INFORMATION NETWORKS: EVIDENCE FROM ILLEGAL INSIDER TRADING TIPS KENNETH R. AHERN† ABSTRACT Thispaperexploits detailed datafromillegal insidertradingcases tostudyhowprivateinformation diffuses across investors through social networks. I find that the majority of inside traders are connected through family and friendship links and a minority are connected through professional relationships. Traders cluster by age, occupation, gender, and location. Using inside information, traders earn prodigious returnsof about35% over 21 days. Traders farther from the original source earn lower percentage returns, but higher dollar gains. More broadly, this paper provides some of the first evidence on the transmission of information between stock market participants using direct observations of person-to-person communication. This version: 25 August 2014 † University of Southern California, Marshall School of Business, 3670 Trousdale Parkway, Los Angeles, CA 90089. E-mail: [email protected]. Information Networks: Evidence from Illegal Insider Trading Tips ABSTRACT Thispaperexploits detailed datafromillegal insidertradingcases tostudyhowprivateinformation diffuses across investors through social networks. I find that the majority of inside traders are connected through family and friendship links and a minority are connected through professional relationships. Traders cluster by age, occupation, gender, and location. Using inside information, traders earn prodigious returnsof about35% over 21 days. Traders farther from the original source earn lower percentage returns, but higher dollar gains. More broadly, this paper provides some of the first evidence on the transmission of information between stock market participants using direct observations of person-to-person communication. INFORMATION NETWORKS 1 Thoughinformationiscentraltounderstandingfinancialmarkets, weknowrelatively littleabout how private information spreads among market participants. A prominent line of research starting with Hayek (1945) argues that private information is revealed through trading and that market prices convey all relevant information.1 An alternative line of research argues that private informa- tion could also be conveyed through non-market social interactions between investors (Hong and Stein, 1999). A serious concern with this claim is that the supporting empirical evidence relies on imperfect proxies for social interaction, such as geographic proximity (Hong, Kubik, and Stein, 2005; Brown, Ivkovi´c, Smith, and Weisbenner, 2008), common educational backgrounds (Cohen, Frazzini, and Malloy, 2010), and correlated trades (Ozsoylev, Walden, Yavuz, and Bildak, 2014). While these proxies may capture social interactions, they could also reflect homophily among peo- ple of similar backgrounds. There is currently almost no direct evidence of actual communication between individual investors. In this paper, I use data from illegal insider trading cases to present some of the first direct evidence on the diffusion of private information across market participants. I provide an in-depth picture of where information originates, to whom and how fast it spreads, how tippers and tippees know each other, and the trading behavior and investment returns of individual investors. Com- bining all of the features of the data, I construct information networks of inside traders connected through word-of-mouth communication. Figure 1 presents an illustrative example of an information network in the data. This network is centered on Raj Rajaratnam, the former hedge fund manager of the Galleon Group. The con- nections between people in the network represent direct observations of word-of-mouth exchange of private information. For example, on March 14, 2007, an unidentified credit analyst at UBS learned through his job that Hellman & Friedman would acquire Kronos. He tipped his friend and roommate, Deep Shah, an analyst at Moody’s, who tipped a friend of the family, Roomy Khan. The following day, Roomy Khan tipped a family friend, Shammara Hussain, two former business associates, Jeffrey Yokuty and his boss, Robert Feinblatt, and another friend, Thomas Hardin. On March 19th, Hardin tipped his friend, Gautham Shankar, who tipped Zvi Goffer, David Plate, and unidentified traders at the investment firm Schottenfeld Group. David Plate subsequently tipped 1Seminal papers in this vein include Fama (1970), Grossman and Stiglitz (1980), and Kyle (1985). 2 INFORMATION NETWORKS others at Schottenfeld and Goffer tipped his long-time friend, Joseph Mancuso. The acquisition was officially announcedat 9 am on March 23, 2007. As a wholethe groupof insidetraders realized gains of $2.9 million on this tip. Illegal insider trading cases provide a number of important advantages for studying information networks. Prosecutors of insidertradingcases mustprovidedetailed recordsto prove thatmaterial, nonpublic information was shared between traders. This means that the information must be factual, rather than speculative. Second, for the trading to be illegal, the information must be private. This rules out public sources of information that would confound the study of private information exchange. Third,prosecutors mustprovide credible evidence of the timing and content of actual information exchange and trading activity. Finally, prosecutors must establish the type of social relationship between inside traders that would facilitate sharing private information. To exploit the richness of insider trading cases, I hand collect data from the narratives recorded in all of the cases filed by the Securities and Exchange Commission (SEC) and the Department of Justice (DOJ) between 2009 and 2013. The data cover 465 corporate events, 351 firms, and 611 inside traders. Using the original source data from the SEC and DOJ, I record the date that information is shared, the amount invested, the timing of trades, and the types of securities traded. I also record thetype of social relationship between insidetraders, such as family, friends,and busi- nessassociates. CombiningtheSECandDOJdatawithdatafromprofessionalnetworkingwebsites and the LexisNexis Public Records database, I fill in missing data on age, location, education, and occupation of the inside traders in the sample. In sum, these data provide an unprecedented view of how investors share private information. Using this comprehensive data, this paper’s main objective is to present a series of stylized facts about the flow of information across inside traders. First, I present a detailed profile of individual traders, the events on which they trade, the firms that are the subject of the information, and the traders’ investment returns. Second, I present a range of findings about how insiders are connected toeach otherthroughsocialrelationships. Third,Ianalyzetheflowofinformationfromtheoriginal source to the final tippee. Finally, I present the characteristics of information networks as a whole. INFORMATION NETWORKS 3 First, the data show that insider tips are about specific corporate events that have meaningful effects on stock prices. Merger-related events account for 51% of the sample, followed by earnings- related events, accounting for 26%. The remaining events include clinical trial and regulatory announcements, sale of new securities, and operational news such as CEO turnovers. Trading in advance of these events yields large returns. Across all types of events, the average stock return from the date of the original leak of information to the official announcement of the event is 34.9% over anaverage holdingperiodof 21tradingdays. Clinicaltrials generate thelargest average gains at 101% in 9 trading days. M&As generate average returns of 43% in 31 trading days. Earnings generate relatively smaller returns of 14% in 11 days. The firms that are the subject of insider trading tend to be large. The median sample firm’s market equity is $1 billion, comparable to the NYSE’s median equity of $1.2 billion in 2011. Comparedtotheuniverseoflistedfirmsandfirmstargetedinacquisitions,hightechpharmaceutical and electronics firms are overweighted and utilities and financial intermediaries are underweighted. Insider trading involves a wide array of people. The average age of inside traders is 43 years and about 10% are women. Compared to the general population, they tend to reside in New York, Florida, and California, and tend not to reside in Texas, Ohio, and Virginia. The median inside trader invests about $200,000 per tip, though some invest as little as a few thousand dollars, and others invest hundredsof millions. For these investments, traders earn about $72,000 per tip at the median. The most common occupation among insiders is top executive, including CEOs, CFOs, anddirectors. Thereareasignificantnumberofbuysideinvestment managersandanalysts, aswell as sell side professions, such as lawyers, accountants, and consultants. The sample also includes a numberofnon-“WallStreet”types,suchassmallbusinessowners,realestateprofessionals,doctors, engineers, nurses, teachers, and physical therapists. The next set of results document the social connections between inside traders. Of the 445 pairs of tippers and tippees in the sample, 23% are family members, 38% are business associates, and 38% are friends, including pairs that are both family members and business associates. Sibling and parental relations are the most common type of family connections. Of business associates, about half of the relationships are between a boss and subordinate or client and agent. The remaining half are associates of equal status. Across the whole sample, I findthat 74% of pairs of insidersmet 4 INFORMATION NETWORKS each other before college, 19% met during college, and 7% met after completing their education. Excluding family members, about 43% of pairs met during college. These results suggest that common educational background is a valid proxy for current information flows (Cohen, Frazzini, and Malloy, 2010), though family ties are a stronger proxy. People who share inside information tend to live close to each other. The median distance between a tipper and his tippee is 26 miles. This result validates the use of local neighborhoods as a proxy for social interaction as used in many papers on peer effects in finance (e.g., Brown, Ivkovi´c, Smith, and Weisbenner, 2008). However, a significant fraction of pairs do not live close to one another. The 75th percentile of distance is 739 miles. Geographic connections between locations are not random. For instance, people in New York are likely to be connected to people in Miami and San Francisco; people in Southern California are likely to be connected to people in Northern California. In contrast, people in Chicago, Dallas, and Atlanta have fewer connections than expected based on the general population. I next investigate the direction of information flows. Though tippers tend to tip other people in their same profession, top executives are three times more likely to be a tipper than a tippee. In contrast, buy side managers and analysts are tippers about half as often as they are tippees. Information tends to flow from subordinates to bosses and from younger tippers to older tippees. Women are more likely to tip and be tipped by other women. The original source of a tip depends upon the type of event that the information is about. Inside information about mergers is leaked by both acquirer and target employees, and external firms, such as law firms and investor relations firms. Earnings information is likely to be leaked by an accounting firm employee. In addition, a significant fraction of leaks are originated by people who secretly misappropriated the information from a friend or family member. To investigate how information flows across a network of traders, I identify “tip chains” in the networks. A tip chain is the ordered set of traders through which a particular tip passes. I find that as information diffuses away from the source, top executives and mid-level managers are less likely to send or receive tips. Instead, after three degrees of separation from the original source, buy side managers and analysts account for the majority of information sharing. The first links in a tip chain are more likely to be friends and family, but as the information diffuses further from INFORMATION NETWORKS 5 the source, business links become more prevalent. People who are closer to the original source of the information earn higher returns, but invest smaller amounts. People further from the source invest larger amounts and make smaller percentage returns, but larger dollar gains. The speed of information also increases as it moves further from the source. Finally, the last set of results document the structure of the networks of inside traders. Of 184 insider networks in the sample, 59 contain only one person. These are people who learn inside information, but do not tip anyone else. On the other end of the spectrum, the largest network has 50 members, and the second largest has 46. In the cross-section of networks, larger networks become less densewith fewer clusters of links. Thisimplies that peripheralmembersare notclosely connected to central members of a network. Instead, information networks sprawl outward. Larger networks have younger members and fewer women who are more likely connected through business relationships, compared to smaller networks. Using illegal insider trading cases to study information networks is not without limitations. The most important limitation is that insider trading cases do not represent a random sample of all information flows. Because insidertradingisillegal, itis potentially costly toshareinformation. To overcome the legal costs of insider trading, investors might only share tips about events with very large payoffs. They might also prefer to share information with trusted confidants, which would not be necessary if insider trading was legal. The second key limitation is that the sample of inside traders is based on those who were caught. This could bias the sample towards the most egregious violations, or alternatively, the least sophisticated inside traders. I discuss these limitations in greater detail later in the paper and conclude that the advantages of the data far outweigh their limitations. This paper contributes to two areas of research. First, to my knowledge, this paper presents the most detailed description of illegal insider trading to date. The most similar paper is Meulbroek (1992), which uses SEC cases from the 1980s to show that insider trades affect takeover prices. Subsequently, a number of other papers test whether illegal insider trading influences stock prices and takeover premia (Meulbroek and Hart, 1997; Chakravarty and McConnell, 1999; Fishe and Robe, 2004). Another set of papers considers whether the intensity of enforcement of insider trading laws affects financial markets (Bhattacharya and Daouk, 2002; Bushman, Piotroski, and 6 INFORMATION NETWORKS Smith, 2005; Bhattacharya and Marshall, 2012; Del Guercio, Odders-White, and Ready, 2013). In contrast, this paper focuses on the flow of information through the social connections of traders. Second, this paper contributes to a broader research agenda on social interactions in finance. Most directly, this paper contributes to the new field of information networks. Theoretical mod- els predict that the structure of information networks affect price informativeness, liquidity, and trading strategies (Colla and Mele, 2010; Ozsoylev and Walden, 2011; Walden, 2013; Han and Hir- shleifer, 2013; Han and Yang, 2013). However, apart from Ozsoylev, Walden, Yavuz, and Bildak (2014), which uses correlated trades to infer social connections, there is little empirical evidence on information networks. This paper also relates to research on peer effects in economics and finance. Manski (2000) argues that information sharing is one of the primary forms of non-market social in- teraction among economic agents. Recent research shows that peersinfluencestock market activity (Hong, Kubik, and Stein, 2004; Ivkovi´c and Weisbenner, 2007; Pool, Stoffman, and Yonker, 2014), CEO compensation and investment (Shue, 2013), and entrepreneurship (Lerner and Malmendier, 2013). Rather than relying on instrumental variables, this paper uses direct observation of social interactions to understand how information travels through a set of peers. I. Legal Environment and Sample Selection AccordingtotheSecuritiesandExchangeCommission(SEC),insidertradingrefersto“buyingor sellingasecurity, inbreachof afiduciarydutyorother relationshipof trustandconfidence, whilein possession of material, nonpublic information about the security.”2 Under U.S. law, insider trading is both a crime punishable by monetary penalties and imprisonment and a civil offense requiring disgorgement of illegal profits and payment of civil penalties. Criminal offenses are charged by the Department of Justice (DOJ) and civil offenses are charged by the SEC. Civil and criminal charges canbemadeatthesametimeforthesameoffense. Criminalchargesaremuchlesscommon,because criminal law requires evidence of guilt beyond a reasonable doubt in order to convict someone of a crime. In contrast, civil cases only require that it shows guilt based on the preponderance of the evidence. 2From theSEC’s website: http://www.sec.gov/answers/insider.htm. INFORMATION NETWORKS 7 Prosecution of illegal insider trading usually falls under Rule 10b-5 of the Securities Act of 1934. Whether a trade is covered by Rule 10b-5 is based on two theories. The classical theory applies to corporate insiders that purchase or sell securities on the basis of material, nonpublic information. Insidersinclude both employees of the firmand others whoreceive temporary access to confidential information,suchasexternallyhiredlawyersandaccountants. Themisappropriationtheoryapplies to anyone whouses confidential information for gain in breach of a fiduciary, contractual, or similar obligation to the rightful owner of the information (typically the firm). For more detail on the legal environment of insider trading see King, Corrigan, and Dukin (2009). For the study of information networks, using data from illegal insider trading cases offers certain advantages and limitations. The primary advantage is the credibility and level of detail provided in the case documentation. To support an accusation of illegal insider trading, the SEC and DOJ mustprovidecredibleevidence of how information is transmitted and how tradersknow each other. Thus, the documents provide explicit records of social relationships and communication, including phone records, emails, and text messages. This means I don’t need to prove that an instrument for social relations is valid —thedata aredirectobservations of socialrelations. Thesecond advantage is that I can directly observe the specific information that is shared between people, rather than guessing at the nature of the information. Third, I observe specific details of trading behavior, including the timing, amount, and type of security purchased. Finally, the case data provide the identities of the insiders. This allows me to trace the information from person to person. It also allows me to match individuals to outside data sources, unlike most data on individual traders, such as that of Barber and Odean (2000). The primary limitation of using case documents is selection bias. In particular, it is reasonable to assume that the magnitude of total insider trading is much larger than the sample of insider trading cases that are prosecuted. It is also reasonable to assume that the insiders that get caught by regulators are not randomly chosen. Instead, there are two forces that likely influence whether an insider is caught. First, sophisticated insiders are less likely to get caught than unsophisticated insiders. In the documents, I find that some insiders are oblivious to the regular monitoring of financial markets by regulators. For instance, in one case, a trader’s purchases accounted for 100% of the volume of out-of-the-money call options in the days before a merger announcement, sending 8 INFORMATION NETWORKS up red flags to regulators. The second force that influences whether an insider is caught is the extent of an insider’s activity. Regulators have a greater incentive to identify insiders who are investing larger sums and making more trades. To understand how these forces influence selection bias, consider Figure 2. This figure presents a stylized view of the population of inside traders and those that are most likely to be caught by regulators. There are four types of inside traders corresponding to the four quadrants in the figure: 1) unsophisticated with small assets under management, 2) unsophisticated with large assets, 3) sophisticated with small assets, and 4) sophisticated with large assets under management. The regulators are most likely to catch unsophisticated traders and traders making large investments. If insiders were evenly distributed across the four quadrants, the selection of insiders that are caught would pose a serious selection bias problem. However, the population of all inside traders is not random either. Instead, it is likely to have the greatest weight on unsophisticated traders with small investments and sophisticated traders with large investments. It is unlikely that there are many unsophisticated inside traders who have large amounts of capital to invest. This means that the type 3 investors (sophisticated investors with small amounts to invest) are the type that is most likely to be underrepresented in a sample of prosecutions. Though it is impossible to know type 3’s fraction of the insider population, it is plausible that sophistication and assets under management (whether in a fund or personal account) are positively related. This means that sophisticated traders likely have large amounts to invest. The empirical evidence supports these assumptions. First, the sample includes traders who trade very small amounts (a few thousand dollars). Thismeans thattheregulators donotonlytarget thebiggest insiders. Second, thesample also includes sophisticated traders who invest millions of dollars. In particular, the sample includes Steven Cohen, one of the most successful hedge fund managers of all time. A final selection issue is whether the people accused of insider trading are found guilty. First, the track record of the DOJ is impressive: since 2009, the DOJ has won 85 cases and lost just once. Therefore, the facts reported in the cases are likely to be true. Second, the SEC cases that are subsequently dropped are not typically dropped because the facts presented in the case are incorrect. They are usually dropped based on technical issues about what constitutes insider trading. For example, in the case of Donald Longueuil, the defendant’s attorneys argued that