Kim, Weikai Li, Felix Meschke, Greg Niehaus, Kevin Pisciotta, Mark Power, Travis Sapp, Christoph Schneider, Xiaolu Wang, seminar participants at Iowa State University, the University of Kansas, the Western Finance Association meetings, the China International Conference in Finance, and the SFS Cavalcade-Asia Conference. Special thanks go to Henk Berkman for his valuable input. Any errors are our own.

I. Introduction

The privilege of corporate insiders’ access to material non-public information attracts the attention of market participants and regulators who wish to know whether insider trades contain predictive information about forthcoming price movements. However, insiders also trade for various reasons not driven by information, including a desire for liquidity, diversification, or corporate control. These alternative rationales make it challenging for market participants and regulators to detect the subset of insiders or their trades that are more likely to signal private information. Thus, while a large body of literature has studied the information content of insider trading, the informational role of insider trades remains an open question.1

Prior studies glean the sample of all publicly disclosed insider trades in an attempt to find groups of unusual trades that contain more predictive information about stock prices. For example, recent work identifies subsets of insiders and their trades that tend to be particularly opportunistic

(i.e., profitable), including non-routine insiders (Cohen, Malloy, and Pomorski, 2012), insiders who trade profitably before earnings announcements (Ali and Hirshleifer, 2017), and insiders with a short investment horizon (Akbas, Jiang, and Koch, 2019).

This article breaks new ground by examining a different group of unusual insider trades.

We compare the information content of direct trades made in the insider’s own account versus indirect trades made in the accounts of family members, trusts, retirement accounts, and foundations. We obtain data on all U.S. corporate insider trades from pdf files containing the individual Form 4s filed by insiders electronically, which are available on the SEC Edgar web site

1 Among others, see Aboody and Lev (2000); Akbas, Jiang, and Koch (2019); Ali and Hirshleifer (2017); Alldredge and Cicero (2015); Cline, Gokkaya, and Liu (2017); Cohen, Malloy, and Pomorski (2012); Fidrmuc, Goergen, and Renneboog (2006); Frankel and Li (2004); Jaffe (1974); Jagolinzer, Larcker, and Taylor (2011); Jeng, Metrick, and Zeckhauser (2003); Jenter (2005); Lakonishok and Lee (2001); Lee, Mikkelson, and Partch (1992); Marin and Olivier (2008); Piotroski and Roulstone (2005); Ravina and Sapienza (2010); Seyhun (1986, 1988, 1992); and Skaife, Veenman, and Wangerin (2013).

1 since July 2003.2 From these pdf files, we also gather information about whether an insider trade is direct (for the insider’s own account), or an indirect trade for a family member, trust, retirement account, or foundation, which is disclosed on the Form 4 in field #7 and its footnotes.

Consider four theoretical rationales for expecting a higher proportion of informed trading through indirect accounts relative to direct accounts. First, indirect accounts represent one way to camouflage opportunistic insider trading. Prior work contends that informed investors disguise their activity by trading when liquidity is high and by splitting large orders into smaller trades.3 In line with this view, insiders might break up their most informed trades and place a disproportionate amount through indirect accounts, to reduce the price impact of such trades. In this case, we would expect indirect trades to be more informed, on average, relative to direct trades.

Second, indirect sales are more likely to contain negative information than direct sales, because insiders are less likely to make uninformed sales through indirect accounts to achieve diversification or liquidity. Insiders typically accumulate large direct shareholdings from their compensation packages. This concentration of wealth motivates direct selling from the insider’s own account to obtain diversification or liquidity, rather than because of bad news. For this reason, most prior work finds that (direct) insider sales are not informed. These motives for uninformed selling are less likely to apply to indirect sales, which are thus more likely to reveal bad news.4

Third, insiders who use their information advantage to build wealth, either directly for themselves or for eventual bequests, may trade through indirect accounts to minimize the impact of personal, estate, or gift taxes. Consistent with this view, Yermack (2009) finds that Chairmen

2 In July 2003, the SEC began requiring insiders to disclose their insider trades by submitting Form 4 electronically. 3 For example, see Admati and Pfleiderer (1988), Barclay and Warner (1993), Chakravarty (2001), Garfinkel and Nimalendran (2003), Hasbrouck (1995), Keim and Madhavan (1995), and Kyle (1985). 4 Some early studies find that insider sales contain negative information about future returns (e.g., see Jaffe, 1974, and Seyhun, 1986). However, more recent work generally finds that only insider purchases contain predictive information about future stock returns (e.g., see Jeng, Metrick, and Zeckhauser, 2003, and Lakonishok and Lee, 2001).

2 and CEO’s donate shares to foundations just before sharp drops in the share price, to maximize their own personal income tax benefits for charitable giving. Dambra, Gustafson, and Quinn

(2019) find that 23 percent of CEOs use tax-advantaged pre-IPO trusts, and share transfers into such trusts are positively associated with CEO equity wealth, estate taxes, and post-IPO stock price appreciation. Brown, Huston, and Wenzel (2018) analyze a small hand-collected sample of CEO stock gifts to family members, and show that the timing of these gifts precedes significant price appreciation over long horizons, thereby avoiding estate and gift taxes. While these three studies only look at limited samples of stock donations, transfers, and gifts, we note that insiders may also use open market purchases and sales through indirect accounts to obtain these tax benefits. For example, they may buy shares for family accounts prior to positive news, to avoid estate or gift taxes. Alternatively, insiders may sell holdings in a trust or retirement account before a price decline, in order to preserve the tax benefits embedded in the accumulated wealth of such accounts.

Fourth, consider indirect trades made through family accounts in particular. The subset of insiders who open accounts and trade on behalf of family members may be more opportunistic than other insiders, and reveal a greater tendency to make informed trades on average. They may also tend to channel a disproportionate amount of these informed trades through the accounts of family.5

We begin by conducting portfolio analysis on all insider purchases or sales, respectively.

Consistent with prior work, calendar time portfolios of stocks bought by all insiders earn a significant four-factor alpha of 0.69 percent in the next month, while insider sales do not outperform. It is noteworthy, however, that the subset of direct purchases made in the insider’s

5 Consistent with this view, Berkman, Koch, and Westerholm (2014) show that the subset of all Finnish investors who open accounts and trade on behalf of young children outperform other retail investors, and tend to make their best trades through the accounts of children. Also, while illegal insider trading through family accounts may seem unlikely, Appendix A documents several recent insider trading cases that involve the accounts of family members.

3 own account have a smaller alpha of 0.65 percent per month, while indirect purchases have a larger alpha of 0.89 percent, significantly outperforming direct purchases by 25 basis points (bps).

When we further analyze different subsets of indirect trades, we find that some outperform the insider’s own direct trades on either the buy side or the sell side, depending on the trade category. For example, portfolios of indirect purchases made in family accounts earn an alpha of

1.09 percent in the following month, significantly outperforming direct purchases by 45 bps, while subsets of family purchases made for a spouse or child earn an even higher alpha of 1.27 or 1.30 percent, respectively, outperforming direct purchases by 63 or 65 bps. On the sell side, indirect sales made in non-family trust or retirement accounts significantly outperform direct sales by -22 or -67 bps.

This evidence is robust when we account for trade size, by considering hedge portfolios that replicate both the large purchases and large sales in each trade category. For example, a hedge portfolio that is long the tercile of stocks with the largest direct insider purchases and short the tercile with the largest direct sales earns 0.35 percent per month. In contrast, the analogous hedge portfolio that replicates large insider purchases and sales made through any indirect account earns

0.85 percent per month, which outperforms the analogous hedge portfolio of direct trades by 49 bps. Furthermore, when we restrict this analysis to large insider purchases and sales made through the account of a spouse or child, this performance increases to 0.90 or 1.05 percent per month, respectively, which outperforms direct trades by 54 or 69 bps.

We next examine the possibility that this outperformance of indirect trades may be due to various attributes of the insider’s firm, which have also been shown to predict returns. We specify a regression model in which the coefficient of the variable for each trade category is analogous to the monthly return on a hedge portfolio that is long the tercile of stocks with the largest insider

4 purchases and short the tercile with the largest sales, for each type of account. The results confirm that different groups of indirect trades continue to significantly outperform direct trades, after controlling for trade size and firm characteristics.

For example, after controlling for firm attributes, a hedge portfolio that replicates large direct purchases and sales earns abnormal returns that are somewhat larger than the alphas from our portfolio analysis, at 0.45 to 0.50 percent per month. However, the analogous hedge portfolio of large purchases and sales made in any indirect account also earns a larger abnormal return of

0.95 percent the following month, which significantly outperforms direct trades by 45 bps.

Furthermore, a similar hedge portfolio of indirect trades made in any family account now earns

1.00 percent per month, which outperforms direct trades by 52 bps, while trades on behalf of a child earn 1.24 percent per month, which outperform by 77 bps. In addition, the hedge portfolio that duplicates large purchases and sales made in a family trust earns 1.05 percent per month, which outperforms direct trades by 54 bps, while the analogous hedge portfolio made through a non- family trust earns 1.28 percent per month, which outperforms by 77 bps.

We next show that this outperformance of indirect trades persists beyond one month and does not reverse over longer horizons. For example, the abnormal one-month-ahead return of 0.50 percent for the hedge portfolio of large direct insider purchases and sales accumulates to 3.10 percent after 36 months. However, the analogous hedge portfolio of indirect family trades accumulates to 7.84 percent after three years, which significantly outperforms the hedge portfolio of direct trades. In addition, after three years, the hedge portfolio of large purchases and sales in non-family trusts earns even more, at 14.95 percent, while the analogous trades made in non-family retirement accounts earns 12.52 percent, both of which significantly outperform direct trades over the same three-year period. Importantly, our analysis reveals no return reversals over this period.

We proceed to consider whether the information content of indirect insider trades reflects information about future firm-specific information events. We find that indirect trades (in particular, family trades) contain more predictive information than direct trades about future quarterly earnings surprises and large idiosyncratic price changes. These results suggest that insiders strategically place their trades through family accounts prior to imminent firm events.

Consequently, the superior performance of indirect trades (especially, trades in family accounts) arises at least partially from inside information about future firm-specific informational events.

Prior research suggests that insiders are less likely to make informed trades when the risk of doing so is made salient, as it is during periods of intense regulatory scrutiny of insider trading by the SEC (Cohen, Malloy, and Pomorski, 2012, Del Guercio, Odders-White, and Ready, 2017).

We thus investigate whether, or not, intensified SEC enforcement activity deters insider trading through indirect accounts or family accounts. We find that insiders make relatively fewer trades in indirect accounts (in particular, family accounts) following months with more releases of litigation cases by the SEC against illegal insider trading. This outcome suggests that insiders are more reluctant to make opportunistic trades through indirect accounts or family accounts when litigation risk is especially salient.

We further explore the attributes of insiders who make direct versus indirect trades, as well as the characteristics of their respective firms. We find that indirect trades are more likely to be associated with insiders from firms that are smaller, and have a higher book-to-market ratio, lower profits, more volatile stock prices, and lower institutional ownership. This evidence is in line with the view that higher information asymmetry in these firms provides more trading opportunities and less scrutiny for insiders who wish to make more informed trades through indirect accounts.

In addition, indirect trades are more likely to be made by insiders who are older, more experienced,

6 and have longer tenure. These insiders are also more likely to be the CEO or Chairman of the

Board, less likely to be the General Counsel, and less likely to be female. We also find that indirect trades are more likely to be made by insiders with attributes that prior research has found to be associated with opportunistic trading in general. Specifically, we find that indirect trades are more likely to be made by insiders who trade profitably prior to earnings announcements (Ali and

Hirshleifer, 2017), or have a short investment horizon (Akbas, Jiang, and Koch, 2019).

Our paper contributes to the bountiful literature on the information content of insider trades with regard to future stock returns.6 This literature generally examines all publicly disclosed open market purchases or sales by various groups of insiders, without distinguishing between direct trades versus indirect trades on behalf of family, trusts, retirement accounts, or foundations. Our study establishes that an insider’s inclination to make different types of indirect trades represents a unique and important attribute that can help both market participants and regulators to focus on subsets of insiders and their trades that are more informative.

We also build upon the work of Berkman, Koch and Westerholm (2014), who show that the subset of all Finnish investors who open accounts and trade on behalf of young children outperform other retail investors, and tend to channel their most informed trades through these under aged accounts. We extend this work in a number of ways. Perhaps most importantly, while they analyze all trades by all Finnish investors (who may or may not be informed), we focus exclusively on the insider trades of U.S. corporate managers who clearly possess non-public information about their own firms. Moreover, unlike the average Finnish investor, U.S. corporate insiders are required to disclose not just their own direct trades, but also indirect trades made on behalf of family members, trusts, retirement accounts, and foundations. In addition, our analysis

6 For a sample of references, see footnote 1.

7 of the broader (and more liquid) U.S. equity market allows us to draw more generalizable inferences about the extent to which informed traders use indirectly controlled family, trust, retirement, and foundation accounts to exploit their access to private information.

II. Data and Description of Variables

II.A. Data on Indirect Trades

U.S. corporate insiders are required to disclose any personal trades made in their company’s stock within two business days of the trade’s execution. Insiders comply with this regulation by electronically submitting a Form 4 that describes the transaction details. Figure 1 provides an example of this document with personal information redacted, to illustrate how we construct the data used in this study. The top portion of the Form 4 provides personal information about the insider and her/his firm. Table I then lists information about the trades disclosed by the insider. Field #6 in Table I requires the insider to indicate whether each transaction is a “direct” trade for the insider’s own account or an “indirect” trade through another account associated with the insider. If a transaction is flagged as an indirect trade, the insider is given the opportunity to elaborate on the nature of this indirect ownership in field #7, as well as in footnotes to this field.

While the Thomson Reuters insider-trading database contains a variable that denotes every trade as direct or indirect, from field #6, this database does not provide any additional information about the nature of indirect ownership that insiders disclose in field #7 or its footnotes.

Since July 2003, the SEC has required insiders to submit the Form 4 electronically, and the

Commission makes these Form 4s available to the public on Edgar. We downloaded the pdf files containing all individual Form 4s submitted by corporate insiders over the period from July 2003

8 through December 2017. From these files, we generate dummy variables to identify different categories of indirect insider trades.7

We accomplish this task by using word processing software to parse any information provided in field #7 of Form 4 and its related footnotes, in a search for words that indicate trades are for a spouse, child, or other family member, as well as trades allocated to a trust, retirement account, or foundation. In the top half of Table 1, we describe the resulting dummy variables for each category of indirect insider trades that we analyze.8 Panel A of Table 2 presents the relative frequency of insider trades made in each category from our sample. For example, Panel A of Table

2 indicates that roughly 18 percent of all insider trades are indirect trades made through another account, with 8 percent allocated to any family member. Similarly, around 8 percent of all insider trades are allocated to a trust account, with around half of these trades made in family trusts.

Finally, a smaller proportion of insider trades are made through retirement accounts (0.4%), or foundations (0.3%).9

II.B. Variables

Our sample of insider trades is limited to open market purchases and sales of common stocks. We examine monthly data on aggregate net purchases or sales by individual insiders for each trade category listed in Tables 1 and 2. During a given month, we sum across all purchases and sales by every insider to obtain net shares traded in each type of account, so the unit of measurement is net shares of type (k) purchased or sold by a given insider (i) at firm (j) during month (t). Following prior work, the final sample excludes small trades of less than 100 shares.

7 We have verified that our sample of insider trades matches the information in the Thomson Reuters database. 8 In footnote a of Table 1, we provide a list of key words used to identify these categories of indirect insider trades. 9 Note that the subsets of family trades assigned to a child, spouse, or other family member sum to more than 100 percent of the entire subgroup of trades for any family member, because some trades apply to more than one trade category (e.g., a trade could be assigned to the accounts of a child and spouse). On the other hand, the subsets of trust trades assigned to a child or spouse (35.7% + 9.1%) sum to less than 100 percent of the entire subgroup of all trust trades made in a family trust (54.4%), because some of these trades do not mention a child or spouse.

We obtain firm financial statement data from Compustat and stock return data are from CRSP.

Our main sample spans the period between July 2003 and December 2017.

The bottom half of Table 1 describes our monthly variables. In our calendar time portfolio analysis we estimate four-factor alphas (αk) based on the one-month-ahead return (RETi,j,t+1) for portfolios of stocks (j) bought or sold by insiders (i) in each trade category (k) during month (t).

Dependent variables for our panel regression approach include the analogous one-month-ahead

Fama French four-factor alphas (ARj,t+1) following a trade of any type (k) by insider (i) in stock (j) during month (t), as well as the analogous n-month-ahead cumulative abnormal return covering the following a months (CARj,t+1,t+a).

Firm-specific control variables include firm size (Size), book-to-market ratio (B/M), short term returns over the prior month (RETj,t-1), momentum returns over the previous eleven months

(RETj,t-12,t-2), asset growth (AssetGr), stock return volatility (Std_Ret), and firm profitability

(Profit). We also construct a variable that measures the relative magnitude of a given insider’s net trading activity of type (k) during the month (Trade_Size_k), as the ratio of net shares purchased or sold on behalf of account type (k), scaled by the total share volume by all investors in firm (j) during month (t).

In Panel B of Table 2, we present descriptive statistics and correlations for our key variables.

The mean of overall insider trade size (Trade_Size) is negative, which suggests that the typical trade is a sale. This outcome reflects the fact that insiders are more likely to sell because they often obtain shares as a part of their compensation. We find similar patterns for all other three trade size measures for direct trades, all indirect trades, and family trades.

III. Calendar Time Portfolios: The Relative Performance of Different Trade Categories

III.A. Univariate Analysis: The Performance of Each Category of Insider Trades

In our first set of tests, we analyze calendar time portfolios of stocks purchased or sold by insiders, either directly for their own accounts or indirectly on behalf of family members, trusts, retirement accounts, or foundations. We also examine portfolios based on subsets of family trades made in the account of a spouse or child, as well as subsets of trades in trust or retirement accounts that are devoted to family members. We begin by building portfolios of stocks based on all insider purchases or sales in each trade category (k) during any given one-month period. We then examine the time series of one-month-ahead portfolio returns for each type of insider transaction.

The first column of Table 3 presents the monthly Carhart (1997) four-factor alpha for every

10 category of insider purchases (αk). The second column compares this performance with the alpha for the portfolio of direct insider purchases (αme). That is, for each type of indirect purchase (k), we also provide the alpha of a hedge portfolio that is long stocks with that type of indirect purchase and short stocks with direct purchases made in the insider’s own account (αk – αme). The third and fourth columns provide the same information for each category of indirect sales. Finally, the fifth column presents the performance of a hedge portfolio that is long insider purchases and short insider sales, for every category of indirect trades (k), while the sixth column compares this performance with the alpha of the analogous long-short hedge portfolio based on direct trades.

In the top row of Table 3, we document the performance of all insider purchases or sales.

Consistent with prior work, calendar time portfolios of all stocks bought by insiders in one month earn a four-factor alpha (αAll) of 0.69 percent in the following month (t-ratio = 4.14), while all stocks sold by insiders yield an insignificant αAll of -0.05 percent (t-ratio = -0.72). As a result, the combined hedge portfolio that is long all stocks purchased and short all stocks sold by insiders is dominated by the performance of insider purchases, earning an αAll of 0.73 percent (t-ratio = 4.61).

10 We find similar results when we use the Fama-French 3-factor or 5-factor model (Fama and French, 1993, 2015).

When we distinguish between direct and indirect insider trades, their performance diverges.

For example, the subset of direct purchases made in the insider’s own account have a slightly smaller alpha (αme) of 0.65 percent per month (t-ratio = 3.63). In contrast, insider purchases made through any indirect account have a significantly larger alpha (αother) of 0.89 percent (t-ratio =

4.99), which outperforms direct purchases by (αother - αme) = 25 bps (t-ratio = 1.97). Once again, insider sales do not significantly outperform, when made either directly through the insider’s own account or indirectly through any other account (αme = -0.02 or αother = -0.10, with t-ratio = -0.32 or -1.11, respectively). As a result, the combined hedge portfolio that replicates both direct purchases and sales earns an alpha (αme) of 0.67 percent (t-ratio = 3.96), while the analogous hedge portfolio of indirect purchases and sales has a significantly larger alpha (αother) of 0.99 percent (t- ratio = 4.98), which outperforms direct trades by (αother - αme) = 33 bps (t-ratio = 2.10).

We next analyze the subset of indirect trades made through family accounts. For example, portfolios of indirect purchases made in any family account earn an alpha (αfam) of 1.09 percent (t- ratio = 4.37) in the following month, significantly outperforming direct purchases by (αfam - αme)

= 45 bps (t-ratio = 3.12). Furthermore, the subset of family purchases made on behalf of a spouse earn an even higher alpha (αspouse) of 1.27 (t-ratio = 4.65), which outperforms direct purchases by

(αspouse - αme) = 63 bps (t-ratio = 2.48), while purchases for a child earn an alpha (αchild) of 1.30 percent (t-ratio = 3.91), which outperforms direct purchases by (αchild - αme) = 65 bps (t-ratio =

2.76). On the sell side, no subset of family trades generates a significant alpha. However, hedge portfolios that the combined hedge portfolio that replicates both purchases and sales in any family account earns an alpha of 1.02 percent, outperforming the analogous hedge portfolio of trades made through direct accounts by 35 bps. When we narrow this further, we find that hedge portfolios replicating both purchases and sales in spouse or child accounts earn alphas of 1.19

12 percent and 1.33 percent respectively, outperforming the analogous hedge portfolio of trades made through direct accounts by 53 and 67 bps respectively.

Consider next the performance of indirect trades made through trust accounts. Purchases in any trust generate a one-month alpha (αtrust) of 1.05% (t-ratio = 4.73), which outperforms direct purchases by (αtrust – αme) = 0.40% (t-ratio = 1.82). This outperformance is larger for the subset of purchases made in family trusts (αtrust-fam = 1.39%, t-ratio = 3.71), which outperforms direct purchases by (αtrust-fam – αme) = 0.74% (t-ratio = 2.08). In contrast, while purchases in non-family trusts also generate a significant alpha (αtrust-notfam = 0.73%, t-ratio = 3.03), this does not significantly outperform direct purchases (αtrust-notfam – αme = 0.09%; t-ratio = 0.32). On the sell side, indirect sales through family trusts do not outperform. On the other hand, sales in non-family trusts generate a marginally significant negative alpha (αtrust-notfam) of -0.25% (t-ratio = -1.77), which outperforms direct insider sales (αtrust-nonfam – αme = -0.22%, t-ratio = -1.68). This latter evidence is consistent with insiders using non-family trust accounts to avoid estate and gift taxes.

Next, we examine the smaller sample of indirect trades made through retirement accounts.

While insiders do not generate significant alphas when they trade in retirement accounts on behalf of family members, they do when they either buy or sell in their own (i.e., non-family) retirement account. For example, purchases in non-family retirement accounts earn an alpha (αret-notfam) of

0.82% (t-ratio = 3.05) which is slightly larger than the alpha for direct purchases (αret-notfam – αme =

0.17%, t-ratio = 0.67). On the sell side, trades in non-family retirement accounts significantly outperform direct sales (i.e., αret-notfam = -0.69%, t-ratio = -2.75; and αret-notfam – αme = 0.67%, t-ratio

= -2.55). As a result, the combined hedge portfolio that duplicates insider purchases and sales in non-family retirement accounts generates a significant alpha (αret-notfam) of 1.51 percent per month

(t-ratio = 4.61), which outperforms the analogous hedge portfolio of direct trades (αret-notfam – αme

= 0.85%, t-ratio = 3.11). This evidence is also consistent with insiders using their own retirement accounts to avoid estate and gift taxes. Finally, there is no evidence to indicate that insider trades made on behalf of a foundation significantly outperform on either the buy side or the sell side.

Taken together, the evidence in Table 3 provides our first piece of compelling evidence that insider trades made in accounts that the insider controls indirectly outperform direct insider trades on either the buy side or the sell side, depending on the trade category. In particular, this evidence supports our conjecture that insider trades made through family, trust, and retirement accounts outperform trades made in the insider’s own account.

III.B. Bivariate Analysis: The Importance of Trade Size

III.B.1. Measuring Insider Trade Size as a Proxy for the Information Content of Insider Trades

While Table 3 provides initial evidence that indirect trades outperform direct trades, this analysis does not account for the potential information embedded in the trade’s size (i.e., all trades of each type are considered to be of equal importance, regardless of trade size). In this section, we account for the possibility that insiders tend to make larger trades when they have more salient information. For each trade category (k), we construct a proxy for the direction and magnitude of the information signal revealed by an insider’s trading activity during any month, by computing the net shares of type k, purchased or sold as a proportion of the total share volume by all investors in firm j, during month t. Specifically, for trades of type k made by insider i of firm j in month t, the insider’s trade size relative to his or her shareholdings is defined as:

푃k,i,j,t − 푆k,i,j,t 푇푟푎푑푒_푆푖푧푒_푘i,j,t = , 푉푂퐿j,t

14 where Pk,i,j,t is the number of shares of account type k purchased by insider i at firm j in month t,

Sk,i,j,t is the number of shares sold, and VOLj,t is the total share volume by all investors in firm j during month t.11

III.B.2. The Performance of Different Categories of Insider Trades, based on Trade Size

For each month (t), we sort all insiders who make trades of a given category (k) into terciles, based on the insider’s trade size within that category (Trade_Size_k), to form three portfolios of stocks that range from large insider sales (LS) to large insider purchases (LP). Table 4 reports the monthly four-factor alphas for these tercile portfolios based on trade size, for several categories of trades analyzed (k). We also report the analogous monthly alphas for hedge portfolios that are long the portfolio of stocks comprised of indirect trades of each type (k), and short the portfolio of direct trades (αk – αme), within each tercile by trade size.

We first consider the results for direct trades in column (1) of Table 4. Consistent with prior work that does not distinguish between direct and indirect trades, the lowest tercile portfolio comprised of large direct sales (LS) does not generate negative abnormal returns (i.e., its alpha is positive but insignificant). Similarly, the middle group of smaller direct trades has an insignificant positive alpha. In contrast, the tercile of large direct purchases (LP) generates a larger and significant one-month alpha of 0.38 percent (t-ratio = 3.49). Furthermore, at the bottom of column

(1), the hedge portfolio that is long the portfolio of large direct purchases and short large direct sales (LP – LS) also yields a significant monthly alpha (αme) of 0.35 percent (t-ratio = 2.71).

Next, we consider the analogous results for the four categories of indirect trades (k) provided in columns (2) – (5) of Table 4. In the top two rows, the lower two terciles by trade size

11 Few other studies examine relative trade size as a proportion of total trade volume or some other scalar, as a potential signal for the information content of insider trades (see Akbas, Jiang, and Koch, 2018; Bettis, Vickrey, and Vickrey, 1997; and Scott and Xu, 2004).

15 yield insignificant alphas for all four categories of indirect trades analyzed. In contrast, the third row of large purchases yields larger monthly alphas that are significantly positive, for all four indirect trade categories. For example, the portfolio of large purchases in any indirect account generates a monthly alpha (αother) of 0.85 percent (t-ratio = 3.38), which outperforms large direct purchases by 46 bps (t-ratio = 2.47). When we consider the subset of large indirect purchases for any family account, we find a larger alpha (αfam) of 0.94% (t-ratio = 3.37), which outperforms large direct purchases by 56 bps (t-ratio = 2.48). Furthermore, large purchases for a spouse or child yield even larger alphas of 1.08 or 1.07 percent, respectively (t-ratio = 3.83 or 3.13), both of which outperform large direct purchases by 70 bps (t-ratio = 2.67 or 2.40).

Finally, at the bottom of columns (2) – (5) in Table 4, the hedge portfolio that is long the tercile of large purchases and short the tercile of large sales (LP – LS) yields a significant monthly alpha for each of the four indirect trade categories, that ranges from 0.75% (t-ratio = 2.37) for any family trades, to 1.05% (t-ratio = 3.01) for the subset of trades made on behalf of a child.

Furthermore, three of these four long-short hedge portfolios significantly outperform the analogous hedge portfolio of direct trades by amounts that range from 49 bps (t-ratio = 2.18) for any indirect trades, to 69 bps (t-ratio = 1.93) for trades made on behalf of a child.

Overall, the portfolio analysis in Table 4 corroborates the evidence in Table 3, to show that insider trades made through different types of accounts that the insider controls indirectly outperform direct insider trades to different degrees, depending on the trade category. Furthermore, these results are robust when we account for trade size by considering subsets of large insider purchases and sales, respectively. This evidence indicates that these subsets of indirect trades are significantly more informative about future stock returns, relative to the insider’s own direct trades.

IV. Monthly Panel Regression: The Relative Performance of Different Trade Categories

In this section, we estimate panel regressions to assess whether the performance of large purchases versus large sales (LP – LS) for each trade category, documented in Table 4, is robust when we control for other firm attributes. We begin by constructing the scaled rank of our proxy for trade size, for each insider trade category (TrSize_k_RK). First, the cross section of all insiders

(i) who make trades of type k in month t is ranked into terciles by the trade size (Trade_Size_ki,j,t), and the observations in each tercile are assigned the values 0, 1, or 2. Second, we divide these tercile ranks by 2 to form the scaled rank, which ranges from 0 (for the lowest tercile with large sales) to +1 (for the highest tercile with large purchases). Note that a one-unit increase in the scaled rank, TrSize_k_RK (from 0 to +1), compares the tercile of insiders who make large purchases with the tercile who make large sales (LP – LS) in month t.

IV.A. Regression Approach: Short Run Trading Performance

This subsection estimates the relative short run trading profitability of large purchases versus large sales (LP – LS), for every category of insider trades (k), in the month following the trades. We regress the future abnormal return on the scaled rank of insider trade size for each trade category (k), along with other control variables, as follows:

15 ARj,t+1 = αt + ∑ βk TrSize_k_RKi,j,t + Other Controlsi,j,t + εi,j,t ; (1) k = 1 where k indexes the fifteen categories of direct and indirect insider trades analyzed here.

The dependent variable, ARj,t+1, is the leading one-month-ahead market-adjusted abnormal

12 return for the firm (j) of insider (i). We multiply ARj,t+1 by 100 to reflect performance in percentage terms. TrSize_k_RK is the scaled rank of trade size for each category of trades analyzed here (k = 1-15). The control variables in Equation (1) include the firm’s book-to-market ratio

12 In Appendix B.1, we also use raw stock returns as the dependent variable and find similar results.

(B/M), firm size, the short term one-month lagged return (RETj,t-1), momentum returns over the previous eleven months (RETj,t-12,t-2), gross profits (Profit), asset growth (AssetGr), and the volatility of daily stock returns (Std_Ret). These controls help to establish whether the predictive information contained in indirect trades of each type (k) remains after accounting for other firm attributes that have also been shown to predict returns. Monthly fixed effects are included.

For each category of direct or indirect trades (k), the coefficient of the scaled rank of trade size (βk) is analogous to the return on a hedge portfolio that is long the tercile of large purchases of type k and short the tercile of large sales (LP – LS). To see this result, consider the association between the scaled rank for each trade size measure and future returns, that is implied by Equation

∂퐴푅(푗,푡+1) (1): = 훽 . This partial derivative shows that a one-unit increase in the scaled ∂푇푟푆푖푧푒_푘_푅퐾(푖,푗,푡) 푘 rank of trade size, which compares large sales with large purchases (i.e., changing TrSize_k_RK from 0 to +1), is associated with a change in ARj,t+1 of βk percent.

In Table 5, we present ten columns of results from estimating ten different specifications of this panel regression model that include trade size variables for various subsets of the fifteen trade categories analyzed here. It is appropriate to consider only subsets of these fifteen trade categories in each specification, because many categories overlap. For example, a single trade may be allocated to a trust account on behalf of the insider’s spouse and children.

The top row of Table 5 presents the coefficient of the scaled rank of trade size for direct insider trades (β1), for the ten specifications analyzed here. These ten coefficients range from .456 to .505. This outcome implies that, after controlling for firm attributes, a hedge portfolio that duplicates large direct purchases and sales (LP – LS) earns abnormal returns that are somewhat larger than the analogous alpha (αme) of 0.35 percent, from the bottom row of Table 4.

The remaining rows in Table 5 similarly indicate that, after controlling for firm attributes, the analogous (LP – LS) hedge portfolios for most indirect trade categories also earn more than their counterparts from the bottom row of Table 4 (i.e., βk > αk). Furthermore, the F-tests at the bottom of each column in Table 5 indicate that, for most categories of indirect trades (k), these regression coefficients are significantly greater than the coefficient for direct trades (i.e., βk > β1).

For example, the (LP – LS) hedge portfolio for all indirect trades earns (β2 =) 0.95 percent per month (t-ratio = 5.11), which significantly outperforms direct trades by (β2 – β1 =) 45 bps (p-value

= 0.01). Furthermore, a similar hedge portfolio of indirect trades made in any family account now earns (β3 =) 1.00 percent per month (t-ratio = 4.12), which significantly outperforms direct trades by (β3 – β1 =) 52 bps (p-value = 0.03), while trades on behalf of a child earn even more at (β5 =)

1.24 percent per month (t-ratio = 3.50), which outperforms by (β5 – β1 =) 77 bps (p-value = 0.04).

In addition, the analogous hedge portfolio that duplicates the large purchases and sales made in a family trust earns (β8 =) 1.05 percent per month (t-ratio = 3.35), which significantly outperforms direct trades by (β8 – β1 =) 54 bps, while analogous trades through a non-family trust earn (β9 =)

1.28 percent (t-ratio = 4.47), which outperforms by (β9 – β1 =) 77 bps (p-value = 0.001).

Finally, the last two columns of Table 5 present our basic models that include multiple scaled rank variables that span the different categories of indirect trades, while limiting the overlap across more than one trade category that could involve family members. In particular, we include the scaled rank of trades for any family member, as well as trades in non-family trust or retirement accounts, along with trades on behalf of foundations. Column (9) reveals that, once again, the performance of direct trades (β1 = 0.50%, t-ratio = 4.14) is dominated by trades on behalf of any family member (β3 = 0.98%, t-ratio = 4.03), non-family trusts (β9 = 1.23%, t-ratio = 4.27), and the insider’s own (non-family) retirement account (β14 = 1.06%, t-ratio = 2.00), while trades on behalf

19 of a foundation do not outperform (β15 = -0.62%, t-ratio = -1.01). Column (10) replaces the single trade size variable in column (9) for any family trades with the three variables that separately consider trades for a spouse, child, or other family member. Once again, this last model shows that the outperformance of family trades is concentrated among trades on behalf of children (β5 =

1.24%, t-ratio = 3.51), which exceeds the performance of trades for the insider’s spouse (β4 =

0.66%, t-ratio = 1.68) or other family members (β6 = -0.11%, t-ratio = -0.24).

This regression analysis confirms that the findings from the portfolio approach remain, after controlling for other firm attributes that also predict stock returns. Moreover, the results in

Table 5 are generally larger than the analogous results from the bottom row of Table 4, which examines the one-month performance of (LP – LS) hedge portfolios for several categories of indirect trades, without controlling for firm attributes. In addition, the coefficients of the control variables are consistent across the columns in Table 5, and generally support our expectations. For example, firms that outperform in month t+1 tend to have smaller size and lower asset growth.

IV.B. Regression Approach: Long Run Trading Performance

In this subsection, we examine the longer run performance of indirect trades versus direct trades. The evidence from Sections III and IV.A indicates a tendency for indirect trades in family accounts, trusts, and retirement accounts to outperform direct trades over the next month, but it is not clear whether this performance is based on information that is short-lived or more long-lasting.

In this section, we investigate whether this short run outperformance continues to accumulate over longer periods that extend up to three years following the insider trades.

For this analysis, we estimate a revised version of Equation (1) that includes the non- overlapping categories of trade size variables from the specification in column (9) of Table 5, and

20 replaces the one-month-ahead abnormal return (ARj,t+1) as dependent variable with the cumulative abnormal return extending further into the future (CARj,t+1,t+a), as follows:

CARj,t+1,t+a = αt + β1 TrSize_me_RKi,j,t + β2 TrSize_fam_RKi,j,t + β3 TrSize_trust-notfam_RKi,j,t

+ β4 TrSize_ret-notfam_RKi,j,t + β5 TrSize_found_RKi,j,t + Controlsi,j,t + εi,j,t , (2)

where CARj,t+1,t+a is the future cumulative market-adjusted abnormal stock return for firm j over months t+1 through t+a, following the month (t) in which insider i trades.13

In Table 6, we present six columns that analyze the performance of (LP – LS) hedge portfolios for different categories of insider trades (k) over future periods that extend up to three years later. The first column reproduces the one-month-ahead results from our main model (9) in

Table 5, while the remaining columns present the analogous results for performance measured over longer periods. We do not present the estimated coefficients of the control variables in this table for brevity.

The top row of Table 6 presents the coefficient of the scaled rank for direct trades (β1) that pertains to future periods that range from one month to three years ahead. These results indicate that the performance of the (LP – LS) hedge portfolio comprised of direct trades grows in magnitude from 0.50% (t-ratio = 4.14) over the next month to 3.10% (t-ratio = 3.78) after three years. This evidence is consistent with prior work that generally finds insider purchases outperform over relatively long horizons that extend up to a year or more.14

In contrast, the second row of coefficients in Table 6 shows that the abnormal performance of any family trades continues to grow monotonically over the following three years, from 0.98%

(t-ratio = 4.03) over the next month to 7.84% (t-ratio = 4.01) after 36 months. The third row

13 We also repeat this analysis using raw returns and find similar results, provided in Appendix B.2. 14 We refer the reader to the references cited in footnote one.

21 similarly reveals that the abnormal performance of indirect trades for non-family trusts grows from

1.23% (t-ratio = 4.27) after one month to 14.95% (t-ratio = 7.01) after 36 months. In the fourth row, the abnormal performance of trades for the insider’s own (non-family) retirement account also grows from 1.06% (t-ratio = 2.00) after one month to 12.52% (t-ratio = 3.31) after three years.

As in Table 5, trades on behalf of a foundation never significantly outperform over any time frame.

The F-tests at the bottom of each column in Table 6 indicate that indirect trades for family members significantly outperform direct trades for all periods over the next three years. Similarly, trades made in a non-family trust outperform over the next one-month, as well as longer periods that extend beyond one year into the future. In contrast, trades made in the insider’s own (non- family) retirement account only significantly outperform direct insider trades over longer periods that extend two or three years, while trades for a foundation never outperform direct trades.

The regression analysis in this section corroborates and extends the calendar time portfolio analyses in Tables 3 and 4 above. These results establish that indirect trades through the accounts of family members, as well as non-family trust and retirement accounts, outperform direct trades in the insider’s own account, after controlling for other firm attributes that prior research has shown to predict returns. This outperformance is both economically and statistically significant, and it extends for up to three years following the trades.

V. Indirect Insider Trading and the Use of Private Information

In this section, we compare the predictive information contained in direct trades to that of indirect trades ahead of firm future informational events, including earnings announcements and large price changes. If the superior performance of indirect trades relative to direct trades arises from strategic timing by insiders prior to imminent corporate events or, we should expect indirect transactions to contain more predictive information about these events than direct trades.

V.A. Quarterly Earnings Surprises

We first relate direct and indirect insider trades to the firm’s forthcoming earnings surprise with the following regression:

Surprisej,q = αt + β1 TrSize_me_RKi,j,t + β2 TrSize_other_RKi,j,t (or β3 TrSize_fam_RKi,j,t

+ β4 TrSize_trust-notfam_RKi,j,t + β5 TrSize_ret_notfam_RKi,j,t

+ β6 TrSize_found_RKi,j,t) + β7 lag1Surprisej,q + Controlsj,t + ε i,j,t. (3)

where Surprisej,q is the earnings surprise of firm j at the next quarterly earnings announcement that occurs in quarter q, following trades made by insider i in month t. We measure Surprise using the standardized unexpected earnings (SUE) model of Bernard and Thomas (1990). TrSize_me_RK,

TrSize_other_RK, TrSize_fam_RK, TrSize_trust-notfam_RK, TrSize_ret_notfam_RK, and

TrSize_found_RK are the scaled rank variables for trade size in a given month (t) based on terciles of direct trades, all indirect trades, family trades, non-family trades in trust accounts, non-family trades in retirement accounts, and foundation trades, respectively, as described in Section IV. In addition to the standard control variables from equation (1), we also control for the lagged-one- quarter earnings surprise. We conjecture that, if the subset of all indirect trades (or other trades) made in a given month contain more predictive information about the firm’s next quarterly earnings surprise, then β2 (or β3 - β6) should be significantly larger than β1.

We present the results of this analysis in Panel A of Table 7. Column (1) compares the information content of direct trades (TrSize_me_RK) with that of indirect trades

(TrSize_other_RK). The results indicate a negative relation between net order flow through direct trades and the next earnings surprise, but a positive relation for indirect trades and the next surprise, although both coefficients (β1 and β2) fall short of statistical significance at conventional levels.

Importantly, however, these two coefficients are significantly different from each other (p-value =

0.05). Consistent with our conjecture, this evidence indicates that net order flow through indirect trades contains significantly more predictive information than direct trades, regarding the firm’s next earnings surprise.

In column (2) of Panel A in Table 7, we present the analogous results comparing the predictive ability of direct trades (TrSize_me_RK) with the subset of indirect trades made through family accounts (TrSize_fam_RK) and non-family indirect accounts (TrSize_trust-notfam_RK,

TrSize_ret_notfam_RK, and TrSize_found_RK). Now the coefficient of the scaled rank of trade size for family trades (β3) is positive and significant (t-ratio = 1.95), which shows that an increase in net purchases made through family accounts during one month is associated with a larger upcoming earnings surprise (SUE). Moreover, these two coefficients (β1 and β3) are again significantly different from each other (p-value = 0.01), indicating that net order flow through family accounts contains more predictive information about future earnings. We do not find similar results for other non-family indirect trades.

V.B. Large Stock Price Changes

Next, we examine the information content of direct trades versus indirect trades by insiders regarding imminent large idiosyncratic stock price changes. We identify large price change events as follows. For each firm we compute the three-day cumulative abnormal return (CAR) around every trading day during a given year. If the CAR for a given day is among the top (bottom) five percent among all trading days in the year, we classify that day as having a large positive (negative) price change. We then create two indicator variables, +∆P or −∆P, for the subset of such large positive or negative price changes that occur shortly after insider trades. Specifically, if a large price increase (decrease) occurs within ten days following an insider trade, the dummy variable

+∆P (−∆P) is assigned a value of one, and zero otherwise. We then perform the following probit regression analysis for trades made by insider i at firm j during month t:

−1 Φ (+/ −∆Pj,t) = αt + β1 TrSize_me_RKi,j,t + β2 TrSize_other_RKi,j,t (or β3 TrSize_fam_RKi,j,t

+ β4 TrSize_trust-notfam_RKi,j,t + β5 TrSize_ret_notfam_RKi,j,t

+ β6TrSize_found_RKi,j,t) + Controlsj,t + εi,j,t. (4) where Φ(.) represents the cumulative distribution function for the standard normal distribution.

We again incorporate all control variables included in model (1).

We present the results from this probit regression in Panel B of Table 7. We also provide the marginal effects implied by the insider trading signals from direct trades and indirect trades, along with the χ2-statistics for the null hypotheses that different pairs of coefficients are equal.

Columns (1) – (4) provide the evidence for large positive price changes, while columns (5) – (8) present the analogous results for large negative price changes.

Across columns (1) – (4) in Panel B of Table 7, the coefficients on net insider trading through direct accounts (β1), any indirect accounts (β2), family accounts (β3), non-family trust accounts (β4), and non-family retirement accounts (β5) are all significantly positive. This outcome indicates that, in general, net purchases (sales) in all five types of insider accounts are associated with a higher (lower) likelihood of an imminent large positive price change. Importantly, however, in columns (1) – (4) the impact of all indirect trades, family trades, and trades through non-family trust accounts (β2, β3, and β4) are significantly larger than that of direct trades (β1). For example, in column (2), the marginal effect of β1 indicates that changing from the tercile of large sales to the tercile of large purchases through the insiders’ own direct account increases the probability of an imminent large positive stock price change by 3.9%. However, the analogous effect for indirect trades is larger, at 7%. Furthermore, the difference between β2 and β1 is statistically significant at

25 the 1% (χ2=32.1 and p-value = 0.00), indicating that large net purchases through any indirect accounts are significantly more likely to precede large stock price increases than large net purchases through the insider’s own direct account.

Columns (5) – (8) in Panel B of Table 7 provide similar results for large stock price declines.

The coefficients of all four variables related to the first four categories of insider trading are significantly negative, indicating that net purchases (sales) of each type predict a lower (higher) likelihood of an imminent large price decline. Once again, the effect is significantly larger for indirect trades, family trades, and trades through non-family trust accounts when compared with direct trades. For instance, in column (6) the marginal effect associated with β1 indicates that changing from large sales to large purchases through the insider’s own direct account reduces the likelihood of an imminent large price decline by 2.2%, while the same evidence for indirect trades

(β2) reduces the chance of a large price decline by 4.1%. Once again, this difference between β2

2 and β1 is statistically significant (χ = 11.9; p-value = 0.00). Finally, we note that the analogous evidence in column (7) indicates that family trades and trades through non-family trust accounts also contain significantly more predictive information than direct trades about large price declines, although the difference between β3 and β1 becomes statistically insignificant when we include the controls in column (8) (i.e., χ2 = 1.7; p-value = 0.19).

Taken together, the analysis in this section documents that net order flow through indirect accounts or family accounts contains higher predictive power on firm future informational events.

These results are consistent with our main findings above that insider trades through various indirect accounts tend to have better performance than trades through insiders’ own accounts. The results suggest that insiders strategically place their trades through indirect accounts or family

26 accounts prior to firm events. Therefore, the superior performance of indirect insider trades arises, at least partially, from their privileged access to private information.

VI. The Intensity of SEC Scrutiny and Indirect Insider Trading

In this section, we investigate whether and how the ratio of indirect (or family) trades to direct trades varies with the intensity of SEC scrutiny against illegal insider trading. Del Guercio,

Odders-White, and Ready (2017) find that aggressive SEC enforcement deters illegal insider trading activity. Similarly, Cohen, Malloy, and Pomorski (2012) show that opportunistic insiders tend to trade less following an increase in SEC investigations regarding illegal insider trading, arguing that such intensified scrutiny makes the risks of illegal insider trading particularly salient.

If insiders tend to channel their best trades through indirect accounts, and these indirect trades tend to be based on non-public information and thus subject to elevated litigation risk, we would expect insiders to trade less through such indirect accounts when litigation risk is particularly salient.

We follow the approach in Cohen, Malloy, and Pomorski (2012), and test this conjecture by examining the relation between the ratio of indirect trades to direct trades and the recent intensity of SEC enforcement activity. The regression model is specified as follows:

Ratiot = α + β1 SECt−1 + β2 SECt−2 + β3 SECt−3 + β4 MKTRETt−1

+ β5 MKTRETt−13, t−2 + εt . (5)

where Ratiot is the ratio of the number of insiders who trade through any indirect account (or through a family account) to the number of insiders who trade directly in their own account, during month t. SECt is the number of SEC investigations against illegal insider trading in month t. We also control for the lagged one-month market return (MKTRETt−1), as well as the cumulative market return over the prior twelve months, from month t-13 through month t-2 (MKTRETt−13, t−2).

The results are provided in Table 8. The left panel presents the analysis of any indirect trades, while the right panel focuses on family trades. The results across all eight columns show that more aggressive SEC enforcement is associated with less trading through indirect accounts or family accounts over the following three months. Furthermore, the impact is economically large.

For example, column (4) indicates that one additional litigation case against illegal insider trading is associated with a 0.30% decline in the ratio of indirect trades to direct trades in the next month, and another decline of 0.50% in two months. We also find that this deterence effect is temporary, since the coefficient of the SEC variable loses statistical power after two months, as shown in columns (4) and (8).

This section provides evidence that insiders reduce their trading activity through indirect accounts or family accounts, relative to direct trading activity, following greater scrutiny by the

SEC. These results suggest that insiders become more conservative when litigation risk is higher, as proxied by more litigation cases against illegal insider trading. This evidence is again in line with our main findings above, which suggest that insiders are more likely to place informed trades through indirect accounts, and this informed trading is deterred by greater SEC enforcement activity.

VII. The Attributes of Insiders and Their Firms, for Subsets of Insiders Who Make Direct Trades versus Indirect Trades

In this section, we explore the differences in the attributes of insiders, and their respective firms, between the subsets of insider trades made each month through direct accounts versus those made through indirect accounts.15 Specifically, within each month (t), we calculate the average measures of various attributes for each subset of direct or indirect trades. We then calculate the

15 In unreported analysis, we also compare the differential characteristics of insiders and their firms across the subsets of insiders who make direct trades versus family trades in any given month, and we find similar results.

28 time series means of these monthly cross-sectional averages. We also test the statistical significance of the differences across these mean attributes associated with direct versus indirect trades.

We begin by examining the differences in firm characteristics insiders who make direct versus indirect trades. The top portion of Table 9 shows that indirect trades are more likely to be made by insiders from firms with smaller size and higher stock return volatility. Since these firm attributes tend to be associated with greater higher information asymmetry, these firms could be more susceptible to mispriced stock prices and thus more trading opportunities for the insiders.

Indirect trades also tend to be from insiders at firms with lower institutional ownership, consistent with less monitoring from institutional investors. Opportunistic insider trading at such companies is also less likely to attract attention and scrutiny from the public.

Next turn to the attributes of the insiders themselves in the bottom portion of Table 9. On average, indirect trades are more likely than direct trades to originate from older insiders, and insiders with more experience on the job, as well as experience in making insider trades. Insiders who trade in indirect accounts also tend to have better access to private information associated with the firm, since they are more likely to be the CEO or Chair of the Board, or serve on the strategy committee. They are less likely to be the firm’s General Counsel or female. Finally, insiders who trade through indirect accounts have higher total compensation and higher pay- performance sensitivity.

Finally, we examine the extent to which indirect traders are more likely to exhibit trading behavior that prior literature has associated with opportunistic insider trading. Along these lines, we find that indirect trading is less likely to be associated with non-routine insiders (Cohen,

Malloy, and Pomorski, 2012).16 On the other hand, indirect trading is more likely to be conducted by insiders who trade profitably prior to earnings announcements (Ali and Hirshleifer, 2017), and who have a short investment horizon in that they trade more frequently in and out of their won firm’s stock (Akbas, Jiang, and Koch, 2019).

VIII. Conclusion

We compare the information content of direct trades made in the insider’s own account with indirect trades made in the accounts of family members, trusts, retirement accounts, and foundations. We hypothesize four economic rationales for insiders to make informed trades through such indirectly controlled accounts. First, insiders may use indirect accounts to camouflage their informed trades, in the spirit of Kyle (1985). Second, insiders often sell shares directly from their own holdings to achieve diversification or liquidity, rather than to act on negative information. These non-informational motives are less likely to apply to indirect sales, which are thus more likely to be based on negative information. Third, insiders may make informed trades through family, trust, or retirement accounts to obtain benefits from managing estate, gift, or income taxes. Finally, insiders who trade on behalf of family may have unique parental instincts or other unobservable attributes that inspire them to make more opportunistic trades, on average, and to save their most informed trades for the accounts of family members.

We document that, on average, insider purchases made through family accounts outperform direct purchases made in the insider’s own account, while insider sales made through non-family trust or retirement accounts outperform direct insider sales. Furthermore, when we

16 One potential reason for this finding is that insiders who trade through multiple accounts are more likely to be mechanically classified as routine insiders, since they trade more. To see this result, note that Cohen, Malloy, and Pomorski (2012) classify an insider as a routine insider if he or she traded in the same month in each of the prior three years. Therefore, if an insider makes trades in multiple months in any given year, he or she is more likely to be identified as a routine insider. It is also possible that our measure captures a different group of opportunistic insiders.

30 account for trade size, hedge portfolios that duplicate the large purchases and sales of insiders through family accounts outperform the analogous hedge portfolios comprised of large direct purchases and sales made in the insider’s own account. These results are robust when we use regression analysis to control for other firm attributes that have also been shown to predict returns.

We also investigate potential sources of the superior performance of indirect trades relative to direct trades. We find that indirect trades convey more information about future firm-specific information events, such as earnings surprises and large idiosyncratic stock price changes.

Furthermore, insiders reduce their trading activity through indirect or family accounts, relative to direct accounts, when there is greater litigation risk associated with more SEC litigation activity against illegal insider trading. This evidence corroborates the conclusion that insider trades through indirect accounts or family accounts are more likely to be based on non-public information.

This paper contributes to the substantial body of prior work on the information content of insider trades with regard to future stock returns, which does not distinguish between direct and indirect insider trades. We show that insider trading through indirect accounts represents another important dimension of managerial opportunism. For example, this analysis extends previous work that documents opportunistic insider trading behavior by different subsets of insiders, such as nonroutine insiders, insiders who trade profitably before earnings announcements, and insiders with a short investment horizon. It also extends prior work that documents informed trading through the accounts of young children.

This paper should be of interest to both market participants and regulators. Insider transactions made through indirect accounts contain more information than direct trades about future stock prices and the prospects for firm performance. Regulators and other market participants are therefore justified in applying more scrutiny to insider trades made through these

31 indirect accounts. Such indirect trades represent a relatively small portion of all insider transactions, but they are more likely to convey private information.

References

Aboody, David, and Baruch Lev, 2000. Information asymmetry, R&D and insider gains. Journal of Finance 55, 2747-2766.

Admati, Anat R., and Paul Pfleiderer, 1988. A Theory of intraday patterns: Volume and price variability. Review of Financial Studies, 1 (1), 3–40.

Akbas, Ferhat, Chao Jiang, and Paul D. Koch, 2019. Insider investment horizon, Journal of Finance, forthcoming.

Ali, Usman, and David A. Hirshleifer, 2017. Opportunism as a managerial trait: Predicting insider trading profits and misconduct. Journal of Financial Economics 126 (3), 490-515.

Alldredge, Dallin M., and David C. Cicero, 2015. Attentive insider trading. Journal of Financial Economics, 115 (2015), 84-101.

Barclay, Michael J. and Jerold B. Warner, 1993. Stealth trading and volatility: Which trades move prices? Journal of Financial Economics, 34 (3), 281-305.

Berkman, Henk, Paul D. Koch, and P. Joakim Westerholm, 2014. Informed trading through the accounts of children. Journal of Finance 69 (1), 363–404.

Berkman, Henk, Paul D. Koch, and P. Joakim Westerholm, 2017. Inside the director network: When insiders trade outside stocks. Working paper.

Bettis, J. Carr, Jeffrey L. Coles, and Michael L. Lemmon, 2000. Corporate policies restricting trading by insiders. Journal of Financial Economics 57 (2), 191-220.

Biggerstaff, Lee, David C. Cicero, and M. Babajide Wintoki, 2019. Insider trading patterns. Working paper.

Brennan, Michael J., Tarun Chordia, and Avanidhar Subrahmanyam.1998. Alternative factor specifications, security characteristics, and the cross-section of expected stock returns. Journal of Financial Economics 49 (3): 345–373.

Brown, Jennifer, G. Ryan Huston, and Brian S. Wenzel, 2018. The gift that keeps on giving: Stock returns around CEO stock gifts to family members. Working Paper, Arizona State University.

Chakravarty, Sugato, 2001, Stealth-trading: Which traders' trades move stock prices? Journal of Financial Economics, 61(2), 289-307.

Chan, Louis, and Josef Lakonishok, 1993. Institutional trades and intra-day stock price behavior. Journal of Financial Economics 33, 173–200.

Cheng, Qiang, and Kin Lo, 2006. Insider trading and voluntary disclosures. Journal of Accounting Research 44 (5), 815-848.

Cline, Brandon N., Sinan Gokkaya, and Xi Liu, 2017. The persistence of opportunistic insider trading. Financial Management 46 (4), 919-964.

Cohen, Lauren, Andrea Frazzini, and Christopher Malloy, 2008. The small world of investing: board connections and mutual fund returns. Journal of Political Economy, 116, 951–979.

Cohen, Lauren, Christopher Malloy, and Lukasz Pomorski, 2012. Decoding inside information. Journal of Finance 67, 1009-1043.

Coles, Jeffrey L., Naveen D. Daniel, and Lalitha Naveen, 2006, Managerial incentives and risk- taking, Journal of Financial Economics 79, 431-468.

Cziraki, Peter, Peter De Goeij, and Luc Renneboog, 2014. Corporate governance rules and insider trading profits. Review of Finance 18 (1), 67-108.

Del Guercio, Diane, Elizabeth R. Odders-White, and Mark J. Ready, 2017. The deterrence effect of the Securities and Exchange Commission’s enforcement intensity on illegal insider trading: Evidence from run-up before news events. Journal of Law and Economics 60 (2), 269-307.

Dambra, Michael, Matthew Gustafson, and Phillip Quinn, 2019. Executives’ tax-advantaged trusts and IPO outcomes. Working Paper.

Fama, Eugene F., and Kenneth R. French, 1993. Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33 (1), 3-56.

Fama, Eugene F., and Kenneth R. French, 2015. A five-factor asset pricing model. Journal of Financial Economics 116 (1), 1-22.

Fidrmuc, Jana P., Marc Goergen, and Luc Renneboog, 2006. Insider trading, new releases and ownership concentration. Journal of Finance 61 (6), 2931-2973.

Frankel, Richard, and Xu Li, 2004. Characteristics of a firm’s information environment and the information asymmetry between insiders and outsiders. Journal of Accounting and Economics 37 (2), 229-259.

Garfinkel, Jon A., and M. Nimalendran, 2003. Market structure and trader anonymity: An analysis of insider trading. Journal of Financial and Quantitative Analysis, 38 (3), 2003, 591–610.

Gompers, Paul A., Joy L. Ishii, and Andrew Metrick, 2003. Corporate governance and equity prices, Quarterly Journal of Economics 118 (1), 107-156.

Grinblatt, Mark, Matti Keloharju, and Juhani Linnainmaa, 2012. IQ, trading behavior, and performance. Journal of Financial Economics, 104, 339–362.

Hasbrouck, Joel, 1995. One security, many markets: Determining the contributions to price discovery. Journal of Finance, 50 (4), 1995, 1175–1199.

Huddart, Steven, Bin Ke, and Charles Shi, 2007. Jeopardy, non-public information, and insider trading around SEC 10-K and 10-Q filings. Journal of Accounting & Economics 43 (1), 3-36.

Jaffe, Jeffrey F., 1974. Special information and insider trading. Journal of Business 47 (3), 410- 428.

Jagolinzer, Alan D., David F. Larcker, and Daniel J. Taylor, 2011. Corporate governance and the information content of insider trades. Journal of Accounting Research 49 (5), 1249-1274.

Jegadeesh, Narasimhan, and Sheridan Titman, 1993. Returns to buying winners and selling losers: implications for stock market efficiency. Journal of Finance 48, 65-91.

Jeng, Leslie A., Andrew Metrick, and Richard J. Zeckhauser, 2003. Estimating the returns to insider trading: A performance-evaluation perspective. Review of Economics and Statistics 85 (2), 453-471.

Jenter, Dirk, 2005. Market timing and managerial portfolio decisions. Journal of Finance 60, 1903-1949.

Jiang, Chao, M. Babajide Wintoki, and Yaoyi Xi, 2018. Insider trading and the legal expertise of corporate executives. Working Paper.

Ke, Bin, and Steven Huddart, 2007. Information asymmetry and cross-sectional determinants of insider trading. Contemporary Accounting Research, 24 (1), 195-232.

Ke, Bin, Steven Huddart, and Charles Shi, 2007. Jeopardy, non-public information, and insider trading around sec 10-K and 10-Q filings. Journal of Accounting and Economics, 43, 3-36.

Keim, Donald B. and Ananth Madhavan, 1995. Anatomy of the trading process: Empirical evidence on the behavior of institutional traders. Journal of Financial Economics, 37 (3), 371- 398.

Kyle, Albert, 1985. Continuous auctions and insider trading. Econometrica, 53 (6), 1315-1335.

Lakonishok, Josef, and Inmoo Lee, 2001. Are insider trades informative? Review of Financial Studies 14 (1), 79-111.

Lee, D. Scott, Wayne H. Mikkelson, and M. Megan Partch, 1992. Managers’ trading around stock repurchases. Journal of Finance 47, 1947–61.

Manne, Henry G., 1966. Insider Trading and the Stock Market. Free Press, New York.

Marin, Jose M., Jacques Olivier, 2008. The dog that did not bark: Insider trading and crashes. Journal of Finance 63 (5), 2429-2476.

Odean, Terrance, 1998. Are investors reluctant to realize their losses? Journal of Finance 53, 1775-1798.

Piotroski, Joseph D., and Darren T. Roulstone, 2005. Do insider trades reflect both contrarian beliefs and superior knowledge about future cash flow realizations? Journal of Accounting and Economics 39 (1), 55-81.

Ravina, Enrichetta, and Paola Sapienza, 2010. What do independent directors know? Evidence from their trading. Review of Financial Studies 23 (3), 962-1003.

Rozeff, Michael S., and Mir A. Zaman, 1988. Market efficiency and insider trading: New evidence. Journal of Business, 61, 25–44.

Rozeff, Michael S., and Mir A. Zaman, 1998. Overreaction and insider trading. Journal of Finance, 53, 701–716.

Saar, Gideon, 2001. Price impact asymmetry and block trades: An institutional trading explanation. Review of Financial Studies, 14, 1153–1181.

Scott, James, and Peter Xu, 2004. Some insider sales are positive signals. Financial Analysts Journal, 60 (3), 44-51.

Seyhun, H. Nejat, 1986. Insiders’ profits, costs of trading, and market efficiency. Journal of Financial Economics 16 (2), 189-212.

Seyhun, H. Nejat, 1988. The information content of aggregate insider trading. Journal of Business 61, 1-24.

Seyhun, H. Nejat, 1992. Why does aggregate insider trading predict future stock returns? Quarterly Journal of Economics 107, 1303-1331.

Skaife, Hollis A., David Veenman, and Daniel Wangerin, 2013. Internal control over financial reporting and managerial rent extraction: evidence from profitability of insider trading. Journal of Accounting and Economics 55, 91-110.

Wang, Weimin, Yong-Chul Shin, and Bill B. Francis, 2012. Are CFOs’ trades more informative than CEOs’ trades? Journal of Financial and Quantitative Analysis 47 (4), 743-762.

Yermack, David, 2009. Deductio' ad absurdum: CEOs donating their own stock to their own family foundations. Journal of Financial Economics, 94(1), 107-123.

Figure 1. Example of Form 4 with Family Trades

Table 1. Categories of Insider Trades and Descriptions of Monthly Variables

The top half of this table defines the different categories of direct and indirect insider trades analyzed in this study. The indicator variable associated with each trade category is assigned a value of one for all insider trades in that category, or zero otherwise. The bottom half of this table describes the monthly variables used in our panel regression analysis.

Name of Categories of Insider Trades Indicator Variable

Insider (me) Direct trades for the insider’s own account.

Other Indirect trades by the insider for any other account controlled by the insider.

Subsets of Indirect Trades:

Any_Family a For a spouse, child, or any other family member. Spouse For a spouse. Child For a child. Oth_Fam For a family member other than a spouse or child, or with no mention of a spouse or child. Trust For a trust account.

TrustFam For the trust account of a family member, including a spouse, child, or any other family.

TrustSpouse For the trust account of a spouse.

TrustChild For the trust account of a child.

TrustNotFam For a trust account with no reference to a family member. Retirement For a retirement account.

RetiremtFam For a retirement account with reference to a family member.

RetiremtNotFam For a retirement account with no reference to a family member. Found For the account of a foundation.

Name of Variable Description of Monthly Dependent and Control Variables

ARj, t+1 Future Fama-French four-factor alphas for firm j in month t+1, following the month (t) in which the insider trades. Following Brennan, Chordia, and Subrahmanyam (1998), we compute the firm’s monthly factor loadings using 60-month rolling windows, while requiring at least 24 non-missing months in each 60-month period.

CARj, t+1, t+a Future cumulative Fama-French four-factor alphas for firm j over months t+1 through t+a, following the month (t) in which the insider trades.

AssetGrj,t Annual asset growth, (ATj,t − ATj,t−1)/ATj,t−1, where ATj,t is total assets of firm j in the most recent fiscal year prior to month t.

B/Mj,t Book-to-market ratio of firm j in the most recent fiscal year prior to month t. We take the natural log of this variable in the analysis.

Profitj,t Firm profitability, measured by the gross profit for firm j during the most recent quarter

prior to month t, (SALESj,t – COGSj,t)/ATj,t.

RETj, t-1 Lagged one-month stock return for firm j in month t-1.

RETj, t-12, t-2 Recent cumulative (i.e., momentum) stock return for firm j from month t−12 through t−2.

Table 1, continued

Sizej,t The firm’s market capitalization, measured as the total number of shares outstanding for

firm j (SHROUTj,t) multiplied by price per share (abs(PRCj,t)) at the end of month t. We take the natural log of this variable in the analysis.

Std_Retj,t Volatility of daily stock returns for firm j during month t, measured as the standard deviation across daily returns during the month.

Trade_Sizei,j,t Our measure of the trade size for every insider trade disclosed on a Form 4, regardless of the trade category (k), by insider i of firm j during month t is: Pi,j,t − Si,j,t Trade_Sizei,j,t = , where Pi,j,t is the number of shares of any trade type purchased VOLj,t by insider i at firm j in month t, Si,j,t is the number of shares of any trade type sold, and

VOLj,t is the total share volume by all investors in firm j during month t. Trade_Size_ki,j,t Our measure of the trade size for trades of type k by insider i of firm j during month t is: Pk,i,j,t − Sk,i,j,t Trade_Size_ki,j,t = , where Pk,i,j,t is the number of shares of each trade type (k) VOLj,t purchased by insider i at firm j in month t, Sk,i,j,t is the number of shares of each trade type b (k) sold, and VOLj,t is the total share volume by all investors in firm j during month t. TrSize_k _RKi,j,t Adjusted rank of Trade_Size_ki,j,t or Trade_Sizei,j,t, for every category (k) of trades by

or TrSize_RKi,j,t insider i of firm j during month t, or for all trade categories, constructed as follows. First, for

each category of trades (k), Trade_Sizei,j,t is ranked across all insiders who buy or sell in month t into terciles, which are assigned values from 0 to 2. Second, these ranked values are scaled by 2 to obtain the adjusted rank measure for insider purchases (or sales), which ranges from 0 to 1. a We search for the following lists of keywords for each respective dummy variable that categorizes different subsets of family trades. This search pertains to the insider’s response to both Question 7 of Table 1 in Form 4, Nature of Indirect Beneficial Ownership, and also in any footnoted responses to the question in the Explanation of Responses, found at the end of Form 4. Our search requires appropriate spaces at the beginning and end of words, is case insensitive, and allows for plural or possessive versions of words where appropriate.

CATEGORY Key Words Child Son, Daughter, Child(ren) Spouse Wife, Husband, Spouse Family Son, Daughter, Child(ren), Wife, Husband, Spouse, Family, Mom, Mother, Dad, Father, Niece, Nephew, Aunt, Uncle, Grandchild(ren), Granddaughter, Grandson Retirement 401, ESOP, Profit Sharing, Pension, Retirement, IRA Foundation Foundation, Charity Trust Trust b In constructing these trade size variables, we have also scaled monthly insider net order flow by two other scalars: the firm’s total shares outstanding and total direct shareholdings by insider i at firm j in month t, with robust results.

Table 2. Sample Composition and Summary Statistics

Panel A of this table presents the total number of trades for each category of insider trades, along with the percentage of trades for each type relative to all trades in that group, and relative to all insider trades. We provide the results for these subsets of all trades, as well as for the finer subsets of insider purchases and sales, separately. Panel B provides summary statistics and correlations for the key variables, including the one-month-ahead abnormal return (AR), size of insider trades regardless of account types (Trade_Size), size of direct insider trades (Trade_Size_me), size of all indirect insider trades (Trade_Size_other), size of insider trades through any family account (Trade_Size_fam), firm size (Size), book-to-market ratio (B/M), short term returns over the prior month (RETt-1), momentum returns over the previous eleven months (RETt-12,t-2), asset growth (AssetGr), firm profitability (Profit), and stock return volatility (Std_Ret). The sample period covers July, 2003 through December, 2017.

Panel A. Relative Frequencies for the Different Categories of Insider Trades

Subsets of All Trades Subsets of All Purchases Subsets of All Sales

% of Number % of % of Group of Number % of All of % of All Number % of All Trades for of Trades Group Trades Purchases Group Buys of Sales Group Sales

Insider (Direct) 1,305,857 100% 81.7% 136,653 100% 66.9% 1,169,204 100% 83.9%

Other (Indirect) 292,165 100% 18.3% 67,674 100% 33.1% 224,491 100% 16.1%

Any Family 131,326 100% 8.2% 28,115 100% 13.8% 103,211 100% 7.4% Spouse 32,627 24.8% 2.0% 5,024 17.9% 2.5% 27,603 26.7% 2.0% Child 82,826 63.1% 5.2% 21,179 75.3% 10.4% 61,647 59.7% 4.4% Other Family 47,076 35.8% 2.9% 6,263 22.3% 3.1% 40,813 39.5% 2.9%

Trust 128,918 100% 8.1% 18,095 100% 8.9% 110,823 100% 8.0% For Family 70,105 54.4% 4.4% 11,998 66.3% 5.9% 58,107 52.4% 4.2% Spouse 11,709 9.1% 0.7% 1,734 9.6% 0.8% 9,975 9.0% 0.7% Child 46,039 35.7% 2.9% 9,496 52.5% 4.6% 36,543 33.0% 2.6% Not Family 58,813 45.6% 3.7% 6,097 33.7% 3.0% 52,716 47.6% 3.8%

Retirement 6,865 100% 0.4% 4,221 100% 2.1% 2,644 100% 0.2% For Family 1,939 28.2% 0.1% 1,236 29.3% 0.6% 703 26.6% 0.1% Not Family 4,926 71.8% 0.3% 2,985 70.7% 1.5% 1,941 73.4% 0.1%

Foundation 4,897 100% 0.3% 222 100% 0.1% 4,675 100% 0.3%

All Trades 1,598,022 - - 204,327 - - 1,393,695 - -

Table 2, continued

Panel B. Summary Statistics and Correlations

Variables Mean Std. Dev. Correlations

arde_

ade_

Size_

RET

Size_fam

ade_

Size_me

AssetGr

StdRet

RET

Profit

B/M

Size

ther 12,t

AR 0.12 11.69 1 0.00 0.00 0.01 0.01 0.00 0.00 0.00 0.00 0.01 -0.01 0.00 Trade_Size -0.17 60.93 0.00 1 0.26 0.27 0.10 0.01 0.01 -0.01 -0.02 -0.01 0.00 0.02 Trade_Size_me -0.26 1.41 0.00 0.26 1 0.01 0.01 -0.01 0.05 -0.04 -0.05 -0.04 0.00 0.05 Trade_Size_other -0.04 1.16 0.01 0.27 0.01 1 0.49 0.01 0.01 -0.02 -0.04 -0.03 0.00 0.01 Trade_Size_fam -0.01 0.25 0.01 0.10 0.01 0.49 1 0.00 0.02 -0.01 -0.03 -0.03 0.00 0.00 B/M 0.56 0.58 0.00 0.01 -0.01 0.01 0.00 1 -0.11 0.00 -0.16 -0.22 -0.11 0.09 Size 8,316 21,437 0.00 0.01 0.05 0.01 0.02 -0.11 1 0.00 -0.01 -0.01 0.00 -0.23

RETt-1 2.47 10.82 0.00 -0.01 -0.04 -0.02 -0.01 0.00 0.00 1 0.04 0.02 -0.01 -0.03

RETt-12,t-2 25.07 48.52 0.00 -0.02 -0.05 -0.04 -0.03 -0.16 -0.01 0.04 1 0.03 -0.02 0.02 Profit 34.50 25.9 0.01 -0.01 -0.04 -0.03 -0.03 -0.22 -0.01 0.02 0.03 1 -0.08 -0.03 AssetGr 15.39 34.99 -0.01 0.00 0.00 0.00 0.00 -0.11 0.00 -0.01 -0.02 -0.08 1 0.04 StdRet 2.55 1.48 0.00 0.02 0.05 0.01 0.00 0.09 -0.23 -0.03 0.02 -0.03 0.04 1 *Numbers appearing in bold are significant at the 0.10 level.

Table 3. Calendar Time Portfolio Approach: Performance of Indirect Insider Trades

This Table analyzes calendar time portfolios of stocks purchased or sold by insiders, either directly for their own accounts or indirectly on behalf of family members, trusts, retirement accounts, or foundations. We also examine portfolios based on subsets of indirect trades made in trust or retirement accounts that are devoted to family members. We begin by building portfolios of stocks based on all insider purchases or sales in each trade category (k) during any given one-month period. We then examine the time series of monthly portfolio returns for every type of insider transaction.

The first column presents the monthly four-factor alpha for every portfolio of insider purchases (αk). The second column compares this performance (αk) with the alpha for a portfolio comprising the insider’s own direct purchases (αme). That is, for each type of indirect purchase (k), we also provide the alpha of a hedge portfolio that is long stocks that experience that type of indirect purchase and short stocks with direct purchases made in the insider’s own account (αk – αme). The third and fourth columns provide the same information for each category of indirect sales. Finally, the fifth column presents the performance of a hedge portfolio that is long insider purchases and short insider sales, for every category of indirect trades (k), while the sixth column compares this performance with the alpha of the analogous long-short hedge portfolio based on direct trades. The t-statistics are based on Newey-West robust standard errors with twelve monthly lags. * indicates significance at the .10 level; ** at the .10 level; and *** at the .01 level.

Table 3, continued

Purchases Sales Purchases - Sales

(1) (2) (3) (4) (5) (6) Insider Trades for (type k): αk (%) αk - αme αk (%) αk - αme αk (%) αk - αme

All Trades (αAll) 0.69*** -0.05 0.73*** (4.14) (-0.72) (4.61) For Me (Direct, αme) 0.65*** -0.02 0.67*** (3.63) (-0.32) (3.96)

Other (Indirect, αother) 0.89*** 0.25* -0.10 -0.08 0.99*** 0.33** (4.99) (1.97) (-1.11) (-0.98) (4.98) (2.10)

Any Family (αfam) 1.09*** 0.45*** 0.07 0.09 1.02*** 0.35* (4.37) (3.12) (0.53) (0.78) (3.68) (1.80) Spouse (αspouse) 1.27*** 0.63** 0.08 0.10 1.19*** 0.53* (4.65) (2.48) (0.48) (0.72) (3.65) (1.94) Child (αchild) 1.30*** 0.65*** -0.03 -0.01 1.33*** 0.67** (3.91) (2.76) (-0.19) (-0.08) (3.83) (2.24) Other Family (αoth-fam) 0.47 -0.18 0.14 0.16 0.33 -0.34 (1.27) (-0.50) (0.72) (0.99) (0.94) (-0.96)

Trust (αtrust) 1.05*** 0.40* -0.05 -0.03 1.10*** 0.44** (4.73) (1.82) (-0.52) (-0.42) (4.75) (2.10) Trust For Family (αtrust-fam) 1.39*** 0.74** 0.16 0.18* 1.23*** 0.56 (3.71) (2.08) (1.19) (1.82) (2.96) (1.43) Trust For Spouse (αtrust-spouse) 1.55** 0.91 -0.22 -0.20 1.78*** 1.11 (2.36) (1.29) (-0.88) (-0.88) (2.74) (1.64)

Trust For Child (αtrust-child) 1.35** 0.71 0.13 0.15 1.22** 0.56 (2.50) (1.29) (0.72) (1.05) (2.06) (0.93) Trust Not Family (αtrust-notfam) 0.73*** 0.09 -0.25* -0.22* 0.98*** 0.31 (3.03) (0.32) (-1.77) (-1.68) (3.84) (1.23)

Retirement (αretiremt) 0.76*** 0.11 -0.73*** -0.71*** 1.49*** 0.82*** (2.81) (0.46) (-2.95) (-2.98) (4.48) (2.98) Retirement For Family (αret-fam) 0.24 -0.40 -0.87 -0.85 1.11 0.44 (0.62) (-1.27) (-1.11) (-1.15) (1.32) (0.55) Retirement Not Family (αret-notfam) 0.82*** 0.17 -0.69*** -0.67** 1.51*** 0.85*** (3.05) (0.67) (-2.75) (-2.55) (4.61) (3.11)

Foundation (αfound) 0.75 0.10 -0.27 -0.24 1.02 0.35 (0.94) (0.12) (-0.46) (-0.45) (1.07) (0.36)

Table 4. Portfolio Approach: Monthly Alphas of Different Categories of Insider Trades, for Tercile Portfolios by Trade Size

This Table examines the trading performance of different categories of direct and indirect insider trades, conditioned on trade size. For each month (t), we sort all insider trades of a given type (k) into three groups, based on the insider’s relative trade size within that group (Trade_Size_k), to form three portfolios that experience insider trading that ranges from large sales (LS) to large purchases (LP). The resulting three portfolios for each trade type (k) are then held for one month, in month t+1. For each tercile portfolio, for every trade category (k), we then report the risk adjusted abnormal returns (i.e., monthly alphas, αk) from the Fama and French four-factor model. We also provide the analogous alphas for hedge portfolios that are long the portfolio of stocks comprised of indirect trades of each type (k) and short the portfolio comprised of direct trades (αk – αme), within each tercile by trade size. The t-statistics are based on Newey-West robust standard errors with twelve lags. * indicates significance at the .10 level; ** at the .10 level; and *** at the .01 level.

Trade Category (k) Hedge Portfolios (1) (2) (3) (4) (5) Long Trade Type (k) and Short Direct Trades Trade Direct Indirect Family Spouse Child Size (2) - (1) (3) - (1) (4) - (1) (5) - (1) (αme) (αother) (αfam) (αspouse) (αchild)

Large 0.02 -0.01 0.19 0.18 0.03 -0.03 0.16 0.16 0.00 Sales (0.25) (-0.09) (1.23) (0.81) (0.11) (-0.36) (1.13) (0.83) (0.01)

Middle 0.04 0.10 0.12 0.38 0.02 0.06 0.08 0.35 -0.02 Group (0.36) (0.78) (0.61) (1.32) (0.10) (0.53) (0.62) (1.47) (-0.17)

Large 0.38*** 0.84*** 0.94*** 1.08*** 1.07*** 0.46** 0.56** 0.70*** 0.70** Purchases (3.49) (3.59) (3.37) (3.83) (3.13) (2.47) (2.48) (2.67) (2.40)

Hedge Portfolios 0.35*** 0.85*** 0.75** 0.90** 1.05*** 0.49** 0.39 0.54* 0.69* LP - LS (2.71) (3.38) (2.37) (2.35) (3.01) (2.18) (1.31) (1.71) (1.93)

Table 5. Panel Regression Approach: The Short Run Relative Performance of Direct versus Indirect Trades

This table estimates the short run (one-month-ahead) relative trading profitability of large purchases versus large sales (LP – LS), for every category of direct or indirect insider trades (k), in the month following the trades. We regress the one-month-ahead market-adjusted abnormal stock return on the scaled rank of insider trade size for each category of insider trades, along with other control variables, as follows: 15 ARj,t+1 = αt + ∑푘=1 훽푘 푇푟푆푖푧푒_푘_푅퐾i,j,t + Controlsi,j,t + εi,j,t ; (1) where k indexes the fifteen categories of direct and indirect trades by insider (i) analyzed here. The control variables are listed in Table 1. The dependent variable, ARj,t+1, is the leading one-month-ahead Fama French 4-factor alpha for the firm (j). We multiply ARj,t+1 by 100 to reflect performance in percentage terms. TrSize_k_RK is the scaled tercile rank of trade size for each category of trades analyzed (k = 1 - 15). For each category of direct or indirect trades (k), the coefficient of the scaled rank of trade size (βk) is analogous to the return on a hedge portfolio that is long the tercile of large purchases of type k and short the tercile of large sales (LP – LS). To see this result, consider the association between the scaled rank of each trade size measure and future returns implied by ∂퐴푅(푗,푡+1) Equation (1): = 훽 . This partial derivative shows that a one-unit increase in the scaled rank of trade size, which compares the tercile of large sales ∂푇푟푆푖푧푒_푘_푅퐾(푗,푡) 푘 with the tercile of large purchases (i.e., changing TrSize_k_RK from 0 to 1), is associated with a change in ARj,t+1 of βk percent. Monthly fixed effects are included, and standard errors are clustered by time at the monthly level. The t-statistics are provided in parentheses below the parameter estimates. At the bottom of each column, we provide F-statistics that test the equality of different pairs of regression coefficients. Throughout the table, * indicates significance at the .10 level; ** at the .05 level; and *** at the .01 level.

Table 5, continued

Trade Category Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

TrSize_me_RK β 1 0.502*** 0.480*** 0.473*** 0.505*** 0.505*** 0.501*** 0.456*** 0.455*** 0.499*** 0.493*** (All Direct Trades) (4.19) (4.04) (3.99) (4.16) (4.16) (4.17) (3.89) (3.89) (4.14) (4.09)

TrSize_other_RK β2 0.950*** (All Indirect Trades) (5.11)

TrSize_fam_RK β3 0.999*** 0.977*** (4.12) (4.03) TrSize_spouse_RK β4 0.723* 0.659* (1.83) (1.68) TrSize_child_RK β5 1.238*** 1.242*** (3.50) (3.51) TrSize_oth-fam_RK β6 -0.097 -0.105 (-0.22) (-0.24)

TrSize_trust_RK β7 1.198*** (4.87) TrSize_trust-fam_RK β8 1.045*** (3.35) TrSize_trust-notfam_RK β9 1.278*** 1.277*** 1.225*** 1.237*** (4.47) (4.45) (4.27) (4.33) TrSize_trust-spouse_RK β10 1.668*** (2.65) TrSize_trust-child_RK β11 1.134** (2.03)

TrSize_retirement_RK β12 1.130** (2.31) TrSize_ret-fam_RK β13 0.922 (0.72) TrSize_ret-notfam_RK β14 1.249** 1.059** 1.066** (2.36) (2.00) (2.01)

TrSize_found_RK β15 -0.615 -0.376 (-1.01) (-0.62)

Table 5, continued

Control Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

B/M β16 0.688 0.631 0.691 0.690 0.711 0.688 0.847 0.842 0.592 0.648 (0.07) (0.06) (0.07) (0.07) (0.07) (0.07) (0.08) (0.08) (0.06) (0.06) Size β17 -15.371*** -15.584*** -15.638*** -15.582*** -15.576*** -15.632*** -15.669*** -15.656*** -15.375*** -15.428*** (-3.42) (-3.47) (-3.48) (-3.47) (-3.46) (-3.48) (-3.48) (-3.47) (-3.42) (-3.43) RETi,j,t-1 β18 -0.509 -0.554 -0.558 -0.551 -0.551 -0.554 -0.582 -0.581 -0.526 -0.530 (-0.45) (-0.48) (-0.49) (-0.48) (-0.48) (-0.48) (-0.51) (-0.51) (-0.46) (-0.46) RETi,j,t-12,t-2 β19 -0.024 -0.034 -0.036 -0.032 -0.033 -0.034 -0.041 -0.041 -0.029 -0.030 (-0.10) (-0.14) (-0.14) (-0.13) (-0.13) (-0.14) (-0.17) (-0.16) (-0.12) (-0.12) Profit β20 0.359 0.344 0.343 0.348 0.348 0.349 0.334 0.334 0.357 0.357 (1.02) (0.97) (0.97) (0.99) (0.99) (0.99) (0.95) (0.95) (1.01) (1.01) AssetGr β21 -0.355** -0.359** -0.361** -0.359** -0.360** -0.362** -0.366** -0.366** -0.358** -0.361** (-2.10) (-2.13) (-2.14) (-2.13) (-2.13) (-2.15) (-2.17) (-2.17) (-2.12) (-2.14) StdRet β22 -4.496 -4.270 -4.244 -4.509 -4.509 -4.518 -4.129 -4.126 -4.510 -4.491 (-0.41) (-0.39) (-0.39) (-0.42) (-0.42) (-0.42) (-0.38) (-0.38) (-0.42) (-0.41) N 323,639 323,639 323,639 323,639 323,639 323,639 323,639 323,639 323,639 323,639 Adj. R2 0.010 0.090 0.010 0.010 0.010 0.010 0.009 0.009 0.010 0.010

Hi: Testing equality β1 = β2 β1 = β3 β1 = β5 β1 = β7 β1 = β8 β1 = β9 β1 = β12 β1 = β13 β1 = β3 β1 = β4 F-Statistic 6.2 5.0 4.5 10.5 3.6 8.4 1.9 0.0 4.2 4.3 (p-value) (.01)** (.03)** (.04)** (.00)*** (.06)* (.00)*** (.18) (.72) (.04)** (.04)** β1 = β9 β1 = β10 β1 = β14 β1 = β9 β1 = β5 8.4(.00)*** 3.4(.07)* 2.2(.14) 7.3(.00)*** 0.2(.67) β1 = β11 β1 = β14 β1 = β6 1.3(.25) 1.1(.30) 2.0(.16) β1 = β15 β1 = β9 3.4(.07)* 7.8(.01)*** β1 = β14 1.1(.30) β1 = β15 2.1(.15)

Table 6. Panel Regression: The Long Run Performance of Direct versus Indirect Trades

This table presents the long run relative trading performance of direct versus indirect trades over different periods that extend from month t+1 through month t+a. In this Table, we analyze whether the short run performance documented in Table 5 extends beyond one month, by considering the cumulative abnormal returns measured over time frames that span different subsets of the three years following insider trades. For this analysis, we estimate a revised version of Equation (1) that replaces ARi,j,t+1 as the dependent variable with CARj,t+1,t+a, as follows:

CARj,t+1,t+a = αt + β1 TrSize_me_RKi,j,t + β2 TrSize_fam_RKi,j,t + β3 TrSize_trust-notfam_RKi,j,t

+ β4 TrSize_ret-notfam_RKi,j,t + β5 TrSize_found_RKi,j,t + Controlsi,j,t + εi,j,t . (2)

The first column reproduces the results for one-month-ahead abnormal returns (ARj,t+1), from our main model in column (9) of Table 5. The remaining columns present the analogous results for performance measured over longer time frames that extend up to 36 months in the future. We include the same controls as our analysis of Equation (1) in Table 5. The coefficients of the controls have similar implications to those presented in Table 5, and are omitted here for brevity. All variables are defined in Table 1. Monthly fixed effects are included, and standard errors are clustered by time at the monthly level. The t-statistics are provided in parentheses below the parameter estimates. At the bottom of each column, we provide F-statistics that test the equality of different pairs of parameter estimates. Throughout this table, * indicates significance at the .10 level; ** at the .05 level; and *** at the .01 level.

(1) (2) (3) (4) (5) (6) VARIABLES ARj,t+1 CARj,t+1,t+3 CARj,t+1,t+6 CARj,t+1,t+12 CARj,t+1,t+24 CARj,t+1,t+36

TrSize_me_RK β1 0.499*** 1.105*** 1.196*** 2.058*** 2.191*** 3.097*** (4.14) (5.59) (4.39) (5.10) (3.62) (3.78)

TrSize_fam_RK β2 0.977*** 2.074*** 3.710*** 4.505*** 6.308*** 7.836*** (4.03) (4.25) (4.71) (4.69) (4.57) (4.01)

TrSize_trust-notfam_RK β3 1.225*** 1.460*** 1.829** 4.244*** 10.392*** 14.952*** (4.27) (3.04) (2.35) (4.29) (6.02) (7.01)

TrSize_ret_notfam_RK β4 1.059** 2.322*** 2.213 5.056*** 7.125** 12.518*** (2.00) (2.63) (1.58) (2.65) (2.55) (3.31)

TrSize_found_RK β5 -0.615 -0.277 0.416 0.205 0.160 7.024 (-1.01) (-0.24) (0.24) (0.08) (0.04) (1.45)

Controls Yes Yes Yes Yes Yes Yes N 323,639 321,949 319,247 301,551 266,396 233,698 Adj. R2 0.010 0.009 0.011 0.011 0.009 0.011

Hi: Testing equality; F-Statistic (p-value)

1. β1 = β2 4.2(.04)** 4.2(.04)* 10.3(.00)*** 6.9(.01)*** 10.0(.00)*** 6.7(.01)** 2. β1 = β3 7.3(.00)*** 0.6(.32) 0.7(.40) 5.3(.02)** 26.1(.000)*** 38.4(.00)*** 3. β1 = β4 1.1(.30) 1.9(.17) 0.5(.47) 2.2(.14) 3.2(.08)* 5.9(.02)**

4. β1 = β5 3.4(.07)* 1.4(.24) 0.2(.66) 0.5(.49) 0.3(.62) 0.7(.42) 5. β2 = β3 0.6(.45) 1.0(.32) 3.8(.05)* 0.0(.83) 4.2(.04)** 7.5(.00)*** 6. β2 = β4 0.0(.89) 0.1(.81) 0.9(.35) 0.1(.80) 0.1(.78) 1.3(.26) 7. β2 = β5 5.3(.02)** 3.2(.08)* 2.7(.10) 2.2(.14) 2.2(.14) 0.0(.88)

Table 7. Direct versus Indirect Trades and Future Firm-Specific Informational Events

Panel A reports the results from the following panel regression analysis that relates direct trades and the non- overlapping categories of indirect trades to the firm’s next earnings surprise:

Surprisej,q = αt + β1 TrSize_me_RKi,j,t + β2 TrSize_other_RK i,j,t (or β3 TrSize_fam_RKi,j,t

+ β4 TrSize_trust-notfam_RKi,j,t + β5 TrSize_ret_notfam_RKi,j,t + β6 TrSize_found_RKi,j,t)

+ β7 lag1Surprisej,q + Controlsj,t + ε i,j,t. (3) The dependent variable is the earnings surprise for the next earnings announcement by firm j in quarter q, following the trades of insider i during month t, measured by standardized unexpected earnings (SUE).

Panel B presents the results of probit regression analysis that relates direct and indirect trades to the likelihood of an imminent large idiosyncratic stock price change within the next 10 trading days following an insider trade:

−1 Φ (+/ −∆Pj,t) = αt + β1 TrSize_me_RKi,j,t + β2 TrSize_other_RK i,j,t (or β3 TrSize_fam_RKi,j,t

+ β4 TrSize_trust-notfam_RKi,j,t + β5 TrSize_ret_notfam_RKi,j,t + β6 TrSize_found_RKi,j,t)

+ Controlsj,t + ε i,j,t. (4) Φ(.) represents the cumulative distribution function for the standard normal distribution. The sample of large price change events is identified as follows. First, for each firm we compute the three-day cumulative abnormal return (CAR) around every trading day during a given year. If the CAR for a given day is among the top (bottom) 5% among all trading days in the year, that day is identified as having a large positive (negative) price change. If such a large price increase (decrease) occurs within 10 days following an insider trade, the dummy variable +∆P (−∆P) is assigned a value of one, and zero otherwise. We also present the marginal effects implied by this probit analysis. We include the same control variables as our analysis of Equation (1) in Table 5. Monthly fixed effects are included, and standard errors are clustered by time at the monthly level. The t-statistics are provided in parentheses below the parameter estimates. At the bottom of each column, we also provide F-statistics or χ2-statistics that test the equality of different pairs of parameter estimates. Throughout this table, * indicates significance at the .10 level; ** at the .05 level; and *** at the .01 level.

Table 7, continued

Panel A. Standardized Unexpected Earnings (SUE)

(1) (2) VARIABLES SUE SUE

TrSize_me_RK β1 -0.014 -0.014 (All Direct Trades) (-1.46) (-1.49)

TrSize_other_RK β2 0.019 (All Indirect Trades) (1.07)

TrSize_fam_RK β3 0.048* (1.95)

TrSize_trust-notfam_RK β4 -0.009 (-0.28)

TrSize_ret_notfam_RK β5 0.020 (0.32)

TrSize_found_RK β6 -0.043 (-0.45)

lag1SUE β7 0.303*** 0.303*** (35.10) (35.10) Controls Yes Yes N 189,729 189,729 Adj. R2 0.120 0.120

H1: Testing Equality β1 = β2 β1 = β3 F-Statistic (p-value) 4.0(.05)* 7.0(.01)***

Table 7, continued

Panel B. Large Idiosyncratic Stock Price Changes (ΔP)

(1) (2) (3) (4) (5) (6) (7) (8) VARIABLES +∆P +∆P +∆P +∆P -∆P -∆P -∆P -∆P

TrSize_me_RK β1 0.268*** 0.142*** 0.267*** 0.140*** -0.047*** -0.082*** -0.045*** -0.080*** (All Direct Trades) (20.53) (12.90) (20.28) (12.72) (-4.02) (-6.84) (-3.89) (-6.74)

TrSize_other_RK β2 0.361*** 0.257*** -0.140*** -0.152*** (All Indirect Trades) (16.97) (12.30) (-6.56) (-7.41)

TrSize_fam_RK β3 0.327*** 0.238*** -0.115*** -0.123*** (10.47) (7.68) (-3.59) (-3.81)

TrSize_trust-notfam_RK β4 0.358*** 0.221*** -0.101*** -0.147*** (10.31) (6.15) (-3.57) (-5.26)

TrSize_ret_notfam_RK β5 0.245*** 0.138** -0.074 -0.080 (4.23) (2.30) (-1.15) (-1.22)

TrSize_found_RK β6 0.209* 0.154 0.096 0.046 (1.87) (1.34) (0.88) (0.42)

Controls No Yes No Yes No Yes No Yes N 325,997 325,997 325,997 325,997 325,997 325,997 325,997 325,997 Pseudo R2 0.036 0.062 0.035 0.062 0.035 0.047 0.035 0.046

Marginal Effects β1 0.075 0.039 0.075 0.038 -0.013 -0.022 -0.012 -0.022 β2 0.101 0.070 - - -0.039 -0.041 - - β3 - - 0.092 0.065 - - -0.032 -0.033 β4 - - 0.100 0.060 - - -0.028 -0.040 β5 - - 0.069 0.038 - - -0.020 -0.022 β6 - - 0.059 0.042 - - 0.027 0.013

Hi: Testing Equality χ2 (p-value)

1. β1 = β2 21.7(.00)*** 32.1(.00)*** - - 21.2(.00)*** 11.9(.00)*** - - 2. β1 = β3 - - 4.2(.04)** 10.1(.00)*** - - 4.8(.03)** 1.7(.19) 3. β1 = β4 - - 7.0(.01)*** 5.3(.02)** - - 3.8(.05)** 5.6(.02)** 4. β1 = β5 - - 0.1(.71) 0.0(.98) - - 0.2(.65) 0.0(.99) 5. β1 = β6 - - 0.3(.61) 0.0(.91) - - 1.7(.20) 1.3(.25)

Table 8. The Intensity of SEC Scrutiny and Indirect Insider Trading

This table estimates a time series regression model to analyze whether the ratio of indirect trades (or family trades) to direct trades in a given month (t) is sensitive to the extent of recent SEC enforcement activity regarding insider trading abuse. The dependent variable is the ratio of the number of insider trades in any indirect account (or a family account) to the number of direct insider trades during month t. The independent variables of interest include three monthly lagged values of the number of SEC investigations against insider trading during months t−1, t−2, and t−3, respectively (i.e., SECt−k, k = 1 − 3). We also control for the one-month lagged stock market return (i.e., MKTRETt−1), and the previous cumulative stock market return from month t−13 through month t−2 (i.e., MKTRETt−13, t−2). The model is specified as follows:

Ratiot = α + β1 SECt−1 + β2 SECt−2 + β3 SECt−3 + β4 MKTRETt−1 + β5 MKTRETt−13, t−2 + ε t . (5) The sample period is determined by the availability of SEC investigation data which spans the period, 2003 through 2012. Newey-West adjusted t-ratios appear in parentheses (with 12 monthly lags). *, **, and *** indicate statistical significance at the .10, .05, and .01 levels, respectively.

(1) (2) (3) (4) (5) (6) (7) (8)

Dependent Var: Ratio = #Indirect Trades / # Direct Trades Ratio = #Family Trades / #Direct Trades

SECt-1 -0.005** -0.003* -0.003** -0.001 (-2.49) (-1.91) (-2.30) (-1.64)

SECt-2 -0.006*** -0.005** -0.003** -0.003** (-2.63) (-2.18) (-2.48) (-2.06)

SECt-3 -0.004** -0.001 -0.002** -0.001 (-2.10) (-0.73) (-2.28) (-0.89)

MKTRETt-1 -0.177*** -0.179*** -0.177*** -0.177*** -0.056** -0.056** -0.056* -0.055** (-3.27) (-3.36) (-2.92) (-3.29) (-2.06) (-2.15) (-1.86) (-2.08)

MKTRETt-13, t-2 -0.057*** -0.058*** -0.056*** -0.058*** -0.018*** -0.018*** -0.017*** -0.018*** (-4.36) (-4.45) (-4.01) (-4.52) (-2.91) (-3.02) (-2.76) (-3.01)

N 114 114 114 114 114 114 114 114 Adj. R2 0.203 0.214 0.186 0.215 0.088 0.109 0.077 0.108

Table 9. The Attributes of Insiders and Their Firms, for Subsets of Insiders Who Make Direct Trades versus Indirect Trades

This Table compares the personal attributes of insiders and their respective firms, across subsets of insiders who make direct trades versus indirect trades. IO is the percentage institutional ownership in the insider’s firm. The other firm attributes are described in Table 1. The remaining variables pertain to the attributes of the insiders themselves, as follows. AGE is the insider’s age. YEAR_ EXP is the number of years of experience since the insider’s first year of insider trading. TRADENO is the total number of previous trading months for the insider. TIMEROLE is the number of years since the insider’s first year in the current position. CEO, CHAIRMAN, GENERAL COUNSEL, and STRATEGY COMMITTEE are indicator variables that take a value of one if the insider is the CEO, Chair of the Board, general counsel, or serves on the strategy committee, respectively. NONROUTINE is an indicator variable for opportunistic insiders, following Cohen, Malloy, and Pomorski (2012). AH_OPPO is an indicator variable that identifies opportunistic insiders, based on Ali and Hirshleifer (2017). SHORT HORIZON is an indicator variable for insiders with a short investment horizon, following Akbas, Jiang, and Koch (2019). TOTAL COMPENSATION is the total compensation of the insider, and DELTA is a measure of pay-performance sensitivity following Coles, Daniel, and Naveen (2006). FEMALE is a dummy variable for female insiders. The sample period covers July 2003 through December 2017. The t-statistics (in parentheses) are based on Newey-West robust standard errors with 12 lags. *, **, and *** indicate significance at the .10, .05, and .01 levels, respectively.

Variable Direct Indirect I − D t-Stat Firm Attributes: SIZE 8516.68 6614.60 -1902.08*** (-7.80) B/M 0.55 0.63 0.08*** (5.15)

RETt-1 2.53 2.09 -0.44*** (-2.75)

RETt-12, t-2 25.27 23.73 -1.54 (-1.20) Profit 34.74 33.14 -1.60*** (-3.68) AssetGr 15.45 16.01 0.56 (1.55) StdRet 2.52 2.81 0.29*** (9.40) IO 0.72 0.64 -0.08*** (-20.06) Insider Attributes: AGE 58.52 59.95 1.42*** (7.28) YEAR_EXP 4.17 4.39 0.22*** (3.15) TRADENO 8.49 10.19 1.69*** (8.60) TIMEROLE 6.25 6.90 0.66*** (5.17) CEO 0.12 0.14 0.02*** (5.28) CB 0.05 0.09 0.04*** (8.35) GENERAL COUNSEL 0.03 0.01 -0.02*** (-9.61) STRATEGY COMMITTEE 0.05 0.06 0.01*** (4.57) NONROUTINE 0.61 0.53 -0.08*** (-9.42) AH_OPPO 1.95 2.00 0.05** (2.33) SHORT_HORIZON 0.55 0.66 0.11*** (4.12) TOTAL COMPENSATION 7.48 7.61 0.14*** (5.50) DELTA 4.43 5.38 0.94*** (24.83) FEMALE 0.08 0.07 -0.01*** (-3.20)

Appendix A

Documented cases of informed insider trading through family accounts

There have been numerous documented cases of illegal insider trading in which the alleged informed trading was carried out through indirectly controlled family accounts. The following discussion summarizes seven such cases that involve corporate executives trading through the accounts of family members, which ultimately led to SEC charges.

1. SEC v. Peter C. Chang (09/20/2017)

This case involved alleged serial insider trading in the securities of Alliance Fiber Optic

Products, Inc. by Defendant Peter C. Chang, who served as the company’s Chairman of the

Board, Chief Executive Officer, and President from its formation in 1995 until its acquisition in

2016 by Corning, Inc.

According to the SEC, “…in 2015 and 2016, Chang, by virtue of his leadership positions at AFOP, acquired material nonpublic information about AFOP’s earnings results and financial performance, as well as the intended acquisition of AFOP by Corning, all of which were significant, market moving information. Chang was the largest holder of AFOP stock and was required to disclose his ownership of AFOP securities as an officer and director in accordance with the federal securities laws. But to capitalize on the highly sensitive information he learned about AFOP without detection, Chang secretly traded AFOP shares in two nominee accounts – one held in his wife’s name, and the other in his brother’s name – in advance of two public earnings announcements and the public acquisition announcement...” In addition to this, the SEC complaint also highlighted that fact that “…Chang allegedly tried to hide his control over one of the accounts by posing as his brother in communications with one of the brokerage firms, and he

54 allegedly obscured his relationship with his wife in response to a market surveillance inquiry by the Financial Industry Regulatory Authority…” The SEC also alleged that “…In total, Chang’s insider trading scheme generated more than $2 million in illicit profits and losses avoided, with at least $1.5 million for Chang’s nominee accounts, and more than $600,000 for Chang’s brother’s account...” On February 21, 2018, Peter C. Chang pleaded guilty to illegal insider trading and tender offer fraud. As part of his guilty plea, Chang admitted that he traded AFOP stock through brokerage accounts in the names of his wife and brother.

(https://www.sec.gov/litigation/litreleases/2017/lr23937.htm)

2. SEC v. Alexander J. Yaroshinsky (06/23/2006)

This case involved transactions in the securities of Connetics Corp. by Alexander J.

Yaroshinsky, the former Vice President of Biostatistics and Clinical Operations (and another defendant, Victor E. Zak). The SEC complaint alleges that, in 2005 “…between April 13 and

June 10, Yaroshinsky and Zak executed numerous trades. Yaroshinsky purchased put contracts in his own account and in a nominee account opened in the name of his mother-in-law and sold shares of Connetics common stock in his own account. Zak purchased put contracts, sold short

Connetics shares, and sold his long position of Connetics shares. All of the trading by defendants was conducted in advance of a June 13, 2005 public announcement by Connetics stating that it had received a "not approvable" letter from the Food and Drug Administration ("FDA") concerning Velac Gel. After the announcement, Connetics' stock price fell 27%...”

In 2008, Alexander J. Yaroshinsky agreed to pay $723,000 to settle charges that he and a neighbor traded on nonpublic information.

(https://www.sec.gov/litigation/litreleases/2006/lr19738.htm)

3. SEC v. Jorge Eduardo Ballesteros (05/08/2001)

The SEC complaint alleges that Jorge Ballesteros (the former Chairman of Grupo

Mexicano de Desarrollo, S.A., a major Mexican construction company) received a tip from his brother, the late Jose Luis Ballesteros, who was then a director of Nalco Chemical Company.

The complaint stated that Jorge Ballesteros was tipped about a possible acquisition of Nalco prior to the June 28, 1999 public announcement that Nalco would be acquired by Suez Lyonnaise des Eaux, S.A. Following the tip, Jorge Ballesteros placed orders to purchase over $5.7 million in Nalco stock through Swiss accounts controlled by offshore trusts and nominee companies in the names of his wife and mother, respectively, resulting in illegal profits of over $2 million. In

2003, without admitting or denying the allegations of the SEC's complaint, Jorge Ballesteros agreed to pay a penalty of $2,573,875, as one part of settling the case.

(https://www.sec.gov/litigation/litreleases/lr18441.htm)

4. SEC v. Frank Hixon, Jr. (02/21/2014)

This allegation claims that Hixon regularly traded through the brokerage account of

Destiny Robinson, the mother of his young child, as well as his father, often within minutes of learning confidential information while working as a New York City-based investment banker.

When confronted with evidence of this illicit trading activity, Hixon pretended to not recognize the names of his father or his child’s mother. The evidence later revealed that he was apparently generating illegal trading profits, partly, in lieu of paying formal child support.

(https://www.sec.gov/news/press-release/2014-40)

5. SEC v. Saleem Khan (06/13/2014)

This case alleges that Kahn was repeatedly tipped by his friend, Roshanlal Chaganlal, a director at headquarters of Ross Stores in Dublin, CA. Khan allegedly traded more than 40 times in advance of the company’s disclosure of its ongoing financial performance. In addition to trading in his own account, Khan traded in the account of his brother-in-law, as well as another acquaintance. He also tipped two work colleagues who also traded on the non-public information. This insider trading activity lead to more than $12 million in collective profits.

(https://www.sec.gov/news/press-release/2014-117)

6. SEC v. Yue Han (11/25/2015)

This case involves the allegation that Han abused his privileged access to Goldman

Sachs’ email system, which he was supposed to be using to conduct routine electronic surveillance of Goldman’s employees. He allegedly used this non-public information about forthcoming M&A deals to buy out-of-the-money call options on at least four companies that were about to be acquired. Han traded not only in his own account, but also in an account belonging to his father Wei Han, who lives in China.

(https://www.sec.gov/news/pressrelease/2015-267.html)

7. SEC v. John Alfriyie (04/13/2016)

In this case, the SEC alleges that John Alfriyie, a research analyst, “reaped more than

$1.5 million … through trades he made in his mother’s brokerage account based on non-public information he learned at work. The SEC alleges that John Afriyie found out about an impending acquisition of home security company The ADT Corporation when prospective acquirer Apollo

Global Management approached the investment firm where he worked and discussed potential debt financing for a public-to-private deal. Afriyie subsequently accessed several highly confidential, deal-related documents on the firm’s computer network and purchased thousands of high-risk, out-of-the-money ADT call options in his mother’s account in anticipation that ADT’s stock price would rise when the transaction was publicly announced. The ADT deal was announced on February 16, and afterwards Afriyie sold all of the ADT options in his mother’s account to obtain his illicit profits … ‘Insider traders should have learned by now that trying to hide their illegal activity in a relative’s account ultimately won’d work,’ said Jina L. Choi,

Director of the SEC’s San Francisco Regional Office.’ ”

(https://www.sec.gov/news/pressrelease/2016-67.html)

Appendix B.1 Raw Returns: The Short Run Performance of Direct versus Indirect Trades

This table estimates the short run (one-month) relative trading profitability of large purchases versus large sales (LP – LS), for every category of direct or indirect insider trades (k), in the month following the trades. We regress the one-month-ahead raw stock return on the scaled rank of insider trade size for each category of insider trades, along with other control variables, as follows: 15 Rj,t+1 = α + ∑푘=1 훽푘 푇푟푆푖푧푒_푘_푅퐾i,j,t + Controlsi,j,t + εi,j,t ; (6) where k indexes the fifteen categories of direct and indirect insider trades analyzed here. The control variables are listed in Table 1. The dependent variable, Rj,t+1, is the leading one-month-ahead raw return for the firm (j) of insider (i). We multiply Rj,t+1 by 100 to reflect performance in percentage terms. TrSize_k_RK is the scaled tercile rank of trade size for each category of trades analyzed here (k = 1 - 15). For each category of direct or indirect trades (k), the coefficient of the scaled rank of trade size (βk) is analogous to the return on a hedge portfolio that is long the tercile of large purchases of type k and short the tercile of large sales (LP – LS). To see this result, consider the association ∂푅(푖,푗,푡+1) between the scaled rank of each trade size measure and future returns implied by Equation (1): = ∂푇푟푆푖푧푒_푘_푅퐾(푖,푗,푡) 훽푘. This partial derivative shows that a one-unit increase in the adjusted scaled rank of trade size, from the tercile of large sales to the tercile of large purchases (i.e., changing TrSize_k_RK from 0 to +1), is associated with a change in Rj,t+1 of βk percent. Monthly fixed effects are included, and standard errors are clustered by time at the monthly level. The t-statistics are provided in parentheses below the parameter estimates. At the bottom of each column, we provide F-statistics that test the equality of different pairs of regression coefficients. Throughout the table, * indicates significance at the .10 level; ** at the .05 level; and *** at the .01 level.

Appendix B.1, continued

Trade Category Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

TrSize_me_RK β 1 0.331*** 0.312*** 0.309*** 0.332*** 0.332*** 0.330*** 0.294*** 0.293*** 0.327*** 0.325*** (All Direct Trades) (2.82) (2.71) (2.68) (2.83) (2.83) (2.84) (2.61) (2.61) (2.79) (2.76)

TrSize_other_RK β2 0.743*** (All Indirect Trades) (3.84)

TrSize_fam_RK β3 0.744*** 0.735*** (3.05) (3.03) TrSize_spouse_RK β4 0.539 0.488 (1.49) (1.35) TrSize_child_RK β5 0.956*** 0.963*** (2.74) (2.76) TrSize_oth-fam_RK β6 0.037 0.052 (0.09) (0.12)

TrSize_trust_RK β7 0.930*** (3.88) TrSize_trust-fam_RK β8 0.750** (2.53) TrSize_trust-notfam_RK β9 1.044*** 1.041*** 1.007*** 1.015*** (3.68) (3.66) (3.56) (3.59) TrSize_trust-spouse_RK β10 1.307** (2.27) TrSize_trust-child_RK β11 0.887* (1.74)

TrSize_retirement_RK β12 0.827* (1.66) TrSize_ret-fam_RK β13 0.175 (0.13) TrSize_ret-notfam_RK β14 1.043* 0.884 0.888 (1.90) (1.62) (1.62)

TrSize_found_RK β15 -0.968 -0.844 (-1.59) (-1.37)

Appendix. B.1, continued

Control Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

B/M β16 17.021 16.998 17.019 17.037 17.062 17.036 17.162 17.150 16.963 16.979 (1.36) (1.36) (1.36) (1.36) (1.37) (1.36) (1.38) (1.37) (1.36) (1.36) Size β17 -9.476 -9.648 -9.675 -9.639 -9.636 -9.673 -9.711 -9.697 -9.473 -9.499 (-1.49) (-1.52) (-1.52) (-1.52) (-1.52) (-1.52) (-1.53) (-1.53) (-1.49) (-1.49) RETi,j,t-1 β18 0.333 0.296 0.294 0.299 0.299 0.298 0.275 0.276 0.319 0.318 (0.35) (0.31) (0.31) (0.32) (0.32) (0.31) (0.29) (0.29) (0.34) (0.33) RETi,j,t-12,t-2 β19 -0.101 -0.108 -0.109 -0.107 -0.107 -0.108 -0.114 -0.114 -0.104 -0.105 (-0.45) (-0.49) (-0.49) (-0.48) (-0.48) (-0.49) (-0.51) (-0.51) (-0.47) (-0.47) Profit β20 0.370 0.357 0.358 0.361 0.361 0.363 0.350 0.350 0.368 0.369 (1.19) (1.15) (1.15) (1.16) (1.16) (1.17) (1.13) (1.13) (1.18) (1.19) AssetGr β21 -0.349** -0.352** -0.354** -0.353** -0.353** -0.355** -0.358** -0.358** -0.352** -0.353** (-2.20) (-2.22) (-2.23) (-2.22) (-2.22) (-2.23) (-2.25) (-2.25) (-2.21) (-2.22) StdRet β22 4.784 4.965 4.974 4.774 4.772 4.758 5.068 5.076 4.775 4.779 (0.28) (0.30) (0.30) (0.28) (0.28) (0.28) (0.30) (0.30) (0.28) (0.28) Constant β23 1.866 1.893* 1.504 1.798 1.367 0.652 1.863 1.666 1.411 0.963 (1.61) (1.67) (1.30) (1.56) (1.17) (0.55) (1.56) (1.32) (1.21) (0.82) N 325,997 325,997 325,997 325,997 325,997 325,997 325,997 325,997 325,997 325,997 Adj. R2 0.154 0.154 0.154 0.154 0.154 0.154 0.154 0.154 0.154 0.154

Hi: Testing equality β1 = β2 β1 = β3 β1 = β4 β1 = β7 β1 = β8 β1 = β9 β1 = β12 β1 = β13 β1 = β3 β1 = β4 F-Statistic 6.3 4.2 3.6 8.6 2.3 7.8 1.1 0.0 3.7 3.4 (p-value) (.01)** (.04)** (.06)* (.00)*** (.13) (.00)*** (.29) (.93) (.06)* (.07)* β1 = β9 β1 = β10 β1 = β14 β1 = β9 β1 = β5 7.9(.01)*** 2.8(.10)* 1.9(.17) 7.1(.00)*** 0.2(.65) β1 = β11 β1 = β14 β1 = β6 1.3(.26) 1.0(.32) 0.5(.50) β1 = β15 β1 = β9 4.7(.03)** 7.4(.31)*** β1 = β14 1.0(.10) β1 = β15 3.8(.05)*

Appendix B.2 Raw Returns: The Long Run Performance of Direct versus Indirect Trades

This table presents the long run relative trading performance of direct versus indirect trades over different periods that extend from month t+1 through month t+a. In this Table, we analyze whether the short run performance extends beyond one month, by considering returns over different time frames that span the three years following insider trades. For this analysis, we estimate a revised version of Equation (6) that replaces Rj,t+1 as the dependent variable with CRj,t+1,t+a, as follows:

CRj,t+1,t+a = α + β1 TrSize_me_RKi,j,t + β2 TrSize_fam_RKi,j,t + β3 TrSize_trust-notfam_RKi,j,t

+ β4 TrSize_ret-notfam_RKi,j,t + β5 TrSize_found_RKi,j,t + Controlsi,j,t + εi,j,t . (7)

The first column shows the results for one-month-ahead returns (Rj,t+1) that are similar to our main model in column (9) of Table 5. The remaining columns present the analogous results for performance measured using raw returns over longer time frames extending 36 months into the future. We include the same controls as our analysis of Equation (1) in Table 5. The coefficients of the controls have similar implications to those presented in Table 6, and are omitted here for brevity. All variables are defined in Table 1. Monthly fixed effects are included, and standard errors are clustered by time at the monthly level. The t-statistics are provided in parentheses below the parameter estimates. F-statistics that test the equality of different pairs of parameter estimates are provided at the bottom of each column. Throughout this table, * indicates significance at the .10 level; ** at the .05 level; and *** at the .01 level.

(1) (2) (3) (4) (5) (6)

VARIABLES Ri,j,t+1 CRj,t+1,t+3 CRj,t+1,t+6 CRj,t+1,t+12 CRj,t+1,t+24 CRj,t+1,t+36

TrSize_me_RK β1 0.327*** 0.905*** 1.161*** 1.957*** 2.979*** 3.413*** (0.117) (0.242) (0.325) (0.463) (0.885) (1.193) TrSize_fam_RK β2 0.735*** 1.455*** 3.063*** 3.802*** 6.583*** 5.041** (0.243) (0.478) (0.867) (1.073) (1.655) (2.080) TrSize_trust-notfam_RK β3 1.007*** 1.333*** 1.999** 4.520*** 11.10*** 14.39*** (0.283) (0.467) (0.771) (1.089) (1.777) (2.445) TrSize_ret_notfam_RK β4 0.884 0.939 -0.0127 1.524 0.758 3.699 (0.547) (0.961) (1.599) (2.256) (3.166) (4.471) TrSize_found_RK β5 -0.968 -0.540 -1.868 -2.585 -1.355 5.018 (0.611) (1.168) (2.141) (3.398) (5.108) (7.595) Controls Yes Yes Yes Yes Yes Yes N 325,997 324,296 321,569 303,716 266,396 235,128 2 Adj. R 0.154 0.169 0.149 0.132 0.179 0.140

Hi: Testing equality; F-Stat (p-value)

1. β1 = β2 3.7(.06)* 1.6(.21) 6.4(.01)** 3.5(.06)* 5.6(.02)** 0.6(.46)

2. β1 = β3 7.1(.00)*** 0.98(.32) 1.5(.23) 6.4(.01)** 22.8(.000)*** 26.4(.00)***

3. β1 = β4 1.0(.32) 0.0(.97) 0.5(.49)* 0.0(.86) 0.4(.52) 0.0(.95)

4. β1 = β5 4.7(.03)* 1.6(.20) 2.0(.16) 1.9(.18) 0.8(.39) 0.0(.83)

5. β2 = β3 0.7(.39) 0.0(.84) 1.2(.27) 0.3(.60) 4.0(.05)** 10.5(.00)***

6. β2 = β4 0.1(.80) 0.21(.65) 2.5(.12) 0.8(.39) 2.9(.09)* 0.0(.78)

7. β2 = β5 6.3(.01)** 2.5(.11) 4.5(.04)* 3.7(.07)* 2.5(.12) 0.0(1.00)